diff options
Diffstat (limited to 'src/video_core')
171 files changed, 7988 insertions, 27151 deletions
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index e4de55f4d..007ecc13e 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -29,7 +29,6 @@ add_library(video_core STATIC dirty_flags.h dma_pusher.cpp dma_pusher.h - engines/const_buffer_engine_interface.h engines/const_buffer_info.h engines/engine_interface.h engines/engine_upload.cpp @@ -44,9 +43,6 @@ add_library(video_core STATIC engines/maxwell_3d.h engines/maxwell_dma.cpp engines/maxwell_dma.h - engines/shader_bytecode.h - engines/shader_header.h - engines/shader_type.h framebuffer_config.h macro/macro.cpp macro/macro.h @@ -61,8 +57,6 @@ add_library(video_core STATIC gpu.h gpu_thread.cpp gpu_thread.h - guest_driver.cpp - guest_driver.h memory_manager.cpp memory_manager.h query_cache.h @@ -71,26 +65,25 @@ add_library(video_core STATIC rasterizer_interface.h renderer_base.cpp renderer_base.h - renderer_opengl/gl_arb_decompiler.cpp - renderer_opengl/gl_arb_decompiler.h renderer_opengl/gl_buffer_cache.cpp renderer_opengl/gl_buffer_cache.h + renderer_opengl/gl_compute_pipeline.cpp + renderer_opengl/gl_compute_pipeline.h renderer_opengl/gl_device.cpp renderer_opengl/gl_device.h renderer_opengl/gl_fence_manager.cpp renderer_opengl/gl_fence_manager.h + renderer_opengl/gl_graphics_pipeline.cpp + renderer_opengl/gl_graphics_pipeline.h renderer_opengl/gl_rasterizer.cpp renderer_opengl/gl_rasterizer.h renderer_opengl/gl_resource_manager.cpp renderer_opengl/gl_resource_manager.h renderer_opengl/gl_shader_cache.cpp renderer_opengl/gl_shader_cache.h - renderer_opengl/gl_shader_decompiler.cpp - renderer_opengl/gl_shader_decompiler.h - renderer_opengl/gl_shader_disk_cache.cpp - renderer_opengl/gl_shader_disk_cache.h renderer_opengl/gl_shader_manager.cpp renderer_opengl/gl_shader_manager.h + renderer_opengl/gl_shader_context.h renderer_opengl/gl_shader_util.cpp renderer_opengl/gl_shader_util.h renderer_opengl/gl_state_tracker.cpp @@ -112,6 +105,7 @@ add_library(video_core STATIC renderer_vulkan/fixed_pipeline_state.h renderer_vulkan/maxwell_to_vk.cpp renderer_vulkan/maxwell_to_vk.h + renderer_vulkan/pipeline_helper.h renderer_vulkan/renderer_vulkan.h renderer_vulkan/renderer_vulkan.cpp renderer_vulkan/vk_blit_screen.cpp @@ -138,12 +132,12 @@ add_library(video_core STATIC renderer_vulkan/vk_query_cache.h renderer_vulkan/vk_rasterizer.cpp renderer_vulkan/vk_rasterizer.h + renderer_vulkan/vk_render_pass_cache.cpp + renderer_vulkan/vk_render_pass_cache.h renderer_vulkan/vk_resource_pool.cpp renderer_vulkan/vk_resource_pool.h renderer_vulkan/vk_scheduler.cpp renderer_vulkan/vk_scheduler.h - renderer_vulkan/vk_shader_decompiler.cpp - renderer_vulkan/vk_shader_decompiler.h renderer_vulkan/vk_shader_util.cpp renderer_vulkan/vk_shader_util.h renderer_vulkan/vk_staging_buffer_pool.cpp @@ -156,60 +150,12 @@ add_library(video_core STATIC renderer_vulkan/vk_texture_cache.h renderer_vulkan/vk_update_descriptor.cpp renderer_vulkan/vk_update_descriptor.h + shader_cache.cpp shader_cache.h + shader_environment.cpp + shader_environment.h shader_notify.cpp shader_notify.h - shader/decode/arithmetic.cpp - shader/decode/arithmetic_immediate.cpp - shader/decode/bfe.cpp - shader/decode/bfi.cpp - shader/decode/shift.cpp - shader/decode/arithmetic_integer.cpp - shader/decode/arithmetic_integer_immediate.cpp - shader/decode/arithmetic_half.cpp - shader/decode/arithmetic_half_immediate.cpp - shader/decode/ffma.cpp - shader/decode/hfma2.cpp - shader/decode/conversion.cpp - shader/decode/memory.cpp - shader/decode/texture.cpp - shader/decode/image.cpp - shader/decode/float_set_predicate.cpp - shader/decode/integer_set_predicate.cpp - shader/decode/half_set_predicate.cpp - shader/decode/predicate_set_register.cpp - shader/decode/predicate_set_predicate.cpp - shader/decode/register_set_predicate.cpp - shader/decode/float_set.cpp - shader/decode/integer_set.cpp - shader/decode/half_set.cpp - shader/decode/video.cpp - shader/decode/warp.cpp - shader/decode/xmad.cpp - shader/decode/other.cpp - shader/ast.cpp - shader/ast.h - shader/async_shaders.cpp - shader/async_shaders.h - shader/compiler_settings.cpp - shader/compiler_settings.h - shader/control_flow.cpp - shader/control_flow.h - shader/decode.cpp - shader/expr.cpp - shader/expr.h - shader/memory_util.cpp - shader/memory_util.h - shader/node_helper.cpp - shader/node_helper.h - shader/node.h - shader/registry.cpp - shader/registry.h - shader/shader_ir.cpp - shader/shader_ir.h - shader/track.cpp - shader/transform_feedback.cpp - shader/transform_feedback.h surface.cpp surface.h texture_cache/accelerated_swizzle.cpp @@ -242,6 +188,8 @@ add_library(video_core STATIC textures/decoders.h textures/texture.cpp textures/texture.h + transform_feedback.cpp + transform_feedback.h video_core.cpp video_core.h vulkan_common/vulkan_debug_callback.cpp @@ -265,7 +213,7 @@ add_library(video_core STATIC create_target_directory_groups(video_core) target_link_libraries(video_core PUBLIC common core) -target_link_libraries(video_core PRIVATE glad xbyak) +target_link_libraries(video_core PUBLIC glad shader_recompiler xbyak) if (YUZU_USE_BUNDLED_FFMPEG AND NOT WIN32) add_dependencies(video_core ffmpeg-build) diff --git a/src/video_core/buffer_cache/buffer_base.h b/src/video_core/buffer_cache/buffer_base.h index b121d36a3..c3318095c 100644 --- a/src/video_core/buffer_cache/buffer_base.h +++ b/src/video_core/buffer_cache/buffer_base.h @@ -226,19 +226,24 @@ public: /// Call 'func' for each CPU modified range and unmark those pages as CPU modified template <typename Func> void ForEachUploadRange(VAddr query_cpu_range, u64 size, Func&& func) { - ForEachModifiedRange<Type::CPU>(query_cpu_range, size, func); + ForEachModifiedRange<Type::CPU>(query_cpu_range, size, true, func); } /// Call 'func' for each GPU modified range and unmark those pages as GPU modified template <typename Func> - void ForEachDownloadRange(VAddr query_cpu_range, u64 size, Func&& func) { - ForEachModifiedRange<Type::GPU>(query_cpu_range, size, func); + void ForEachDownloadRange(VAddr query_cpu_range, u64 size, bool clear, Func&& func) { + ForEachModifiedRange<Type::GPU>(query_cpu_range, size, clear, func); + } + + template <typename Func> + void ForEachDownloadRangeAndClear(VAddr query_cpu_range, u64 size, Func&& func) { + ForEachModifiedRange<Type::GPU>(query_cpu_range, size, true, func); } /// Call 'func' for each GPU modified range and unmark those pages as GPU modified template <typename Func> void ForEachDownloadRange(Func&& func) { - ForEachModifiedRange<Type::GPU>(cpu_addr, SizeBytes(), func); + ForEachModifiedRange<Type::GPU>(cpu_addr, SizeBytes(), true, func); } /// Mark buffer as picked @@ -415,7 +420,7 @@ private: * @param func Function to call for each turned off region */ template <Type type, typename Func> - void ForEachModifiedRange(VAddr query_cpu_range, s64 size, Func&& func) { + void ForEachModifiedRange(VAddr query_cpu_range, s64 size, bool clear, Func&& func) { static_assert(type != Type::Untracked); const s64 difference = query_cpu_range - cpu_addr; @@ -467,7 +472,9 @@ private: bits = (bits << left_offset) >> left_offset; const u64 current_word = state_words[word_index] & bits; - state_words[word_index] &= ~bits; + if (clear) { + state_words[word_index] &= ~bits; + } if constexpr (type == Type::CPU) { const u64 current_bits = untracked_words[word_index] & bits; diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index cad7f902d..24c858104 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -15,6 +15,7 @@ #include <vector> #include <boost/container/small_vector.hpp> +#include <boost/icl/interval_set.hpp> #include "common/common_types.h" #include "common/div_ceil.h" @@ -30,6 +31,7 @@ #include "video_core/engines/maxwell_3d.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" +#include "video_core/surface.h" #include "video_core/texture_cache/slot_vector.h" #include "video_core/texture_cache/types.h" @@ -41,14 +43,19 @@ MICROPROFILE_DECLARE(GPU_DownloadMemory); using BufferId = SlotId; +using VideoCore::Surface::PixelFormat; +using namespace Common::Literals; + constexpr u32 NUM_VERTEX_BUFFERS = 32; constexpr u32 NUM_TRANSFORM_FEEDBACK_BUFFERS = 4; constexpr u32 NUM_GRAPHICS_UNIFORM_BUFFERS = 18; constexpr u32 NUM_COMPUTE_UNIFORM_BUFFERS = 8; constexpr u32 NUM_STORAGE_BUFFERS = 16; +constexpr u32 NUM_TEXTURE_BUFFERS = 16; constexpr u32 NUM_STAGES = 5; -using namespace Common::Literals; +using UniformBufferSizes = std::array<std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES>; +using ComputeUniformBufferSizes = std::array<u32, NUM_COMPUTE_UNIFORM_BUFFERS>; template <typename P> class BufferCache { @@ -66,6 +73,7 @@ class BufferCache { static constexpr bool NEEDS_BIND_UNIFORM_INDEX = P::NEEDS_BIND_UNIFORM_INDEX; static constexpr bool NEEDS_BIND_STORAGE_INDEX = P::NEEDS_BIND_STORAGE_INDEX; static constexpr bool USE_MEMORY_MAPS = P::USE_MEMORY_MAPS; + static constexpr bool SEPARATE_IMAGE_BUFFERS_BINDINGS = P::SEPARATE_IMAGE_BUFFER_BINDINGS; static constexpr BufferId NULL_BUFFER_ID{0}; @@ -77,6 +85,9 @@ class BufferCache { using Runtime = typename P::Runtime; using Buffer = typename P::Buffer; + using IntervalSet = boost::icl::interval_set<VAddr>; + using IntervalType = typename IntervalSet::interval_type; + struct Empty {}; struct OverlapResult { @@ -92,6 +103,10 @@ class BufferCache { BufferId buffer_id; }; + struct TextureBufferBinding : Binding { + PixelFormat format; + }; + static constexpr Binding NULL_BINDING{ .cpu_addr = 0, .size = 0, @@ -129,38 +144,63 @@ public: void BindHostComputeBuffers(); - void SetEnabledUniformBuffers(size_t stage, u32 enabled); + void SetUniformBuffersState(const std::array<u32, NUM_STAGES>& mask, + const UniformBufferSizes* sizes); - void SetEnabledComputeUniformBuffers(u32 enabled); + void SetComputeUniformBufferState(u32 mask, const ComputeUniformBufferSizes* sizes); void UnbindGraphicsStorageBuffers(size_t stage); void BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset, bool is_written); + void UnbindGraphicsTextureBuffers(size_t stage); + + void BindGraphicsTextureBuffer(size_t stage, size_t tbo_index, GPUVAddr gpu_addr, u32 size, + PixelFormat format, bool is_written, bool is_image); + void UnbindComputeStorageBuffers(); void BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset, bool is_written); + void UnbindComputeTextureBuffers(); + + void BindComputeTextureBuffer(size_t tbo_index, GPUVAddr gpu_addr, u32 size, PixelFormat format, + bool is_written, bool is_image); + void FlushCachedWrites(); /// Return true when there are uncommitted buffers to be downloaded [[nodiscard]] bool HasUncommittedFlushes() const noexcept; + void AccumulateFlushes(); + /// Return true when the caller should wait for async downloads [[nodiscard]] bool ShouldWaitAsyncFlushes() const noexcept; /// Commit asynchronous downloads void CommitAsyncFlushes(); + void CommitAsyncFlushesHigh(); /// Pop asynchronous downloads void PopAsyncFlushes(); + bool DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount); + + bool DMAClear(GPUVAddr src_address, u64 amount, u32 value); + /// Return true when a CPU region is modified from the GPU [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size); + /// Return true when a region is registered on the cache + [[nodiscard]] bool IsRegionRegistered(VAddr addr, size_t size); + + /// Return true when a CPU region is modified from the CPU + [[nodiscard]] bool IsRegionCpuModified(VAddr addr, size_t size); + std::mutex mutex; + Runtime& runtime; private: template <typename Func> @@ -190,6 +230,36 @@ private: } } + template <typename Func> + void ForEachWrittenRange(VAddr cpu_addr, u64 size, Func&& func) { + const VAddr start_address = cpu_addr; + const VAddr end_address = start_address + size; + const VAddr search_base = + static_cast<VAddr>(std::min<s64>(0LL, static_cast<s64>(start_address - size))); + const IntervalType search_interval{search_base, search_base + 1}; + auto it = common_ranges.lower_bound(search_interval); + if (it == common_ranges.end()) { + it = common_ranges.begin(); + } + for (; it != common_ranges.end(); it++) { + VAddr inter_addr_end = it->upper(); + VAddr inter_addr = it->lower(); + if (inter_addr >= end_address) { + break; + } + if (inter_addr_end <= start_address) { + continue; + } + if (inter_addr_end > end_address) { + inter_addr_end = end_address; + } + if (inter_addr < start_address) { + inter_addr = start_address; + } + func(inter_addr, inter_addr_end); + } + } + static bool IsRangeGranular(VAddr cpu_addr, size_t size) { return (cpu_addr & ~Core::Memory::PAGE_MASK) == ((cpu_addr + size) & ~Core::Memory::PAGE_MASK); @@ -207,12 +277,16 @@ private: void BindHostGraphicsStorageBuffers(size_t stage); + void BindHostGraphicsTextureBuffers(size_t stage); + void BindHostTransformFeedbackBuffers(); void BindHostComputeUniformBuffers(); void BindHostComputeStorageBuffers(); + void BindHostComputeTextureBuffers(); + void DoUpdateGraphicsBuffers(bool is_indexed); void DoUpdateComputeBuffers(); @@ -227,6 +301,8 @@ private: void UpdateStorageBuffers(size_t stage); + void UpdateTextureBuffers(size_t stage); + void UpdateTransformFeedbackBuffers(); void UpdateTransformFeedbackBuffer(u32 index); @@ -235,6 +311,8 @@ private: void UpdateComputeStorageBuffers(); + void UpdateComputeTextureBuffers(); + void MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 size); [[nodiscard]] BufferId FindBuffer(VAddr cpu_addr, u32 size); @@ -272,24 +350,26 @@ private: void DeleteBuffer(BufferId buffer_id); - void ReplaceBufferDownloads(BufferId old_buffer_id, BufferId new_buffer_id); - void NotifyBufferDeletion(); [[nodiscard]] Binding StorageBufferBinding(GPUVAddr ssbo_addr) const; + [[nodiscard]] TextureBufferBinding GetTextureBufferBinding(GPUVAddr gpu_addr, u32 size, + PixelFormat format); + [[nodiscard]] std::span<const u8> ImmediateBufferWithData(VAddr cpu_addr, size_t size); [[nodiscard]] std::span<u8> ImmediateBuffer(size_t wanted_capacity); [[nodiscard]] bool HasFastUniformBufferBound(size_t stage, u32 binding_index) const noexcept; + void ClearDownload(IntervalType subtract_interval); + VideoCore::RasterizerInterface& rasterizer; Tegra::Engines::Maxwell3D& maxwell3d; Tegra::Engines::KeplerCompute& kepler_compute; Tegra::MemoryManager& gpu_memory; Core::Memory::Memory& cpu_memory; - Runtime& runtime; SlotVector<Buffer> slot_buffers; DelayedDestructionRing<Buffer, 8> delayed_destruction_ring; @@ -300,20 +380,30 @@ private: std::array<Binding, NUM_VERTEX_BUFFERS> vertex_buffers; std::array<std::array<Binding, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES> uniform_buffers; std::array<std::array<Binding, NUM_STORAGE_BUFFERS>, NUM_STAGES> storage_buffers; + std::array<std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS>, NUM_STAGES> texture_buffers; std::array<Binding, NUM_TRANSFORM_FEEDBACK_BUFFERS> transform_feedback_buffers; std::array<Binding, NUM_COMPUTE_UNIFORM_BUFFERS> compute_uniform_buffers; std::array<Binding, NUM_STORAGE_BUFFERS> compute_storage_buffers; + std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS> compute_texture_buffers; + + std::array<u32, NUM_STAGES> enabled_uniform_buffer_masks{}; + u32 enabled_compute_uniform_buffer_mask = 0; - std::array<u32, NUM_STAGES> enabled_uniform_buffers{}; - u32 enabled_compute_uniform_buffers = 0; + const UniformBufferSizes* uniform_buffer_sizes{}; + const ComputeUniformBufferSizes* compute_uniform_buffer_sizes{}; std::array<u32, NUM_STAGES> enabled_storage_buffers{}; std::array<u32, NUM_STAGES> written_storage_buffers{}; u32 enabled_compute_storage_buffers = 0; u32 written_compute_storage_buffers = 0; - std::array<u32, NUM_STAGES> fast_bound_uniform_buffers{}; + std::array<u32, NUM_STAGES> enabled_texture_buffers{}; + std::array<u32, NUM_STAGES> written_texture_buffers{}; + std::array<u32, NUM_STAGES> image_texture_buffers{}; + u32 enabled_compute_texture_buffers = 0; + u32 written_compute_texture_buffers = 0; + u32 image_compute_texture_buffers = 0; std::array<u32, 16> uniform_cache_hits{}; std::array<u32, 16> uniform_cache_shots{}; @@ -324,12 +414,16 @@ private: std::conditional_t<HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS, std::array<u32, NUM_STAGES>, Empty> dirty_uniform_buffers{}; + std::conditional_t<IS_OPENGL, std::array<u32, NUM_STAGES>, Empty> fast_bound_uniform_buffers{}; + std::conditional_t<HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS, + std::array<std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES>, Empty> + uniform_buffer_binding_sizes{}; std::vector<BufferId> cached_write_buffer_ids; - // TODO: This data structure is not optimal and it should be reworked - std::vector<BufferId> uncommitted_downloads; - std::deque<std::vector<BufferId>> committed_downloads; + IntervalSet uncommitted_ranges; + IntervalSet common_ranges; + std::deque<IntervalSet> committed_ranges; size_t immediate_buffer_capacity = 0; std::unique_ptr<u8[]> immediate_buffer_alloc; @@ -347,11 +441,12 @@ BufferCache<P>::BufferCache(VideoCore::RasterizerInterface& rasterizer_, Tegra::Engines::KeplerCompute& kepler_compute_, Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, Runtime& runtime_) - : rasterizer{rasterizer_}, maxwell3d{maxwell3d_}, kepler_compute{kepler_compute_}, - gpu_memory{gpu_memory_}, cpu_memory{cpu_memory_}, runtime{runtime_} { + : runtime{runtime_}, rasterizer{rasterizer_}, maxwell3d{maxwell3d_}, + kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_}, cpu_memory{cpu_memory_} { // Ensure the first slot is used for the null buffer void(slot_buffers.insert(runtime, NullBufferParams{})); deletion_iterator = slot_buffers.end(); + common_ranges.clear(); } template <class P> @@ -422,6 +517,97 @@ void BufferCache<P>::DownloadMemory(VAddr cpu_addr, u64 size) { } template <class P> +void BufferCache<P>::ClearDownload(IntervalType subtract_interval) { + uncommitted_ranges.subtract(subtract_interval); + for (auto& interval_set : committed_ranges) { + interval_set.subtract(subtract_interval); + } +} + +template <class P> +bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) { + const std::optional<VAddr> cpu_src_address = gpu_memory.GpuToCpuAddress(src_address); + const std::optional<VAddr> cpu_dest_address = gpu_memory.GpuToCpuAddress(dest_address); + if (!cpu_src_address || !cpu_dest_address) { + return false; + } + const bool source_dirty = IsRegionRegistered(*cpu_src_address, amount); + const bool dest_dirty = IsRegionRegistered(*cpu_dest_address, amount); + if (!source_dirty && !dest_dirty) { + return false; + } + + const IntervalType subtract_interval{*cpu_dest_address, *cpu_dest_address + amount}; + ClearDownload(subtract_interval); + + BufferId buffer_a; + BufferId buffer_b; + do { + has_deleted_buffers = false; + buffer_a = FindBuffer(*cpu_src_address, static_cast<u32>(amount)); + buffer_b = FindBuffer(*cpu_dest_address, static_cast<u32>(amount)); + } while (has_deleted_buffers); + auto& src_buffer = slot_buffers[buffer_a]; + auto& dest_buffer = slot_buffers[buffer_b]; + SynchronizeBuffer(src_buffer, *cpu_src_address, static_cast<u32>(amount)); + SynchronizeBuffer(dest_buffer, *cpu_dest_address, static_cast<u32>(amount)); + std::array copies{BufferCopy{ + .src_offset = src_buffer.Offset(*cpu_src_address), + .dst_offset = dest_buffer.Offset(*cpu_dest_address), + .size = amount, + }}; + + boost::container::small_vector<IntervalType, 4> tmp_intervals; + auto mirror = [&](VAddr base_address, VAddr base_address_end) { + const u64 size = base_address_end - base_address; + const VAddr diff = base_address - *cpu_src_address; + const VAddr new_base_address = *cpu_dest_address + diff; + const IntervalType add_interval{new_base_address, new_base_address + size}; + uncommitted_ranges.add(add_interval); + tmp_intervals.push_back(add_interval); + }; + ForEachWrittenRange(*cpu_src_address, amount, mirror); + // This subtraction in this order is important for overlapping copies. + common_ranges.subtract(subtract_interval); + bool atleast_1_download = tmp_intervals.size() != 0; + for (const IntervalType add_interval : tmp_intervals) { + common_ranges.add(add_interval); + } + + runtime.CopyBuffer(dest_buffer, src_buffer, copies); + if (atleast_1_download) { + dest_buffer.MarkRegionAsGpuModified(*cpu_dest_address, amount); + } + std::vector<u8> tmp_buffer(amount); + cpu_memory.ReadBlockUnsafe(*cpu_src_address, tmp_buffer.data(), amount); + cpu_memory.WriteBlockUnsafe(*cpu_dest_address, tmp_buffer.data(), amount); + return true; +} + +template <class P> +bool BufferCache<P>::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) { + const std::optional<VAddr> cpu_dst_address = gpu_memory.GpuToCpuAddress(dst_address); + if (!cpu_dst_address) { + return false; + } + const bool dest_dirty = IsRegionRegistered(*cpu_dst_address, amount); + if (!dest_dirty) { + return false; + } + + const size_t size = amount * sizeof(u32); + const IntervalType subtract_interval{*cpu_dst_address, *cpu_dst_address + size}; + ClearDownload(subtract_interval); + common_ranges.subtract(subtract_interval); + + const BufferId buffer = FindBuffer(*cpu_dst_address, static_cast<u32>(size)); + auto& dest_buffer = slot_buffers[buffer]; + const u32 offset = dest_buffer.Offset(*cpu_dst_address); + runtime.ClearBuffer(dest_buffer, offset, size, value); + return true; +} + +template <class P> void BufferCache<P>::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) { const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); @@ -476,6 +662,7 @@ void BufferCache<P>::BindHostStageBuffers(size_t stage) { MICROPROFILE_SCOPE(GPU_BindUploadBuffers); BindHostGraphicsUniformBuffers(stage); BindHostGraphicsStorageBuffers(stage); + BindHostGraphicsTextureBuffers(stage); } template <class P> @@ -483,21 +670,30 @@ void BufferCache<P>::BindHostComputeBuffers() { MICROPROFILE_SCOPE(GPU_BindUploadBuffers); BindHostComputeUniformBuffers(); BindHostComputeStorageBuffers(); + BindHostComputeTextureBuffers(); } template <class P> -void BufferCache<P>::SetEnabledUniformBuffers(size_t stage, u32 enabled) { +void BufferCache<P>::SetUniformBuffersState(const std::array<u32, NUM_STAGES>& mask, + const UniformBufferSizes* sizes) { if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { - if (enabled_uniform_buffers[stage] != enabled) { - dirty_uniform_buffers[stage] = ~u32{0}; + if (enabled_uniform_buffer_masks != mask) { + if constexpr (IS_OPENGL) { + fast_bound_uniform_buffers.fill(0); + } + dirty_uniform_buffers.fill(~u32{0}); + uniform_buffer_binding_sizes.fill({}); } } - enabled_uniform_buffers[stage] = enabled; + enabled_uniform_buffer_masks = mask; + uniform_buffer_sizes = sizes; } template <class P> -void BufferCache<P>::SetEnabledComputeUniformBuffers(u32 enabled) { - enabled_compute_uniform_buffers = enabled; +void BufferCache<P>::SetComputeUniformBufferState(u32 mask, + const ComputeUniformBufferSizes* sizes) { + enabled_compute_uniform_buffer_mask = mask; + compute_uniform_buffer_sizes = sizes; } template <class P> @@ -518,9 +714,29 @@ void BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, } template <class P> +void BufferCache<P>::UnbindGraphicsTextureBuffers(size_t stage) { + enabled_texture_buffers[stage] = 0; + written_texture_buffers[stage] = 0; + image_texture_buffers[stage] = 0; +} + +template <class P> +void BufferCache<P>::BindGraphicsTextureBuffer(size_t stage, size_t tbo_index, GPUVAddr gpu_addr, + u32 size, PixelFormat format, bool is_written, + bool is_image) { + enabled_texture_buffers[stage] |= 1U << tbo_index; + written_texture_buffers[stage] |= (is_written ? 1U : 0U) << tbo_index; + if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) { + image_texture_buffers[stage] |= (is_image ? 1U : 0U) << tbo_index; + } + texture_buffers[stage][tbo_index] = GetTextureBufferBinding(gpu_addr, size, format); +} + +template <class P> void BufferCache<P>::UnbindComputeStorageBuffers() { enabled_compute_storage_buffers = 0; written_compute_storage_buffers = 0; + image_compute_texture_buffers = 0; } template <class P> @@ -538,6 +754,24 @@ void BufferCache<P>::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, } template <class P> +void BufferCache<P>::UnbindComputeTextureBuffers() { + enabled_compute_texture_buffers = 0; + written_compute_texture_buffers = 0; + image_compute_texture_buffers = 0; +} + +template <class P> +void BufferCache<P>::BindComputeTextureBuffer(size_t tbo_index, GPUVAddr gpu_addr, u32 size, + PixelFormat format, bool is_written, bool is_image) { + enabled_compute_texture_buffers |= 1U << tbo_index; + written_compute_texture_buffers |= (is_written ? 1U : 0U) << tbo_index; + if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) { + image_compute_texture_buffers |= (is_image ? 1U : 0U) << tbo_index; + } + compute_texture_buffers[tbo_index] = GetTextureBufferBinding(gpu_addr, size, format); +} + +template <class P> void BufferCache<P>::FlushCachedWrites() { for (const BufferId buffer_id : cached_write_buffer_ids) { slot_buffers[buffer_id].FlushCachedWrites(); @@ -547,29 +781,30 @@ void BufferCache<P>::FlushCachedWrites() { template <class P> bool BufferCache<P>::HasUncommittedFlushes() const noexcept { - return !uncommitted_downloads.empty(); + return !uncommitted_ranges.empty() || !committed_ranges.empty(); } template <class P> -bool BufferCache<P>::ShouldWaitAsyncFlushes() const noexcept { - return !committed_downloads.empty() && !committed_downloads.front().empty(); +void BufferCache<P>::AccumulateFlushes() { + if (Settings::values.gpu_accuracy.GetValue() != Settings::GPUAccuracy::High) { + uncommitted_ranges.clear(); + return; + } + if (uncommitted_ranges.empty()) { + return; + } + committed_ranges.emplace_back(std::move(uncommitted_ranges)); } template <class P> -void BufferCache<P>::CommitAsyncFlushes() { - // This is intentionally passing the value by copy - committed_downloads.push_front(uncommitted_downloads); - uncommitted_downloads.clear(); +bool BufferCache<P>::ShouldWaitAsyncFlushes() const noexcept { + return false; } template <class P> -void BufferCache<P>::PopAsyncFlushes() { - if (committed_downloads.empty()) { - return; - } - auto scope_exit_pop_download = detail::ScopeExit([this] { committed_downloads.pop_back(); }); - const std::span<const BufferId> download_ids = committed_downloads.back(); - if (download_ids.empty()) { +void BufferCache<P>::CommitAsyncFlushesHigh() { + AccumulateFlushes(); + if (committed_ranges.empty()) { return; } MICROPROFILE_SCOPE(GPU_DownloadMemory); @@ -577,20 +812,43 @@ void BufferCache<P>::PopAsyncFlushes() { boost::container::small_vector<std::pair<BufferCopy, BufferId>, 1> downloads; u64 total_size_bytes = 0; u64 largest_copy = 0; - for (const BufferId buffer_id : download_ids) { - slot_buffers[buffer_id].ForEachDownloadRange([&](u64 range_offset, u64 range_size) { - downloads.push_back({ - BufferCopy{ - .src_offset = range_offset, - .dst_offset = total_size_bytes, - .size = range_size, - }, - buffer_id, + for (const IntervalSet& intervals : committed_ranges) { + for (auto& interval : intervals) { + const std::size_t size = interval.upper() - interval.lower(); + const VAddr cpu_addr = interval.lower(); + ForEachBufferInRange(cpu_addr, size, [&](BufferId buffer_id, Buffer& buffer) { + boost::container::small_vector<BufferCopy, 1> copies; + buffer.ForEachDownloadRangeAndClear( + cpu_addr, size, [&](u64 range_offset, u64 range_size) { + const VAddr buffer_addr = buffer.CpuAddr(); + const auto add_download = [&](VAddr start, VAddr end) { + const u64 new_offset = start - buffer_addr; + const u64 new_size = end - start; + downloads.push_back({ + BufferCopy{ + .src_offset = new_offset, + .dst_offset = total_size_bytes, + .size = new_size, + }, + buffer_id, + }); + // Align up to avoid cache conflicts + constexpr u64 align = 256ULL; + constexpr u64 mask = ~(align - 1ULL); + total_size_bytes += (new_size + align - 1) & mask; + largest_copy = std::max(largest_copy, new_size); + }; + + const VAddr start_address = buffer_addr + range_offset; + const VAddr end_address = start_address + range_size; + ForEachWrittenRange(start_address, range_size, add_download); + const IntervalType subtract_interval{start_address, end_address}; + common_ranges.subtract(subtract_interval); + }); }); - total_size_bytes += range_size; - largest_copy = std::max(largest_copy, range_size); - }); + } } + committed_ranges.clear(); if (downloads.empty()) { return; } @@ -623,6 +881,19 @@ void BufferCache<P>::PopAsyncFlushes() { } template <class P> +void BufferCache<P>::CommitAsyncFlushes() { + if (Settings::values.gpu_accuracy.GetValue() == Settings::GPUAccuracy::High) { + CommitAsyncFlushesHigh(); + } else { + uncommitted_ranges.clear(); + committed_ranges.clear(); + } +} + +template <class P> +void BufferCache<P>::PopAsyncFlushes() {} + +template <class P> bool BufferCache<P>::IsRegionGpuModified(VAddr addr, size_t size) { const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE); for (u64 page = addr >> PAGE_BITS; page < page_end;) { @@ -642,6 +913,46 @@ bool BufferCache<P>::IsRegionGpuModified(VAddr addr, size_t size) { } template <class P> +bool BufferCache<P>::IsRegionRegistered(VAddr addr, size_t size) { + const VAddr end_addr = addr + size; + const u64 page_end = Common::DivCeil(end_addr, PAGE_SIZE); + for (u64 page = addr >> PAGE_BITS; page < page_end;) { + const BufferId buffer_id = page_table[page]; + if (!buffer_id) { + ++page; + continue; + } + Buffer& buffer = slot_buffers[buffer_id]; + const VAddr buf_start_addr = buffer.CpuAddr(); + const VAddr buf_end_addr = buf_start_addr + buffer.SizeBytes(); + if (buf_start_addr < end_addr && addr < buf_end_addr) { + return true; + } + page = Common::DivCeil(end_addr, PAGE_SIZE); + } + return false; +} + +template <class P> +bool BufferCache<P>::IsRegionCpuModified(VAddr addr, size_t size) { + const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE); + for (u64 page = addr >> PAGE_BITS; page < page_end;) { + const BufferId image_id = page_table[page]; + if (!image_id) { + ++page; + continue; + } + Buffer& buffer = slot_buffers[image_id]; + if (buffer.IsRegionCpuModified(addr, size)) { + return true; + } + const VAddr end_addr = buffer.CpuAddr() + buffer.SizeBytes(); + page = Common::DivCeil(end_addr, PAGE_SIZE); + } + return false; +} + +template <class P> void BufferCache<P>::BindHostIndexBuffer() { Buffer& buffer = slot_buffers[index_buffer.buffer_id]; TouchBuffer(buffer); @@ -649,7 +960,9 @@ void BufferCache<P>::BindHostIndexBuffer() { const u32 size = index_buffer.size; SynchronizeBuffer(buffer, index_buffer.cpu_addr, size); if constexpr (HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) { - runtime.BindIndexBuffer(buffer, offset, size); + const u32 new_offset = offset + maxwell3d.regs.index_array.first * + maxwell3d.regs.index_array.FormatSizeInBytes(); + runtime.BindIndexBuffer(buffer, new_offset, size); } else { runtime.BindIndexBuffer(maxwell3d.regs.draw.topology, maxwell3d.regs.index_array.format, maxwell3d.regs.index_array.first, maxwell3d.regs.index_array.count, @@ -683,7 +996,7 @@ void BufferCache<P>::BindHostGraphicsUniformBuffers(size_t stage) { dirty = std::exchange(dirty_uniform_buffers[stage], 0); } u32 binding_index = 0; - ForEachEnabledBit(enabled_uniform_buffers[stage], [&](u32 index) { + ForEachEnabledBit(enabled_uniform_buffer_masks[stage], [&](u32 index) { const bool needs_bind = ((dirty >> index) & 1) != 0; BindHostGraphicsUniformBuffer(stage, index, binding_index, needs_bind); if constexpr (NEEDS_BIND_UNIFORM_INDEX) { @@ -697,7 +1010,7 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 bool needs_bind) { const Binding& binding = uniform_buffers[stage][index]; const VAddr cpu_addr = binding.cpu_addr; - const u32 size = binding.size; + const u32 size = std::min(binding.size, (*uniform_buffer_sizes)[stage][index]); Buffer& buffer = slot_buffers[binding.buffer_id]; TouchBuffer(buffer); const bool use_fast_buffer = binding.buffer_id != NULL_BUFFER_ID && @@ -707,8 +1020,13 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 if constexpr (IS_OPENGL) { if (runtime.HasFastBufferSubData()) { // Fast path for Nvidia - if (!HasFastUniformBufferBound(stage, binding_index)) { + const bool should_fast_bind = + !HasFastUniformBufferBound(stage, binding_index) || + uniform_buffer_binding_sizes[stage][binding_index] != size; + if (should_fast_bind) { // We only have to bind when the currently bound buffer is not the fast version + fast_bound_uniform_buffers[stage] |= 1U << binding_index; + uniform_buffer_binding_sizes[stage][binding_index] = size; runtime.BindFastUniformBuffer(stage, binding_index, size); } const auto span = ImmediateBufferWithData(cpu_addr, size); @@ -716,8 +1034,10 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 return; } } - fast_bound_uniform_buffers[stage] |= 1U << binding_index; - + if constexpr (IS_OPENGL) { + fast_bound_uniform_buffers[stage] |= 1U << binding_index; + uniform_buffer_binding_sizes[stage][binding_index] = size; + } // Stream buffer path to avoid stalling on non-Nvidia drivers or Vulkan const std::span<u8> span = runtime.BindMappedUniformBuffer(stage, binding_index, size); cpu_memory.ReadBlockUnsafe(cpu_addr, span.data(), size); @@ -730,14 +1050,27 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 } ++uniform_cache_shots[0]; - if (!needs_bind && !HasFastUniformBufferBound(stage, binding_index)) { - // Skip binding if it's not needed and if the bound buffer is not the fast version - // This exists to avoid instances where the fast buffer is bound and a GPU write happens + // Skip binding if it's not needed and if the bound buffer is not the fast version + // This exists to avoid instances where the fast buffer is bound and a GPU write happens + needs_bind |= HasFastUniformBufferBound(stage, binding_index); + if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { + needs_bind |= uniform_buffer_binding_sizes[stage][binding_index] != size; + } + if (!needs_bind) { return; } - fast_bound_uniform_buffers[stage] &= ~(1U << binding_index); - const u32 offset = buffer.Offset(cpu_addr); + if constexpr (IS_OPENGL) { + // Fast buffer will be unbound + fast_bound_uniform_buffers[stage] &= ~(1U << binding_index); + + // Mark the index as dirty if offset doesn't match + const bool is_copy_bind = offset != 0 && !runtime.SupportsNonZeroUniformOffset(); + dirty_uniform_buffers[stage] |= (is_copy_bind ? 1U : 0U) << index; + } + if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { + uniform_buffer_binding_sizes[stage][binding_index] = size; + } if constexpr (NEEDS_BIND_UNIFORM_INDEX) { runtime.BindUniformBuffer(stage, binding_index, buffer, offset, size); } else { @@ -767,6 +1100,28 @@ void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) { } template <class P> +void BufferCache<P>::BindHostGraphicsTextureBuffers(size_t stage) { + ForEachEnabledBit(enabled_texture_buffers[stage], [&](u32 index) { + const TextureBufferBinding& binding = texture_buffers[stage][index]; + Buffer& buffer = slot_buffers[binding.buffer_id]; + const u32 size = binding.size; + SynchronizeBuffer(buffer, binding.cpu_addr, size); + + const u32 offset = buffer.Offset(binding.cpu_addr); + const PixelFormat format = binding.format; + if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) { + if (((image_texture_buffers[stage] >> index) & 1) != 0) { + runtime.BindImageBuffer(buffer, offset, size, format); + } else { + runtime.BindTextureBuffer(buffer, offset, size, format); + } + } else { + runtime.BindTextureBuffer(buffer, offset, size, format); + } + }); +} + +template <class P> void BufferCache<P>::BindHostTransformFeedbackBuffers() { if (maxwell3d.regs.tfb_enabled == 0) { return; @@ -788,13 +1143,14 @@ void BufferCache<P>::BindHostComputeUniformBuffers() { if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { // Mark all uniform buffers as dirty dirty_uniform_buffers.fill(~u32{0}); + fast_bound_uniform_buffers.fill(0); } u32 binding_index = 0; - ForEachEnabledBit(enabled_compute_uniform_buffers, [&](u32 index) { + ForEachEnabledBit(enabled_compute_uniform_buffer_mask, [&](u32 index) { const Binding& binding = compute_uniform_buffers[index]; Buffer& buffer = slot_buffers[binding.buffer_id]; TouchBuffer(buffer); - const u32 size = binding.size; + const u32 size = std::min(binding.size, (*compute_uniform_buffer_sizes)[index]); SynchronizeBuffer(buffer, binding.cpu_addr, size); const u32 offset = buffer.Offset(binding.cpu_addr); @@ -829,6 +1185,28 @@ void BufferCache<P>::BindHostComputeStorageBuffers() { } template <class P> +void BufferCache<P>::BindHostComputeTextureBuffers() { + ForEachEnabledBit(enabled_compute_texture_buffers, [&](u32 index) { + const TextureBufferBinding& binding = compute_texture_buffers[index]; + Buffer& buffer = slot_buffers[binding.buffer_id]; + const u32 size = binding.size; + SynchronizeBuffer(buffer, binding.cpu_addr, size); + + const u32 offset = buffer.Offset(binding.cpu_addr); + const PixelFormat format = binding.format; + if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) { + if (((image_compute_texture_buffers >> index) & 1) != 0) { + runtime.BindImageBuffer(buffer, offset, size, format); + } else { + runtime.BindTextureBuffer(buffer, offset, size, format); + } + } else { + runtime.BindTextureBuffer(buffer, offset, size, format); + } + }); +} + +template <class P> void BufferCache<P>::DoUpdateGraphicsBuffers(bool is_indexed) { if (is_indexed) { UpdateIndexBuffer(); @@ -838,6 +1216,7 @@ void BufferCache<P>::DoUpdateGraphicsBuffers(bool is_indexed) { for (size_t stage = 0; stage < NUM_STAGES; ++stage) { UpdateUniformBuffers(stage); UpdateStorageBuffers(stage); + UpdateTextureBuffers(stage); } } @@ -845,6 +1224,7 @@ template <class P> void BufferCache<P>::DoUpdateComputeBuffers() { UpdateComputeUniformBuffers(); UpdateComputeStorageBuffers(); + UpdateComputeTextureBuffers(); } template <class P> @@ -863,7 +1243,7 @@ void BufferCache<P>::UpdateIndexBuffer() { const GPUVAddr gpu_addr_end = index_array.EndAddress(); const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr_begin); const u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin); - const u32 draw_size = index_array.count * index_array.FormatSizeInBytes(); + const u32 draw_size = (index_array.count + index_array.first) * index_array.FormatSizeInBytes(); const u32 size = std::min(address_size, draw_size); if (size == 0 || !cpu_addr) { index_buffer = NULL_BINDING; @@ -914,7 +1294,7 @@ void BufferCache<P>::UpdateVertexBuffer(u32 index) { template <class P> void BufferCache<P>::UpdateUniformBuffers(size_t stage) { - ForEachEnabledBit(enabled_uniform_buffers[stage], [&](u32 index) { + ForEachEnabledBit(enabled_uniform_buffer_masks[stage], [&](u32 index) { Binding& binding = uniform_buffers[stage][index]; if (binding.buffer_id) { // Already updated @@ -945,6 +1325,18 @@ void BufferCache<P>::UpdateStorageBuffers(size_t stage) { } template <class P> +void BufferCache<P>::UpdateTextureBuffers(size_t stage) { + ForEachEnabledBit(enabled_texture_buffers[stage], [&](u32 index) { + Binding& binding = texture_buffers[stage][index]; + binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size); + // Mark buffer as written if needed + if (((written_texture_buffers[stage] >> index) & 1) != 0) { + MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size); + } + }); +} + +template <class P> void BufferCache<P>::UpdateTransformFeedbackBuffers() { if (maxwell3d.regs.tfb_enabled == 0) { return; @@ -975,7 +1367,7 @@ void BufferCache<P>::UpdateTransformFeedbackBuffer(u32 index) { template <class P> void BufferCache<P>::UpdateComputeUniformBuffers() { - ForEachEnabledBit(enabled_compute_uniform_buffers, [&](u32 index) { + ForEachEnabledBit(enabled_compute_uniform_buffer_mask, [&](u32 index) { Binding& binding = compute_uniform_buffers[index]; binding = NULL_BINDING; const auto& launch_desc = kepler_compute.launch_description; @@ -996,11 +1388,22 @@ void BufferCache<P>::UpdateComputeStorageBuffers() { ForEachEnabledBit(enabled_compute_storage_buffers, [&](u32 index) { // Resolve buffer Binding& binding = compute_storage_buffers[index]; - const BufferId buffer_id = FindBuffer(binding.cpu_addr, binding.size); - binding.buffer_id = buffer_id; + binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size); // Mark as written if needed if (((written_compute_storage_buffers >> index) & 1) != 0) { - MarkWrittenBuffer(buffer_id, binding.cpu_addr, binding.size); + MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size); + } + }); +} + +template <class P> +void BufferCache<P>::UpdateComputeTextureBuffers() { + ForEachEnabledBit(enabled_compute_texture_buffers, [&](u32 index) { + Binding& binding = compute_texture_buffers[index]; + binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size); + // Mark as written if needed + if (((written_compute_texture_buffers >> index) & 1) != 0) { + MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size); } }); } @@ -1010,16 +1413,16 @@ void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 s Buffer& buffer = slot_buffers[buffer_id]; buffer.MarkRegionAsGpuModified(cpu_addr, size); - const bool is_accuracy_high = Settings::IsGPULevelHigh(); + const IntervalType base_interval{cpu_addr, cpu_addr + size}; + common_ranges.add(base_interval); + + const bool is_accuracy_high = + Settings::values.gpu_accuracy.GetValue() == Settings::GPUAccuracy::High; const bool is_async = Settings::values.use_asynchronous_gpu_emulation.GetValue(); - if (!is_accuracy_high || !is_async) { + if (!is_async && !is_accuracy_high) { return; } - if (std::ranges::find(uncommitted_downloads, buffer_id) != uncommitted_downloads.end()) { - // Already inserted - return; - } - uncommitted_downloads.push_back(buffer_id); + uncommitted_ranges.add(base_interval); } template <class P> @@ -1103,7 +1506,6 @@ void BufferCache<P>::JoinOverlap(BufferId new_buffer_id, BufferId overlap_id, if (!copies.empty()) { runtime.CopyBuffer(slot_buffers[new_buffer_id], overlap, copies); } - ReplaceBufferDownloads(overlap_id, new_buffer_id); DeleteBuffer(overlap_id); } @@ -1244,14 +1646,29 @@ void BufferCache<P>::DownloadBufferMemory(Buffer& buffer, VAddr cpu_addr, u64 si boost::container::small_vector<BufferCopy, 1> copies; u64 total_size_bytes = 0; u64 largest_copy = 0; - buffer.ForEachDownloadRange(cpu_addr, size, [&](u64 range_offset, u64 range_size) { - copies.push_back(BufferCopy{ - .src_offset = range_offset, - .dst_offset = total_size_bytes, - .size = range_size, - }); - total_size_bytes += range_size; - largest_copy = std::max(largest_copy, range_size); + buffer.ForEachDownloadRangeAndClear(cpu_addr, size, [&](u64 range_offset, u64 range_size) { + const VAddr buffer_addr = buffer.CpuAddr(); + const auto add_download = [&](VAddr start, VAddr end) { + const u64 new_offset = start - buffer_addr; + const u64 new_size = end - start; + copies.push_back(BufferCopy{ + .src_offset = new_offset, + .dst_offset = total_size_bytes, + .size = new_size, + }); + // Align up to avoid cache conflicts + constexpr u64 align = 256ULL; + constexpr u64 mask = ~(align - 1ULL); + total_size_bytes += (new_size + align - 1) & mask; + largest_copy = std::max(largest_copy, new_size); + }; + + const VAddr start_address = buffer_addr + range_offset; + const VAddr end_address = start_address + range_size; + ForEachWrittenRange(start_address, range_size, add_download); + const IntervalType subtract_interval{start_address, end_address}; + ClearDownload(subtract_interval); + common_ranges.subtract(subtract_interval); }); if (total_size_bytes == 0) { return; @@ -1316,21 +1733,10 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id) { } template <class P> -void BufferCache<P>::ReplaceBufferDownloads(BufferId old_buffer_id, BufferId new_buffer_id) { - const auto replace = [old_buffer_id, new_buffer_id](std::vector<BufferId>& buffers) { - std::ranges::replace(buffers, old_buffer_id, new_buffer_id); - if (auto it = std::ranges::find(buffers, new_buffer_id); it != buffers.end()) { - buffers.erase(std::remove(it + 1, buffers.end(), new_buffer_id), buffers.end()); - } - }; - replace(uncommitted_downloads); - std::ranges::for_each(committed_downloads, replace); -} - -template <class P> void BufferCache<P>::NotifyBufferDeletion() { if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { dirty_uniform_buffers.fill(~u32{0}); + uniform_buffer_binding_sizes.fill({}); } auto& flags = maxwell3d.dirty.flags; flags[Dirty::IndexBuffer] = true; @@ -1349,21 +1755,34 @@ typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr s if (!cpu_addr || size == 0) { return NULL_BINDING; } - // HACK(Rodrigo): This is the number of bytes bound in host beyond the guest API's range. - // It exists due to some games like Astral Chain operate out of bounds. - // Binding the whole map range would be technically correct, but games have large maps that make - // this approach unaffordable for now. - static constexpr u32 arbitrary_extra_bytes = 0xc000; - const u32 bytes_to_map_end = static_cast<u32>(gpu_memory.BytesToMapEnd(gpu_addr)); const Binding binding{ .cpu_addr = *cpu_addr, - .size = std::min(size + arbitrary_extra_bytes, bytes_to_map_end), + .size = size, .buffer_id = BufferId{}, }; return binding; } template <class P> +typename BufferCache<P>::TextureBufferBinding BufferCache<P>::GetTextureBufferBinding( + GPUVAddr gpu_addr, u32 size, PixelFormat format) { + const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); + TextureBufferBinding binding; + if (!cpu_addr || size == 0) { + binding.cpu_addr = 0; + binding.size = 0; + binding.buffer_id = NULL_BUFFER_ID; + binding.format = PixelFormat::Invalid; + } else { + binding.cpu_addr = *cpu_addr; + binding.size = size; + binding.buffer_id = BufferId{}; + binding.format = format; + } + return binding; +} + +template <class P> std::span<const u8> BufferCache<P>::ImmediateBufferWithData(VAddr cpu_addr, size_t size) { u8* const base_pointer = cpu_memory.GetPointer(cpu_addr); if (IsRangeGranular(cpu_addr, size) || diff --git a/src/video_core/command_classes/vic.cpp b/src/video_core/command_classes/vic.cpp index ff3db0aee..ffb7c82a1 100644 --- a/src/video_core/command_classes/vic.cpp +++ b/src/video_core/command_classes/vic.cpp @@ -129,28 +129,27 @@ void Vic::Execute() { const std::size_t surface_width = config.surface_width_minus1 + 1; const std::size_t surface_height = config.surface_height_minus1 + 1; - const std::size_t half_width = surface_width / 2; - const std::size_t half_height = config.surface_height_minus1 / 2; + const auto frame_width = std::min(surface_width, static_cast<size_t>(frame->width)); + const auto frame_height = std::min(surface_height, static_cast<size_t>(frame->height)); + const std::size_t half_width = frame_width / 2; + const std::size_t half_height = frame_height / 2; const std::size_t aligned_width = (surface_width + 0xff) & ~0xff; const auto* luma_ptr = frame->data[0]; const auto* chroma_b_ptr = frame->data[1]; const auto* chroma_r_ptr = frame->data[2]; - const auto stride = frame->linesize[0]; - const auto half_stride = frame->linesize[1]; + const auto stride = static_cast<size_t>(frame->linesize[0]); + const auto half_stride = static_cast<size_t>(frame->linesize[1]); luma_buffer.resize(aligned_width * surface_height); - chroma_buffer.resize(aligned_width * half_height); + chroma_buffer.resize(aligned_width * surface_height / 2); // Populate luma buffer - for (std::size_t y = 0; y < surface_height - 1; ++y) { + for (std::size_t y = 0; y < frame_height; ++y) { const std::size_t src = y * stride; const std::size_t dst = y * aligned_width; - - const std::size_t size = surface_width; - - for (std::size_t offset = 0; offset < size; ++offset) { - luma_buffer[dst + offset] = luma_ptr[src + offset]; + for (std::size_t x = 0; x < frame_width; ++x) { + luma_buffer[dst + x] = luma_ptr[src + x]; } } gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), diff --git a/src/video_core/dirty_flags.cpp b/src/video_core/dirty_flags.cpp index 7149af290..b1be065c3 100644 --- a/src/video_core/dirty_flags.cpp +++ b/src/video_core/dirty_flags.cpp @@ -58,6 +58,11 @@ void SetupDirtyRenderTargets(Maxwell3D::DirtyState::Tables& tables) { FillBlock(table, OFF(zeta), NUM(zeta), flag); } } + +void SetupDirtyShaders(Maxwell3D::DirtyState::Tables& tables) { + FillBlock(tables[0], OFF(shader_config[0]), + NUM(shader_config[0]) * Maxwell3D::Regs::MaxShaderProgram, Shaders); +} } // Anonymous namespace void SetupDirtyFlags(Maxwell3D::DirtyState::Tables& tables) { @@ -65,6 +70,7 @@ void SetupDirtyFlags(Maxwell3D::DirtyState::Tables& tables) { SetupIndexBuffer(tables); SetupDirtyDescriptors(tables); SetupDirtyRenderTargets(tables); + SetupDirtyShaders(tables); } } // namespace VideoCommon::Dirty diff --git a/src/video_core/dirty_flags.h b/src/video_core/dirty_flags.h index 702688ace..504465d3f 100644 --- a/src/video_core/dirty_flags.h +++ b/src/video_core/dirty_flags.h @@ -36,6 +36,8 @@ enum : u8 { IndexBuffer, + Shaders, + LastCommonEntry, }; diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index 8b33c04ab..8d28bd884 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -4,6 +4,7 @@ #include "common/cityhash.h" #include "common/microprofile.h" +#include "common/settings.h" #include "core/core.h" #include "core/memory.h" #include "video_core/dma_pusher.h" @@ -76,8 +77,13 @@ bool DmaPusher::Step() { // Push buffer non-empty, read a word command_headers.resize(command_list_header.size); - gpu.MemoryManager().ReadBlockUnsafe(dma_get, command_headers.data(), - command_list_header.size * sizeof(u32)); + if (Settings::IsGPULevelHigh()) { + gpu.MemoryManager().ReadBlock(dma_get, command_headers.data(), + command_list_header.size * sizeof(u32)); + } else { + gpu.MemoryManager().ReadBlockUnsafe(dma_get, command_headers.data(), + command_list_header.size * sizeof(u32)); + } } for (std::size_t index = 0; index < command_headers.size();) { const CommandHeader& command_header = command_headers[index]; diff --git a/src/video_core/engines/const_buffer_engine_interface.h b/src/video_core/engines/const_buffer_engine_interface.h deleted file mode 100644 index f46e81bb7..000000000 --- a/src/video_core/engines/const_buffer_engine_interface.h +++ /dev/null @@ -1,103 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <type_traits> -#include "common/bit_field.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/engines/shader_type.h" -#include "video_core/guest_driver.h" -#include "video_core/textures/texture.h" - -namespace Tegra::Engines { - -struct SamplerDescriptor { - union { - u32 raw = 0; - BitField<0, 2, Tegra::Shader::TextureType> texture_type; - BitField<2, 3, Tegra::Texture::ComponentType> r_type; - BitField<5, 1, u32> is_array; - BitField<6, 1, u32> is_buffer; - BitField<7, 1, u32> is_shadow; - BitField<8, 3, Tegra::Texture::ComponentType> g_type; - BitField<11, 3, Tegra::Texture::ComponentType> b_type; - BitField<14, 3, Tegra::Texture::ComponentType> a_type; - BitField<17, 7, Tegra::Texture::TextureFormat> format; - }; - - bool operator==(const SamplerDescriptor& rhs) const noexcept { - return raw == rhs.raw; - } - - bool operator!=(const SamplerDescriptor& rhs) const noexcept { - return !operator==(rhs); - } - - static SamplerDescriptor FromTIC(const Tegra::Texture::TICEntry& tic) { - using Tegra::Shader::TextureType; - SamplerDescriptor result; - - result.format.Assign(tic.format.Value()); - result.r_type.Assign(tic.r_type.Value()); - result.g_type.Assign(tic.g_type.Value()); - result.b_type.Assign(tic.b_type.Value()); - result.a_type.Assign(tic.a_type.Value()); - - switch (tic.texture_type.Value()) { - case Tegra::Texture::TextureType::Texture1D: - result.texture_type.Assign(TextureType::Texture1D); - return result; - case Tegra::Texture::TextureType::Texture2D: - result.texture_type.Assign(TextureType::Texture2D); - return result; - case Tegra::Texture::TextureType::Texture3D: - result.texture_type.Assign(TextureType::Texture3D); - return result; - case Tegra::Texture::TextureType::TextureCubemap: - result.texture_type.Assign(TextureType::TextureCube); - return result; - case Tegra::Texture::TextureType::Texture1DArray: - result.texture_type.Assign(TextureType::Texture1D); - result.is_array.Assign(1); - return result; - case Tegra::Texture::TextureType::Texture2DArray: - result.texture_type.Assign(TextureType::Texture2D); - result.is_array.Assign(1); - return result; - case Tegra::Texture::TextureType::Texture1DBuffer: - result.texture_type.Assign(TextureType::Texture1D); - result.is_buffer.Assign(1); - return result; - case Tegra::Texture::TextureType::Texture2DNoMipmap: - result.texture_type.Assign(TextureType::Texture2D); - return result; - case Tegra::Texture::TextureType::TextureCubeArray: - result.texture_type.Assign(TextureType::TextureCube); - result.is_array.Assign(1); - return result; - default: - result.texture_type.Assign(TextureType::Texture2D); - return result; - } - } -}; -static_assert(std::is_trivially_copyable_v<SamplerDescriptor>); - -class ConstBufferEngineInterface { -public: - virtual ~ConstBufferEngineInterface() = default; - virtual u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const = 0; - virtual SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const = 0; - virtual SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer, - u64 offset) const = 0; - virtual SamplerDescriptor AccessSampler(u32 handle) const = 0; - virtual u32 GetBoundBuffer() const = 0; - - virtual VideoCore::GuestDriverProfile& AccessGuestDriverProfile() = 0; - virtual const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const = 0; -}; - -} // namespace Tegra::Engines diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp index a9b75091e..492b4c5a3 100644 --- a/src/video_core/engines/kepler_compute.cpp +++ b/src/video_core/engines/kepler_compute.cpp @@ -8,7 +8,6 @@ #include "core/core.h" #include "video_core/engines/kepler_compute.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/shader_type.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" #include "video_core/renderer_base.h" @@ -57,53 +56,11 @@ void KeplerCompute::CallMultiMethod(u32 method, const u32* base_start, u32 amoun } } -u32 KeplerCompute::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const { - ASSERT(stage == ShaderType::Compute); - const auto& buffer = launch_description.const_buffer_config[const_buffer]; - u32 result; - std::memcpy(&result, memory_manager.GetPointer(buffer.Address() + offset), sizeof(u32)); - return result; -} - -SamplerDescriptor KeplerCompute::AccessBoundSampler(ShaderType stage, u64 offset) const { - return AccessBindlessSampler(stage, regs.tex_cb_index, offset * sizeof(Texture::TextureHandle)); -} - -SamplerDescriptor KeplerCompute::AccessBindlessSampler(ShaderType stage, u64 const_buffer, - u64 offset) const { - ASSERT(stage == ShaderType::Compute); - const auto& tex_info_buffer = launch_description.const_buffer_config[const_buffer]; - const GPUVAddr tex_info_address = tex_info_buffer.Address() + offset; - return AccessSampler(memory_manager.Read<u32>(tex_info_address)); -} - -SamplerDescriptor KeplerCompute::AccessSampler(u32 handle) const { - const Texture::TextureHandle tex_handle{handle}; - const Texture::TICEntry tic = GetTICEntry(tex_handle.tic_id); - const Texture::TSCEntry tsc = GetTSCEntry(tex_handle.tsc_id); - - SamplerDescriptor result = SamplerDescriptor::FromTIC(tic); - result.is_shadow.Assign(tsc.depth_compare_enabled.Value()); - return result; -} - -VideoCore::GuestDriverProfile& KeplerCompute::AccessGuestDriverProfile() { - return rasterizer->AccessGuestDriverProfile(); -} - -const VideoCore::GuestDriverProfile& KeplerCompute::AccessGuestDriverProfile() const { - return rasterizer->AccessGuestDriverProfile(); -} - void KeplerCompute::ProcessLaunch() { const GPUVAddr launch_desc_loc = regs.launch_desc_loc.Address(); memory_manager.ReadBlockUnsafe(launch_desc_loc, &launch_description, LaunchParams::NUM_LAUNCH_PARAMETERS * sizeof(u32)); - - const GPUVAddr code_addr = regs.code_loc.Address() + launch_description.program_start; - LOG_TRACE(HW_GPU, "Compute invocation launched at address 0x{:016x}", code_addr); - - rasterizer->DispatchCompute(code_addr); + rasterizer->DispatchCompute(); } Texture::TICEntry KeplerCompute::GetTICEntry(u32 tic_index) const { diff --git a/src/video_core/engines/kepler_compute.h b/src/video_core/engines/kepler_compute.h index 7c40cba38..f8b8d06ac 100644 --- a/src/video_core/engines/kepler_compute.h +++ b/src/video_core/engines/kepler_compute.h @@ -10,10 +10,8 @@ #include "common/bit_field.h" #include "common/common_funcs.h" #include "common/common_types.h" -#include "video_core/engines/const_buffer_engine_interface.h" #include "video_core/engines/engine_interface.h" #include "video_core/engines/engine_upload.h" -#include "video_core/engines/shader_type.h" #include "video_core/gpu.h" #include "video_core/textures/texture.h" @@ -40,7 +38,7 @@ namespace Tegra::Engines { #define KEPLER_COMPUTE_REG_INDEX(field_name) \ (offsetof(Tegra::Engines::KeplerCompute::Regs, field_name) / sizeof(u32)) -class KeplerCompute final : public ConstBufferEngineInterface, public EngineInterface { +class KeplerCompute final : public EngineInterface { public: explicit KeplerCompute(Core::System& system, MemoryManager& memory_manager); ~KeplerCompute(); @@ -209,23 +207,6 @@ public: void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending) override; - u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const override; - - SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const override; - - SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer, - u64 offset) const override; - - SamplerDescriptor AccessSampler(u32 handle) const override; - - u32 GetBoundBuffer() const override { - return regs.tex_cb_index; - } - - VideoCore::GuestDriverProfile& AccessGuestDriverProfile() override; - - const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override; - private: void ProcessLaunch(); diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index aab6b8f7a..b18b8a02a 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -8,7 +8,6 @@ #include "core/core.h" #include "core/core_timing.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/shader_type.h" #include "video_core/gpu.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" @@ -670,42 +669,4 @@ void Maxwell3D::ProcessClearBuffers() { rasterizer->Clear(); } -u32 Maxwell3D::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const { - ASSERT(stage != ShaderType::Compute); - const auto& shader_stage = state.shader_stages[static_cast<std::size_t>(stage)]; - const auto& buffer = shader_stage.const_buffers[const_buffer]; - return memory_manager.Read<u32>(buffer.address + offset); -} - -SamplerDescriptor Maxwell3D::AccessBoundSampler(ShaderType stage, u64 offset) const { - return AccessBindlessSampler(stage, regs.tex_cb_index, offset * sizeof(Texture::TextureHandle)); -} - -SamplerDescriptor Maxwell3D::AccessBindlessSampler(ShaderType stage, u64 const_buffer, - u64 offset) const { - ASSERT(stage != ShaderType::Compute); - const auto& shader = state.shader_stages[static_cast<std::size_t>(stage)]; - const auto& tex_info_buffer = shader.const_buffers[const_buffer]; - const GPUVAddr tex_info_address = tex_info_buffer.address + offset; - return AccessSampler(memory_manager.Read<u32>(tex_info_address)); -} - -SamplerDescriptor Maxwell3D::AccessSampler(u32 handle) const { - const Texture::TextureHandle tex_handle{handle}; - const Texture::TICEntry tic = GetTICEntry(tex_handle.tic_id); - const Texture::TSCEntry tsc = GetTSCEntry(tex_handle.tsc_id); - - SamplerDescriptor result = SamplerDescriptor::FromTIC(tic); - result.is_shadow.Assign(tsc.depth_compare_enabled.Value()); - return result; -} - -VideoCore::GuestDriverProfile& Maxwell3D::AccessGuestDriverProfile() { - return rasterizer->AccessGuestDriverProfile(); -} - -const VideoCore::GuestDriverProfile& Maxwell3D::AccessGuestDriverProfile() const { - return rasterizer->AccessGuestDriverProfile(); -} - } // namespace Tegra::Engines diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 335383955..1aa43523a 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -17,11 +17,9 @@ #include "common/common_funcs.h" #include "common/common_types.h" #include "common/math_util.h" -#include "video_core/engines/const_buffer_engine_interface.h" #include "video_core/engines/const_buffer_info.h" #include "video_core/engines/engine_interface.h" #include "video_core/engines/engine_upload.h" -#include "video_core/engines/shader_type.h" #include "video_core/gpu.h" #include "video_core/macro/macro.h" #include "video_core/textures/texture.h" @@ -49,7 +47,7 @@ namespace Tegra::Engines { #define MAXWELL3D_REG_INDEX(field_name) \ (offsetof(Tegra::Engines::Maxwell3D::Regs, field_name) / sizeof(u32)) -class Maxwell3D final : public ConstBufferEngineInterface, public EngineInterface { +class Maxwell3D final : public EngineInterface { public: explicit Maxwell3D(Core::System& system, MemoryManager& memory_manager); ~Maxwell3D(); @@ -307,10 +305,6 @@ public: return (type == Type::SignedNorm) || (type == Type::UnsignedNorm); } - bool IsConstant() const { - return constant; - } - bool IsValid() const { return size != Size::Invalid; } @@ -912,7 +906,11 @@ public: u32 fill_rectangle; - INSERT_PADDING_WORDS_NOINIT(0x8); + INSERT_PADDING_WORDS_NOINIT(0x2); + + u32 conservative_raster_enable; + + INSERT_PADDING_WORDS_NOINIT(0x5); std::array<VertexAttribute, NumVertexAttributes> vertex_attrib_format; @@ -959,7 +957,11 @@ public: SamplerIndex sampler_index; - INSERT_PADDING_WORDS_NOINIT(0x25); + INSERT_PADDING_WORDS_NOINIT(0x2); + + std::array<u32, 8> gp_passthrough_mask; + + INSERT_PADDING_WORDS_NOINIT(0x1B); u32 depth_test_enable; @@ -1152,7 +1154,11 @@ public: u32 index; } primitive_restart; - INSERT_PADDING_WORDS_NOINIT(0x5F); + INSERT_PADDING_WORDS_NOINIT(0xE); + + u32 provoking_vertex_last; + + INSERT_PADDING_WORDS_NOINIT(0x50); struct { u32 start_addr_high; @@ -1424,23 +1430,6 @@ public: void FlushMMEInlineDraw(); - u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const override; - - SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const override; - - SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer, - u64 offset) const override; - - SamplerDescriptor AccessSampler(u32 handle) const override; - - u32 GetBoundBuffer() const override { - return regs.tex_cb_index; - } - - VideoCore::GuestDriverProfile& AccessGuestDriverProfile() override; - - const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override; - bool ShouldExecute() const { return execute_on; } @@ -1630,6 +1619,7 @@ ASSERT_REG_POSITION(zeta, 0x3F8); ASSERT_REG_POSITION(render_area, 0x3FD); ASSERT_REG_POSITION(clear_flags, 0x43E); ASSERT_REG_POSITION(fill_rectangle, 0x44F); +ASSERT_REG_POSITION(conservative_raster_enable, 0x452); ASSERT_REG_POSITION(vertex_attrib_format, 0x458); ASSERT_REG_POSITION(multisample_sample_locations, 0x478); ASSERT_REG_POSITION(multisample_coverage_to_color, 0x47E); @@ -1638,6 +1628,7 @@ ASSERT_REG_POSITION(zeta_width, 0x48a); ASSERT_REG_POSITION(zeta_height, 0x48b); ASSERT_REG_POSITION(zeta_depth, 0x48c); ASSERT_REG_POSITION(sampler_index, 0x48D); +ASSERT_REG_POSITION(gp_passthrough_mask, 0x490); ASSERT_REG_POSITION(depth_test_enable, 0x4B3); ASSERT_REG_POSITION(independent_blend_enable, 0x4B9); ASSERT_REG_POSITION(depth_write_enabled, 0x4BA); @@ -1690,6 +1681,7 @@ ASSERT_REG_POSITION(point_coord_replace, 0x581); ASSERT_REG_POSITION(code_address, 0x582); ASSERT_REG_POSITION(draw, 0x585); ASSERT_REG_POSITION(primitive_restart, 0x591); +ASSERT_REG_POSITION(provoking_vertex_last, 0x5A1); ASSERT_REG_POSITION(index_array, 0x5F2); ASSERT_REG_POSITION(polygon_offset_clamp, 0x61F); ASSERT_REG_POSITION(instanced_arrays, 0x620); diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index 2ee980bab..c7ec1eac9 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -4,6 +4,7 @@ #include "common/assert.h" #include "common/logging/log.h" +#include "common/microprofile.h" #include "common/settings.h" #include "core/core.h" #include "video_core/engines/maxwell_3d.h" @@ -12,6 +13,9 @@ #include "video_core/renderer_base.h" #include "video_core/textures/decoders.h" +MICROPROFILE_DECLARE(GPU_DMAEngine); +MICROPROFILE_DEFINE(GPU_DMAEngine, "GPU", "DMA Engine", MP_RGB(224, 224, 128)); + namespace Tegra::Engines { using namespace Texture; @@ -21,6 +25,10 @@ MaxwellDMA::MaxwellDMA(Core::System& system_, MemoryManager& memory_manager_) MaxwellDMA::~MaxwellDMA() = default; +void MaxwellDMA::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) { + rasterizer = rasterizer_; +} + void MaxwellDMA::CallMethod(u32 method, u32 method_argument, bool is_last_call) { ASSERT_MSG(method < NUM_REGS, "Invalid MaxwellDMA register"); @@ -39,12 +47,12 @@ void MaxwellDMA::CallMultiMethod(u32 method, const u32* base_start, u32 amount, } void MaxwellDMA::Launch() { + MICROPROFILE_SCOPE(GPU_DMAEngine); LOG_TRACE(Render_OpenGL, "DMA copy 0x{:x} -> 0x{:x}", static_cast<GPUVAddr>(regs.offset_in), static_cast<GPUVAddr>(regs.offset_out)); // TODO(Subv): Perform more research and implement all features of this engine. const LaunchDMA& launch = regs.launch_dma; - ASSERT(launch.remap_enable == 0); ASSERT(launch.semaphore_type == LaunchDMA::SemaphoreType::NONE); ASSERT(launch.interrupt_type == LaunchDMA::InterruptType::NONE); ASSERT(launch.data_transfer_type == LaunchDMA::DataTransferType::NON_PIPELINED); @@ -77,11 +85,31 @@ void MaxwellDMA::CopyPitchToPitch() { // When `multi_line_enable` bit is disabled the copy is performed as if we were copying a 1D // buffer of length `line_length_in`. // Otherwise we copy a 2D image of dimensions (line_length_in, line_count). + auto& accelerate = rasterizer->AccessAccelerateDMA(); if (!regs.launch_dma.multi_line_enable) { - memory_manager.CopyBlock(regs.offset_out, regs.offset_in, regs.line_length_in); + const bool is_buffer_clear = regs.launch_dma.remap_enable != 0 && + regs.remap_const.dst_x == RemapConst::Swizzle::CONST_A; + // TODO: allow multisized components. + if (is_buffer_clear) { + ASSERT(regs.remap_const.component_size_minus_one == 3); + accelerate.BufferClear(regs.offset_out, regs.line_length_in, regs.remap_consta_value); + std::vector<u32> tmp_buffer(regs.line_length_in, regs.remap_consta_value); + memory_manager.WriteBlockUnsafe(regs.offset_out, + reinterpret_cast<u8*>(tmp_buffer.data()), + regs.line_length_in * sizeof(u32)); + return; + } + UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0); + if (!accelerate.BufferCopy(regs.offset_in, regs.offset_out, regs.line_length_in)) { + std::vector<u8> tmp_buffer(regs.line_length_in); + memory_manager.ReadBlockUnsafe(regs.offset_in, tmp_buffer.data(), regs.line_length_in); + memory_manager.WriteBlock(regs.offset_out, tmp_buffer.data(), regs.line_length_in); + } return; } + UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0); + // Perform a line-by-line copy. // We're going to take a subrect of size (line_length_in, line_count) from the source rectangle. // There is no need to manually flush/invalidate the regions because CopyBlock does that for us. @@ -99,12 +127,14 @@ void MaxwellDMA::CopyBlockLinearToPitch() { // Optimized path for micro copies. const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count; - if (dst_size < GOB_SIZE && regs.pitch_out <= GOB_SIZE_X) { + if (dst_size < GOB_SIZE && regs.pitch_out <= GOB_SIZE_X && + regs.src_params.height > GOB_SIZE_Y) { FastCopyBlockLinearToPitch(); return; } // Deswizzle the input and copy it over. + UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0); const u32 bytes_per_pixel = regs.pitch_out / regs.line_length_in; const Parameters& src_params = regs.src_params; const u32 width = src_params.width; @@ -134,6 +164,7 @@ void MaxwellDMA::CopyBlockLinearToPitch() { void MaxwellDMA::CopyPitchToBlockLinear() { UNIMPLEMENTED_IF_MSG(regs.dst_params.block_size.width != 0, "Block width is not one"); + UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0); const auto& dst_params = regs.dst_params; const u32 bytes_per_pixel = regs.pitch_in / regs.line_length_in; diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h index c77f02a22..d3329b0f8 100644 --- a/src/video_core/engines/maxwell_dma.h +++ b/src/video_core/engines/maxwell_dma.h @@ -21,8 +21,20 @@ namespace Tegra { class MemoryManager; } +namespace VideoCore { +class RasterizerInterface; +} + namespace Tegra::Engines { +class AccelerateDMAInterface { +public: + /// Write the value to the register identified by method. + virtual bool BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) = 0; + + virtual bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) = 0; +}; + /** * This engine is known as gk104_copy. Documentation can be found in: * https://github.com/NVIDIA/open-gpu-doc/blob/master/classes/dma-copy/clb0b5.h @@ -187,6 +199,8 @@ public: }; static_assert(sizeof(RemapConst) == 12); + void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); + explicit MaxwellDMA(Core::System& system_, MemoryManager& memory_manager_); ~MaxwellDMA() override; @@ -213,6 +227,7 @@ private: Core::System& system; MemoryManager& memory_manager; + VideoCore::RasterizerInterface* rasterizer; std::vector<u8> read_buffer; std::vector<u8> write_buffer; @@ -240,7 +255,9 @@ private: u32 pitch_out; u32 line_length_in; u32 line_count; - u32 reserved06[0xb8]; + u32 reserved06[0xb6]; + u32 remap_consta_value; + u32 remap_constb_value; RemapConst remap_const; Parameters dst_params; u32 reserved07[0x1]; diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h deleted file mode 100644 index 8b45f1b62..000000000 --- a/src/video_core/engines/shader_bytecode.h +++ /dev/null @@ -1,2298 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <array> -#include <bitset> -#include <optional> -#include <tuple> -#include <vector> - -#include "common/assert.h" -#include "common/bit_field.h" -#include "common/common_types.h" - -namespace Tegra::Shader { - -struct Register { - /// Number of registers - static constexpr std::size_t NumRegisters = 256; - - /// Register 255 is special cased to always be 0 - static constexpr std::size_t ZeroIndex = 255; - - enum class Size : u64 { - Byte = 0, - Short = 1, - Word = 2, - Long = 3, - }; - - constexpr Register() = default; - - constexpr Register(u64 value_) : value(value_) {} - - [[nodiscard]] constexpr operator u64() const { - return value; - } - - template <typename T> - [[nodiscard]] constexpr u64 operator-(const T& oth) const { - return value - oth; - } - - template <typename T> - [[nodiscard]] constexpr u64 operator&(const T& oth) const { - return value & oth; - } - - [[nodiscard]] constexpr u64 operator&(const Register& oth) const { - return value & oth.value; - } - - [[nodiscard]] constexpr u64 operator~() const { - return ~value; - } - - [[nodiscard]] u64 GetSwizzledIndex(u64 elem) const { - elem = (value + elem) & 3; - return (value & ~3) + elem; - } - -private: - u64 value{}; -}; - -enum class AttributeSize : u64 { - Word = 0, - DoubleWord = 1, - TripleWord = 2, - QuadWord = 3, -}; - -union Attribute { - Attribute() = default; - - constexpr explicit Attribute(u64 value_) : value(value_) {} - - enum class Index : u64 { - LayerViewportPointSize = 6, - Position = 7, - Attribute_0 = 8, - Attribute_31 = 39, - FrontColor = 40, - FrontSecondaryColor = 41, - BackColor = 42, - BackSecondaryColor = 43, - ClipDistances0123 = 44, - ClipDistances4567 = 45, - PointCoord = 46, - // This attribute contains a tuple of (~, ~, InstanceId, VertexId) when inside a vertex - // shader, and a tuple of (TessCoord.x, TessCoord.y, TessCoord.z, ~) when inside a Tess Eval - // shader. - TessCoordInstanceIDVertexID = 47, - TexCoord_0 = 48, - TexCoord_7 = 55, - // This attribute contains a tuple of (Unk, Unk, Unk, gl_FrontFacing) when inside a fragment - // shader. It is unknown what the other values contain. - FrontFacing = 63, - }; - - union { - BitField<20, 10, u64> immediate; - BitField<22, 2, u64> element; - BitField<24, 6, Index> index; - BitField<31, 1, u64> patch; - BitField<47, 3, AttributeSize> size; - - [[nodiscard]] bool IsPhysical() const { - return patch == 0 && element == 0 && static_cast<u64>(index.Value()) == 0; - } - } fmt20; - - union { - BitField<30, 2, u64> element; - BitField<32, 6, Index> index; - } fmt28; - - BitField<39, 8, u64> reg; - u64 value{}; -}; - -union Sampler { - Sampler() = default; - - constexpr explicit Sampler(u64 value_) : value(value_) {} - - enum class Index : u64 { - Sampler_0 = 8, - }; - - BitField<36, 13, Index> index; - u64 value{}; -}; - -union Image { - Image() = default; - - constexpr explicit Image(u64 value_) : value{value_} {} - - BitField<36, 13, u64> index; - u64 value; -}; - -} // namespace Tegra::Shader - -namespace std { - -// TODO(bunnei): The below is forbidden by the C++ standard, but works fine. See #330. -template <> -struct make_unsigned<Tegra::Shader::Attribute> { - using type = Tegra::Shader::Attribute; -}; - -template <> -struct make_unsigned<Tegra::Shader::Register> { - using type = Tegra::Shader::Register; -}; - -} // namespace std - -namespace Tegra::Shader { - -enum class Pred : u64 { - UnusedIndex = 0x7, - NeverExecute = 0xF, -}; - -enum class PredCondition : u64 { - F = 0, // Always false - LT = 1, // Ordered less than - EQ = 2, // Ordered equal - LE = 3, // Ordered less than or equal - GT = 4, // Ordered greater than - NE = 5, // Ordered not equal - GE = 6, // Ordered greater than or equal - NUM = 7, // Ordered - NAN_ = 8, // Unordered - LTU = 9, // Unordered less than - EQU = 10, // Unordered equal - LEU = 11, // Unordered less than or equal - GTU = 12, // Unordered greater than - NEU = 13, // Unordered not equal - GEU = 14, // Unordered greater than or equal - T = 15, // Always true -}; - -enum class PredOperation : u64 { - And = 0, - Or = 1, - Xor = 2, -}; - -enum class LogicOperation : u64 { - And = 0, - Or = 1, - Xor = 2, - PassB = 3, -}; - -enum class SubOp : u64 { - Cos = 0x0, - Sin = 0x1, - Ex2 = 0x2, - Lg2 = 0x3, - Rcp = 0x4, - Rsq = 0x5, - Sqrt = 0x8, -}; - -enum class F2iRoundingOp : u64 { - RoundEven = 0, - Floor = 1, - Ceil = 2, - Trunc = 3, -}; - -enum class F2fRoundingOp : u64 { - None = 0, - Pass = 3, - Round = 8, - Floor = 9, - Ceil = 10, - Trunc = 11, -}; - -enum class AtomicOp : u64 { - Add = 0, - Min = 1, - Max = 2, - Inc = 3, - Dec = 4, - And = 5, - Or = 6, - Xor = 7, - Exch = 8, - SafeAdd = 10, -}; - -enum class GlobalAtomicType : u64 { - U32 = 0, - S32 = 1, - U64 = 2, - F32_FTZ_RN = 3, - F16x2_FTZ_RN = 4, - S64 = 5, -}; - -enum class UniformType : u64 { - UnsignedByte = 0, - SignedByte = 1, - UnsignedShort = 2, - SignedShort = 3, - Single = 4, - Double = 5, - Quad = 6, - UnsignedQuad = 7, -}; - -enum class StoreType : u64 { - Unsigned8 = 0, - Signed8 = 1, - Unsigned16 = 2, - Signed16 = 3, - Bits32 = 4, - Bits64 = 5, - Bits128 = 6, -}; - -enum class AtomicType : u64 { - U32 = 0, - S32 = 1, - U64 = 2, - S64 = 3, -}; - -enum class IMinMaxExchange : u64 { - None = 0, - XLo = 1, - XMed = 2, - XHi = 3, -}; - -enum class VideoType : u64 { - Size16_Low = 0, - Size16_High = 1, - Size32 = 2, - Invalid = 3, -}; - -enum class VmadShr : u64 { - Shr7 = 1, - Shr15 = 2, -}; - -enum class VmnmxType : u64 { - Bits8, - Bits16, - Bits32, -}; - -enum class VmnmxOperation : u64 { - Mrg_16H = 0, - Mrg_16L = 1, - Mrg_8B0 = 2, - Mrg_8B2 = 3, - Acc = 4, - Min = 5, - Max = 6, - Nop = 7, -}; - -enum class XmadMode : u64 { - None = 0, - CLo = 1, - CHi = 2, - CSfu = 3, - CBcc = 4, -}; - -enum class IAdd3Mode : u64 { - None = 0, - RightShift = 1, - LeftShift = 2, -}; - -enum class IAdd3Height : u64 { - None = 0, - LowerHalfWord = 1, - UpperHalfWord = 2, -}; - -enum class FlowCondition : u64 { - Always = 0xF, - Fcsm_Tr = 0x1C, // TODO(bunnei): What is this used for? -}; - -enum class ConditionCode : u64 { - F = 0, - LT = 1, - EQ = 2, - LE = 3, - GT = 4, - NE = 5, - GE = 6, - Num = 7, - Nan = 8, - LTU = 9, - EQU = 10, - LEU = 11, - GTU = 12, - NEU = 13, - GEU = 14, - T = 15, - OFF = 16, - LO = 17, - SFF = 18, - LS = 19, - HI = 20, - SFT = 21, - HS = 22, - OFT = 23, - CSM_TA = 24, - CSM_TR = 25, - CSM_MX = 26, - FCSM_TA = 27, - FCSM_TR = 28, - FCSM_MX = 29, - RLE = 30, - RGT = 31, -}; - -enum class PredicateResultMode : u64 { - None = 0x0, - NotZero = 0x3, -}; - -enum class TextureType : u64 { - Texture1D = 0, - Texture2D = 1, - Texture3D = 2, - TextureCube = 3, -}; - -enum class TextureQueryType : u64 { - Dimension = 1, - TextureType = 2, - SamplePosition = 5, - Filter = 16, - LevelOfDetail = 18, - Wrap = 20, - BorderColor = 22, -}; - -enum class TextureProcessMode : u64 { - None = 0, - LZ = 1, // Load LOD of zero. - LB = 2, // Load Bias. - LL = 3, // Load LOD. - LBA = 6, // Load Bias. The A is unknown, does not appear to differ with LB. - LLA = 7 // Load LOD. The A is unknown, does not appear to differ with LL. -}; - -enum class TextureMiscMode : u64 { - DC, - AOFFI, // Uses Offset - NDV, - NODEP, - MZ, - PTP, -}; - -enum class SurfaceDataMode : u64 { - P = 0, - D_BA = 1, -}; - -enum class OutOfBoundsStore : u64 { - Ignore = 0, - Clamp = 1, - Trap = 2, -}; - -enum class ImageType : u64 { - Texture1D = 0, - TextureBuffer = 1, - Texture1DArray = 2, - Texture2D = 3, - Texture2DArray = 4, - Texture3D = 5, -}; - -enum class IsberdMode : u64 { - None = 0, - Patch = 1, - Prim = 2, - Attr = 3, -}; - -enum class IsberdShift : u64 { None = 0, U16 = 1, B32 = 2 }; - -enum class MembarType : u64 { - CTA = 0, - GL = 1, - SYS = 2, - VC = 3, -}; - -enum class MembarUnknown : u64 { Default = 0, IVALLD = 1, IVALLT = 2, IVALLTD = 3 }; - -enum class HalfType : u64 { - H0_H1 = 0, - F32 = 1, - H0_H0 = 2, - H1_H1 = 3, -}; - -enum class HalfMerge : u64 { - H0_H1 = 0, - F32 = 1, - Mrg_H0 = 2, - Mrg_H1 = 3, -}; - -enum class HalfPrecision : u64 { - None = 0, - FTZ = 1, - FMZ = 2, -}; - -enum class R2pMode : u64 { - Pr = 0, - Cc = 1, -}; - -enum class IpaInterpMode : u64 { - Pass = 0, - Multiply = 1, - Constant = 2, - Sc = 3, -}; - -enum class IpaSampleMode : u64 { - Default = 0, - Centroid = 1, - Offset = 2, -}; - -enum class LmemLoadCacheManagement : u64 { - Default = 0, - LU = 1, - CI = 2, - CV = 3, -}; - -enum class StoreCacheManagement : u64 { - Default = 0, - CG = 1, - CS = 2, - WT = 3, -}; - -struct IpaMode { - IpaInterpMode interpolation_mode; - IpaSampleMode sampling_mode; - - [[nodiscard]] bool operator==(const IpaMode& a) const { - return std::tie(interpolation_mode, sampling_mode) == - std::tie(a.interpolation_mode, a.sampling_mode); - } - [[nodiscard]] bool operator!=(const IpaMode& a) const { - return !operator==(a); - } - [[nodiscard]] bool operator<(const IpaMode& a) const { - return std::tie(interpolation_mode, sampling_mode) < - std::tie(a.interpolation_mode, a.sampling_mode); - } -}; - -enum class SystemVariable : u64 { - LaneId = 0x00, - VirtCfg = 0x02, - VirtId = 0x03, - Pm0 = 0x04, - Pm1 = 0x05, - Pm2 = 0x06, - Pm3 = 0x07, - Pm4 = 0x08, - Pm5 = 0x09, - Pm6 = 0x0a, - Pm7 = 0x0b, - OrderingTicket = 0x0f, - PrimType = 0x10, - InvocationId = 0x11, - Ydirection = 0x12, - ThreadKill = 0x13, - ShaderType = 0x14, - DirectBeWriteAddressLow = 0x15, - DirectBeWriteAddressHigh = 0x16, - DirectBeWriteEnabled = 0x17, - MachineId0 = 0x18, - MachineId1 = 0x19, - MachineId2 = 0x1a, - MachineId3 = 0x1b, - Affinity = 0x1c, - InvocationInfo = 0x1d, - WscaleFactorXY = 0x1e, - WscaleFactorZ = 0x1f, - Tid = 0x20, - TidX = 0x21, - TidY = 0x22, - TidZ = 0x23, - CtaParam = 0x24, - CtaIdX = 0x25, - CtaIdY = 0x26, - CtaIdZ = 0x27, - NtId = 0x28, - CirQueueIncrMinusOne = 0x29, - Nlatc = 0x2a, - SmSpaVersion = 0x2c, - MultiPassShaderInfo = 0x2d, - LwinHi = 0x2e, - SwinHi = 0x2f, - SwinLo = 0x30, - SwinSz = 0x31, - SmemSz = 0x32, - SmemBanks = 0x33, - LwinLo = 0x34, - LwinSz = 0x35, - LmemLosz = 0x36, - LmemHioff = 0x37, - EqMask = 0x38, - LtMask = 0x39, - LeMask = 0x3a, - GtMask = 0x3b, - GeMask = 0x3c, - RegAlloc = 0x3d, - CtxAddr = 0x3e, // .fmask = F_SM50 - BarrierAlloc = 0x3e, // .fmask = F_SM60 - GlobalErrorStatus = 0x40, - WarpErrorStatus = 0x42, - WarpErrorStatusClear = 0x43, - PmHi0 = 0x48, - PmHi1 = 0x49, - PmHi2 = 0x4a, - PmHi3 = 0x4b, - PmHi4 = 0x4c, - PmHi5 = 0x4d, - PmHi6 = 0x4e, - PmHi7 = 0x4f, - ClockLo = 0x50, - ClockHi = 0x51, - GlobalTimerLo = 0x52, - GlobalTimerHi = 0x53, - HwTaskId = 0x60, - CircularQueueEntryIndex = 0x61, - CircularQueueEntryAddressLow = 0x62, - CircularQueueEntryAddressHigh = 0x63, -}; - -enum class PhysicalAttributeDirection : u64 { - Input = 0, - Output = 1, -}; - -enum class VoteOperation : u64 { - All = 0, // allThreadsNV - Any = 1, // anyThreadNV - Eq = 2, // allThreadsEqualNV -}; - -enum class ImageAtomicOperationType : u64 { - U32 = 0, - S32 = 1, - U64 = 2, - F32 = 3, - S64 = 5, - SD32 = 6, - SD64 = 7, -}; - -enum class ImageAtomicOperation : u64 { - Add = 0, - Min = 1, - Max = 2, - Inc = 3, - Dec = 4, - And = 5, - Or = 6, - Xor = 7, - Exch = 8, -}; - -enum class ShuffleOperation : u64 { - Idx = 0, // shuffleNV - Up = 1, // shuffleUpNV - Down = 2, // shuffleDownNV - Bfly = 3, // shuffleXorNV -}; - -enum class ShfType : u64 { - Bits32 = 0, - U64 = 2, - S64 = 3, -}; - -enum class ShfXmode : u64 { - None = 0, - HI = 1, - X = 2, - XHI = 3, -}; - -union Instruction { - constexpr Instruction& operator=(const Instruction& instr) { - value = instr.value; - return *this; - } - - constexpr Instruction(u64 value_) : value{value_} {} - constexpr Instruction(const Instruction& instr) : value(instr.value) {} - - [[nodiscard]] constexpr bool Bit(u64 offset) const { - return ((value >> offset) & 1) != 0; - } - - BitField<0, 8, Register> gpr0; - BitField<8, 8, Register> gpr8; - union { - BitField<16, 4, Pred> full_pred; - BitField<16, 3, u64> pred_index; - } pred; - BitField<19, 1, u64> negate_pred; - BitField<20, 8, Register> gpr20; - BitField<20, 4, SubOp> sub_op; - BitField<28, 8, Register> gpr28; - BitField<39, 8, Register> gpr39; - BitField<48, 16, u64> opcode; - - union { - BitField<8, 5, ConditionCode> cc; - BitField<13, 1, u64> trigger; - } nop; - - union { - BitField<48, 2, VoteOperation> operation; - BitField<45, 3, u64> dest_pred; - BitField<39, 3, u64> value; - BitField<42, 1, u64> negate_value; - } vote; - - union { - BitField<30, 2, ShuffleOperation> operation; - BitField<48, 3, u64> pred48; - BitField<28, 1, u64> is_index_imm; - BitField<29, 1, u64> is_mask_imm; - BitField<20, 5, u64> index_imm; - BitField<34, 13, u64> mask_imm; - } shfl; - - union { - BitField<44, 1, u64> ftz; - BitField<39, 2, u64> tab5cb8_2; - BitField<38, 1, u64> ndv; - BitField<47, 1, u64> cc; - BitField<28, 8, u64> swizzle; - } fswzadd; - - union { - BitField<8, 8, Register> gpr; - BitField<20, 24, s64> offset; - } gmem; - - union { - BitField<20, 16, u64> imm20_16; - BitField<20, 19, u64> imm20_19; - BitField<20, 32, s64> imm20_32; - BitField<45, 1, u64> negate_b; - BitField<46, 1, u64> abs_a; - BitField<48, 1, u64> negate_a; - BitField<49, 1, u64> abs_b; - BitField<50, 1, u64> saturate_d; - BitField<56, 1, u64> negate_imm; - - union { - BitField<39, 3, u64> pred; - BitField<42, 1, u64> negate_pred; - } fmnmx; - - union { - BitField<39, 1, u64> invert_a; - BitField<40, 1, u64> invert_b; - BitField<41, 2, LogicOperation> operation; - BitField<44, 2, PredicateResultMode> pred_result_mode; - BitField<48, 3, Pred> pred48; - } lop; - - union { - BitField<53, 2, LogicOperation> operation; - BitField<55, 1, u64> invert_a; - BitField<56, 1, u64> invert_b; - } lop32i; - - union { - BitField<28, 8, u64> imm_lut28; - BitField<48, 8, u64> imm_lut48; - - [[nodiscard]] u32 GetImmLut28() const { - return static_cast<u32>(imm_lut28); - } - - [[nodiscard]] u32 GetImmLut48() const { - return static_cast<u32>(imm_lut48); - } - } lop3; - - [[nodiscard]] u16 GetImm20_16() const { - return static_cast<u16>(imm20_16); - } - - [[nodiscard]] u32 GetImm20_19() const { - u32 imm{static_cast<u32>(imm20_19)}; - imm <<= 12; - imm |= negate_imm ? 0x80000000 : 0; - return imm; - } - - [[nodiscard]] u32 GetImm20_32() const { - return static_cast<u32>(imm20_32); - } - - [[nodiscard]] s32 GetSignedImm20_20() const { - const auto immediate = static_cast<u32>(imm20_19 | (negate_imm << 19)); - // Sign extend the 20-bit value. - const auto mask = 1U << (20 - 1); - return static_cast<s32>((immediate ^ mask) - mask); - } - } alu; - - union { - BitField<38, 1, u64> idx; - BitField<51, 1, u64> saturate; - BitField<52, 2, IpaSampleMode> sample_mode; - BitField<54, 2, IpaInterpMode> interp_mode; - } ipa; - - union { - BitField<39, 2, u64> tab5cb8_2; - BitField<41, 3, u64> postfactor; - BitField<44, 2, u64> tab5c68_0; - BitField<48, 1, u64> negate_b; - } fmul; - - union { - BitField<55, 1, u64> saturate; - } fmul32; - - union { - BitField<52, 1, u64> generates_cc; - } op_32; - - union { - BitField<48, 1, u64> is_signed; - } shift; - - union { - BitField<39, 1, u64> wrap; - } shr; - - union { - BitField<37, 2, ShfType> type; - BitField<48, 2, ShfXmode> xmode; - BitField<50, 1, u64> wrap; - BitField<20, 6, u64> immediate; - } shf; - - union { - BitField<39, 5, u64> shift_amount; - BitField<48, 1, u64> negate_b; - BitField<49, 1, u64> negate_a; - } alu_integer; - - union { - BitField<43, 1, u64> x; - } iadd; - - union { - BitField<39, 1, u64> ftz; - BitField<32, 1, u64> saturate; - BitField<49, 2, HalfMerge> merge; - - BitField<44, 1, u64> abs_a; - BitField<47, 2, HalfType> type_a; - - BitField<30, 1, u64> abs_b; - BitField<28, 2, HalfType> type_b; - - BitField<35, 2, HalfType> type_c; - } alu_half; - - union { - BitField<39, 2, HalfPrecision> precision; - BitField<39, 1, u64> ftz; - BitField<52, 1, u64> saturate; - BitField<49, 2, HalfMerge> merge; - - BitField<43, 1, u64> negate_a; - BitField<44, 1, u64> abs_a; - BitField<47, 2, HalfType> type_a; - } alu_half_imm; - - union { - BitField<29, 1, u64> first_negate; - BitField<20, 9, u64> first; - - BitField<56, 1, u64> second_negate; - BitField<30, 9, u64> second; - - [[nodiscard]] u32 PackImmediates() const { - // Immediates are half floats shifted. - constexpr u32 imm_shift = 6; - return static_cast<u32>((first << imm_shift) | (second << (16 + imm_shift))); - } - } half_imm; - - union { - union { - BitField<37, 2, HalfPrecision> precision; - BitField<32, 1, u64> saturate; - - BitField<31, 1, u64> negate_b; - BitField<30, 1, u64> negate_c; - BitField<35, 2, HalfType> type_c; - } rr; - - BitField<57, 2, HalfPrecision> precision; - BitField<52, 1, u64> saturate; - - BitField<49, 2, HalfMerge> merge; - - BitField<47, 2, HalfType> type_a; - - BitField<56, 1, u64> negate_b; - BitField<28, 2, HalfType> type_b; - - BitField<51, 1, u64> negate_c; - BitField<53, 2, HalfType> type_reg39; - } hfma2; - - union { - BitField<40, 1, u64> invert; - } popc; - - union { - BitField<41, 1, u64> sh; - BitField<40, 1, u64> invert; - BitField<48, 1, u64> is_signed; - } flo; - - union { - BitField<39, 3, u64> pred; - BitField<42, 1, u64> neg_pred; - } sel; - - union { - BitField<39, 3, u64> pred; - BitField<42, 1, u64> negate_pred; - BitField<43, 2, IMinMaxExchange> exchange; - BitField<48, 1, u64> is_signed; - } imnmx; - - union { - BitField<31, 2, IAdd3Height> height_c; - BitField<33, 2, IAdd3Height> height_b; - BitField<35, 2, IAdd3Height> height_a; - BitField<37, 2, IAdd3Mode> mode; - BitField<49, 1, u64> neg_c; - BitField<50, 1, u64> neg_b; - BitField<51, 1, u64> neg_a; - } iadd3; - - union { - BitField<54, 1, u64> saturate; - BitField<56, 1, u64> negate_a; - } iadd32i; - - union { - BitField<53, 1, u64> negate_b; - BitField<54, 1, u64> abs_a; - BitField<56, 1, u64> negate_a; - BitField<57, 1, u64> abs_b; - } fadd32i; - - union { - BitField<40, 1, u64> brev; - BitField<47, 1, u64> rd_cc; - BitField<48, 1, u64> is_signed; - } bfe; - - union { - BitField<48, 3, u64> pred48; - - union { - BitField<20, 20, u64> entry_a; - BitField<39, 5, u64> entry_b; - BitField<45, 1, u64> neg; - BitField<46, 1, u64> uses_cc; - } imm; - - union { - BitField<20, 14, u64> cb_index; - BitField<34, 5, u64> cb_offset; - BitField<56, 1, u64> neg; - BitField<57, 1, u64> uses_cc; - } hi; - - union { - BitField<20, 14, u64> cb_index; - BitField<34, 5, u64> cb_offset; - BitField<39, 5, u64> entry_a; - BitField<45, 1, u64> neg; - BitField<46, 1, u64> uses_cc; - } rz; - - union { - BitField<39, 5, u64> entry_a; - BitField<45, 1, u64> neg; - BitField<46, 1, u64> uses_cc; - } r1; - - union { - BitField<28, 8, u64> entry_a; - BitField<37, 1, u64> neg; - BitField<38, 1, u64> uses_cc; - } r2; - - } lea; - - union { - BitField<0, 5, FlowCondition> cond; - } flow; - - union { - BitField<47, 1, u64> cc; - BitField<48, 1, u64> negate_b; - BitField<49, 1, u64> negate_c; - BitField<51, 2, u64> tab5980_1; - BitField<53, 2, u64> tab5980_0; - } ffma; - - union { - BitField<48, 3, UniformType> type; - BitField<44, 2, u64> unknown; - } ld_c; - - union { - BitField<48, 3, StoreType> type; - } ldst_sl; - - union { - BitField<44, 2, u64> unknown; - } ld_l; - - union { - BitField<44, 2, StoreCacheManagement> cache_management; - } st_l; - - union { - BitField<48, 3, UniformType> type; - BitField<46, 2, u64> cache_mode; - } ldg; - - union { - BitField<48, 3, UniformType> type; - BitField<46, 2, u64> cache_mode; - } stg; - - union { - BitField<23, 3, AtomicOp> operation; - BitField<48, 1, u64> extended; - BitField<20, 3, GlobalAtomicType> type; - } red; - - union { - BitField<52, 4, AtomicOp> operation; - BitField<49, 3, GlobalAtomicType> type; - BitField<28, 20, s64> offset; - } atom; - - union { - BitField<52, 4, AtomicOp> operation; - BitField<28, 2, AtomicType> type; - BitField<30, 22, s64> offset; - - [[nodiscard]] s32 GetImmediateOffset() const { - return static_cast<s32>(offset << 2); - } - } atoms; - - union { - BitField<32, 1, PhysicalAttributeDirection> direction; - BitField<47, 3, AttributeSize> size; - BitField<20, 11, u64> address; - } al2p; - - union { - BitField<53, 3, UniformType> type; - BitField<52, 1, u64> extended; - } generic; - - union { - BitField<0, 3, u64> pred0; - BitField<3, 3, u64> pred3; - BitField<6, 1, u64> neg_b; - BitField<7, 1, u64> abs_a; - BitField<39, 3, u64> pred39; - BitField<42, 1, u64> neg_pred; - BitField<43, 1, u64> neg_a; - BitField<44, 1, u64> abs_b; - BitField<45, 2, PredOperation> op; - BitField<47, 1, u64> ftz; - BitField<48, 4, PredCondition> cond; - } fsetp; - - union { - BitField<0, 3, u64> pred0; - BitField<3, 3, u64> pred3; - BitField<39, 3, u64> pred39; - BitField<42, 1, u64> neg_pred; - BitField<45, 2, PredOperation> op; - BitField<48, 1, u64> is_signed; - BitField<49, 3, PredCondition> cond; - } isetp; - - union { - BitField<48, 1, u64> is_signed; - BitField<49, 3, PredCondition> cond; - } icmp; - - union { - BitField<0, 3, u64> pred0; - BitField<3, 3, u64> pred3; - BitField<12, 3, u64> pred12; - BitField<15, 1, u64> neg_pred12; - BitField<24, 2, PredOperation> cond; - BitField<29, 3, u64> pred29; - BitField<32, 1, u64> neg_pred29; - BitField<39, 3, u64> pred39; - BitField<42, 1, u64> neg_pred39; - BitField<45, 2, PredOperation> op; - } psetp; - - union { - BitField<43, 4, PredCondition> cond; - BitField<45, 2, PredOperation> op; - BitField<3, 3, u64> pred3; - BitField<0, 3, u64> pred0; - BitField<39, 3, u64> pred39; - } vsetp; - - union { - BitField<12, 3, u64> pred12; - BitField<15, 1, u64> neg_pred12; - BitField<24, 2, PredOperation> cond; - BitField<29, 3, u64> pred29; - BitField<32, 1, u64> neg_pred29; - BitField<39, 3, u64> pred39; - BitField<42, 1, u64> neg_pred39; - BitField<44, 1, u64> bf; - BitField<45, 2, PredOperation> op; - } pset; - - union { - BitField<0, 3, u64> pred0; - BitField<3, 3, u64> pred3; - BitField<8, 5, ConditionCode> cc; // flag in cc - BitField<39, 3, u64> pred39; - BitField<42, 1, u64> neg_pred39; - BitField<45, 4, PredOperation> op; // op with pred39 - } csetp; - - union { - BitField<6, 1, u64> ftz; - BitField<45, 2, PredOperation> op; - BitField<3, 3, u64> pred3; - BitField<0, 3, u64> pred0; - BitField<43, 1, u64> negate_a; - BitField<44, 1, u64> abs_a; - BitField<47, 2, HalfType> type_a; - union { - BitField<35, 4, PredCondition> cond; - BitField<49, 1, u64> h_and; - BitField<31, 1, u64> negate_b; - BitField<30, 1, u64> abs_b; - BitField<28, 2, HalfType> type_b; - } reg; - union { - BitField<56, 1, u64> negate_b; - BitField<54, 1, u64> abs_b; - } cbuf; - union { - BitField<49, 4, PredCondition> cond; - BitField<53, 1, u64> h_and; - } cbuf_and_imm; - BitField<42, 1, u64> neg_pred; - BitField<39, 3, u64> pred39; - } hsetp2; - - union { - BitField<40, 1, R2pMode> mode; - BitField<41, 2, u64> byte; - BitField<20, 7, u64> immediate_mask; - } p2r_r2p; - - union { - BitField<39, 3, u64> pred39; - BitField<42, 1, u64> neg_pred; - BitField<43, 1, u64> neg_a; - BitField<44, 1, u64> abs_b; - BitField<45, 2, PredOperation> op; - BitField<48, 4, PredCondition> cond; - BitField<52, 1, u64> bf; - BitField<53, 1, u64> neg_b; - BitField<54, 1, u64> abs_a; - BitField<55, 1, u64> ftz; - } fset; - - union { - BitField<47, 1, u64> ftz; - BitField<48, 4, PredCondition> cond; - } fcmp; - - union { - BitField<49, 1, u64> bf; - BitField<35, 3, PredCondition> cond; - BitField<50, 1, u64> ftz; - BitField<45, 2, PredOperation> op; - BitField<43, 1, u64> negate_a; - BitField<44, 1, u64> abs_a; - BitField<47, 2, HalfType> type_a; - BitField<31, 1, u64> negate_b; - BitField<30, 1, u64> abs_b; - BitField<28, 2, HalfType> type_b; - BitField<42, 1, u64> neg_pred; - BitField<39, 3, u64> pred39; - } hset2; - - union { - BitField<39, 3, u64> pred39; - BitField<42, 1, u64> neg_pred; - BitField<44, 1, u64> bf; - BitField<45, 2, PredOperation> op; - BitField<48, 1, u64> is_signed; - BitField<49, 3, PredCondition> cond; - } iset; - - union { - BitField<45, 1, u64> negate_a; - BitField<49, 1, u64> abs_a; - BitField<10, 2, Register::Size> src_size; - BitField<13, 1, u64> is_input_signed; - BitField<8, 2, Register::Size> dst_size; - BitField<12, 1, u64> is_output_signed; - - union { - BitField<39, 2, u64> tab5cb8_2; - } i2f; - - union { - BitField<39, 2, F2iRoundingOp> rounding; - } f2i; - - union { - BitField<39, 4, u64> rounding; - // H0, H1 extract for F16 missing - BitField<41, 1, u64> selector; // Guessed as some games set it, TODO: reverse this value - [[nodiscard]] F2fRoundingOp GetRoundingMode() const { - constexpr u64 rounding_mask = 0x0B; - return static_cast<F2fRoundingOp>(rounding.Value() & rounding_mask); - } - } f2f; - - union { - BitField<41, 2, u64> selector; - } int_src; - - union { - BitField<41, 1, u64> selector; - } float_src; - } conversion; - - union { - BitField<28, 1, u64> array; - BitField<29, 2, TextureType> texture_type; - BitField<31, 4, u64> component_mask; - BitField<49, 1, u64> nodep_flag; - BitField<50, 1, u64> dc_flag; - BitField<54, 1, u64> aoffi_flag; - BitField<55, 3, TextureProcessMode> process_mode; - - [[nodiscard]] bool IsComponentEnabled(std::size_t component) const { - return ((1ULL << component) & component_mask) != 0; - } - - [[nodiscard]] TextureProcessMode GetTextureProcessMode() const { - return process_mode; - } - - [[nodiscard]] bool UsesMiscMode(TextureMiscMode mode) const { - switch (mode) { - case TextureMiscMode::DC: - return dc_flag != 0; - case TextureMiscMode::NODEP: - return nodep_flag != 0; - case TextureMiscMode::AOFFI: - return aoffi_flag != 0; - default: - break; - } - return false; - } - } tex; - - union { - BitField<28, 1, u64> array; - BitField<29, 2, TextureType> texture_type; - BitField<31, 4, u64> component_mask; - BitField<49, 1, u64> nodep_flag; - BitField<50, 1, u64> dc_flag; - BitField<36, 1, u64> aoffi_flag; - BitField<37, 3, TextureProcessMode> process_mode; - - [[nodiscard]] bool IsComponentEnabled(std::size_t component) const { - return ((1ULL << component) & component_mask) != 0; - } - - [[nodiscard]] TextureProcessMode GetTextureProcessMode() const { - return process_mode; - } - - [[nodiscard]] bool UsesMiscMode(TextureMiscMode mode) const { - switch (mode) { - case TextureMiscMode::DC: - return dc_flag != 0; - case TextureMiscMode::NODEP: - return nodep_flag != 0; - case TextureMiscMode::AOFFI: - return aoffi_flag != 0; - default: - break; - } - return false; - } - } tex_b; - - union { - BitField<22, 6, TextureQueryType> query_type; - BitField<31, 4, u64> component_mask; - BitField<49, 1, u64> nodep_flag; - - [[nodiscard]] bool UsesMiscMode(TextureMiscMode mode) const { - switch (mode) { - case TextureMiscMode::NODEP: - return nodep_flag != 0; - default: - break; - } - return false; - } - - [[nodiscard]] bool IsComponentEnabled(std::size_t component) const { - return ((1ULL << component) & component_mask) != 0; - } - } txq; - - union { - BitField<28, 1, u64> array; - BitField<29, 2, TextureType> texture_type; - BitField<31, 4, u64> component_mask; - BitField<35, 1, u64> ndv_flag; - BitField<49, 1, u64> nodep_flag; - - [[nodiscard]] bool IsComponentEnabled(std::size_t component) const { - return ((1ULL << component) & component_mask) != 0; - } - - [[nodiscard]] bool UsesMiscMode(TextureMiscMode mode) const { - switch (mode) { - case TextureMiscMode::NDV: - return (ndv_flag != 0); - case TextureMiscMode::NODEP: - return (nodep_flag != 0); - default: - break; - } - return false; - } - } tmml; - - union { - BitField<28, 1, u64> array; - BitField<29, 2, TextureType> texture_type; - BitField<35, 1, u64> ndv_flag; - BitField<49, 1, u64> nodep_flag; - BitField<50, 1, u64> dc_flag; - BitField<54, 2, u64> offset_mode; - BitField<56, 2, u64> component; - - [[nodiscard]] bool UsesMiscMode(TextureMiscMode mode) const { - switch (mode) { - case TextureMiscMode::NDV: - return ndv_flag != 0; - case TextureMiscMode::NODEP: - return nodep_flag != 0; - case TextureMiscMode::DC: - return dc_flag != 0; - case TextureMiscMode::AOFFI: - return offset_mode == 1; - case TextureMiscMode::PTP: - return offset_mode == 2; - default: - break; - } - return false; - } - } tld4; - - union { - BitField<35, 1, u64> ndv_flag; - BitField<49, 1, u64> nodep_flag; - BitField<50, 1, u64> dc_flag; - BitField<33, 2, u64> offset_mode; - BitField<37, 2, u64> component; - - [[nodiscard]] bool UsesMiscMode(TextureMiscMode mode) const { - switch (mode) { - case TextureMiscMode::NDV: - return ndv_flag != 0; - case TextureMiscMode::NODEP: - return nodep_flag != 0; - case TextureMiscMode::DC: - return dc_flag != 0; - case TextureMiscMode::AOFFI: - return offset_mode == 1; - case TextureMiscMode::PTP: - return offset_mode == 2; - default: - break; - } - return false; - } - } tld4_b; - - union { - BitField<49, 1, u64> nodep_flag; - BitField<50, 1, u64> dc_flag; - BitField<51, 1, u64> aoffi_flag; - BitField<52, 2, u64> component; - BitField<55, 1, u64> fp16_flag; - - [[nodiscard]] bool UsesMiscMode(TextureMiscMode mode) const { - switch (mode) { - case TextureMiscMode::DC: - return dc_flag != 0; - case TextureMiscMode::NODEP: - return nodep_flag != 0; - case TextureMiscMode::AOFFI: - return aoffi_flag != 0; - default: - break; - } - return false; - } - } tld4s; - - union { - BitField<0, 8, Register> gpr0; - BitField<28, 8, Register> gpr28; - BitField<49, 1, u64> nodep_flag; - BitField<50, 3, u64> component_mask_selector; - BitField<53, 4, u64> texture_info; - BitField<59, 1, u64> fp32_flag; - - [[nodiscard]] TextureType GetTextureType() const { - // The TEXS instruction has a weird encoding for the texture type. - if (texture_info == 0) { - return TextureType::Texture1D; - } - if (texture_info >= 1 && texture_info <= 9) { - return TextureType::Texture2D; - } - if (texture_info >= 10 && texture_info <= 11) { - return TextureType::Texture3D; - } - if (texture_info >= 12 && texture_info <= 13) { - return TextureType::TextureCube; - } - - LOG_CRITICAL(HW_GPU, "Unhandled texture_info: {}", texture_info.Value()); - UNREACHABLE(); - return TextureType::Texture1D; - } - - [[nodiscard]] TextureProcessMode GetTextureProcessMode() const { - switch (texture_info) { - case 0: - case 2: - case 6: - case 8: - case 9: - case 11: - return TextureProcessMode::LZ; - case 3: - case 5: - case 13: - return TextureProcessMode::LL; - default: - break; - } - return TextureProcessMode::None; - } - - [[nodiscard]] bool UsesMiscMode(TextureMiscMode mode) const { - switch (mode) { - case TextureMiscMode::DC: - return (texture_info >= 4 && texture_info <= 6) || texture_info == 9; - case TextureMiscMode::NODEP: - return nodep_flag != 0; - default: - break; - } - return false; - } - - [[nodiscard]] bool IsArrayTexture() const { - // TEXS only supports Texture2D arrays. - return texture_info >= 7 && texture_info <= 9; - } - - [[nodiscard]] bool HasTwoDestinations() const { - return gpr28.Value() != Register::ZeroIndex; - } - - [[nodiscard]] bool IsComponentEnabled(std::size_t component) const { - static constexpr std::array<std::array<u32, 8>, 4> mask_lut{{ - {}, - {0x1, 0x2, 0x4, 0x8, 0x3, 0x9, 0xa, 0xc}, - {0x1, 0x2, 0x4, 0x8, 0x3, 0x9, 0xa, 0xc}, - {0x7, 0xb, 0xd, 0xe, 0xf}, - }}; - - std::size_t index{gpr0.Value() != Register::ZeroIndex ? 1U : 0U}; - index |= gpr28.Value() != Register::ZeroIndex ? 2 : 0; - - u32 mask = mask_lut[index][component_mask_selector]; - // A mask of 0 means this instruction uses an unimplemented mask. - ASSERT(mask != 0); - return ((1ull << component) & mask) != 0; - } - } texs; - - union { - BitField<28, 1, u64> is_array; - BitField<29, 2, TextureType> texture_type; - BitField<35, 1, u64> aoffi; - BitField<49, 1, u64> nodep_flag; - BitField<50, 1, u64> ms; // Multisample? - BitField<54, 1, u64> cl; - BitField<55, 1, u64> process_mode; - - [[nodiscard]] TextureProcessMode GetTextureProcessMode() const { - return process_mode == 0 ? TextureProcessMode::LZ : TextureProcessMode::LL; - } - } tld; - - union { - BitField<49, 1, u64> nodep_flag; - BitField<53, 4, u64> texture_info; - BitField<59, 1, u64> fp32_flag; - - [[nodiscard]] TextureType GetTextureType() const { - // The TLDS instruction has a weird encoding for the texture type. - if (texture_info <= 1) { - return TextureType::Texture1D; - } - if (texture_info == 2 || texture_info == 8 || texture_info == 12 || - (texture_info >= 4 && texture_info <= 6)) { - return TextureType::Texture2D; - } - if (texture_info == 7) { - return TextureType::Texture3D; - } - - LOG_CRITICAL(HW_GPU, "Unhandled texture_info: {}", texture_info.Value()); - UNREACHABLE(); - return TextureType::Texture1D; - } - - [[nodiscard]] TextureProcessMode GetTextureProcessMode() const { - if (texture_info == 1 || texture_info == 5 || texture_info == 12) { - return TextureProcessMode::LL; - } - return TextureProcessMode::LZ; - } - - [[nodiscard]] bool UsesMiscMode(TextureMiscMode mode) const { - switch (mode) { - case TextureMiscMode::AOFFI: - return texture_info == 12 || texture_info == 4; - case TextureMiscMode::MZ: - return texture_info == 5; - case TextureMiscMode::NODEP: - return nodep_flag != 0; - default: - break; - } - return false; - } - - [[nodiscard]] bool IsArrayTexture() const { - // TEXS only supports Texture2D arrays. - return texture_info == 8; - } - } tlds; - - union { - BitField<28, 1, u64> is_array; - BitField<29, 2, TextureType> texture_type; - BitField<35, 1, u64> aoffi_flag; - BitField<49, 1, u64> nodep_flag; - - [[nodiscard]] bool UsesMiscMode(TextureMiscMode mode) const { - switch (mode) { - case TextureMiscMode::AOFFI: - return aoffi_flag != 0; - case TextureMiscMode::NODEP: - return nodep_flag != 0; - default: - break; - } - return false; - } - - } txd; - - union { - BitField<24, 2, StoreCacheManagement> cache_management; - BitField<33, 3, ImageType> image_type; - BitField<49, 2, OutOfBoundsStore> out_of_bounds_store; - BitField<51, 1, u64> is_immediate; - BitField<52, 1, SurfaceDataMode> mode; - - BitField<20, 3, StoreType> store_data_layout; - BitField<20, 4, u64> component_mask_selector; - - [[nodiscard]] bool IsComponentEnabled(std::size_t component) const { - ASSERT(mode == SurfaceDataMode::P); - constexpr u8 R = 0b0001; - constexpr u8 G = 0b0010; - constexpr u8 B = 0b0100; - constexpr u8 A = 0b1000; - constexpr std::array<u8, 16> mask = { - 0, (R), (G), (R | G), (B), (R | B), - (G | B), (R | G | B), (A), (R | A), (G | A), (R | G | A), - (B | A), (R | B | A), (G | B | A), (R | G | B | A)}; - return std::bitset<4>{mask.at(component_mask_selector)}.test(component); - } - - [[nodiscard]] StoreType GetStoreDataLayout() const { - ASSERT(mode == SurfaceDataMode::D_BA); - return store_data_layout; - } - } suldst; - - union { - BitField<28, 1, u64> is_ba; - BitField<51, 3, ImageAtomicOperationType> operation_type; - BitField<33, 3, ImageType> image_type; - BitField<29, 4, ImageAtomicOperation> operation; - BitField<49, 2, OutOfBoundsStore> out_of_bounds_store; - } suatom_d; - - union { - BitField<20, 24, u64> target; - BitField<5, 1, u64> constant_buffer; - - [[nodiscard]] s32 GetBranchTarget() const { - // Sign extend the branch target offset - const auto mask = 1U << (24 - 1); - const auto target_value = static_cast<u32>(target); - constexpr auto instruction_size = static_cast<s32>(sizeof(Instruction)); - - // The branch offset is relative to the next instruction and is stored in bytes, so - // divide it by the size of an instruction and add 1 to it. - return static_cast<s32>((target_value ^ mask) - mask) / instruction_size + 1; - } - } bra; - - union { - BitField<20, 24, u64> target; - BitField<5, 1, u64> constant_buffer; - - [[nodiscard]] s32 GetBranchExtend() const { - // Sign extend the branch target offset - const auto mask = 1U << (24 - 1); - const auto target_value = static_cast<u32>(target); - constexpr auto instruction_size = static_cast<s32>(sizeof(Instruction)); - - // The branch offset is relative to the next instruction and is stored in bytes, so - // divide it by the size of an instruction and add 1 to it. - return static_cast<s32>((target_value ^ mask) - mask) / instruction_size + 1; - } - } brx; - - union { - BitField<39, 1, u64> emit; // EmitVertex - BitField<40, 1, u64> cut; // EndPrimitive - } out; - - union { - BitField<31, 1, u64> skew; - BitField<32, 1, u64> o; - BitField<33, 2, IsberdMode> mode; - BitField<47, 2, IsberdShift> shift; - } isberd; - - union { - BitField<8, 2, MembarType> type; - BitField<0, 2, MembarUnknown> unknown; - } membar; - - union { - BitField<48, 1, u64> signed_a; - BitField<38, 1, u64> is_byte_chunk_a; - BitField<36, 2, VideoType> type_a; - BitField<36, 2, u64> byte_height_a; - - BitField<49, 1, u64> signed_b; - BitField<50, 1, u64> use_register_b; - BitField<30, 1, u64> is_byte_chunk_b; - BitField<28, 2, VideoType> type_b; - BitField<28, 2, u64> byte_height_b; - } video; - - union { - BitField<51, 2, VmadShr> shr; - BitField<55, 1, u64> saturate; // Saturates the result (a * b + c) - BitField<47, 1, u64> cc; - } vmad; - - union { - BitField<54, 1, u64> is_dest_signed; - BitField<48, 1, u64> is_src_a_signed; - BitField<49, 1, u64> is_src_b_signed; - BitField<37, 2, u64> src_format_a; - BitField<29, 2, u64> src_format_b; - BitField<56, 1, u64> mx; - BitField<55, 1, u64> sat; - BitField<36, 2, u64> selector_a; - BitField<28, 2, u64> selector_b; - BitField<50, 1, u64> is_op_b_register; - BitField<51, 3, VmnmxOperation> operation; - - [[nodiscard]] VmnmxType SourceFormatA() const { - switch (src_format_a) { - case 0b11: - return VmnmxType::Bits32; - case 0b10: - return VmnmxType::Bits16; - default: - return VmnmxType::Bits8; - } - } - - [[nodiscard]] VmnmxType SourceFormatB() const { - switch (src_format_b) { - case 0b11: - return VmnmxType::Bits32; - case 0b10: - return VmnmxType::Bits16; - default: - return VmnmxType::Bits8; - } - } - } vmnmx; - - union { - BitField<20, 16, u64> imm20_16; - BitField<35, 1, u64> high_b_rr; // used on RR - BitField<36, 1, u64> product_shift_left; - BitField<37, 1, u64> merge_37; - BitField<48, 1, u64> sign_a; - BitField<49, 1, u64> sign_b; - BitField<50, 2, XmadMode> mode_cbf; // used by CR, RC - BitField<50, 3, XmadMode> mode; - BitField<52, 1, u64> high_b; - BitField<53, 1, u64> high_a; - BitField<55, 1, u64> product_shift_left_second; // used on CR - BitField<56, 1, u64> merge_56; - } xmad; - - union { - BitField<20, 14, u64> shifted_offset; - BitField<34, 5, u64> index; - - [[nodiscard]] u64 GetOffset() const { - return shifted_offset * 4; - } - } cbuf34; - - union { - BitField<20, 16, s64> offset; - BitField<36, 5, u64> index; - - [[nodiscard]] s64 GetOffset() const { - return offset; - } - } cbuf36; - - // Unsure about the size of this one. - // It's always used with a gpr0, so any size should be fine. - BitField<20, 8, SystemVariable> sys20; - - BitField<47, 1, u64> generates_cc; - BitField<61, 1, u64> is_b_imm; - BitField<60, 1, u64> is_b_gpr; - BitField<59, 1, u64> is_c_gpr; - BitField<20, 24, s64> smem_imm; - BitField<0, 5, ConditionCode> flow_condition_code; - - Attribute attribute; - Sampler sampler; - Image image; - - u64 value; -}; -static_assert(sizeof(Instruction) == 0x8, "Incorrect structure size"); -static_assert(std::is_standard_layout_v<Instruction>, "Instruction is not standard layout"); - -class OpCode { -public: - enum class Id { - KIL, - SSY, - SYNC, - BRK, - DEPBAR, - VOTE, - VOTE_VTG, - SHFL, - FSWZADD, - BFE_C, - BFE_R, - BFE_IMM, - BFI_RC, - BFI_IMM_R, - BRA, - BRX, - PBK, - LD_A, - LD_L, - LD_S, - LD_C, - LD, // Load from generic memory - LDG, // Load from global memory - ST_A, - ST_L, - ST_S, - ST, // Store in generic memory - STG, // Store in global memory - RED, // Reduction operation - ATOM, // Atomic operation on global memory - ATOMS, // Atomic operation on shared memory - AL2P, // Transforms attribute memory into physical memory - TEX, - TEX_B, // Texture Load Bindless - TXQ, // Texture Query - TXQ_B, // Texture Query Bindless - TEXS, // Texture Fetch with scalar/non-vec4 source/destinations - TLD, // Texture Load - TLDS, // Texture Load with scalar/non-vec4 source/destinations - TLD4, // Texture Gather 4 - TLD4_B, // Texture Gather 4 Bindless - TLD4S, // Texture Load 4 with scalar / non - vec4 source / destinations - TMML_B, // Texture Mip Map Level - TMML, // Texture Mip Map Level - TXD, // Texture Gradient/Load with Derivates - TXD_B, // Texture Gradient/Load with Derivates Bindless - SUST, // Surface Store - SULD, // Surface Load - SUATOM, // Surface Atomic Operation - EXIT, - NOP, - IPA, - OUT_R, // Emit vertex/primitive - ISBERD, - BAR, - MEMBAR, - VMAD, - VSETP, - VMNMX, - FFMA_IMM, // Fused Multiply and Add - FFMA_CR, - FFMA_RC, - FFMA_RR, - FADD_C, - FADD_R, - FADD_IMM, - FADD32I, - FMUL_C, - FMUL_R, - FMUL_IMM, - FMUL32_IMM, - IADD_C, - IADD_R, - IADD_IMM, - IADD3_C, // Add 3 Integers - IADD3_R, - IADD3_IMM, - IADD32I, - ISCADD_C, // Scale and Add - ISCADD_R, - ISCADD_IMM, - FLO_R, - FLO_C, - FLO_IMM, - LEA_R1, - LEA_R2, - LEA_RZ, - LEA_IMM, - LEA_HI, - HADD2_C, - HADD2_R, - HADD2_IMM, - HMUL2_C, - HMUL2_R, - HMUL2_IMM, - HFMA2_CR, - HFMA2_RC, - HFMA2_RR, - HFMA2_IMM_R, - HSETP2_C, - HSETP2_R, - HSETP2_IMM, - HSET2_C, - HSET2_R, - HSET2_IMM, - POPC_C, - POPC_R, - POPC_IMM, - SEL_C, - SEL_R, - SEL_IMM, - ICMP_RC, - ICMP_R, - ICMP_CR, - ICMP_IMM, - FCMP_RR, - FCMP_RC, - FCMP_IMMR, - MUFU, // Multi-Function Operator - RRO_C, // Range Reduction Operator - RRO_R, - RRO_IMM, - F2F_C, - F2F_R, - F2F_IMM, - F2I_C, - F2I_R, - F2I_IMM, - I2F_C, - I2F_R, - I2F_IMM, - I2I_C, - I2I_R, - I2I_IMM, - LOP_C, - LOP_R, - LOP_IMM, - LOP32I, - LOP3_C, - LOP3_R, - LOP3_IMM, - MOV_C, - MOV_R, - MOV_IMM, - S2R, - MOV32_IMM, - SHL_C, - SHL_R, - SHL_IMM, - SHR_C, - SHR_R, - SHR_IMM, - SHF_RIGHT_R, - SHF_RIGHT_IMM, - SHF_LEFT_R, - SHF_LEFT_IMM, - FMNMX_C, - FMNMX_R, - FMNMX_IMM, - IMNMX_C, - IMNMX_R, - IMNMX_IMM, - FSETP_C, // Set Predicate - FSETP_R, - FSETP_IMM, - FSET_C, - FSET_R, - FSET_IMM, - ISETP_C, - ISETP_IMM, - ISETP_R, - ISET_R, - ISET_C, - ISET_IMM, - PSETP, - PSET, - CSETP, - R2P_IMM, - P2R_IMM, - XMAD_IMM, - XMAD_CR, - XMAD_RC, - XMAD_RR, - }; - - enum class Type { - Trivial, - Arithmetic, - ArithmeticImmediate, - ArithmeticInteger, - ArithmeticIntegerImmediate, - ArithmeticHalf, - ArithmeticHalfImmediate, - Bfe, - Bfi, - Shift, - Ffma, - Hfma2, - Flow, - Synch, - Warp, - Memory, - Texture, - Image, - FloatSet, - FloatSetPredicate, - IntegerSet, - IntegerSetPredicate, - HalfSet, - HalfSetPredicate, - PredicateSetPredicate, - PredicateSetRegister, - RegisterSetPredicate, - Conversion, - Video, - Xmad, - Unknown, - }; - - /// Returns whether an opcode has an execution predicate field or not (ie, whether it can be - /// conditionally executed). - [[nodiscard]] static bool IsPredicatedInstruction(Id opcode) { - // TODO(Subv): Add the rest of unpredicated instructions. - return opcode != Id::SSY && opcode != Id::PBK; - } - - class Matcher { - public: - constexpr Matcher(const char* const name_, u16 mask_, u16 expected_, Id id_, Type type_) - : name{name_}, mask{mask_}, expected{expected_}, id{id_}, type{type_} {} - - [[nodiscard]] constexpr const char* GetName() const { - return name; - } - - [[nodiscard]] constexpr u16 GetMask() const { - return mask; - } - - [[nodiscard]] constexpr Id GetId() const { - return id; - } - - [[nodiscard]] constexpr Type GetType() const { - return type; - } - - /** - * Tests to see if the given instruction is the instruction this matcher represents. - * @param instruction The instruction to test - * @returns true if the given instruction matches. - */ - [[nodiscard]] constexpr bool Matches(u16 instruction) const { - return (instruction & mask) == expected; - } - - private: - const char* name; - u16 mask; - u16 expected; - Id id; - Type type; - }; - - using DecodeResult = std::optional<std::reference_wrapper<const Matcher>>; - [[nodiscard]] static DecodeResult Decode(Instruction instr) { - static const auto table{GetDecodeTable()}; - - const auto matches_instruction = [instr](const auto& matcher) { - return matcher.Matches(static_cast<u16>(instr.opcode)); - }; - - auto iter = std::find_if(table.begin(), table.end(), matches_instruction); - return iter != table.end() ? std::optional<std::reference_wrapper<const Matcher>>(*iter) - : std::nullopt; - } - -private: - struct Detail { - private: - static constexpr std::size_t opcode_bitsize = 16; - - /** - * Generates the mask and the expected value after masking from a given bitstring. - * A '0' in a bitstring indicates that a zero must be present at that bit position. - * A '1' in a bitstring indicates that a one must be present at that bit position. - */ - [[nodiscard]] static constexpr auto GetMaskAndExpect(const char* const bitstring) { - u16 mask = 0, expect = 0; - for (std::size_t i = 0; i < opcode_bitsize; i++) { - const std::size_t bit_position = opcode_bitsize - i - 1; - switch (bitstring[i]) { - case '0': - mask |= static_cast<u16>(1U << bit_position); - break; - case '1': - expect |= static_cast<u16>(1U << bit_position); - mask |= static_cast<u16>(1U << bit_position); - break; - default: - // Ignore - break; - } - } - return std::make_pair(mask, expect); - } - - public: - /// Creates a matcher that can match and parse instructions based on bitstring. - [[nodiscard]] static constexpr auto GetMatcher(const char* const bitstring, Id op, - Type type, const char* const name) { - const auto [mask, expected] = GetMaskAndExpect(bitstring); - return Matcher(name, mask, expected, op, type); - } - }; - - [[nodiscard]] static std::vector<Matcher> GetDecodeTable() { - std::vector<Matcher> table = { -#define INST(bitstring, op, type, name) Detail::GetMatcher(bitstring, op, type, name) - INST("111000110011----", Id::KIL, Type::Flow, "KIL"), - INST("111000101001----", Id::SSY, Type::Flow, "SSY"), - INST("111000101010----", Id::PBK, Type::Flow, "PBK"), - INST("111000100100----", Id::BRA, Type::Flow, "BRA"), - INST("111000100101----", Id::BRX, Type::Flow, "BRX"), - INST("1111000011111---", Id::SYNC, Type::Flow, "SYNC"), - INST("111000110100----", Id::BRK, Type::Flow, "BRK"), - INST("111000110000----", Id::EXIT, Type::Flow, "EXIT"), - INST("1111000011110---", Id::DEPBAR, Type::Synch, "DEPBAR"), - INST("0101000011011---", Id::VOTE, Type::Warp, "VOTE"), - INST("0101000011100---", Id::VOTE_VTG, Type::Warp, "VOTE_VTG"), - INST("1110111100010---", Id::SHFL, Type::Warp, "SHFL"), - INST("0101000011111---", Id::FSWZADD, Type::Warp, "FSWZADD"), - INST("1110111111011---", Id::LD_A, Type::Memory, "LD_A"), - INST("1110111101001---", Id::LD_S, Type::Memory, "LD_S"), - INST("1110111101000---", Id::LD_L, Type::Memory, "LD_L"), - INST("1110111110010---", Id::LD_C, Type::Memory, "LD_C"), - INST("100-------------", Id::LD, Type::Memory, "LD"), - INST("1110111011010---", Id::LDG, Type::Memory, "LDG"), - INST("1110111111110---", Id::ST_A, Type::Memory, "ST_A"), - INST("1110111101011---", Id::ST_S, Type::Memory, "ST_S"), - INST("1110111101010---", Id::ST_L, Type::Memory, "ST_L"), - INST("101-------------", Id::ST, Type::Memory, "ST"), - INST("1110111011011---", Id::STG, Type::Memory, "STG"), - INST("1110101111111---", Id::RED, Type::Memory, "RED"), - INST("11101101--------", Id::ATOM, Type::Memory, "ATOM"), - INST("11101100--------", Id::ATOMS, Type::Memory, "ATOMS"), - INST("1110111110100---", Id::AL2P, Type::Memory, "AL2P"), - INST("110000----111---", Id::TEX, Type::Texture, "TEX"), - INST("1101111010111---", Id::TEX_B, Type::Texture, "TEX_B"), - INST("1101111101001---", Id::TXQ, Type::Texture, "TXQ"), - INST("1101111101010---", Id::TXQ_B, Type::Texture, "TXQ_B"), - INST("1101-00---------", Id::TEXS, Type::Texture, "TEXS"), - INST("11011100--11----", Id::TLD, Type::Texture, "TLD"), - INST("1101-01---------", Id::TLDS, Type::Texture, "TLDS"), - INST("110010----111---", Id::TLD4, Type::Texture, "TLD4"), - INST("1101111011111---", Id::TLD4_B, Type::Texture, "TLD4_B"), - INST("11011111-0------", Id::TLD4S, Type::Texture, "TLD4S"), - INST("110111110110----", Id::TMML_B, Type::Texture, "TMML_B"), - INST("1101111101011---", Id::TMML, Type::Texture, "TMML"), - INST("11011110011110--", Id::TXD_B, Type::Texture, "TXD_B"), - INST("11011110001110--", Id::TXD, Type::Texture, "TXD"), - INST("11101011001-----", Id::SUST, Type::Image, "SUST"), - INST("11101011000-----", Id::SULD, Type::Image, "SULD"), - INST("1110101000------", Id::SUATOM, Type::Image, "SUATOM_D"), - INST("0101000010110---", Id::NOP, Type::Trivial, "NOP"), - INST("11100000--------", Id::IPA, Type::Trivial, "IPA"), - INST("1111101111100---", Id::OUT_R, Type::Trivial, "OUT_R"), - INST("1110111111010---", Id::ISBERD, Type::Trivial, "ISBERD"), - INST("1111000010101---", Id::BAR, Type::Trivial, "BAR"), - INST("1110111110011---", Id::MEMBAR, Type::Trivial, "MEMBAR"), - INST("01011111--------", Id::VMAD, Type::Video, "VMAD"), - INST("0101000011110---", Id::VSETP, Type::Video, "VSETP"), - INST("0011101---------", Id::VMNMX, Type::Video, "VMNMX"), - INST("0011001-1-------", Id::FFMA_IMM, Type::Ffma, "FFMA_IMM"), - INST("010010011-------", Id::FFMA_CR, Type::Ffma, "FFMA_CR"), - INST("010100011-------", Id::FFMA_RC, Type::Ffma, "FFMA_RC"), - INST("010110011-------", Id::FFMA_RR, Type::Ffma, "FFMA_RR"), - INST("0100110001011---", Id::FADD_C, Type::Arithmetic, "FADD_C"), - INST("0101110001011---", Id::FADD_R, Type::Arithmetic, "FADD_R"), - INST("0011100-01011---", Id::FADD_IMM, Type::Arithmetic, "FADD_IMM"), - INST("000010----------", Id::FADD32I, Type::ArithmeticImmediate, "FADD32I"), - INST("0100110001101---", Id::FMUL_C, Type::Arithmetic, "FMUL_C"), - INST("0101110001101---", Id::FMUL_R, Type::Arithmetic, "FMUL_R"), - INST("0011100-01101---", Id::FMUL_IMM, Type::Arithmetic, "FMUL_IMM"), - INST("00011110--------", Id::FMUL32_IMM, Type::ArithmeticImmediate, "FMUL32_IMM"), - INST("0100110000010---", Id::IADD_C, Type::ArithmeticInteger, "IADD_C"), - INST("0101110000010---", Id::IADD_R, Type::ArithmeticInteger, "IADD_R"), - INST("0011100-00010---", Id::IADD_IMM, Type::ArithmeticInteger, "IADD_IMM"), - INST("010011001100----", Id::IADD3_C, Type::ArithmeticInteger, "IADD3_C"), - INST("010111001100----", Id::IADD3_R, Type::ArithmeticInteger, "IADD3_R"), - INST("0011100-1100----", Id::IADD3_IMM, Type::ArithmeticInteger, "IADD3_IMM"), - INST("0001110---------", Id::IADD32I, Type::ArithmeticIntegerImmediate, "IADD32I"), - INST("0100110000011---", Id::ISCADD_C, Type::ArithmeticInteger, "ISCADD_C"), - INST("0101110000011---", Id::ISCADD_R, Type::ArithmeticInteger, "ISCADD_R"), - INST("0011100-00011---", Id::ISCADD_IMM, Type::ArithmeticInteger, "ISCADD_IMM"), - INST("0100110000001---", Id::POPC_C, Type::ArithmeticInteger, "POPC_C"), - INST("0101110000001---", Id::POPC_R, Type::ArithmeticInteger, "POPC_R"), - INST("0011100-00001---", Id::POPC_IMM, Type::ArithmeticInteger, "POPC_IMM"), - INST("0100110010100---", Id::SEL_C, Type::ArithmeticInteger, "SEL_C"), - INST("0101110010100---", Id::SEL_R, Type::ArithmeticInteger, "SEL_R"), - INST("0011100-10100---", Id::SEL_IMM, Type::ArithmeticInteger, "SEL_IMM"), - INST("010100110100----", Id::ICMP_RC, Type::ArithmeticInteger, "ICMP_RC"), - INST("010110110100----", Id::ICMP_R, Type::ArithmeticInteger, "ICMP_R"), - INST("010010110100----", Id::ICMP_CR, Type::ArithmeticInteger, "ICMP_CR"), - INST("0011011-0100----", Id::ICMP_IMM, Type::ArithmeticInteger, "ICMP_IMM"), - INST("0101110000110---", Id::FLO_R, Type::ArithmeticInteger, "FLO_R"), - INST("0100110000110---", Id::FLO_C, Type::ArithmeticInteger, "FLO_C"), - INST("0011100-00110---", Id::FLO_IMM, Type::ArithmeticInteger, "FLO_IMM"), - INST("0101101111011---", Id::LEA_R2, Type::ArithmeticInteger, "LEA_R2"), - INST("0101101111010---", Id::LEA_R1, Type::ArithmeticInteger, "LEA_R1"), - INST("001101101101----", Id::LEA_IMM, Type::ArithmeticInteger, "LEA_IMM"), - INST("010010111101----", Id::LEA_RZ, Type::ArithmeticInteger, "LEA_RZ"), - INST("00011000--------", Id::LEA_HI, Type::ArithmeticInteger, "LEA_HI"), - INST("0111101-1-------", Id::HADD2_C, Type::ArithmeticHalf, "HADD2_C"), - INST("0101110100010---", Id::HADD2_R, Type::ArithmeticHalf, "HADD2_R"), - INST("0111101-0-------", Id::HADD2_IMM, Type::ArithmeticHalfImmediate, "HADD2_IMM"), - INST("0111100-1-------", Id::HMUL2_C, Type::ArithmeticHalf, "HMUL2_C"), - INST("0101110100001---", Id::HMUL2_R, Type::ArithmeticHalf, "HMUL2_R"), - INST("0111100-0-------", Id::HMUL2_IMM, Type::ArithmeticHalfImmediate, "HMUL2_IMM"), - INST("01110---1-------", Id::HFMA2_CR, Type::Hfma2, "HFMA2_CR"), - INST("01100---1-------", Id::HFMA2_RC, Type::Hfma2, "HFMA2_RC"), - INST("0101110100000---", Id::HFMA2_RR, Type::Hfma2, "HFMA2_RR"), - INST("01110---0-------", Id::HFMA2_IMM_R, Type::Hfma2, "HFMA2_R_IMM"), - INST("0111111-1-------", Id::HSETP2_C, Type::HalfSetPredicate, "HSETP2_C"), - INST("0101110100100---", Id::HSETP2_R, Type::HalfSetPredicate, "HSETP2_R"), - INST("0111111-0-------", Id::HSETP2_IMM, Type::HalfSetPredicate, "HSETP2_IMM"), - INST("0111110-1-------", Id::HSET2_C, Type::HalfSet, "HSET2_C"), - INST("0101110100011---", Id::HSET2_R, Type::HalfSet, "HSET2_R"), - INST("0111110-0-------", Id::HSET2_IMM, Type::HalfSet, "HSET2_IMM"), - INST("010110111010----", Id::FCMP_RR, Type::Arithmetic, "FCMP_RR"), - INST("010010111010----", Id::FCMP_RC, Type::Arithmetic, "FCMP_RC"), - INST("0011011-1010----", Id::FCMP_IMMR, Type::Arithmetic, "FCMP_IMMR"), - INST("0101000010000---", Id::MUFU, Type::Arithmetic, "MUFU"), - INST("0100110010010---", Id::RRO_C, Type::Arithmetic, "RRO_C"), - INST("0101110010010---", Id::RRO_R, Type::Arithmetic, "RRO_R"), - INST("0011100-10010---", Id::RRO_IMM, Type::Arithmetic, "RRO_IMM"), - INST("0100110010101---", Id::F2F_C, Type::Conversion, "F2F_C"), - INST("0101110010101---", Id::F2F_R, Type::Conversion, "F2F_R"), - INST("0011100-10101---", Id::F2F_IMM, Type::Conversion, "F2F_IMM"), - INST("0100110010110---", Id::F2I_C, Type::Conversion, "F2I_C"), - INST("0101110010110---", Id::F2I_R, Type::Conversion, "F2I_R"), - INST("0011100-10110---", Id::F2I_IMM, Type::Conversion, "F2I_IMM"), - INST("0100110010011---", Id::MOV_C, Type::Arithmetic, "MOV_C"), - INST("0101110010011---", Id::MOV_R, Type::Arithmetic, "MOV_R"), - INST("0011100-10011---", Id::MOV_IMM, Type::Arithmetic, "MOV_IMM"), - INST("1111000011001---", Id::S2R, Type::Trivial, "S2R"), - INST("000000010000----", Id::MOV32_IMM, Type::ArithmeticImmediate, "MOV32_IMM"), - INST("0100110001100---", Id::FMNMX_C, Type::Arithmetic, "FMNMX_C"), - INST("0101110001100---", Id::FMNMX_R, Type::Arithmetic, "FMNMX_R"), - INST("0011100-01100---", Id::FMNMX_IMM, Type::Arithmetic, "FMNMX_IMM"), - INST("0100110000100---", Id::IMNMX_C, Type::ArithmeticInteger, "IMNMX_C"), - INST("0101110000100---", Id::IMNMX_R, Type::ArithmeticInteger, "IMNMX_R"), - INST("0011100-00100---", Id::IMNMX_IMM, Type::ArithmeticInteger, "IMNMX_IMM"), - INST("0100110000000---", Id::BFE_C, Type::Bfe, "BFE_C"), - INST("0101110000000---", Id::BFE_R, Type::Bfe, "BFE_R"), - INST("0011100-00000---", Id::BFE_IMM, Type::Bfe, "BFE_IMM"), - INST("0101001111110---", Id::BFI_RC, Type::Bfi, "BFI_RC"), - INST("0011011-11110---", Id::BFI_IMM_R, Type::Bfi, "BFI_IMM_R"), - INST("0100110001000---", Id::LOP_C, Type::ArithmeticInteger, "LOP_C"), - INST("0101110001000---", Id::LOP_R, Type::ArithmeticInteger, "LOP_R"), - INST("0011100-01000---", Id::LOP_IMM, Type::ArithmeticInteger, "LOP_IMM"), - INST("000001----------", Id::LOP32I, Type::ArithmeticIntegerImmediate, "LOP32I"), - INST("0000001---------", Id::LOP3_C, Type::ArithmeticInteger, "LOP3_C"), - INST("0101101111100---", Id::LOP3_R, Type::ArithmeticInteger, "LOP3_R"), - INST("0011110---------", Id::LOP3_IMM, Type::ArithmeticInteger, "LOP3_IMM"), - INST("0100110001001---", Id::SHL_C, Type::Shift, "SHL_C"), - INST("0101110001001---", Id::SHL_R, Type::Shift, "SHL_R"), - INST("0011100-01001---", Id::SHL_IMM, Type::Shift, "SHL_IMM"), - INST("0100110000101---", Id::SHR_C, Type::Shift, "SHR_C"), - INST("0101110000101---", Id::SHR_R, Type::Shift, "SHR_R"), - INST("0011100-00101---", Id::SHR_IMM, Type::Shift, "SHR_IMM"), - INST("0101110011111---", Id::SHF_RIGHT_R, Type::Shift, "SHF_RIGHT_R"), - INST("0011100-11111---", Id::SHF_RIGHT_IMM, Type::Shift, "SHF_RIGHT_IMM"), - INST("0101101111111---", Id::SHF_LEFT_R, Type::Shift, "SHF_LEFT_R"), - INST("0011011-11111---", Id::SHF_LEFT_IMM, Type::Shift, "SHF_LEFT_IMM"), - INST("0100110011100---", Id::I2I_C, Type::Conversion, "I2I_C"), - INST("0101110011100---", Id::I2I_R, Type::Conversion, "I2I_R"), - INST("0011100-11100---", Id::I2I_IMM, Type::Conversion, "I2I_IMM"), - INST("0100110010111---", Id::I2F_C, Type::Conversion, "I2F_C"), - INST("0101110010111---", Id::I2F_R, Type::Conversion, "I2F_R"), - INST("0011100-10111---", Id::I2F_IMM, Type::Conversion, "I2F_IMM"), - INST("01011000--------", Id::FSET_R, Type::FloatSet, "FSET_R"), - INST("0100100---------", Id::FSET_C, Type::FloatSet, "FSET_C"), - INST("0011000---------", Id::FSET_IMM, Type::FloatSet, "FSET_IMM"), - INST("010010111011----", Id::FSETP_C, Type::FloatSetPredicate, "FSETP_C"), - INST("010110111011----", Id::FSETP_R, Type::FloatSetPredicate, "FSETP_R"), - INST("0011011-1011----", Id::FSETP_IMM, Type::FloatSetPredicate, "FSETP_IMM"), - INST("010010110110----", Id::ISETP_C, Type::IntegerSetPredicate, "ISETP_C"), - INST("010110110110----", Id::ISETP_R, Type::IntegerSetPredicate, "ISETP_R"), - INST("0011011-0110----", Id::ISETP_IMM, Type::IntegerSetPredicate, "ISETP_IMM"), - INST("010110110101----", Id::ISET_R, Type::IntegerSet, "ISET_R"), - INST("010010110101----", Id::ISET_C, Type::IntegerSet, "ISET_C"), - INST("0011011-0101----", Id::ISET_IMM, Type::IntegerSet, "ISET_IMM"), - INST("0101000010001---", Id::PSET, Type::PredicateSetRegister, "PSET"), - INST("0101000010010---", Id::PSETP, Type::PredicateSetPredicate, "PSETP"), - INST("010100001010----", Id::CSETP, Type::PredicateSetPredicate, "CSETP"), - INST("0011100-11110---", Id::R2P_IMM, Type::RegisterSetPredicate, "R2P_IMM"), - INST("0011100-11101---", Id::P2R_IMM, Type::RegisterSetPredicate, "P2R_IMM"), - INST("0011011-00------", Id::XMAD_IMM, Type::Xmad, "XMAD_IMM"), - INST("0100111---------", Id::XMAD_CR, Type::Xmad, "XMAD_CR"), - INST("010100010-------", Id::XMAD_RC, Type::Xmad, "XMAD_RC"), - INST("0101101100------", Id::XMAD_RR, Type::Xmad, "XMAD_RR"), - }; -#undef INST - std::stable_sort(table.begin(), table.end(), [](const auto& a, const auto& b) { - // If a matcher has more bits in its mask it is more specific, so it - // should come first. - return std::bitset<16>(a.GetMask()).count() > std::bitset<16>(b.GetMask()).count(); - }); - - return table; - } -}; - -} // namespace Tegra::Shader diff --git a/src/video_core/engines/shader_header.h b/src/video_core/engines/shader_header.h deleted file mode 100644 index e0d7b89c5..000000000 --- a/src/video_core/engines/shader_header.h +++ /dev/null @@ -1,158 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <array> -#include <optional> - -#include "common/bit_field.h" -#include "common/common_funcs.h" -#include "common/common_types.h" - -namespace Tegra::Shader { - -enum class OutputTopology : u32 { - PointList = 1, - LineStrip = 6, - TriangleStrip = 7, -}; - -enum class PixelImap : u8 { - Unused = 0, - Constant = 1, - Perspective = 2, - ScreenLinear = 3, -}; - -// Documentation in: -// http://download.nvidia.com/open-gpu-doc/Shader-Program-Header/1/Shader-Program-Header.html -struct Header { - union { - BitField<0, 5, u32> sph_type; - BitField<5, 5, u32> version; - BitField<10, 4, u32> shader_type; - BitField<14, 1, u32> mrt_enable; - BitField<15, 1, u32> kills_pixels; - BitField<16, 1, u32> does_global_store; - BitField<17, 4, u32> sass_version; - BitField<21, 5, u32> reserved; - BitField<26, 1, u32> does_load_or_store; - BitField<27, 1, u32> does_fp64; - BitField<28, 4, u32> stream_out_mask; - } common0; - - union { - BitField<0, 24, u32> shader_local_memory_low_size; - BitField<24, 8, u32> per_patch_attribute_count; - } common1; - - union { - BitField<0, 24, u32> shader_local_memory_high_size; - BitField<24, 8, u32> threads_per_input_primitive; - } common2; - - union { - BitField<0, 24, u32> shader_local_memory_crs_size; - BitField<24, 4, OutputTopology> output_topology; - BitField<28, 4, u32> reserved; - } common3; - - union { - BitField<0, 12, u32> max_output_vertices; - BitField<12, 8, u32> store_req_start; // NOTE: not used by geometry shaders. - BitField<20, 4, u32> reserved; - BitField<24, 8, u32> store_req_end; // NOTE: not used by geometry shaders. - } common4; - - union { - struct { - INSERT_PADDING_BYTES_NOINIT(3); // ImapSystemValuesA - INSERT_PADDING_BYTES_NOINIT(1); // ImapSystemValuesB - INSERT_PADDING_BYTES_NOINIT(16); // ImapGenericVector[32] - INSERT_PADDING_BYTES_NOINIT(2); // ImapColor - union { - BitField<0, 8, u16> clip_distances; - BitField<8, 1, u16> point_sprite_s; - BitField<9, 1, u16> point_sprite_t; - BitField<10, 1, u16> fog_coordinate; - BitField<12, 1, u16> tessellation_eval_point_u; - BitField<13, 1, u16> tessellation_eval_point_v; - BitField<14, 1, u16> instance_id; - BitField<15, 1, u16> vertex_id; - }; - INSERT_PADDING_BYTES_NOINIT(5); // ImapFixedFncTexture[10] - INSERT_PADDING_BYTES_NOINIT(1); // ImapReserved - INSERT_PADDING_BYTES_NOINIT(3); // OmapSystemValuesA - INSERT_PADDING_BYTES_NOINIT(1); // OmapSystemValuesB - INSERT_PADDING_BYTES_NOINIT(16); // OmapGenericVector[32] - INSERT_PADDING_BYTES_NOINIT(2); // OmapColor - INSERT_PADDING_BYTES_NOINIT(2); // OmapSystemValuesC - INSERT_PADDING_BYTES_NOINIT(5); // OmapFixedFncTexture[10] - INSERT_PADDING_BYTES_NOINIT(1); // OmapReserved - } vtg; - - struct { - INSERT_PADDING_BYTES_NOINIT(3); // ImapSystemValuesA - INSERT_PADDING_BYTES_NOINIT(1); // ImapSystemValuesB - - union { - BitField<0, 2, PixelImap> x; - BitField<2, 2, PixelImap> y; - BitField<4, 2, PixelImap> z; - BitField<6, 2, PixelImap> w; - u8 raw; - } imap_generic_vector[32]; - - INSERT_PADDING_BYTES_NOINIT(2); // ImapColor - INSERT_PADDING_BYTES_NOINIT(2); // ImapSystemValuesC - INSERT_PADDING_BYTES_NOINIT(10); // ImapFixedFncTexture[10] - INSERT_PADDING_BYTES_NOINIT(2); // ImapReserved - - struct { - u32 target; - union { - BitField<0, 1, u32> sample_mask; - BitField<1, 1, u32> depth; - BitField<2, 30, u32> reserved; - }; - } omap; - - bool IsColorComponentOutputEnabled(u32 render_target, u32 component) const { - const u32 bit = render_target * 4 + component; - return omap.target & (1 << bit); - } - - PixelImap GetPixelImap(u32 attribute) const { - const auto get_index = [this, attribute](u32 index) { - return static_cast<PixelImap>( - (imap_generic_vector[attribute].raw >> (index * 2)) & 3); - }; - - std::optional<PixelImap> result; - for (u32 component = 0; component < 4; ++component) { - const PixelImap index = get_index(component); - if (index == PixelImap::Unused) { - continue; - } - if (result && result != index) { - LOG_CRITICAL(HW_GPU, "Generic attribute conflict in interpolation mode"); - } - result = index; - } - return result.value_or(PixelImap::Unused); - } - } ps; - - std::array<u32, 0xF> raw; - }; - - u64 GetLocalMemorySize() const { - return (common1.shader_local_memory_low_size | - (common2.shader_local_memory_high_size << 24)); - } -}; -static_assert(sizeof(Header) == 0x50, "Incorrect structure size"); - -} // namespace Tegra::Shader diff --git a/src/video_core/engines/shader_type.h b/src/video_core/engines/shader_type.h deleted file mode 100644 index 49ce5cde5..000000000 --- a/src/video_core/engines/shader_type.h +++ /dev/null @@ -1,21 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include "common/common_types.h" - -namespace Tegra::Engines { - -enum class ShaderType : u32 { - Vertex = 0, - TesselationControl = 1, - TesselationEval = 2, - Geometry = 3, - Fragment = 4, - Compute = 5, -}; -static constexpr std::size_t MaxShaderTypes = 6; - -} // namespace Tegra::Engines diff --git a/src/video_core/fence_manager.h b/src/video_core/fence_manager.h index f055b61e9..34dc6c596 100644 --- a/src/video_core/fence_manager.h +++ b/src/video_core/fence_manager.h @@ -8,6 +8,7 @@ #include <queue> #include "common/common_types.h" +#include "common/settings.h" #include "core/core.h" #include "video_core/delayed_destruction_ring.h" #include "video_core/gpu.h" @@ -53,6 +54,12 @@ public: delayed_destruction_ring.Tick(); } + // Unlike other fences, this one doesn't + void SignalOrdering() { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.AccumulateFlushes(); + } + void SignalSemaphore(GPUVAddr addr, u32 value) { TryReleasePendingFences(); const bool should_flush = ShouldFlush(); diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 35cc561be..ff024f530 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -50,6 +50,7 @@ void GPU::BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer_) { maxwell_3d->BindRasterizer(rasterizer); fermi_2d->BindRasterizer(rasterizer); kepler_compute->BindRasterizer(rasterizer); + maxwell_dma->BindRasterizer(rasterizer); } Engines::Maxwell3D& GPU::Maxwell3D() { @@ -268,11 +269,13 @@ void GPU::CallPullerMethod(const MethodCall& method_call) { case BufferMethods::SemaphoreAddressHigh: case BufferMethods::SemaphoreAddressLow: case BufferMethods::SemaphoreSequence: - case BufferMethods::RefCnt: case BufferMethods::UnkCacheFlush: case BufferMethods::WrcacheFlush: case BufferMethods::FenceValue: break; + case BufferMethods::RefCnt: + rasterizer->SignalReference(); + break; case BufferMethods::FenceAction: ProcessFenceActionMethod(); break; diff --git a/src/video_core/guest_driver.cpp b/src/video_core/guest_driver.cpp deleted file mode 100644 index f058f2744..000000000 --- a/src/video_core/guest_driver.cpp +++ /dev/null @@ -1,37 +0,0 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <limits> -#include <vector> - -#include "common/common_types.h" -#include "video_core/guest_driver.h" - -namespace VideoCore { - -void GuestDriverProfile::DeduceTextureHandlerSize(std::vector<u32> bound_offsets) { - if (texture_handler_size) { - return; - } - const std::size_t size = bound_offsets.size(); - if (size < 2) { - return; - } - std::sort(bound_offsets.begin(), bound_offsets.end(), std::less{}); - u32 min_val = std::numeric_limits<u32>::max(); - for (std::size_t i = 1; i < size; ++i) { - if (bound_offsets[i] == bound_offsets[i - 1]) { - continue; - } - const u32 new_min = bound_offsets[i] - bound_offsets[i - 1]; - min_val = std::min(min_val, new_min); - } - if (min_val > 2) { - return; - } - texture_handler_size = min_texture_handler_size * min_val; -} - -} // namespace VideoCore diff --git a/src/video_core/guest_driver.h b/src/video_core/guest_driver.h deleted file mode 100644 index 21e569ba1..000000000 --- a/src/video_core/guest_driver.h +++ /dev/null @@ -1,46 +0,0 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <optional> -#include <vector> - -#include "common/common_types.h" - -namespace VideoCore { - -/** - * The GuestDriverProfile class is used to learn about the GPU drivers behavior and collect - * information necessary for impossible to avoid HLE methods like shader tracks as they are - * Entscheidungsproblems. - */ -class GuestDriverProfile { -public: - explicit GuestDriverProfile() = default; - explicit GuestDriverProfile(std::optional<u32> texture_handler_size_) - : texture_handler_size{texture_handler_size_} {} - - void DeduceTextureHandlerSize(std::vector<u32> bound_offsets); - - u32 GetTextureHandlerSize() const { - return texture_handler_size.value_or(default_texture_handler_size); - } - - bool IsTextureHandlerSizeKnown() const { - return texture_handler_size.has_value(); - } - -private: - // Minimum size of texture handler any driver can use. - static constexpr u32 min_texture_handler_size = 4; - - // This goes with Vulkan and OpenGL standards but Nvidia GPUs can easily use 4 bytes instead. - // Thus, certain drivers may squish the size. - static constexpr u32 default_texture_handler_size = 8; - - std::optional<u32> texture_handler_size = default_texture_handler_size; -}; - -} // namespace VideoCore diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp index d2b9d5f2b..882eff880 100644 --- a/src/video_core/memory_manager.cpp +++ b/src/video_core/memory_manager.cpp @@ -69,7 +69,6 @@ void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) { } else { UNREACHABLE_MSG("Unmapping non-existent GPU address=0x{:x}", gpu_addr); } - const auto submapped_ranges = GetSubmappedRange(gpu_addr, size); for (const auto& map : submapped_ranges) { diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index 0cec4225b..b094fc064 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -11,11 +11,13 @@ #include "common/common_types.h" #include "video_core/engines/fermi_2d.h" #include "video_core/gpu.h" -#include "video_core/guest_driver.h" namespace Tegra { class MemoryManager; +namespace Engines { +class AccelerateDMAInterface; } +} // namespace Tegra namespace VideoCore { @@ -42,7 +44,7 @@ public: virtual void Clear() = 0; /// Dispatches a compute shader invocation - virtual void DispatchCompute(GPUVAddr code_addr) = 0; + virtual void DispatchCompute() = 0; /// Resets the counter of a query virtual void ResetCounter(QueryType type) = 0; @@ -63,6 +65,9 @@ public: /// Signal a GPU based syncpoint as a fence virtual void SignalSyncPoint(u32 value) = 0; + /// Signal a GPU based reference as point + virtual void SignalReference() = 0; + /// Release all pending fences. virtual void ReleaseFences() = 0; @@ -116,6 +121,8 @@ public: return false; } + [[nodiscard]] virtual Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() = 0; + /// Attempt to use a faster method to display the framebuffer to screen [[nodiscard]] virtual bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) { @@ -128,18 +135,5 @@ public: /// Initialize disk cached resources for the game being emulated virtual void LoadDiskResources(u64 title_id, std::stop_token stop_loading, const DiskResourceLoadCallback& callback) {} - - /// Grant access to the Guest Driver Profile for recording/obtaining info on the guest driver. - [[nodiscard]] GuestDriverProfile& AccessGuestDriverProfile() { - return guest_driver_profile; - } - - /// Grant access to the Guest Driver Profile for recording/obtaining info on the guest driver. - [[nodiscard]] const GuestDriverProfile& AccessGuestDriverProfile() const { - return guest_driver_profile; - } - -private: - GuestDriverProfile guest_driver_profile{}; }; } // namespace VideoCore diff --git a/src/video_core/renderer_opengl/gl_arb_decompiler.cpp b/src/video_core/renderer_opengl/gl_arb_decompiler.cpp deleted file mode 100644 index e8d8d2aa5..000000000 --- a/src/video_core/renderer_opengl/gl_arb_decompiler.cpp +++ /dev/null @@ -1,2124 +0,0 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <array> -#include <cstddef> -#include <string> -#include <string_view> -#include <utility> -#include <variant> - -#include <fmt/format.h> - -#include "common/alignment.h" -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/renderer_opengl/gl_arb_decompiler.h" -#include "video_core/renderer_opengl/gl_device.h" -#include "video_core/shader/registry.h" -#include "video_core/shader/shader_ir.h" - -// Predicates in the decompiled code follow the convention that -1 means true and 0 means false. -// GLASM lacks booleans, so they have to be implemented as integers. -// Using -1 for true is useful because both CMP.S and NOT.U can negate it, and CMP.S can be used to -// select between two values, because -1 will be evaluated as true and 0 as false. - -namespace OpenGL { - -namespace { - -using Tegra::Engines::ShaderType; -using Tegra::Shader::Attribute; -using Tegra::Shader::PixelImap; -using Tegra::Shader::Register; -using namespace VideoCommon::Shader; -using Operation = const OperationNode&; - -constexpr std::array INTERNAL_FLAG_NAMES = {"ZERO", "SIGN", "CARRY", "OVERFLOW"}; - -char Swizzle(std::size_t component) { - static constexpr std::string_view SWIZZLE{"xyzw"}; - return SWIZZLE.at(component); -} - -constexpr bool IsGenericAttribute(Attribute::Index index) { - return index >= Attribute::Index::Attribute_0 && index <= Attribute::Index::Attribute_31; -} - -u32 GetGenericAttributeIndex(Attribute::Index index) { - ASSERT(IsGenericAttribute(index)); - return static_cast<u32>(index) - static_cast<u32>(Attribute::Index::Attribute_0); -} - -std::string_view Modifiers(Operation operation) { - const auto meta = std::get_if<MetaArithmetic>(&operation.GetMeta()); - if (meta && meta->precise) { - return ".PREC"; - } - return ""; -} - -std::string_view GetInputFlags(PixelImap attribute) { - switch (attribute) { - case PixelImap::Perspective: - return ""; - case PixelImap::Constant: - return "FLAT "; - case PixelImap::ScreenLinear: - return "NOPERSPECTIVE "; - case PixelImap::Unused: - break; - } - UNIMPLEMENTED_MSG("Unknown attribute usage index={}", attribute); - return {}; -} - -std::string_view ImageType(Tegra::Shader::ImageType image_type) { - switch (image_type) { - case Tegra::Shader::ImageType::Texture1D: - return "1D"; - case Tegra::Shader::ImageType::TextureBuffer: - return "BUFFER"; - case Tegra::Shader::ImageType::Texture1DArray: - return "ARRAY1D"; - case Tegra::Shader::ImageType::Texture2D: - return "2D"; - case Tegra::Shader::ImageType::Texture2DArray: - return "ARRAY2D"; - case Tegra::Shader::ImageType::Texture3D: - return "3D"; - } - UNREACHABLE(); - return {}; -} - -std::string_view StackName(MetaStackClass stack) { - switch (stack) { - case MetaStackClass::Ssy: - return "SSY"; - case MetaStackClass::Pbk: - return "PBK"; - } - UNREACHABLE(); - return ""; -}; - -std::string_view PrimitiveDescription(Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology topology) { - switch (topology) { - case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::Points: - return "POINTS"; - case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::Lines: - case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::LineStrip: - return "LINES"; - case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::LinesAdjacency: - case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::LineStripAdjacency: - return "LINES_ADJACENCY"; - case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::Triangles: - case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::TriangleStrip: - case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::TriangleFan: - return "TRIANGLES"; - case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::TrianglesAdjacency: - case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::TriangleStripAdjacency: - return "TRIANGLES_ADJACENCY"; - default: - UNIMPLEMENTED_MSG("topology={}", topology); - return "POINTS"; - } -} - -std::string_view TopologyName(Tegra::Shader::OutputTopology topology) { - switch (topology) { - case Tegra::Shader::OutputTopology::PointList: - return "POINTS"; - case Tegra::Shader::OutputTopology::LineStrip: - return "LINE_STRIP"; - case Tegra::Shader::OutputTopology::TriangleStrip: - return "TRIANGLE_STRIP"; - default: - UNIMPLEMENTED_MSG("Unknown output topology: {}", topology); - return "points"; - } -} - -std::string_view StageInputName(ShaderType stage) { - switch (stage) { - case ShaderType::Vertex: - case ShaderType::Geometry: - return "vertex"; - case ShaderType::Fragment: - return "fragment"; - case ShaderType::Compute: - return "invocation"; - default: - UNREACHABLE(); - return ""; - } -} - -std::string TextureType(const MetaTexture& meta) { - if (meta.sampler.is_buffer) { - return "BUFFER"; - } - std::string type; - if (meta.sampler.is_shadow) { - type += "SHADOW"; - } - if (meta.sampler.is_array) { - type += "ARRAY"; - } - type += [&meta] { - switch (meta.sampler.type) { - case Tegra::Shader::TextureType::Texture1D: - return "1D"; - case Tegra::Shader::TextureType::Texture2D: - return "2D"; - case Tegra::Shader::TextureType::Texture3D: - return "3D"; - case Tegra::Shader::TextureType::TextureCube: - return "CUBE"; - } - UNREACHABLE(); - return "2D"; - }(); - return type; -} - -class ARBDecompiler final { -public: - explicit ARBDecompiler(const Device& device_, const ShaderIR& ir_, const Registry& registry_, - ShaderType stage_, std::string_view identifier); - - std::string Code() const { - return shader_source; - } - -private: - void DefineGlobalMemory(); - - void DeclareHeader(); - void DeclareVertex(); - void DeclareGeometry(); - void DeclareFragment(); - void DeclareCompute(); - void DeclareInputAttributes(); - void DeclareOutputAttributes(); - void DeclareLocalMemory(); - void DeclareGlobalMemory(); - void DeclareConstantBuffers(); - void DeclareRegisters(); - void DeclareTemporaries(); - void DeclarePredicates(); - void DeclareInternalFlags(); - - void InitializeVariables(); - - void DecompileAST(); - void DecompileBranchMode(); - - void VisitAST(const ASTNode& node); - std::string VisitExpression(const Expr& node); - - void VisitBlock(const NodeBlock& bb); - - std::string Visit(const Node& node); - - std::tuple<std::string, std::string, std::size_t> BuildCoords(Operation); - std::string BuildAoffi(Operation); - std::string GlobalMemoryPointer(const GmemNode& gmem); - void Exit(); - - std::string Assign(Operation); - std::string Select(Operation); - std::string FClamp(Operation); - std::string FCastHalf0(Operation); - std::string FCastHalf1(Operation); - std::string FSqrt(Operation); - std::string FSwizzleAdd(Operation); - std::string HAdd2(Operation); - std::string HMul2(Operation); - std::string HFma2(Operation); - std::string HAbsolute(Operation); - std::string HNegate(Operation); - std::string HClamp(Operation); - std::string HCastFloat(Operation); - std::string HUnpack(Operation); - std::string HMergeF32(Operation); - std::string HMergeH0(Operation); - std::string HMergeH1(Operation); - std::string HPack2(Operation); - std::string LogicalAssign(Operation); - std::string LogicalPick2(Operation); - std::string LogicalAnd2(Operation); - std::string FloatOrdered(Operation); - std::string FloatUnordered(Operation); - std::string LogicalAddCarry(Operation); - std::string Texture(Operation); - std::string TextureGather(Operation); - std::string TextureQueryDimensions(Operation); - std::string TextureQueryLod(Operation); - std::string TexelFetch(Operation); - std::string TextureGradient(Operation); - std::string ImageLoad(Operation); - std::string ImageStore(Operation); - std::string Branch(Operation); - std::string BranchIndirect(Operation); - std::string PushFlowStack(Operation); - std::string PopFlowStack(Operation); - std::string Exit(Operation); - std::string Discard(Operation); - std::string EmitVertex(Operation); - std::string EndPrimitive(Operation); - std::string InvocationId(Operation); - std::string YNegate(Operation); - std::string ThreadId(Operation); - std::string ShuffleIndexed(Operation); - std::string Barrier(Operation); - std::string MemoryBarrierGroup(Operation); - std::string MemoryBarrierGlobal(Operation); - - template <const std::string_view& op> - std::string Unary(Operation operation) { - std::string temporary = AllocTemporary(); - AddLine("{}{} {}, {};", op, Modifiers(operation), temporary, Visit(operation[0])); - return temporary; - } - - template <const std::string_view& op> - std::string Binary(Operation operation) { - std::string temporary = AllocTemporary(); - AddLine("{}{} {}, {}, {};", op, Modifiers(operation), temporary, Visit(operation[0]), - Visit(operation[1])); - return temporary; - } - - template <const std::string_view& op> - std::string Trinary(Operation operation) { - std::string temporary = AllocTemporary(); - AddLine("{}{} {}, {}, {}, {};", op, Modifiers(operation), temporary, Visit(operation[0]), - Visit(operation[1]), Visit(operation[2])); - return temporary; - } - - template <const std::string_view& op, bool unordered> - std::string FloatComparison(Operation operation) { - std::string temporary = AllocTemporary(); - AddLine("TRUNC.U.CC RC.x, {};", Binary<op>(operation)); - AddLine("MOV.S {}, 0;", temporary); - AddLine("MOV.S {} (NE.x), -1;", temporary); - - const std::string op_a = Visit(operation[0]); - const std::string op_b = Visit(operation[1]); - if constexpr (unordered) { - AddLine("SNE.F RC.x, {}, {};", op_a, op_a); - AddLine("TRUNC.U.CC RC.x, RC.x;"); - AddLine("MOV.S {} (NE.x), -1;", temporary); - AddLine("SNE.F RC.x, {}, {};", op_b, op_b); - AddLine("TRUNC.U.CC RC.x, RC.x;"); - AddLine("MOV.S {} (NE.x), -1;", temporary); - } else if (op == SNE_F) { - AddLine("SNE.F RC.x, {}, {};", op_a, op_a); - AddLine("TRUNC.U.CC RC.x, RC.x;"); - AddLine("MOV.S {} (NE.x), 0;", temporary); - AddLine("SNE.F RC.x, {}, {};", op_b, op_b); - AddLine("TRUNC.U.CC RC.x, RC.x;"); - AddLine("MOV.S {} (NE.x), 0;", temporary); - } - return temporary; - } - - template <const std::string_view& op, bool is_nan> - std::string HalfComparison(Operation operation) { - std::string tmp1 = AllocVectorTemporary(); - const std::string tmp2 = AllocVectorTemporary(); - const std::string op_a = Visit(operation[0]); - const std::string op_b = Visit(operation[1]); - AddLine("UP2H.F {}, {};", tmp1, op_a); - AddLine("UP2H.F {}, {};", tmp2, op_b); - AddLine("{} {}, {}, {};", op, tmp1, tmp1, tmp2); - AddLine("TRUNC.U.CC RC.xy, {};", tmp1); - AddLine("MOV.S {}.xy, {{0, 0, 0, 0}};", tmp1); - AddLine("MOV.S {}.x (NE.x), -1;", tmp1); - AddLine("MOV.S {}.y (NE.y), -1;", tmp1); - if constexpr (is_nan) { - AddLine("MOVC.F RC.x, {};", op_a); - AddLine("MOV.S {}.x (NAN.x), -1;", tmp1); - AddLine("MOVC.F RC.x, {};", op_b); - AddLine("MOV.S {}.y (NAN.x), -1;", tmp1); - } - return tmp1; - } - - template <const std::string_view& op, const std::string_view& type> - std::string AtomicImage(Operation operation) { - const auto& meta = std::get<MetaImage>(operation.GetMeta()); - const u32 image_id = device.GetBaseBindings(stage).image + meta.image.index; - const std::size_t num_coords = operation.GetOperandsCount(); - const std::size_t num_values = meta.values.size(); - - const std::string coord = AllocVectorTemporary(); - const std::string value = AllocVectorTemporary(); - for (std::size_t i = 0; i < num_coords; ++i) { - AddLine("MOV.S {}.{}, {};", coord, Swizzle(i), Visit(operation[i])); - } - for (std::size_t i = 0; i < num_values; ++i) { - AddLine("MOV.F {}.{}, {};", value, Swizzle(i), Visit(meta.values[i])); - } - - AddLine("ATOMIM.{}.{} {}.x, {}, {}, image[{}], {};", op, type, coord, value, coord, - image_id, ImageType(meta.image.type)); - return fmt::format("{}.x", coord); - } - - template <const std::string_view& op, const std::string_view& type> - std::string Atomic(Operation operation) { - std::string temporary = AllocTemporary(); - std::string address; - std::string_view opname; - bool robust = false; - if (const auto gmem = std::get_if<GmemNode>(&*operation[0])) { - address = GlobalMemoryPointer(*gmem); - opname = "ATOM"; - robust = true; - } else if (const auto smem = std::get_if<SmemNode>(&*operation[0])) { - address = fmt::format("shared_mem[{}]", Visit(smem->GetAddress())); - opname = "ATOMS"; - } else { - UNREACHABLE(); - return "{0, 0, 0, 0}"; - } - if (robust) { - AddLine("IF NE.x;"); - } - AddLine("{}.{}.{} {}, {}, {};", opname, op, type, temporary, Visit(operation[1]), address); - if (robust) { - AddLine("ELSE;"); - AddLine("MOV.S {}, 0;", temporary); - AddLine("ENDIF;"); - } - return temporary; - } - - template <char type> - std::string Negate(Operation operation) { - std::string temporary = AllocTemporary(); - if constexpr (type == 'F') { - AddLine("MOV.F32 {}, -{};", temporary, Visit(operation[0])); - } else { - AddLine("MOV.{} {}, -{};", type, temporary, Visit(operation[0])); - } - return temporary; - } - - template <char type> - std::string Absolute(Operation operation) { - std::string temporary = AllocTemporary(); - AddLine("MOV.{} {}, |{}|;", type, temporary, Visit(operation[0])); - return temporary; - } - - template <char type> - std::string BitfieldInsert(Operation operation) { - const std::string temporary = AllocVectorTemporary(); - AddLine("MOV.{} {}.x, {};", type, temporary, Visit(operation[3])); - AddLine("MOV.{} {}.y, {};", type, temporary, Visit(operation[2])); - AddLine("BFI.{} {}.x, {}, {}, {};", type, temporary, temporary, Visit(operation[1]), - Visit(operation[0])); - return fmt::format("{}.x", temporary); - } - - template <char type> - std::string BitfieldExtract(Operation operation) { - const std::string temporary = AllocVectorTemporary(); - AddLine("MOV.{} {}.x, {};", type, temporary, Visit(operation[2])); - AddLine("MOV.{} {}.y, {};", type, temporary, Visit(operation[1])); - AddLine("BFE.{} {}.x, {}, {};", type, temporary, temporary, Visit(operation[0])); - return fmt::format("{}.x", temporary); - } - - template <char swizzle> - std::string LocalInvocationId(Operation) { - return fmt::format("invocation.localid.{}", swizzle); - } - - template <char swizzle> - std::string WorkGroupId(Operation) { - return fmt::format("invocation.groupid.{}", swizzle); - } - - template <char c1, char c2> - std::string ThreadMask(Operation) { - return fmt::format("{}.thread{}{}mask", StageInputName(stage), c1, c2); - } - - template <typename... Args> - void AddExpression(std::string_view text, Args&&... args) { - shader_source += fmt::format(fmt::runtime(text), std::forward<Args>(args)...); - } - - template <typename... Args> - void AddLine(std::string_view text, Args&&... args) { - AddExpression(text, std::forward<Args>(args)...); - shader_source += '\n'; - } - - std::string AllocLongVectorTemporary() { - max_long_temporaries = std::max(max_long_temporaries, num_long_temporaries + 1); - return fmt::format("L{}", num_long_temporaries++); - } - - std::string AllocLongTemporary() { - return fmt::format("{}.x", AllocLongVectorTemporary()); - } - - std::string AllocVectorTemporary() { - max_temporaries = std::max(max_temporaries, num_temporaries + 1); - return fmt::format("T{}", num_temporaries++); - } - - std::string AllocTemporary() { - return fmt::format("{}.x", AllocVectorTemporary()); - } - - void ResetTemporaries() noexcept { - num_temporaries = 0; - num_long_temporaries = 0; - } - - const Device& device; - const ShaderIR& ir; - const Registry& registry; - const ShaderType stage; - - std::size_t num_temporaries = 0; - std::size_t max_temporaries = 0; - - std::size_t num_long_temporaries = 0; - std::size_t max_long_temporaries = 0; - - std::map<GlobalMemoryBase, u32> global_memory_names; - - std::string shader_source; - - static constexpr std::string_view ADD_F32 = "ADD.F32"; - static constexpr std::string_view ADD_S = "ADD.S"; - static constexpr std::string_view ADD_U = "ADD.U"; - static constexpr std::string_view MUL_F32 = "MUL.F32"; - static constexpr std::string_view MUL_S = "MUL.S"; - static constexpr std::string_view MUL_U = "MUL.U"; - static constexpr std::string_view DIV_F32 = "DIV.F32"; - static constexpr std::string_view DIV_S = "DIV.S"; - static constexpr std::string_view DIV_U = "DIV.U"; - static constexpr std::string_view MAD_F32 = "MAD.F32"; - static constexpr std::string_view RSQ_F32 = "RSQ.F32"; - static constexpr std::string_view COS_F32 = "COS.F32"; - static constexpr std::string_view SIN_F32 = "SIN.F32"; - static constexpr std::string_view EX2_F32 = "EX2.F32"; - static constexpr std::string_view LG2_F32 = "LG2.F32"; - static constexpr std::string_view SLT_F = "SLT.F32"; - static constexpr std::string_view SLT_S = "SLT.S"; - static constexpr std::string_view SLT_U = "SLT.U"; - static constexpr std::string_view SEQ_F = "SEQ.F32"; - static constexpr std::string_view SEQ_S = "SEQ.S"; - static constexpr std::string_view SEQ_U = "SEQ.U"; - static constexpr std::string_view SLE_F = "SLE.F32"; - static constexpr std::string_view SLE_S = "SLE.S"; - static constexpr std::string_view SLE_U = "SLE.U"; - static constexpr std::string_view SGT_F = "SGT.F32"; - static constexpr std::string_view SGT_S = "SGT.S"; - static constexpr std::string_view SGT_U = "SGT.U"; - static constexpr std::string_view SNE_F = "SNE.F32"; - static constexpr std::string_view SNE_S = "SNE.S"; - static constexpr std::string_view SNE_U = "SNE.U"; - static constexpr std::string_view SGE_F = "SGE.F32"; - static constexpr std::string_view SGE_S = "SGE.S"; - static constexpr std::string_view SGE_U = "SGE.U"; - static constexpr std::string_view AND_S = "AND.S"; - static constexpr std::string_view AND_U = "AND.U"; - static constexpr std::string_view TRUNC_F = "TRUNC.F"; - static constexpr std::string_view TRUNC_S = "TRUNC.S"; - static constexpr std::string_view TRUNC_U = "TRUNC.U"; - static constexpr std::string_view SHL_S = "SHL.S"; - static constexpr std::string_view SHL_U = "SHL.U"; - static constexpr std::string_view SHR_S = "SHR.S"; - static constexpr std::string_view SHR_U = "SHR.U"; - static constexpr std::string_view OR_S = "OR.S"; - static constexpr std::string_view OR_U = "OR.U"; - static constexpr std::string_view XOR_S = "XOR.S"; - static constexpr std::string_view XOR_U = "XOR.U"; - static constexpr std::string_view NOT_S = "NOT.S"; - static constexpr std::string_view NOT_U = "NOT.U"; - static constexpr std::string_view BTC_S = "BTC.S"; - static constexpr std::string_view BTC_U = "BTC.U"; - static constexpr std::string_view BTFM_S = "BTFM.S"; - static constexpr std::string_view BTFM_U = "BTFM.U"; - static constexpr std::string_view ROUND_F = "ROUND.F"; - static constexpr std::string_view CEIL_F = "CEIL.F"; - static constexpr std::string_view FLR_F = "FLR.F"; - static constexpr std::string_view I2F_S = "I2F.S"; - static constexpr std::string_view I2F_U = "I2F.U"; - static constexpr std::string_view MIN_F = "MIN.F"; - static constexpr std::string_view MIN_S = "MIN.S"; - static constexpr std::string_view MIN_U = "MIN.U"; - static constexpr std::string_view MAX_F = "MAX.F"; - static constexpr std::string_view MAX_S = "MAX.S"; - static constexpr std::string_view MAX_U = "MAX.U"; - static constexpr std::string_view MOV_U = "MOV.U"; - static constexpr std::string_view TGBALLOT_U = "TGBALLOT.U"; - static constexpr std::string_view TGALL_U = "TGALL.U"; - static constexpr std::string_view TGANY_U = "TGANY.U"; - static constexpr std::string_view TGEQ_U = "TGEQ.U"; - static constexpr std::string_view EXCH = "EXCH"; - static constexpr std::string_view ADD = "ADD"; - static constexpr std::string_view MIN = "MIN"; - static constexpr std::string_view MAX = "MAX"; - static constexpr std::string_view AND = "AND"; - static constexpr std::string_view OR = "OR"; - static constexpr std::string_view XOR = "XOR"; - static constexpr std::string_view U32 = "U32"; - static constexpr std::string_view S32 = "S32"; - - static constexpr std::size_t NUM_ENTRIES = static_cast<std::size_t>(OperationCode::Amount); - using DecompilerType = std::string (ARBDecompiler::*)(Operation); - static constexpr std::array<DecompilerType, NUM_ENTRIES> OPERATION_DECOMPILERS = { - &ARBDecompiler::Assign, - - &ARBDecompiler::Select, - - &ARBDecompiler::Binary<ADD_F32>, - &ARBDecompiler::Binary<MUL_F32>, - &ARBDecompiler::Binary<DIV_F32>, - &ARBDecompiler::Trinary<MAD_F32>, - &ARBDecompiler::Negate<'F'>, - &ARBDecompiler::Absolute<'F'>, - &ARBDecompiler::FClamp, - &ARBDecompiler::FCastHalf0, - &ARBDecompiler::FCastHalf1, - &ARBDecompiler::Binary<MIN_F>, - &ARBDecompiler::Binary<MAX_F>, - &ARBDecompiler::Unary<COS_F32>, - &ARBDecompiler::Unary<SIN_F32>, - &ARBDecompiler::Unary<EX2_F32>, - &ARBDecompiler::Unary<LG2_F32>, - &ARBDecompiler::Unary<RSQ_F32>, - &ARBDecompiler::FSqrt, - &ARBDecompiler::Unary<ROUND_F>, - &ARBDecompiler::Unary<FLR_F>, - &ARBDecompiler::Unary<CEIL_F>, - &ARBDecompiler::Unary<TRUNC_F>, - &ARBDecompiler::Unary<I2F_S>, - &ARBDecompiler::Unary<I2F_U>, - &ARBDecompiler::FSwizzleAdd, - - &ARBDecompiler::Binary<ADD_S>, - &ARBDecompiler::Binary<MUL_S>, - &ARBDecompiler::Binary<DIV_S>, - &ARBDecompiler::Negate<'S'>, - &ARBDecompiler::Absolute<'S'>, - &ARBDecompiler::Binary<MIN_S>, - &ARBDecompiler::Binary<MAX_S>, - - &ARBDecompiler::Unary<TRUNC_S>, - &ARBDecompiler::Unary<MOV_U>, - &ARBDecompiler::Binary<SHL_S>, - &ARBDecompiler::Binary<SHR_U>, - &ARBDecompiler::Binary<SHR_S>, - &ARBDecompiler::Binary<AND_S>, - &ARBDecompiler::Binary<OR_S>, - &ARBDecompiler::Binary<XOR_S>, - &ARBDecompiler::Unary<NOT_S>, - &ARBDecompiler::BitfieldInsert<'S'>, - &ARBDecompiler::BitfieldExtract<'S'>, - &ARBDecompiler::Unary<BTC_S>, - &ARBDecompiler::Unary<BTFM_S>, - - &ARBDecompiler::Binary<ADD_U>, - &ARBDecompiler::Binary<MUL_U>, - &ARBDecompiler::Binary<DIV_U>, - &ARBDecompiler::Binary<MIN_U>, - &ARBDecompiler::Binary<MAX_U>, - &ARBDecompiler::Unary<TRUNC_U>, - &ARBDecompiler::Unary<MOV_U>, - &ARBDecompiler::Binary<SHL_U>, - &ARBDecompiler::Binary<SHR_U>, - &ARBDecompiler::Binary<SHR_U>, - &ARBDecompiler::Binary<AND_U>, - &ARBDecompiler::Binary<OR_U>, - &ARBDecompiler::Binary<XOR_U>, - &ARBDecompiler::Unary<NOT_U>, - &ARBDecompiler::BitfieldInsert<'U'>, - &ARBDecompiler::BitfieldExtract<'U'>, - &ARBDecompiler::Unary<BTC_U>, - &ARBDecompiler::Unary<BTFM_U>, - - &ARBDecompiler::HAdd2, - &ARBDecompiler::HMul2, - &ARBDecompiler::HFma2, - &ARBDecompiler::HAbsolute, - &ARBDecompiler::HNegate, - &ARBDecompiler::HClamp, - &ARBDecompiler::HCastFloat, - &ARBDecompiler::HUnpack, - &ARBDecompiler::HMergeF32, - &ARBDecompiler::HMergeH0, - &ARBDecompiler::HMergeH1, - &ARBDecompiler::HPack2, - - &ARBDecompiler::LogicalAssign, - &ARBDecompiler::Binary<AND_U>, - &ARBDecompiler::Binary<OR_U>, - &ARBDecompiler::Binary<XOR_U>, - &ARBDecompiler::Unary<NOT_U>, - &ARBDecompiler::LogicalPick2, - &ARBDecompiler::LogicalAnd2, - - &ARBDecompiler::FloatComparison<SLT_F, false>, - &ARBDecompiler::FloatComparison<SEQ_F, false>, - &ARBDecompiler::FloatComparison<SLE_F, false>, - &ARBDecompiler::FloatComparison<SGT_F, false>, - &ARBDecompiler::FloatComparison<SNE_F, false>, - &ARBDecompiler::FloatComparison<SGE_F, false>, - &ARBDecompiler::FloatOrdered, - &ARBDecompiler::FloatUnordered, - &ARBDecompiler::FloatComparison<SLT_F, true>, - &ARBDecompiler::FloatComparison<SEQ_F, true>, - &ARBDecompiler::FloatComparison<SLE_F, true>, - &ARBDecompiler::FloatComparison<SGT_F, true>, - &ARBDecompiler::FloatComparison<SNE_F, true>, - &ARBDecompiler::FloatComparison<SGE_F, true>, - - &ARBDecompiler::Binary<SLT_S>, - &ARBDecompiler::Binary<SEQ_S>, - &ARBDecompiler::Binary<SLE_S>, - &ARBDecompiler::Binary<SGT_S>, - &ARBDecompiler::Binary<SNE_S>, - &ARBDecompiler::Binary<SGE_S>, - - &ARBDecompiler::Binary<SLT_U>, - &ARBDecompiler::Binary<SEQ_U>, - &ARBDecompiler::Binary<SLE_U>, - &ARBDecompiler::Binary<SGT_U>, - &ARBDecompiler::Binary<SNE_U>, - &ARBDecompiler::Binary<SGE_U>, - - &ARBDecompiler::LogicalAddCarry, - - &ARBDecompiler::HalfComparison<SLT_F, false>, - &ARBDecompiler::HalfComparison<SEQ_F, false>, - &ARBDecompiler::HalfComparison<SLE_F, false>, - &ARBDecompiler::HalfComparison<SGT_F, false>, - &ARBDecompiler::HalfComparison<SNE_F, false>, - &ARBDecompiler::HalfComparison<SGE_F, false>, - &ARBDecompiler::HalfComparison<SLT_F, true>, - &ARBDecompiler::HalfComparison<SEQ_F, true>, - &ARBDecompiler::HalfComparison<SLE_F, true>, - &ARBDecompiler::HalfComparison<SGT_F, true>, - &ARBDecompiler::HalfComparison<SNE_F, true>, - &ARBDecompiler::HalfComparison<SGE_F, true>, - - &ARBDecompiler::Texture, - &ARBDecompiler::Texture, - &ARBDecompiler::TextureGather, - &ARBDecompiler::TextureQueryDimensions, - &ARBDecompiler::TextureQueryLod, - &ARBDecompiler::TexelFetch, - &ARBDecompiler::TextureGradient, - - &ARBDecompiler::ImageLoad, - &ARBDecompiler::ImageStore, - - &ARBDecompiler::AtomicImage<ADD, U32>, - &ARBDecompiler::AtomicImage<AND, U32>, - &ARBDecompiler::AtomicImage<OR, U32>, - &ARBDecompiler::AtomicImage<XOR, U32>, - &ARBDecompiler::AtomicImage<EXCH, U32>, - - &ARBDecompiler::Atomic<EXCH, U32>, - &ARBDecompiler::Atomic<ADD, U32>, - &ARBDecompiler::Atomic<MIN, U32>, - &ARBDecompiler::Atomic<MAX, U32>, - &ARBDecompiler::Atomic<AND, U32>, - &ARBDecompiler::Atomic<OR, U32>, - &ARBDecompiler::Atomic<XOR, U32>, - - &ARBDecompiler::Atomic<EXCH, S32>, - &ARBDecompiler::Atomic<ADD, S32>, - &ARBDecompiler::Atomic<MIN, S32>, - &ARBDecompiler::Atomic<MAX, S32>, - &ARBDecompiler::Atomic<AND, S32>, - &ARBDecompiler::Atomic<OR, S32>, - &ARBDecompiler::Atomic<XOR, S32>, - - &ARBDecompiler::Atomic<ADD, U32>, - &ARBDecompiler::Atomic<MIN, U32>, - &ARBDecompiler::Atomic<MAX, U32>, - &ARBDecompiler::Atomic<AND, U32>, - &ARBDecompiler::Atomic<OR, U32>, - &ARBDecompiler::Atomic<XOR, U32>, - - &ARBDecompiler::Atomic<ADD, S32>, - &ARBDecompiler::Atomic<MIN, S32>, - &ARBDecompiler::Atomic<MAX, S32>, - &ARBDecompiler::Atomic<AND, S32>, - &ARBDecompiler::Atomic<OR, S32>, - &ARBDecompiler::Atomic<XOR, S32>, - - &ARBDecompiler::Branch, - &ARBDecompiler::BranchIndirect, - &ARBDecompiler::PushFlowStack, - &ARBDecompiler::PopFlowStack, - &ARBDecompiler::Exit, - &ARBDecompiler::Discard, - - &ARBDecompiler::EmitVertex, - &ARBDecompiler::EndPrimitive, - - &ARBDecompiler::InvocationId, - &ARBDecompiler::YNegate, - &ARBDecompiler::LocalInvocationId<'x'>, - &ARBDecompiler::LocalInvocationId<'y'>, - &ARBDecompiler::LocalInvocationId<'z'>, - &ARBDecompiler::WorkGroupId<'x'>, - &ARBDecompiler::WorkGroupId<'y'>, - &ARBDecompiler::WorkGroupId<'z'>, - - &ARBDecompiler::Unary<TGBALLOT_U>, - &ARBDecompiler::Unary<TGALL_U>, - &ARBDecompiler::Unary<TGANY_U>, - &ARBDecompiler::Unary<TGEQ_U>, - - &ARBDecompiler::ThreadId, - &ARBDecompiler::ThreadMask<'e', 'q'>, - &ARBDecompiler::ThreadMask<'g', 'e'>, - &ARBDecompiler::ThreadMask<'g', 't'>, - &ARBDecompiler::ThreadMask<'l', 'e'>, - &ARBDecompiler::ThreadMask<'l', 't'>, - &ARBDecompiler::ShuffleIndexed, - - &ARBDecompiler::Barrier, - &ARBDecompiler::MemoryBarrierGroup, - &ARBDecompiler::MemoryBarrierGlobal, - }; -}; - -ARBDecompiler::ARBDecompiler(const Device& device_, const ShaderIR& ir_, const Registry& registry_, - ShaderType stage_, std::string_view identifier) - : device{device_}, ir{ir_}, registry{registry_}, stage{stage_} { - DefineGlobalMemory(); - - AddLine("TEMP RC;"); - AddLine("TEMP FSWZA[4];"); - AddLine("TEMP FSWZB[4];"); - if (ir.IsDecompiled()) { - DecompileAST(); - } else { - DecompileBranchMode(); - } - AddLine("END"); - - const std::string code = std::move(shader_source); - DeclareHeader(); - DeclareVertex(); - DeclareGeometry(); - DeclareFragment(); - DeclareCompute(); - DeclareInputAttributes(); - DeclareOutputAttributes(); - DeclareLocalMemory(); - DeclareGlobalMemory(); - DeclareConstantBuffers(); - DeclareRegisters(); - DeclareTemporaries(); - DeclarePredicates(); - DeclareInternalFlags(); - - shader_source += code; -} - -std::string_view HeaderStageName(ShaderType stage) { - switch (stage) { - case ShaderType::Vertex: - return "vp"; - case ShaderType::Geometry: - return "gp"; - case ShaderType::Fragment: - return "fp"; - case ShaderType::Compute: - return "cp"; - default: - UNREACHABLE(); - return ""; - } -} - -void ARBDecompiler::DefineGlobalMemory() { - u32 binding = 0; - for (const auto& pair : ir.GetGlobalMemory()) { - const GlobalMemoryBase base = pair.first; - global_memory_names.emplace(base, binding); - ++binding; - } -} - -void ARBDecompiler::DeclareHeader() { - AddLine("!!NV{}5.0", HeaderStageName(stage)); - // Enabling this allows us to cheat on some instructions like TXL with SHADOWARRAY2D - AddLine("OPTION NV_internal;"); - AddLine("OPTION NV_gpu_program_fp64;"); - AddLine("OPTION NV_shader_thread_group;"); - if (ir.UsesWarps() && device.HasWarpIntrinsics()) { - AddLine("OPTION NV_shader_thread_shuffle;"); - } - if (stage == ShaderType::Vertex) { - if (device.HasNvViewportArray2()) { - AddLine("OPTION NV_viewport_array2;"); - } - } - if (stage == ShaderType::Fragment) { - AddLine("OPTION ARB_draw_buffers;"); - } - if (device.HasImageLoadFormatted()) { - AddLine("OPTION EXT_shader_image_load_formatted;"); - } -} - -void ARBDecompiler::DeclareVertex() { - if (stage != ShaderType::Vertex) { - return; - } - AddLine("OUTPUT result_clip[] = {{ result.clip[0..7] }};"); -} - -void ARBDecompiler::DeclareGeometry() { - if (stage != ShaderType::Geometry) { - return; - } - const auto& info = registry.GetGraphicsInfo(); - const auto& header = ir.GetHeader(); - AddLine("PRIMITIVE_IN {};", PrimitiveDescription(info.primitive_topology)); - AddLine("PRIMITIVE_OUT {};", TopologyName(header.common3.output_topology)); - AddLine("VERTICES_OUT {};", header.common4.max_output_vertices.Value()); - AddLine("ATTRIB vertex_position = vertex.position;"); -} - -void ARBDecompiler::DeclareFragment() { - if (stage != ShaderType::Fragment) { - return; - } - AddLine("OUTPUT result_color7 = result.color[7];"); - AddLine("OUTPUT result_color6 = result.color[6];"); - AddLine("OUTPUT result_color5 = result.color[5];"); - AddLine("OUTPUT result_color4 = result.color[4];"); - AddLine("OUTPUT result_color3 = result.color[3];"); - AddLine("OUTPUT result_color2 = result.color[2];"); - AddLine("OUTPUT result_color1 = result.color[1];"); - AddLine("OUTPUT result_color0 = result.color;"); -} - -void ARBDecompiler::DeclareCompute() { - if (stage != ShaderType::Compute) { - return; - } - const ComputeInfo& info = registry.GetComputeInfo(); - AddLine("GROUP_SIZE {} {} {};", info.workgroup_size[0], info.workgroup_size[1], - info.workgroup_size[2]); - if (info.shared_memory_size_in_words == 0) { - return; - } - const u32 limit = device.GetMaxComputeSharedMemorySize(); - u32 size_in_bytes = info.shared_memory_size_in_words * 4; - if (size_in_bytes > limit) { - LOG_ERROR(Render_OpenGL, "Shared memory size {} is clamped to host's limit {}", - size_in_bytes, limit); - size_in_bytes = limit; - } - - AddLine("SHARED_MEMORY {};", size_in_bytes); - AddLine("SHARED shared_mem[] = {{program.sharedmem}};"); -} - -void ARBDecompiler::DeclareInputAttributes() { - if (stage == ShaderType::Compute) { - return; - } - const std::string_view stage_name = StageInputName(stage); - for (const auto attribute : ir.GetInputAttributes()) { - if (!IsGenericAttribute(attribute)) { - continue; - } - const u32 index = GetGenericAttributeIndex(attribute); - - std::string_view suffix; - if (stage == ShaderType::Fragment) { - const auto input_mode{ir.GetHeader().ps.GetPixelImap(index)}; - if (input_mode == PixelImap::Unused) { - return; - } - suffix = GetInputFlags(input_mode); - } - AddLine("{}ATTRIB in_attr{}[] = {{ {}.attrib[{}..{}] }};", suffix, index, stage_name, index, - index); - } -} - -void ARBDecompiler::DeclareOutputAttributes() { - if (stage == ShaderType::Compute) { - return; - } - for (const auto attribute : ir.GetOutputAttributes()) { - if (!IsGenericAttribute(attribute)) { - continue; - } - const u32 index = GetGenericAttributeIndex(attribute); - AddLine("OUTPUT out_attr{}[] = {{ result.attrib[{}..{}] }};", index, index, index); - } -} - -void ARBDecompiler::DeclareLocalMemory() { - u64 size = 0; - if (stage == ShaderType::Compute) { - size = registry.GetComputeInfo().local_memory_size_in_words * 4ULL; - } else { - size = ir.GetHeader().GetLocalMemorySize(); - } - if (size == 0) { - return; - } - const u64 element_count = Common::AlignUp(size, 4) / 4; - AddLine("TEMP lmem[{}];", element_count); -} - -void ARBDecompiler::DeclareGlobalMemory() { - const size_t num_entries = ir.GetGlobalMemory().size(); - if (num_entries > 0) { - AddLine("PARAM c[{}] = {{ program.local[0..{}] }};", num_entries, num_entries - 1); - } -} - -void ARBDecompiler::DeclareConstantBuffers() { - u32 binding = 0; - for (const auto& cbuf : ir.GetConstantBuffers()) { - AddLine("CBUFFER cbuf{}[] = {{ program.buffer[{}] }};", cbuf.first, binding); - ++binding; - } -} - -void ARBDecompiler::DeclareRegisters() { - for (const u32 gpr : ir.GetRegisters()) { - AddLine("TEMP R{};", gpr); - } -} - -void ARBDecompiler::DeclareTemporaries() { - for (std::size_t i = 0; i < max_temporaries; ++i) { - AddLine("TEMP T{};", i); - } - for (std::size_t i = 0; i < max_long_temporaries; ++i) { - AddLine("LONG TEMP L{};", i); - } -} - -void ARBDecompiler::DeclarePredicates() { - for (const Tegra::Shader::Pred pred : ir.GetPredicates()) { - AddLine("TEMP P{};", static_cast<u64>(pred)); - } -} - -void ARBDecompiler::DeclareInternalFlags() { - for (const char* name : INTERNAL_FLAG_NAMES) { - AddLine("TEMP {};", name); - } -} - -void ARBDecompiler::InitializeVariables() { - AddLine("MOV.F32 FSWZA[0], -1;"); - AddLine("MOV.F32 FSWZA[1], 1;"); - AddLine("MOV.F32 FSWZA[2], -1;"); - AddLine("MOV.F32 FSWZA[3], 0;"); - AddLine("MOV.F32 FSWZB[0], -1;"); - AddLine("MOV.F32 FSWZB[1], -1;"); - AddLine("MOV.F32 FSWZB[2], 1;"); - AddLine("MOV.F32 FSWZB[3], -1;"); - - if (stage == ShaderType::Vertex || stage == ShaderType::Geometry) { - AddLine("MOV.F result.position, {{0, 0, 0, 1}};"); - } - for (const auto attribute : ir.GetOutputAttributes()) { - if (!IsGenericAttribute(attribute)) { - continue; - } - const u32 index = GetGenericAttributeIndex(attribute); - AddLine("MOV.F result.attrib[{}], {{0, 0, 0, 1}};", index); - } - for (const u32 gpr : ir.GetRegisters()) { - AddLine("MOV.F R{}, {{0, 0, 0, 0}};", gpr); - } - for (const Tegra::Shader::Pred pred : ir.GetPredicates()) { - AddLine("MOV.U P{}, {{0, 0, 0, 0}};", static_cast<u64>(pred)); - } -} - -void ARBDecompiler::DecompileAST() { - const u32 num_flow_variables = ir.GetASTNumVariables(); - for (u32 i = 0; i < num_flow_variables; ++i) { - AddLine("TEMP F{};", i); - } - for (u32 i = 0; i < num_flow_variables; ++i) { - AddLine("MOV.U F{}, {{0, 0, 0, 0}};", i); - } - - InitializeVariables(); - - VisitAST(ir.GetASTProgram()); -} - -void ARBDecompiler::DecompileBranchMode() { - static constexpr u32 FLOW_STACK_SIZE = 20; - if (!ir.IsFlowStackDisabled()) { - AddLine("TEMP SSY[{}];", FLOW_STACK_SIZE); - AddLine("TEMP PBK[{}];", FLOW_STACK_SIZE); - AddLine("TEMP SSY_TOP;"); - AddLine("TEMP PBK_TOP;"); - } - - AddLine("TEMP PC;"); - - if (!ir.IsFlowStackDisabled()) { - AddLine("MOV.U SSY_TOP.x, 0;"); - AddLine("MOV.U PBK_TOP.x, 0;"); - } - - InitializeVariables(); - - const auto basic_block_end = ir.GetBasicBlocks().end(); - auto basic_block_it = ir.GetBasicBlocks().begin(); - const u32 first_address = basic_block_it->first; - AddLine("MOV.U PC.x, {};", first_address); - - AddLine("REP;"); - - std::size_t num_blocks = 0; - while (basic_block_it != basic_block_end) { - const auto& [address, bb] = *basic_block_it; - ++num_blocks; - - AddLine("SEQ.S.CC RC.x, PC.x, {};", address); - AddLine("IF NE.x;"); - - VisitBlock(bb); - - ++basic_block_it; - - if (basic_block_it != basic_block_end) { - const auto op = std::get_if<OperationNode>(&*bb[bb.size() - 1]); - if (!op || op->GetCode() != OperationCode::Branch) { - const u32 next_address = basic_block_it->first; - AddLine("MOV.U PC.x, {};", next_address); - AddLine("CONT;"); - } - } - - AddLine("ELSE;"); - } - AddLine("RET;"); - while (num_blocks--) { - AddLine("ENDIF;"); - } - - AddLine("ENDREP;"); -} - -void ARBDecompiler::VisitAST(const ASTNode& node) { - if (const auto ast = std::get_if<ASTProgram>(&*node->GetInnerData())) { - for (ASTNode current = ast->nodes.GetFirst(); current; current = current->GetNext()) { - VisitAST(current); - } - } else if (const auto if_then = std::get_if<ASTIfThen>(&*node->GetInnerData())) { - const std::string condition = VisitExpression(if_then->condition); - ResetTemporaries(); - - AddLine("MOVC.U RC.x, {};", condition); - AddLine("IF NE.x;"); - for (ASTNode current = if_then->nodes.GetFirst(); current; current = current->GetNext()) { - VisitAST(current); - } - AddLine("ENDIF;"); - } else if (const auto if_else = std::get_if<ASTIfElse>(&*node->GetInnerData())) { - AddLine("ELSE;"); - for (ASTNode current = if_else->nodes.GetFirst(); current; current = current->GetNext()) { - VisitAST(current); - } - } else if (const auto decoded = std::get_if<ASTBlockDecoded>(&*node->GetInnerData())) { - VisitBlock(decoded->nodes); - } else if (const auto var_set = std::get_if<ASTVarSet>(&*node->GetInnerData())) { - AddLine("MOV.U F{}, {};", var_set->index, VisitExpression(var_set->condition)); - ResetTemporaries(); - } else if (const auto do_while = std::get_if<ASTDoWhile>(&*node->GetInnerData())) { - const std::string condition = VisitExpression(do_while->condition); - ResetTemporaries(); - AddLine("REP;"); - for (ASTNode current = do_while->nodes.GetFirst(); current; current = current->GetNext()) { - VisitAST(current); - } - AddLine("MOVC.U RC.x, {};", condition); - AddLine("BRK (NE.x);"); - AddLine("ENDREP;"); - } else if (const auto ast_return = std::get_if<ASTReturn>(&*node->GetInnerData())) { - const bool is_true = ExprIsTrue(ast_return->condition); - if (!is_true) { - AddLine("MOVC.U RC.x, {};", VisitExpression(ast_return->condition)); - AddLine("IF NE.x;"); - ResetTemporaries(); - } - if (ast_return->kills) { - AddLine("KIL TR;"); - } else { - Exit(); - } - if (!is_true) { - AddLine("ENDIF;"); - } - } else if (const auto ast_break = std::get_if<ASTBreak>(&*node->GetInnerData())) { - if (ExprIsTrue(ast_break->condition)) { - AddLine("BRK;"); - } else { - AddLine("MOVC.U RC.x, {};", VisitExpression(ast_break->condition)); - AddLine("BRK (NE.x);"); - ResetTemporaries(); - } - } else if (std::holds_alternative<ASTLabel>(*node->GetInnerData())) { - // Nothing to do - } else { - UNREACHABLE(); - } -} - -std::string ARBDecompiler::VisitExpression(const Expr& node) { - if (const auto expr = std::get_if<ExprAnd>(&*node)) { - std::string result = AllocTemporary(); - AddLine("AND.U {}, {}, {};", result, VisitExpression(expr->operand1), - VisitExpression(expr->operand2)); - return result; - } - if (const auto expr = std::get_if<ExprOr>(&*node)) { - std::string result = AllocTemporary(); - AddLine("OR.U {}, {}, {};", result, VisitExpression(expr->operand1), - VisitExpression(expr->operand2)); - return result; - } - if (const auto expr = std::get_if<ExprNot>(&*node)) { - std::string result = AllocTemporary(); - AddLine("CMP.S {}, {}, 0, -1;", result, VisitExpression(expr->operand1)); - return result; - } - if (const auto expr = std::get_if<ExprPredicate>(&*node)) { - return fmt::format("P{}.x", static_cast<u64>(expr->predicate)); - } - if (const auto expr = std::get_if<ExprCondCode>(&*node)) { - return Visit(ir.GetConditionCode(expr->cc)); - } - if (const auto expr = std::get_if<ExprVar>(&*node)) { - return fmt::format("F{}.x", expr->var_index); - } - if (const auto expr = std::get_if<ExprBoolean>(&*node)) { - return expr->value ? "0xffffffff" : "0"; - } - if (const auto expr = std::get_if<ExprGprEqual>(&*node)) { - std::string result = AllocTemporary(); - AddLine("SEQ.U {}, R{}.x, {};", result, expr->gpr, expr->value); - return result; - } - UNREACHABLE(); - return "0"; -} - -void ARBDecompiler::VisitBlock(const NodeBlock& bb) { - for (const auto& node : bb) { - Visit(node); - } -} - -std::string ARBDecompiler::Visit(const Node& node) { - if (const auto operation = std::get_if<OperationNode>(&*node)) { - if (const auto amend_index = operation->GetAmendIndex()) { - Visit(ir.GetAmendNode(*amend_index)); - } - const std::size_t index = static_cast<std::size_t>(operation->GetCode()); - if (index >= OPERATION_DECOMPILERS.size()) { - UNREACHABLE_MSG("Out of bounds operation: {}", index); - return {}; - } - const auto decompiler = OPERATION_DECOMPILERS[index]; - if (decompiler == nullptr) { - UNREACHABLE_MSG("Undefined operation: {}", index); - return {}; - } - return (this->*decompiler)(*operation); - } - - if (const auto gpr = std::get_if<GprNode>(&*node)) { - const u32 index = gpr->GetIndex(); - if (index == Register::ZeroIndex) { - return "{0, 0, 0, 0}.x"; - } - return fmt::format("R{}.x", index); - } - - if (const auto cv = std::get_if<CustomVarNode>(&*node)) { - return fmt::format("CV{}.x", cv->GetIndex()); - } - - if (const auto immediate = std::get_if<ImmediateNode>(&*node)) { - std::string temporary = AllocTemporary(); - AddLine("MOV.U {}, {};", temporary, immediate->GetValue()); - return temporary; - } - - if (const auto predicate = std::get_if<PredicateNode>(&*node)) { - std::string temporary = AllocTemporary(); - switch (const auto index = predicate->GetIndex(); index) { - case Tegra::Shader::Pred::UnusedIndex: - AddLine("MOV.S {}, -1;", temporary); - break; - case Tegra::Shader::Pred::NeverExecute: - AddLine("MOV.S {}, 0;", temporary); - break; - default: - AddLine("MOV.S {}, P{}.x;", temporary, static_cast<u64>(index)); - break; - } - if (predicate->IsNegated()) { - AddLine("CMP.S {}, {}, 0, -1;", temporary, temporary); - } - return temporary; - } - - if (const auto abuf = std::get_if<AbufNode>(&*node)) { - if (abuf->IsPhysicalBuffer()) { - UNIMPLEMENTED_MSG("Physical buffers are not implemented"); - return "{0, 0, 0, 0}.x"; - } - - const Attribute::Index index = abuf->GetIndex(); - const u32 element = abuf->GetElement(); - const char swizzle = Swizzle(element); - switch (index) { - case Attribute::Index::Position: { - if (stage == ShaderType::Geometry) { - return fmt::format("{}_position[{}].{}", StageInputName(stage), - Visit(abuf->GetBuffer()), swizzle); - } else { - return fmt::format("{}.position.{}", StageInputName(stage), swizzle); - } - } - case Attribute::Index::TessCoordInstanceIDVertexID: - ASSERT(stage == ShaderType::Vertex); - switch (element) { - case 2: - return "vertex.instance"; - case 3: - return "vertex.id"; - } - UNIMPLEMENTED_MSG("Unmanaged TessCoordInstanceIDVertexID element={}", element); - break; - case Attribute::Index::PointCoord: - switch (element) { - case 0: - return "fragment.pointcoord.x"; - case 1: - return "fragment.pointcoord.y"; - } - UNIMPLEMENTED(); - break; - case Attribute::Index::FrontFacing: { - ASSERT(stage == ShaderType::Fragment); - ASSERT(element == 3); - const std::string temporary = AllocVectorTemporary(); - AddLine("SGT.S RC.x, fragment.facing, {{0, 0, 0, 0}};"); - AddLine("MOV.U.CC RC.x, -RC;"); - AddLine("MOV.S {}.x, 0;", temporary); - AddLine("MOV.S {}.x (NE.x), -1;", temporary); - return fmt::format("{}.x", temporary); - } - default: - if (IsGenericAttribute(index)) { - if (stage == ShaderType::Geometry) { - return fmt::format("in_attr{}[{}][0].{}", GetGenericAttributeIndex(index), - Visit(abuf->GetBuffer()), swizzle); - } else { - return fmt::format("{}.attrib[{}].{}", StageInputName(stage), - GetGenericAttributeIndex(index), swizzle); - } - } - UNIMPLEMENTED_MSG("Unimplemented input attribute={}", index); - break; - } - return "{0, 0, 0, 0}.x"; - } - - if (const auto cbuf = std::get_if<CbufNode>(&*node)) { - std::string offset_string; - const auto& offset = cbuf->GetOffset(); - if (const auto imm = std::get_if<ImmediateNode>(&*offset)) { - offset_string = std::to_string(imm->GetValue()); - } else { - offset_string = Visit(offset); - } - std::string temporary = AllocTemporary(); - AddLine("LDC.F32 {}, cbuf{}[{}];", temporary, cbuf->GetIndex(), offset_string); - return temporary; - } - - if (const auto gmem = std::get_if<GmemNode>(&*node)) { - std::string temporary = AllocTemporary(); - AddLine("MOV {}, 0;", temporary); - AddLine("LOAD.U32 {} (NE.x), {};", temporary, GlobalMemoryPointer(*gmem)); - return temporary; - } - - if (const auto lmem = std::get_if<LmemNode>(&*node)) { - std::string temporary = Visit(lmem->GetAddress()); - AddLine("SHR.U {}, {}, 2;", temporary, temporary); - AddLine("MOV.U {}, lmem[{}].x;", temporary, temporary); - return temporary; - } - - if (const auto smem = std::get_if<SmemNode>(&*node)) { - std::string temporary = Visit(smem->GetAddress()); - AddLine("LDS.U32 {}, shared_mem[{}];", temporary, temporary); - return temporary; - } - - if (const auto internal_flag = std::get_if<InternalFlagNode>(&*node)) { - const std::size_t index = static_cast<std::size_t>(internal_flag->GetFlag()); - return fmt::format("{}.x", INTERNAL_FLAG_NAMES[index]); - } - - if (const auto conditional = std::get_if<ConditionalNode>(&*node)) { - if (const auto amend_index = conditional->GetAmendIndex()) { - Visit(ir.GetAmendNode(*amend_index)); - } - AddLine("MOVC.U RC.x, {};", Visit(conditional->GetCondition())); - AddLine("IF NE.x;"); - VisitBlock(conditional->GetCode()); - AddLine("ENDIF;"); - return {}; - } - - if ([[maybe_unused]] const auto cmt = std::get_if<CommentNode>(&*node)) { - // Uncommenting this will generate invalid code. GLASM lacks comments. - // AddLine("// {}", cmt->GetText()); - return {}; - } - - UNIMPLEMENTED(); - return {}; -} - -std::tuple<std::string, std::string, std::size_t> ARBDecompiler::BuildCoords(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - UNIMPLEMENTED_IF(meta.sampler.is_indexed); - - const bool is_extended = meta.sampler.is_shadow && meta.sampler.is_array && - meta.sampler.type == Tegra::Shader::TextureType::TextureCube; - const std::size_t count = operation.GetOperandsCount(); - std::string temporary = AllocVectorTemporary(); - std::size_t i = 0; - for (; i < count; ++i) { - AddLine("MOV.F {}.{}, {};", temporary, Swizzle(i), Visit(operation[i])); - } - if (meta.sampler.is_array) { - AddLine("I2F.S {}.{}, {};", temporary, Swizzle(i), Visit(meta.array)); - ++i; - } - if (meta.sampler.is_shadow) { - std::string compare = Visit(meta.depth_compare); - if (is_extended) { - ASSERT(i == 4); - std::string extra_coord = AllocVectorTemporary(); - AddLine("MOV.F {}.x, {};", extra_coord, compare); - return {fmt::format("{}, {}", temporary, extra_coord), extra_coord, 0}; - } - AddLine("MOV.F {}.{}, {};", temporary, Swizzle(i), compare); - ++i; - } - return {temporary, temporary, i}; -} - -std::string ARBDecompiler::BuildAoffi(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - if (meta.aoffi.empty()) { - return {}; - } - const std::string temporary = AllocVectorTemporary(); - std::size_t i = 0; - for (auto& node : meta.aoffi) { - AddLine("MOV.S {}.{}, {};", temporary, Swizzle(i++), Visit(node)); - } - return fmt::format(", offset({})", temporary); -} - -std::string ARBDecompiler::GlobalMemoryPointer(const GmemNode& gmem) { - // Read a bindless SSBO, return its address and set CC accordingly - // address = c[binding].xy - // length = c[binding].z - const u32 binding = global_memory_names.at(gmem.GetDescriptor()); - - const std::string pointer = AllocLongVectorTemporary(); - std::string temporary = AllocTemporary(); - - AddLine("PK64.U {}, c[{}];", pointer, binding); - AddLine("SUB.U {}, {}, {};", temporary, Visit(gmem.GetRealAddress()), - Visit(gmem.GetBaseAddress())); - AddLine("CVT.U64.U32 {}.z, {};", pointer, temporary); - AddLine("ADD.U64 {}.x, {}.x, {}.z;", pointer, pointer, pointer); - // Compare offset to length and set CC - AddLine("SLT.U.CC RC.x, {}, c[{}].z;", temporary, binding); - return fmt::format("{}.x", pointer); -} - -void ARBDecompiler::Exit() { - if (stage != ShaderType::Fragment) { - AddLine("RET;"); - return; - } - - const auto safe_get_register = [this](u32 reg) -> std::string { - if (ir.GetRegisters().contains(reg)) { - return fmt::format("R{}.x", reg); - } - return "{0, 0, 0, 0}.x"; - }; - - const auto& header = ir.GetHeader(); - u32 current_reg = 0; - for (u32 rt = 0; rt < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; ++rt) { - for (u32 component = 0; component < 4; ++component) { - if (!header.ps.IsColorComponentOutputEnabled(rt, component)) { - continue; - } - AddLine("MOV.F result_color{}.{}, {};", rt, Swizzle(component), - safe_get_register(current_reg)); - ++current_reg; - } - } - if (header.ps.omap.depth) { - AddLine("MOV.F result.depth.z, {};", safe_get_register(current_reg + 1)); - } - - AddLine("RET;"); -} - -std::string ARBDecompiler::Assign(Operation operation) { - const Node& dest = operation[0]; - const Node& src = operation[1]; - - std::string dest_name; - if (const auto gpr = std::get_if<GprNode>(&*dest)) { - if (gpr->GetIndex() == Register::ZeroIndex) { - // Writing to Register::ZeroIndex is a no op - return {}; - } - dest_name = fmt::format("R{}.x", gpr->GetIndex()); - } else if (const auto abuf = std::get_if<AbufNode>(&*dest)) { - const u32 element = abuf->GetElement(); - const char swizzle = Swizzle(element); - switch (const Attribute::Index index = abuf->GetIndex()) { - case Attribute::Index::Position: - dest_name = fmt::format("result.position.{}", swizzle); - break; - case Attribute::Index::LayerViewportPointSize: - switch (element) { - case 0: - UNIMPLEMENTED(); - return {}; - case 1: - case 2: - if (!device.HasNvViewportArray2()) { - LOG_ERROR( - Render_OpenGL, - "NV_viewport_array2 is missing. Maxwell gen 2 or better is required."); - return {}; - } - dest_name = element == 1 ? "result.layer.x" : "result.viewport.x"; - break; - case 3: - dest_name = "result.pointsize.x"; - break; - } - break; - case Attribute::Index::ClipDistances0123: - dest_name = fmt::format("result.clip[{}].x", element); - break; - case Attribute::Index::ClipDistances4567: - dest_name = fmt::format("result.clip[{}].x", element + 4); - break; - default: - if (!IsGenericAttribute(index)) { - UNREACHABLE(); - return {}; - } - dest_name = - fmt::format("result.attrib[{}].{}", GetGenericAttributeIndex(index), swizzle); - break; - } - } else if (const auto lmem = std::get_if<LmemNode>(&*dest)) { - const std::string address = Visit(lmem->GetAddress()); - AddLine("SHR.U {}, {}, 2;", address, address); - dest_name = fmt::format("lmem[{}].x", address); - } else if (const auto smem = std::get_if<SmemNode>(&*dest)) { - AddLine("STS.U32 {}, shared_mem[{}];", Visit(src), Visit(smem->GetAddress())); - ResetTemporaries(); - return {}; - } else if (const auto gmem = std::get_if<GmemNode>(&*dest)) { - AddLine("IF NE.x;"); - AddLine("STORE.U32 {}, {};", Visit(src), GlobalMemoryPointer(*gmem)); - AddLine("ENDIF;"); - ResetTemporaries(); - return {}; - } else { - UNREACHABLE(); - ResetTemporaries(); - return {}; - } - - AddLine("MOV.U {}, {};", dest_name, Visit(src)); - ResetTemporaries(); - return {}; -} - -std::string ARBDecompiler::Select(Operation operation) { - std::string temporary = AllocTemporary(); - AddLine("CMP.S {}, {}, {}, {};", temporary, Visit(operation[0]), Visit(operation[1]), - Visit(operation[2])); - return temporary; -} - -std::string ARBDecompiler::FClamp(Operation operation) { - // 1.0f in hex, replace with std::bit_cast on C++20 - static constexpr u32 POSITIVE_ONE = 0x3f800000; - - std::string temporary = AllocTemporary(); - const Node& value = operation[0]; - const Node& low = operation[1]; - const Node& high = operation[2]; - const auto* const imm_low = std::get_if<ImmediateNode>(&*low); - const auto* const imm_high = std::get_if<ImmediateNode>(&*high); - if (imm_low && imm_high && imm_low->GetValue() == 0 && imm_high->GetValue() == POSITIVE_ONE) { - AddLine("MOV.F32.SAT {}, {};", temporary, Visit(value)); - } else { - AddLine("MIN.F {}, {}, {};", temporary, Visit(value), Visit(high)); - AddLine("MAX.F {}, {}, {};", temporary, temporary, Visit(low)); - } - return temporary; -} - -std::string ARBDecompiler::FCastHalf0(Operation operation) { - const std::string temporary = AllocVectorTemporary(); - AddLine("UP2H.F {}.x, {};", temporary, Visit(operation[0])); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::FCastHalf1(Operation operation) { - const std::string temporary = AllocVectorTemporary(); - AddLine("UP2H.F {}.y, {};", temporary, Visit(operation[0])); - AddLine("MOV {}.x, {}.y;", temporary, temporary); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::FSqrt(Operation operation) { - std::string temporary = AllocTemporary(); - AddLine("RSQ.F32 {}, {};", temporary, Visit(operation[0])); - AddLine("RCP.F32 {}, {};", temporary, temporary); - return temporary; -} - -std::string ARBDecompiler::FSwizzleAdd(Operation operation) { - const std::string temporary = AllocVectorTemporary(); - if (!device.HasWarpIntrinsics()) { - LOG_ERROR(Render_OpenGL, - "NV_shader_thread_shuffle is missing. Kepler or better is required."); - AddLine("ADD.F {}.x, {}, {};", temporary, Visit(operation[0]), Visit(operation[1])); - return fmt::format("{}.x", temporary); - } - - AddLine("AND.U {}.z, {}.threadid, 3;", temporary, StageInputName(stage)); - AddLine("SHL.U {}.z, {}.z, 1;", temporary, temporary); - AddLine("SHR.U {}.z, {}, {}.z;", temporary, Visit(operation[2]), temporary); - AddLine("AND.U {}.z, {}.z, 3;", temporary, temporary); - AddLine("MUL.F32 {}.x, {}, FSWZA[{}.z];", temporary, Visit(operation[0]), temporary); - AddLine("MUL.F32 {}.y, {}, FSWZB[{}.z];", temporary, Visit(operation[1]), temporary); - AddLine("ADD.F32 {}.x, {}.x, {}.y;", temporary, temporary, temporary); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::HAdd2(Operation operation) { - const std::string tmp1 = AllocVectorTemporary(); - const std::string tmp2 = AllocVectorTemporary(); - AddLine("UP2H.F {}.xy, {};", tmp1, Visit(operation[0])); - AddLine("UP2H.F {}.xy, {};", tmp2, Visit(operation[1])); - AddLine("ADD.F16 {}, {}, {};", tmp1, tmp1, tmp2); - AddLine("PK2H.F {}.x, {};", tmp1, tmp1); - return fmt::format("{}.x", tmp1); -} - -std::string ARBDecompiler::HMul2(Operation operation) { - const std::string tmp1 = AllocVectorTemporary(); - const std::string tmp2 = AllocVectorTemporary(); - AddLine("UP2H.F {}.xy, {};", tmp1, Visit(operation[0])); - AddLine("UP2H.F {}.xy, {};", tmp2, Visit(operation[1])); - AddLine("MUL.F16 {}, {}, {};", tmp1, tmp1, tmp2); - AddLine("PK2H.F {}.x, {};", tmp1, tmp1); - return fmt::format("{}.x", tmp1); -} - -std::string ARBDecompiler::HFma2(Operation operation) { - const std::string tmp1 = AllocVectorTemporary(); - const std::string tmp2 = AllocVectorTemporary(); - const std::string tmp3 = AllocVectorTemporary(); - AddLine("UP2H.F {}.xy, {};", tmp1, Visit(operation[0])); - AddLine("UP2H.F {}.xy, {};", tmp2, Visit(operation[1])); - AddLine("UP2H.F {}.xy, {};", tmp3, Visit(operation[2])); - AddLine("MAD.F16 {}, {}, {}, {};", tmp1, tmp1, tmp2, tmp3); - AddLine("PK2H.F {}.x, {};", tmp1, tmp1); - return fmt::format("{}.x", tmp1); -} - -std::string ARBDecompiler::HAbsolute(Operation operation) { - const std::string temporary = AllocVectorTemporary(); - AddLine("UP2H.F {}.xy, {};", temporary, Visit(operation[0])); - AddLine("PK2H.F {}.x, |{}|;", temporary, temporary); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::HNegate(Operation operation) { - const std::string temporary = AllocVectorTemporary(); - AddLine("UP2H.F {}.xy, {};", temporary, Visit(operation[0])); - AddLine("MOVC.S RC.x, {};", Visit(operation[1])); - AddLine("MOV.F {}.x (NE.x), -{}.x;", temporary, temporary); - AddLine("MOVC.S RC.x, {};", Visit(operation[2])); - AddLine("MOV.F {}.y (NE.x), -{}.y;", temporary, temporary); - AddLine("PK2H.F {}.x, {};", temporary, temporary); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::HClamp(Operation operation) { - const std::string tmp1 = AllocVectorTemporary(); - const std::string tmp2 = AllocVectorTemporary(); - AddLine("UP2H.F {}.xy, {};", tmp1, Visit(operation[0])); - AddLine("MOV.U {}.x, {};", tmp2, Visit(operation[1])); - AddLine("MOV.U {}.y, {}.x;", tmp2, tmp2); - AddLine("MAX.F {}, {}, {};", tmp1, tmp1, tmp2); - AddLine("MOV.U {}.x, {};", tmp2, Visit(operation[2])); - AddLine("MOV.U {}.y, {}.x;", tmp2, tmp2); - AddLine("MIN.F {}, {}, {};", tmp1, tmp1, tmp2); - AddLine("PK2H.F {}.x, {};", tmp1, tmp1); - return fmt::format("{}.x", tmp1); -} - -std::string ARBDecompiler::HCastFloat(Operation operation) { - const std::string temporary = AllocVectorTemporary(); - AddLine("MOV.F {}.y, {{0, 0, 0, 0}};", temporary); - AddLine("MOV.F {}.x, {};", temporary, Visit(operation[0])); - AddLine("PK2H.F {}.x, {};", temporary, temporary); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::HUnpack(Operation operation) { - std::string operand = Visit(operation[0]); - switch (std::get<Tegra::Shader::HalfType>(operation.GetMeta())) { - case Tegra::Shader::HalfType::H0_H1: - return operand; - case Tegra::Shader::HalfType::F32: { - const std::string temporary = AllocVectorTemporary(); - AddLine("MOV.U {}.x, {};", temporary, operand); - AddLine("MOV.U {}.y, {}.x;", temporary, temporary); - AddLine("PK2H.F {}.x, {};", temporary, temporary); - return fmt::format("{}.x", temporary); - } - case Tegra::Shader::HalfType::H0_H0: { - const std::string temporary = AllocVectorTemporary(); - AddLine("UP2H.F {}.xy, {};", temporary, operand); - AddLine("MOV.U {}.y, {}.x;", temporary, temporary); - AddLine("PK2H.F {}.x, {};", temporary, temporary); - return fmt::format("{}.x", temporary); - } - case Tegra::Shader::HalfType::H1_H1: { - const std::string temporary = AllocVectorTemporary(); - AddLine("UP2H.F {}.xy, {};", temporary, operand); - AddLine("MOV.U {}.x, {}.y;", temporary, temporary); - AddLine("PK2H.F {}.x, {};", temporary, temporary); - return fmt::format("{}.x", temporary); - } - } - UNREACHABLE(); - return "{0, 0, 0, 0}.x"; -} - -std::string ARBDecompiler::HMergeF32(Operation operation) { - const std::string temporary = AllocVectorTemporary(); - AddLine("UP2H.F {}.xy, {};", temporary, Visit(operation[0])); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::HMergeH0(Operation operation) { - const std::string temporary = AllocVectorTemporary(); - AddLine("UP2H.F {}.xy, {};", temporary, Visit(operation[0])); - AddLine("UP2H.F {}.zw, {};", temporary, Visit(operation[1])); - AddLine("MOV.U {}.x, {}.z;", temporary, temporary); - AddLine("PK2H.F {}.x, {};", temporary, temporary); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::HMergeH1(Operation operation) { - const std::string temporary = AllocVectorTemporary(); - AddLine("UP2H.F {}.xy, {};", temporary, Visit(operation[0])); - AddLine("UP2H.F {}.zw, {};", temporary, Visit(operation[1])); - AddLine("MOV.U {}.y, {}.w;", temporary, temporary); - AddLine("PK2H.F {}.x, {};", temporary, temporary); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::HPack2(Operation operation) { - const std::string temporary = AllocVectorTemporary(); - AddLine("MOV.U {}.x, {};", temporary, Visit(operation[0])); - AddLine("MOV.U {}.y, {};", temporary, Visit(operation[1])); - AddLine("PK2H.F {}.x, {};", temporary, temporary); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::LogicalAssign(Operation operation) { - const Node& dest = operation[0]; - const Node& src = operation[1]; - - std::string target; - - if (const auto pred = std::get_if<PredicateNode>(&*dest)) { - ASSERT_MSG(!pred->IsNegated(), "Negating logical assignment"); - - const Tegra::Shader::Pred index = pred->GetIndex(); - switch (index) { - case Tegra::Shader::Pred::NeverExecute: - case Tegra::Shader::Pred::UnusedIndex: - // Writing to these predicates is a no-op - return {}; - } - target = fmt::format("P{}.x", static_cast<u64>(index)); - } else if (const auto internal_flag = std::get_if<InternalFlagNode>(&*dest)) { - const std::size_t index = static_cast<std::size_t>(internal_flag->GetFlag()); - target = fmt::format("{}.x", INTERNAL_FLAG_NAMES[index]); - } else { - UNREACHABLE(); - ResetTemporaries(); - return {}; - } - - AddLine("MOV.U {}, {};", target, Visit(src)); - ResetTemporaries(); - return {}; -} - -std::string ARBDecompiler::LogicalPick2(Operation operation) { - std::string temporary = AllocTemporary(); - const u32 index = std::get<ImmediateNode>(*operation[1]).GetValue(); - AddLine("MOV.U {}, {}.{};", temporary, Visit(operation[0]), Swizzle(index)); - return temporary; -} - -std::string ARBDecompiler::LogicalAnd2(Operation operation) { - std::string temporary = AllocTemporary(); - const std::string op = Visit(operation[0]); - AddLine("AND.U {}, {}.x, {}.y;", temporary, op, op); - return temporary; -} - -std::string ARBDecompiler::FloatOrdered(Operation operation) { - std::string temporary = AllocTemporary(); - AddLine("MOVC.F32 RC.x, {};", Visit(operation[0])); - AddLine("MOVC.F32 RC.y, {};", Visit(operation[1])); - AddLine("MOV.S {}, -1;", temporary); - AddLine("MOV.S {} (NAN.x), 0;", temporary); - AddLine("MOV.S {} (NAN.y), 0;", temporary); - return temporary; -} - -std::string ARBDecompiler::FloatUnordered(Operation operation) { - std::string temporary = AllocTemporary(); - AddLine("MOVC.F32 RC.x, {};", Visit(operation[0])); - AddLine("MOVC.F32 RC.y, {};", Visit(operation[1])); - AddLine("MOV.S {}, 0;", temporary); - AddLine("MOV.S {} (NAN.x), -1;", temporary); - AddLine("MOV.S {} (NAN.y), -1;", temporary); - return temporary; -} - -std::string ARBDecompiler::LogicalAddCarry(Operation operation) { - std::string temporary = AllocTemporary(); - AddLine("ADDC.U RC, {}, {};", Visit(operation[0]), Visit(operation[1])); - AddLine("MOV.S {}, 0;", temporary); - AddLine("IF CF.x;"); - AddLine("MOV.S {}, -1;", temporary); - AddLine("ENDIF;"); - return temporary; -} - -std::string ARBDecompiler::Texture(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index; - const auto [coords, temporary, swizzle] = BuildCoords(operation); - - std::string_view opcode = "TEX"; - std::string extra; - if (meta.bias) { - ASSERT(!meta.lod); - opcode = "TXB"; - - if (swizzle < 4) { - AddLine("MOV.F {}.w, {};", temporary, Visit(meta.bias)); - } else { - const std::string bias = AllocTemporary(); - AddLine("MOV.F {}, {};", bias, Visit(meta.bias)); - extra = fmt::format(" {},", bias); - } - } - if (meta.lod) { - ASSERT(!meta.bias); - opcode = "TXL"; - - if (swizzle < 4) { - AddLine("MOV.F {}.w, {};", temporary, Visit(meta.lod)); - } else { - const std::string lod = AllocTemporary(); - AddLine("MOV.F {}, {};", lod, Visit(meta.lod)); - extra = fmt::format(" {},", lod); - } - } - - AddLine("{}.F {}, {},{} texture[{}], {}{};", opcode, temporary, coords, extra, sampler_id, - TextureType(meta), BuildAoffi(operation)); - AddLine("MOV.U {}.x, {}.{};", temporary, temporary, Swizzle(meta.element)); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::TextureGather(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index; - const auto [coords, temporary, swizzle] = BuildCoords(operation); - - std::string comp; - if (!meta.sampler.is_shadow) { - const auto& immediate = std::get<ImmediateNode>(*meta.component); - comp = fmt::format(".{}", Swizzle(immediate.GetValue())); - } - - AddLine("TXG.F {}, {}, texture[{}]{}, {}{};", temporary, temporary, sampler_id, comp, - TextureType(meta), BuildAoffi(operation)); - AddLine("MOV.U {}.x, {}.{};", temporary, coords, Swizzle(meta.element)); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::TextureQueryDimensions(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - const std::string temporary = AllocVectorTemporary(); - const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index; - - ASSERT(!meta.sampler.is_array); - - const std::string lod = operation.GetOperandsCount() > 0 ? Visit(operation[0]) : "0"; - AddLine("TXQ {}, {}, texture[{}], {};", temporary, lod, sampler_id, TextureType(meta)); - AddLine("MOV.U {}.x, {}.{};", temporary, temporary, Swizzle(meta.element)); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::TextureQueryLod(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - const std::string temporary = AllocVectorTemporary(); - const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index; - - ASSERT(!meta.sampler.is_array); - - const std::size_t count = operation.GetOperandsCount(); - for (std::size_t i = 0; i < count; ++i) { - AddLine("MOV.F {}.{}, {};", temporary, Swizzle(i), Visit(operation[i])); - } - AddLine("LOD.F {}, {}, texture[{}], {};", temporary, temporary, sampler_id, TextureType(meta)); - AddLine("MUL.F32 {}, {}, {{256, 256, 0, 0}};", temporary, temporary); - AddLine("TRUNC.S {}, {};", temporary, temporary); - AddLine("MOV.U {}.x, {}.{};", temporary, temporary, Swizzle(meta.element)); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::TexelFetch(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index; - const auto [coords, temporary, swizzle] = BuildCoords(operation); - - if (!meta.sampler.is_buffer) { - ASSERT(swizzle < 4); - AddLine("MOV.F {}.w, {};", temporary, Visit(meta.lod)); - } - AddLine("TXF.F {}, {}, texture[{}], {}{};", temporary, coords, sampler_id, TextureType(meta), - BuildAoffi(operation)); - AddLine("MOV.U {}.x, {}.{};", temporary, temporary, Swizzle(meta.element)); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::TextureGradient(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index; - const std::string ddx = AllocVectorTemporary(); - const std::string ddy = AllocVectorTemporary(); - const std::string coord = std::get<1>(BuildCoords(operation)); - - const std::size_t num_components = meta.derivates.size() / 2; - for (std::size_t index = 0; index < num_components; ++index) { - const char swizzle = Swizzle(index); - AddLine("MOV.F {}.{}, {};", ddx, swizzle, Visit(meta.derivates[index * 2])); - AddLine("MOV.F {}.{}, {};", ddy, swizzle, Visit(meta.derivates[index * 2 + 1])); - } - - const std::string_view result = coord; - AddLine("TXD.F {}, {}, {}, {}, texture[{}], {}{};", result, coord, ddx, ddy, sampler_id, - TextureType(meta), BuildAoffi(operation)); - AddLine("MOV.F {}.x, {}.{};", result, result, Swizzle(meta.element)); - return fmt::format("{}.x", result); -} - -std::string ARBDecompiler::ImageLoad(Operation operation) { - const auto& meta = std::get<MetaImage>(operation.GetMeta()); - const u32 image_id = device.GetBaseBindings(stage).image + meta.image.index; - const std::size_t count = operation.GetOperandsCount(); - const std::string_view type = ImageType(meta.image.type); - - const std::string temporary = AllocVectorTemporary(); - for (std::size_t i = 0; i < count; ++i) { - AddLine("MOV.S {}.{}, {};", temporary, Swizzle(i), Visit(operation[i])); - } - AddLine("LOADIM.F {}, {}, image[{}], {};", temporary, temporary, image_id, type); - AddLine("MOV.F {}.x, {}.{};", temporary, temporary, Swizzle(meta.element)); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::ImageStore(Operation operation) { - const auto& meta = std::get<MetaImage>(operation.GetMeta()); - const u32 image_id = device.GetBaseBindings(stage).image + meta.image.index; - const std::size_t num_coords = operation.GetOperandsCount(); - const std::size_t num_values = meta.values.size(); - const std::string_view type = ImageType(meta.image.type); - - const std::string coord = AllocVectorTemporary(); - const std::string value = AllocVectorTemporary(); - for (std::size_t i = 0; i < num_coords; ++i) { - AddLine("MOV.S {}.{}, {};", coord, Swizzle(i), Visit(operation[i])); - } - for (std::size_t i = 0; i < num_values; ++i) { - AddLine("MOV.F {}.{}, {};", value, Swizzle(i), Visit(meta.values[i])); - } - AddLine("STOREIM.F image[{}], {}, {}, {};", image_id, value, coord, type); - return {}; -} - -std::string ARBDecompiler::Branch(Operation operation) { - const auto target = std::get<ImmediateNode>(*operation[0]); - AddLine("MOV.U PC.x, {};", target.GetValue()); - AddLine("CONT;"); - return {}; -} - -std::string ARBDecompiler::BranchIndirect(Operation operation) { - AddLine("MOV.U PC.x, {};", Visit(operation[0])); - AddLine("CONT;"); - return {}; -} - -std::string ARBDecompiler::PushFlowStack(Operation operation) { - const auto stack = std::get<MetaStackClass>(operation.GetMeta()); - const u32 target = std::get<ImmediateNode>(*operation[0]).GetValue(); - const std::string_view stack_name = StackName(stack); - AddLine("MOV.U {}[{}_TOP.x].x, {};", stack_name, stack_name, target); - AddLine("ADD.S {}_TOP.x, {}_TOP.x, 1;", stack_name, stack_name); - return {}; -} - -std::string ARBDecompiler::PopFlowStack(Operation operation) { - const auto stack = std::get<MetaStackClass>(operation.GetMeta()); - const std::string_view stack_name = StackName(stack); - AddLine("SUB.S {}_TOP.x, {}_TOP.x, 1;", stack_name, stack_name); - AddLine("MOV.U PC.x, {}[{}_TOP.x].x;", stack_name, stack_name); - AddLine("CONT;"); - return {}; -} - -std::string ARBDecompiler::Exit(Operation) { - Exit(); - return {}; -} - -std::string ARBDecompiler::Discard(Operation) { - AddLine("KIL TR;"); - return {}; -} - -std::string ARBDecompiler::EmitVertex(Operation) { - AddLine("EMIT;"); - return {}; -} - -std::string ARBDecompiler::EndPrimitive(Operation) { - AddLine("ENDPRIM;"); - return {}; -} - -std::string ARBDecompiler::InvocationId(Operation) { - return "primitive.invocation"; -} - -std::string ARBDecompiler::YNegate(Operation) { - LOG_WARNING(Render_OpenGL, "(STUBBED)"); - std::string temporary = AllocTemporary(); - AddLine("MOV.F {}, 1;", temporary); - return temporary; -} - -std::string ARBDecompiler::ThreadId(Operation) { - return fmt::format("{}.threadid", StageInputName(stage)); -} - -std::string ARBDecompiler::ShuffleIndexed(Operation operation) { - if (!device.HasWarpIntrinsics()) { - LOG_ERROR(Render_OpenGL, - "NV_shader_thread_shuffle is missing. Kepler or better is required."); - return Visit(operation[0]); - } - const std::string temporary = AllocVectorTemporary(); - AddLine("SHFIDX.U {}, {}, {}, {{31, 0, 0, 0}};", temporary, Visit(operation[0]), - Visit(operation[1])); - AddLine("MOV.U {}.x, {}.y;", temporary, temporary); - return fmt::format("{}.x", temporary); -} - -std::string ARBDecompiler::Barrier(Operation) { - AddLine("BAR;"); - return {}; -} - -std::string ARBDecompiler::MemoryBarrierGroup(Operation) { - AddLine("MEMBAR.CTA;"); - return {}; -} - -std::string ARBDecompiler::MemoryBarrierGlobal(Operation) { - AddLine("MEMBAR;"); - return {}; -} - -} // Anonymous namespace - -std::string DecompileAssemblyShader(const Device& device, const VideoCommon::Shader::ShaderIR& ir, - const VideoCommon::Shader::Registry& registry, - Tegra::Engines::ShaderType stage, std::string_view identifier) { - return ARBDecompiler(device, ir, registry, stage, identifier).Code(); -} - -} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_arb_decompiler.h b/src/video_core/renderer_opengl/gl_arb_decompiler.h deleted file mode 100644 index 6afc87220..000000000 --- a/src/video_core/renderer_opengl/gl_arb_decompiler.h +++ /dev/null @@ -1,29 +0,0 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <string> -#include <string_view> - -#include "common/common_types.h" - -namespace Tegra::Engines { -enum class ShaderType : u32; -} - -namespace VideoCommon::Shader { -class ShaderIR; -class Registry; -} // namespace VideoCommon::Shader - -namespace OpenGL { - -class Device; - -std::string DecompileAssemblyShader(const Device& device, const VideoCommon::Shader::ShaderIR& ir, - const VideoCommon::Shader::Registry& registry, - Tegra::Engines::ShaderType stage, std::string_view identifier); - -} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp index c225d1fc9..07a995f7d 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -2,14 +2,18 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <algorithm> #include <span> #include "video_core/buffer_cache/buffer_cache.h" #include "video_core/renderer_opengl/gl_buffer_cache.h" #include "video_core/renderer_opengl/gl_device.h" +#include "video_core/renderer_opengl/maxwell_to_gl.h" namespace OpenGL { namespace { +using VideoCore::Surface::PixelFormat; + struct BindlessSSBO { GLuint64EXT address; GLsizei length; @@ -21,6 +25,25 @@ constexpr std::array PROGRAM_LUT{ GL_VERTEX_PROGRAM_NV, GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV, GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV, }; + +[[nodiscard]] GLenum GetTextureBufferFormat(GLenum gl_format) { + switch (gl_format) { + case GL_RGBA8_SNORM: + return GL_RGBA8; + case GL_R8_SNORM: + return GL_R8; + case GL_RGBA16_SNORM: + return GL_RGBA16; + case GL_R16_SNORM: + return GL_R16; + case GL_RG16_SNORM: + return GL_RG16; + case GL_RG8_SNORM: + return GL_RG8; + default: + return gl_format; + } +} } // Anonymous namespace Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params) @@ -62,6 +85,30 @@ void Buffer::MakeResident(GLenum access) noexcept { glMakeNamedBufferResidentNV(buffer.handle, access); } +GLuint Buffer::View(u32 offset, u32 size, PixelFormat format) { + const auto it{std::ranges::find_if(views, [offset, size, format](const BufferView& view) { + return offset == view.offset && size == view.size && format == view.format; + })}; + if (it != views.end()) { + return it->texture.handle; + } + OGLTexture texture; + texture.Create(GL_TEXTURE_BUFFER); + const GLenum gl_format{MaxwellToGL::GetFormatTuple(format).internal_format}; + const GLenum texture_format{GetTextureBufferFormat(gl_format)}; + if (texture_format != gl_format) { + LOG_WARNING(Render_OpenGL, "Emulating SNORM texture buffer with UNORM."); + } + glTextureBufferRange(texture.handle, texture_format, buffer.handle, offset, size); + views.push_back({ + .offset = offset, + .size = size, + .format = format, + .texture = std::move(texture), + }); + return views.back().texture.handle; +} + BufferCacheRuntime::BufferCacheRuntime(const Device& device_) : device{device_}, has_fast_buffer_sub_data{device.HasFastBufferSubData()}, use_assembly_shaders{device.UseAssemblyShaders()}, @@ -98,6 +145,12 @@ void BufferCacheRuntime::CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer, } } +void BufferCacheRuntime::ClearBuffer(Buffer& dest_buffer, u32 offset, size_t size, u32 value) { + glClearNamedBufferSubData(dest_buffer.Handle(), GL_R32UI, static_cast<GLintptr>(offset), + static_cast<GLsizeiptr>(size / sizeof(u32)), GL_RED, GL_UNSIGNED_INT, + &value); +} + void BufferCacheRuntime::BindIndexBuffer(Buffer& buffer, u32 offset, u32 size) { if (has_unified_vertex_buffers) { buffer.MakeResident(GL_READ_ONLY); @@ -138,7 +191,7 @@ void BufferCacheRuntime::BindUniformBuffer(size_t stage, u32 binding_index, Buff glBindBufferRangeNV(PABO_LUT[stage], binding_index, handle, 0, static_cast<GLsizeiptr>(size)); } else { - const GLuint base_binding = device.GetBaseBindings(stage).uniform_buffer; + const GLuint base_binding = graphics_base_uniform_bindings[stage]; const GLuint binding = base_binding + binding_index; glBindBufferRange(GL_UNIFORM_BUFFER, binding, buffer.Handle(), static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); @@ -165,7 +218,12 @@ void BufferCacheRuntime::BindComputeUniformBuffer(u32 binding_index, Buffer& buf void BufferCacheRuntime::BindStorageBuffer(size_t stage, u32 binding_index, Buffer& buffer, u32 offset, u32 size, bool is_written) { - if (use_assembly_shaders) { + if (use_storage_buffers) { + const GLuint base_binding = graphics_base_storage_bindings[stage]; + const GLuint binding = base_binding + binding_index; + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, buffer.Handle(), + static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); + } else { const BindlessSSBO ssbo{ .address = buffer.HostGpuAddr() + offset, .length = static_cast<GLsizei>(size), @@ -174,17 +232,19 @@ void BufferCacheRuntime::BindStorageBuffer(size_t stage, u32 binding_index, Buff buffer.MakeResident(is_written ? GL_READ_WRITE : GL_READ_ONLY); glProgramLocalParametersI4uivNV(PROGRAM_LUT[stage], binding_index, 1, reinterpret_cast<const GLuint*>(&ssbo)); - } else { - const GLuint base_binding = device.GetBaseBindings(stage).shader_storage_buffer; - const GLuint binding = base_binding + binding_index; - glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, buffer.Handle(), - static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); } } void BufferCacheRuntime::BindComputeStorageBuffer(u32 binding_index, Buffer& buffer, u32 offset, u32 size, bool is_written) { - if (use_assembly_shaders) { + if (use_storage_buffers) { + if (size != 0) { + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding_index, buffer.Handle(), + static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); + } else { + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding_index, 0, 0, 0); + } + } else { const BindlessSSBO ssbo{ .address = buffer.HostGpuAddr() + offset, .length = static_cast<GLsizei>(size), @@ -193,11 +253,6 @@ void BufferCacheRuntime::BindComputeStorageBuffer(u32 binding_index, Buffer& buf buffer.MakeResident(is_written ? GL_READ_WRITE : GL_READ_ONLY); glProgramLocalParametersI4uivNV(GL_COMPUTE_PROGRAM_NV, binding_index, 1, reinterpret_cast<const GLuint*>(&ssbo)); - } else if (size == 0) { - glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding_index, 0, 0, 0); - } else { - glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding_index, buffer.Handle(), - static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); } } @@ -207,4 +262,13 @@ void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, Buffer& buffer, static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); } +void BufferCacheRuntime::BindTextureBuffer(Buffer& buffer, u32 offset, u32 size, + PixelFormat format) { + *texture_handles++ = buffer.View(offset, size, format); +} + +void BufferCacheRuntime::BindImageBuffer(Buffer& buffer, u32 offset, u32 size, PixelFormat format) { + *image_handles++ = buffer.View(offset, size, format); +} + } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h index d8b20a9af..060d36427 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.h +++ b/src/video_core/renderer_opengl/gl_buffer_cache.h @@ -32,6 +32,8 @@ public: void MakeResident(GLenum access) noexcept; + [[nodiscard]] GLuint View(u32 offset, u32 size, VideoCore::Surface::PixelFormat format); + [[nodiscard]] GLuint64EXT HostGpuAddr() const noexcept { return address; } @@ -41,9 +43,17 @@ public: } private: + struct BufferView { + u32 offset; + u32 size; + VideoCore::Surface::PixelFormat format; + OGLTexture texture; + }; + GLuint64EXT address = 0; OGLBuffer buffer; GLenum current_residency_access = GL_NONE; + std::vector<BufferView> views; }; class BufferCacheRuntime { @@ -57,6 +67,8 @@ public: void CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer, std::span<const VideoCommon::BufferCopy> copies); + void ClearBuffer(Buffer& dest_buffer, u32 offset, size_t size, u32 value); + void BindIndexBuffer(Buffer& buffer, u32 offset, u32 size); void BindVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size, u32 stride); @@ -73,17 +85,21 @@ public: void BindTransformFeedbackBuffer(u32 index, Buffer& buffer, u32 offset, u32 size); + void BindTextureBuffer(Buffer& buffer, u32 offset, u32 size, + VideoCore::Surface::PixelFormat format); + + void BindImageBuffer(Buffer& buffer, u32 offset, u32 size, + VideoCore::Surface::PixelFormat format); + void BindFastUniformBuffer(size_t stage, u32 binding_index, u32 size) { + const GLuint handle = fast_uniforms[stage][binding_index].handle; + const GLsizeiptr gl_size = static_cast<GLsizeiptr>(size); if (use_assembly_shaders) { - const GLuint handle = fast_uniforms[stage][binding_index].handle; - const GLsizeiptr gl_size = static_cast<GLsizeiptr>(size); glBindBufferRangeNV(PABO_LUT[stage], binding_index, handle, 0, gl_size); } else { - const GLuint base_binding = device.GetBaseBindings(stage).uniform_buffer; + const GLuint base_binding = graphics_base_uniform_bindings[stage]; const GLuint binding = base_binding + binding_index; - glBindBufferRange(GL_UNIFORM_BUFFER, binding, - fast_uniforms[stage][binding_index].handle, 0, - static_cast<GLsizeiptr>(size)); + glBindBufferRange(GL_UNIFORM_BUFFER, binding, handle, 0, gl_size); } } @@ -101,7 +117,7 @@ public: std::span<u8> BindMappedUniformBuffer(size_t stage, u32 binding_index, u32 size) noexcept { const auto [mapped_span, offset] = stream_buffer->Request(static_cast<size_t>(size)); - const GLuint base_binding = device.GetBaseBindings(stage).uniform_buffer; + const GLuint base_binding = graphics_base_uniform_bindings[stage]; const GLuint binding = base_binding + binding_index; glBindBufferRange(GL_UNIFORM_BUFFER, binding, stream_buffer->Handle(), static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); @@ -116,6 +132,27 @@ public: return has_fast_buffer_sub_data; } + [[nodiscard]] bool SupportsNonZeroUniformOffset() const noexcept { + return !use_assembly_shaders; + } + + void SetBaseUniformBindings(const std::array<GLuint, 5>& bindings) { + graphics_base_uniform_bindings = bindings; + } + + void SetBaseStorageBindings(const std::array<GLuint, 5>& bindings) { + graphics_base_storage_bindings = bindings; + } + + void SetImagePointers(GLuint* texture_handles_, GLuint* image_handles_) { + texture_handles = texture_handles_; + image_handles = image_handles_; + } + + void SetEnableStorageBuffers(bool use_storage_buffers_) { + use_storage_buffers = use_storage_buffers_; + } + private: static constexpr std::array PABO_LUT{ GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV, @@ -129,8 +166,15 @@ private: bool use_assembly_shaders = false; bool has_unified_vertex_buffers = false; + bool use_storage_buffers = false; + u32 max_attributes = 0; + std::array<GLuint, 5> graphics_base_uniform_bindings{}; + std::array<GLuint, 5> graphics_base_storage_bindings{}; + GLuint* texture_handles = nullptr; + GLuint* image_handles = nullptr; + std::optional<StreamBuffer> stream_buffer; std::array<std::array<OGLBuffer, VideoCommon::NUM_GRAPHICS_UNIFORM_BUFFERS>, @@ -154,6 +198,7 @@ struct BufferCacheParams { static constexpr bool NEEDS_BIND_UNIFORM_INDEX = true; static constexpr bool NEEDS_BIND_STORAGE_INDEX = true; static constexpr bool USE_MEMORY_MAPS = false; + static constexpr bool SEPARATE_IMAGE_BUFFER_BINDINGS = true; }; using BufferCache = VideoCommon::BufferCache<BufferCacheParams>; diff --git a/src/video_core/renderer_opengl/gl_compute_pipeline.cpp b/src/video_core/renderer_opengl/gl_compute_pipeline.cpp new file mode 100644 index 000000000..aa1cc592f --- /dev/null +++ b/src/video_core/renderer_opengl/gl_compute_pipeline.cpp @@ -0,0 +1,209 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <cstring> + +#include "common/cityhash.h" +#include "common/settings.h" // for enum class Settings::ShaderBackend +#include "video_core/renderer_opengl/gl_compute_pipeline.h" +#include "video_core/renderer_opengl/gl_shader_manager.h" +#include "video_core/renderer_opengl/gl_shader_util.h" + +namespace OpenGL { + +using Shader::ImageBufferDescriptor; +using Tegra::Texture::TexturePair; +using VideoCommon::ImageId; + +constexpr u32 MAX_TEXTURES = 64; +constexpr u32 MAX_IMAGES = 16; + +template <typename Range> +u32 AccumulateCount(const Range& range) { + u32 num{}; + for (const auto& desc : range) { + num += desc.count; + } + return num; +} + +size_t ComputePipelineKey::Hash() const noexcept { + return static_cast<size_t>( + Common::CityHash64(reinterpret_cast<const char*>(this), sizeof *this)); +} + +bool ComputePipelineKey::operator==(const ComputePipelineKey& rhs) const noexcept { + return std::memcmp(this, &rhs, sizeof *this) == 0; +} + +ComputePipeline::ComputePipeline(const Device& device, TextureCache& texture_cache_, + BufferCache& buffer_cache_, Tegra::MemoryManager& gpu_memory_, + Tegra::Engines::KeplerCompute& kepler_compute_, + ProgramManager& program_manager_, const Shader::Info& info_, + std::string code, std::vector<u32> code_v) + : texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, gpu_memory{gpu_memory_}, + kepler_compute{kepler_compute_}, program_manager{program_manager_}, info{info_} { + switch (device.GetShaderBackend()) { + case Settings::ShaderBackend::GLSL: + source_program = CreateProgram(code, GL_COMPUTE_SHADER); + break; + case Settings::ShaderBackend::GLASM: + assembly_program = CompileProgram(code, GL_COMPUTE_PROGRAM_NV); + break; + case Settings::ShaderBackend::SPIRV: + source_program = CreateProgram(code_v, GL_COMPUTE_SHADER); + break; + } + std::copy_n(info.constant_buffer_used_sizes.begin(), uniform_buffer_sizes.size(), + uniform_buffer_sizes.begin()); + + num_texture_buffers = AccumulateCount(info.texture_buffer_descriptors); + num_image_buffers = AccumulateCount(info.image_buffer_descriptors); + + const u32 num_textures{num_texture_buffers + AccumulateCount(info.texture_descriptors)}; + ASSERT(num_textures <= MAX_TEXTURES); + + const u32 num_images{num_image_buffers + AccumulateCount(info.image_descriptors)}; + ASSERT(num_images <= MAX_IMAGES); + + const bool is_glasm{assembly_program.handle != 0}; + const u32 num_storage_buffers{AccumulateCount(info.storage_buffers_descriptors)}; + use_storage_buffers = + !is_glasm || num_storage_buffers < device.GetMaxGLASMStorageBufferBlocks(); + writes_global_memory = !use_storage_buffers && + std::ranges::any_of(info.storage_buffers_descriptors, + [](const auto& desc) { return desc.is_written; }); +} + +void ComputePipeline::Configure() { + buffer_cache.SetComputeUniformBufferState(info.constant_buffer_mask, &uniform_buffer_sizes); + buffer_cache.UnbindComputeStorageBuffers(); + size_t ssbo_index{}; + for (const auto& desc : info.storage_buffers_descriptors) { + ASSERT(desc.count == 1); + buffer_cache.BindComputeStorageBuffer(ssbo_index, desc.cbuf_index, desc.cbuf_offset, + desc.is_written); + ++ssbo_index; + } + texture_cache.SynchronizeComputeDescriptors(); + + std::array<ImageViewId, MAX_TEXTURES + MAX_IMAGES> image_view_ids; + boost::container::static_vector<u32, MAX_TEXTURES + MAX_IMAGES> image_view_indices; + std::array<GLuint, MAX_TEXTURES> samplers; + std::array<GLuint, MAX_TEXTURES> textures; + std::array<GLuint, MAX_IMAGES> images; + GLsizei sampler_binding{}; + GLsizei texture_binding{}; + GLsizei image_binding{}; + + const auto& qmd{kepler_compute.launch_description}; + const auto& cbufs{qmd.const_buffer_config}; + const bool via_header_index{qmd.linked_tsc != 0}; + const auto read_handle{[&](const auto& desc, u32 index) { + ASSERT(((qmd.const_buffer_enable_mask >> desc.cbuf_index) & 1) != 0); + const u32 index_offset{index << desc.size_shift}; + const u32 offset{desc.cbuf_offset + index_offset}; + const GPUVAddr addr{cbufs[desc.cbuf_index].Address() + offset}; + if constexpr (std::is_same_v<decltype(desc), const Shader::TextureDescriptor&> || + std::is_same_v<decltype(desc), const Shader::TextureBufferDescriptor&>) { + if (desc.has_secondary) { + ASSERT(((qmd.const_buffer_enable_mask >> desc.secondary_cbuf_index) & 1) != 0); + const u32 secondary_offset{desc.secondary_cbuf_offset + index_offset}; + const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].Address() + + secondary_offset}; + const u32 lhs_raw{gpu_memory.Read<u32>(addr)}; + const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)}; + return TexturePair(lhs_raw | rhs_raw, via_header_index); + } + } + return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); + }}; + const auto add_image{[&](const auto& desc) { + for (u32 index = 0; index < desc.count; ++index) { + const auto handle{read_handle(desc, index)}; + image_view_indices.push_back(handle.first); + } + }}; + for (const auto& desc : info.texture_buffer_descriptors) { + for (u32 index = 0; index < desc.count; ++index) { + const auto handle{read_handle(desc, index)}; + image_view_indices.push_back(handle.first); + samplers[sampler_binding++] = 0; + } + } + std::ranges::for_each(info.image_buffer_descriptors, add_image); + for (const auto& desc : info.texture_descriptors) { + for (u32 index = 0; index < desc.count; ++index) { + const auto handle{read_handle(desc, index)}; + image_view_indices.push_back(handle.first); + + Sampler* const sampler = texture_cache.GetComputeSampler(handle.second); + samplers[sampler_binding++] = sampler->Handle(); + } + } + std::ranges::for_each(info.image_descriptors, add_image); + + const std::span indices_span(image_view_indices.data(), image_view_indices.size()); + texture_cache.FillComputeImageViews(indices_span, image_view_ids); + + if (assembly_program.handle != 0) { + program_manager.BindComputeAssemblyProgram(assembly_program.handle); + } else { + program_manager.BindComputeProgram(source_program.handle); + } + buffer_cache.UnbindComputeTextureBuffers(); + size_t texbuf_index{}; + const auto add_buffer{[&](const auto& desc) { + constexpr bool is_image = std::is_same_v<decltype(desc), const ImageBufferDescriptor&>; + for (u32 i = 0; i < desc.count; ++i) { + bool is_written{false}; + if constexpr (is_image) { + is_written = desc.is_written; + } + ImageView& image_view{texture_cache.GetImageView(image_view_ids[texbuf_index])}; + buffer_cache.BindComputeTextureBuffer(texbuf_index, image_view.GpuAddr(), + image_view.BufferSize(), image_view.format, + is_written, is_image); + ++texbuf_index; + } + }}; + std::ranges::for_each(info.texture_buffer_descriptors, add_buffer); + std::ranges::for_each(info.image_buffer_descriptors, add_buffer); + + buffer_cache.UpdateComputeBuffers(); + + buffer_cache.runtime.SetEnableStorageBuffers(use_storage_buffers); + buffer_cache.runtime.SetImagePointers(textures.data(), images.data()); + buffer_cache.BindHostComputeBuffers(); + + const ImageId* views_it{image_view_ids.data() + num_texture_buffers + num_image_buffers}; + texture_binding += num_texture_buffers; + image_binding += num_image_buffers; + + for (const auto& desc : info.texture_descriptors) { + for (u32 index = 0; index < desc.count; ++index) { + ImageView& image_view{texture_cache.GetImageView(*(views_it++))}; + textures[texture_binding++] = image_view.Handle(desc.type); + } + } + for (const auto& desc : info.image_descriptors) { + for (u32 index = 0; index < desc.count; ++index) { + ImageView& image_view{texture_cache.GetImageView(*(views_it++))}; + if (desc.is_written) { + texture_cache.MarkModification(image_view.image_id); + } + images[image_binding++] = image_view.StorageView(desc.type, desc.format); + } + } + if (texture_binding != 0) { + ASSERT(texture_binding == sampler_binding); + glBindTextures(0, texture_binding, textures.data()); + glBindSamplers(0, sampler_binding, samplers.data()); + } + if (image_binding != 0) { + glBindImageTextures(0, image_binding, images.data()); + } +} + +} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_compute_pipeline.h b/src/video_core/renderer_opengl/gl_compute_pipeline.h new file mode 100644 index 000000000..50c676365 --- /dev/null +++ b/src/video_core/renderer_opengl/gl_compute_pipeline.h @@ -0,0 +1,93 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <array> +#include <type_traits> +#include <utility> + +#include "common/common_types.h" +#include "shader_recompiler/shader_info.h" +#include "video_core/renderer_opengl/gl_buffer_cache.h" +#include "video_core/renderer_opengl/gl_resource_manager.h" +#include "video_core/renderer_opengl/gl_texture_cache.h" + +namespace Tegra { +class MemoryManager; +} + +namespace Tegra::Engines { +class KeplerCompute; +} + +namespace Shader { +struct Info; +} + +namespace OpenGL { + +class Device; +class ProgramManager; + +struct ComputePipelineKey { + u64 unique_hash; + u32 shared_memory_size; + std::array<u32, 3> workgroup_size; + + size_t Hash() const noexcept; + + bool operator==(const ComputePipelineKey&) const noexcept; + + bool operator!=(const ComputePipelineKey& rhs) const noexcept { + return !operator==(rhs); + } +}; +static_assert(std::has_unique_object_representations_v<ComputePipelineKey>); +static_assert(std::is_trivially_copyable_v<ComputePipelineKey>); +static_assert(std::is_trivially_constructible_v<ComputePipelineKey>); + +class ComputePipeline { +public: + explicit ComputePipeline(const Device& device, TextureCache& texture_cache_, + BufferCache& buffer_cache_, Tegra::MemoryManager& gpu_memory_, + Tegra::Engines::KeplerCompute& kepler_compute_, + ProgramManager& program_manager_, const Shader::Info& info_, + std::string code, std::vector<u32> code_v); + + void Configure(); + + [[nodiscard]] bool WritesGlobalMemory() const noexcept { + return writes_global_memory; + } + +private: + TextureCache& texture_cache; + BufferCache& buffer_cache; + Tegra::MemoryManager& gpu_memory; + Tegra::Engines::KeplerCompute& kepler_compute; + ProgramManager& program_manager; + + Shader::Info info; + OGLProgram source_program; + OGLAssemblyProgram assembly_program; + VideoCommon::ComputeUniformBufferSizes uniform_buffer_sizes{}; + + u32 num_texture_buffers{}; + u32 num_image_buffers{}; + + bool use_storage_buffers{}; + bool writes_global_memory{}; +}; + +} // namespace OpenGL + +namespace std { +template <> +struct hash<OpenGL::ComputePipelineKey> { + size_t operator()(const OpenGL::ComputePipelineKey& k) const noexcept { + return k.Hash(); + } +}; +} // namespace std diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp index 3b00614e7..9692b8e94 100644 --- a/src/video_core/renderer_opengl/gl_device.cpp +++ b/src/video_core/renderer_opengl/gl_device.cpp @@ -17,39 +17,17 @@ #include "common/logging/log.h" #include "common/scope_exit.h" #include "common/settings.h" +#include "shader_recompiler/stage.h" #include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_resource_manager.h" namespace OpenGL { namespace { -// One uniform block is reserved for emulation purposes -constexpr u32 ReservedUniformBlocks = 1; - -constexpr u32 NumStages = 5; - constexpr std::array LIMIT_UBOS = { GL_MAX_VERTEX_UNIFORM_BLOCKS, GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS, GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS, GL_MAX_GEOMETRY_UNIFORM_BLOCKS, GL_MAX_FRAGMENT_UNIFORM_BLOCKS, GL_MAX_COMPUTE_UNIFORM_BLOCKS, }; -constexpr std::array LIMIT_SSBOS = { - GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS, GL_MAX_TESS_CONTROL_SHADER_STORAGE_BLOCKS, - GL_MAX_TESS_EVALUATION_SHADER_STORAGE_BLOCKS, GL_MAX_GEOMETRY_SHADER_STORAGE_BLOCKS, - GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS, GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS, -}; -constexpr std::array LIMIT_SAMPLERS = { - GL_MAX_VERTEX_TEXTURE_IMAGE_UNITS, - GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS, - GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS, - GL_MAX_GEOMETRY_TEXTURE_IMAGE_UNITS, - GL_MAX_TEXTURE_IMAGE_UNITS, - GL_MAX_COMPUTE_TEXTURE_IMAGE_UNITS, -}; -constexpr std::array LIMIT_IMAGES = { - GL_MAX_VERTEX_IMAGE_UNIFORMS, GL_MAX_TESS_CONTROL_IMAGE_UNIFORMS, - GL_MAX_TESS_EVALUATION_IMAGE_UNIFORMS, GL_MAX_GEOMETRY_IMAGE_UNIFORMS, - GL_MAX_FRAGMENT_IMAGE_UNIFORMS, GL_MAX_COMPUTE_IMAGE_UNIFORMS, -}; template <typename T> T GetInteger(GLenum pname) { @@ -82,81 +60,18 @@ bool HasExtension(std::span<const std::string_view> extensions, std::string_view return std::ranges::find(extensions, extension) != extensions.end(); } -u32 Extract(u32& base, u32& num, u32 amount, std::optional<GLenum> limit = {}) { - ASSERT(num >= amount); - if (limit) { - amount = std::min(amount, GetInteger<u32>(*limit)); - } - num -= amount; - return std::exchange(base, base + amount); -} - -std::array<u32, Tegra::Engines::MaxShaderTypes> BuildMaxUniformBuffers() noexcept { - std::array<u32, Tegra::Engines::MaxShaderTypes> max; - std::ranges::transform(LIMIT_UBOS, max.begin(), - [](GLenum pname) { return GetInteger<u32>(pname); }); +std::array<u32, Shader::MaxStageTypes> BuildMaxUniformBuffers() noexcept { + std::array<u32, Shader::MaxStageTypes> max; + std::ranges::transform(LIMIT_UBOS, max.begin(), &GetInteger<u32>); return max; } -std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindings() noexcept { - std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> bindings; - - static constexpr std::array<std::size_t, 5> stage_swizzle{0, 1, 2, 3, 4}; - const u32 total_ubos = GetInteger<u32>(GL_MAX_UNIFORM_BUFFER_BINDINGS); - const u32 total_ssbos = GetInteger<u32>(GL_MAX_SHADER_STORAGE_BUFFER_BINDINGS); - const u32 total_samplers = GetInteger<u32>(GL_MAX_COMBINED_TEXTURE_IMAGE_UNITS); - - u32 num_ubos = total_ubos - ReservedUniformBlocks; - u32 num_ssbos = total_ssbos; - u32 num_samplers = total_samplers; - - u32 base_ubo = ReservedUniformBlocks; - u32 base_ssbo = 0; - u32 base_samplers = 0; - - for (std::size_t i = 0; i < NumStages; ++i) { - const std::size_t stage = stage_swizzle[i]; - bindings[stage] = { - Extract(base_ubo, num_ubos, total_ubos / NumStages, LIMIT_UBOS[stage]), - Extract(base_ssbo, num_ssbos, total_ssbos / NumStages, LIMIT_SSBOS[stage]), - Extract(base_samplers, num_samplers, total_samplers / NumStages, - LIMIT_SAMPLERS[stage])}; - } - - u32 num_images = GetInteger<u32>(GL_MAX_IMAGE_UNITS); - u32 base_images = 0; - - // GL_MAX_IMAGE_UNITS is guaranteed by the spec to have a minimum value of 8. - // Due to the limitation of GL_MAX_IMAGE_UNITS, reserve at least 4 image bindings on the - // fragment stage, and at least 1 for the rest of the stages. - // So far games are observed to use 1 image binding on vertex and 4 on fragment stages. - - // Reserve at least 4 image bindings on the fragment stage. - bindings[4].image = - Extract(base_images, num_images, std::max(4U, num_images / NumStages), LIMIT_IMAGES[4]); - - // This is guaranteed to be at least 1. - const u32 total_extracted_images = num_images / (NumStages - 1); - - // Reserve the other image bindings. - for (std::size_t i = 0; i < NumStages; ++i) { - const std::size_t stage = stage_swizzle[i]; - if (stage == 4) { - continue; - } - bindings[stage].image = - Extract(base_images, num_images, total_extracted_images, LIMIT_IMAGES[stage]); - } - - // Compute doesn't care about any of this. - bindings[5] = {0, 0, 0, 0}; - - return bindings; -} - bool IsASTCSupported() { - static constexpr std::array targets = {GL_TEXTURE_2D, GL_TEXTURE_2D_ARRAY}; - static constexpr std::array formats = { + static constexpr std::array targets{ + GL_TEXTURE_2D, + GL_TEXTURE_2D_ARRAY, + }; + static constexpr std::array formats{ GL_COMPRESSED_RGBA_ASTC_4x4_KHR, GL_COMPRESSED_RGBA_ASTC_5x4_KHR, GL_COMPRESSED_RGBA_ASTC_5x5_KHR, GL_COMPRESSED_RGBA_ASTC_6x5_KHR, GL_COMPRESSED_RGBA_ASTC_6x6_KHR, GL_COMPRESSED_RGBA_ASTC_8x5_KHR, @@ -172,11 +87,10 @@ bool IsASTCSupported() { GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR, GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR, GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x10_KHR, GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x12_KHR, }; - static constexpr std::array required_support = { + static constexpr std::array required_support{ GL_VERTEX_TEXTURE, GL_TESS_CONTROL_TEXTURE, GL_TESS_EVALUATION_TEXTURE, GL_GEOMETRY_TEXTURE, GL_FRAGMENT_TEXTURE, GL_COMPUTE_TEXTURE, }; - for (const GLenum target : targets) { for (const GLenum format : formats) { for (const GLenum support : required_support) { @@ -223,14 +137,13 @@ Device::Device() { "Beta driver 443.24 is known to have issues. There might be performance issues."); disable_fast_buffer_sub_data = true; } - max_uniform_buffers = BuildMaxUniformBuffers(); - base_bindings = BuildBaseBindings(); uniform_buffer_alignment = GetInteger<size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT); shader_storage_alignment = GetInteger<size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT); max_vertex_attributes = GetInteger<u32>(GL_MAX_VERTEX_ATTRIBS); max_varyings = GetInteger<u32>(GL_MAX_VARYING_VECTORS); max_compute_shared_memory_size = GetInteger<u32>(GL_MAX_COMPUTE_SHARED_MEMORY_SIZE); + max_glasm_storage_buffer_blocks = GetInteger<u32>(GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS); has_warp_intrinsics = GLAD_GL_NV_gpu_shader5 && GLAD_GL_NV_shader_thread_group && GLAD_GL_NV_shader_thread_shuffle; has_shader_ballot = GLAD_GL_ARB_shader_ballot; @@ -243,18 +156,30 @@ Device::Device() { has_precise_bug = TestPreciseBug(); has_broken_texture_view_formats = is_amd || (!is_linux && is_intel); has_nv_viewport_array2 = GLAD_GL_NV_viewport_array2; + has_derivative_control = GLAD_GL_ARB_derivative_control; has_vertex_buffer_unified_memory = GLAD_GL_NV_vertex_buffer_unified_memory; has_debugging_tool_attached = IsDebugToolAttached(extensions); has_depth_buffer_float = HasExtension(extensions, "GL_NV_depth_buffer_float"); + has_geometry_shader_passthrough = GLAD_GL_NV_geometry_shader_passthrough; + has_nv_gpu_shader_5 = GLAD_GL_NV_gpu_shader5; + has_shader_int64 = HasExtension(extensions, "GL_ARB_gpu_shader_int64"); + has_amd_shader_half_float = GLAD_GL_AMD_gpu_shader_half_float; + has_sparse_texture_2 = GLAD_GL_ARB_sparse_texture2; + warp_size_potentially_larger_than_guest = !is_nvidia && !is_intel; + need_fastmath_off = is_nvidia; // At the moment of writing this, only Nvidia's driver optimizes BufferSubData on exclusive // uniform buffers as "push constants" has_fast_buffer_sub_data = is_nvidia && !disable_fast_buffer_sub_data; - use_assembly_shaders = Settings::values.use_assembly_shaders.GetValue() && + shader_backend = Settings::values.shader_backend.GetValue(); + use_assembly_shaders = shader_backend == Settings::ShaderBackend::GLASM && GLAD_GL_NV_gpu_program5 && GLAD_GL_NV_compute_program5 && GLAD_GL_NV_transform_feedback && GLAD_GL_NV_transform_feedback2; - + if (shader_backend == Settings::ShaderBackend::GLASM && !use_assembly_shaders) { + LOG_ERROR(Render_OpenGL, "Assembly shaders enabled but not supported"); + shader_backend = Settings::ShaderBackend::GLSL; + } // Blocks AMD and Intel OpenGL drivers on Windows from using asynchronous shader compilation. use_asynchronous_shaders = Settings::values.use_asynchronous_shaders.GetValue() && !(is_amd || (is_intel && !is_linux)); @@ -265,11 +190,6 @@ Device::Device() { LOG_INFO(Render_OpenGL, "Renderer_PreciseBug: {}", has_precise_bug); LOG_INFO(Render_OpenGL, "Renderer_BrokenTextureViewFormats: {}", has_broken_texture_view_formats); - - if (Settings::values.use_assembly_shaders.GetValue() && !use_assembly_shaders) { - LOG_ERROR(Render_OpenGL, "Assembly shaders enabled but not supported"); - } - if (Settings::values.use_asynchronous_shaders.GetValue() && !use_asynchronous_shaders) { LOG_WARNING(Render_OpenGL, "Asynchronous shader compilation enabled but not supported"); } @@ -325,22 +245,6 @@ std::string Device::GetVendorName() const { return vendor_name; } -Device::Device(std::nullptr_t) { - max_uniform_buffers.fill(std::numeric_limits<u32>::max()); - uniform_buffer_alignment = 4; - shader_storage_alignment = 4; - max_vertex_attributes = 16; - max_varyings = 15; - max_compute_shared_memory_size = 0x10000; - has_warp_intrinsics = true; - has_shader_ballot = true; - has_vertex_viewport_layer = true; - has_image_load_formatted = true; - has_texture_shadow_lod = true; - has_variable_aoffi = true; - has_depth_buffer_float = true; -} - bool Device::TestVariableAoffi() { return TestProgram(R"(#version 430 core // This is a unit test, please ignore me on apitrace bug reports. diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h index 2c2b13767..ee992aed4 100644 --- a/src/video_core/renderer_opengl/gl_device.h +++ b/src/video_core/renderer_opengl/gl_device.h @@ -6,34 +6,22 @@ #include <cstddef> #include "common/common_types.h" -#include "video_core/engines/shader_type.h" +#include "shader_recompiler/stage.h" + +namespace Settings { +enum class ShaderBackend : u32; +}; namespace OpenGL { class Device { public: - struct BaseBindings { - u32 uniform_buffer{}; - u32 shader_storage_buffer{}; - u32 sampler{}; - u32 image{}; - }; - explicit Device(); - explicit Device(std::nullptr_t); [[nodiscard]] std::string GetVendorName() const; - u32 GetMaxUniformBuffers(Tegra::Engines::ShaderType shader_type) const noexcept { - return max_uniform_buffers[static_cast<std::size_t>(shader_type)]; - } - - const BaseBindings& GetBaseBindings(std::size_t stage_index) const noexcept { - return base_bindings[stage_index]; - } - - const BaseBindings& GetBaseBindings(Tegra::Engines::ShaderType shader_type) const noexcept { - return GetBaseBindings(static_cast<std::size_t>(shader_type)); + u32 GetMaxUniformBuffers(Shader::Stage stage) const noexcept { + return max_uniform_buffers[static_cast<size_t>(stage)]; } size_t GetUniformBufferAlignment() const { @@ -56,6 +44,10 @@ public: return max_compute_shared_memory_size; } + u32 GetMaxGLASMStorageBufferBlocks() const { + return max_glasm_storage_buffer_blocks; + } + bool HasWarpIntrinsics() const { return has_warp_intrinsics; } @@ -108,6 +100,10 @@ public: return has_nv_viewport_array2; } + bool HasDerivativeControl() const { + return has_derivative_control; + } + bool HasDebuggingToolAttached() const { return has_debugging_tool_attached; } @@ -128,18 +124,52 @@ public: return has_depth_buffer_float; } + bool HasGeometryShaderPassthrough() const { + return has_geometry_shader_passthrough; + } + + bool HasNvGpuShader5() const { + return has_nv_gpu_shader_5; + } + + bool HasShaderInt64() const { + return has_shader_int64; + } + + bool HasAmdShaderHalfFloat() const { + return has_amd_shader_half_float; + } + + bool HasSparseTexture2() const { + return has_sparse_texture_2; + } + + bool IsWarpSizePotentiallyLargerThanGuest() const { + return warp_size_potentially_larger_than_guest; + } + + bool NeedsFastmathOff() const { + return need_fastmath_off; + } + + Settings::ShaderBackend GetShaderBackend() const { + return shader_backend; + } + private: static bool TestVariableAoffi(); static bool TestPreciseBug(); - std::string vendor_name; - std::array<u32, Tegra::Engines::MaxShaderTypes> max_uniform_buffers{}; - std::array<BaseBindings, Tegra::Engines::MaxShaderTypes> base_bindings{}; + std::array<u32, Shader::MaxStageTypes> max_uniform_buffers{}; size_t uniform_buffer_alignment{}; size_t shader_storage_alignment{}; u32 max_vertex_attributes{}; u32 max_varyings{}; u32 max_compute_shared_memory_size{}; + u32 max_glasm_storage_buffer_blocks{}; + + Settings::ShaderBackend shader_backend{}; + bool has_warp_intrinsics{}; bool has_shader_ballot{}; bool has_vertex_viewport_layer{}; @@ -153,11 +183,21 @@ private: bool has_broken_texture_view_formats{}; bool has_fast_buffer_sub_data{}; bool has_nv_viewport_array2{}; + bool has_derivative_control{}; bool has_debugging_tool_attached{}; bool use_assembly_shaders{}; bool use_asynchronous_shaders{}; bool use_driver_cache{}; bool has_depth_buffer_float{}; + bool has_geometry_shader_passthrough{}; + bool has_nv_gpu_shader_5{}; + bool has_shader_int64{}; + bool has_amd_shader_half_float{}; + bool has_sparse_texture_2{}; + bool warp_size_potentially_larger_than_guest{}; + bool need_fastmath_off{}; + + std::string vendor_name; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp new file mode 100644 index 000000000..fac0034fb --- /dev/null +++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp @@ -0,0 +1,572 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <algorithm> +#include <array> +#include <string> +#include <vector> + +#include "common/settings.h" // for enum class Settings::ShaderBackend +#include "common/thread_worker.h" +#include "shader_recompiler/shader_info.h" +#include "video_core/renderer_opengl/gl_graphics_pipeline.h" +#include "video_core/renderer_opengl/gl_shader_manager.h" +#include "video_core/renderer_opengl/gl_shader_util.h" +#include "video_core/renderer_opengl/gl_state_tracker.h" +#include "video_core/shader_notify.h" +#include "video_core/texture_cache/texture_cache.h" + +#if defined(_MSC_VER) && defined(NDEBUG) +#define LAMBDA_FORCEINLINE [[msvc::forceinline]] +#else +#define LAMBDA_FORCEINLINE +#endif + +namespace OpenGL { +namespace { +using Shader::ImageBufferDescriptor; +using Shader::ImageDescriptor; +using Shader::TextureBufferDescriptor; +using Shader::TextureDescriptor; +using Tegra::Texture::TexturePair; +using VideoCommon::ImageId; + +constexpr u32 MAX_TEXTURES = 64; +constexpr u32 MAX_IMAGES = 8; + +template <typename Range> +u32 AccumulateCount(const Range& range) { + u32 num{}; + for (const auto& desc : range) { + num += desc.count; + } + return num; +} + +GLenum Stage(size_t stage_index) { + switch (stage_index) { + case 0: + return GL_VERTEX_SHADER; + case 1: + return GL_TESS_CONTROL_SHADER; + case 2: + return GL_TESS_EVALUATION_SHADER; + case 3: + return GL_GEOMETRY_SHADER; + case 4: + return GL_FRAGMENT_SHADER; + } + UNREACHABLE_MSG("{}", stage_index); + return GL_NONE; +} + +GLenum AssemblyStage(size_t stage_index) { + switch (stage_index) { + case 0: + return GL_VERTEX_PROGRAM_NV; + case 1: + return GL_TESS_CONTROL_PROGRAM_NV; + case 2: + return GL_TESS_EVALUATION_PROGRAM_NV; + case 3: + return GL_GEOMETRY_PROGRAM_NV; + case 4: + return GL_FRAGMENT_PROGRAM_NV; + } + UNREACHABLE_MSG("{}", stage_index); + return GL_NONE; +} + +/// Translates hardware transform feedback indices +/// @param location Hardware location +/// @return Pair of ARB_transform_feedback3 token stream first and third arguments +/// @note Read https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_transform_feedback3.txt +std::pair<GLint, GLint> TransformFeedbackEnum(u8 location) { + const u8 index = location / 4; + if (index >= 8 && index <= 39) { + return {GL_GENERIC_ATTRIB_NV, index - 8}; + } + if (index >= 48 && index <= 55) { + return {GL_TEXTURE_COORD_NV, index - 48}; + } + switch (index) { + case 7: + return {GL_POSITION, 0}; + case 40: + return {GL_PRIMARY_COLOR_NV, 0}; + case 41: + return {GL_SECONDARY_COLOR_NV, 0}; + case 42: + return {GL_BACK_PRIMARY_COLOR_NV, 0}; + case 43: + return {GL_BACK_SECONDARY_COLOR_NV, 0}; + } + UNIMPLEMENTED_MSG("index={}", index); + return {GL_POSITION, 0}; +} + +template <typename Spec> +bool Passes(const std::array<Shader::Info, 5>& stage_infos, u32 enabled_mask) { + for (size_t stage = 0; stage < stage_infos.size(); ++stage) { + if (!Spec::enabled_stages[stage] && ((enabled_mask >> stage) & 1) != 0) { + return false; + } + const auto& info{stage_infos[stage]}; + if constexpr (!Spec::has_storage_buffers) { + if (!info.storage_buffers_descriptors.empty()) { + return false; + } + } + if constexpr (!Spec::has_texture_buffers) { + if (!info.texture_buffer_descriptors.empty()) { + return false; + } + } + if constexpr (!Spec::has_image_buffers) { + if (!info.image_buffer_descriptors.empty()) { + return false; + } + } + if constexpr (!Spec::has_images) { + if (!info.image_descriptors.empty()) { + return false; + } + } + } + return true; +} + +using ConfigureFuncPtr = void (*)(GraphicsPipeline*, bool); + +template <typename Spec, typename... Specs> +ConfigureFuncPtr FindSpec(const std::array<Shader::Info, 5>& stage_infos, u32 enabled_mask) { + if constexpr (sizeof...(Specs) > 0) { + if (!Passes<Spec>(stage_infos, enabled_mask)) { + return FindSpec<Specs...>(stage_infos, enabled_mask); + } + } + return GraphicsPipeline::MakeConfigureSpecFunc<Spec>(); +} + +struct SimpleVertexFragmentSpec { + static constexpr std::array<bool, 5> enabled_stages{true, false, false, false, true}; + static constexpr bool has_storage_buffers = false; + static constexpr bool has_texture_buffers = false; + static constexpr bool has_image_buffers = false; + static constexpr bool has_images = false; +}; + +struct SimpleVertexSpec { + static constexpr std::array<bool, 5> enabled_stages{true, false, false, false, false}; + static constexpr bool has_storage_buffers = false; + static constexpr bool has_texture_buffers = false; + static constexpr bool has_image_buffers = false; + static constexpr bool has_images = false; +}; + +struct DefaultSpec { + static constexpr std::array<bool, 5> enabled_stages{true, true, true, true, true}; + static constexpr bool has_storage_buffers = true; + static constexpr bool has_texture_buffers = true; + static constexpr bool has_image_buffers = true; + static constexpr bool has_images = true; +}; + +ConfigureFuncPtr ConfigureFunc(const std::array<Shader::Info, 5>& infos, u32 enabled_mask) { + return FindSpec<SimpleVertexSpec, SimpleVertexFragmentSpec, DefaultSpec>(infos, enabled_mask); +} +} // Anonymous namespace + +GraphicsPipeline::GraphicsPipeline( + const Device& device, TextureCache& texture_cache_, BufferCache& buffer_cache_, + Tegra::MemoryManager& gpu_memory_, Tegra::Engines::Maxwell3D& maxwell3d_, + ProgramManager& program_manager_, StateTracker& state_tracker_, ShaderWorker* thread_worker, + VideoCore::ShaderNotify* shader_notify, std::array<std::string, 5> sources, + std::array<std::vector<u32>, 5> sources_spirv, const std::array<const Shader::Info*, 5>& infos, + const GraphicsPipelineKey& key_) + : texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, + gpu_memory{gpu_memory_}, maxwell3d{maxwell3d_}, program_manager{program_manager_}, + state_tracker{state_tracker_}, key{key_} { + if (shader_notify) { + shader_notify->MarkShaderBuilding(); + } + u32 num_textures{}; + u32 num_images{}; + u32 num_storage_buffers{}; + for (size_t stage = 0; stage < base_uniform_bindings.size(); ++stage) { + auto& info{stage_infos[stage]}; + if (infos[stage]) { + info = *infos[stage]; + enabled_stages_mask |= 1u << stage; + } + if (stage < 4) { + base_uniform_bindings[stage + 1] = base_uniform_bindings[stage]; + base_storage_bindings[stage + 1] = base_storage_bindings[stage]; + + base_uniform_bindings[stage + 1] += AccumulateCount(info.constant_buffer_descriptors); + base_storage_bindings[stage + 1] += AccumulateCount(info.storage_buffers_descriptors); + } + enabled_uniform_buffer_masks[stage] = info.constant_buffer_mask; + std::ranges::copy(info.constant_buffer_used_sizes, uniform_buffer_sizes[stage].begin()); + + const u32 num_tex_buffer_bindings{AccumulateCount(info.texture_buffer_descriptors)}; + num_texture_buffers[stage] += num_tex_buffer_bindings; + num_textures += num_tex_buffer_bindings; + + const u32 num_img_buffers_bindings{AccumulateCount(info.image_buffer_descriptors)}; + num_image_buffers[stage] += num_img_buffers_bindings; + num_images += num_img_buffers_bindings; + + num_textures += AccumulateCount(info.texture_descriptors); + num_images += AccumulateCount(info.image_descriptors); + num_storage_buffers += AccumulateCount(info.storage_buffers_descriptors); + + writes_global_memory |= std::ranges::any_of( + info.storage_buffers_descriptors, [](const auto& desc) { return desc.is_written; }); + } + ASSERT(num_textures <= MAX_TEXTURES); + ASSERT(num_images <= MAX_IMAGES); + + const bool assembly_shaders{assembly_programs[0].handle != 0}; + use_storage_buffers = + !assembly_shaders || num_storage_buffers <= device.GetMaxGLASMStorageBufferBlocks(); + writes_global_memory &= !use_storage_buffers; + configure_func = ConfigureFunc(stage_infos, enabled_stages_mask); + + if (key.xfb_enabled && device.UseAssemblyShaders()) { + GenerateTransformFeedbackState(); + } + const bool in_parallel = thread_worker != nullptr; + const auto backend = device.GetShaderBackend(); + auto func{[this, sources = std::move(sources), sources_spirv = std::move(sources_spirv), + shader_notify, backend, in_parallel](ShaderContext::Context*) mutable { + for (size_t stage = 0; stage < 5; ++stage) { + switch (backend) { + case Settings::ShaderBackend::GLSL: + if (!sources[stage].empty()) { + source_programs[stage] = CreateProgram(sources[stage], Stage(stage)); + } + break; + case Settings::ShaderBackend::GLASM: + if (!sources[stage].empty()) { + assembly_programs[stage] = CompileProgram(sources[stage], AssemblyStage(stage)); + if (in_parallel) { + // Make sure program is built before continuing when building in parallel + glGetString(GL_PROGRAM_ERROR_STRING_NV); + } + } + break; + case Settings::ShaderBackend::SPIRV: + if (!sources_spirv[stage].empty()) { + source_programs[stage] = CreateProgram(sources_spirv[stage], Stage(stage)); + } + break; + } + } + if (in_parallel && backend != Settings::ShaderBackend::GLASM) { + // Make sure programs have built if we are building shaders in parallel + for (OGLProgram& program : source_programs) { + if (program.handle != 0) { + GLint status{}; + glGetProgramiv(program.handle, GL_LINK_STATUS, &status); + } + } + } + if (shader_notify) { + shader_notify->MarkShaderComplete(); + } + is_built = true; + built_condvar.notify_one(); + }}; + if (thread_worker) { + thread_worker->QueueWork(std::move(func)); + } else { + func(nullptr); + } +} + +template <typename Spec> +void GraphicsPipeline::ConfigureImpl(bool is_indexed) { + std::array<ImageId, MAX_TEXTURES + MAX_IMAGES> image_view_ids; + std::array<u32, MAX_TEXTURES + MAX_IMAGES> image_view_indices; + std::array<GLuint, MAX_TEXTURES> samplers; + size_t image_view_index{}; + GLsizei sampler_binding{}; + + texture_cache.SynchronizeGraphicsDescriptors(); + + buffer_cache.SetUniformBuffersState(enabled_uniform_buffer_masks, &uniform_buffer_sizes); + buffer_cache.runtime.SetBaseUniformBindings(base_uniform_bindings); + buffer_cache.runtime.SetBaseStorageBindings(base_storage_bindings); + buffer_cache.runtime.SetEnableStorageBuffers(use_storage_buffers); + + const auto& regs{maxwell3d.regs}; + const bool via_header_index{regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex}; + const auto config_stage{[&](size_t stage) LAMBDA_FORCEINLINE { + const Shader::Info& info{stage_infos[stage]}; + buffer_cache.UnbindGraphicsStorageBuffers(stage); + if constexpr (Spec::has_storage_buffers) { + size_t ssbo_index{}; + for (const auto& desc : info.storage_buffers_descriptors) { + ASSERT(desc.count == 1); + buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, desc.cbuf_index, + desc.cbuf_offset, desc.is_written); + ++ssbo_index; + } + } + const auto& cbufs{maxwell3d.state.shader_stages[stage].const_buffers}; + const auto read_handle{[&](const auto& desc, u32 index) { + ASSERT(cbufs[desc.cbuf_index].enabled); + const u32 index_offset{index << desc.size_shift}; + const u32 offset{desc.cbuf_offset + index_offset}; + const GPUVAddr addr{cbufs[desc.cbuf_index].address + offset}; + if constexpr (std::is_same_v<decltype(desc), const TextureDescriptor&> || + std::is_same_v<decltype(desc), const TextureBufferDescriptor&>) { + if (desc.has_secondary) { + ASSERT(cbufs[desc.secondary_cbuf_index].enabled); + const u32 second_offset{desc.secondary_cbuf_offset + index_offset}; + const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].address + + second_offset}; + const u32 lhs_raw{gpu_memory.Read<u32>(addr)}; + const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)}; + const u32 raw{lhs_raw | rhs_raw}; + return TexturePair(raw, via_header_index); + } + } + return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); + }}; + const auto add_image{[&](const auto& desc) { + for (u32 index = 0; index < desc.count; ++index) { + const auto handle{read_handle(desc, index)}; + image_view_indices[image_view_index++] = handle.first; + } + }}; + if constexpr (Spec::has_texture_buffers) { + for (const auto& desc : info.texture_buffer_descriptors) { + for (u32 index = 0; index < desc.count; ++index) { + const auto handle{read_handle(desc, index)}; + image_view_indices[image_view_index++] = handle.first; + samplers[sampler_binding++] = 0; + } + } + } + if constexpr (Spec::has_image_buffers) { + for (const auto& desc : info.image_buffer_descriptors) { + add_image(desc); + } + } + for (const auto& desc : info.texture_descriptors) { + for (u32 index = 0; index < desc.count; ++index) { + const auto handle{read_handle(desc, index)}; + image_view_indices[image_view_index++] = handle.first; + + Sampler* const sampler{texture_cache.GetGraphicsSampler(handle.second)}; + samplers[sampler_binding++] = sampler->Handle(); + } + } + if constexpr (Spec::has_images) { + for (const auto& desc : info.image_descriptors) { + add_image(desc); + } + } + }}; + if constexpr (Spec::enabled_stages[0]) { + config_stage(0); + } + if constexpr (Spec::enabled_stages[1]) { + config_stage(1); + } + if constexpr (Spec::enabled_stages[2]) { + config_stage(2); + } + if constexpr (Spec::enabled_stages[3]) { + config_stage(3); + } + if constexpr (Spec::enabled_stages[4]) { + config_stage(4); + } + const std::span indices_span(image_view_indices.data(), image_view_index); + texture_cache.FillGraphicsImageViews(indices_span, image_view_ids); + + texture_cache.UpdateRenderTargets(false); + state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle()); + + ImageId* texture_buffer_index{image_view_ids.data()}; + const auto bind_stage_info{[&](size_t stage) LAMBDA_FORCEINLINE { + size_t index{}; + const auto add_buffer{[&](const auto& desc) { + constexpr bool is_image = std::is_same_v<decltype(desc), const ImageBufferDescriptor&>; + for (u32 i = 0; i < desc.count; ++i) { + bool is_written{false}; + if constexpr (is_image) { + is_written = desc.is_written; + } + ImageView& image_view{texture_cache.GetImageView(*texture_buffer_index)}; + buffer_cache.BindGraphicsTextureBuffer(stage, index, image_view.GpuAddr(), + image_view.BufferSize(), image_view.format, + is_written, is_image); + ++index; + ++texture_buffer_index; + } + }}; + const Shader::Info& info{stage_infos[stage]}; + buffer_cache.UnbindGraphicsTextureBuffers(stage); + + if constexpr (Spec::has_texture_buffers) { + for (const auto& desc : info.texture_buffer_descriptors) { + add_buffer(desc); + } + } + if constexpr (Spec::has_image_buffers) { + for (const auto& desc : info.image_buffer_descriptors) { + add_buffer(desc); + } + } + for (const auto& desc : info.texture_descriptors) { + texture_buffer_index += desc.count; + } + if constexpr (Spec::has_images) { + for (const auto& desc : info.image_descriptors) { + texture_buffer_index += desc.count; + } + } + }}; + if constexpr (Spec::enabled_stages[0]) { + bind_stage_info(0); + } + if constexpr (Spec::enabled_stages[1]) { + bind_stage_info(1); + } + if constexpr (Spec::enabled_stages[2]) { + bind_stage_info(2); + } + if constexpr (Spec::enabled_stages[3]) { + bind_stage_info(3); + } + if constexpr (Spec::enabled_stages[4]) { + bind_stage_info(4); + } + buffer_cache.UpdateGraphicsBuffers(is_indexed); + buffer_cache.BindHostGeometryBuffers(is_indexed); + + if (!is_built.load(std::memory_order::relaxed)) { + WaitForBuild(); + } + if (assembly_programs[0].handle != 0) { + program_manager.BindAssemblyPrograms(assembly_programs, enabled_stages_mask); + } else { + program_manager.BindSourcePrograms(source_programs); + } + const ImageId* views_it{image_view_ids.data()}; + GLsizei texture_binding = 0; + GLsizei image_binding = 0; + std::array<GLuint, MAX_TEXTURES> textures; + std::array<GLuint, MAX_IMAGES> images; + const auto prepare_stage{[&](size_t stage) { + buffer_cache.runtime.SetImagePointers(&textures[texture_binding], &images[image_binding]); + buffer_cache.BindHostStageBuffers(stage); + + texture_binding += num_texture_buffers[stage]; + image_binding += num_image_buffers[stage]; + + views_it += num_texture_buffers[stage]; + views_it += num_image_buffers[stage]; + + const auto& info{stage_infos[stage]}; + for (const auto& desc : info.texture_descriptors) { + for (u32 index = 0; index < desc.count; ++index) { + ImageView& image_view{texture_cache.GetImageView(*(views_it++))}; + textures[texture_binding++] = image_view.Handle(desc.type); + } + } + for (const auto& desc : info.image_descriptors) { + for (u32 index = 0; index < desc.count; ++index) { + ImageView& image_view{texture_cache.GetImageView(*(views_it++))}; + if (desc.is_written) { + texture_cache.MarkModification(image_view.image_id); + } + images[image_binding++] = image_view.StorageView(desc.type, desc.format); + } + } + }}; + if constexpr (Spec::enabled_stages[0]) { + prepare_stage(0); + } + if constexpr (Spec::enabled_stages[1]) { + prepare_stage(1); + } + if constexpr (Spec::enabled_stages[2]) { + prepare_stage(2); + } + if constexpr (Spec::enabled_stages[3]) { + prepare_stage(3); + } + if constexpr (Spec::enabled_stages[4]) { + prepare_stage(4); + } + if (texture_binding != 0) { + ASSERT(texture_binding == sampler_binding); + glBindTextures(0, texture_binding, textures.data()); + glBindSamplers(0, sampler_binding, samplers.data()); + } + if (image_binding != 0) { + glBindImageTextures(0, image_binding, images.data()); + } +} + +void GraphicsPipeline::ConfigureTransformFeedbackImpl() const { + glTransformFeedbackStreamAttribsNV(num_xfb_attribs, xfb_attribs.data(), num_xfb_strides, + xfb_streams.data(), GL_INTERLEAVED_ATTRIBS); +} + +void GraphicsPipeline::GenerateTransformFeedbackState() { + // TODO(Rodrigo): Inject SKIP_COMPONENTS*_NV when required. An unimplemented message will signal + // when this is required. + GLint* cursor{xfb_attribs.data()}; + GLint* current_stream{xfb_streams.data()}; + + for (size_t feedback = 0; feedback < Maxwell::NumTransformFeedbackBuffers; ++feedback) { + const auto& layout = key.xfb_state.layouts[feedback]; + UNIMPLEMENTED_IF_MSG(layout.stride != layout.varying_count * 4, "Stride padding"); + if (layout.varying_count == 0) { + continue; + } + *current_stream = static_cast<GLint>(feedback); + if (current_stream != xfb_streams.data()) { + // When stepping one stream, push the expected token + cursor[0] = GL_NEXT_BUFFER_NV; + cursor[1] = 0; + cursor[2] = 0; + cursor += XFB_ENTRY_STRIDE; + } + ++current_stream; + + const auto& locations = key.xfb_state.varyings[feedback]; + std::optional<u8> current_index; + for (u32 offset = 0; offset < layout.varying_count; ++offset) { + const u8 location = locations[offset]; + const u8 index = location / 4; + + if (current_index == index) { + // Increase number of components of the previous attachment + ++cursor[-2]; + continue; + } + current_index = index; + + std::tie(cursor[0], cursor[2]) = TransformFeedbackEnum(location); + cursor[1] = 1; + cursor += XFB_ENTRY_STRIDE; + } + } + num_xfb_attribs = static_cast<GLsizei>((cursor - xfb_attribs.data()) / XFB_ENTRY_STRIDE); + num_xfb_strides = static_cast<GLsizei>(current_stream - xfb_streams.data()); +} + +void GraphicsPipeline::WaitForBuild() { + std::unique_lock lock{built_mutex}; + built_condvar.wait(lock, [this] { return is_built.load(std::memory_order::relaxed); }); +} + +} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.h b/src/video_core/renderer_opengl/gl_graphics_pipeline.h new file mode 100644 index 000000000..4e28d9a42 --- /dev/null +++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.h @@ -0,0 +1,169 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <array> +#include <cstring> +#include <type_traits> +#include <utility> + +#include "common/bit_field.h" +#include "common/cityhash.h" +#include "common/common_types.h" +#include "shader_recompiler/shader_info.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/memory_manager.h" +#include "video_core/renderer_opengl/gl_buffer_cache.h" +#include "video_core/renderer_opengl/gl_resource_manager.h" +#include "video_core/renderer_opengl/gl_texture_cache.h" +#include "video_core/transform_feedback.h" + +namespace OpenGL { + +namespace ShaderContext { +struct Context; +} + +class Device; +class ProgramManager; + +using Maxwell = Tegra::Engines::Maxwell3D::Regs; +using ShaderWorker = Common::StatefulThreadWorker<ShaderContext::Context>; + +struct GraphicsPipelineKey { + std::array<u64, 6> unique_hashes; + union { + u32 raw; + BitField<0, 1, u32> xfb_enabled; + BitField<1, 1, u32> early_z; + BitField<2, 4, Maxwell::PrimitiveTopology> gs_input_topology; + BitField<6, 2, Maxwell::TessellationPrimitive> tessellation_primitive; + BitField<8, 2, Maxwell::TessellationSpacing> tessellation_spacing; + BitField<10, 1, u32> tessellation_clockwise; + }; + std::array<u32, 3> padding; + VideoCommon::TransformFeedbackState xfb_state; + + size_t Hash() const noexcept { + return static_cast<size_t>(Common::CityHash64(reinterpret_cast<const char*>(this), Size())); + } + + bool operator==(const GraphicsPipelineKey& rhs) const noexcept { + return std::memcmp(this, &rhs, Size()) == 0; + } + + bool operator!=(const GraphicsPipelineKey& rhs) const noexcept { + return !operator==(rhs); + } + + [[nodiscard]] size_t Size() const noexcept { + if (xfb_enabled != 0) { + return sizeof(GraphicsPipelineKey); + } else { + return offsetof(GraphicsPipelineKey, padding); + } + } +}; +static_assert(std::has_unique_object_representations_v<GraphicsPipelineKey>); +static_assert(std::is_trivially_copyable_v<GraphicsPipelineKey>); +static_assert(std::is_trivially_constructible_v<GraphicsPipelineKey>); + +class GraphicsPipeline { +public: + explicit GraphicsPipeline(const Device& device, TextureCache& texture_cache_, + BufferCache& buffer_cache_, Tegra::MemoryManager& gpu_memory_, + Tegra::Engines::Maxwell3D& maxwell3d_, + ProgramManager& program_manager_, StateTracker& state_tracker_, + ShaderWorker* thread_worker, VideoCore::ShaderNotify* shader_notify, + std::array<std::string, 5> sources, + std::array<std::vector<u32>, 5> sources_spirv, + const std::array<const Shader::Info*, 5>& infos, + const GraphicsPipelineKey& key_); + + void Configure(bool is_indexed) { + configure_func(this, is_indexed); + } + + void ConfigureTransformFeedback() const { + if (num_xfb_attribs != 0) { + ConfigureTransformFeedbackImpl(); + } + } + + [[nodiscard]] const GraphicsPipelineKey& Key() const noexcept { + return key; + } + + [[nodiscard]] bool WritesGlobalMemory() const noexcept { + return writes_global_memory; + } + + [[nodiscard]] bool IsBuilt() const noexcept { + return is_built.load(std::memory_order::relaxed); + } + + template <typename Spec> + static auto MakeConfigureSpecFunc() { + return [](GraphicsPipeline* pipeline, bool is_indexed) { + pipeline->ConfigureImpl<Spec>(is_indexed); + }; + } + +private: + template <typename Spec> + void ConfigureImpl(bool is_indexed); + + void ConfigureTransformFeedbackImpl() const; + + void GenerateTransformFeedbackState(); + + void WaitForBuild(); + + TextureCache& texture_cache; + BufferCache& buffer_cache; + Tegra::MemoryManager& gpu_memory; + Tegra::Engines::Maxwell3D& maxwell3d; + ProgramManager& program_manager; + StateTracker& state_tracker; + const GraphicsPipelineKey key; + + void (*configure_func)(GraphicsPipeline*, bool){}; + + std::array<OGLProgram, 5> source_programs; + std::array<OGLAssemblyProgram, 5> assembly_programs; + u32 enabled_stages_mask{}; + + std::array<Shader::Info, 5> stage_infos{}; + std::array<u32, 5> enabled_uniform_buffer_masks{}; + VideoCommon::UniformBufferSizes uniform_buffer_sizes{}; + std::array<u32, 5> base_uniform_bindings{}; + std::array<u32, 5> base_storage_bindings{}; + std::array<u32, 5> num_texture_buffers{}; + std::array<u32, 5> num_image_buffers{}; + + bool use_storage_buffers{}; + bool writes_global_memory{}; + + static constexpr std::size_t XFB_ENTRY_STRIDE = 3; + GLsizei num_xfb_attribs{}; + GLsizei num_xfb_strides{}; + std::array<GLint, 128 * XFB_ENTRY_STRIDE * Maxwell::NumTransformFeedbackBuffers> xfb_attribs{}; + std::array<GLint, Maxwell::NumTransformFeedbackBuffers> xfb_streams{}; + + std::mutex built_mutex; + std::condition_variable built_condvar; + std::atomic_bool is_built{false}; +}; + +} // namespace OpenGL + +namespace std { +template <> +struct hash<OpenGL::GraphicsPipelineKey> { + size_t operator()(const OpenGL::GraphicsPipelineKey& k) const noexcept { + return k.Hash(); + } +}; +} // namespace std diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 07ad0e205..41d2b73f4 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -23,7 +23,6 @@ #include "core/memory.h" #include "video_core/engines/kepler_compute.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/shader_type.h" #include "video_core/memory_manager.h" #include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_query_cache.h" @@ -40,7 +39,6 @@ namespace OpenGL { using Maxwell = Tegra::Engines::Maxwell3D::Regs; using GLvec4 = std::array<GLfloat, 4>; -using Tegra::Engines::ShaderType; using VideoCore::Surface::PixelFormat; using VideoCore::Surface::SurfaceTarget; using VideoCore::Surface::SurfaceType; @@ -51,112 +49,11 @@ MICROPROFILE_DEFINE(OpenGL_Blits, "OpenGL", "Blits", MP_RGB(128, 128, 192)); MICROPROFILE_DEFINE(OpenGL_CacheManagement, "OpenGL", "Cache Management", MP_RGB(100, 255, 100)); namespace { - constexpr size_t NUM_SUPPORTED_VERTEX_ATTRIBUTES = 16; -struct TextureHandle { - constexpr TextureHandle(u32 data, bool via_header_index) { - const Tegra::Texture::TextureHandle handle{data}; - image = handle.tic_id; - sampler = via_header_index ? image : handle.tsc_id.Value(); - } - - u32 image; - u32 sampler; -}; - -template <typename Engine, typename Entry> -TextureHandle GetTextureInfo(const Engine& engine, bool via_header_index, const Entry& entry, - ShaderType shader_type, size_t index = 0) { - if constexpr (std::is_same_v<Entry, SamplerEntry>) { - if (entry.is_separated) { - const u32 buffer_1 = entry.buffer; - const u32 buffer_2 = entry.secondary_buffer; - const u32 offset_1 = entry.offset; - const u32 offset_2 = entry.secondary_offset; - const u32 handle_1 = engine.AccessConstBuffer32(shader_type, buffer_1, offset_1); - const u32 handle_2 = engine.AccessConstBuffer32(shader_type, buffer_2, offset_2); - return TextureHandle(handle_1 | handle_2, via_header_index); - } - } - if (entry.is_bindless) { - const u32 raw = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset); - return TextureHandle(raw, via_header_index); - } - const u32 buffer = engine.GetBoundBuffer(); - const u64 offset = (entry.offset + index) * sizeof(u32); - return TextureHandle(engine.AccessConstBuffer32(shader_type, buffer, offset), via_header_index); -} - -/// Translates hardware transform feedback indices -/// @param location Hardware location -/// @return Pair of ARB_transform_feedback3 token stream first and third arguments -/// @note Read https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_transform_feedback3.txt -std::pair<GLint, GLint> TransformFeedbackEnum(u8 location) { - const u8 index = location / 4; - if (index >= 8 && index <= 39) { - return {GL_GENERIC_ATTRIB_NV, index - 8}; - } - if (index >= 48 && index <= 55) { - return {GL_TEXTURE_COORD_NV, index - 48}; - } - switch (index) { - case 7: - return {GL_POSITION, 0}; - case 40: - return {GL_PRIMARY_COLOR_NV, 0}; - case 41: - return {GL_SECONDARY_COLOR_NV, 0}; - case 42: - return {GL_BACK_PRIMARY_COLOR_NV, 0}; - case 43: - return {GL_BACK_SECONDARY_COLOR_NV, 0}; - } - UNIMPLEMENTED_MSG("index={}", index); - return {GL_POSITION, 0}; -} - void oglEnable(GLenum cap, bool state) { (state ? glEnable : glDisable)(cap); } - -ImageViewType ImageViewTypeFromEntry(const SamplerEntry& entry) { - if (entry.is_buffer) { - return ImageViewType::Buffer; - } - switch (entry.type) { - case Tegra::Shader::TextureType::Texture1D: - return entry.is_array ? ImageViewType::e1DArray : ImageViewType::e1D; - case Tegra::Shader::TextureType::Texture2D: - return entry.is_array ? ImageViewType::e2DArray : ImageViewType::e2D; - case Tegra::Shader::TextureType::Texture3D: - return ImageViewType::e3D; - case Tegra::Shader::TextureType::TextureCube: - return entry.is_array ? ImageViewType::CubeArray : ImageViewType::Cube; - } - UNREACHABLE(); - return ImageViewType::e2D; -} - -ImageViewType ImageViewTypeFromEntry(const ImageEntry& entry) { - switch (entry.type) { - case Tegra::Shader::ImageType::Texture1D: - return ImageViewType::e1D; - case Tegra::Shader::ImageType::Texture1DArray: - return ImageViewType::e1DArray; - case Tegra::Shader::ImageType::Texture2D: - return ImageViewType::e2D; - case Tegra::Shader::ImageType::Texture2DArray: - return ImageViewType::e2DArray; - case Tegra::Shader::ImageType::Texture3D: - return ImageViewType::e3D; - case Tegra::Shader::ImageType::TextureBuffer: - return ImageViewType::Buffer; - } - UNREACHABLE(); - return ImageViewType::e2D; -} - } // Anonymous namespace RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, @@ -170,14 +67,10 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory), buffer_cache_runtime(device), buffer_cache(*this, maxwell3d, kepler_compute, gpu_memory, cpu_memory_, buffer_cache_runtime), - shader_cache(*this, emu_window_, gpu, maxwell3d, kepler_compute, gpu_memory, device), - query_cache(*this, maxwell3d, gpu_memory), - fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache), - async_shaders(emu_window_) { - if (device.UseAsynchronousShaders()) { - async_shaders.AllocateWorkers(); - } -} + shader_cache(*this, emu_window_, maxwell3d, kepler_compute, gpu_memory, device, texture_cache, + buffer_cache, program_manager, state_tracker, gpu.ShaderNotify()), + query_cache(*this, maxwell3d, gpu_memory), accelerate_dma(buffer_cache), + fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache) {} RasterizerOpenGL::~RasterizerOpenGL() = default; @@ -204,7 +97,7 @@ void RasterizerOpenGL::SyncVertexFormats() { const auto gl_index = static_cast<GLuint>(index); // Disable constant attributes. - if (attrib.IsConstant()) { + if (attrib.constant) { glDisableVertexAttribArray(gl_index); continue; } @@ -244,116 +137,9 @@ void RasterizerOpenGL::SyncVertexInstances() { } } -void RasterizerOpenGL::SetupShaders(bool is_indexed) { - u32 clip_distances = 0; - - std::array<Shader*, Maxwell::MaxShaderStage> shaders{}; - image_view_indices.clear(); - sampler_handles.clear(); - - texture_cache.SynchronizeGraphicsDescriptors(); - - for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { - const auto& shader_config = maxwell3d.regs.shader_config[index]; - const auto program{static_cast<Maxwell::ShaderProgram>(index)}; - - // Skip stages that are not enabled - if (!maxwell3d.regs.IsShaderConfigEnabled(index)) { - switch (program) { - case Maxwell::ShaderProgram::Geometry: - program_manager.UseGeometryShader(0); - break; - case Maxwell::ShaderProgram::Fragment: - program_manager.UseFragmentShader(0); - break; - default: - break; - } - continue; - } - // Currently this stages are not supported in the OpenGL backend. - // TODO(Blinkhawk): Port tesselation shaders from Vulkan to OpenGL - if (program == Maxwell::ShaderProgram::TesselationControl || - program == Maxwell::ShaderProgram::TesselationEval) { - continue; - } - - Shader* const shader = shader_cache.GetStageProgram(program, async_shaders); - const GLuint program_handle = shader->IsBuilt() ? shader->GetHandle() : 0; - switch (program) { - case Maxwell::ShaderProgram::VertexA: - case Maxwell::ShaderProgram::VertexB: - program_manager.UseVertexShader(program_handle); - break; - case Maxwell::ShaderProgram::Geometry: - program_manager.UseGeometryShader(program_handle); - break; - case Maxwell::ShaderProgram::Fragment: - program_manager.UseFragmentShader(program_handle); - break; - default: - UNIMPLEMENTED_MSG("Unimplemented shader index={}, enable={}, offset=0x{:08X}", index, - shader_config.enable.Value(), shader_config.offset); - break; - } - - // Stage indices are 0 - 5 - const size_t stage = index == 0 ? 0 : index - 1; - shaders[stage] = shader; - - SetupDrawTextures(shader, stage); - SetupDrawImages(shader, stage); - - buffer_cache.SetEnabledUniformBuffers(stage, shader->GetEntries().enabled_uniform_buffers); - - buffer_cache.UnbindGraphicsStorageBuffers(stage); - u32 ssbo_index = 0; - for (const auto& buffer : shader->GetEntries().global_memory_entries) { - buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, buffer.cbuf_index, - buffer.cbuf_offset, buffer.is_written); - ++ssbo_index; - } - - // Workaround for Intel drivers. - // When a clip distance is enabled but not set in the shader it crops parts of the screen - // (sometimes it's half the screen, sometimes three quarters). To avoid this, enable the - // clip distances only when it's written by a shader stage. - clip_distances |= shader->GetEntries().clip_distances; - - // When VertexA is enabled, we have dual vertex shaders - if (program == Maxwell::ShaderProgram::VertexA) { - // VertexB was combined with VertexA, so we skip the VertexB iteration - ++index; - } - } - SyncClipEnabled(clip_distances); - maxwell3d.dirty.flags[Dirty::Shaders] = false; - - buffer_cache.UpdateGraphicsBuffers(is_indexed); - - const std::span indices_span(image_view_indices.data(), image_view_indices.size()); - texture_cache.FillGraphicsImageViews(indices_span, image_view_ids); - - buffer_cache.BindHostGeometryBuffers(is_indexed); - - size_t image_view_index = 0; - size_t texture_index = 0; - size_t image_index = 0; - for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) { - const Shader* const shader = shaders[stage]; - if (!shader) { - continue; - } - buffer_cache.BindHostStageBuffers(stage); - const auto& base = device.GetBaseBindings(stage); - BindTextures(shader->GetEntries(), base.sampler, base.image, image_view_index, - texture_index, image_index); - } -} - void RasterizerOpenGL::LoadDiskResources(u64 title_id, std::stop_token stop_loading, const VideoCore::DiskResourceLoadCallback& callback) { - shader_cache.LoadDiskCache(title_id, stop_loading, callback); + shader_cache.LoadDiskResources(title_id, stop_loading, callback); } void RasterizerOpenGL::Clear() { @@ -432,16 +218,15 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { SyncState(); - // Setup shaders and their used resources. + GraphicsPipeline* const pipeline{shader_cache.CurrentGraphicsPipeline()}; + if (!pipeline) { + return; + } std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - SetupShaders(is_indexed); - - texture_cache.UpdateRenderTargets(false); - state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle()); - program_manager.BindGraphicsPipeline(); + pipeline->Configure(is_indexed); const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d.regs.draw.topology); - BeginTransformFeedback(primitive_mode); + BeginTransformFeedback(pipeline, primitive_mode); const GLuint base_instance = static_cast<GLuint>(maxwell3d.regs.vb_base_instance); const GLsizei num_instances = @@ -480,35 +265,24 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { num_instances, base_instance); } } - EndTransformFeedback(); ++num_queued_commands; + has_written_global_memory |= pipeline->WritesGlobalMemory(); gpu.TickWork(); } -void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { - Shader* const kernel = shader_cache.GetComputeKernel(code_addr); - - std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - BindComputeTextures(kernel); - - const auto& entries = kernel->GetEntries(); - buffer_cache.SetEnabledComputeUniformBuffers(entries.enabled_uniform_buffers); - buffer_cache.UnbindComputeStorageBuffers(); - u32 ssbo_index = 0; - for (const auto& buffer : entries.global_memory_entries) { - buffer_cache.BindComputeStorageBuffer(ssbo_index, buffer.cbuf_index, buffer.cbuf_offset, - buffer.is_written); - ++ssbo_index; - } - buffer_cache.UpdateComputeBuffers(); - buffer_cache.BindHostComputeBuffers(); - - const auto& launch_desc = kepler_compute.launch_description; - glDispatchCompute(launch_desc.grid_dim_x, launch_desc.grid_dim_y, launch_desc.grid_dim_z); +void RasterizerOpenGL::DispatchCompute() { + ComputePipeline* const pipeline{shader_cache.CurrentComputePipeline()}; + if (!pipeline) { + return; + } + pipeline->Configure(); + const auto& qmd{kepler_compute.launch_description}; + glDispatchCompute(qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z); ++num_queued_commands; + has_written_global_memory |= pipeline->WritesGlobalMemory(); } void RasterizerOpenGL::ResetCounter(VideoCore::QueryType type) { @@ -634,6 +408,13 @@ void RasterizerOpenGL::SignalSyncPoint(u32 value) { fence_manager.SignalSyncPoint(value); } +void RasterizerOpenGL::SignalReference() { + if (!gpu.IsAsync()) { + return; + } + fence_manager.SignalOrdering(); +} + void RasterizerOpenGL::ReleaseFences() { if (!gpu.IsAsync()) { return; @@ -650,10 +431,11 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) { void RasterizerOpenGL::WaitForIdle() { glMemoryBarrier(GL_ALL_BARRIER_BITS); + SignalReference(); } void RasterizerOpenGL::FragmentBarrier() { - glMemoryBarrier(GL_FRAMEBUFFER_BARRIER_BIT); + glMemoryBarrier(GL_FRAMEBUFFER_BARRIER_BIT | GL_TEXTURE_FETCH_BARRIER_BIT); } void RasterizerOpenGL::TiledCacheBarrier() { @@ -666,6 +448,13 @@ void RasterizerOpenGL::FlushCommands() { return; } num_queued_commands = 0; + + // Make sure memory stored from the previous GL command stream is visible + // This is only needed on assembly shaders where we write to GPU memory with raw pointers + if (has_written_global_memory) { + has_written_global_memory = false; + glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT); + } glFlush(); } @@ -693,6 +482,10 @@ bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surf return true; } +Tegra::Engines::AccelerateDMAInterface& RasterizerOpenGL::AccessAccelerateDMA() { + return accelerate_dma; +} + bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) { if (framebuffer_addr == 0) { @@ -709,111 +502,11 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config, // ASSERT_MSG(image_view->size.width == config.width, "Framebuffer width is different"); // ASSERT_MSG(image_view->size.height == config.height, "Framebuffer height is different"); - screen_info.display_texture = image_view->Handle(ImageViewType::e2D); + screen_info.display_texture = image_view->Handle(Shader::TextureType::Color2D); screen_info.display_srgb = VideoCore::Surface::IsPixelFormatSRGB(image_view->format); return true; } -void RasterizerOpenGL::BindComputeTextures(Shader* kernel) { - image_view_indices.clear(); - sampler_handles.clear(); - - texture_cache.SynchronizeComputeDescriptors(); - - SetupComputeTextures(kernel); - SetupComputeImages(kernel); - - const std::span indices_span(image_view_indices.data(), image_view_indices.size()); - texture_cache.FillComputeImageViews(indices_span, image_view_ids); - - program_manager.BindCompute(kernel->GetHandle()); - size_t image_view_index = 0; - size_t texture_index = 0; - size_t image_index = 0; - BindTextures(kernel->GetEntries(), 0, 0, image_view_index, texture_index, image_index); -} - -void RasterizerOpenGL::BindTextures(const ShaderEntries& entries, GLuint base_texture, - GLuint base_image, size_t& image_view_index, - size_t& texture_index, size_t& image_index) { - const GLuint* const samplers = sampler_handles.data() + texture_index; - const GLuint* const textures = texture_handles.data() + texture_index; - const GLuint* const images = image_handles.data() + image_index; - - const size_t num_samplers = entries.samplers.size(); - for (const auto& sampler : entries.samplers) { - for (size_t i = 0; i < sampler.size; ++i) { - const ImageViewId image_view_id = image_view_ids[image_view_index++]; - const ImageView& image_view = texture_cache.GetImageView(image_view_id); - const GLuint handle = image_view.Handle(ImageViewTypeFromEntry(sampler)); - texture_handles[texture_index++] = handle; - } - } - const size_t num_images = entries.images.size(); - for (size_t unit = 0; unit < num_images; ++unit) { - // TODO: Mark as modified - const ImageViewId image_view_id = image_view_ids[image_view_index++]; - const ImageView& image_view = texture_cache.GetImageView(image_view_id); - const GLuint handle = image_view.Handle(ImageViewTypeFromEntry(entries.images[unit])); - image_handles[image_index] = handle; - ++image_index; - } - if (num_samplers > 0) { - glBindSamplers(base_texture, static_cast<GLsizei>(num_samplers), samplers); - glBindTextures(base_texture, static_cast<GLsizei>(num_samplers), textures); - } - if (num_images > 0) { - glBindImageTextures(base_image, static_cast<GLsizei>(num_images), images); - } -} - -void RasterizerOpenGL::SetupDrawTextures(const Shader* shader, size_t stage_index) { - const bool via_header_index = - maxwell3d.regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; - for (const auto& entry : shader->GetEntries().samplers) { - const auto shader_type = static_cast<ShaderType>(stage_index); - for (size_t index = 0; index < entry.size; ++index) { - const auto handle = - GetTextureInfo(maxwell3d, via_header_index, entry, shader_type, index); - const Sampler* const sampler = texture_cache.GetGraphicsSampler(handle.sampler); - sampler_handles.push_back(sampler->Handle()); - image_view_indices.push_back(handle.image); - } - } -} - -void RasterizerOpenGL::SetupComputeTextures(const Shader* kernel) { - const bool via_header_index = kepler_compute.launch_description.linked_tsc; - for (const auto& entry : kernel->GetEntries().samplers) { - for (size_t i = 0; i < entry.size; ++i) { - const auto handle = - GetTextureInfo(kepler_compute, via_header_index, entry, ShaderType::Compute, i); - const Sampler* const sampler = texture_cache.GetComputeSampler(handle.sampler); - sampler_handles.push_back(sampler->Handle()); - image_view_indices.push_back(handle.image); - } - } -} - -void RasterizerOpenGL::SetupDrawImages(const Shader* shader, size_t stage_index) { - const bool via_header_index = - maxwell3d.regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; - for (const auto& entry : shader->GetEntries().images) { - const auto shader_type = static_cast<ShaderType>(stage_index); - const auto handle = GetTextureInfo(maxwell3d, via_header_index, entry, shader_type); - image_view_indices.push_back(handle.image); - } -} - -void RasterizerOpenGL::SetupComputeImages(const Shader* shader) { - const bool via_header_index = kepler_compute.launch_description.linked_tsc; - for (const auto& entry : shader->GetEntries().images) { - const auto handle = - GetTextureInfo(kepler_compute, via_header_index, entry, ShaderType::Compute); - image_view_indices.push_back(handle.image); - } -} - void RasterizerOpenGL::SyncState() { SyncViewport(); SyncRasterizeEnable(); @@ -929,7 +622,7 @@ void RasterizerOpenGL::SyncDepthClamp() { void RasterizerOpenGL::SyncClipEnabled(u32 clip_mask) { auto& flags = maxwell3d.dirty.flags; - if (!flags[Dirty::ClipDistances] && !flags[Dirty::Shaders]) { + if (!flags[Dirty::ClipDistances] && !flags[VideoCommon::Dirty::Shaders]) { return; } flags[Dirty::ClipDistances] = false; @@ -1306,68 +999,13 @@ void RasterizerOpenGL::SyncFramebufferSRGB() { oglEnable(GL_FRAMEBUFFER_SRGB, maxwell3d.regs.framebuffer_srgb); } -void RasterizerOpenGL::SyncTransformFeedback() { - // TODO(Rodrigo): Inject SKIP_COMPONENTS*_NV when required. An unimplemented message will signal - // when this is required. - const auto& regs = maxwell3d.regs; - - static constexpr std::size_t STRIDE = 3; - std::array<GLint, 128 * STRIDE * Maxwell::NumTransformFeedbackBuffers> attribs; - std::array<GLint, Maxwell::NumTransformFeedbackBuffers> streams; - - GLint* cursor = attribs.data(); - GLint* current_stream = streams.data(); - - for (std::size_t feedback = 0; feedback < Maxwell::NumTransformFeedbackBuffers; ++feedback) { - const auto& layout = regs.tfb_layouts[feedback]; - UNIMPLEMENTED_IF_MSG(layout.stride != layout.varying_count * 4, "Stride padding"); - if (layout.varying_count == 0) { - continue; - } - - *current_stream = static_cast<GLint>(feedback); - if (current_stream != streams.data()) { - // When stepping one stream, push the expected token - cursor[0] = GL_NEXT_BUFFER_NV; - cursor[1] = 0; - cursor[2] = 0; - cursor += STRIDE; - } - ++current_stream; - - const auto& locations = regs.tfb_varying_locs[feedback]; - std::optional<u8> current_index; - for (u32 offset = 0; offset < layout.varying_count; ++offset) { - const u8 location = locations[offset]; - const u8 index = location / 4; - - if (current_index == index) { - // Increase number of components of the previous attachment - ++cursor[-2]; - continue; - } - current_index = index; - - std::tie(cursor[0], cursor[2]) = TransformFeedbackEnum(location); - cursor[1] = 1; - cursor += STRIDE; - } - } - - const GLsizei num_attribs = static_cast<GLsizei>((cursor - attribs.data()) / STRIDE); - const GLsizei num_strides = static_cast<GLsizei>(current_stream - streams.data()); - glTransformFeedbackStreamAttribsNV(num_attribs, attribs.data(), num_strides, streams.data(), - GL_INTERLEAVED_ATTRIBS); -} - -void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) { +void RasterizerOpenGL::BeginTransformFeedback(GraphicsPipeline* program, GLenum primitive_mode) { const auto& regs = maxwell3d.regs; if (regs.tfb_enabled == 0) { return; } - if (device.UseAssemblyShaders()) { - SyncTransformFeedback(); - } + program->ConfigureTransformFeedback(); + UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) || regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) || regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::Geometry)); @@ -1381,11 +1019,21 @@ void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) { } void RasterizerOpenGL::EndTransformFeedback() { - const auto& regs = maxwell3d.regs; - if (regs.tfb_enabled == 0) { - return; + if (maxwell3d.regs.tfb_enabled != 0) { + glEndTransformFeedback(); } - glEndTransformFeedback(); +} + +AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_) : buffer_cache{buffer_cache_} {} + +bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) { + std::scoped_lock lock{buffer_cache.mutex}; + return buffer_cache.DMACopy(src_address, dest_address, amount); +} + +bool AccelerateDMA::BufferClear(GPUVAddr src_address, u64 amount, u32 value) { + std::scoped_lock lock{buffer_cache.mutex}; + return buffer_cache.DMAClear(src_address, amount, value); } } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index 482efed7a..d0397b745 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -19,6 +19,7 @@ #include "common/common_types.h" #include "video_core/engines/const_buffer_info.h" #include "video_core/engines/maxwell_3d.h" +#include "video_core/engines/maxwell_dma.h" #include "video_core/rasterizer_accelerated.h" #include "video_core/rasterizer_interface.h" #include "video_core/renderer_opengl/gl_buffer_cache.h" @@ -27,11 +28,9 @@ #include "video_core/renderer_opengl/gl_query_cache.h" #include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/gl_shader_cache.h" -#include "video_core/renderer_opengl/gl_shader_decompiler.h" #include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/gl_state_tracker.h" #include "video_core/renderer_opengl/gl_texture_cache.h" -#include "video_core/shader/async_shaders.h" #include "video_core/textures/texture.h" namespace Core::Memory { @@ -58,6 +57,18 @@ struct BindlessSSBO { }; static_assert(sizeof(BindlessSSBO) * CHAR_BIT == 128); +class AccelerateDMA : public Tegra::Engines::AccelerateDMAInterface { +public: + explicit AccelerateDMA(BufferCache& buffer_cache); + + bool BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) override; + + bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) override; + +private: + BufferCache& buffer_cache; +}; + class RasterizerOpenGL : public VideoCore::RasterizerAccelerated { public: explicit RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, @@ -68,7 +79,7 @@ public: void Draw(bool is_indexed, bool is_instanced) override; void Clear() override; - void DispatchCompute(GPUVAddr code_addr) override; + void DispatchCompute() override; void ResetCounter(VideoCore::QueryType type) override; void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override; void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override; @@ -83,6 +94,7 @@ public: void ModifyGPUMemory(GPUVAddr addr, u64 size) override; void SignalSemaphore(GPUVAddr addr, u32 value) override; void SignalSyncPoint(u32 value) override; + void SignalReference() override; void ReleaseFences() override; void FlushAndInvalidateRegion(VAddr addr, u64 size) override; void WaitForIdle() override; @@ -93,6 +105,7 @@ public: bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surface& src, const Tegra::Engines::Fermi2D::Surface& dst, const Tegra::Engines::Fermi2D::Config& copy_config) override; + Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() override; bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) override; void LoadDiskResources(u64 title_id, std::stop_token stop_loading, @@ -103,36 +116,11 @@ public: return num_queued_commands > 0; } - VideoCommon::Shader::AsyncShaders& GetAsyncShaders() { - return async_shaders; - } - - const VideoCommon::Shader::AsyncShaders& GetAsyncShaders() const { - return async_shaders; - } - private: static constexpr size_t MAX_TEXTURES = 192; static constexpr size_t MAX_IMAGES = 48; static constexpr size_t MAX_IMAGE_VIEWS = MAX_TEXTURES + MAX_IMAGES; - void BindComputeTextures(Shader* kernel); - - void BindTextures(const ShaderEntries& entries, GLuint base_texture, GLuint base_image, - size_t& image_view_index, size_t& texture_index, size_t& image_index); - - /// Configures the current textures to use for the draw command. - void SetupDrawTextures(const Shader* shader, size_t stage_index); - - /// Configures the textures used in a compute shader. - void SetupComputeTextures(const Shader* kernel); - - /// Configures images in a graphics shader. - void SetupDrawImages(const Shader* shader, size_t stage_index); - - /// Configures images in a compute shader. - void SetupComputeImages(const Shader* shader); - /// Syncs state to match guest's void SyncState(); @@ -205,18 +193,12 @@ private: /// Syncs vertex instances to match the guest state void SyncVertexInstances(); - /// Syncs transform feedback state to match guest state - /// @note Only valid on assembly shaders - void SyncTransformFeedback(); - /// Begin a transform feedback - void BeginTransformFeedback(GLenum primitive_mode); + void BeginTransformFeedback(GraphicsPipeline* pipeline, GLenum primitive_mode); /// End a transform feedback void EndTransformFeedback(); - void SetupShaders(bool is_indexed); - Tegra::GPU& gpu; Tegra::Engines::Maxwell3D& maxwell3d; Tegra::Engines::KeplerCompute& kepler_compute; @@ -231,12 +213,11 @@ private: TextureCache texture_cache; BufferCacheRuntime buffer_cache_runtime; BufferCache buffer_cache; - ShaderCacheOpenGL shader_cache; + ShaderCache shader_cache; QueryCache query_cache; + AccelerateDMA accelerate_dma; FenceManagerOpenGL fence_manager; - VideoCommon::Shader::AsyncShaders async_shaders; - boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices; std::array<ImageViewId, MAX_IMAGE_VIEWS> image_view_ids; boost::container::static_vector<GLuint, MAX_TEXTURES> sampler_handles; @@ -244,7 +225,8 @@ private: std::array<GLuint, MAX_IMAGES> image_handles{}; /// Number of commands queued to the OpenGL driver. Resetted on flush. - std::size_t num_queued_commands = 0; + size_t num_queued_commands = 0; + bool has_written_global_memory = false; u32 last_clip_distance_mask = 0; }; diff --git a/src/video_core/renderer_opengl/gl_resource_manager.cpp b/src/video_core/renderer_opengl/gl_resource_manager.cpp index 3428e5e21..8695c29e3 100644 --- a/src/video_core/renderer_opengl/gl_resource_manager.cpp +++ b/src/video_core/renderer_opengl/gl_resource_manager.cpp @@ -83,18 +83,6 @@ void OGLSampler::Release() { handle = 0; } -void OGLShader::Create(std::string_view source, GLenum type) { - if (handle != 0) { - return; - } - if (source.empty()) { - return; - } - - MICROPROFILE_SCOPE(OpenGL_ResourceCreation); - handle = GLShader::LoadShader(source, type); -} - void OGLShader::Release() { if (handle == 0) return; @@ -104,21 +92,6 @@ void OGLShader::Release() { handle = 0; } -void OGLProgram::CreateFromSource(const char* vert_shader, const char* geo_shader, - const char* frag_shader, bool separable_program, - bool hint_retrievable) { - OGLShader vert, geo, frag; - if (vert_shader) - vert.Create(vert_shader, GL_VERTEX_SHADER); - if (geo_shader) - geo.Create(geo_shader, GL_GEOMETRY_SHADER); - if (frag_shader) - frag.Create(frag_shader, GL_FRAGMENT_SHADER); - - MICROPROFILE_SCOPE(OpenGL_ResourceCreation); - Create(separable_program, hint_retrievable, vert.handle, geo.handle, frag.handle); -} - void OGLProgram::Release() { if (handle == 0) return; diff --git a/src/video_core/renderer_opengl/gl_resource_manager.h b/src/video_core/renderer_opengl/gl_resource_manager.h index 552d79db4..b2d5bfd3b 100644 --- a/src/video_core/renderer_opengl/gl_resource_manager.h +++ b/src/video_core/renderer_opengl/gl_resource_manager.h @@ -8,7 +8,6 @@ #include <utility> #include <glad/glad.h> #include "common/common_types.h" -#include "video_core/renderer_opengl/gl_shader_util.h" namespace OpenGL { @@ -128,8 +127,6 @@ public: return *this; } - void Create(std::string_view source, GLenum type); - void Release(); GLuint handle = 0; @@ -151,17 +148,6 @@ public: return *this; } - template <typename... T> - void Create(bool separable_program, bool hint_retrievable, T... shaders) { - if (handle != 0) - return; - handle = GLShader::LoadProgram(separable_program, hint_retrievable, shaders...); - } - - /// Creates a new internal OpenGL resource and stores the handle - void CreateFromSource(const char* vert_shader, const char* geo_shader, const char* frag_shader, - bool separable_program = false, bool hint_retrievable = false); - /// Deletes the internal OpenGL resource void Release(); diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 5a01c59ec..8d6cc074c 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -3,606 +3,544 @@ // Refer to the license.txt file included. #include <atomic> +#include <fstream> #include <functional> #include <mutex> -#include <optional> #include <string> #include <thread> -#include <unordered_set> #include "common/alignment.h" #include "common/assert.h" +#include "common/fs/fs.h" +#include "common/fs/path_util.h" #include "common/logging/log.h" #include "common/scope_exit.h" +#include "common/settings.h" +#include "common/thread_worker.h" #include "core/core.h" -#include "core/frontend/emu_window.h" +#include "shader_recompiler/backend/glasm/emit_glasm.h" +#include "shader_recompiler/backend/glsl/emit_glsl.h" +#include "shader_recompiler/backend/spirv/emit_spirv.h" +#include "shader_recompiler/frontend/ir/program.h" +#include "shader_recompiler/frontend/maxwell/control_flow.h" +#include "shader_recompiler/frontend/maxwell/translate_program.h" +#include "shader_recompiler/profile.h" #include "video_core/engines/kepler_compute.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/shader_type.h" #include "video_core/memory_manager.h" -#include "video_core/renderer_opengl/gl_arb_decompiler.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/gl_shader_cache.h" -#include "video_core/renderer_opengl/gl_shader_decompiler.h" -#include "video_core/renderer_opengl/gl_shader_disk_cache.h" +#include "video_core/renderer_opengl/gl_shader_util.h" #include "video_core/renderer_opengl/gl_state_tracker.h" -#include "video_core/shader/memory_util.h" -#include "video_core/shader/registry.h" -#include "video_core/shader/shader_ir.h" #include "video_core/shader_cache.h" +#include "video_core/shader_environment.h" #include "video_core/shader_notify.h" namespace OpenGL { - -using Tegra::Engines::ShaderType; -using VideoCommon::Shader::GetShaderAddress; -using VideoCommon::Shader::GetShaderCode; -using VideoCommon::Shader::GetUniqueIdentifier; -using VideoCommon::Shader::KERNEL_MAIN_OFFSET; -using VideoCommon::Shader::ProgramCode; -using VideoCommon::Shader::Registry; -using VideoCommon::Shader::ShaderIR; -using VideoCommon::Shader::STAGE_MAIN_OFFSET; - namespace { - -constexpr VideoCommon::Shader::CompilerSettings COMPILER_SETTINGS{}; - -/// Gets the shader type from a Maxwell program type -constexpr GLenum GetGLShaderType(ShaderType shader_type) { - switch (shader_type) { - case ShaderType::Vertex: - return GL_VERTEX_SHADER; - case ShaderType::Geometry: - return GL_GEOMETRY_SHADER; - case ShaderType::Fragment: - return GL_FRAGMENT_SHADER; - case ShaderType::Compute: - return GL_COMPUTE_SHADER; - default: - return GL_NONE; - } +using Shader::Backend::GLASM::EmitGLASM; +using Shader::Backend::GLSL::EmitGLSL; +using Shader::Backend::SPIRV::EmitSPIRV; +using Shader::Maxwell::MergeDualVertexPrograms; +using Shader::Maxwell::TranslateProgram; +using VideoCommon::ComputeEnvironment; +using VideoCommon::FileEnvironment; +using VideoCommon::GenericEnvironment; +using VideoCommon::GraphicsEnvironment; +using VideoCommon::LoadPipelines; +using VideoCommon::SerializePipeline; +using Context = ShaderContext::Context; + +constexpr u32 CACHE_VERSION = 5; + +template <typename Container> +auto MakeSpan(Container& container) { + return std::span(container.data(), container.size()); } -constexpr const char* GetShaderTypeName(ShaderType shader_type) { - switch (shader_type) { - case ShaderType::Vertex: - return "VS"; - case ShaderType::TesselationControl: - return "HS"; - case ShaderType::TesselationEval: - return "DS"; - case ShaderType::Geometry: - return "GS"; - case ShaderType::Fragment: - return "FS"; - case ShaderType::Compute: - return "CS"; - } - return "UNK"; +Shader::RuntimeInfo MakeRuntimeInfo(const GraphicsPipelineKey& key, + const Shader::IR::Program& program, + const Shader::IR::Program* previous_program, + bool glasm_use_storage_buffers, bool use_assembly_shaders) { + Shader::RuntimeInfo info; + if (previous_program) { + info.previous_stage_stores = previous_program->info.stores; + } else { + // Mark all stores as available for vertex shaders + info.previous_stage_stores.mask.set(); + } + switch (program.stage) { + case Shader::Stage::VertexB: + case Shader::Stage::Geometry: + if (!use_assembly_shaders && key.xfb_enabled != 0) { + info.xfb_varyings = VideoCommon::MakeTransformFeedbackVaryings(key.xfb_state); + } + break; + case Shader::Stage::TessellationEval: + info.tess_clockwise = key.tessellation_clockwise != 0; + info.tess_primitive = [&key] { + switch (key.tessellation_primitive) { + case Maxwell::TessellationPrimitive::Isolines: + return Shader::TessPrimitive::Isolines; + case Maxwell::TessellationPrimitive::Triangles: + return Shader::TessPrimitive::Triangles; + case Maxwell::TessellationPrimitive::Quads: + return Shader::TessPrimitive::Quads; + } + UNREACHABLE(); + return Shader::TessPrimitive::Triangles; + }(); + info.tess_spacing = [&] { + switch (key.tessellation_spacing) { + case Maxwell::TessellationSpacing::Equal: + return Shader::TessSpacing::Equal; + case Maxwell::TessellationSpacing::FractionalOdd: + return Shader::TessSpacing::FractionalOdd; + case Maxwell::TessellationSpacing::FractionalEven: + return Shader::TessSpacing::FractionalEven; + } + UNREACHABLE(); + return Shader::TessSpacing::Equal; + }(); + break; + case Shader::Stage::Fragment: + info.force_early_z = key.early_z != 0; + break; + default: + break; + } + switch (key.gs_input_topology) { + case Maxwell::PrimitiveTopology::Points: + info.input_topology = Shader::InputTopology::Points; + break; + case Maxwell::PrimitiveTopology::Lines: + case Maxwell::PrimitiveTopology::LineLoop: + case Maxwell::PrimitiveTopology::LineStrip: + info.input_topology = Shader::InputTopology::Lines; + break; + case Maxwell::PrimitiveTopology::Triangles: + case Maxwell::PrimitiveTopology::TriangleStrip: + case Maxwell::PrimitiveTopology::TriangleFan: + case Maxwell::PrimitiveTopology::Quads: + case Maxwell::PrimitiveTopology::QuadStrip: + case Maxwell::PrimitiveTopology::Polygon: + case Maxwell::PrimitiveTopology::Patches: + info.input_topology = Shader::InputTopology::Triangles; + break; + case Maxwell::PrimitiveTopology::LinesAdjacency: + case Maxwell::PrimitiveTopology::LineStripAdjacency: + info.input_topology = Shader::InputTopology::LinesAdjacency; + break; + case Maxwell::PrimitiveTopology::TrianglesAdjacency: + case Maxwell::PrimitiveTopology::TriangleStripAdjacency: + info.input_topology = Shader::InputTopology::TrianglesAdjacency; + break; + } + info.glasm_use_storage_buffers = glasm_use_storage_buffers; + return info; } -constexpr ShaderType GetShaderType(Maxwell::ShaderProgram program_type) { - switch (program_type) { - case Maxwell::ShaderProgram::VertexA: - case Maxwell::ShaderProgram::VertexB: - return ShaderType::Vertex; - case Maxwell::ShaderProgram::TesselationControl: - return ShaderType::TesselationControl; - case Maxwell::ShaderProgram::TesselationEval: - return ShaderType::TesselationEval; - case Maxwell::ShaderProgram::Geometry: - return ShaderType::Geometry; - case Maxwell::ShaderProgram::Fragment: - return ShaderType::Fragment; - } - return {}; +void SetXfbState(VideoCommon::TransformFeedbackState& state, const Maxwell& regs) { + std::ranges::transform(regs.tfb_layouts, state.layouts.begin(), [](const auto& layout) { + return VideoCommon::TransformFeedbackState::Layout{ + .stream = layout.stream, + .varying_count = layout.varying_count, + .stride = layout.stride, + }; + }); + state.varyings = regs.tfb_varying_locs; } +} // Anonymous namespace -constexpr GLenum AssemblyEnum(ShaderType shader_type) { - switch (shader_type) { - case ShaderType::Vertex: - return GL_VERTEX_PROGRAM_NV; - case ShaderType::TesselationControl: - return GL_TESS_CONTROL_PROGRAM_NV; - case ShaderType::TesselationEval: - return GL_TESS_EVALUATION_PROGRAM_NV; - case ShaderType::Geometry: - return GL_GEOMETRY_PROGRAM_NV; - case ShaderType::Fragment: - return GL_FRAGMENT_PROGRAM_NV; - case ShaderType::Compute: - return GL_COMPUTE_PROGRAM_NV; +ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindow& emu_window_, + Tegra::Engines::Maxwell3D& maxwell3d_, + Tegra::Engines::KeplerCompute& kepler_compute_, + Tegra::MemoryManager& gpu_memory_, const Device& device_, + TextureCache& texture_cache_, BufferCache& buffer_cache_, + ProgramManager& program_manager_, StateTracker& state_tracker_, + VideoCore::ShaderNotify& shader_notify_) + : VideoCommon::ShaderCache{rasterizer_, gpu_memory_, maxwell3d_, kepler_compute_}, + emu_window{emu_window_}, device{device_}, texture_cache{texture_cache_}, + buffer_cache{buffer_cache_}, program_manager{program_manager_}, state_tracker{state_tracker_}, + shader_notify{shader_notify_}, use_asynchronous_shaders{device.UseAsynchronousShaders()}, + profile{ + .supported_spirv = 0x00010000, + + .unified_descriptor_binding = false, + .support_descriptor_aliasing = false, + .support_int8 = false, + .support_int16 = false, + .support_int64 = device.HasShaderInt64(), + .support_vertex_instance_id = true, + .support_float_controls = false, + .support_separate_denorm_behavior = false, + .support_separate_rounding_mode = false, + .support_fp16_denorm_preserve = false, + .support_fp32_denorm_preserve = false, + .support_fp16_denorm_flush = false, + .support_fp32_denorm_flush = false, + .support_fp16_signed_zero_nan_preserve = false, + .support_fp32_signed_zero_nan_preserve = false, + .support_fp64_signed_zero_nan_preserve = false, + .support_explicit_workgroup_layout = false, + .support_vote = true, + .support_viewport_index_layer_non_geometry = + device.HasNvViewportArray2() || device.HasVertexViewportLayer(), + .support_viewport_mask = device.HasNvViewportArray2(), + .support_typeless_image_loads = device.HasImageLoadFormatted(), + .support_demote_to_helper_invocation = false, + .support_int64_atomics = false, + .support_derivative_control = device.HasDerivativeControl(), + .support_geometry_shader_passthrough = device.HasGeometryShaderPassthrough(), + .support_gl_nv_gpu_shader_5 = device.HasNvGpuShader5(), + .support_gl_amd_gpu_shader_half_float = device.HasAmdShaderHalfFloat(), + .support_gl_texture_shadow_lod = device.HasTextureShadowLod(), + .support_gl_warp_intrinsics = false, + .support_gl_variable_aoffi = device.HasVariableAoffi(), + .support_gl_sparse_textures = device.HasSparseTexture2(), + .support_gl_derivative_control = device.HasDerivativeControl(), + + .warp_size_potentially_larger_than_guest = device.IsWarpSizePotentiallyLargerThanGuest(), + + .lower_left_origin_mode = true, + .need_declared_frag_colors = true, + .need_fastmath_off = device.NeedsFastmathOff(), + + .has_broken_spirv_clamp = true, + .has_broken_unsigned_image_offsets = true, + .has_broken_signed_operations = true, + .has_broken_fp16_float_controls = false, + .has_gl_component_indexing_bug = device.HasComponentIndexingBug(), + .has_gl_precise_bug = device.HasPreciseBug(), + .ignore_nan_fp_comparisons = true, + .gl_max_compute_smem_size = device.GetMaxComputeSharedMemorySize(), + }, + host_info{ + .support_float16 = false, + .support_int64 = device.HasShaderInt64(), + } { + if (use_asynchronous_shaders) { + workers = CreateWorkers(); } - return {}; } -std::string MakeShaderID(u64 unique_identifier, ShaderType shader_type) { - return fmt::format("{}{:016X}", GetShaderTypeName(shader_type), unique_identifier); -} +ShaderCache::~ShaderCache() = default; -std::shared_ptr<Registry> MakeRegistry(const ShaderDiskCacheEntry& entry) { - const VideoCore::GuestDriverProfile guest_profile{entry.texture_handler_size}; - const VideoCommon::Shader::SerializedRegistryInfo info{guest_profile, entry.bound_buffer, - entry.graphics_info, entry.compute_info}; - auto registry = std::make_shared<Registry>(entry.type, info); - for (const auto& [address, value] : entry.keys) { - const auto [buffer, offset] = address; - registry->InsertKey(buffer, offset, value); - } - for (const auto& [offset, sampler] : entry.bound_samplers) { - registry->InsertBoundSampler(offset, sampler); +void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading, + const VideoCore::DiskResourceLoadCallback& callback) { + if (title_id == 0) { + return; } - for (const auto& [key, sampler] : entry.bindless_samplers) { - const auto [buffer, offset] = key; - registry->InsertBindlessSampler(buffer, offset, sampler); + const auto shader_dir{Common::FS::GetYuzuPath(Common::FS::YuzuPath::ShaderDir)}; + const auto base_dir{shader_dir / fmt::format("{:016x}", title_id)}; + if (!Common::FS::CreateDir(shader_dir) || !Common::FS::CreateDir(base_dir)) { + LOG_ERROR(Common_Filesystem, "Failed to create shader cache directories"); + return; } - return registry; -} - -std::unordered_set<GLenum> GetSupportedFormats() { - GLint num_formats; - glGetIntegerv(GL_NUM_PROGRAM_BINARY_FORMATS, &num_formats); + shader_cache_filename = base_dir / "opengl.bin"; + + if (!workers) { + workers = CreateWorkers(); + } + struct { + std::mutex mutex; + size_t total{}; + size_t built{}; + bool has_loaded{}; + } state; + + const auto load_compute{[&](std::ifstream& file, FileEnvironment env) { + ComputePipelineKey key; + file.read(reinterpret_cast<char*>(&key), sizeof(key)); + workers->QueueWork( + [this, key, env = std::move(env), &state, &callback](Context* ctx) mutable { + ctx->pools.ReleaseContents(); + auto pipeline{CreateComputePipeline(ctx->pools, key, env)}; + std::lock_guard lock{state.mutex}; + if (pipeline) { + compute_cache.emplace(key, std::move(pipeline)); + } + ++state.built; + if (state.has_loaded) { + callback(VideoCore::LoadCallbackStage::Build, state.built, state.total); + } + }); + ++state.total; + }}; + const auto load_graphics{[&](std::ifstream& file, std::vector<FileEnvironment> envs) { + GraphicsPipelineKey key; + file.read(reinterpret_cast<char*>(&key), sizeof(key)); + workers->QueueWork( + [this, key, envs = std::move(envs), &state, &callback](Context* ctx) mutable { + boost::container::static_vector<Shader::Environment*, 5> env_ptrs; + for (auto& env : envs) { + env_ptrs.push_back(&env); + } + ctx->pools.ReleaseContents(); + auto pipeline{CreateGraphicsPipeline(ctx->pools, key, MakeSpan(env_ptrs), false)}; + std::lock_guard lock{state.mutex}; + if (pipeline) { + graphics_cache.emplace(key, std::move(pipeline)); + } + ++state.built; + if (state.has_loaded) { + callback(VideoCore::LoadCallbackStage::Build, state.built, state.total); + } + }); + ++state.total; + }}; + LoadPipelines(stop_loading, shader_cache_filename, CACHE_VERSION, load_compute, load_graphics); - std::vector<GLint> formats(num_formats); - glGetIntegerv(GL_PROGRAM_BINARY_FORMATS, formats.data()); + std::unique_lock lock{state.mutex}; + callback(VideoCore::LoadCallbackStage::Build, 0, state.total); + state.has_loaded = true; + lock.unlock(); - std::unordered_set<GLenum> supported_formats; - for (const GLint format : formats) { - supported_formats.insert(static_cast<GLenum>(format)); + workers->WaitForRequests(); + if (!use_asynchronous_shaders) { + workers.reset(); } - return supported_formats; } -} // Anonymous namespace - -ProgramSharedPtr BuildShader(const Device& device, ShaderType shader_type, u64 unique_identifier, - const ShaderIR& ir, const Registry& registry, bool hint_retrievable) { - if (device.UseDriverCache()) { - // Ignore hint retrievable if we are using the driver cache - hint_retrievable = false; - } - const std::string shader_id = MakeShaderID(unique_identifier, shader_type); - LOG_INFO(Render_OpenGL, "{}", shader_id); - - auto program = std::make_shared<ProgramHandle>(); - - if (device.UseAssemblyShaders()) { - const std::string arb = - DecompileAssemblyShader(device, ir, registry, shader_type, shader_id); - - GLuint& arb_prog = program->assembly_program.handle; - -// Commented out functions signal OpenGL errors but are compatible with apitrace. -// Use them only to capture and replay on apitrace. -#if 0 - glGenProgramsNV(1, &arb_prog); - glLoadProgramNV(AssemblyEnum(shader_type), arb_prog, static_cast<GLsizei>(arb.size()), - reinterpret_cast<const GLubyte*>(arb.data())); -#else - glGenProgramsARB(1, &arb_prog); - glNamedProgramStringEXT(arb_prog, AssemblyEnum(shader_type), GL_PROGRAM_FORMAT_ASCII_ARB, - static_cast<GLsizei>(arb.size()), arb.data()); -#endif - const auto err = reinterpret_cast<const char*>(glGetString(GL_PROGRAM_ERROR_STRING_NV)); - if (err && *err) { - LOG_CRITICAL(Render_OpenGL, "{}", err); - LOG_INFO(Render_OpenGL, "\n{}", arb); - } - } else { - const std::string glsl = DecompileShader(device, ir, registry, shader_type, shader_id); - OGLShader shader; - shader.Create(glsl.c_str(), GetGLShaderType(shader_type)); - - program->source_program.Create(true, hint_retrievable, shader.handle); - } - - return program; +GraphicsPipeline* ShaderCache::CurrentGraphicsPipeline() { + if (!RefreshStages(graphics_key.unique_hashes)) { + current_pipeline = nullptr; + return nullptr; + } + const auto& regs{maxwell3d.regs}; + graphics_key.raw = 0; + graphics_key.early_z.Assign(regs.force_early_fragment_tests != 0 ? 1 : 0); + graphics_key.gs_input_topology.Assign(graphics_key.unique_hashes[4] != 0 + ? regs.draw.topology.Value() + : Maxwell::PrimitiveTopology{}); + graphics_key.tessellation_primitive.Assign(regs.tess_mode.prim.Value()); + graphics_key.tessellation_spacing.Assign(regs.tess_mode.spacing.Value()); + graphics_key.tessellation_clockwise.Assign(regs.tess_mode.cw.Value()); + graphics_key.xfb_enabled.Assign(regs.tfb_enabled != 0 ? 1 : 0); + if (graphics_key.xfb_enabled) { + SetXfbState(graphics_key.xfb_state, regs); + } + if (current_pipeline && graphics_key == current_pipeline->Key()) { + return BuiltPipeline(current_pipeline); + } + return CurrentGraphicsPipelineSlowPath(); } -Shader::Shader(std::shared_ptr<Registry> registry_, ShaderEntries entries_, - ProgramSharedPtr program_, bool is_built_) - : registry{std::move(registry_)}, entries{std::move(entries_)}, program{std::move(program_)}, - is_built{is_built_} { - handle = program->assembly_program.handle; - if (handle == 0) { - handle = program->source_program.handle; +GraphicsPipeline* ShaderCache::CurrentGraphicsPipelineSlowPath() { + const auto [pair, is_new]{graphics_cache.try_emplace(graphics_key)}; + auto& pipeline{pair->second}; + if (is_new) { + pipeline = CreateGraphicsPipeline(); } - if (is_built) { - ASSERT(handle != 0); + if (!pipeline) { + return nullptr; } + current_pipeline = pipeline.get(); + return BuiltPipeline(current_pipeline); } -Shader::~Shader() = default; - -GLuint Shader::GetHandle() const { - DEBUG_ASSERT(registry->IsConsistent()); - return handle; -} - -bool Shader::IsBuilt() const { - return is_built; -} - -void Shader::AsyncOpenGLBuilt(OGLProgram new_program) { - program->source_program = std::move(new_program); - handle = program->source_program.handle; - is_built = true; -} - -void Shader::AsyncGLASMBuilt(OGLAssemblyProgram new_program) { - program->assembly_program = std::move(new_program); - handle = program->assembly_program.handle; - is_built = true; -} - -std::unique_ptr<Shader> Shader::CreateStageFromMemory( - const ShaderParameters& params, Maxwell::ShaderProgram program_type, ProgramCode code, - ProgramCode code_b, VideoCommon::Shader::AsyncShaders& async_shaders, VAddr cpu_addr) { - const auto shader_type = GetShaderType(program_type); - - auto& gpu = params.gpu; - gpu.ShaderNotify().MarkSharderBuilding(); - - auto registry = std::make_shared<Registry>(shader_type, gpu.Maxwell3D()); - if (!async_shaders.IsShaderAsync(gpu) || !params.device.UseAsynchronousShaders()) { - const ShaderIR ir(code, STAGE_MAIN_OFFSET, COMPILER_SETTINGS, *registry); - // TODO(Rodrigo): Handle VertexA shaders - // std::optional<ShaderIR> ir_b; - // if (!code_b.empty()) { - // ir_b.emplace(code_b, STAGE_MAIN_OFFSET); - // } - auto program = - BuildShader(params.device, shader_type, params.unique_identifier, ir, *registry); - ShaderDiskCacheEntry entry; - entry.type = shader_type; - entry.code = std::move(code); - entry.code_b = std::move(code_b); - entry.unique_identifier = params.unique_identifier; - entry.bound_buffer = registry->GetBoundBuffer(); - entry.graphics_info = registry->GetGraphicsInfo(); - entry.keys = registry->GetKeys(); - entry.bound_samplers = registry->GetBoundSamplers(); - entry.bindless_samplers = registry->GetBindlessSamplers(); - params.disk_cache.SaveEntry(std::move(entry)); - - gpu.ShaderNotify().MarkShaderComplete(); - - return std::unique_ptr<Shader>(new Shader(std::move(registry), - MakeEntries(params.device, ir, shader_type), - std::move(program), true)); - } else { - // Required for entries - const ShaderIR ir(code, STAGE_MAIN_OFFSET, COMPILER_SETTINGS, *registry); - auto entries = MakeEntries(params.device, ir, shader_type); - - async_shaders.QueueOpenGLShader(params.device, shader_type, params.unique_identifier, - std::move(code), std::move(code_b), STAGE_MAIN_OFFSET, - COMPILER_SETTINGS, *registry, cpu_addr); - - auto program = std::make_shared<ProgramHandle>(); - return std::unique_ptr<Shader>( - new Shader(std::move(registry), std::move(entries), std::move(program), false)); +GraphicsPipeline* ShaderCache::BuiltPipeline(GraphicsPipeline* pipeline) const noexcept { + if (pipeline->IsBuilt()) { + return pipeline; } -} - -std::unique_ptr<Shader> Shader::CreateKernelFromMemory(const ShaderParameters& params, - ProgramCode code) { - auto& gpu = params.gpu; - gpu.ShaderNotify().MarkSharderBuilding(); - - auto registry = std::make_shared<Registry>(ShaderType::Compute, params.engine); - const ShaderIR ir(code, KERNEL_MAIN_OFFSET, COMPILER_SETTINGS, *registry); - const u64 uid = params.unique_identifier; - auto program = BuildShader(params.device, ShaderType::Compute, uid, ir, *registry); - - ShaderDiskCacheEntry entry; - entry.type = ShaderType::Compute; - entry.code = std::move(code); - entry.unique_identifier = uid; - entry.bound_buffer = registry->GetBoundBuffer(); - entry.compute_info = registry->GetComputeInfo(); - entry.keys = registry->GetKeys(); - entry.bound_samplers = registry->GetBoundSamplers(); - entry.bindless_samplers = registry->GetBindlessSamplers(); - params.disk_cache.SaveEntry(std::move(entry)); - - gpu.ShaderNotify().MarkShaderComplete(); - - return std::unique_ptr<Shader>(new Shader(std::move(registry), - MakeEntries(params.device, ir, ShaderType::Compute), - std::move(program))); -} - -std::unique_ptr<Shader> Shader::CreateFromCache(const ShaderParameters& params, - const PrecompiledShader& precompiled_shader) { - return std::unique_ptr<Shader>(new Shader( - precompiled_shader.registry, precompiled_shader.entries, precompiled_shader.program)); -} - -ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer_, - Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, const Device& device_) - : ShaderCache{rasterizer_}, emu_window{emu_window_}, gpu{gpu_}, gpu_memory{gpu_memory_}, - maxwell3d{maxwell3d_}, kepler_compute{kepler_compute_}, device{device_} {} - -ShaderCacheOpenGL::~ShaderCacheOpenGL() = default; - -void ShaderCacheOpenGL::LoadDiskCache(u64 title_id, std::stop_token stop_loading, - const VideoCore::DiskResourceLoadCallback& callback) { - disk_cache.BindTitleID(title_id); - const std::optional transferable = disk_cache.LoadTransferable(); - - LOG_INFO(Render_OpenGL, "Total Shader Count: {}", - transferable.has_value() ? transferable->size() : 0); - - if (!transferable) { - return; + if (!use_asynchronous_shaders) { + return pipeline; } - - std::vector<ShaderDiskCachePrecompiled> gl_cache; - if (!device.UseAssemblyShaders() && !device.UseDriverCache()) { - // Only load precompiled cache when we are not using assembly shaders - gl_cache = disk_cache.LoadPrecompiled(); + // If something is using depth, we can assume that games are not rendering anything which + // will be used one time. + if (maxwell3d.regs.zeta_enable) { + return nullptr; } - const auto supported_formats = GetSupportedFormats(); - - // Track if precompiled cache was altered during loading to know if we have to - // serialize the virtual precompiled cache file back to the hard drive - bool precompiled_cache_altered = false; - - // Inform the frontend about shader build initialization - if (callback) { - callback(VideoCore::LoadCallbackStage::Build, 0, transferable->size()); + // If games are using a small index count, we can assume these are full screen quads. + // Usually these shaders are only used once for building textures so we can assume they + // can't be built async + if (maxwell3d.regs.index_array.count <= 6 || maxwell3d.regs.vertex_buffer.count <= 6) { + return pipeline; } + return nullptr; +} - std::mutex mutex; - std::size_t built_shaders = 0; // It doesn't have be atomic since it's used behind a mutex - std::atomic_bool gl_cache_failed = false; - - const auto find_precompiled = [&gl_cache](u64 id) { - return std::ranges::find(gl_cache, id, &ShaderDiskCachePrecompiled::unique_identifier); - }; - - const auto worker = [&](Core::Frontend::GraphicsContext* context, std::size_t begin, - std::size_t end) { - const auto scope = context->Acquire(); - - for (std::size_t i = begin; i < end; ++i) { - if (stop_loading.stop_requested()) { - return; - } - const auto& entry = (*transferable)[i]; - const u64 uid = entry.unique_identifier; - const auto it = find_precompiled(uid); - const auto precompiled_entry = it != gl_cache.end() ? &*it : nullptr; - - const bool is_compute = entry.type == ShaderType::Compute; - const u32 main_offset = is_compute ? KERNEL_MAIN_OFFSET : STAGE_MAIN_OFFSET; - auto registry = MakeRegistry(entry); - const ShaderIR ir(entry.code, main_offset, COMPILER_SETTINGS, *registry); - - ProgramSharedPtr program; - if (precompiled_entry) { - // If the shader is precompiled, attempt to load it with - program = GeneratePrecompiledProgram(entry, *precompiled_entry, supported_formats); - if (!program) { - gl_cache_failed = true; - } - } - if (!program) { - // Otherwise compile it from GLSL - program = BuildShader(device, entry.type, uid, ir, *registry, true); - } - - PrecompiledShader shader; - shader.program = std::move(program); - shader.registry = std::move(registry); - shader.entries = MakeEntries(device, ir, entry.type); - - std::scoped_lock lock{mutex}; - if (callback) { - callback(VideoCore::LoadCallbackStage::Build, ++built_shaders, - transferable->size()); - } - runtime_cache.emplace(entry.unique_identifier, std::move(shader)); - } - }; - - const std::size_t num_workers{std::max(1U, std::thread::hardware_concurrency())}; - const std::size_t bucket_size{transferable->size() / num_workers}; - std::vector<std::unique_ptr<Core::Frontend::GraphicsContext>> contexts(num_workers); - std::vector<std::thread> threads(num_workers); - for (std::size_t i = 0; i < num_workers; ++i) { - const bool is_last_worker = i + 1 == num_workers; - const std::size_t start{bucket_size * i}; - const std::size_t end{is_last_worker ? transferable->size() : start + bucket_size}; - - // On some platforms the shared context has to be created from the GUI thread - contexts[i] = emu_window.CreateSharedContext(); - threads[i] = std::thread(worker, contexts[i].get(), start, end); +ComputePipeline* ShaderCache::CurrentComputePipeline() { + const VideoCommon::ShaderInfo* const shader{ComputeShader()}; + if (!shader) { + return nullptr; } - for (auto& thread : threads) { - thread.join(); + const auto& qmd{kepler_compute.launch_description}; + const ComputePipelineKey key{ + .unique_hash = shader->unique_hash, + .shared_memory_size = qmd.shared_alloc, + .workgroup_size{qmd.block_dim_x, qmd.block_dim_y, qmd.block_dim_z}, + }; + const auto [pair, is_new]{compute_cache.try_emplace(key)}; + auto& pipeline{pair->second}; + if (!is_new) { + return pipeline.get(); } + pipeline = CreateComputePipeline(key, shader); + return pipeline.get(); +} - if (gl_cache_failed) { - // Invalidate the precompiled cache if a shader dumped shader was rejected - disk_cache.InvalidatePrecompiled(); - precompiled_cache_altered = true; - return; - } - if (stop_loading.stop_requested()) { - return; - } +std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline() { + GraphicsEnvironments environments; + GetGraphicsEnvironments(environments, graphics_key.unique_hashes); - if (device.UseAssemblyShaders() || device.UseDriverCache()) { - // Don't store precompiled binaries for assembly shaders or when using the driver cache - return; + main_pools.ReleaseContents(); + auto pipeline{CreateGraphicsPipeline(main_pools, graphics_key, environments.Span(), + use_asynchronous_shaders)}; + if (!pipeline || shader_cache_filename.empty()) { + return pipeline; } - - // TODO(Rodrigo): Do state tracking for transferable shaders and do a dummy draw - // before precompiling them - - for (std::size_t i = 0; i < transferable->size(); ++i) { - const u64 id = (*transferable)[i].unique_identifier; - const auto it = find_precompiled(id); - if (it == gl_cache.end()) { - const GLuint program = runtime_cache.at(id).program->source_program.handle; - disk_cache.SavePrecompiled(id, program); - precompiled_cache_altered = true; + boost::container::static_vector<const GenericEnvironment*, Maxwell::MaxShaderProgram> env_ptrs; + for (size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { + if (graphics_key.unique_hashes[index] != 0) { + env_ptrs.push_back(&environments.envs[index]); } } - - if (precompiled_cache_altered) { - disk_cache.SaveVirtualPrecompiledFile(); - } -} - -ProgramSharedPtr ShaderCacheOpenGL::GeneratePrecompiledProgram( - const ShaderDiskCacheEntry& entry, const ShaderDiskCachePrecompiled& precompiled_entry, - const std::unordered_set<GLenum>& supported_formats) { - if (!supported_formats.contains(precompiled_entry.binary_format)) { - LOG_INFO(Render_OpenGL, "Precompiled cache entry with unsupported format, removing"); - return {}; - } - - auto program = std::make_shared<ProgramHandle>(); - GLuint& handle = program->source_program.handle; - handle = glCreateProgram(); - glProgramParameteri(handle, GL_PROGRAM_SEPARABLE, GL_TRUE); - glProgramBinary(handle, precompiled_entry.binary_format, precompiled_entry.binary.data(), - static_cast<GLsizei>(precompiled_entry.binary.size())); - - GLint link_status; - glGetProgramiv(handle, GL_LINK_STATUS, &link_status); - if (link_status == GL_FALSE) { - LOG_INFO(Render_OpenGL, "Precompiled cache rejected by the driver, removing"); - return {}; - } - - return program; + SerializePipeline(graphics_key, env_ptrs, shader_cache_filename, CACHE_VERSION); + return pipeline; } -Shader* ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program, - VideoCommon::Shader::AsyncShaders& async_shaders) { - if (!maxwell3d.dirty.flags[Dirty::Shaders]) { - auto* last_shader = last_shaders[static_cast<std::size_t>(program)]; - if (last_shader->IsBuilt()) { - return last_shader; +std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline( + ShaderContext::ShaderPools& pools, const GraphicsPipelineKey& key, + std::span<Shader::Environment* const> envs, bool build_in_parallel) try { + LOG_INFO(Render_OpenGL, "0x{:016x}", key.Hash()); + size_t env_index{}; + u32 total_storage_buffers{}; + std::array<Shader::IR::Program, Maxwell::MaxShaderProgram> programs; + const bool uses_vertex_a{key.unique_hashes[0] != 0}; + const bool uses_vertex_b{key.unique_hashes[1] != 0}; + for (size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { + if (key.unique_hashes[index] == 0) { + continue; } - } + Shader::Environment& env{*envs[env_index]}; + ++env_index; - const GPUVAddr address{GetShaderAddress(maxwell3d, program)}; + const u32 cfg_offset{static_cast<u32>(env.StartAddress() + sizeof(Shader::ProgramHeader))}; + Shader::Maxwell::Flow::CFG cfg(env, pools.flow_block, cfg_offset, index == 0); + if (!uses_vertex_a || index != 1) { + // Normal path + programs[index] = TranslateProgram(pools.inst, pools.block, env, cfg, host_info); - if (device.UseAsynchronousShaders() && async_shaders.HasCompletedWork()) { - auto completed_work = async_shaders.GetCompletedWork(); - for (auto& work : completed_work) { - Shader* shader = TryGet(work.cpu_address); - gpu.ShaderNotify().MarkShaderComplete(); - if (shader == nullptr) { - continue; + for (const auto& desc : programs[index].info.storage_buffers_descriptors) { + total_storage_buffers += desc.count; } - using namespace VideoCommon::Shader; - if (work.backend == AsyncShaders::Backend::OpenGL) { - shader->AsyncOpenGLBuilt(std::move(work.program.opengl)); - } else if (work.backend == AsyncShaders::Backend::GLASM) { - shader->AsyncGLASMBuilt(std::move(work.program.glasm)); + } else { + // VertexB path when VertexA is present. + auto& program_va{programs[0]}; + auto program_vb{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)}; + for (const auto& desc : program_vb.info.storage_buffers_descriptors) { + total_storage_buffers += desc.count; } - - auto& registry = shader->GetRegistry(); - - ShaderDiskCacheEntry entry; - entry.type = work.shader_type; - entry.code = std::move(work.code); - entry.code_b = std::move(work.code_b); - entry.unique_identifier = work.uid; - entry.bound_buffer = registry.GetBoundBuffer(); - entry.graphics_info = registry.GetGraphicsInfo(); - entry.keys = registry.GetKeys(); - entry.bound_samplers = registry.GetBoundSamplers(); - entry.bindless_samplers = registry.GetBindlessSamplers(); - disk_cache.SaveEntry(std::move(entry)); + programs[index] = MergeDualVertexPrograms(program_va, program_vb, env); } } - - // Look up shader in the cache based on address - const std::optional<VAddr> cpu_addr{gpu_memory.GpuToCpuAddress(address)}; - if (Shader* const shader{cpu_addr ? TryGet(*cpu_addr) : null_shader.get()}) { - return last_shaders[static_cast<std::size_t>(program)] = shader; - } - - const u8* const host_ptr{gpu_memory.GetPointer(address)}; - - // No shader found - create a new one - ProgramCode code{GetShaderCode(gpu_memory, address, host_ptr, false)}; - ProgramCode code_b; - if (program == Maxwell::ShaderProgram::VertexA) { - const GPUVAddr address_b{GetShaderAddress(maxwell3d, Maxwell::ShaderProgram::VertexB)}; - const u8* host_ptr_b = gpu_memory.GetPointer(address_b); - code_b = GetShaderCode(gpu_memory, address_b, host_ptr_b, false); - } - const std::size_t code_size = code.size() * sizeof(u64); - - const u64 unique_identifier = GetUniqueIdentifier( - GetShaderType(program), program == Maxwell::ShaderProgram::VertexA, code, code_b); - - const ShaderParameters params{gpu, maxwell3d, disk_cache, device, - *cpu_addr, host_ptr, unique_identifier}; - - std::unique_ptr<Shader> shader; - const auto found = runtime_cache.find(unique_identifier); - if (found == runtime_cache.end()) { - shader = Shader::CreateStageFromMemory(params, program, std::move(code), std::move(code_b), - async_shaders, cpu_addr.value_or(0)); - } else { - shader = Shader::CreateFromCache(params, found->second); - } - - Shader* const result = shader.get(); - if (cpu_addr) { - Register(std::move(shader), *cpu_addr, code_size); - } else { - null_shader = std::move(shader); + const u32 glasm_storage_buffer_limit{device.GetMaxGLASMStorageBufferBlocks()}; + const bool glasm_use_storage_buffers{total_storage_buffers <= glasm_storage_buffer_limit}; + + std::array<const Shader::Info*, Maxwell::MaxShaderStage> infos{}; + + OGLProgram source_program; + std::array<std::string, 5> sources; + std::array<std::vector<u32>, 5> sources_spirv; + Shader::Backend::Bindings binding; + Shader::IR::Program* previous_program{}; + const bool use_glasm{device.UseAssemblyShaders()}; + const size_t first_index = uses_vertex_a && uses_vertex_b ? 1 : 0; + for (size_t index = first_index; index < Maxwell::MaxShaderProgram; ++index) { + if (key.unique_hashes[index] == 0) { + continue; + } + UNIMPLEMENTED_IF(index == 0); + + Shader::IR::Program& program{programs[index]}; + const size_t stage_index{index - 1}; + infos[stage_index] = &program.info; + + const auto runtime_info{ + MakeRuntimeInfo(key, program, previous_program, glasm_use_storage_buffers, use_glasm)}; + switch (device.GetShaderBackend()) { + case Settings::ShaderBackend::GLSL: + sources[stage_index] = EmitGLSL(profile, runtime_info, program, binding); + break; + case Settings::ShaderBackend::GLASM: + sources[stage_index] = EmitGLASM(profile, runtime_info, program, binding); + break; + case Settings::ShaderBackend::SPIRV: + sources_spirv[stage_index] = EmitSPIRV(profile, runtime_info, program, binding); + break; + } + previous_program = &program; } + auto* const thread_worker{build_in_parallel ? workers.get() : nullptr}; + return std::make_unique<GraphicsPipeline>( + device, texture_cache, buffer_cache, gpu_memory, maxwell3d, program_manager, state_tracker, + thread_worker, &shader_notify, sources, sources_spirv, infos, key); - return last_shaders[static_cast<std::size_t>(program)] = result; +} catch (Shader::Exception& exception) { + LOG_ERROR(Render_OpenGL, "{}", exception.what()); + return nullptr; } -Shader* ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) { - const std::optional<VAddr> cpu_addr{gpu_memory.GpuToCpuAddress(code_addr)}; - - if (Shader* const kernel = cpu_addr ? TryGet(*cpu_addr) : null_kernel.get()) { - return kernel; - } - - // No kernel found, create a new one - const u8* host_ptr{gpu_memory.GetPointer(code_addr)}; - ProgramCode code{GetShaderCode(gpu_memory, code_addr, host_ptr, true)}; - const std::size_t code_size{code.size() * sizeof(u64)}; - const u64 unique_identifier{GetUniqueIdentifier(ShaderType::Compute, false, code)}; - - const ShaderParameters params{gpu, kepler_compute, disk_cache, device, - *cpu_addr, host_ptr, unique_identifier}; +std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline( + const ComputePipelineKey& key, const VideoCommon::ShaderInfo* shader) { + const GPUVAddr program_base{kepler_compute.regs.code_loc.Address()}; + const auto& qmd{kepler_compute.launch_description}; + ComputeEnvironment env{kepler_compute, gpu_memory, program_base, qmd.program_start}; + env.SetCachedSize(shader->size_bytes); + + main_pools.ReleaseContents(); + auto pipeline{CreateComputePipeline(main_pools, key, env)}; + if (!pipeline || shader_cache_filename.empty()) { + return pipeline; + } + SerializePipeline(key, std::array<const GenericEnvironment*, 1>{&env}, shader_cache_filename, + CACHE_VERSION); + return pipeline; +} - std::unique_ptr<Shader> kernel; - const auto found = runtime_cache.find(unique_identifier); - if (found == runtime_cache.end()) { - kernel = Shader::CreateKernelFromMemory(params, std::move(code)); - } else { - kernel = Shader::CreateFromCache(params, found->second); - } +std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline( + ShaderContext::ShaderPools& pools, const ComputePipelineKey& key, + Shader::Environment& env) try { + LOG_INFO(Render_OpenGL, "0x{:016x}", key.Hash()); + + Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()}; + auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)}; + + u32 num_storage_buffers{}; + for (const auto& desc : program.info.storage_buffers_descriptors) { + num_storage_buffers += desc.count; + } + Shader::RuntimeInfo info; + info.glasm_use_storage_buffers = num_storage_buffers <= device.GetMaxGLASMStorageBufferBlocks(); + + std::string code{}; + std::vector<u32> code_spirv; + switch (device.GetShaderBackend()) { + case Settings::ShaderBackend::GLSL: + code = EmitGLSL(profile, program); + break; + case Settings::ShaderBackend::GLASM: + code = EmitGLASM(profile, info, program); + break; + case Settings::ShaderBackend::SPIRV: + code_spirv = EmitSPIRV(profile, program); + break; + } + + return std::make_unique<ComputePipeline>(device, texture_cache, buffer_cache, gpu_memory, + kepler_compute, program_manager, program.info, code, + code_spirv); +} catch (Shader::Exception& exception) { + LOG_ERROR(Render_OpenGL, "{}", exception.what()); + return nullptr; +} - Shader* const result = kernel.get(); - if (cpu_addr) { - Register(std::move(kernel), *cpu_addr, code_size); - } else { - null_kernel = std::move(kernel); - } - return result; +std::unique_ptr<ShaderWorker> ShaderCache::CreateWorkers() const { + return std::make_unique<ShaderWorker>(std::max(std::thread::hardware_concurrency(), 2U) - 1, + "yuzu:ShaderBuilder", + [this] { return Context{emu_window}; }); } } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h index b30308b6f..a34110b37 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_cache.h @@ -5,157 +5,93 @@ #pragma once #include <array> -#include <atomic> -#include <bitset> -#include <memory> -#include <string> -#include <tuple> +#include <filesystem> +#include <stop_token> #include <unordered_map> -#include <unordered_set> -#include <vector> #include <glad/glad.h> #include "common/common_types.h" -#include "video_core/engines/shader_type.h" -#include "video_core/renderer_opengl/gl_resource_manager.h" -#include "video_core/renderer_opengl/gl_shader_decompiler.h" -#include "video_core/renderer_opengl/gl_shader_disk_cache.h" -#include "video_core/shader/registry.h" -#include "video_core/shader/shader_ir.h" +#include "common/thread_worker.h" +#include "shader_recompiler/frontend/ir/value.h" +#include "shader_recompiler/host_translate_info.h" +#include "shader_recompiler/object_pool.h" +#include "shader_recompiler/profile.h" +#include "video_core/renderer_opengl/gl_compute_pipeline.h" +#include "video_core/renderer_opengl/gl_graphics_pipeline.h" +#include "video_core/renderer_opengl/gl_shader_context.h" #include "video_core/shader_cache.h" namespace Tegra { class MemoryManager; } -namespace Core::Frontend { -class EmuWindow; -} - -namespace VideoCommon::Shader { -class AsyncShaders; -} - namespace OpenGL { class Device; +class ProgramManager; class RasterizerOpenGL; +using ShaderWorker = Common::StatefulThreadWorker<ShaderContext::Context>; -using Maxwell = Tegra::Engines::Maxwell3D::Regs; - -struct ProgramHandle { - OGLProgram source_program; - OGLAssemblyProgram assembly_program; -}; -using ProgramSharedPtr = std::shared_ptr<ProgramHandle>; - -struct PrecompiledShader { - ProgramSharedPtr program; - std::shared_ptr<VideoCommon::Shader::Registry> registry; - ShaderEntries entries; -}; - -struct ShaderParameters { - Tegra::GPU& gpu; - Tegra::Engines::ConstBufferEngineInterface& engine; - ShaderDiskCacheOpenGL& disk_cache; - const Device& device; - VAddr cpu_addr; - const u8* host_ptr; - u64 unique_identifier; -}; - -ProgramSharedPtr BuildShader(const Device& device, Tegra::Engines::ShaderType shader_type, - u64 unique_identifier, const VideoCommon::Shader::ShaderIR& ir, - const VideoCommon::Shader::Registry& registry, - bool hint_retrievable = false); - -class Shader final { +class ShaderCache : public VideoCommon::ShaderCache { public: - ~Shader(); - - /// Gets the GL program handle for the shader - GLuint GetHandle() const; - - bool IsBuilt() const; - - /// Gets the shader entries for the shader - const ShaderEntries& GetEntries() const { - return entries; - } - - const VideoCommon::Shader::Registry& GetRegistry() const { - return *registry; - } - - /// Mark a OpenGL shader as built - void AsyncOpenGLBuilt(OGLProgram new_program); + explicit ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindow& emu_window_, + Tegra::Engines::Maxwell3D& maxwell3d_, + Tegra::Engines::KeplerCompute& kepler_compute_, + Tegra::MemoryManager& gpu_memory_, const Device& device_, + TextureCache& texture_cache_, BufferCache& buffer_cache_, + ProgramManager& program_manager_, StateTracker& state_tracker_, + VideoCore::ShaderNotify& shader_notify_); + ~ShaderCache(); - /// Mark a GLASM shader as built - void AsyncGLASMBuilt(OGLAssemblyProgram new_program); + void LoadDiskResources(u64 title_id, std::stop_token stop_loading, + const VideoCore::DiskResourceLoadCallback& callback); - static std::unique_ptr<Shader> CreateStageFromMemory( - const ShaderParameters& params, Maxwell::ShaderProgram program_type, - ProgramCode program_code, ProgramCode program_code_b, - VideoCommon::Shader::AsyncShaders& async_shaders, VAddr cpu_addr); + [[nodiscard]] GraphicsPipeline* CurrentGraphicsPipeline(); - static std::unique_ptr<Shader> CreateKernelFromMemory(const ShaderParameters& params, - ProgramCode code); - - static std::unique_ptr<Shader> CreateFromCache(const ShaderParameters& params, - const PrecompiledShader& precompiled_shader); + [[nodiscard]] ComputePipeline* CurrentComputePipeline(); private: - explicit Shader(std::shared_ptr<VideoCommon::Shader::Registry> registry, ShaderEntries entries, - ProgramSharedPtr program, bool is_built_ = true); - - std::shared_ptr<VideoCommon::Shader::Registry> registry; - ShaderEntries entries; - ProgramSharedPtr program; - GLuint handle = 0; - bool is_built{}; -}; + GraphicsPipeline* CurrentGraphicsPipelineSlowPath(); -class ShaderCacheOpenGL final : public VideoCommon::ShaderCache<Shader> { -public: - explicit ShaderCacheOpenGL(RasterizerOpenGL& rasterizer_, - Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, const Device& device_); - ~ShaderCacheOpenGL() override; + [[nodiscard]] GraphicsPipeline* BuiltPipeline(GraphicsPipeline* pipeline) const noexcept; - /// Loads disk cache for the current game - void LoadDiskCache(u64 title_id, std::stop_token stop_loading, - const VideoCore::DiskResourceLoadCallback& callback); + std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline(); - /// Gets the current specified shader stage program - Shader* GetStageProgram(Maxwell::ShaderProgram program, - VideoCommon::Shader::AsyncShaders& async_shaders); + std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline( + ShaderContext::ShaderPools& pools, const GraphicsPipelineKey& key, + std::span<Shader::Environment* const> envs, bool build_in_parallel); - /// Gets a compute kernel in the passed address - Shader* GetComputeKernel(GPUVAddr code_addr); + std::unique_ptr<ComputePipeline> CreateComputePipeline(const ComputePipelineKey& key, + const VideoCommon::ShaderInfo* shader); -private: - ProgramSharedPtr GeneratePrecompiledProgram( - const ShaderDiskCacheEntry& entry, const ShaderDiskCachePrecompiled& precompiled_entry, - const std::unordered_set<GLenum>& supported_formats); + std::unique_ptr<ComputePipeline> CreateComputePipeline(ShaderContext::ShaderPools& pools, + const ComputePipelineKey& key, + Shader::Environment& env); + + std::unique_ptr<ShaderWorker> CreateWorkers() const; Core::Frontend::EmuWindow& emu_window; - Tegra::GPU& gpu; - Tegra::MemoryManager& gpu_memory; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::Engines::KeplerCompute& kepler_compute; const Device& device; + TextureCache& texture_cache; + BufferCache& buffer_cache; + ProgramManager& program_manager; + StateTracker& state_tracker; + VideoCore::ShaderNotify& shader_notify; + const bool use_asynchronous_shaders; + + GraphicsPipelineKey graphics_key{}; + GraphicsPipeline* current_pipeline{}; - ShaderDiskCacheOpenGL disk_cache; - std::unordered_map<u64, PrecompiledShader> runtime_cache; + ShaderContext::ShaderPools main_pools; + std::unordered_map<GraphicsPipelineKey, std::unique_ptr<GraphicsPipeline>> graphics_cache; + std::unordered_map<ComputePipelineKey, std::unique_ptr<ComputePipeline>> compute_cache; - std::unique_ptr<Shader> null_shader; - std::unique_ptr<Shader> null_kernel; + Shader::Profile profile; + Shader::HostTranslateInfo host_info; - std::array<Shader*, Maxwell::MaxShaderProgram> last_shaders{}; + std::filesystem::path shader_cache_filename; + std::unique_ptr<ShaderWorker> workers; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_context.h b/src/video_core/renderer_opengl/gl_shader_context.h new file mode 100644 index 000000000..6ff34e5d6 --- /dev/null +++ b/src/video_core/renderer_opengl/gl_shader_context.h @@ -0,0 +1,33 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include "core/frontend/emu_window.h" +#include "shader_recompiler/frontend/ir/basic_block.h" +#include "shader_recompiler/frontend/maxwell/control_flow.h" + +namespace OpenGL::ShaderContext { +struct ShaderPools { + void ReleaseContents() { + flow_block.ReleaseContents(); + block.ReleaseContents(); + inst.ReleaseContents(); + } + + Shader::ObjectPool<Shader::IR::Inst> inst; + Shader::ObjectPool<Shader::IR::Block> block; + Shader::ObjectPool<Shader::Maxwell::Flow::Block> flow_block; +}; + +struct Context { + explicit Context(Core::Frontend::EmuWindow& emu_window) + : gl_context{emu_window.CreateSharedContext()}, scoped{*gl_context} {} + + std::unique_ptr<Core::Frontend::GraphicsContext> gl_context; + Core::Frontend::GraphicsContext::Scoped scoped; + ShaderPools pools; +}; + +} // namespace OpenGL::ShaderContext diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp deleted file mode 100644 index 9c28498e8..000000000 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ /dev/null @@ -1,2986 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <array> -#include <string> -#include <string_view> -#include <utility> -#include <variant> -#include <vector> - -#include <fmt/format.h> - -#include "common/alignment.h" -#include "common/assert.h" -#include "common/common_types.h" -#include "common/div_ceil.h" -#include "common/logging/log.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/shader_type.h" -#include "video_core/renderer_opengl/gl_device.h" -#include "video_core/renderer_opengl/gl_rasterizer.h" -#include "video_core/renderer_opengl/gl_shader_decompiler.h" -#include "video_core/shader/ast.h" -#include "video_core/shader/node.h" -#include "video_core/shader/shader_ir.h" -#include "video_core/shader/transform_feedback.h" - -namespace OpenGL { - -namespace { - -using Tegra::Engines::ShaderType; -using Tegra::Shader::Attribute; -using Tegra::Shader::Header; -using Tegra::Shader::IpaInterpMode; -using Tegra::Shader::IpaMode; -using Tegra::Shader::IpaSampleMode; -using Tegra::Shader::PixelImap; -using Tegra::Shader::Register; -using Tegra::Shader::TextureType; - -using namespace VideoCommon::Shader; -using namespace std::string_literals; - -using Maxwell = Tegra::Engines::Maxwell3D::Regs; -using Operation = const OperationNode&; - -class ASTDecompiler; -class ExprDecompiler; - -enum class Type { Void, Bool, Bool2, Float, Int, Uint, HalfFloat }; - -constexpr std::array FLOAT_TYPES{"float", "vec2", "vec3", "vec4"}; - -constexpr std::string_view INPUT_ATTRIBUTE_NAME = "in_attr"; -constexpr std::string_view OUTPUT_ATTRIBUTE_NAME = "out_attr"; - -struct TextureOffset {}; -struct TextureDerivates {}; -using TextureArgument = std::pair<Type, Node>; -using TextureIR = std::variant<TextureOffset, TextureDerivates, TextureArgument>; - -constexpr u32 MAX_CONSTBUFFER_SCALARS = static_cast<u32>(Maxwell::MaxConstBufferSize) / sizeof(u32); -constexpr u32 MAX_CONSTBUFFER_ELEMENTS = MAX_CONSTBUFFER_SCALARS / sizeof(u32); - -constexpr std::string_view COMMON_DECLARATIONS = R"(#define ftoi floatBitsToInt -#define ftou floatBitsToUint -#define itof intBitsToFloat -#define utof uintBitsToFloat - -bvec2 HalfFloatNanComparison(bvec2 comparison, vec2 pair1, vec2 pair2) {{ - bvec2 is_nan1 = isnan(pair1); - bvec2 is_nan2 = isnan(pair2); - return bvec2(comparison.x || is_nan1.x || is_nan2.x, comparison.y || is_nan1.y || is_nan2.y); -}} - -const float fswzadd_modifiers_a[] = float[4](-1.0f, 1.0f, -1.0f, 0.0f ); -const float fswzadd_modifiers_b[] = float[4](-1.0f, -1.0f, 1.0f, -1.0f ); -)"; - -class ShaderWriter final { -public: - void AddExpression(std::string_view text) { - DEBUG_ASSERT(scope >= 0); - if (!text.empty()) { - AppendIndentation(); - } - shader_source += text; - } - - // Forwards all arguments directly to libfmt. - // Note that all formatting requirements for fmt must be - // obeyed when using this function. (e.g. {{ must be used - // printing the character '{' is desirable. Ditto for }} and '}', - // etc). - template <typename... Args> - void AddLine(std::string_view text, Args&&... args) { - AddExpression(fmt::format(fmt::runtime(text), std::forward<Args>(args)...)); - AddNewLine(); - } - - void AddNewLine() { - DEBUG_ASSERT(scope >= 0); - shader_source += '\n'; - } - - std::string GenerateTemporary() { - return fmt::format("tmp{}", temporary_index++); - } - - std::string GetResult() { - return std::move(shader_source); - } - - s32 scope = 0; - -private: - void AppendIndentation() { - shader_source.append(static_cast<std::size_t>(scope) * 4, ' '); - } - - std::string shader_source; - u32 temporary_index = 1; -}; - -class Expression final { -public: - Expression(std::string code_, Type type_) : code{std::move(code_)}, type{type_} { - ASSERT(type != Type::Void); - } - Expression() : type{Type::Void} {} - - Type GetType() const { - return type; - } - - std::string GetCode() const { - return code; - } - - void CheckVoid() const { - ASSERT(type == Type::Void); - } - - std::string As(Type type_) const { - switch (type_) { - case Type::Bool: - return AsBool(); - case Type::Bool2: - return AsBool2(); - case Type::Float: - return AsFloat(); - case Type::Int: - return AsInt(); - case Type::Uint: - return AsUint(); - case Type::HalfFloat: - return AsHalfFloat(); - default: - UNREACHABLE_MSG("Invalid type"); - return code; - } - } - - std::string AsBool() const { - switch (type) { - case Type::Bool: - return code; - default: - UNREACHABLE_MSG("Incompatible types"); - return code; - } - } - - std::string AsBool2() const { - switch (type) { - case Type::Bool2: - return code; - default: - UNREACHABLE_MSG("Incompatible types"); - return code; - } - } - - std::string AsFloat() const { - switch (type) { - case Type::Float: - return code; - case Type::Uint: - return fmt::format("utof({})", code); - case Type::Int: - return fmt::format("itof({})", code); - case Type::HalfFloat: - return fmt::format("utof(packHalf2x16({}))", code); - default: - UNREACHABLE_MSG("Incompatible types"); - return code; - } - } - - std::string AsInt() const { - switch (type) { - case Type::Float: - return fmt::format("ftoi({})", code); - case Type::Uint: - return fmt::format("int({})", code); - case Type::Int: - return code; - case Type::HalfFloat: - return fmt::format("int(packHalf2x16({}))", code); - default: - UNREACHABLE_MSG("Incompatible types"); - return code; - } - } - - std::string AsUint() const { - switch (type) { - case Type::Float: - return fmt::format("ftou({})", code); - case Type::Uint: - return code; - case Type::Int: - return fmt::format("uint({})", code); - case Type::HalfFloat: - return fmt::format("packHalf2x16({})", code); - default: - UNREACHABLE_MSG("Incompatible types"); - return code; - } - } - - std::string AsHalfFloat() const { - switch (type) { - case Type::Float: - return fmt::format("unpackHalf2x16(ftou({}))", code); - case Type::Uint: - return fmt::format("unpackHalf2x16({})", code); - case Type::Int: - return fmt::format("unpackHalf2x16(int({}))", code); - case Type::HalfFloat: - return code; - default: - UNREACHABLE_MSG("Incompatible types"); - return code; - } - } - -private: - std::string code; - Type type{}; -}; - -const char* GetTypeString(Type type) { - switch (type) { - case Type::Bool: - return "bool"; - case Type::Bool2: - return "bvec2"; - case Type::Float: - return "float"; - case Type::Int: - return "int"; - case Type::Uint: - return "uint"; - case Type::HalfFloat: - return "vec2"; - default: - UNREACHABLE_MSG("Invalid type"); - return "<invalid type>"; - } -} - -const char* GetImageTypeDeclaration(Tegra::Shader::ImageType image_type) { - switch (image_type) { - case Tegra::Shader::ImageType::Texture1D: - return "1D"; - case Tegra::Shader::ImageType::TextureBuffer: - return "Buffer"; - case Tegra::Shader::ImageType::Texture1DArray: - return "1DArray"; - case Tegra::Shader::ImageType::Texture2D: - return "2D"; - case Tegra::Shader::ImageType::Texture2DArray: - return "2DArray"; - case Tegra::Shader::ImageType::Texture3D: - return "3D"; - default: - UNREACHABLE(); - return "1D"; - } -} - -/// Describes primitive behavior on geometry shaders -std::pair<const char*, u32> GetPrimitiveDescription(Maxwell::PrimitiveTopology topology) { - switch (topology) { - case Maxwell::PrimitiveTopology::Points: - return {"points", 1}; - case Maxwell::PrimitiveTopology::Lines: - case Maxwell::PrimitiveTopology::LineStrip: - return {"lines", 2}; - case Maxwell::PrimitiveTopology::LinesAdjacency: - case Maxwell::PrimitiveTopology::LineStripAdjacency: - return {"lines_adjacency", 4}; - case Maxwell::PrimitiveTopology::Triangles: - case Maxwell::PrimitiveTopology::TriangleStrip: - case Maxwell::PrimitiveTopology::TriangleFan: - return {"triangles", 3}; - case Maxwell::PrimitiveTopology::TrianglesAdjacency: - case Maxwell::PrimitiveTopology::TriangleStripAdjacency: - return {"triangles_adjacency", 6}; - default: - UNIMPLEMENTED_MSG("topology={}", topology); - return {"points", 1}; - } -} - -/// Generates code to use for a swizzle operation. -constexpr const char* GetSwizzle(std::size_t element) { - constexpr std::array swizzle = {".x", ".y", ".z", ".w"}; - return swizzle.at(element); -} - -constexpr const char* GetColorSwizzle(std::size_t element) { - constexpr std::array swizzle = {".r", ".g", ".b", ".a"}; - return swizzle.at(element); -} - -/// Translate topology -std::string GetTopologyName(Tegra::Shader::OutputTopology topology) { - switch (topology) { - case Tegra::Shader::OutputTopology::PointList: - return "points"; - case Tegra::Shader::OutputTopology::LineStrip: - return "line_strip"; - case Tegra::Shader::OutputTopology::TriangleStrip: - return "triangle_strip"; - default: - UNIMPLEMENTED_MSG("Unknown output topology: {}", topology); - return "points"; - } -} - -/// Returns true if an object has to be treated as precise -bool IsPrecise(Operation operand) { - const auto& meta{operand.GetMeta()}; - if (const auto arithmetic = std::get_if<MetaArithmetic>(&meta)) { - return arithmetic->precise; - } - return false; -} - -bool IsPrecise(const Node& node) { - if (const auto operation = std::get_if<OperationNode>(&*node)) { - return IsPrecise(*operation); - } - return false; -} - -constexpr bool IsGenericAttribute(Attribute::Index index) { - return index >= Attribute::Index::Attribute_0 && index <= Attribute::Index::Attribute_31; -} - -constexpr bool IsLegacyTexCoord(Attribute::Index index) { - return static_cast<int>(index) >= static_cast<int>(Attribute::Index::TexCoord_0) && - static_cast<int>(index) <= static_cast<int>(Attribute::Index::TexCoord_7); -} - -constexpr Attribute::Index ToGenericAttribute(u64 value) { - return static_cast<Attribute::Index>(value + static_cast<u64>(Attribute::Index::Attribute_0)); -} - -constexpr int GetLegacyTexCoordIndex(Attribute::Index index) { - return static_cast<int>(index) - static_cast<int>(Attribute::Index::TexCoord_0); -} - -u32 GetGenericAttributeIndex(Attribute::Index index) { - ASSERT(IsGenericAttribute(index)); - return static_cast<u32>(index) - static_cast<u32>(Attribute::Index::Attribute_0); -} - -constexpr const char* GetFlowStackPrefix(MetaStackClass stack) { - switch (stack) { - case MetaStackClass::Ssy: - return "ssy"; - case MetaStackClass::Pbk: - return "pbk"; - } - return {}; -} - -std::string FlowStackName(MetaStackClass stack) { - return fmt::format("{}_flow_stack", GetFlowStackPrefix(stack)); -} - -std::string FlowStackTopName(MetaStackClass stack) { - return fmt::format("{}_flow_stack_top", GetFlowStackPrefix(stack)); -} - -struct GenericVaryingDescription { - std::string name; - u8 first_element = 0; - bool is_scalar = false; -}; - -class GLSLDecompiler final { -public: - explicit GLSLDecompiler(const Device& device_, const ShaderIR& ir_, const Registry& registry_, - ShaderType stage_, std::string_view identifier_, - std::string_view suffix_) - : device{device_}, ir{ir_}, registry{registry_}, stage{stage_}, - identifier{identifier_}, suffix{suffix_}, header{ir.GetHeader()} { - if (stage != ShaderType::Compute) { - transform_feedback = BuildTransformFeedback(registry.GetGraphicsInfo()); - } - } - - void Decompile() { - DeclareHeader(); - DeclareVertex(); - DeclareGeometry(); - DeclareFragment(); - DeclareCompute(); - DeclareInputAttributes(); - DeclareOutputAttributes(); - DeclareImages(); - DeclareSamplers(); - DeclareGlobalMemory(); - DeclareConstantBuffers(); - DeclareLocalMemory(); - DeclareRegisters(); - DeclarePredicates(); - DeclareInternalFlags(); - DeclareCustomVariables(); - DeclarePhysicalAttributeReader(); - - code.AddLine("void main() {{"); - ++code.scope; - - if (stage == ShaderType::Vertex) { - code.AddLine("gl_Position = vec4(0.0f, 0.0f, 0.0f, 1.0f);"); - } - - if (ir.IsDecompiled()) { - DecompileAST(); - } else { - DecompileBranchMode(); - } - - --code.scope; - code.AddLine("}}"); - } - - std::string GetResult() { - return code.GetResult(); - } - -private: - friend class ASTDecompiler; - friend class ExprDecompiler; - - void DecompileBranchMode() { - // VM's program counter - const auto first_address = ir.GetBasicBlocks().begin()->first; - code.AddLine("uint jmp_to = {}U;", first_address); - - // TODO(Subv): Figure out the actual depth of the flow stack, for now it seems - // unlikely that shaders will use 20 nested SSYs and PBKs. - constexpr u32 FLOW_STACK_SIZE = 20; - if (!ir.IsFlowStackDisabled()) { - for (const auto stack : std::array{MetaStackClass::Ssy, MetaStackClass::Pbk}) { - code.AddLine("uint {}[{}];", FlowStackName(stack), FLOW_STACK_SIZE); - code.AddLine("uint {} = 0U;", FlowStackTopName(stack)); - } - } - - code.AddLine("while (true) {{"); - ++code.scope; - - code.AddLine("switch (jmp_to) {{"); - - for (const auto& pair : ir.GetBasicBlocks()) { - const auto& [address, bb] = pair; - code.AddLine("case 0x{:X}U: {{", address); - ++code.scope; - - VisitBlock(bb); - - --code.scope; - code.AddLine("}}"); - } - - code.AddLine("default: return;"); - code.AddLine("}}"); - - --code.scope; - code.AddLine("}}"); - } - - void DecompileAST(); - - void DeclareHeader() { - if (!identifier.empty()) { - code.AddLine("// {}", identifier); - } - const bool use_compatibility = ir.UsesLegacyVaryings() || ir.UsesYNegate(); - code.AddLine("#version 440 {}", use_compatibility ? "compatibility" : "core"); - code.AddLine("#extension GL_ARB_separate_shader_objects : enable"); - if (device.HasShaderBallot()) { - code.AddLine("#extension GL_ARB_shader_ballot : require"); - } - if (device.HasVertexViewportLayer()) { - code.AddLine("#extension GL_ARB_shader_viewport_layer_array : require"); - } - if (device.HasImageLoadFormatted()) { - code.AddLine("#extension GL_EXT_shader_image_load_formatted : require"); - } - if (device.HasTextureShadowLod()) { - code.AddLine("#extension GL_EXT_texture_shadow_lod : require"); - } - if (device.HasWarpIntrinsics()) { - code.AddLine("#extension GL_NV_gpu_shader5 : require"); - code.AddLine("#extension GL_NV_shader_thread_group : require"); - code.AddLine("#extension GL_NV_shader_thread_shuffle : require"); - } - // This pragma stops Nvidia's driver from over optimizing math (probably using fp16 - // operations) on places where we don't want to. - // Thanks to Ryujinx for finding this workaround. - code.AddLine("#pragma optionNV(fastmath off)"); - - code.AddNewLine(); - - code.AddLine(COMMON_DECLARATIONS); - } - - void DeclareVertex() { - if (stage != ShaderType::Vertex) { - return; - } - - DeclareVertexRedeclarations(); - } - - void DeclareGeometry() { - if (stage != ShaderType::Geometry) { - return; - } - - const auto& info = registry.GetGraphicsInfo(); - const auto input_topology = info.primitive_topology; - const auto [glsl_topology, max_vertices] = GetPrimitiveDescription(input_topology); - max_input_vertices = max_vertices; - code.AddLine("layout ({}) in;", glsl_topology); - - const auto topology = GetTopologyName(header.common3.output_topology); - const auto max_output_vertices = header.common4.max_output_vertices.Value(); - code.AddLine("layout ({}, max_vertices = {}) out;", topology, max_output_vertices); - code.AddNewLine(); - - code.AddLine("in gl_PerVertex {{"); - ++code.scope; - code.AddLine("vec4 gl_Position;"); - --code.scope; - code.AddLine("}} gl_in[];"); - - DeclareVertexRedeclarations(); - } - - void DeclareFragment() { - if (stage != ShaderType::Fragment) { - return; - } - if (ir.UsesLegacyVaryings()) { - code.AddLine("in gl_PerFragment {{"); - ++code.scope; - code.AddLine("vec4 gl_TexCoord[8];"); - code.AddLine("vec4 gl_Color;"); - code.AddLine("vec4 gl_SecondaryColor;"); - --code.scope; - code.AddLine("}};"); - } - - for (u32 rt = 0; rt < Maxwell::NumRenderTargets; ++rt) { - code.AddLine("layout (location = {}) out vec4 frag_color{};", rt, rt); - } - } - - void DeclareCompute() { - if (stage != ShaderType::Compute) { - return; - } - const auto& info = registry.GetComputeInfo(); - if (u32 size = info.shared_memory_size_in_words * 4; size > 0) { - const u32 limit = device.GetMaxComputeSharedMemorySize(); - if (size > limit) { - LOG_ERROR(Render_OpenGL, "Shared memory size {} is clamped to host's limit {}", - size, limit); - size = limit; - } - - code.AddLine("shared uint smem[{}];", size / 4); - code.AddNewLine(); - } - code.AddLine("layout (local_size_x = {}, local_size_y = {}, local_size_z = {}) in;", - info.workgroup_size[0], info.workgroup_size[1], info.workgroup_size[2]); - code.AddNewLine(); - } - - void DeclareVertexRedeclarations() { - code.AddLine("out gl_PerVertex {{"); - ++code.scope; - - auto pos_xfb = GetTransformFeedbackDecoration(Attribute::Index::Position); - if (!pos_xfb.empty()) { - pos_xfb = fmt::format("layout ({}) ", pos_xfb); - } - const char* pos_type = - FLOAT_TYPES.at(GetNumComponents(Attribute::Index::Position).value_or(4) - 1); - code.AddLine("{}{} gl_Position;", pos_xfb, pos_type); - - for (const auto attribute : ir.GetOutputAttributes()) { - if (attribute == Attribute::Index::ClipDistances0123 || - attribute == Attribute::Index::ClipDistances4567) { - code.AddLine("float gl_ClipDistance[];"); - break; - } - } - - if (stage != ShaderType::Geometry && - (stage != ShaderType::Vertex || device.HasVertexViewportLayer())) { - if (ir.UsesLayer()) { - code.AddLine("int gl_Layer;"); - } - if (ir.UsesViewportIndex()) { - code.AddLine("int gl_ViewportIndex;"); - } - } else if ((ir.UsesLayer() || ir.UsesViewportIndex()) && stage == ShaderType::Vertex && - !device.HasVertexViewportLayer()) { - LOG_ERROR( - Render_OpenGL, - "GL_ARB_shader_viewport_layer_array is not available and its required by a shader"); - } - - if (ir.UsesPointSize()) { - code.AddLine("float gl_PointSize;"); - } - - if (ir.UsesLegacyVaryings()) { - code.AddLine("vec4 gl_TexCoord[8];"); - code.AddLine("vec4 gl_FrontColor;"); - code.AddLine("vec4 gl_FrontSecondaryColor;"); - code.AddLine("vec4 gl_BackColor;"); - code.AddLine("vec4 gl_BackSecondaryColor;"); - } - - --code.scope; - code.AddLine("}};"); - code.AddNewLine(); - - if (stage == ShaderType::Geometry) { - if (ir.UsesLayer()) { - code.AddLine("out int gl_Layer;"); - } - if (ir.UsesViewportIndex()) { - code.AddLine("out int gl_ViewportIndex;"); - } - } - code.AddNewLine(); - } - - void DeclareRegisters() { - const auto& registers = ir.GetRegisters(); - for (const u32 gpr : registers) { - code.AddLine("float {} = 0.0f;", GetRegister(gpr)); - } - if (!registers.empty()) { - code.AddNewLine(); - } - } - - void DeclareCustomVariables() { - const u32 num_custom_variables = ir.GetNumCustomVariables(); - for (u32 i = 0; i < num_custom_variables; ++i) { - code.AddLine("float {} = 0.0f;", GetCustomVariable(i)); - } - if (num_custom_variables > 0) { - code.AddNewLine(); - } - } - - void DeclarePredicates() { - const auto& predicates = ir.GetPredicates(); - for (const auto pred : predicates) { - code.AddLine("bool {} = false;", GetPredicate(pred)); - } - if (!predicates.empty()) { - code.AddNewLine(); - } - } - - void DeclareLocalMemory() { - u64 local_memory_size = 0; - if (stage == ShaderType::Compute) { - local_memory_size = registry.GetComputeInfo().local_memory_size_in_words * 4ULL; - } else { - local_memory_size = header.GetLocalMemorySize(); - } - if (local_memory_size == 0) { - return; - } - const u64 element_count = Common::AlignUp(local_memory_size, 4) / 4; - code.AddLine("uint {}[{}];", GetLocalMemory(), element_count); - code.AddNewLine(); - } - - void DeclareInternalFlags() { - for (u32 flag = 0; flag < static_cast<u32>(InternalFlag::Amount); flag++) { - const auto flag_code = static_cast<InternalFlag>(flag); - code.AddLine("bool {} = false;", GetInternalFlag(flag_code)); - } - code.AddNewLine(); - } - - const char* GetInputFlags(PixelImap attribute) { - switch (attribute) { - case PixelImap::Perspective: - return "smooth"; - case PixelImap::Constant: - return "flat"; - case PixelImap::ScreenLinear: - return "noperspective"; - case PixelImap::Unused: - break; - } - UNIMPLEMENTED_MSG("Unknown attribute usage index={}", attribute); - return {}; - } - - void DeclareInputAttributes() { - if (ir.HasPhysicalAttributes()) { - const u32 num_inputs{GetNumPhysicalInputAttributes()}; - for (u32 i = 0; i < num_inputs; ++i) { - DeclareInputAttribute(ToGenericAttribute(i), true); - } - code.AddNewLine(); - return; - } - - const auto& attributes = ir.GetInputAttributes(); - for (const auto index : attributes) { - if (IsGenericAttribute(index)) { - DeclareInputAttribute(index, false); - } - } - if (!attributes.empty()) { - code.AddNewLine(); - } - } - - void DeclareInputAttribute(Attribute::Index index, bool skip_unused) { - const u32 location{GetGenericAttributeIndex(index)}; - - std::string name{GetGenericInputAttribute(index)}; - if (stage == ShaderType::Geometry) { - name = "gs_" + name + "[]"; - } - - std::string suffix_; - if (stage == ShaderType::Fragment) { - const auto input_mode{header.ps.GetPixelImap(location)}; - if (input_mode == PixelImap::Unused) { - return; - } - suffix_ = GetInputFlags(input_mode); - } - - code.AddLine("layout (location = {}) {} in vec4 {};", location, suffix_, name); - } - - void DeclareOutputAttributes() { - if (ir.HasPhysicalAttributes() && stage != ShaderType::Fragment) { - for (u32 i = 0; i < GetNumPhysicalVaryings(); ++i) { - DeclareOutputAttribute(ToGenericAttribute(i)); - } - code.AddNewLine(); - return; - } - - const auto& attributes = ir.GetOutputAttributes(); - for (const auto index : attributes) { - if (IsGenericAttribute(index)) { - DeclareOutputAttribute(index); - } - } - if (!attributes.empty()) { - code.AddNewLine(); - } - } - - std::optional<std::size_t> GetNumComponents(Attribute::Index index, u8 element = 0) const { - const u8 location = static_cast<u8>(static_cast<u32>(index) * 4 + element); - const auto it = transform_feedback.find(location); - if (it == transform_feedback.end()) { - return std::nullopt; - } - return it->second.components; - } - - std::string GetTransformFeedbackDecoration(Attribute::Index index, u8 element = 0) const { - const u8 location = static_cast<u8>(static_cast<u32>(index) * 4 + element); - const auto it = transform_feedback.find(location); - if (it == transform_feedback.end()) { - return {}; - } - - const VaryingTFB& tfb = it->second; - return fmt::format("xfb_buffer = {}, xfb_offset = {}, xfb_stride = {}", tfb.buffer, - tfb.offset, tfb.stride); - } - - void DeclareOutputAttribute(Attribute::Index index) { - static constexpr std::string_view swizzle = "xyzw"; - u8 element = 0; - while (element < 4) { - auto xfb = GetTransformFeedbackDecoration(index, element); - if (!xfb.empty()) { - xfb = fmt::format(", {}", xfb); - } - const std::size_t remainder = 4 - element; - const std::size_t num_components = GetNumComponents(index, element).value_or(remainder); - const char* const type = FLOAT_TYPES.at(num_components - 1); - - const u32 location = GetGenericAttributeIndex(index); - - GenericVaryingDescription description; - description.first_element = static_cast<u8>(element); - description.is_scalar = num_components == 1; - description.name = AppendSuffix(location, OUTPUT_ATTRIBUTE_NAME); - if (element != 0 || num_components != 4) { - const std::string_view name_swizzle = swizzle.substr(element, num_components); - description.name = fmt::format("{}_{}", description.name, name_swizzle); - } - for (std::size_t i = 0; i < num_components; ++i) { - const u8 offset = static_cast<u8>(location * 4 + element + i); - varying_description.insert({offset, description}); - } - - code.AddLine("layout (location = {}, component = {}{}) out {} {};", location, element, - xfb, type, description.name); - - element = static_cast<u8>(static_cast<std::size_t>(element) + num_components); - } - } - - void DeclareConstantBuffers() { - u32 binding = device.GetBaseBindings(stage).uniform_buffer; - for (const auto& [index, info] : ir.GetConstantBuffers()) { - const u32 num_elements = Common::DivCeil(info.GetSize(), 4 * sizeof(u32)); - const u32 size = info.IsIndirect() ? MAX_CONSTBUFFER_ELEMENTS : num_elements; - code.AddLine("layout (std140, binding = {}) uniform {} {{", binding++, - GetConstBufferBlock(index)); - code.AddLine(" uvec4 {}[{}];", GetConstBuffer(index), size); - code.AddLine("}};"); - code.AddNewLine(); - } - } - - void DeclareGlobalMemory() { - u32 binding = device.GetBaseBindings(stage).shader_storage_buffer; - for (const auto& [base, usage] : ir.GetGlobalMemory()) { - // Since we don't know how the shader will use the shader, hint the driver to disable as - // much optimizations as possible - std::string qualifier = "coherent volatile"; - if (usage.is_read && !usage.is_written) { - qualifier += " readonly"; - } else if (usage.is_written && !usage.is_read) { - qualifier += " writeonly"; - } - - code.AddLine("layout (std430, binding = {}) {} buffer {} {{", binding++, qualifier, - GetGlobalMemoryBlock(base)); - code.AddLine(" uint {}[];", GetGlobalMemory(base)); - code.AddLine("}};"); - code.AddNewLine(); - } - } - - void DeclareSamplers() { - u32 binding = device.GetBaseBindings(stage).sampler; - for (const auto& sampler : ir.GetSamplers()) { - const std::string name = GetSampler(sampler); - const std::string description = fmt::format("layout (binding = {}) uniform", binding); - binding += sampler.is_indexed ? sampler.size : 1; - - std::string sampler_type = [&]() { - if (sampler.is_buffer) { - return "samplerBuffer"; - } - switch (sampler.type) { - case TextureType::Texture1D: - return "sampler1D"; - case TextureType::Texture2D: - return "sampler2D"; - case TextureType::Texture3D: - return "sampler3D"; - case TextureType::TextureCube: - return "samplerCube"; - default: - UNREACHABLE(); - return "sampler2D"; - } - }(); - if (sampler.is_array) { - sampler_type += "Array"; - } - if (sampler.is_shadow) { - sampler_type += "Shadow"; - } - - if (!sampler.is_indexed) { - code.AddLine("{} {} {};", description, sampler_type, name); - } else { - code.AddLine("{} {} {}[{}];", description, sampler_type, name, sampler.size); - } - } - if (!ir.GetSamplers().empty()) { - code.AddNewLine(); - } - } - - void DeclarePhysicalAttributeReader() { - if (!ir.HasPhysicalAttributes()) { - return; - } - code.AddLine("float ReadPhysicalAttribute(uint physical_address) {{"); - ++code.scope; - code.AddLine("switch (physical_address) {{"); - - // Just declare generic attributes for now. - const auto num_attributes{static_cast<u32>(GetNumPhysicalInputAttributes())}; - for (u32 index = 0; index < num_attributes; ++index) { - const auto attribute{ToGenericAttribute(index)}; - for (u32 element = 0; element < 4; ++element) { - constexpr u32 generic_base = 0x80; - constexpr u32 generic_stride = 16; - constexpr u32 element_stride = 4; - const u32 address{generic_base + index * generic_stride + element * element_stride}; - - const bool declared = stage != ShaderType::Fragment || - header.ps.GetPixelImap(index) != PixelImap::Unused; - const std::string value = - declared ? ReadAttribute(attribute, element).AsFloat() : "0.0f"; - code.AddLine("case 0x{:X}U: return {};", address, value); - } - } - - code.AddLine("default: return 0;"); - - code.AddLine("}}"); - --code.scope; - code.AddLine("}}"); - code.AddNewLine(); - } - - void DeclareImages() { - u32 binding = device.GetBaseBindings(stage).image; - for (const auto& image : ir.GetImages()) { - std::string qualifier = "coherent volatile"; - if (image.is_read && !image.is_written) { - qualifier += " readonly"; - } else if (image.is_written && !image.is_read) { - qualifier += " writeonly"; - } - - const char* format = image.is_atomic ? "r32ui, " : ""; - const char* type_declaration = GetImageTypeDeclaration(image.type); - code.AddLine("layout ({}binding = {}) {} uniform uimage{} {};", format, binding++, - qualifier, type_declaration, GetImage(image)); - } - if (!ir.GetImages().empty()) { - code.AddNewLine(); - } - } - - void VisitBlock(const NodeBlock& bb) { - for (const auto& node : bb) { - Visit(node).CheckVoid(); - } - } - - Expression Visit(const Node& node) { - if (const auto operation = std::get_if<OperationNode>(&*node)) { - if (const auto amend_index = operation->GetAmendIndex()) { - Visit(ir.GetAmendNode(*amend_index)).CheckVoid(); - } - const auto operation_index = static_cast<std::size_t>(operation->GetCode()); - if (operation_index >= operation_decompilers.size()) { - UNREACHABLE_MSG("Out of bounds operation: {}", operation_index); - return {}; - } - const auto decompiler = operation_decompilers[operation_index]; - if (decompiler == nullptr) { - UNREACHABLE_MSG("Undefined operation: {}", operation_index); - return {}; - } - return (this->*decompiler)(*operation); - } - - if (const auto gpr = std::get_if<GprNode>(&*node)) { - const u32 index = gpr->GetIndex(); - if (index == Register::ZeroIndex) { - return {"0U", Type::Uint}; - } - return {GetRegister(index), Type::Float}; - } - - if (const auto cv = std::get_if<CustomVarNode>(&*node)) { - const u32 index = cv->GetIndex(); - return {GetCustomVariable(index), Type::Float}; - } - - if (const auto immediate = std::get_if<ImmediateNode>(&*node)) { - const u32 value = immediate->GetValue(); - if (value < 10) { - // For eyecandy avoid using hex numbers on single digits - return {fmt::format("{}U", immediate->GetValue()), Type::Uint}; - } - return {fmt::format("0x{:X}U", immediate->GetValue()), Type::Uint}; - } - - if (const auto predicate = std::get_if<PredicateNode>(&*node)) { - const auto value = [&]() -> std::string { - switch (const auto index = predicate->GetIndex(); index) { - case Tegra::Shader::Pred::UnusedIndex: - return "true"; - case Tegra::Shader::Pred::NeverExecute: - return "false"; - default: - return GetPredicate(index); - } - }(); - if (predicate->IsNegated()) { - return {fmt::format("!({})", value), Type::Bool}; - } - return {value, Type::Bool}; - } - - if (const auto abuf = std::get_if<AbufNode>(&*node)) { - UNIMPLEMENTED_IF_MSG(abuf->IsPhysicalBuffer() && stage == ShaderType::Geometry, - "Physical attributes in geometry shaders are not implemented"); - if (abuf->IsPhysicalBuffer()) { - return {fmt::format("ReadPhysicalAttribute({})", - Visit(abuf->GetPhysicalAddress()).AsUint()), - Type::Float}; - } - return ReadAttribute(abuf->GetIndex(), abuf->GetElement(), abuf->GetBuffer()); - } - - if (const auto cbuf = std::get_if<CbufNode>(&*node)) { - const Node offset = cbuf->GetOffset(); - - if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) { - // Direct access - const u32 offset_imm = immediate->GetValue(); - ASSERT_MSG(offset_imm % 4 == 0, "Unaligned cbuf direct access"); - return {fmt::format("{}[{}][{}]", GetConstBuffer(cbuf->GetIndex()), - offset_imm / (4 * 4), (offset_imm / 4) % 4), - Type::Uint}; - } - - // Indirect access - const std::string final_offset = code.GenerateTemporary(); - code.AddLine("uint {} = {} >> 2;", final_offset, Visit(offset).AsUint()); - - if (!device.HasComponentIndexingBug()) { - return {fmt::format("{}[{} >> 2][{} & 3]", GetConstBuffer(cbuf->GetIndex()), - final_offset, final_offset), - Type::Uint}; - } - - // AMD's proprietary GLSL compiler emits ill code for variable component access. - // To bypass this driver bug generate 4 ifs, one per each component. - const std::string pack = code.GenerateTemporary(); - code.AddLine("uvec4 {} = {}[{} >> 2];", pack, GetConstBuffer(cbuf->GetIndex()), - final_offset); - - const std::string result = code.GenerateTemporary(); - code.AddLine("uint {};", result); - for (u32 swizzle = 0; swizzle < 4; ++swizzle) { - code.AddLine("if (({} & 3) == {}) {} = {}{};", final_offset, swizzle, result, pack, - GetSwizzle(swizzle)); - } - return {result, Type::Uint}; - } - - if (const auto gmem = std::get_if<GmemNode>(&*node)) { - const std::string real = Visit(gmem->GetRealAddress()).AsUint(); - const std::string base = Visit(gmem->GetBaseAddress()).AsUint(); - const std::string final_offset = fmt::format("({} - {}) >> 2", real, base); - return {fmt::format("{}[{}]", GetGlobalMemory(gmem->GetDescriptor()), final_offset), - Type::Uint}; - } - - if (const auto lmem = std::get_if<LmemNode>(&*node)) { - return { - fmt::format("{}[{} >> 2]", GetLocalMemory(), Visit(lmem->GetAddress()).AsUint()), - Type::Uint}; - } - - if (const auto smem = std::get_if<SmemNode>(&*node)) { - return {fmt::format("smem[{} >> 2]", Visit(smem->GetAddress()).AsUint()), Type::Uint}; - } - - if (const auto internal_flag = std::get_if<InternalFlagNode>(&*node)) { - return {GetInternalFlag(internal_flag->GetFlag()), Type::Bool}; - } - - if (const auto conditional = std::get_if<ConditionalNode>(&*node)) { - if (const auto amend_index = conditional->GetAmendIndex()) { - Visit(ir.GetAmendNode(*amend_index)).CheckVoid(); - } - // It's invalid to call conditional on nested nodes, use an operation instead - code.AddLine("if ({}) {{", Visit(conditional->GetCondition()).AsBool()); - ++code.scope; - - VisitBlock(conditional->GetCode()); - - --code.scope; - code.AddLine("}}"); - return {}; - } - - if (const auto comment = std::get_if<CommentNode>(&*node)) { - code.AddLine("// " + comment->GetText()); - return {}; - } - - UNREACHABLE(); - return {}; - } - - Expression ReadAttribute(Attribute::Index attribute, u32 element, const Node& buffer = {}) { - const auto GeometryPass = [&](std::string_view name) { - if (stage == ShaderType::Geometry && buffer) { - // TODO(Rodrigo): Guard geometry inputs against out of bound reads. Some games - // set an 0x80000000 index for those and the shader fails to build. Find out why - // this happens and what's its intent. - return fmt::format("gs_{}[{} % {}]", name, Visit(buffer).AsUint(), - max_input_vertices.value()); - } - return std::string(name); - }; - - switch (attribute) { - case Attribute::Index::Position: - switch (stage) { - case ShaderType::Geometry: - return {fmt::format("gl_in[{}].gl_Position{}", Visit(buffer).AsUint(), - GetSwizzle(element)), - Type::Float}; - case ShaderType::Fragment: - return {"gl_FragCoord"s + GetSwizzle(element), Type::Float}; - default: - UNREACHABLE(); - return {"0", Type::Int}; - } - case Attribute::Index::FrontColor: - return {"gl_Color"s + GetSwizzle(element), Type::Float}; - case Attribute::Index::FrontSecondaryColor: - return {"gl_SecondaryColor"s + GetSwizzle(element), Type::Float}; - case Attribute::Index::PointCoord: - switch (element) { - case 0: - return {"gl_PointCoord.x", Type::Float}; - case 1: - return {"gl_PointCoord.y", Type::Float}; - case 2: - case 3: - return {"0.0f", Type::Float}; - } - UNREACHABLE(); - return {"0", Type::Int}; - case Attribute::Index::TessCoordInstanceIDVertexID: - // TODO(Subv): Find out what the values are for the first two elements when inside a - // vertex shader, and what's the value of the fourth element when inside a Tess Eval - // shader. - ASSERT(stage == ShaderType::Vertex); - switch (element) { - case 2: - // Config pack's first value is instance_id. - return {"gl_InstanceID", Type::Int}; - case 3: - return {"gl_VertexID", Type::Int}; - } - UNIMPLEMENTED_MSG("Unmanaged TessCoordInstanceIDVertexID element={}", element); - return {"0", Type::Int}; - case Attribute::Index::FrontFacing: - // TODO(Subv): Find out what the values are for the other elements. - ASSERT(stage == ShaderType::Fragment); - switch (element) { - case 3: - return {"(gl_FrontFacing ? -1 : 0)", Type::Int}; - } - UNIMPLEMENTED_MSG("Unmanaged FrontFacing element={}", element); - return {"0", Type::Int}; - default: - if (IsGenericAttribute(attribute)) { - return {GeometryPass(GetGenericInputAttribute(attribute)) + GetSwizzle(element), - Type::Float}; - } - if (IsLegacyTexCoord(attribute)) { - UNIMPLEMENTED_IF(stage == ShaderType::Geometry); - return {fmt::format("gl_TexCoord[{}]{}", GetLegacyTexCoordIndex(attribute), - GetSwizzle(element)), - Type::Float}; - } - break; - } - UNIMPLEMENTED_MSG("Unhandled input attribute: {}", attribute); - return {"0", Type::Int}; - } - - Expression ApplyPrecise(Operation operation, std::string value, Type type) { - if (!IsPrecise(operation)) { - return {std::move(value), type}; - } - // Old Nvidia drivers have a bug with precise and texture sampling. These are more likely to - // be found in fragment shaders, so we disable precise there. There are vertex shaders that - // also fail to build but nobody seems to care about those. - // Note: Only bugged drivers will skip precise. - const bool disable_precise = device.HasPreciseBug() && stage == ShaderType::Fragment; - - std::string temporary = code.GenerateTemporary(); - code.AddLine("{}{} {} = {};", disable_precise ? "" : "precise ", GetTypeString(type), - temporary, value); - return {std::move(temporary), type}; - } - - Expression VisitOperand(Operation operation, std::size_t operand_index) { - const auto& operand = operation[operand_index]; - const bool parent_precise = IsPrecise(operation); - const bool child_precise = IsPrecise(operand); - const bool child_trivial = !std::holds_alternative<OperationNode>(*operand); - if (!parent_precise || child_precise || child_trivial) { - return Visit(operand); - } - - Expression value = Visit(operand); - std::string temporary = code.GenerateTemporary(); - code.AddLine("{} {} = {};", GetTypeString(value.GetType()), temporary, value.GetCode()); - return {std::move(temporary), value.GetType()}; - } - - std::optional<Expression> GetOutputAttribute(const AbufNode* abuf) { - const u32 element = abuf->GetElement(); - switch (const auto attribute = abuf->GetIndex()) { - case Attribute::Index::Position: - return {{"gl_Position"s + GetSwizzle(element), Type::Float}}; - case Attribute::Index::LayerViewportPointSize: - switch (element) { - case 0: - UNIMPLEMENTED(); - return std::nullopt; - case 1: - if (stage == ShaderType::Vertex && !device.HasVertexViewportLayer()) { - return std::nullopt; - } - return {{"gl_Layer", Type::Int}}; - case 2: - if (stage == ShaderType::Vertex && !device.HasVertexViewportLayer()) { - return std::nullopt; - } - return {{"gl_ViewportIndex", Type::Int}}; - case 3: - return {{"gl_PointSize", Type::Float}}; - } - return std::nullopt; - case Attribute::Index::FrontColor: - return {{"gl_FrontColor"s + GetSwizzle(element), Type::Float}}; - case Attribute::Index::FrontSecondaryColor: - return {{"gl_FrontSecondaryColor"s + GetSwizzle(element), Type::Float}}; - case Attribute::Index::BackColor: - return {{"gl_BackColor"s + GetSwizzle(element), Type::Float}}; - case Attribute::Index::BackSecondaryColor: - return {{"gl_BackSecondaryColor"s + GetSwizzle(element), Type::Float}}; - case Attribute::Index::ClipDistances0123: - return {{fmt::format("gl_ClipDistance[{}]", element), Type::Float}}; - case Attribute::Index::ClipDistances4567: - return {{fmt::format("gl_ClipDistance[{}]", element + 4), Type::Float}}; - default: - if (IsGenericAttribute(attribute)) { - return {{GetGenericOutputAttribute(attribute, element), Type::Float}}; - } - if (IsLegacyTexCoord(attribute)) { - return {{fmt::format("gl_TexCoord[{}]{}", GetLegacyTexCoordIndex(attribute), - GetSwizzle(element)), - Type::Float}}; - } - UNIMPLEMENTED_MSG("Unhandled output attribute: {}", attribute); - return std::nullopt; - } - } - - Expression GenerateUnary(Operation operation, std::string_view func, Type result_type, - Type type_a) { - std::string op_str = fmt::format("{}({})", func, VisitOperand(operation, 0).As(type_a)); - return ApplyPrecise(operation, std::move(op_str), result_type); - } - - Expression GenerateBinaryInfix(Operation operation, std::string_view func, Type result_type, - Type type_a, Type type_b) { - const std::string op_a = VisitOperand(operation, 0).As(type_a); - const std::string op_b = VisitOperand(operation, 1).As(type_b); - std::string op_str = fmt::format("({} {} {})", op_a, func, op_b); - - return ApplyPrecise(operation, std::move(op_str), result_type); - } - - Expression GenerateBinaryCall(Operation operation, std::string_view func, Type result_type, - Type type_a, Type type_b) { - const std::string op_a = VisitOperand(operation, 0).As(type_a); - const std::string op_b = VisitOperand(operation, 1).As(type_b); - std::string op_str = fmt::format("{}({}, {})", func, op_a, op_b); - - return ApplyPrecise(operation, std::move(op_str), result_type); - } - - Expression GenerateTernary(Operation operation, std::string_view func, Type result_type, - Type type_a, Type type_b, Type type_c) { - const std::string op_a = VisitOperand(operation, 0).As(type_a); - const std::string op_b = VisitOperand(operation, 1).As(type_b); - const std::string op_c = VisitOperand(operation, 2).As(type_c); - std::string op_str = fmt::format("{}({}, {}, {})", func, op_a, op_b, op_c); - - return ApplyPrecise(operation, std::move(op_str), result_type); - } - - Expression GenerateQuaternary(Operation operation, const std::string& func, Type result_type, - Type type_a, Type type_b, Type type_c, Type type_d) { - const std::string op_a = VisitOperand(operation, 0).As(type_a); - const std::string op_b = VisitOperand(operation, 1).As(type_b); - const std::string op_c = VisitOperand(operation, 2).As(type_c); - const std::string op_d = VisitOperand(operation, 3).As(type_d); - std::string op_str = fmt::format("{}({}, {}, {}, {})", func, op_a, op_b, op_c, op_d); - - return ApplyPrecise(operation, std::move(op_str), result_type); - } - - std::string GenerateTexture(Operation operation, const std::string& function_suffix, - const std::vector<TextureIR>& extras, bool separate_dc = false) { - constexpr std::array coord_constructors = {"float", "vec2", "vec3", "vec4"}; - - const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); - ASSERT(meta); - - const std::size_t count = operation.GetOperandsCount(); - const bool has_array = meta->sampler.is_array; - const bool has_shadow = meta->sampler.is_shadow; - const bool workaround_lod_array_shadow_as_grad = - !device.HasTextureShadowLod() && function_suffix == "Lod" && meta->sampler.is_shadow && - ((meta->sampler.type == TextureType::Texture2D && meta->sampler.is_array) || - meta->sampler.type == TextureType::TextureCube); - - std::string expr = "texture"; - - if (workaround_lod_array_shadow_as_grad) { - expr += "Grad"; - } else { - expr += function_suffix; - } - - if (!meta->aoffi.empty()) { - expr += "Offset"; - } else if (!meta->ptp.empty()) { - expr += "Offsets"; - } - if (!meta->sampler.is_indexed) { - expr += '(' + GetSampler(meta->sampler) + ", "; - } else { - expr += '(' + GetSampler(meta->sampler) + '[' + Visit(meta->index).AsUint() + "], "; - } - expr += coord_constructors.at(count + (has_array ? 1 : 0) + - (has_shadow && !separate_dc ? 1 : 0) - 1); - expr += '('; - for (std::size_t i = 0; i < count; ++i) { - expr += Visit(operation[i]).AsFloat(); - - const std::size_t next = i + 1; - if (next < count) - expr += ", "; - } - if (has_array) { - expr += ", float(" + Visit(meta->array).AsInt() + ')'; - } - if (has_shadow) { - if (separate_dc) { - expr += "), " + Visit(meta->depth_compare).AsFloat(); - } else { - expr += ", " + Visit(meta->depth_compare).AsFloat() + ')'; - } - } else { - expr += ')'; - } - - if (workaround_lod_array_shadow_as_grad) { - switch (meta->sampler.type) { - case TextureType::Texture2D: - return expr + ", vec2(0.0), vec2(0.0))"; - case TextureType::TextureCube: - return expr + ", vec3(0.0), vec3(0.0))"; - default: - UNREACHABLE(); - break; - } - } - - for (const auto& variant : extras) { - if (const auto argument = std::get_if<TextureArgument>(&variant)) { - expr += GenerateTextureArgument(*argument); - } else if (std::holds_alternative<TextureOffset>(variant)) { - if (!meta->aoffi.empty()) { - expr += GenerateTextureAoffi(meta->aoffi); - } else if (!meta->ptp.empty()) { - expr += GenerateTexturePtp(meta->ptp); - } - } else if (std::holds_alternative<TextureDerivates>(variant)) { - expr += GenerateTextureDerivates(meta->derivates); - } else { - UNREACHABLE(); - } - } - - return expr + ')'; - } - - std::string GenerateTextureArgument(const TextureArgument& argument) { - const auto& [type, operand] = argument; - if (operand == nullptr) { - return {}; - } - - std::string expr = ", "; - switch (type) { - case Type::Int: - if (const auto immediate = std::get_if<ImmediateNode>(&*operand)) { - // Inline the string as an immediate integer in GLSL (some extra arguments are - // required to be constant) - expr += std::to_string(static_cast<s32>(immediate->GetValue())); - } else { - expr += Visit(operand).AsInt(); - } - break; - case Type::Float: - expr += Visit(operand).AsFloat(); - break; - default: { - const auto type_int = static_cast<u32>(type); - UNIMPLEMENTED_MSG("Unimplemented extra type={}", type_int); - expr += '0'; - break; - } - } - return expr; - } - - std::string ReadTextureOffset(const Node& value) { - if (const auto immediate = std::get_if<ImmediateNode>(&*value)) { - // Inline the string as an immediate integer in GLSL (AOFFI arguments are required - // to be constant by the standard). - return std::to_string(static_cast<s32>(immediate->GetValue())); - } else if (device.HasVariableAoffi()) { - // Avoid using variable AOFFI on unsupported devices. - return Visit(value).AsInt(); - } else { - // Insert 0 on devices not supporting variable AOFFI. - return "0"; - } - } - - std::string GenerateTextureAoffi(const std::vector<Node>& aoffi) { - if (aoffi.empty()) { - return {}; - } - constexpr std::array coord_constructors = {"int", "ivec2", "ivec3"}; - std::string expr = ", "; - expr += coord_constructors.at(aoffi.size() - 1); - expr += '('; - - for (std::size_t index = 0; index < aoffi.size(); ++index) { - expr += ReadTextureOffset(aoffi.at(index)); - if (index + 1 < aoffi.size()) { - expr += ", "; - } - } - expr += ')'; - - return expr; - } - - std::string GenerateTexturePtp(const std::vector<Node>& ptp) { - static constexpr std::size_t num_vectors = 4; - ASSERT(ptp.size() == num_vectors * 2); - - std::string expr = ", ivec2[]("; - for (std::size_t vector = 0; vector < num_vectors; ++vector) { - const bool has_next = vector + 1 < num_vectors; - expr += fmt::format("ivec2({}, {}){}", ReadTextureOffset(ptp.at(vector * 2)), - ReadTextureOffset(ptp.at(vector * 2 + 1)), has_next ? ", " : ""); - } - expr += ')'; - return expr; - } - - std::string GenerateTextureDerivates(const std::vector<Node>& derivates) { - if (derivates.empty()) { - return {}; - } - constexpr std::array coord_constructors = {"float", "vec2", "vec3"}; - std::string expr = ", "; - const std::size_t components = derivates.size() / 2; - std::string dx = coord_constructors.at(components - 1); - std::string dy = coord_constructors.at(components - 1); - dx += '('; - dy += '('; - - for (std::size_t index = 0; index < components; ++index) { - const auto& operand_x{derivates.at(index * 2)}; - const auto& operand_y{derivates.at(index * 2 + 1)}; - dx += Visit(operand_x).AsFloat(); - dy += Visit(operand_y).AsFloat(); - - if (index + 1 < components) { - dx += ", "; - dy += ", "; - } - } - dx += ')'; - dy += ')'; - expr += dx + ", " + dy; - - return expr; - } - - std::string BuildIntegerCoordinates(Operation operation) { - constexpr std::array constructors{"int(", "ivec2(", "ivec3(", "ivec4("}; - const std::size_t coords_count{operation.GetOperandsCount()}; - std::string expr = constructors.at(coords_count - 1); - for (std::size_t i = 0; i < coords_count; ++i) { - expr += VisitOperand(operation, i).AsInt(); - if (i + 1 < coords_count) { - expr += ", "; - } - } - expr += ')'; - return expr; - } - - std::string BuildImageValues(Operation operation) { - constexpr std::array constructors{"uint", "uvec2", "uvec3", "uvec4"}; - const auto& meta{std::get<MetaImage>(operation.GetMeta())}; - - const std::size_t values_count{meta.values.size()}; - std::string expr = fmt::format("{}(", constructors.at(values_count - 1)); - for (std::size_t i = 0; i < values_count; ++i) { - expr += Visit(meta.values.at(i)).AsUint(); - if (i + 1 < values_count) { - expr += ", "; - } - } - expr += ')'; - return expr; - } - - Expression Assign(Operation operation) { - const Node& dest = operation[0]; - const Node& src = operation[1]; - - Expression target; - if (const auto gpr = std::get_if<GprNode>(&*dest)) { - if (gpr->GetIndex() == Register::ZeroIndex) { - // Writing to Register::ZeroIndex is a no op but we still have to visit the source - // as it might have side effects. - code.AddLine("{};", Visit(src).GetCode()); - return {}; - } - target = {GetRegister(gpr->GetIndex()), Type::Float}; - } else if (const auto abuf = std::get_if<AbufNode>(&*dest)) { - UNIMPLEMENTED_IF(abuf->IsPhysicalBuffer()); - auto output = GetOutputAttribute(abuf); - if (!output) { - return {}; - } - target = std::move(*output); - } else if (const auto lmem = std::get_if<LmemNode>(&*dest)) { - target = { - fmt::format("{}[{} >> 2]", GetLocalMemory(), Visit(lmem->GetAddress()).AsUint()), - Type::Uint}; - } else if (const auto smem = std::get_if<SmemNode>(&*dest)) { - ASSERT(stage == ShaderType::Compute); - target = {fmt::format("smem[{} >> 2]", Visit(smem->GetAddress()).AsUint()), Type::Uint}; - } else if (const auto gmem = std::get_if<GmemNode>(&*dest)) { - const std::string real = Visit(gmem->GetRealAddress()).AsUint(); - const std::string base = Visit(gmem->GetBaseAddress()).AsUint(); - const std::string final_offset = fmt::format("({} - {}) >> 2", real, base); - target = {fmt::format("{}[{}]", GetGlobalMemory(gmem->GetDescriptor()), final_offset), - Type::Uint}; - } else if (const auto cv = std::get_if<CustomVarNode>(&*dest)) { - target = {GetCustomVariable(cv->GetIndex()), Type::Float}; - } else { - UNREACHABLE_MSG("Assign called without a proper target"); - } - - code.AddLine("{} = {};", target.GetCode(), Visit(src).As(target.GetType())); - return {}; - } - - template <Type type> - Expression Add(Operation operation) { - return GenerateBinaryInfix(operation, "+", type, type, type); - } - - template <Type type> - Expression Mul(Operation operation) { - return GenerateBinaryInfix(operation, "*", type, type, type); - } - - template <Type type> - Expression Div(Operation operation) { - return GenerateBinaryInfix(operation, "/", type, type, type); - } - - template <Type type> - Expression Fma(Operation operation) { - return GenerateTernary(operation, "fma", type, type, type, type); - } - - template <Type type> - Expression Negate(Operation operation) { - return GenerateUnary(operation, "-", type, type); - } - - template <Type type> - Expression Absolute(Operation operation) { - return GenerateUnary(operation, "abs", type, type); - } - - Expression FClamp(Operation operation) { - return GenerateTernary(operation, "clamp", Type::Float, Type::Float, Type::Float, - Type::Float); - } - - Expression FCastHalf0(Operation operation) { - return {fmt::format("({})[0]", VisitOperand(operation, 0).AsHalfFloat()), Type::Float}; - } - - Expression FCastHalf1(Operation operation) { - return {fmt::format("({})[1]", VisitOperand(operation, 0).AsHalfFloat()), Type::Float}; - } - - template <Type type> - Expression Min(Operation operation) { - return GenerateBinaryCall(operation, "min", type, type, type); - } - - template <Type type> - Expression Max(Operation operation) { - return GenerateBinaryCall(operation, "max", type, type, type); - } - - Expression Select(Operation operation) { - const std::string condition = Visit(operation[0]).AsBool(); - const std::string true_case = Visit(operation[1]).AsUint(); - const std::string false_case = Visit(operation[2]).AsUint(); - std::string op_str = fmt::format("({} ? {} : {})", condition, true_case, false_case); - - return ApplyPrecise(operation, std::move(op_str), Type::Uint); - } - - Expression FCos(Operation operation) { - return GenerateUnary(operation, "cos", Type::Float, Type::Float); - } - - Expression FSin(Operation operation) { - return GenerateUnary(operation, "sin", Type::Float, Type::Float); - } - - Expression FExp2(Operation operation) { - return GenerateUnary(operation, "exp2", Type::Float, Type::Float); - } - - Expression FLog2(Operation operation) { - return GenerateUnary(operation, "log2", Type::Float, Type::Float); - } - - Expression FInverseSqrt(Operation operation) { - return GenerateUnary(operation, "inversesqrt", Type::Float, Type::Float); - } - - Expression FSqrt(Operation operation) { - return GenerateUnary(operation, "sqrt", Type::Float, Type::Float); - } - - Expression FRoundEven(Operation operation) { - return GenerateUnary(operation, "roundEven", Type::Float, Type::Float); - } - - Expression FFloor(Operation operation) { - return GenerateUnary(operation, "floor", Type::Float, Type::Float); - } - - Expression FCeil(Operation operation) { - return GenerateUnary(operation, "ceil", Type::Float, Type::Float); - } - - Expression FTrunc(Operation operation) { - return GenerateUnary(operation, "trunc", Type::Float, Type::Float); - } - - template <Type type> - Expression FCastInteger(Operation operation) { - return GenerateUnary(operation, "float", Type::Float, type); - } - - Expression FSwizzleAdd(Operation operation) { - const std::string op_a = VisitOperand(operation, 0).AsFloat(); - const std::string op_b = VisitOperand(operation, 1).AsFloat(); - - if (!device.HasShaderBallot()) { - LOG_ERROR(Render_OpenGL, "Shader ballot is unavailable but required by the shader"); - return {fmt::format("{} + {}", op_a, op_b), Type::Float}; - } - - const std::string instr_mask = VisitOperand(operation, 2).AsUint(); - const std::string mask = code.GenerateTemporary(); - code.AddLine("uint {} = ({} >> ((gl_SubGroupInvocationARB & 3) << 1)) & 3;", mask, - instr_mask); - - const std::string modifier_a = fmt::format("fswzadd_modifiers_a[{}]", mask); - const std::string modifier_b = fmt::format("fswzadd_modifiers_b[{}]", mask); - return {fmt::format("(({} * {}) + ({} * {}))", op_a, modifier_a, op_b, modifier_b), - Type::Float}; - } - - Expression ICastFloat(Operation operation) { - return GenerateUnary(operation, "int", Type::Int, Type::Float); - } - - Expression ICastUnsigned(Operation operation) { - return GenerateUnary(operation, "int", Type::Int, Type::Uint); - } - - template <Type type> - Expression LogicalShiftLeft(Operation operation) { - return GenerateBinaryInfix(operation, "<<", type, type, Type::Uint); - } - - Expression ILogicalShiftRight(Operation operation) { - const std::string op_a = VisitOperand(operation, 0).AsUint(); - const std::string op_b = VisitOperand(operation, 1).AsUint(); - std::string op_str = fmt::format("int({} >> {})", op_a, op_b); - - return ApplyPrecise(operation, std::move(op_str), Type::Int); - } - - Expression IArithmeticShiftRight(Operation operation) { - return GenerateBinaryInfix(operation, ">>", Type::Int, Type::Int, Type::Uint); - } - - template <Type type> - Expression BitwiseAnd(Operation operation) { - return GenerateBinaryInfix(operation, "&", type, type, type); - } - - template <Type type> - Expression BitwiseOr(Operation operation) { - return GenerateBinaryInfix(operation, "|", type, type, type); - } - - template <Type type> - Expression BitwiseXor(Operation operation) { - return GenerateBinaryInfix(operation, "^", type, type, type); - } - - template <Type type> - Expression BitwiseNot(Operation operation) { - return GenerateUnary(operation, "~", type, type); - } - - Expression UCastFloat(Operation operation) { - return GenerateUnary(operation, "uint", Type::Uint, Type::Float); - } - - Expression UCastSigned(Operation operation) { - return GenerateUnary(operation, "uint", Type::Uint, Type::Int); - } - - Expression UShiftRight(Operation operation) { - return GenerateBinaryInfix(operation, ">>", Type::Uint, Type::Uint, Type::Uint); - } - - template <Type type> - Expression BitfieldInsert(Operation operation) { - return GenerateQuaternary(operation, "bitfieldInsert", type, type, type, Type::Int, - Type::Int); - } - - template <Type type> - Expression BitfieldExtract(Operation operation) { - return GenerateTernary(operation, "bitfieldExtract", type, type, Type::Int, Type::Int); - } - - template <Type type> - Expression BitCount(Operation operation) { - return GenerateUnary(operation, "bitCount", type, type); - } - - template <Type type> - Expression BitMSB(Operation operation) { - return GenerateUnary(operation, "findMSB", type, type); - } - - Expression HNegate(Operation operation) { - const auto GetNegate = [&](std::size_t index) { - return VisitOperand(operation, index).AsBool() + " ? -1 : 1"; - }; - return {fmt::format("({} * vec2({}, {}))", VisitOperand(operation, 0).AsHalfFloat(), - GetNegate(1), GetNegate(2)), - Type::HalfFloat}; - } - - Expression HClamp(Operation operation) { - const std::string value = VisitOperand(operation, 0).AsHalfFloat(); - const std::string min = VisitOperand(operation, 1).AsFloat(); - const std::string max = VisitOperand(operation, 2).AsFloat(); - std::string clamped = fmt::format("clamp({}, vec2({}), vec2({}))", value, min, max); - - return ApplyPrecise(operation, std::move(clamped), Type::HalfFloat); - } - - Expression HCastFloat(Operation operation) { - return {fmt::format("vec2({}, 0.0f)", VisitOperand(operation, 0).AsFloat()), - Type::HalfFloat}; - } - - Expression HUnpack(Operation operation) { - Expression operand = VisitOperand(operation, 0); - switch (std::get<Tegra::Shader::HalfType>(operation.GetMeta())) { - case Tegra::Shader::HalfType::H0_H1: - return operand; - case Tegra::Shader::HalfType::F32: - return {fmt::format("vec2({})", operand.AsFloat()), Type::HalfFloat}; - case Tegra::Shader::HalfType::H0_H0: - return {fmt::format("vec2({}[0])", operand.AsHalfFloat()), Type::HalfFloat}; - case Tegra::Shader::HalfType::H1_H1: - return {fmt::format("vec2({}[1])", operand.AsHalfFloat()), Type::HalfFloat}; - } - UNREACHABLE(); - return {"0", Type::Int}; - } - - Expression HMergeF32(Operation operation) { - return {fmt::format("float({}[0])", VisitOperand(operation, 0).AsHalfFloat()), Type::Float}; - } - - Expression HMergeH0(Operation operation) { - const std::string dest = VisitOperand(operation, 0).AsUint(); - const std::string src = VisitOperand(operation, 1).AsUint(); - return {fmt::format("vec2(unpackHalf2x16({}).x, unpackHalf2x16({}).y)", src, dest), - Type::HalfFloat}; - } - - Expression HMergeH1(Operation operation) { - const std::string dest = VisitOperand(operation, 0).AsUint(); - const std::string src = VisitOperand(operation, 1).AsUint(); - return {fmt::format("vec2(unpackHalf2x16({}).x, unpackHalf2x16({}).y)", dest, src), - Type::HalfFloat}; - } - - Expression HPack2(Operation operation) { - return {fmt::format("vec2({}, {})", VisitOperand(operation, 0).AsFloat(), - VisitOperand(operation, 1).AsFloat()), - Type::HalfFloat}; - } - - template <const std::string_view& op, Type type, bool unordered = false> - Expression Comparison(Operation operation) { - static_assert(!unordered || type == Type::Float); - - Expression expr = GenerateBinaryInfix(operation, op, Type::Bool, type, type); - - if constexpr (op.compare("!=") == 0 && type == Type::Float && !unordered) { - // GLSL's operator!=(float, float) doesn't seem be ordered. This happens on both AMD's - // and Nvidia's proprietary stacks. Manually force an ordered comparison. - return {fmt::format("({} && !isnan({}) && !isnan({}))", expr.AsBool(), - VisitOperand(operation, 0).AsFloat(), - VisitOperand(operation, 1).AsFloat()), - Type::Bool}; - } - if constexpr (!unordered) { - return expr; - } - // Unordered comparisons are always true for NaN operands. - return {fmt::format("({} || isnan({}) || isnan({}))", expr.AsBool(), - VisitOperand(operation, 0).AsFloat(), - VisitOperand(operation, 1).AsFloat()), - Type::Bool}; - } - - Expression FOrdered(Operation operation) { - return {fmt::format("(!isnan({}) && !isnan({}))", VisitOperand(operation, 0).AsFloat(), - VisitOperand(operation, 1).AsFloat()), - Type::Bool}; - } - - Expression FUnordered(Operation operation) { - return {fmt::format("(isnan({}) || isnan({}))", VisitOperand(operation, 0).AsFloat(), - VisitOperand(operation, 1).AsFloat()), - Type::Bool}; - } - - Expression LogicalAddCarry(Operation operation) { - const std::string carry = code.GenerateTemporary(); - code.AddLine("uint {};", carry); - code.AddLine("uaddCarry({}, {}, {});", VisitOperand(operation, 0).AsUint(), - VisitOperand(operation, 1).AsUint(), carry); - return {fmt::format("({} != 0)", carry), Type::Bool}; - } - - Expression LogicalAssign(Operation operation) { - const Node& dest = operation[0]; - const Node& src = operation[1]; - - std::string target; - - if (const auto pred = std::get_if<PredicateNode>(&*dest)) { - ASSERT_MSG(!pred->IsNegated(), "Negating logical assignment"); - - const auto index = pred->GetIndex(); - switch (index) { - case Tegra::Shader::Pred::NeverExecute: - case Tegra::Shader::Pred::UnusedIndex: - // Writing to these predicates is a no-op - return {}; - } - target = GetPredicate(index); - } else if (const auto flag = std::get_if<InternalFlagNode>(&*dest)) { - target = GetInternalFlag(flag->GetFlag()); - } - - code.AddLine("{} = {};", target, Visit(src).AsBool()); - return {}; - } - - Expression LogicalAnd(Operation operation) { - return GenerateBinaryInfix(operation, "&&", Type::Bool, Type::Bool, Type::Bool); - } - - Expression LogicalOr(Operation operation) { - return GenerateBinaryInfix(operation, "||", Type::Bool, Type::Bool, Type::Bool); - } - - Expression LogicalXor(Operation operation) { - return GenerateBinaryInfix(operation, "^^", Type::Bool, Type::Bool, Type::Bool); - } - - Expression LogicalNegate(Operation operation) { - return GenerateUnary(operation, "!", Type::Bool, Type::Bool); - } - - Expression LogicalPick2(Operation operation) { - return {fmt::format("{}[{}]", VisitOperand(operation, 0).AsBool2(), - VisitOperand(operation, 1).AsUint()), - Type::Bool}; - } - - Expression LogicalAnd2(Operation operation) { - return GenerateUnary(operation, "all", Type::Bool, Type::Bool2); - } - - template <bool with_nan> - Expression GenerateHalfComparison(Operation operation, std::string_view compare_op) { - Expression comparison = GenerateBinaryCall(operation, compare_op, Type::Bool2, - Type::HalfFloat, Type::HalfFloat); - if constexpr (!with_nan) { - return comparison; - } - return {fmt::format("HalfFloatNanComparison({}, {}, {})", comparison.AsBool2(), - VisitOperand(operation, 0).AsHalfFloat(), - VisitOperand(operation, 1).AsHalfFloat()), - Type::Bool2}; - } - - template <bool with_nan> - Expression Logical2HLessThan(Operation operation) { - return GenerateHalfComparison<with_nan>(operation, "lessThan"); - } - - template <bool with_nan> - Expression Logical2HEqual(Operation operation) { - return GenerateHalfComparison<with_nan>(operation, "equal"); - } - - template <bool with_nan> - Expression Logical2HLessEqual(Operation operation) { - return GenerateHalfComparison<with_nan>(operation, "lessThanEqual"); - } - - template <bool with_nan> - Expression Logical2HGreaterThan(Operation operation) { - return GenerateHalfComparison<with_nan>(operation, "greaterThan"); - } - - template <bool with_nan> - Expression Logical2HNotEqual(Operation operation) { - return GenerateHalfComparison<with_nan>(operation, "notEqual"); - } - - template <bool with_nan> - Expression Logical2HGreaterEqual(Operation operation) { - return GenerateHalfComparison<with_nan>(operation, "greaterThanEqual"); - } - - Expression Texture(Operation operation) { - const auto meta = std::get<MetaTexture>(operation.GetMeta()); - const bool separate_dc = meta.sampler.type == TextureType::TextureCube && - meta.sampler.is_array && meta.sampler.is_shadow; - // TODO: Replace this with an array and make GenerateTexture use C++20 std::span - const std::vector<TextureIR> extras{ - TextureOffset{}, - TextureArgument{Type::Float, meta.bias}, - }; - std::string expr = GenerateTexture(operation, "", extras, separate_dc); - if (meta.sampler.is_shadow) { - expr = fmt::format("vec4({})", expr); - } - return {expr + GetSwizzle(meta.element), Type::Float}; - } - - Expression TextureLod(Operation operation) { - const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); - ASSERT(meta); - - std::string expr{}; - - if (!device.HasTextureShadowLod() && meta->sampler.is_shadow && - ((meta->sampler.type == TextureType::Texture2D && meta->sampler.is_array) || - meta->sampler.type == TextureType::TextureCube)) { - LOG_ERROR(Render_OpenGL, - "Device lacks GL_EXT_texture_shadow_lod, using textureGrad as a workaround"); - expr = GenerateTexture(operation, "Lod", {}); - } else { - expr = GenerateTexture(operation, "Lod", - {TextureArgument{Type::Float, meta->lod}, TextureOffset{}}); - } - - if (meta->sampler.is_shadow) { - expr = "vec4(" + expr + ')'; - } - return {expr + GetSwizzle(meta->element), Type::Float}; - } - - Expression TextureGather(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - - const auto type = meta.sampler.is_shadow ? Type::Float : Type::Int; - const bool separate_dc = meta.sampler.is_shadow; - - std::vector<TextureIR> ir_; - if (meta.sampler.is_shadow) { - ir_ = {TextureOffset{}}; - } else { - ir_ = {TextureOffset{}, TextureArgument{type, meta.component}}; - } - return {GenerateTexture(operation, "Gather", ir_, separate_dc) + GetSwizzle(meta.element), - Type::Float}; - } - - Expression TextureQueryDimensions(Operation operation) { - const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); - ASSERT(meta); - - const std::string sampler = GetSampler(meta->sampler); - const std::string lod = VisitOperand(operation, 0).AsInt(); - - switch (meta->element) { - case 0: - case 1: - return {fmt::format("textureSize({}, {}){}", sampler, lod, GetSwizzle(meta->element)), - Type::Int}; - case 3: - return {fmt::format("textureQueryLevels({})", sampler), Type::Int}; - } - UNREACHABLE(); - return {"0", Type::Int}; - } - - Expression TextureQueryLod(Operation operation) { - const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); - ASSERT(meta); - - if (meta->element < 2) { - return {fmt::format("int(({} * vec2(256)){})", - GenerateTexture(operation, "QueryLod", {}), - GetSwizzle(meta->element)), - Type::Int}; - } - return {"0", Type::Int}; - } - - Expression TexelFetch(Operation operation) { - constexpr std::array constructors = {"int", "ivec2", "ivec3", "ivec4"}; - const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); - ASSERT(meta); - UNIMPLEMENTED_IF(meta->sampler.is_array); - const std::size_t count = operation.GetOperandsCount(); - - std::string expr = "texelFetch("; - expr += GetSampler(meta->sampler); - expr += ", "; - - expr += constructors.at(operation.GetOperandsCount() + (meta->array ? 1 : 0) - 1); - expr += '('; - for (std::size_t i = 0; i < count; ++i) { - if (i > 0) { - expr += ", "; - } - expr += VisitOperand(operation, i).AsInt(); - } - if (meta->array) { - expr += ", "; - expr += Visit(meta->array).AsInt(); - } - expr += ')'; - - if (meta->lod && !meta->sampler.is_buffer) { - expr += ", "; - expr += Visit(meta->lod).AsInt(); - } - expr += ')'; - expr += GetSwizzle(meta->element); - - return {std::move(expr), Type::Float}; - } - - Expression TextureGradient(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - std::string expr = - GenerateTexture(operation, "Grad", {TextureDerivates{}, TextureOffset{}}); - return {std::move(expr) + GetSwizzle(meta.element), Type::Float}; - } - - Expression ImageLoad(Operation operation) { - if (!device.HasImageLoadFormatted()) { - LOG_ERROR(Render_OpenGL, - "Device lacks GL_EXT_shader_image_load_formatted, stubbing image load"); - return {"0", Type::Int}; - } - - const auto& meta{std::get<MetaImage>(operation.GetMeta())}; - return {fmt::format("imageLoad({}, {}){}", GetImage(meta.image), - BuildIntegerCoordinates(operation), GetSwizzle(meta.element)), - Type::Uint}; - } - - Expression ImageStore(Operation operation) { - const auto& meta{std::get<MetaImage>(operation.GetMeta())}; - code.AddLine("imageStore({}, {}, {});", GetImage(meta.image), - BuildIntegerCoordinates(operation), BuildImageValues(operation)); - return {}; - } - - template <const std::string_view& opname> - Expression AtomicImage(Operation operation) { - const auto& meta{std::get<MetaImage>(operation.GetMeta())}; - ASSERT(meta.values.size() == 1); - - return {fmt::format("imageAtomic{}({}, {}, {})", opname, GetImage(meta.image), - BuildIntegerCoordinates(operation), Visit(meta.values[0]).AsUint()), - Type::Uint}; - } - - template <const std::string_view& opname, Type type> - Expression Atomic(Operation operation) { - if ((opname == Func::Min || opname == Func::Max) && type == Type::Int) { - UNIMPLEMENTED_MSG("Unimplemented Min & Max for atomic operations"); - return {}; - } - return {fmt::format("atomic{}({}, {})", opname, Visit(operation[0]).GetCode(), - Visit(operation[1]).AsUint()), - Type::Uint}; - } - - template <const std::string_view& opname, Type type> - Expression Reduce(Operation operation) { - code.AddLine("{};", Atomic<opname, type>(operation).GetCode()); - return {}; - } - - Expression Branch(Operation operation) { - const auto target = std::get_if<ImmediateNode>(&*operation[0]); - UNIMPLEMENTED_IF(!target); - - code.AddLine("jmp_to = 0x{:X}U;", target->GetValue()); - code.AddLine("break;"); - return {}; - } - - Expression BranchIndirect(Operation operation) { - const std::string op_a = VisitOperand(operation, 0).AsUint(); - - code.AddLine("jmp_to = {};", op_a); - code.AddLine("break;"); - return {}; - } - - Expression PushFlowStack(Operation operation) { - const auto stack = std::get<MetaStackClass>(operation.GetMeta()); - const auto target = std::get_if<ImmediateNode>(&*operation[0]); - UNIMPLEMENTED_IF(!target); - - code.AddLine("{}[{}++] = 0x{:X}U;", FlowStackName(stack), FlowStackTopName(stack), - target->GetValue()); - return {}; - } - - Expression PopFlowStack(Operation operation) { - const auto stack = std::get<MetaStackClass>(operation.GetMeta()); - code.AddLine("jmp_to = {}[--{}];", FlowStackName(stack), FlowStackTopName(stack)); - code.AddLine("break;"); - return {}; - } - - void PreExit() { - if (stage != ShaderType::Fragment) { - return; - } - const auto& used_registers = ir.GetRegisters(); - const auto SafeGetRegister = [&](u32 reg) -> Expression { - // TODO(Rodrigo): Replace with contains once C++20 releases - if (used_registers.find(reg) != used_registers.end()) { - return {GetRegister(reg), Type::Float}; - } - return {"0.0f", Type::Float}; - }; - - UNIMPLEMENTED_IF_MSG(header.ps.omap.sample_mask != 0, "Sample mask write is unimplemented"); - - // Write the color outputs using the data in the shader registers, disabled - // rendertargets/components are skipped in the register assignment. - u32 current_reg = 0; - for (u32 render_target = 0; render_target < Maxwell::NumRenderTargets; ++render_target) { - // TODO(Subv): Figure out how dual-source blending is configured in the Switch. - for (u32 component = 0; component < 4; ++component) { - if (header.ps.IsColorComponentOutputEnabled(render_target, component)) { - code.AddLine("frag_color{}{} = {};", render_target, GetColorSwizzle(component), - SafeGetRegister(current_reg).AsFloat()); - ++current_reg; - } - } - } - if (header.ps.omap.depth) { - // The depth output is always 2 registers after the last color output, and current_reg - // already contains one past the last color register. - code.AddLine("gl_FragDepth = {};", SafeGetRegister(current_reg + 1).AsFloat()); - } - } - - Expression Exit(Operation operation) { - PreExit(); - code.AddLine("return;"); - return {}; - } - - Expression Discard(Operation operation) { - // Enclose "discard" in a conditional, so that GLSL compilation does not complain - // about unexecuted instructions that may follow this. - code.AddLine("if (true) {{"); - ++code.scope; - code.AddLine("discard;"); - --code.scope; - code.AddLine("}}"); - return {}; - } - - Expression EmitVertex(Operation operation) { - ASSERT_MSG(stage == ShaderType::Geometry, - "EmitVertex is expected to be used in a geometry shader."); - code.AddLine("EmitVertex();"); - return {}; - } - - Expression EndPrimitive(Operation operation) { - ASSERT_MSG(stage == ShaderType::Geometry, - "EndPrimitive is expected to be used in a geometry shader."); - code.AddLine("EndPrimitive();"); - return {}; - } - - Expression InvocationId(Operation operation) { - return {"gl_InvocationID", Type::Int}; - } - - Expression YNegate(Operation operation) { - // Y_NEGATE is mapped to this uniform value - return {"gl_FrontMaterial.ambient.a", Type::Float}; - } - - template <u32 element> - Expression LocalInvocationId(Operation) { - return {"gl_LocalInvocationID"s + GetSwizzle(element), Type::Uint}; - } - - template <u32 element> - Expression WorkGroupId(Operation) { - return {"gl_WorkGroupID"s + GetSwizzle(element), Type::Uint}; - } - - Expression BallotThread(Operation operation) { - const std::string value = VisitOperand(operation, 0).AsBool(); - if (!device.HasWarpIntrinsics()) { - LOG_ERROR(Render_OpenGL, "Nvidia vote intrinsics are required by this shader"); - // Stub on non-Nvidia devices by simulating all threads voting the same as the active - // one. - return {fmt::format("({} ? 0xFFFFFFFFU : 0U)", value), Type::Uint}; - } - return {fmt::format("ballotThreadNV({})", value), Type::Uint}; - } - - Expression Vote(Operation operation, const char* func) { - const std::string value = VisitOperand(operation, 0).AsBool(); - if (!device.HasWarpIntrinsics()) { - LOG_ERROR(Render_OpenGL, "Nvidia vote intrinsics are required by this shader"); - // Stub with a warp size of one. - return {value, Type::Bool}; - } - return {fmt::format("{}({})", func, value), Type::Bool}; - } - - Expression VoteAll(Operation operation) { - return Vote(operation, "allThreadsNV"); - } - - Expression VoteAny(Operation operation) { - return Vote(operation, "anyThreadNV"); - } - - Expression VoteEqual(Operation operation) { - if (!device.HasWarpIntrinsics()) { - LOG_ERROR(Render_OpenGL, "Nvidia vote intrinsics are required by this shader"); - // We must return true here since a stub for a theoretical warp size of 1. - // This will always return an equal result across all votes. - return {"true", Type::Bool}; - } - return Vote(operation, "allThreadsEqualNV"); - } - - Expression ThreadId(Operation operation) { - if (!device.HasShaderBallot()) { - LOG_ERROR(Render_OpenGL, "Shader ballot is unavailable but required by the shader"); - return {"0U", Type::Uint}; - } - return {"gl_SubGroupInvocationARB", Type::Uint}; - } - - template <const std::string_view& comparison> - Expression ThreadMask(Operation) { - if (device.HasWarpIntrinsics()) { - return {fmt::format("gl_Thread{}MaskNV", comparison), Type::Uint}; - } - if (device.HasShaderBallot()) { - return {fmt::format("uint(gl_SubGroup{}MaskARB)", comparison), Type::Uint}; - } - LOG_ERROR(Render_OpenGL, "Thread mask intrinsics are required by the shader"); - return {"0U", Type::Uint}; - } - - Expression ShuffleIndexed(Operation operation) { - std::string value = VisitOperand(operation, 0).AsFloat(); - - if (!device.HasShaderBallot()) { - LOG_ERROR(Render_OpenGL, "Shader ballot is unavailable but required by the shader"); - return {std::move(value), Type::Float}; - } - - const std::string index = VisitOperand(operation, 1).AsUint(); - return {fmt::format("readInvocationARB({}, {})", value, index), Type::Float}; - } - - Expression Barrier(Operation) { - if (!ir.IsDecompiled()) { - LOG_ERROR(Render_OpenGL, "barrier() used but shader is not decompiled"); - return {}; - } - code.AddLine("barrier();"); - return {}; - } - - Expression MemoryBarrierGroup(Operation) { - code.AddLine("groupMemoryBarrier();"); - return {}; - } - - Expression MemoryBarrierGlobal(Operation) { - code.AddLine("memoryBarrier();"); - return {}; - } - - struct Func final { - Func() = delete; - ~Func() = delete; - - static constexpr std::string_view LessThan = "<"; - static constexpr std::string_view Equal = "=="; - static constexpr std::string_view LessEqual = "<="; - static constexpr std::string_view GreaterThan = ">"; - static constexpr std::string_view NotEqual = "!="; - static constexpr std::string_view GreaterEqual = ">="; - - static constexpr std::string_view Eq = "Eq"; - static constexpr std::string_view Ge = "Ge"; - static constexpr std::string_view Gt = "Gt"; - static constexpr std::string_view Le = "Le"; - static constexpr std::string_view Lt = "Lt"; - - static constexpr std::string_view Add = "Add"; - static constexpr std::string_view Min = "Min"; - static constexpr std::string_view Max = "Max"; - static constexpr std::string_view And = "And"; - static constexpr std::string_view Or = "Or"; - static constexpr std::string_view Xor = "Xor"; - static constexpr std::string_view Exchange = "Exchange"; - }; - - static constexpr std::array operation_decompilers = { - &GLSLDecompiler::Assign, - - &GLSLDecompiler::Select, - - &GLSLDecompiler::Add<Type::Float>, - &GLSLDecompiler::Mul<Type::Float>, - &GLSLDecompiler::Div<Type::Float>, - &GLSLDecompiler::Fma<Type::Float>, - &GLSLDecompiler::Negate<Type::Float>, - &GLSLDecompiler::Absolute<Type::Float>, - &GLSLDecompiler::FClamp, - &GLSLDecompiler::FCastHalf0, - &GLSLDecompiler::FCastHalf1, - &GLSLDecompiler::Min<Type::Float>, - &GLSLDecompiler::Max<Type::Float>, - &GLSLDecompiler::FCos, - &GLSLDecompiler::FSin, - &GLSLDecompiler::FExp2, - &GLSLDecompiler::FLog2, - &GLSLDecompiler::FInverseSqrt, - &GLSLDecompiler::FSqrt, - &GLSLDecompiler::FRoundEven, - &GLSLDecompiler::FFloor, - &GLSLDecompiler::FCeil, - &GLSLDecompiler::FTrunc, - &GLSLDecompiler::FCastInteger<Type::Int>, - &GLSLDecompiler::FCastInteger<Type::Uint>, - &GLSLDecompiler::FSwizzleAdd, - - &GLSLDecompiler::Add<Type::Int>, - &GLSLDecompiler::Mul<Type::Int>, - &GLSLDecompiler::Div<Type::Int>, - &GLSLDecompiler::Negate<Type::Int>, - &GLSLDecompiler::Absolute<Type::Int>, - &GLSLDecompiler::Min<Type::Int>, - &GLSLDecompiler::Max<Type::Int>, - - &GLSLDecompiler::ICastFloat, - &GLSLDecompiler::ICastUnsigned, - &GLSLDecompiler::LogicalShiftLeft<Type::Int>, - &GLSLDecompiler::ILogicalShiftRight, - &GLSLDecompiler::IArithmeticShiftRight, - &GLSLDecompiler::BitwiseAnd<Type::Int>, - &GLSLDecompiler::BitwiseOr<Type::Int>, - &GLSLDecompiler::BitwiseXor<Type::Int>, - &GLSLDecompiler::BitwiseNot<Type::Int>, - &GLSLDecompiler::BitfieldInsert<Type::Int>, - &GLSLDecompiler::BitfieldExtract<Type::Int>, - &GLSLDecompiler::BitCount<Type::Int>, - &GLSLDecompiler::BitMSB<Type::Int>, - - &GLSLDecompiler::Add<Type::Uint>, - &GLSLDecompiler::Mul<Type::Uint>, - &GLSLDecompiler::Div<Type::Uint>, - &GLSLDecompiler::Min<Type::Uint>, - &GLSLDecompiler::Max<Type::Uint>, - &GLSLDecompiler::UCastFloat, - &GLSLDecompiler::UCastSigned, - &GLSLDecompiler::LogicalShiftLeft<Type::Uint>, - &GLSLDecompiler::UShiftRight, - &GLSLDecompiler::UShiftRight, - &GLSLDecompiler::BitwiseAnd<Type::Uint>, - &GLSLDecompiler::BitwiseOr<Type::Uint>, - &GLSLDecompiler::BitwiseXor<Type::Uint>, - &GLSLDecompiler::BitwiseNot<Type::Uint>, - &GLSLDecompiler::BitfieldInsert<Type::Uint>, - &GLSLDecompiler::BitfieldExtract<Type::Uint>, - &GLSLDecompiler::BitCount<Type::Uint>, - &GLSLDecompiler::BitMSB<Type::Uint>, - - &GLSLDecompiler::Add<Type::HalfFloat>, - &GLSLDecompiler::Mul<Type::HalfFloat>, - &GLSLDecompiler::Fma<Type::HalfFloat>, - &GLSLDecompiler::Absolute<Type::HalfFloat>, - &GLSLDecompiler::HNegate, - &GLSLDecompiler::HClamp, - &GLSLDecompiler::HCastFloat, - &GLSLDecompiler::HUnpack, - &GLSLDecompiler::HMergeF32, - &GLSLDecompiler::HMergeH0, - &GLSLDecompiler::HMergeH1, - &GLSLDecompiler::HPack2, - - &GLSLDecompiler::LogicalAssign, - &GLSLDecompiler::LogicalAnd, - &GLSLDecompiler::LogicalOr, - &GLSLDecompiler::LogicalXor, - &GLSLDecompiler::LogicalNegate, - &GLSLDecompiler::LogicalPick2, - &GLSLDecompiler::LogicalAnd2, - - &GLSLDecompiler::Comparison<Func::LessThan, Type::Float, false>, - &GLSLDecompiler::Comparison<Func::Equal, Type::Float, false>, - &GLSLDecompiler::Comparison<Func::LessEqual, Type::Float, false>, - &GLSLDecompiler::Comparison<Func::GreaterThan, Type::Float, false>, - &GLSLDecompiler::Comparison<Func::NotEqual, Type::Float, false>, - &GLSLDecompiler::Comparison<Func::GreaterEqual, Type::Float, false>, - &GLSLDecompiler::FOrdered, - &GLSLDecompiler::FUnordered, - &GLSLDecompiler::Comparison<Func::LessThan, Type::Float, true>, - &GLSLDecompiler::Comparison<Func::Equal, Type::Float, true>, - &GLSLDecompiler::Comparison<Func::LessEqual, Type::Float, true>, - &GLSLDecompiler::Comparison<Func::GreaterThan, Type::Float, true>, - &GLSLDecompiler::Comparison<Func::NotEqual, Type::Float, true>, - &GLSLDecompiler::Comparison<Func::GreaterEqual, Type::Float, true>, - - &GLSLDecompiler::Comparison<Func::LessThan, Type::Int>, - &GLSLDecompiler::Comparison<Func::Equal, Type::Int>, - &GLSLDecompiler::Comparison<Func::LessEqual, Type::Int>, - &GLSLDecompiler::Comparison<Func::GreaterThan, Type::Int>, - &GLSLDecompiler::Comparison<Func::NotEqual, Type::Int>, - &GLSLDecompiler::Comparison<Func::GreaterEqual, Type::Int>, - - &GLSLDecompiler::Comparison<Func::LessThan, Type::Uint>, - &GLSLDecompiler::Comparison<Func::Equal, Type::Uint>, - &GLSLDecompiler::Comparison<Func::LessEqual, Type::Uint>, - &GLSLDecompiler::Comparison<Func::GreaterThan, Type::Uint>, - &GLSLDecompiler::Comparison<Func::NotEqual, Type::Uint>, - &GLSLDecompiler::Comparison<Func::GreaterEqual, Type::Uint>, - - &GLSLDecompiler::LogicalAddCarry, - - &GLSLDecompiler::Logical2HLessThan<false>, - &GLSLDecompiler::Logical2HEqual<false>, - &GLSLDecompiler::Logical2HLessEqual<false>, - &GLSLDecompiler::Logical2HGreaterThan<false>, - &GLSLDecompiler::Logical2HNotEqual<false>, - &GLSLDecompiler::Logical2HGreaterEqual<false>, - &GLSLDecompiler::Logical2HLessThan<true>, - &GLSLDecompiler::Logical2HEqual<true>, - &GLSLDecompiler::Logical2HLessEqual<true>, - &GLSLDecompiler::Logical2HGreaterThan<true>, - &GLSLDecompiler::Logical2HNotEqual<true>, - &GLSLDecompiler::Logical2HGreaterEqual<true>, - - &GLSLDecompiler::Texture, - &GLSLDecompiler::TextureLod, - &GLSLDecompiler::TextureGather, - &GLSLDecompiler::TextureQueryDimensions, - &GLSLDecompiler::TextureQueryLod, - &GLSLDecompiler::TexelFetch, - &GLSLDecompiler::TextureGradient, - - &GLSLDecompiler::ImageLoad, - &GLSLDecompiler::ImageStore, - - &GLSLDecompiler::AtomicImage<Func::Add>, - &GLSLDecompiler::AtomicImage<Func::And>, - &GLSLDecompiler::AtomicImage<Func::Or>, - &GLSLDecompiler::AtomicImage<Func::Xor>, - &GLSLDecompiler::AtomicImage<Func::Exchange>, - - &GLSLDecompiler::Atomic<Func::Exchange, Type::Uint>, - &GLSLDecompiler::Atomic<Func::Add, Type::Uint>, - &GLSLDecompiler::Atomic<Func::Min, Type::Uint>, - &GLSLDecompiler::Atomic<Func::Max, Type::Uint>, - &GLSLDecompiler::Atomic<Func::And, Type::Uint>, - &GLSLDecompiler::Atomic<Func::Or, Type::Uint>, - &GLSLDecompiler::Atomic<Func::Xor, Type::Uint>, - - &GLSLDecompiler::Atomic<Func::Exchange, Type::Int>, - &GLSLDecompiler::Atomic<Func::Add, Type::Int>, - &GLSLDecompiler::Atomic<Func::Min, Type::Int>, - &GLSLDecompiler::Atomic<Func::Max, Type::Int>, - &GLSLDecompiler::Atomic<Func::And, Type::Int>, - &GLSLDecompiler::Atomic<Func::Or, Type::Int>, - &GLSLDecompiler::Atomic<Func::Xor, Type::Int>, - - &GLSLDecompiler::Reduce<Func::Add, Type::Uint>, - &GLSLDecompiler::Reduce<Func::Min, Type::Uint>, - &GLSLDecompiler::Reduce<Func::Max, Type::Uint>, - &GLSLDecompiler::Reduce<Func::And, Type::Uint>, - &GLSLDecompiler::Reduce<Func::Or, Type::Uint>, - &GLSLDecompiler::Reduce<Func::Xor, Type::Uint>, - - &GLSLDecompiler::Reduce<Func::Add, Type::Int>, - &GLSLDecompiler::Reduce<Func::Min, Type::Int>, - &GLSLDecompiler::Reduce<Func::Max, Type::Int>, - &GLSLDecompiler::Reduce<Func::And, Type::Int>, - &GLSLDecompiler::Reduce<Func::Or, Type::Int>, - &GLSLDecompiler::Reduce<Func::Xor, Type::Int>, - - &GLSLDecompiler::Branch, - &GLSLDecompiler::BranchIndirect, - &GLSLDecompiler::PushFlowStack, - &GLSLDecompiler::PopFlowStack, - &GLSLDecompiler::Exit, - &GLSLDecompiler::Discard, - - &GLSLDecompiler::EmitVertex, - &GLSLDecompiler::EndPrimitive, - - &GLSLDecompiler::InvocationId, - &GLSLDecompiler::YNegate, - &GLSLDecompiler::LocalInvocationId<0>, - &GLSLDecompiler::LocalInvocationId<1>, - &GLSLDecompiler::LocalInvocationId<2>, - &GLSLDecompiler::WorkGroupId<0>, - &GLSLDecompiler::WorkGroupId<1>, - &GLSLDecompiler::WorkGroupId<2>, - - &GLSLDecompiler::BallotThread, - &GLSLDecompiler::VoteAll, - &GLSLDecompiler::VoteAny, - &GLSLDecompiler::VoteEqual, - - &GLSLDecompiler::ThreadId, - &GLSLDecompiler::ThreadMask<Func::Eq>, - &GLSLDecompiler::ThreadMask<Func::Ge>, - &GLSLDecompiler::ThreadMask<Func::Gt>, - &GLSLDecompiler::ThreadMask<Func::Le>, - &GLSLDecompiler::ThreadMask<Func::Lt>, - &GLSLDecompiler::ShuffleIndexed, - - &GLSLDecompiler::Barrier, - &GLSLDecompiler::MemoryBarrierGroup, - &GLSLDecompiler::MemoryBarrierGlobal, - }; - static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount)); - - std::string GetRegister(u32 index) const { - return AppendSuffix(index, "gpr"); - } - - std::string GetCustomVariable(u32 index) const { - return AppendSuffix(index, "custom_var"); - } - - std::string GetPredicate(Tegra::Shader::Pred pred) const { - return AppendSuffix(static_cast<u32>(pred), "pred"); - } - - std::string GetGenericInputAttribute(Attribute::Index attribute) const { - return AppendSuffix(GetGenericAttributeIndex(attribute), INPUT_ATTRIBUTE_NAME); - } - - std::unordered_map<u8, GenericVaryingDescription> varying_description; - - std::string GetGenericOutputAttribute(Attribute::Index attribute, std::size_t element) const { - const u8 offset = static_cast<u8>(GetGenericAttributeIndex(attribute) * 4 + element); - const auto& description = varying_description.at(offset); - if (description.is_scalar) { - return description.name; - } - return fmt::format("{}[{}]", description.name, element - description.first_element); - } - - std::string GetConstBuffer(u32 index) const { - return AppendSuffix(index, "cbuf"); - } - - std::string GetGlobalMemory(const GlobalMemoryBase& descriptor) const { - return fmt::format("gmem_{}_{}_{}", descriptor.cbuf_index, descriptor.cbuf_offset, suffix); - } - - std::string GetGlobalMemoryBlock(const GlobalMemoryBase& descriptor) const { - return fmt::format("gmem_block_{}_{}_{}", descriptor.cbuf_index, descriptor.cbuf_offset, - suffix); - } - - std::string GetConstBufferBlock(u32 index) const { - return AppendSuffix(index, "cbuf_block"); - } - - std::string GetLocalMemory() const { - if (suffix.empty()) { - return "lmem"; - } else { - return "lmem_" + std::string{suffix}; - } - } - - std::string GetInternalFlag(InternalFlag flag) const { - constexpr std::array InternalFlagNames = {"zero_flag", "sign_flag", "carry_flag", - "overflow_flag"}; - const auto index = static_cast<u32>(flag); - ASSERT(index < static_cast<u32>(InternalFlag::Amount)); - - if (suffix.empty()) { - return InternalFlagNames[index]; - } else { - return fmt::format("{}_{}", InternalFlagNames[index], suffix); - } - } - - std::string GetSampler(const SamplerEntry& sampler) const { - return AppendSuffix(sampler.index, "sampler"); - } - - std::string GetImage(const ImageEntry& image) const { - return AppendSuffix(image.index, "image"); - } - - std::string AppendSuffix(u32 index, std::string_view name) const { - if (suffix.empty()) { - return fmt::format("{}{}", name, index); - } else { - return fmt::format("{}{}_{}", name, index, suffix); - } - } - - u32 GetNumPhysicalInputAttributes() const { - return stage == ShaderType::Vertex ? GetNumPhysicalAttributes() : GetNumPhysicalVaryings(); - } - - u32 GetNumPhysicalAttributes() const { - return std::min<u32>(device.GetMaxVertexAttributes(), Maxwell::NumVertexAttributes); - } - - u32 GetNumPhysicalVaryings() const { - return std::min<u32>(device.GetMaxVaryings(), Maxwell::NumVaryings); - } - - const Device& device; - const ShaderIR& ir; - const Registry& registry; - const ShaderType stage; - const std::string_view identifier; - const std::string_view suffix; - const Header header; - std::unordered_map<u8, VaryingTFB> transform_feedback; - - ShaderWriter code; - - std::optional<u32> max_input_vertices; -}; - -std::string GetFlowVariable(u32 index) { - return fmt::format("flow_var{}", index); -} - -class ExprDecompiler { -public: - explicit ExprDecompiler(GLSLDecompiler& decomp_) : decomp{decomp_} {} - - void operator()(const ExprAnd& expr) { - inner += '('; - std::visit(*this, *expr.operand1); - inner += " && "; - std::visit(*this, *expr.operand2); - inner += ')'; - } - - void operator()(const ExprOr& expr) { - inner += '('; - std::visit(*this, *expr.operand1); - inner += " || "; - std::visit(*this, *expr.operand2); - inner += ')'; - } - - void operator()(const ExprNot& expr) { - inner += '!'; - std::visit(*this, *expr.operand1); - } - - void operator()(const ExprPredicate& expr) { - const auto pred = static_cast<Tegra::Shader::Pred>(expr.predicate); - inner += decomp.GetPredicate(pred); - } - - void operator()(const ExprCondCode& expr) { - inner += decomp.Visit(decomp.ir.GetConditionCode(expr.cc)).AsBool(); - } - - void operator()(const ExprVar& expr) { - inner += GetFlowVariable(expr.var_index); - } - - void operator()(const ExprBoolean& expr) { - inner += expr.value ? "true" : "false"; - } - - void operator()(VideoCommon::Shader::ExprGprEqual& expr) { - inner += fmt::format("(ftou({}) == {})", decomp.GetRegister(expr.gpr), expr.value); - } - - const std::string& GetResult() const { - return inner; - } - -private: - GLSLDecompiler& decomp; - std::string inner; -}; - -class ASTDecompiler { -public: - explicit ASTDecompiler(GLSLDecompiler& decomp_) : decomp{decomp_} {} - - void operator()(const ASTProgram& ast) { - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - } - - void operator()(const ASTIfThen& ast) { - ExprDecompiler expr_parser{decomp}; - std::visit(expr_parser, *ast.condition); - decomp.code.AddLine("if ({}) {{", expr_parser.GetResult()); - decomp.code.scope++; - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - decomp.code.scope--; - decomp.code.AddLine("}}"); - } - - void operator()(const ASTIfElse& ast) { - decomp.code.AddLine("else {{"); - decomp.code.scope++; - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - decomp.code.scope--; - decomp.code.AddLine("}}"); - } - - void operator()([[maybe_unused]] const ASTBlockEncoded& ast) { - UNREACHABLE(); - } - - void operator()(const ASTBlockDecoded& ast) { - decomp.VisitBlock(ast.nodes); - } - - void operator()(const ASTVarSet& ast) { - ExprDecompiler expr_parser{decomp}; - std::visit(expr_parser, *ast.condition); - decomp.code.AddLine("{} = {};", GetFlowVariable(ast.index), expr_parser.GetResult()); - } - - void operator()(const ASTLabel& ast) { - decomp.code.AddLine("// Label_{}:", ast.index); - } - - void operator()([[maybe_unused]] const ASTGoto& ast) { - UNREACHABLE(); - } - - void operator()(const ASTDoWhile& ast) { - ExprDecompiler expr_parser{decomp}; - std::visit(expr_parser, *ast.condition); - decomp.code.AddLine("do {{"); - decomp.code.scope++; - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - decomp.code.scope--; - decomp.code.AddLine("}} while({});", expr_parser.GetResult()); - } - - void operator()(const ASTReturn& ast) { - const bool is_true = VideoCommon::Shader::ExprIsTrue(ast.condition); - if (!is_true) { - ExprDecompiler expr_parser{decomp}; - std::visit(expr_parser, *ast.condition); - decomp.code.AddLine("if ({}) {{", expr_parser.GetResult()); - decomp.code.scope++; - } - if (ast.kills) { - decomp.code.AddLine("discard;"); - } else { - decomp.PreExit(); - decomp.code.AddLine("return;"); - } - if (!is_true) { - decomp.code.scope--; - decomp.code.AddLine("}}"); - } - } - - void operator()(const ASTBreak& ast) { - const bool is_true = VideoCommon::Shader::ExprIsTrue(ast.condition); - if (!is_true) { - ExprDecompiler expr_parser{decomp}; - std::visit(expr_parser, *ast.condition); - decomp.code.AddLine("if ({}) {{", expr_parser.GetResult()); - decomp.code.scope++; - } - decomp.code.AddLine("break;"); - if (!is_true) { - decomp.code.scope--; - decomp.code.AddLine("}}"); - } - } - - void Visit(const ASTNode& node) { - std::visit(*this, *node->GetInnerData()); - } - -private: - GLSLDecompiler& decomp; -}; - -void GLSLDecompiler::DecompileAST() { - const u32 num_flow_variables = ir.GetASTNumVariables(); - for (u32 i = 0; i < num_flow_variables; i++) { - code.AddLine("bool {} = false;", GetFlowVariable(i)); - } - - ASTDecompiler decompiler{*this}; - decompiler.Visit(ir.GetASTProgram()); -} - -} // Anonymous namespace - -ShaderEntries MakeEntries(const Device& device, const ShaderIR& ir, ShaderType stage) { - ShaderEntries entries; - for (const auto& cbuf : ir.GetConstantBuffers()) { - entries.const_buffers.emplace_back(cbuf.second.GetMaxOffset(), cbuf.second.IsIndirect(), - cbuf.first); - } - for (const auto& [base, usage] : ir.GetGlobalMemory()) { - entries.global_memory_entries.emplace_back(base.cbuf_index, base.cbuf_offset, usage.is_read, - usage.is_written); - } - for (const auto& sampler : ir.GetSamplers()) { - entries.samplers.emplace_back(sampler); - } - for (const auto& image : ir.GetImages()) { - entries.images.emplace_back(image); - } - const auto clip_distances = ir.GetClipDistances(); - for (std::size_t i = 0; i < std::size(clip_distances); ++i) { - entries.clip_distances = (clip_distances[i] ? 1U : 0U) << i; - } - for (const auto& buffer : entries.const_buffers) { - entries.enabled_uniform_buffers |= 1U << buffer.GetIndex(); - } - entries.shader_length = ir.GetLength(); - return entries; -} - -std::string DecompileShader(const Device& device, const ShaderIR& ir, const Registry& registry, - ShaderType stage, std::string_view identifier, - std::string_view suffix) { - GLSLDecompiler decompiler(device, ir, registry, stage, identifier, suffix); - decompiler.Decompile(); - return decompiler.GetResult(); -} - -} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.h b/src/video_core/renderer_opengl/gl_shader_decompiler.h deleted file mode 100644 index 0397a000c..000000000 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.h +++ /dev/null @@ -1,69 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <array> -#include <string> -#include <string_view> -#include <utility> -#include <vector> -#include "common/common_types.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/shader_type.h" -#include "video_core/shader/registry.h" -#include "video_core/shader/shader_ir.h" - -namespace OpenGL { - -class Device; - -using Maxwell = Tegra::Engines::Maxwell3D::Regs; -using SamplerEntry = VideoCommon::Shader::SamplerEntry; -using ImageEntry = VideoCommon::Shader::ImageEntry; - -class ConstBufferEntry : public VideoCommon::Shader::ConstBuffer { -public: - explicit ConstBufferEntry(u32 max_offset_, bool is_indirect_, u32 index_) - : ConstBuffer{max_offset_, is_indirect_}, index{index_} {} - - u32 GetIndex() const { - return index; - } - -private: - u32 index = 0; -}; - -struct GlobalMemoryEntry { - constexpr explicit GlobalMemoryEntry(u32 cbuf_index_, u32 cbuf_offset_, bool is_read_, - bool is_written_) - : cbuf_index{cbuf_index_}, cbuf_offset{cbuf_offset_}, is_read{is_read_}, is_written{ - is_written_} {} - - u32 cbuf_index = 0; - u32 cbuf_offset = 0; - bool is_read = false; - bool is_written = false; -}; - -struct ShaderEntries { - std::vector<ConstBufferEntry> const_buffers; - std::vector<GlobalMemoryEntry> global_memory_entries; - std::vector<SamplerEntry> samplers; - std::vector<ImageEntry> images; - std::size_t shader_length{}; - u32 clip_distances{}; - u32 enabled_uniform_buffers{}; -}; - -ShaderEntries MakeEntries(const Device& device, const VideoCommon::Shader::ShaderIR& ir, - Tegra::Engines::ShaderType stage); - -std::string DecompileShader(const Device& device, const VideoCommon::Shader::ShaderIR& ir, - const VideoCommon::Shader::Registry& registry, - Tegra::Engines::ShaderType stage, std::string_view identifier, - std::string_view suffix = {}); - -} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp deleted file mode 100644 index 0deb86517..000000000 --- a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp +++ /dev/null @@ -1,482 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <cstring> - -#include <fmt/format.h> - -#include "common/assert.h" -#include "common/common_types.h" -#include "common/fs/file.h" -#include "common/fs/fs.h" -#include "common/fs/path_util.h" -#include "common/logging/log.h" -#include "common/scm_rev.h" -#include "common/settings.h" -#include "common/zstd_compression.h" -#include "core/core.h" -#include "core/hle/kernel/k_process.h" -#include "video_core/engines/shader_type.h" -#include "video_core/renderer_opengl/gl_shader_cache.h" -#include "video_core/renderer_opengl/gl_shader_disk_cache.h" - -namespace OpenGL { - -using Tegra::Engines::ShaderType; -using VideoCommon::Shader::BindlessSamplerMap; -using VideoCommon::Shader::BoundSamplerMap; -using VideoCommon::Shader::KeyMap; -using VideoCommon::Shader::SeparateSamplerKey; -using ShaderCacheVersionHash = std::array<u8, 64>; - -struct ConstBufferKey { - u32 cbuf = 0; - u32 offset = 0; - u32 value = 0; -}; - -struct BoundSamplerEntry { - u32 offset = 0; - Tegra::Engines::SamplerDescriptor sampler; -}; - -struct SeparateSamplerEntry { - u32 cbuf1 = 0; - u32 cbuf2 = 0; - u32 offset1 = 0; - u32 offset2 = 0; - Tegra::Engines::SamplerDescriptor sampler; -}; - -struct BindlessSamplerEntry { - u32 cbuf = 0; - u32 offset = 0; - Tegra::Engines::SamplerDescriptor sampler; -}; - -namespace { - -constexpr u32 NativeVersion = 21; - -ShaderCacheVersionHash GetShaderCacheVersionHash() { - ShaderCacheVersionHash hash{}; - const std::size_t length = std::min(std::strlen(Common::g_shader_cache_version), hash.size()); - std::memcpy(hash.data(), Common::g_shader_cache_version, length); - return hash; -} - -} // Anonymous namespace - -ShaderDiskCacheEntry::ShaderDiskCacheEntry() = default; - -ShaderDiskCacheEntry::~ShaderDiskCacheEntry() = default; - -bool ShaderDiskCacheEntry::Load(Common::FS::IOFile& file) { - if (!file.ReadObject(type)) { - return false; - } - u32 code_size; - u32 code_size_b; - if (!file.ReadObject(code_size) || !file.ReadObject(code_size_b)) { - return false; - } - code.resize(code_size); - code_b.resize(code_size_b); - if (file.Read(code) != code_size) { - return false; - } - if (HasProgramA() && file.Read(code_b) != code_size_b) { - return false; - } - - u8 is_texture_handler_size_known; - u32 texture_handler_size_value; - u32 num_keys; - u32 num_bound_samplers; - u32 num_separate_samplers; - u32 num_bindless_samplers; - if (!file.ReadObject(unique_identifier) || !file.ReadObject(bound_buffer) || - !file.ReadObject(is_texture_handler_size_known) || - !file.ReadObject(texture_handler_size_value) || !file.ReadObject(graphics_info) || - !file.ReadObject(compute_info) || !file.ReadObject(num_keys) || - !file.ReadObject(num_bound_samplers) || !file.ReadObject(num_separate_samplers) || - !file.ReadObject(num_bindless_samplers)) { - return false; - } - if (is_texture_handler_size_known) { - texture_handler_size = texture_handler_size_value; - } - - std::vector<ConstBufferKey> flat_keys(num_keys); - std::vector<BoundSamplerEntry> flat_bound_samplers(num_bound_samplers); - std::vector<SeparateSamplerEntry> flat_separate_samplers(num_separate_samplers); - std::vector<BindlessSamplerEntry> flat_bindless_samplers(num_bindless_samplers); - if (file.Read(flat_keys) != flat_keys.size() || - file.Read(flat_bound_samplers) != flat_bound_samplers.size() || - file.Read(flat_separate_samplers) != flat_separate_samplers.size() || - file.Read(flat_bindless_samplers) != flat_bindless_samplers.size()) { - return false; - } - for (const auto& entry : flat_keys) { - keys.insert({{entry.cbuf, entry.offset}, entry.value}); - } - for (const auto& entry : flat_bound_samplers) { - bound_samplers.emplace(entry.offset, entry.sampler); - } - for (const auto& entry : flat_separate_samplers) { - SeparateSamplerKey key; - key.buffers = {entry.cbuf1, entry.cbuf2}; - key.offsets = {entry.offset1, entry.offset2}; - separate_samplers.emplace(key, entry.sampler); - } - for (const auto& entry : flat_bindless_samplers) { - bindless_samplers.insert({{entry.cbuf, entry.offset}, entry.sampler}); - } - - return true; -} - -bool ShaderDiskCacheEntry::Save(Common::FS::IOFile& file) const { - if (!file.WriteObject(static_cast<u32>(type)) || - !file.WriteObject(static_cast<u32>(code.size())) || - !file.WriteObject(static_cast<u32>(code_b.size()))) { - return false; - } - if (file.Write(code) != code.size()) { - return false; - } - if (HasProgramA() && file.Write(code_b) != code_b.size()) { - return false; - } - - if (!file.WriteObject(unique_identifier) || !file.WriteObject(bound_buffer) || - !file.WriteObject(static_cast<u8>(texture_handler_size.has_value())) || - !file.WriteObject(texture_handler_size.value_or(0)) || !file.WriteObject(graphics_info) || - !file.WriteObject(compute_info) || !file.WriteObject(static_cast<u32>(keys.size())) || - !file.WriteObject(static_cast<u32>(bound_samplers.size())) || - !file.WriteObject(static_cast<u32>(separate_samplers.size())) || - !file.WriteObject(static_cast<u32>(bindless_samplers.size()))) { - return false; - } - - std::vector<ConstBufferKey> flat_keys; - flat_keys.reserve(keys.size()); - for (const auto& [address, value] : keys) { - flat_keys.push_back(ConstBufferKey{address.first, address.second, value}); - } - - std::vector<BoundSamplerEntry> flat_bound_samplers; - flat_bound_samplers.reserve(bound_samplers.size()); - for (const auto& [address, sampler] : bound_samplers) { - flat_bound_samplers.push_back(BoundSamplerEntry{address, sampler}); - } - - std::vector<SeparateSamplerEntry> flat_separate_samplers; - flat_separate_samplers.reserve(separate_samplers.size()); - for (const auto& [key, sampler] : separate_samplers) { - SeparateSamplerEntry entry; - std::tie(entry.cbuf1, entry.cbuf2) = key.buffers; - std::tie(entry.offset1, entry.offset2) = key.offsets; - entry.sampler = sampler; - flat_separate_samplers.push_back(entry); - } - - std::vector<BindlessSamplerEntry> flat_bindless_samplers; - flat_bindless_samplers.reserve(bindless_samplers.size()); - for (const auto& [address, sampler] : bindless_samplers) { - flat_bindless_samplers.push_back( - BindlessSamplerEntry{address.first, address.second, sampler}); - } - - return file.Write(flat_keys) == flat_keys.size() && - file.Write(flat_bound_samplers) == flat_bound_samplers.size() && - file.Write(flat_separate_samplers) == flat_separate_samplers.size() && - file.Write(flat_bindless_samplers) == flat_bindless_samplers.size(); -} - -ShaderDiskCacheOpenGL::ShaderDiskCacheOpenGL() = default; - -ShaderDiskCacheOpenGL::~ShaderDiskCacheOpenGL() = default; - -void ShaderDiskCacheOpenGL::BindTitleID(u64 title_id_) { - title_id = title_id_; -} - -std::optional<std::vector<ShaderDiskCacheEntry>> ShaderDiskCacheOpenGL::LoadTransferable() { - // Skip games without title id - const bool has_title_id = title_id != 0; - if (!Settings::values.use_disk_shader_cache.GetValue() || !has_title_id) { - return std::nullopt; - } - - Common::FS::IOFile file{GetTransferablePath(), Common::FS::FileAccessMode::Read, - Common::FS::FileType::BinaryFile}; - if (!file.IsOpen()) { - LOG_INFO(Render_OpenGL, "No transferable shader cache found"); - is_usable = true; - return std::nullopt; - } - - u32 version{}; - if (!file.ReadObject(version)) { - LOG_ERROR(Render_OpenGL, "Failed to get transferable cache version, skipping it"); - return std::nullopt; - } - - if (version < NativeVersion) { - LOG_INFO(Render_OpenGL, "Transferable shader cache is old, removing"); - file.Close(); - InvalidateTransferable(); - is_usable = true; - return std::nullopt; - } - if (version > NativeVersion) { - LOG_WARNING(Render_OpenGL, "Transferable shader cache was generated with a newer version " - "of the emulator, skipping"); - return std::nullopt; - } - - // Version is valid, load the shaders - std::vector<ShaderDiskCacheEntry> entries; - while (static_cast<u64>(file.Tell()) < file.GetSize()) { - ShaderDiskCacheEntry& entry = entries.emplace_back(); - if (!entry.Load(file)) { - LOG_ERROR(Render_OpenGL, "Failed to load transferable raw entry, skipping"); - return std::nullopt; - } - } - - is_usable = true; - return {std::move(entries)}; -} - -std::vector<ShaderDiskCachePrecompiled> ShaderDiskCacheOpenGL::LoadPrecompiled() { - if (!is_usable) { - return {}; - } - - Common::FS::IOFile file{GetPrecompiledPath(), Common::FS::FileAccessMode::Read, - Common::FS::FileType::BinaryFile}; - if (!file.IsOpen()) { - LOG_INFO(Render_OpenGL, "No precompiled shader cache found"); - return {}; - } - - if (const auto result = LoadPrecompiledFile(file)) { - return *result; - } - - LOG_INFO(Render_OpenGL, "Failed to load precompiled cache"); - file.Close(); - InvalidatePrecompiled(); - return {}; -} - -std::optional<std::vector<ShaderDiskCachePrecompiled>> ShaderDiskCacheOpenGL::LoadPrecompiledFile( - Common::FS::IOFile& file) { - // Read compressed file from disk and decompress to virtual precompiled cache file - std::vector<u8> compressed(file.GetSize()); - if (file.Read(compressed) != file.GetSize()) { - return std::nullopt; - } - const std::vector<u8> decompressed = Common::Compression::DecompressDataZSTD(compressed); - SaveArrayToPrecompiled(decompressed.data(), decompressed.size()); - precompiled_cache_virtual_file_offset = 0; - - ShaderCacheVersionHash file_hash{}; - if (!LoadArrayFromPrecompiled(file_hash.data(), file_hash.size())) { - precompiled_cache_virtual_file_offset = 0; - return std::nullopt; - } - if (GetShaderCacheVersionHash() != file_hash) { - LOG_INFO(Render_OpenGL, "Precompiled cache is from another version of the emulator"); - precompiled_cache_virtual_file_offset = 0; - return std::nullopt; - } - - std::vector<ShaderDiskCachePrecompiled> entries; - while (precompiled_cache_virtual_file_offset < precompiled_cache_virtual_file.GetSize()) { - u32 binary_size; - auto& entry = entries.emplace_back(); - if (!LoadObjectFromPrecompiled(entry.unique_identifier) || - !LoadObjectFromPrecompiled(entry.binary_format) || - !LoadObjectFromPrecompiled(binary_size)) { - return std::nullopt; - } - - entry.binary.resize(binary_size); - if (!LoadArrayFromPrecompiled(entry.binary.data(), entry.binary.size())) { - return std::nullopt; - } - } - return entries; -} - -void ShaderDiskCacheOpenGL::InvalidateTransferable() { - if (!Common::FS::RemoveFile(GetTransferablePath())) { - LOG_ERROR(Render_OpenGL, "Failed to invalidate transferable file={}", - Common::FS::PathToUTF8String(GetTransferablePath())); - } - InvalidatePrecompiled(); -} - -void ShaderDiskCacheOpenGL::InvalidatePrecompiled() { - // Clear virtaul precompiled cache file - precompiled_cache_virtual_file.Resize(0); - - if (!Common::FS::RemoveFile(GetPrecompiledPath())) { - LOG_ERROR(Render_OpenGL, "Failed to invalidate precompiled file={}", - Common::FS::PathToUTF8String(GetPrecompiledPath())); - } -} - -void ShaderDiskCacheOpenGL::SaveEntry(const ShaderDiskCacheEntry& entry) { - if (!is_usable) { - return; - } - - const u64 id = entry.unique_identifier; - if (stored_transferable.contains(id)) { - // The shader already exists - return; - } - - Common::FS::IOFile file = AppendTransferableFile(); - if (!file.IsOpen()) { - return; - } - if (!entry.Save(file)) { - LOG_ERROR(Render_OpenGL, "Failed to save raw transferable cache entry, removing"); - file.Close(); - InvalidateTransferable(); - return; - } - - stored_transferable.insert(id); -} - -void ShaderDiskCacheOpenGL::SavePrecompiled(u64 unique_identifier, GLuint program) { - if (!is_usable) { - return; - } - - // TODO(Rodrigo): This is a design smell. I shouldn't be having to manually write the header - // when writing the dump. This should be done the moment I get access to write to the virtual - // file. - if (precompiled_cache_virtual_file.GetSize() == 0) { - SavePrecompiledHeaderToVirtualPrecompiledCache(); - } - - GLint binary_length; - glGetProgramiv(program, GL_PROGRAM_BINARY_LENGTH, &binary_length); - - GLenum binary_format; - std::vector<u8> binary(binary_length); - glGetProgramBinary(program, binary_length, nullptr, &binary_format, binary.data()); - - if (!SaveObjectToPrecompiled(unique_identifier) || !SaveObjectToPrecompiled(binary_format) || - !SaveObjectToPrecompiled(static_cast<u32>(binary.size())) || - !SaveArrayToPrecompiled(binary.data(), binary.size())) { - LOG_ERROR(Render_OpenGL, "Failed to save binary program file in shader={:016X}, removing", - unique_identifier); - InvalidatePrecompiled(); - } -} - -Common::FS::IOFile ShaderDiskCacheOpenGL::AppendTransferableFile() const { - if (!EnsureDirectories()) { - return {}; - } - - const auto transferable_path{GetTransferablePath()}; - const bool existed = Common::FS::Exists(transferable_path); - - Common::FS::IOFile file{transferable_path, Common::FS::FileAccessMode::Append, - Common::FS::FileType::BinaryFile}; - if (!file.IsOpen()) { - LOG_ERROR(Render_OpenGL, "Failed to open transferable cache in path={}", - Common::FS::PathToUTF8String(transferable_path)); - return {}; - } - if (!existed || file.GetSize() == 0) { - // If the file didn't exist, write its version - if (!file.WriteObject(NativeVersion)) { - LOG_ERROR(Render_OpenGL, "Failed to write transferable cache version in path={}", - Common::FS::PathToUTF8String(transferable_path)); - return {}; - } - } - return file; -} - -void ShaderDiskCacheOpenGL::SavePrecompiledHeaderToVirtualPrecompiledCache() { - const auto hash{GetShaderCacheVersionHash()}; - if (!SaveArrayToPrecompiled(hash.data(), hash.size())) { - LOG_ERROR( - Render_OpenGL, - "Failed to write precompiled cache version hash to virtual precompiled cache file"); - } -} - -void ShaderDiskCacheOpenGL::SaveVirtualPrecompiledFile() { - precompiled_cache_virtual_file_offset = 0; - const std::vector<u8> uncompressed = precompiled_cache_virtual_file.ReadAllBytes(); - const std::vector<u8> compressed = - Common::Compression::CompressDataZSTDDefault(uncompressed.data(), uncompressed.size()); - - const auto precompiled_path = GetPrecompiledPath(); - Common::FS::IOFile file{precompiled_path, Common::FS::FileAccessMode::Write, - Common::FS::FileType::BinaryFile}; - - if (!file.IsOpen()) { - LOG_ERROR(Render_OpenGL, "Failed to open precompiled cache in path={}", - Common::FS::PathToUTF8String(precompiled_path)); - return; - } - if (file.Write(compressed) != compressed.size()) { - LOG_ERROR(Render_OpenGL, "Failed to write precompiled cache version in path={}", - Common::FS::PathToUTF8String(precompiled_path)); - } -} - -bool ShaderDiskCacheOpenGL::EnsureDirectories() const { - const auto CreateDir = [](const std::filesystem::path& dir) { - if (!Common::FS::CreateDir(dir)) { - LOG_ERROR(Render_OpenGL, "Failed to create directory={}", - Common::FS::PathToUTF8String(dir)); - return false; - } - return true; - }; - - return CreateDir(Common::FS::GetYuzuPath(Common::FS::YuzuPath::ShaderDir)) && - CreateDir(GetBaseDir()) && CreateDir(GetTransferableDir()) && - CreateDir(GetPrecompiledDir()); -} - -std::filesystem::path ShaderDiskCacheOpenGL::GetTransferablePath() const { - return GetTransferableDir() / fmt::format("{}.bin", GetTitleID()); -} - -std::filesystem::path ShaderDiskCacheOpenGL::GetPrecompiledPath() const { - return GetPrecompiledDir() / fmt::format("{}.bin", GetTitleID()); -} - -std::filesystem::path ShaderDiskCacheOpenGL::GetTransferableDir() const { - return GetBaseDir() / "transferable"; -} - -std::filesystem::path ShaderDiskCacheOpenGL::GetPrecompiledDir() const { - return GetBaseDir() / "precompiled"; -} - -std::filesystem::path ShaderDiskCacheOpenGL::GetBaseDir() const { - return Common::FS::GetYuzuPath(Common::FS::YuzuPath::ShaderDir) / "opengl"; -} - -std::string ShaderDiskCacheOpenGL::GetTitleID() const { - return fmt::format("{:016X}", title_id); -} - -} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.h b/src/video_core/renderer_opengl/gl_shader_disk_cache.h deleted file mode 100644 index f8bc23868..000000000 --- a/src/video_core/renderer_opengl/gl_shader_disk_cache.h +++ /dev/null @@ -1,176 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <filesystem> -#include <optional> -#include <string> -#include <tuple> -#include <type_traits> -#include <unordered_map> -#include <unordered_set> -#include <utility> -#include <vector> - -#include <glad/glad.h> - -#include "common/assert.h" -#include "common/common_types.h" -#include "core/file_sys/vfs_vector.h" -#include "video_core/engines/shader_type.h" -#include "video_core/shader/registry.h" - -namespace Common::FS { -class IOFile; -} - -namespace OpenGL { - -using ProgramCode = std::vector<u64>; - -/// Describes a shader and how it's used by the guest GPU -struct ShaderDiskCacheEntry { - ShaderDiskCacheEntry(); - ~ShaderDiskCacheEntry(); - - bool Load(Common::FS::IOFile& file); - - bool Save(Common::FS::IOFile& file) const; - - bool HasProgramA() const { - return !code.empty() && !code_b.empty(); - } - - Tegra::Engines::ShaderType type{}; - ProgramCode code; - ProgramCode code_b; - - u64 unique_identifier = 0; - std::optional<u32> texture_handler_size; - u32 bound_buffer = 0; - VideoCommon::Shader::GraphicsInfo graphics_info; - VideoCommon::Shader::ComputeInfo compute_info; - VideoCommon::Shader::KeyMap keys; - VideoCommon::Shader::BoundSamplerMap bound_samplers; - VideoCommon::Shader::SeparateSamplerMap separate_samplers; - VideoCommon::Shader::BindlessSamplerMap bindless_samplers; -}; - -/// Contains an OpenGL dumped binary program -struct ShaderDiskCachePrecompiled { - u64 unique_identifier = 0; - GLenum binary_format = 0; - std::vector<u8> binary; -}; - -class ShaderDiskCacheOpenGL { -public: - explicit ShaderDiskCacheOpenGL(); - ~ShaderDiskCacheOpenGL(); - - /// Binds a title ID for all future operations. - void BindTitleID(u64 title_id); - - /// Loads transferable cache. If file has a old version or on failure, it deletes the file. - std::optional<std::vector<ShaderDiskCacheEntry>> LoadTransferable(); - - /// Loads current game's precompiled cache. Invalidates on failure. - std::vector<ShaderDiskCachePrecompiled> LoadPrecompiled(); - - /// Removes the transferable (and precompiled) cache file. - void InvalidateTransferable(); - - /// Removes the precompiled cache file and clears virtual precompiled cache file. - void InvalidatePrecompiled(); - - /// Saves a raw dump to the transferable file. Checks for collisions. - void SaveEntry(const ShaderDiskCacheEntry& entry); - - /// Saves a dump entry to the precompiled file. Does not check for collisions. - void SavePrecompiled(u64 unique_identifier, GLuint program); - - /// Serializes virtual precompiled shader cache file to real file - void SaveVirtualPrecompiledFile(); - -private: - /// Loads the transferable cache. Returns empty on failure. - std::optional<std::vector<ShaderDiskCachePrecompiled>> LoadPrecompiledFile( - Common::FS::IOFile& file); - - /// Opens current game's transferable file and write it's header if it doesn't exist - Common::FS::IOFile AppendTransferableFile() const; - - /// Save precompiled header to precompiled_cache_in_memory - void SavePrecompiledHeaderToVirtualPrecompiledCache(); - - /// Create shader disk cache directories. Returns true on success. - bool EnsureDirectories() const; - - /// Gets current game's transferable file path - std::filesystem::path GetTransferablePath() const; - - /// Gets current game's precompiled file path - std::filesystem::path GetPrecompiledPath() const; - - /// Get user's transferable directory path - std::filesystem::path GetTransferableDir() const; - - /// Get user's precompiled directory path - std::filesystem::path GetPrecompiledDir() const; - - /// Get user's shader directory path - std::filesystem::path GetBaseDir() const; - - /// Get current game's title id - std::string GetTitleID() const; - - template <typename T> - bool SaveArrayToPrecompiled(const T* data, std::size_t length) { - const std::size_t write_length = precompiled_cache_virtual_file.WriteArray( - data, length, precompiled_cache_virtual_file_offset); - precompiled_cache_virtual_file_offset += write_length; - return write_length == sizeof(T) * length; - } - - template <typename T> - bool LoadArrayFromPrecompiled(T* data, std::size_t length) { - const std::size_t read_length = precompiled_cache_virtual_file.ReadArray( - data, length, precompiled_cache_virtual_file_offset); - precompiled_cache_virtual_file_offset += read_length; - return read_length == sizeof(T) * length; - } - - template <typename T> - bool SaveObjectToPrecompiled(const T& object) { - return SaveArrayToPrecompiled(&object, 1); - } - - bool SaveObjectToPrecompiled(bool object) { - const auto value = static_cast<u8>(object); - return SaveArrayToPrecompiled(&value, 1); - } - - template <typename T> - bool LoadObjectFromPrecompiled(T& object) { - return LoadArrayFromPrecompiled(&object, 1); - } - - // Stores whole precompiled cache which will be read from or saved to the precompiled chache - // file - FileSys::VectorVfsFile precompiled_cache_virtual_file; - // Stores the current offset of the precompiled cache file for IO purposes - std::size_t precompiled_cache_virtual_file_offset = 0; - - // Stored transferable shaders - std::unordered_set<u64> stored_transferable; - - /// Title ID to operate on - u64 title_id = 0; - - // The cache has been loaded at boot - bool is_usable = false; -}; - -} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_manager.cpp b/src/video_core/renderer_opengl/gl_shader_manager.cpp index 553e6e8d6..399959afb 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.cpp +++ b/src/video_core/renderer_opengl/gl_shader_manager.cpp @@ -1,149 +1,3 @@ // Copyright 2018 yuzu Emulator Project // Licensed under GPLv2 or any later version // Refer to the license.txt file included. - -#include <glad/glad.h> - -#include "common/common_types.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/renderer_opengl/gl_device.h" -#include "video_core/renderer_opengl/gl_shader_manager.h" - -namespace OpenGL { - -namespace { - -void BindProgram(GLenum stage, GLuint current, GLuint old, bool& enabled) { - if (current == old) { - return; - } - if (current == 0) { - if (enabled) { - enabled = false; - glDisable(stage); - } - return; - } - if (!enabled) { - enabled = true; - glEnable(stage); - } - glBindProgramARB(stage, current); -} - -} // Anonymous namespace - -ProgramManager::ProgramManager(const Device& device) - : use_assembly_programs{device.UseAssemblyShaders()} { - if (use_assembly_programs) { - glEnable(GL_COMPUTE_PROGRAM_NV); - } else { - graphics_pipeline.Create(); - glBindProgramPipeline(graphics_pipeline.handle); - } -} - -ProgramManager::~ProgramManager() = default; - -void ProgramManager::BindCompute(GLuint program) { - if (use_assembly_programs) { - glBindProgramARB(GL_COMPUTE_PROGRAM_NV, program); - } else { - is_graphics_bound = false; - glUseProgram(program); - } -} - -void ProgramManager::BindGraphicsPipeline() { - if (!use_assembly_programs) { - UpdateSourcePrograms(); - } -} - -void ProgramManager::BindHostPipeline(GLuint pipeline) { - if (use_assembly_programs) { - if (geometry_enabled) { - geometry_enabled = false; - old_state.geometry = 0; - glDisable(GL_GEOMETRY_PROGRAM_NV); - } - } else { - if (!is_graphics_bound) { - glUseProgram(0); - } - } - glBindProgramPipeline(pipeline); -} - -void ProgramManager::RestoreGuestPipeline() { - if (use_assembly_programs) { - glBindProgramPipeline(0); - } else { - glBindProgramPipeline(graphics_pipeline.handle); - } -} - -void ProgramManager::BindHostCompute(GLuint program) { - if (use_assembly_programs) { - glDisable(GL_COMPUTE_PROGRAM_NV); - } - glUseProgram(program); - is_graphics_bound = false; -} - -void ProgramManager::RestoreGuestCompute() { - if (use_assembly_programs) { - glEnable(GL_COMPUTE_PROGRAM_NV); - glUseProgram(0); - } -} - -void ProgramManager::UseVertexShader(GLuint program) { - if (use_assembly_programs) { - BindProgram(GL_VERTEX_PROGRAM_NV, program, current_state.vertex, vertex_enabled); - } - current_state.vertex = program; -} - -void ProgramManager::UseGeometryShader(GLuint program) { - if (use_assembly_programs) { - BindProgram(GL_GEOMETRY_PROGRAM_NV, program, current_state.vertex, geometry_enabled); - } - current_state.geometry = program; -} - -void ProgramManager::UseFragmentShader(GLuint program) { - if (use_assembly_programs) { - BindProgram(GL_FRAGMENT_PROGRAM_NV, program, current_state.vertex, fragment_enabled); - } - current_state.fragment = program; -} - -void ProgramManager::UpdateSourcePrograms() { - if (!is_graphics_bound) { - is_graphics_bound = true; - glUseProgram(0); - } - - const GLuint handle = graphics_pipeline.handle; - const auto update_state = [handle](GLenum stage, GLuint current, GLuint old) { - if (current == old) { - return; - } - glUseProgramStages(handle, stage, current); - }; - update_state(GL_VERTEX_SHADER_BIT, current_state.vertex, old_state.vertex); - update_state(GL_GEOMETRY_SHADER_BIT, current_state.geometry, old_state.geometry); - update_state(GL_FRAGMENT_SHADER_BIT, current_state.fragment, old_state.fragment); - - old_state = current_state; -} - -void MaxwellUniformData::SetFromRegs(const Tegra::Engines::Maxwell3D& maxwell) { - const auto& regs = maxwell.regs; - - // Y_NEGATE controls what value S2R returns for the Y_DIRECTION system value. - y_direction = regs.screen_y_control.y_negate == 0 ? 1.0f : -1.0f; -} - -} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_manager.h b/src/video_core/renderer_opengl/gl_shader_manager.h index ad42cce74..d7ef0775d 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.h +++ b/src/video_core/renderer_opengl/gl_shader_manager.h @@ -4,79 +4,142 @@ #pragma once -#include <cstddef> +#include <array> +#include <span> #include <glad/glad.h> +#include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_resource_manager.h" -#include "video_core/renderer_opengl/maxwell_to_gl.h" namespace OpenGL { -class Device; - -/// Uniform structure for the Uniform Buffer Object, all vectors must be 16-byte aligned -/// @note Always keep a vec4 at the end. The GL spec is not clear whether the alignment at -/// the end of a uniform block is included in UNIFORM_BLOCK_DATA_SIZE or not. -/// Not following that rule will cause problems on some AMD drivers. -struct alignas(16) MaxwellUniformData { - void SetFromRegs(const Tegra::Engines::Maxwell3D& maxwell); - - GLfloat y_direction; -}; -static_assert(sizeof(MaxwellUniformData) == 16, "MaxwellUniformData structure size is incorrect"); -static_assert(sizeof(MaxwellUniformData) < 16384, - "MaxwellUniformData structure must be less than 16kb as per the OpenGL spec"); - class ProgramManager { -public: - explicit ProgramManager(const Device& device); - ~ProgramManager(); - - /// Binds a compute program - void BindCompute(GLuint program); - - /// Updates bound programs. - void BindGraphicsPipeline(); - - /// Binds an OpenGL pipeline object unsynchronized with the guest state. - void BindHostPipeline(GLuint pipeline); - - /// Rewinds BindHostPipeline state changes. - void RestoreGuestPipeline(); - - /// Binds an OpenGL GLSL program object unsynchronized with the guest state. - void BindHostCompute(GLuint program); + static constexpr size_t NUM_STAGES = 5; - /// Rewinds BindHostCompute state changes. - void RestoreGuestCompute(); - - void UseVertexShader(GLuint program); - void UseGeometryShader(GLuint program); - void UseFragmentShader(GLuint program); - -private: - struct PipelineState { - GLuint vertex = 0; - GLuint geometry = 0; - GLuint fragment = 0; + static constexpr std::array ASSEMBLY_PROGRAM_ENUMS{ + GL_VERTEX_PROGRAM_NV, GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV, + GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV, }; - /// Update GLSL programs. - void UpdateSourcePrograms(); - - OGLPipeline graphics_pipeline; - - PipelineState current_state; - PipelineState old_state; - - bool use_assembly_programs = false; - - bool is_graphics_bound = true; +public: + explicit ProgramManager(const Device& device) { + glCreateProgramPipelines(1, &pipeline.handle); + if (device.UseAssemblyShaders()) { + glEnable(GL_COMPUTE_PROGRAM_NV); + } + } + + void BindComputeProgram(GLuint program) { + glUseProgram(program); + is_compute_bound = true; + } + + void BindComputeAssemblyProgram(GLuint program) { + if (current_assembly_compute_program != program) { + current_assembly_compute_program = program; + glBindProgramARB(GL_COMPUTE_PROGRAM_NV, program); + } + UnbindPipeline(); + } + + void BindSourcePrograms(std::span<const OGLProgram, NUM_STAGES> programs) { + static constexpr std::array<GLenum, 5> stage_enums{ + GL_VERTEX_SHADER_BIT, GL_TESS_CONTROL_SHADER_BIT, GL_TESS_EVALUATION_SHADER_BIT, + GL_GEOMETRY_SHADER_BIT, GL_FRAGMENT_SHADER_BIT, + }; + for (size_t stage = 0; stage < NUM_STAGES; ++stage) { + if (current_programs[stage] != programs[stage].handle) { + current_programs[stage] = programs[stage].handle; + glUseProgramStages(pipeline.handle, stage_enums[stage], programs[stage].handle); + } + } + BindPipeline(); + } + + void BindPresentPrograms(GLuint vertex, GLuint fragment) { + if (current_programs[0] != vertex) { + current_programs[0] = vertex; + glUseProgramStages(pipeline.handle, GL_VERTEX_SHADER_BIT, vertex); + } + if (current_programs[4] != fragment) { + current_programs[4] = fragment; + glUseProgramStages(pipeline.handle, GL_FRAGMENT_SHADER_BIT, fragment); + } + glUseProgramStages( + pipeline.handle, + GL_TESS_CONTROL_SHADER_BIT | GL_TESS_EVALUATION_SHADER_BIT | GL_GEOMETRY_SHADER_BIT, 0); + current_programs[1] = 0; + current_programs[2] = 0; + current_programs[3] = 0; + + if (current_stage_mask != 0) { + current_stage_mask = 0; + for (const GLenum program_type : ASSEMBLY_PROGRAM_ENUMS) { + glDisable(program_type); + } + } + BindPipeline(); + } + + void BindAssemblyPrograms(std::span<const OGLAssemblyProgram, NUM_STAGES> programs, + u32 stage_mask) { + const u32 changed_mask = current_stage_mask ^ stage_mask; + current_stage_mask = stage_mask; + + if (changed_mask != 0) { + for (size_t stage = 0; stage < NUM_STAGES; ++stage) { + if (((changed_mask >> stage) & 1) != 0) { + if (((stage_mask >> stage) & 1) != 0) { + glEnable(ASSEMBLY_PROGRAM_ENUMS[stage]); + } else { + glDisable(ASSEMBLY_PROGRAM_ENUMS[stage]); + } + } + } + } + for (size_t stage = 0; stage < NUM_STAGES; ++stage) { + if (current_programs[stage] != programs[stage].handle) { + current_programs[stage] = programs[stage].handle; + glBindProgramARB(ASSEMBLY_PROGRAM_ENUMS[stage], programs[stage].handle); + } + } + UnbindPipeline(); + } + + void RestoreGuestCompute() {} - bool vertex_enabled = false; - bool geometry_enabled = false; - bool fragment_enabled = false; +private: + void BindPipeline() { + if (!is_pipeline_bound) { + is_pipeline_bound = true; + glBindProgramPipeline(pipeline.handle); + } + UnbindCompute(); + } + + void UnbindPipeline() { + if (is_pipeline_bound) { + is_pipeline_bound = false; + glBindProgramPipeline(0); + } + UnbindCompute(); + } + + void UnbindCompute() { + if (is_compute_bound) { + is_compute_bound = false; + glUseProgram(0); + } + } + + OGLPipeline pipeline; + bool is_pipeline_bound{}; + bool is_compute_bound{}; + + u32 current_stage_mask = 0; + std::array<GLuint, NUM_STAGES> current_programs{}; + GLuint current_assembly_compute_program = 0; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_util.cpp b/src/video_core/renderer_opengl/gl_shader_util.cpp index 4bf0d6090..d432072ad 100644 --- a/src/video_core/renderer_opengl/gl_shader_util.cpp +++ b/src/video_core/renderer_opengl/gl_shader_util.cpp @@ -5,57 +5,108 @@ #include <string_view> #include <vector> #include <glad/glad.h> + #include "common/assert.h" #include "common/logging/log.h" +#include "common/settings.h" #include "video_core/renderer_opengl/gl_shader_util.h" -namespace OpenGL::GLShader { +namespace OpenGL { -namespace { +static OGLProgram LinkSeparableProgram(GLuint shader) { + OGLProgram program; + program.handle = glCreateProgram(); + glProgramParameteri(program.handle, GL_PROGRAM_SEPARABLE, GL_TRUE); + glAttachShader(program.handle, shader); + glLinkProgram(program.handle); + if (!Settings::values.renderer_debug) { + return program; + } + GLint link_status{}; + glGetProgramiv(program.handle, GL_LINK_STATUS, &link_status); -std::string_view StageDebugName(GLenum type) { - switch (type) { - case GL_VERTEX_SHADER: - return "vertex"; - case GL_GEOMETRY_SHADER: - return "geometry"; - case GL_FRAGMENT_SHADER: - return "fragment"; - case GL_COMPUTE_SHADER: - return "compute"; + GLint log_length{}; + glGetProgramiv(program.handle, GL_INFO_LOG_LENGTH, &log_length); + if (log_length == 0) { + return program; + } + std::string log(log_length, 0); + glGetProgramInfoLog(program.handle, log_length, nullptr, log.data()); + if (link_status == GL_FALSE) { + LOG_ERROR(Render_OpenGL, "{}", log); + } else { + LOG_WARNING(Render_OpenGL, "{}", log); } - UNIMPLEMENTED(); - return "unknown"; + return program; } -} // Anonymous namespace +static void LogShader(GLuint shader, std::string_view code = {}) { + GLint shader_status{}; + glGetShaderiv(shader, GL_COMPILE_STATUS, &shader_status); + if (shader_status == GL_FALSE) { + LOG_ERROR(Render_OpenGL, "Failed to build shader"); + } + GLint log_length{}; + glGetShaderiv(shader, GL_INFO_LOG_LENGTH, &log_length); + if (log_length == 0) { + return; + } + std::string log(log_length, 0); + glGetShaderInfoLog(shader, log_length, nullptr, log.data()); + if (shader_status == GL_FALSE) { + LOG_ERROR(Render_OpenGL, "{}", log); + if (!code.empty()) { + LOG_INFO(Render_OpenGL, "\n{}", code); + } + } else { + LOG_WARNING(Render_OpenGL, "{}", log); + } +} -GLuint LoadShader(std::string_view source, GLenum type) { - const std::string_view debug_type = StageDebugName(type); - const GLuint shader_id = glCreateShader(type); +OGLProgram CreateProgram(std::string_view code, GLenum stage) { + OGLShader shader; + shader.handle = glCreateShader(stage); - const GLchar* source_string = source.data(); - const GLint source_length = static_cast<GLint>(source.size()); + const GLint length = static_cast<GLint>(code.size()); + const GLchar* const code_ptr = code.data(); + glShaderSource(shader.handle, 1, &code_ptr, &length); + glCompileShader(shader.handle); + if (Settings::values.renderer_debug) { + LogShader(shader.handle, code); + } + return LinkSeparableProgram(shader.handle); +} - glShaderSource(shader_id, 1, &source_string, &source_length); - LOG_DEBUG(Render_OpenGL, "Compiling {} shader...", debug_type); - glCompileShader(shader_id); +OGLProgram CreateProgram(std::span<const u32> code, GLenum stage) { + OGLShader shader; + shader.handle = glCreateShader(stage); - GLint result = GL_FALSE; - GLint info_log_length; - glGetShaderiv(shader_id, GL_COMPILE_STATUS, &result); - glGetShaderiv(shader_id, GL_INFO_LOG_LENGTH, &info_log_length); + glShaderBinary(1, &shader.handle, GL_SHADER_BINARY_FORMAT_SPIR_V_ARB, code.data(), + static_cast<GLsizei>(code.size_bytes())); + glSpecializeShader(shader.handle, "main", 0, nullptr, nullptr); + if (Settings::values.renderer_debug) { + LogShader(shader.handle); + } + return LinkSeparableProgram(shader.handle); +} - if (info_log_length > 1) { - std::string shader_error(info_log_length, ' '); - glGetShaderInfoLog(shader_id, info_log_length, nullptr, &shader_error[0]); - if (result == GL_TRUE) { - LOG_DEBUG(Render_OpenGL, "{}", shader_error); - } else { - LOG_ERROR(Render_OpenGL, "Error compiling {} shader:\n{}", debug_type, shader_error); +OGLAssemblyProgram CompileProgram(std::string_view code, GLenum target) { + OGLAssemblyProgram program; + glGenProgramsARB(1, &program.handle); + glNamedProgramStringEXT(program.handle, target, GL_PROGRAM_FORMAT_ASCII_ARB, + static_cast<GLsizei>(code.size()), code.data()); + if (Settings::values.renderer_debug) { + const auto err = reinterpret_cast<const char*>(glGetString(GL_PROGRAM_ERROR_STRING_NV)); + if (err && *err) { + if (std::strstr(err, "error")) { + LOG_CRITICAL(Render_OpenGL, "\n{}", err); + LOG_INFO(Render_OpenGL, "\n{}", code); + } else { + LOG_WARNING(Render_OpenGL, "\n{}", err); + } } } - return shader_id; + return program; } -} // namespace OpenGL::GLShader +} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_util.h b/src/video_core/renderer_opengl/gl_shader_util.h index 1b770532e..4e1a2a8e1 100644 --- a/src/video_core/renderer_opengl/gl_shader_util.h +++ b/src/video_core/renderer_opengl/gl_shader_util.h @@ -4,92 +4,23 @@ #pragma once +#include <span> #include <string> +#include <string_view> #include <vector> + #include <glad/glad.h> + #include "common/assert.h" #include "common/logging/log.h" +#include "video_core/renderer_opengl/gl_resource_manager.h" -namespace OpenGL::GLShader { - -/** - * Utility function to log the source code of a list of shaders. - * @param shaders The OpenGL shaders whose source we will print. - */ -template <typename... T> -void LogShaderSource(T... shaders) { - auto shader_list = {shaders...}; - - for (const auto& shader : shader_list) { - if (shader == 0) - continue; - - GLint source_length; - glGetShaderiv(shader, GL_SHADER_SOURCE_LENGTH, &source_length); - - std::string source(source_length, ' '); - glGetShaderSource(shader, source_length, nullptr, &source[0]); - LOG_INFO(Render_OpenGL, "Shader source {}", source); - } -} - -/** - * Utility function to create and compile an OpenGL GLSL shader - * @param source String of the GLSL shader program - * @param type Type of the shader (GL_VERTEX_SHADER, GL_GEOMETRY_SHADER or GL_FRAGMENT_SHADER) - */ -GLuint LoadShader(std::string_view source, GLenum type); - -/** - * Utility function to create and compile an OpenGL GLSL shader program (vertex + fragment shader) - * @param separable_program whether to create a separable program - * @param shaders ID of shaders to attach to the program - * @returns Handle of the newly created OpenGL program object - */ -template <typename... T> -GLuint LoadProgram(bool separable_program, bool hint_retrievable, T... shaders) { - // Link the program - LOG_DEBUG(Render_OpenGL, "Linking program..."); - - GLuint program_id = glCreateProgram(); - - ((shaders == 0 ? (void)0 : glAttachShader(program_id, shaders)), ...); - - if (separable_program) { - glProgramParameteri(program_id, GL_PROGRAM_SEPARABLE, GL_TRUE); - } - if (hint_retrievable) { - glProgramParameteri(program_id, GL_PROGRAM_BINARY_RETRIEVABLE_HINT, GL_TRUE); - } - - glLinkProgram(program_id); - - // Check the program - GLint result = GL_FALSE; - GLint info_log_length; - glGetProgramiv(program_id, GL_LINK_STATUS, &result); - glGetProgramiv(program_id, GL_INFO_LOG_LENGTH, &info_log_length); - - if (info_log_length > 1) { - std::string program_error(info_log_length, ' '); - glGetProgramInfoLog(program_id, info_log_length, nullptr, &program_error[0]); - if (result == GL_TRUE) { - LOG_DEBUG(Render_OpenGL, "{}", program_error); - } else { - LOG_ERROR(Render_OpenGL, "Error linking shader:\n{}", program_error); - } - } - - if (result == GL_FALSE) { - // There was a problem linking the shader, print the source for debugging purposes. - LogShaderSource(shaders...); - } +namespace OpenGL { - ASSERT_MSG(result == GL_TRUE, "Shader not linked"); +OGLProgram CreateProgram(std::string_view code, GLenum stage); - ((shaders == 0 ? (void)0 : glDetachShader(program_id, shaders)), ...); +OGLProgram CreateProgram(std::span<const u32> code, GLenum stage); - return program_id; -} +OGLAssemblyProgram CompileProgram(std::string_view code, GLenum target); -} // namespace OpenGL::GLShader +} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_state_tracker.cpp b/src/video_core/renderer_opengl/gl_state_tracker.cpp index dbdf5230f..586da84e3 100644 --- a/src/video_core/renderer_opengl/gl_state_tracker.cpp +++ b/src/video_core/renderer_opengl/gl_state_tracker.cpp @@ -83,11 +83,6 @@ void SetupDirtyScissors(Tables& tables) { FillBlock(tables[1], OFF(scissor_test), NUM(scissor_test), Scissors); } -void SetupDirtyShaders(Tables& tables) { - FillBlock(tables[0], OFF(shader_config[0]), NUM(shader_config[0]) * Regs::MaxShaderProgram, - Shaders); -} - void SetupDirtyPolygonModes(Tables& tables) { tables[0][OFF(polygon_mode_front)] = PolygonModeFront; tables[0][OFF(polygon_mode_back)] = PolygonModeBack; @@ -217,7 +212,6 @@ StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags} SetupDirtyScissors(tables); SetupDirtyVertexInstances(tables); SetupDirtyVertexFormat(tables); - SetupDirtyShaders(tables); SetupDirtyPolygonModes(tables); SetupDirtyDepthTest(tables); SetupDirtyStencilTest(tables); diff --git a/src/video_core/renderer_opengl/gl_state_tracker.h b/src/video_core/renderer_opengl/gl_state_tracker.h index 94c905116..5864c7c07 100644 --- a/src/video_core/renderer_opengl/gl_state_tracker.h +++ b/src/video_core/renderer_opengl/gl_state_tracker.h @@ -52,7 +52,6 @@ enum : u8 { BlendState0, BlendState7 = BlendState0 + 7, - Shaders, ClipDistances, PolygonModes, diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index ff0f03e99..c373c9cb4 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -24,9 +24,7 @@ #include "video_core/textures/decoders.h" namespace OpenGL { - namespace { - using Tegra::Texture::SwizzleSource; using Tegra::Texture::TextureMipmapFilter; using Tegra::Texture::TextureType; @@ -59,107 +57,6 @@ struct CopyRegion { GLsizei depth; }; -struct FormatTuple { - GLenum internal_format; - GLenum format = GL_NONE; - GLenum type = GL_NONE; -}; - -constexpr std::array<FormatTuple, MaxPixelFormat> FORMAT_TABLE = {{ - {GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV}, // A8B8G8R8_UNORM - {GL_RGBA8_SNORM, GL_RGBA, GL_BYTE}, // A8B8G8R8_SNORM - {GL_RGBA8I, GL_RGBA_INTEGER, GL_BYTE}, // A8B8G8R8_SINT - {GL_RGBA8UI, GL_RGBA_INTEGER, GL_UNSIGNED_BYTE}, // A8B8G8R8_UINT - {GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5}, // R5G6B5_UNORM - {GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV}, // B5G6R5_UNORM - {GL_RGB5_A1, GL_BGRA, GL_UNSIGNED_SHORT_1_5_5_5_REV}, // A1R5G5B5_UNORM - {GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV}, // A2B10G10R10_UNORM - {GL_RGB10_A2UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT_2_10_10_10_REV}, // A2B10G10R10_UINT - {GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_1_5_5_5_REV}, // A1B5G5R5_UNORM - {GL_R8, GL_RED, GL_UNSIGNED_BYTE}, // R8_UNORM - {GL_R8_SNORM, GL_RED, GL_BYTE}, // R8_SNORM - {GL_R8I, GL_RED_INTEGER, GL_BYTE}, // R8_SINT - {GL_R8UI, GL_RED_INTEGER, GL_UNSIGNED_BYTE}, // R8_UINT - {GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT}, // R16G16B16A16_FLOAT - {GL_RGBA16, GL_RGBA, GL_UNSIGNED_SHORT}, // R16G16B16A16_UNORM - {GL_RGBA16_SNORM, GL_RGBA, GL_SHORT}, // R16G16B16A16_SNORM - {GL_RGBA16I, GL_RGBA_INTEGER, GL_SHORT}, // R16G16B16A16_SINT - {GL_RGBA16UI, GL_RGBA_INTEGER, GL_UNSIGNED_SHORT}, // R16G16B16A16_UINT - {GL_R11F_G11F_B10F, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV}, // B10G11R11_FLOAT - {GL_RGBA32UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT}, // R32G32B32A32_UINT - {GL_COMPRESSED_RGBA_S3TC_DXT1_EXT}, // BC1_RGBA_UNORM - {GL_COMPRESSED_RGBA_S3TC_DXT3_EXT}, // BC2_UNORM - {GL_COMPRESSED_RGBA_S3TC_DXT5_EXT}, // BC3_UNORM - {GL_COMPRESSED_RED_RGTC1}, // BC4_UNORM - {GL_COMPRESSED_SIGNED_RED_RGTC1}, // BC4_SNORM - {GL_COMPRESSED_RG_RGTC2}, // BC5_UNORM - {GL_COMPRESSED_SIGNED_RG_RGTC2}, // BC5_SNORM - {GL_COMPRESSED_RGBA_BPTC_UNORM}, // BC7_UNORM - {GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT}, // BC6H_UFLOAT - {GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT}, // BC6H_SFLOAT - {GL_COMPRESSED_RGBA_ASTC_4x4_KHR}, // ASTC_2D_4X4_UNORM - {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE}, // B8G8R8A8_UNORM - {GL_RGBA32F, GL_RGBA, GL_FLOAT}, // R32G32B32A32_FLOAT - {GL_RGBA32I, GL_RGBA_INTEGER, GL_INT}, // R32G32B32A32_SINT - {GL_RG32F, GL_RG, GL_FLOAT}, // R32G32_FLOAT - {GL_RG32I, GL_RG_INTEGER, GL_INT}, // R32G32_SINT - {GL_R32F, GL_RED, GL_FLOAT}, // R32_FLOAT - {GL_R16F, GL_RED, GL_HALF_FLOAT}, // R16_FLOAT - {GL_R16, GL_RED, GL_UNSIGNED_SHORT}, // R16_UNORM - {GL_R16_SNORM, GL_RED, GL_SHORT}, // R16_SNORM - {GL_R16UI, GL_RED_INTEGER, GL_UNSIGNED_SHORT}, // R16_UINT - {GL_R16I, GL_RED_INTEGER, GL_SHORT}, // R16_SINT - {GL_RG16, GL_RG, GL_UNSIGNED_SHORT}, // R16G16_UNORM - {GL_RG16F, GL_RG, GL_HALF_FLOAT}, // R16G16_FLOAT - {GL_RG16UI, GL_RG_INTEGER, GL_UNSIGNED_SHORT}, // R16G16_UINT - {GL_RG16I, GL_RG_INTEGER, GL_SHORT}, // R16G16_SINT - {GL_RG16_SNORM, GL_RG, GL_SHORT}, // R16G16_SNORM - {GL_RGB32F, GL_RGB, GL_FLOAT}, // R32G32B32_FLOAT - {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV}, // A8B8G8R8_SRGB - {GL_RG8, GL_RG, GL_UNSIGNED_BYTE}, // R8G8_UNORM - {GL_RG8_SNORM, GL_RG, GL_BYTE}, // R8G8_SNORM - {GL_RG8I, GL_RG_INTEGER, GL_BYTE}, // R8G8_SINT - {GL_RG8UI, GL_RG_INTEGER, GL_UNSIGNED_BYTE}, // R8G8_UINT - {GL_RG32UI, GL_RG_INTEGER, GL_UNSIGNED_INT}, // R32G32_UINT - {GL_RGB16F, GL_RGBA, GL_HALF_FLOAT}, // R16G16B16X16_FLOAT - {GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT}, // R32_UINT - {GL_R32I, GL_RED_INTEGER, GL_INT}, // R32_SINT - {GL_COMPRESSED_RGBA_ASTC_8x8_KHR}, // ASTC_2D_8X8_UNORM - {GL_COMPRESSED_RGBA_ASTC_8x5_KHR}, // ASTC_2D_8X5_UNORM - {GL_COMPRESSED_RGBA_ASTC_5x4_KHR}, // ASTC_2D_5X4_UNORM - {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE}, // B8G8R8A8_SRGB - {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT}, // BC1_RGBA_SRGB - {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT}, // BC2_SRGB - {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT}, // BC3_SRGB - {GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM}, // BC7_SRGB - {GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4_REV}, // A4B4G4R4_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_4x4_KHR}, // ASTC_2D_4X4_SRGB - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x8_KHR}, // ASTC_2D_8X8_SRGB - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x5_KHR}, // ASTC_2D_8X5_SRGB - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x4_KHR}, // ASTC_2D_5X4_SRGB - {GL_COMPRESSED_RGBA_ASTC_5x5_KHR}, // ASTC_2D_5X5_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x5_KHR}, // ASTC_2D_5X5_SRGB - {GL_COMPRESSED_RGBA_ASTC_10x8_KHR}, // ASTC_2D_10X8_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR}, // ASTC_2D_10X8_SRGB - {GL_COMPRESSED_RGBA_ASTC_6x6_KHR}, // ASTC_2D_6X6_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x6_KHR}, // ASTC_2D_6X6_SRGB - {GL_COMPRESSED_RGBA_ASTC_10x10_KHR}, // ASTC_2D_10X10_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR}, // ASTC_2D_10X10_SRGB - {GL_COMPRESSED_RGBA_ASTC_12x12_KHR}, // ASTC_2D_12X12_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x12_KHR}, // ASTC_2D_12X12_SRGB - {GL_COMPRESSED_RGBA_ASTC_8x6_KHR}, // ASTC_2D_8X6_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x6_KHR}, // ASTC_2D_8X6_SRGB - {GL_COMPRESSED_RGBA_ASTC_6x5_KHR}, // ASTC_2D_6X5_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x5_KHR}, // ASTC_2D_6X5_SRGB - {GL_RGB9_E5, GL_RGB, GL_UNSIGNED_INT_5_9_9_9_REV}, // E5B9G9R9_FLOAT - {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, // D32_FLOAT - {GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16_UNORM - {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // D24_UNORM_S8_UINT - {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // S8_UINT_D24_UNORM - {GL_DEPTH32F_STENCIL8, GL_DEPTH_STENCIL, - GL_FLOAT_32_UNSIGNED_INT_24_8_REV}, // D32_FLOAT_S8_UINT -}}; - constexpr std::array ACCELERATED_FORMATS{ GL_RGBA32F, GL_RGBA16F, GL_RG32F, GL_RG16F, GL_R11F_G11F_B10F, GL_R32F, GL_R16F, GL_RGBA32UI, GL_RGBA16UI, GL_RGB10_A2UI, GL_RGBA8UI, GL_RG32UI, @@ -170,11 +67,6 @@ constexpr std::array ACCELERATED_FORMATS{ GL_RG8_SNORM, GL_R16_SNORM, GL_R8_SNORM, }; -const FormatTuple& GetFormatTuple(PixelFormat pixel_format) { - ASSERT(static_cast<size_t>(pixel_format) < FORMAT_TABLE.size()); - return FORMAT_TABLE[static_cast<size_t>(pixel_format)]; -} - GLenum ImageTarget(const VideoCommon::ImageInfo& info) { switch (info.type) { case ImageType::e1D: @@ -195,26 +87,24 @@ GLenum ImageTarget(const VideoCommon::ImageInfo& info) { return GL_NONE; } -GLenum ImageTarget(ImageViewType type, int num_samples = 1) { +GLenum ImageTarget(Shader::TextureType type, int num_samples = 1) { const bool is_multisampled = num_samples > 1; switch (type) { - case ImageViewType::e1D: + case Shader::TextureType::Color1D: return GL_TEXTURE_1D; - case ImageViewType::e2D: + case Shader::TextureType::Color2D: return is_multisampled ? GL_TEXTURE_2D_MULTISAMPLE : GL_TEXTURE_2D; - case ImageViewType::Cube: + case Shader::TextureType::ColorCube: return GL_TEXTURE_CUBE_MAP; - case ImageViewType::e3D: + case Shader::TextureType::Color3D: return GL_TEXTURE_3D; - case ImageViewType::e1DArray: + case Shader::TextureType::ColorArray1D: return GL_TEXTURE_1D_ARRAY; - case ImageViewType::e2DArray: + case Shader::TextureType::ColorArray2D: return is_multisampled ? GL_TEXTURE_2D_MULTISAMPLE_ARRAY : GL_TEXTURE_2D_ARRAY; - case ImageViewType::CubeArray: + case Shader::TextureType::ColorArrayCube: return GL_TEXTURE_CUBE_MAP_ARRAY; - case ImageViewType::Rect: - return GL_TEXTURE_RECTANGLE; - case ImageViewType::Buffer: + case Shader::TextureType::Buffer: return GL_TEXTURE_BUFFER; } UNREACHABLE_MSG("Invalid image view type={}", type); @@ -322,7 +212,7 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4 default: return false; } - const GLenum internal_format = GetFormatTuple(info.format).internal_format; + const GLenum internal_format = MaxwellToGL::GetFormatTuple(info.format).internal_format; const auto& format_info = runtime.FormatInfo(info.type, internal_format); if (format_info.is_compressed) { return false; @@ -414,11 +304,10 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4 void AttachTexture(GLuint fbo, GLenum attachment, const ImageView* image_view) { if (False(image_view->flags & VideoCommon::ImageViewFlagBits::Slice)) { - const GLuint texture = image_view->DefaultHandle(); - glNamedFramebufferTexture(fbo, attachment, texture, 0); + glNamedFramebufferTexture(fbo, attachment, image_view->DefaultHandle(), 0); return; } - const GLuint texture = image_view->Handle(ImageViewType::e3D); + const GLuint texture = image_view->Handle(Shader::TextureType::Color3D); if (image_view->range.extent.layers > 1) { // TODO: OpenGL doesn't support rendering to a fixed number of slices glNamedFramebufferTexture(fbo, attachment, texture, 0); @@ -439,6 +328,28 @@ void AttachTexture(GLuint fbo, GLenum attachment, const ImageView* image_view) { } } +[[nodiscard]] GLenum ShaderFormat(Shader::ImageFormat format) { + switch (format) { + case Shader::ImageFormat::Typeless: + break; + case Shader::ImageFormat::R8_SINT: + return GL_R8I; + case Shader::ImageFormat::R8_UINT: + return GL_R8UI; + case Shader::ImageFormat::R16_UINT: + return GL_R16UI; + case Shader::ImageFormat::R16_SINT: + return GL_R16I; + case Shader::ImageFormat::R32_UINT: + return GL_R32UI; + case Shader::ImageFormat::R32G32_UINT: + return GL_RG32UI; + case Shader::ImageFormat::R32G32B32A32_UINT: + return GL_RGBA32UI; + } + UNREACHABLE_MSG("Invalid image format={}", format); + return GL_R32UI; +} } // Anonymous namespace ImageBufferMap::~ImageBufferMap() { @@ -453,7 +364,7 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager& static constexpr std::array TARGETS{GL_TEXTURE_1D_ARRAY, GL_TEXTURE_2D_ARRAY, GL_TEXTURE_3D}; for (size_t i = 0; i < TARGETS.size(); ++i) { const GLenum target = TARGETS[i]; - for (const FormatTuple& tuple : FORMAT_TABLE) { + for (const MaxwellToGL::FormatTuple& tuple : MaxwellToGL::FORMAT_TABLE) { const GLenum format = tuple.internal_format; GLint compat_class; GLint compat_type; @@ -475,11 +386,9 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager& null_image_1d_array.Create(GL_TEXTURE_1D_ARRAY); null_image_cube_array.Create(GL_TEXTURE_CUBE_MAP_ARRAY); null_image_3d.Create(GL_TEXTURE_3D); - null_image_rect.Create(GL_TEXTURE_RECTANGLE); glTextureStorage2D(null_image_1d_array.handle, 1, GL_R8, 1, 1); glTextureStorage3D(null_image_cube_array.handle, 1, GL_R8, 1, 1, 6); glTextureStorage3D(null_image_3d.handle, 1, GL_R8, 1, 1, 1); - glTextureStorage2D(null_image_rect.handle, 1, GL_R8, 1, 1); std::array<GLuint, 4> new_handles; glGenTextures(static_cast<GLsizei>(new_handles.size()), new_handles.data()); @@ -496,29 +405,28 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager& glTextureView(null_image_view_cube.handle, GL_TEXTURE_CUBE_MAP, null_image_cube_array.handle, GL_R8, 0, 1, 0, 6); const std::array texture_handles{ - null_image_1d_array.handle, null_image_cube_array.handle, null_image_3d.handle, - null_image_rect.handle, null_image_view_1d.handle, null_image_view_2d.handle, - null_image_view_2d_array.handle, null_image_view_cube.handle, + null_image_1d_array.handle, null_image_cube_array.handle, null_image_3d.handle, + null_image_view_1d.handle, null_image_view_2d.handle, null_image_view_2d_array.handle, + null_image_view_cube.handle, }; for (const GLuint handle : texture_handles) { static constexpr std::array NULL_SWIZZLE{GL_ZERO, GL_ZERO, GL_ZERO, GL_ZERO}; glTextureParameteriv(handle, GL_TEXTURE_SWIZZLE_RGBA, NULL_SWIZZLE.data()); } - const auto set_view = [this](ImageViewType type, GLuint handle) { + const auto set_view = [this](Shader::TextureType type, GLuint handle) { if (device.HasDebuggingToolAttached()) { const std::string name = fmt::format("NullImage {}", type); glObjectLabel(GL_TEXTURE, handle, static_cast<GLsizei>(name.size()), name.data()); } null_image_views[static_cast<size_t>(type)] = handle; }; - set_view(ImageViewType::e1D, null_image_view_1d.handle); - set_view(ImageViewType::e2D, null_image_view_2d.handle); - set_view(ImageViewType::Cube, null_image_view_cube.handle); - set_view(ImageViewType::e3D, null_image_3d.handle); - set_view(ImageViewType::e1DArray, null_image_1d_array.handle); - set_view(ImageViewType::e2DArray, null_image_view_2d_array.handle); - set_view(ImageViewType::CubeArray, null_image_cube_array.handle); - set_view(ImageViewType::Rect, null_image_rect.handle); + set_view(Shader::TextureType::Color1D, null_image_view_1d.handle); + set_view(Shader::TextureType::Color2D, null_image_view_2d.handle); + set_view(Shader::TextureType::ColorCube, null_image_view_cube.handle); + set_view(Shader::TextureType::Color3D, null_image_3d.handle); + set_view(Shader::TextureType::ColorArray1D, null_image_1d_array.handle); + set_view(Shader::TextureType::ColorArray2D, null_image_view_2d_array.handle); + set_view(Shader::TextureType::ColorArrayCube, null_image_cube_array.handle); } TextureCacheRuntime::~TextureCacheRuntime() = default; @@ -710,7 +618,7 @@ Image::Image(TextureCacheRuntime& runtime, const VideoCommon::ImageInfo& info_, gl_format = GL_RGBA; gl_type = GL_UNSIGNED_INT_8_8_8_8_REV; } else { - const auto& tuple = GetFormatTuple(info.format); + const auto& tuple = MaxwellToGL::GetFormatTuple(info.format); gl_internal_format = tuple.internal_format; gl_format = tuple.format; gl_type = tuple.type; @@ -750,8 +658,7 @@ Image::Image(TextureCacheRuntime& runtime, const VideoCommon::ImageInfo& info_, glTextureStorage3D(handle, num_levels, gl_internal_format, width, height, depth); break; case GL_TEXTURE_BUFFER: - buffer.Create(); - glNamedBufferStorage(buffer.handle, guest_size_bytes, nullptr, 0); + UNREACHABLE(); break; default: UNREACHABLE_MSG("Invalid target=0x{:x}", target); @@ -789,14 +696,6 @@ void Image::UploadMemory(const ImageBufferMap& map, } } -void Image::UploadMemory(const ImageBufferMap& map, - std::span<const VideoCommon::BufferCopy> copies) { - for (const VideoCommon::BufferCopy& copy : copies) { - glCopyNamedBufferSubData(map.buffer, buffer.handle, copy.src_offset + map.offset, - copy.dst_offset, copy.size); - } -} - void Image::DownloadMemory(ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies) { glMemoryBarrier(GL_PIXEL_BUFFER_BARRIER_BIT); // TODO: Move this to its own API @@ -958,23 +857,30 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI if (True(image.flags & ImageFlagBits::Converted)) { internal_format = IsPixelFormatSRGB(info.format) ? GL_SRGB8_ALPHA8 : GL_RGBA8; } else { - internal_format = GetFormatTuple(format).internal_format; + internal_format = MaxwellToGL::GetFormatTuple(format).internal_format; + } + full_range = info.range; + flat_range = info.range; + set_object_label = device.HasDebuggingToolAttached(); + is_render_target = info.IsRenderTarget(); + original_texture = image.texture.handle; + num_samples = image.info.num_samples; + if (!is_render_target) { + swizzle[0] = info.x_source; + swizzle[1] = info.y_source; + swizzle[2] = info.z_source; + swizzle[3] = info.w_source; } - VideoCommon::SubresourceRange flatten_range = info.range; - std::array<GLuint, 2> handles; - stored_views.reserve(2); - switch (info.type) { case ImageViewType::e1DArray: - flatten_range.extent.layers = 1; + flat_range.extent.layers = 1; [[fallthrough]]; case ImageViewType::e1D: - glGenTextures(2, handles.data()); - SetupView(device, image, ImageViewType::e1D, handles[0], info, flatten_range); - SetupView(device, image, ImageViewType::e1DArray, handles[1], info, info.range); + SetupView(Shader::TextureType::Color1D); + SetupView(Shader::TextureType::ColorArray1D); break; case ImageViewType::e2DArray: - flatten_range.extent.layers = 1; + flat_range.extent.layers = 1; [[fallthrough]]; case ImageViewType::e2D: if (True(flags & VideoCommon::ImageViewFlagBits::Slice)) { @@ -984,63 +890,126 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI .base = {.level = info.range.base.level, .layer = 0}, .extent = {.levels = 1, .layers = 1}, }; - glGenTextures(1, handles.data()); - SetupView(device, image, ImageViewType::e3D, handles[0], info, slice_range); - break; + full_range = slice_range; + + SetupView(Shader::TextureType::Color3D); + } else { + SetupView(Shader::TextureType::Color2D); + SetupView(Shader::TextureType::ColorArray2D); } - glGenTextures(2, handles.data()); - SetupView(device, image, ImageViewType::e2D, handles[0], info, flatten_range); - SetupView(device, image, ImageViewType::e2DArray, handles[1], info, info.range); break; case ImageViewType::e3D: - glGenTextures(1, handles.data()); - SetupView(device, image, ImageViewType::e3D, handles[0], info, info.range); + SetupView(Shader::TextureType::Color3D); break; case ImageViewType::CubeArray: - flatten_range.extent.layers = 6; + flat_range.extent.layers = 6; [[fallthrough]]; case ImageViewType::Cube: - glGenTextures(2, handles.data()); - SetupView(device, image, ImageViewType::Cube, handles[0], info, flatten_range); - SetupView(device, image, ImageViewType::CubeArray, handles[1], info, info.range); + SetupView(Shader::TextureType::ColorCube); + SetupView(Shader::TextureType::ColorArrayCube); break; case ImageViewType::Rect: - glGenTextures(1, handles.data()); - SetupView(device, image, ImageViewType::Rect, handles[0], info, info.range); + UNIMPLEMENTED(); break; case ImageViewType::Buffer: - glCreateTextures(GL_TEXTURE_BUFFER, 1, handles.data()); - SetupView(device, image, ImageViewType::Buffer, handles[0], info, info.range); + UNREACHABLE(); + break; + } + switch (info.type) { + case ImageViewType::e1D: + default_handle = Handle(Shader::TextureType::Color1D); + break; + case ImageViewType::e1DArray: + default_handle = Handle(Shader::TextureType::ColorArray1D); + break; + case ImageViewType::e2D: + default_handle = Handle(Shader::TextureType::Color2D); + break; + case ImageViewType::e2DArray: + default_handle = Handle(Shader::TextureType::ColorArray2D); + break; + case ImageViewType::e3D: + default_handle = Handle(Shader::TextureType::Color3D); + break; + case ImageViewType::Cube: + default_handle = Handle(Shader::TextureType::ColorCube); + break; + case ImageViewType::CubeArray: + default_handle = Handle(Shader::TextureType::ColorArrayCube); + break; + default: break; } - default_handle = Handle(info.type); } +ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, + const VideoCommon::ImageViewInfo& view_info, GPUVAddr gpu_addr_) + : VideoCommon::ImageViewBase{info, view_info}, gpu_addr{gpu_addr_}, + buffer_size{VideoCommon::CalculateGuestSizeInBytes(info)} {} + +ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, + const VideoCommon::ImageViewInfo& view_info) + : VideoCommon::ImageViewBase{info, view_info} {} + ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::NullImageParams& params) : VideoCommon::ImageViewBase{params}, views{runtime.null_image_views} {} -void ImageView::SetupView(const Device& device, Image& image, ImageViewType view_type, - GLuint handle, const VideoCommon::ImageViewInfo& info, - VideoCommon::SubresourceRange view_range) { - if (info.type == ImageViewType::Buffer) { - // TODO: Take offset from buffer cache - glTextureBufferRange(handle, internal_format, image.buffer.handle, 0, - image.guest_size_bytes); - } else { - const GLuint parent = image.texture.handle; - const GLenum target = ImageTarget(view_type, image.info.num_samples); - glTextureView(handle, target, parent, internal_format, view_range.base.level, - view_range.extent.levels, view_range.base.layer, view_range.extent.layers); - if (!info.IsRenderTarget()) { - ApplySwizzle(handle, format, info.Swizzle()); - } +GLuint ImageView::StorageView(Shader::TextureType texture_type, Shader::ImageFormat image_format) { + if (image_format == Shader::ImageFormat::Typeless) { + return Handle(texture_type); + } + const bool is_signed{image_format == Shader::ImageFormat::R8_SINT || + image_format == Shader::ImageFormat::R16_SINT}; + if (!storage_views) { + storage_views = std::make_unique<StorageViews>(); + } + auto& type_views{is_signed ? storage_views->signeds : storage_views->unsigneds}; + GLuint& view{type_views[static_cast<size_t>(texture_type)]}; + if (view == 0) { + view = MakeView(texture_type, ShaderFormat(image_format)); + } + return view; +} + +void ImageView::SetupView(Shader::TextureType view_type) { + views[static_cast<size_t>(view_type)] = MakeView(view_type, internal_format); +} + +GLuint ImageView::MakeView(Shader::TextureType view_type, GLenum view_format) { + VideoCommon::SubresourceRange view_range; + switch (view_type) { + case Shader::TextureType::Color1D: + case Shader::TextureType::Color2D: + case Shader::TextureType::ColorCube: + view_range = flat_range; + break; + case Shader::TextureType::ColorArray1D: + case Shader::TextureType::ColorArray2D: + case Shader::TextureType::Color3D: + case Shader::TextureType::ColorArrayCube: + view_range = full_range; + break; + default: + UNREACHABLE(); } - if (device.HasDebuggingToolAttached()) { - const std::string name = VideoCommon::Name(*this, view_type); - glObjectLabel(GL_TEXTURE, handle, static_cast<GLsizei>(name.size()), name.data()); + OGLTextureView& view = stored_views.emplace_back(); + view.Create(); + + const GLenum target = ImageTarget(view_type, num_samples); + glTextureView(view.handle, target, original_texture, view_format, view_range.base.level, + view_range.extent.levels, view_range.base.layer, view_range.extent.layers); + if (!is_render_target) { + std::array<SwizzleSource, 4> casted_swizzle; + std::ranges::transform(swizzle, casted_swizzle.begin(), [](u8 component_swizzle) { + return static_cast<SwizzleSource>(component_swizzle); + }); + ApplySwizzle(view.handle, format, casted_swizzle); + } + if (set_object_label) { + const std::string name = VideoCommon::Name(*this); + glObjectLabel(GL_TEXTURE, view.handle, static_cast<GLsizei>(name.size()), name.data()); } - stored_views.emplace_back().handle = handle; - views[static_cast<size_t>(view_type)] = handle; + return view.handle; } Sampler::Sampler(TextureCacheRuntime& runtime, const TSCEntry& config) { diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index 25fe61566..921072ebe 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -9,6 +9,7 @@ #include <glad/glad.h> +#include "shader_recompiler/shader_info.h" #include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/util_shaders.h" #include "video_core/texture_cache/texture_cache.h" @@ -122,18 +123,17 @@ private: bool has_broken_texture_view_formats = false; StagingBuffers upload_buffers{GL_MAP_WRITE_BIT, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT}; - StagingBuffers download_buffers{GL_MAP_READ_BIT, GL_MAP_READ_BIT}; + StagingBuffers download_buffers{GL_MAP_READ_BIT | GL_CLIENT_STORAGE_BIT, GL_MAP_READ_BIT}; OGLTexture null_image_1d_array; OGLTexture null_image_cube_array; OGLTexture null_image_3d; - OGLTexture null_image_rect; OGLTextureView null_image_view_1d; OGLTextureView null_image_view_2d; OGLTextureView null_image_view_2d_array; OGLTextureView null_image_view_cube; - std::array<GLuint, VideoCommon::NUM_IMAGE_VIEW_TYPES> null_image_views; + std::array<GLuint, Shader::NUM_TEXTURE_TYPES> null_image_views{}; }; class Image : public VideoCommon::ImageBase { @@ -154,8 +154,6 @@ public: void UploadMemory(const ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies); - void UploadMemory(const ImageBufferMap& map, std::span<const VideoCommon::BufferCopy> copies); - void DownloadMemory(ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies); GLuint StorageHandle() noexcept; @@ -170,7 +168,6 @@ private: void CopyImageToBuffer(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset); OGLTexture texture; - OGLBuffer buffer; OGLTextureView store_view; GLenum gl_internal_format = GL_NONE; GLenum gl_format = GL_NONE; @@ -182,10 +179,17 @@ class ImageView : public VideoCommon::ImageViewBase { public: explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageViewInfo&, ImageId, Image&); + explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo&, + const VideoCommon::ImageViewInfo&, GPUVAddr); + explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, + const VideoCommon::ImageViewInfo& view_info); explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams&); - [[nodiscard]] GLuint Handle(ImageViewType query_type) const noexcept { - return views[static_cast<size_t>(query_type)]; + [[nodiscard]] GLuint StorageView(Shader::TextureType texture_type, + Shader::ImageFormat image_format); + + [[nodiscard]] GLuint Handle(Shader::TextureType handle_type) const noexcept { + return views[static_cast<size_t>(handle_type)]; } [[nodiscard]] GLuint DefaultHandle() const noexcept { @@ -196,15 +200,38 @@ public: return internal_format; } + [[nodiscard]] GPUVAddr GpuAddr() const noexcept { + return gpu_addr; + } + + [[nodiscard]] u32 BufferSize() const noexcept { + return buffer_size; + } + private: - void SetupView(const Device& device, Image& image, ImageViewType view_type, GLuint handle, - const VideoCommon::ImageViewInfo& info, - VideoCommon::SubresourceRange view_range); + struct StorageViews { + std::array<GLuint, Shader::NUM_TEXTURE_TYPES> signeds{}; + std::array<GLuint, Shader::NUM_TEXTURE_TYPES> unsigneds{}; + }; + + void SetupView(Shader::TextureType view_type); + + GLuint MakeView(Shader::TextureType view_type, GLenum view_format); - std::array<GLuint, VideoCommon::NUM_IMAGE_VIEW_TYPES> views{}; + std::array<GLuint, Shader::NUM_TEXTURE_TYPES> views{}; std::vector<OGLTextureView> stored_views; - GLuint default_handle = 0; + std::unique_ptr<StorageViews> storage_views; GLenum internal_format = GL_NONE; + GLuint default_handle = 0; + GPUVAddr gpu_addr = 0; + u32 buffer_size = 0; + GLuint original_texture = 0; + int num_samples = 0; + VideoCommon::SubresourceRange flat_range; + VideoCommon::SubresourceRange full_range; + std::array<u8, 4> swizzle{}; + bool set_object_label = false; + bool is_render_target = false; }; class ImageAlloc : public VideoCommon::ImageAllocBase {}; diff --git a/src/video_core/renderer_opengl/maxwell_to_gl.h b/src/video_core/renderer_opengl/maxwell_to_gl.h index f7ad8f370..672f94bfc 100644 --- a/src/video_core/renderer_opengl/maxwell_to_gl.h +++ b/src/video_core/renderer_opengl/maxwell_to_gl.h @@ -5,12 +5,120 @@ #pragma once #include <glad/glad.h> + #include "video_core/engines/maxwell_3d.h" +#include "video_core/surface.h" namespace OpenGL::MaxwellToGL { using Maxwell = Tegra::Engines::Maxwell3D::Regs; +struct FormatTuple { + GLenum internal_format; + GLenum format = GL_NONE; + GLenum type = GL_NONE; +}; + +constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> FORMAT_TABLE = {{ + {GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV}, // A8B8G8R8_UNORM + {GL_RGBA8_SNORM, GL_RGBA, GL_BYTE}, // A8B8G8R8_SNORM + {GL_RGBA8I, GL_RGBA_INTEGER, GL_BYTE}, // A8B8G8R8_SINT + {GL_RGBA8UI, GL_RGBA_INTEGER, GL_UNSIGNED_BYTE}, // A8B8G8R8_UINT + {GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5}, // R5G6B5_UNORM + {GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV}, // B5G6R5_UNORM + {GL_RGB5_A1, GL_BGRA, GL_UNSIGNED_SHORT_1_5_5_5_REV}, // A1R5G5B5_UNORM + {GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV}, // A2B10G10R10_UNORM + {GL_RGB10_A2UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT_2_10_10_10_REV}, // A2B10G10R10_UINT + {GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_1_5_5_5_REV}, // A1B5G5R5_UNORM + {GL_R8, GL_RED, GL_UNSIGNED_BYTE}, // R8_UNORM + {GL_R8_SNORM, GL_RED, GL_BYTE}, // R8_SNORM + {GL_R8I, GL_RED_INTEGER, GL_BYTE}, // R8_SINT + {GL_R8UI, GL_RED_INTEGER, GL_UNSIGNED_BYTE}, // R8_UINT + {GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT}, // R16G16B16A16_FLOAT + {GL_RGBA16, GL_RGBA, GL_UNSIGNED_SHORT}, // R16G16B16A16_UNORM + {GL_RGBA16_SNORM, GL_RGBA, GL_SHORT}, // R16G16B16A16_SNORM + {GL_RGBA16I, GL_RGBA_INTEGER, GL_SHORT}, // R16G16B16A16_SINT + {GL_RGBA16UI, GL_RGBA_INTEGER, GL_UNSIGNED_SHORT}, // R16G16B16A16_UINT + {GL_R11F_G11F_B10F, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV}, // B10G11R11_FLOAT + {GL_RGBA32UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT}, // R32G32B32A32_UINT + {GL_COMPRESSED_RGBA_S3TC_DXT1_EXT}, // BC1_RGBA_UNORM + {GL_COMPRESSED_RGBA_S3TC_DXT3_EXT}, // BC2_UNORM + {GL_COMPRESSED_RGBA_S3TC_DXT5_EXT}, // BC3_UNORM + {GL_COMPRESSED_RED_RGTC1}, // BC4_UNORM + {GL_COMPRESSED_SIGNED_RED_RGTC1}, // BC4_SNORM + {GL_COMPRESSED_RG_RGTC2}, // BC5_UNORM + {GL_COMPRESSED_SIGNED_RG_RGTC2}, // BC5_SNORM + {GL_COMPRESSED_RGBA_BPTC_UNORM}, // BC7_UNORM + {GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT}, // BC6H_UFLOAT + {GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT}, // BC6H_SFLOAT + {GL_COMPRESSED_RGBA_ASTC_4x4_KHR}, // ASTC_2D_4X4_UNORM + {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE}, // B8G8R8A8_UNORM + {GL_RGBA32F, GL_RGBA, GL_FLOAT}, // R32G32B32A32_FLOAT + {GL_RGBA32I, GL_RGBA_INTEGER, GL_INT}, // R32G32B32A32_SINT + {GL_RG32F, GL_RG, GL_FLOAT}, // R32G32_FLOAT + {GL_RG32I, GL_RG_INTEGER, GL_INT}, // R32G32_SINT + {GL_R32F, GL_RED, GL_FLOAT}, // R32_FLOAT + {GL_R16F, GL_RED, GL_HALF_FLOAT}, // R16_FLOAT + {GL_R16, GL_RED, GL_UNSIGNED_SHORT}, // R16_UNORM + {GL_R16_SNORM, GL_RED, GL_SHORT}, // R16_SNORM + {GL_R16UI, GL_RED_INTEGER, GL_UNSIGNED_SHORT}, // R16_UINT + {GL_R16I, GL_RED_INTEGER, GL_SHORT}, // R16_SINT + {GL_RG16, GL_RG, GL_UNSIGNED_SHORT}, // R16G16_UNORM + {GL_RG16F, GL_RG, GL_HALF_FLOAT}, // R16G16_FLOAT + {GL_RG16UI, GL_RG_INTEGER, GL_UNSIGNED_SHORT}, // R16G16_UINT + {GL_RG16I, GL_RG_INTEGER, GL_SHORT}, // R16G16_SINT + {GL_RG16_SNORM, GL_RG, GL_SHORT}, // R16G16_SNORM + {GL_RGB32F, GL_RGB, GL_FLOAT}, // R32G32B32_FLOAT + {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV}, // A8B8G8R8_SRGB + {GL_RG8, GL_RG, GL_UNSIGNED_BYTE}, // R8G8_UNORM + {GL_RG8_SNORM, GL_RG, GL_BYTE}, // R8G8_SNORM + {GL_RG8I, GL_RG_INTEGER, GL_BYTE}, // R8G8_SINT + {GL_RG8UI, GL_RG_INTEGER, GL_UNSIGNED_BYTE}, // R8G8_UINT + {GL_RG32UI, GL_RG_INTEGER, GL_UNSIGNED_INT}, // R32G32_UINT + {GL_RGB16F, GL_RGBA, GL_HALF_FLOAT}, // R16G16B16X16_FLOAT + {GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT}, // R32_UINT + {GL_R32I, GL_RED_INTEGER, GL_INT}, // R32_SINT + {GL_COMPRESSED_RGBA_ASTC_8x8_KHR}, // ASTC_2D_8X8_UNORM + {GL_COMPRESSED_RGBA_ASTC_8x5_KHR}, // ASTC_2D_8X5_UNORM + {GL_COMPRESSED_RGBA_ASTC_5x4_KHR}, // ASTC_2D_5X4_UNORM + {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE}, // B8G8R8A8_SRGB + {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT}, // BC1_RGBA_SRGB + {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT}, // BC2_SRGB + {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT}, // BC3_SRGB + {GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM}, // BC7_SRGB + {GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4_REV}, // A4B4G4R4_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_4x4_KHR}, // ASTC_2D_4X4_SRGB + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x8_KHR}, // ASTC_2D_8X8_SRGB + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x5_KHR}, // ASTC_2D_8X5_SRGB + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x4_KHR}, // ASTC_2D_5X4_SRGB + {GL_COMPRESSED_RGBA_ASTC_5x5_KHR}, // ASTC_2D_5X5_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x5_KHR}, // ASTC_2D_5X5_SRGB + {GL_COMPRESSED_RGBA_ASTC_10x8_KHR}, // ASTC_2D_10X8_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR}, // ASTC_2D_10X8_SRGB + {GL_COMPRESSED_RGBA_ASTC_6x6_KHR}, // ASTC_2D_6X6_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x6_KHR}, // ASTC_2D_6X6_SRGB + {GL_COMPRESSED_RGBA_ASTC_10x10_KHR}, // ASTC_2D_10X10_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR}, // ASTC_2D_10X10_SRGB + {GL_COMPRESSED_RGBA_ASTC_12x12_KHR}, // ASTC_2D_12X12_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x12_KHR}, // ASTC_2D_12X12_SRGB + {GL_COMPRESSED_RGBA_ASTC_8x6_KHR}, // ASTC_2D_8X6_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x6_KHR}, // ASTC_2D_8X6_SRGB + {GL_COMPRESSED_RGBA_ASTC_6x5_KHR}, // ASTC_2D_6X5_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x5_KHR}, // ASTC_2D_6X5_SRGB + {GL_RGB9_E5, GL_RGB, GL_UNSIGNED_INT_5_9_9_9_REV}, // E5B9G9R9_FLOAT + {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, // D32_FLOAT + {GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16_UNORM + {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // D24_UNORM_S8_UINT + {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // S8_UINT_D24_UNORM + {GL_DEPTH32F_STENCIL8, GL_DEPTH_STENCIL, + GL_FLOAT_32_UNSIGNED_INT_24_8_REV}, // D32_FLOAT_S8_UINT +}}; + +inline const FormatTuple& GetFormatTuple(VideoCore::Surface::PixelFormat pixel_format) { + ASSERT(static_cast<size_t>(pixel_format) < FORMAT_TABLE.size()); + return FORMAT_TABLE[static_cast<size_t>(pixel_format)]; +} + inline GLenum VertexFormat(Maxwell::VertexAttribute attrib) { switch (attrib.type) { case Maxwell::VertexAttribute::Type::UnsignedNorm: diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index a718bff7a..285e78384 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -25,6 +25,7 @@ #include "video_core/host_shaders/opengl_present_vert.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_shader_manager.h" +#include "video_core/renderer_opengl/gl_shader_util.h" #include "video_core/renderer_opengl/renderer_opengl.h" #include "video_core/textures/decoders.h" @@ -139,6 +140,26 @@ RendererOpenGL::RendererOpenGL(Core::TelemetrySession& telemetry_session_, } AddTelemetryFields(); InitOpenGLObjects(); + + // Initialize default attributes to match hardware's disabled attributes + GLint max_attribs{}; + glGetIntegerv(GL_MAX_VERTEX_ATTRIBS, &max_attribs); + for (GLint attrib = 0; attrib < max_attribs; ++attrib) { + glVertexAttrib4f(attrib, 0.0f, 0.0f, 0.0f, 1.0f); + } + // Enable seamless cubemaps when per texture parameters are not available + if (!GLAD_GL_ARB_seamless_cubemap_per_texture && !GLAD_GL_AMD_seamless_cubemap_per_texture) { + glEnable(GL_TEXTURE_CUBE_MAP_SEAMLESS); + } + // Enable unified vertex attributes and query vertex buffer address when the driver supports it + if (device.HasVertexBufferUnifiedMemory()) { + glEnableClientState(GL_VERTEX_ATTRIB_ARRAY_UNIFIED_NV); + glEnableClientState(GL_ELEMENT_ARRAY_UNIFIED_NV); + + glMakeNamedBufferResidentNV(vertex_buffer.handle, GL_READ_ONLY); + glGetNamedBufferParameterui64vNV(vertex_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, + &vertex_buffer_address); + } } RendererOpenGL::~RendererOpenGL() = default; @@ -229,22 +250,9 @@ void RendererOpenGL::LoadColorToActiveGLTexture(u8 color_r, u8 color_g, u8 color } void RendererOpenGL::InitOpenGLObjects() { - glClearColor(Settings::values.bg_red.GetValue(), Settings::values.bg_green.GetValue(), - Settings::values.bg_blue.GetValue(), 0.0f); - // Create shader programs - OGLShader vertex_shader; - vertex_shader.Create(HostShaders::OPENGL_PRESENT_VERT, GL_VERTEX_SHADER); - - OGLShader fragment_shader; - fragment_shader.Create(HostShaders::OPENGL_PRESENT_FRAG, GL_FRAGMENT_SHADER); - - vertex_program.Create(true, false, vertex_shader.handle); - fragment_program.Create(true, false, fragment_shader.handle); - - pipeline.Create(); - glUseProgramStages(pipeline.handle, GL_VERTEX_SHADER_BIT, vertex_program.handle); - glUseProgramStages(pipeline.handle, GL_FRAGMENT_SHADER_BIT, fragment_program.handle); + present_vertex = CreateProgram(HostShaders::OPENGL_PRESENT_VERT, GL_VERTEX_SHADER); + present_fragment = CreateProgram(HostShaders::OPENGL_PRESENT_FRAG, GL_FRAGMENT_SHADER); // Generate presentation sampler present_sampler.Create(); @@ -266,21 +274,6 @@ void RendererOpenGL::InitOpenGLObjects() { // Clear screen to black LoadColorToActiveGLTexture(0, 0, 0, 0, screen_info.texture); - - // Enable seamless cubemaps when per texture parameters are not available - if (!GLAD_GL_ARB_seamless_cubemap_per_texture && !GLAD_GL_AMD_seamless_cubemap_per_texture) { - glEnable(GL_TEXTURE_CUBE_MAP_SEAMLESS); - } - - // Enable unified vertex attributes and query vertex buffer address when the driver supports it - if (device.HasVertexBufferUnifiedMemory()) { - glEnableClientState(GL_VERTEX_ATTRIB_ARRAY_UNIFIED_NV); - glEnableClientState(GL_ELEMENT_ARRAY_UNIFIED_NV); - - glMakeNamedBufferResidentNV(vertex_buffer.handle, GL_READ_ONLY); - glGetNamedBufferParameterui64vNV(vertex_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, - &vertex_buffer_address); - } } void RendererOpenGL::AddTelemetryFields() { @@ -337,15 +330,17 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture, void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { if (renderer_settings.set_background_color) { // Update background color before drawing - glClearColor(Settings::values.bg_red.GetValue(), Settings::values.bg_green.GetValue(), - Settings::values.bg_blue.GetValue(), 0.0f); + glClearColor(Settings::values.bg_red.GetValue() / 255.0f, + Settings::values.bg_green.GetValue() / 255.0f, + Settings::values.bg_blue.GetValue() / 255.0f, 1.0f); } // Set projection matrix const std::array ortho_matrix = MakeOrthographicMatrix(static_cast<float>(layout.width), static_cast<float>(layout.height)); - glProgramUniformMatrix3x2fv(vertex_program.handle, ModelViewMatrixLocation, 1, GL_FALSE, - std::data(ortho_matrix)); + program_manager.BindPresentPrograms(present_vertex.handle, present_fragment.handle); + glProgramUniformMatrix3x2fv(present_vertex.handle, ModelViewMatrixLocation, 1, GL_FALSE, + ortho_matrix.data()); const auto& texcoords = screen_info.display_texcoords; auto left = texcoords.left; @@ -406,8 +401,6 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { state_tracker.NotifyClipControl(); state_tracker.NotifyAlphaTest(); - program_manager.BindHostPipeline(pipeline.handle); - state_tracker.ClipControl(GL_LOWER_LEFT, GL_ZERO_TO_ONE); glEnable(GL_CULL_FACE); if (screen_info.display_srgb) { @@ -455,7 +448,8 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { glClear(GL_COLOR_BUFFER_BIT); glDrawArrays(GL_TRIANGLE_STRIP, 0, 4); - program_manager.RestoreGuestPipeline(); + // TODO + // program_manager.RestoreGuestPipeline(); } void RendererOpenGL::RenderScreenshot() { diff --git a/src/video_core/renderer_opengl/renderer_opengl.h b/src/video_core/renderer_opengl/renderer_opengl.h index 0b66f8332..d455f572f 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.h +++ b/src/video_core/renderer_opengl/renderer_opengl.h @@ -12,7 +12,6 @@ #include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_resource_manager.h" -#include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/gl_state_tracker.h" namespace Core { @@ -111,9 +110,8 @@ private: // OpenGL object IDs OGLSampler present_sampler; OGLBuffer vertex_buffer; - OGLProgram vertex_program; - OGLProgram fragment_program; - OGLPipeline pipeline; + OGLProgram present_vertex; + OGLProgram present_fragment; OGLFramebuffer screenshot_framebuffer; // GPU address of the vertex buffer diff --git a/src/video_core/renderer_opengl/util_shaders.cpp b/src/video_core/renderer_opengl/util_shaders.cpp index 8fb5be393..37a4d1d9d 100644 --- a/src/video_core/renderer_opengl/util_shaders.cpp +++ b/src/video_core/renderer_opengl/util_shaders.cpp @@ -16,8 +16,8 @@ #include "video_core/host_shaders/opengl_copy_bc4_comp.h" #include "video_core/host_shaders/opengl_copy_bgra_comp.h" #include "video_core/host_shaders/pitch_unswizzle_comp.h" -#include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/gl_shader_manager.h" +#include "video_core/renderer_opengl/gl_shader_util.h" #include "video_core/renderer_opengl/gl_texture_cache.h" #include "video_core/renderer_opengl/util_shaders.h" #include "video_core/texture_cache/accelerated_swizzle.h" @@ -41,21 +41,14 @@ using VideoCommon::Accelerated::MakeBlockLinearSwizzle3DParams; using VideoCore::Surface::BytesPerBlock; namespace { - OGLProgram MakeProgram(std::string_view source) { - OGLShader shader; - shader.Create(source, GL_COMPUTE_SHADER); - - OGLProgram program; - program.Create(true, false, shader.handle); - return program; + return CreateProgram(source, GL_COMPUTE_SHADER); } size_t NumPixelsInCopy(const VideoCommon::ImageCopy& copy) { return static_cast<size_t>(copy.extent.width * copy.extent.height * copy.src_subresource.num_layers); } - } // Anonymous namespace UtilShaders::UtilShaders(ProgramManager& program_manager_) @@ -86,7 +79,7 @@ void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map, .width = VideoCore::Surface::DefaultBlockWidth(image.info.format), .height = VideoCore::Surface::DefaultBlockHeight(image.info.format), }; - program_manager.BindHostCompute(astc_decoder_program.handle); + program_manager.BindComputeProgram(astc_decoder_program.handle); glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle); glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_ENC_BUFFER, astc_buffer.handle); @@ -134,7 +127,7 @@ void UtilShaders::BlockLinearUpload2D(Image& image, const ImageBufferMap& map, static constexpr GLuint BINDING_INPUT_BUFFER = 1; static constexpr GLuint BINDING_OUTPUT_IMAGE = 0; - program_manager.BindHostCompute(block_linear_unswizzle_2d_program.handle); + program_manager.BindComputeProgram(block_linear_unswizzle_2d_program.handle); glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes); glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle); @@ -173,7 +166,7 @@ void UtilShaders::BlockLinearUpload3D(Image& image, const ImageBufferMap& map, static constexpr GLuint BINDING_OUTPUT_IMAGE = 0; glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes); - program_manager.BindHostCompute(block_linear_unswizzle_3d_program.handle); + program_manager.BindComputeProgram(block_linear_unswizzle_3d_program.handle); glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle); const GLenum store_format = StoreFormat(BytesPerBlock(image.info.format)); @@ -222,7 +215,7 @@ void UtilShaders::PitchUpload(Image& image, const ImageBufferMap& map, UNIMPLEMENTED_IF_MSG(!std::has_single_bit(bytes_per_block), "Non-power of two images are not implemented"); - program_manager.BindHostCompute(pitch_unswizzle_program.handle); + program_manager.BindComputeProgram(pitch_unswizzle_program.handle); glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes); glUniform2ui(LOC_ORIGIN, 0, 0); glUniform2i(LOC_DESTINATION, 0, 0); @@ -250,7 +243,7 @@ void UtilShaders::CopyBC4(Image& dst_image, Image& src_image, std::span<const Im static constexpr GLuint LOC_SRC_OFFSET = 0; static constexpr GLuint LOC_DST_OFFSET = 1; - program_manager.BindHostCompute(copy_bc4_program.handle); + program_manager.BindComputeProgram(copy_bc4_program.handle); for (const ImageCopy& copy : copies) { ASSERT(copy.src_subresource.base_layer == 0); @@ -286,7 +279,7 @@ void UtilShaders::CopyBGR(Image& dst_image, Image& src_image, break; case 4: { // BGRA8 copy - program_manager.BindHostCompute(copy_bgra_program.handle); + program_manager.BindComputeProgram(copy_bgra_program.handle); constexpr GLenum FORMAT = GL_RGBA8; for (const ImageCopy& copy : copies) { ASSERT(copy.src_offset == zero_offset); diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp index b7f5b8bc2..6c1b2f063 100644 --- a/src/video_core/renderer_vulkan/blit_image.cpp +++ b/src/video_core/renderer_vulkan/blit_image.cpp @@ -49,6 +49,16 @@ constexpr VkDescriptorSetLayoutCreateInfo ONE_TEXTURE_DESCRIPTOR_SET_LAYOUT_CREA .bindingCount = 1, .pBindings = &TEXTURE_DESCRIPTOR_SET_LAYOUT_BINDING<0>, }; +template <u32 num_textures> +inline constexpr DescriptorBankInfo TEXTURE_DESCRIPTOR_BANK_INFO{ + .uniform_buffers = 0, + .storage_buffers = 0, + .texture_buffers = 0, + .image_buffers = 0, + .textures = num_textures, + .images = 0, + .score = 2, +}; constexpr VkDescriptorSetLayoutCreateInfo TWO_TEXTURES_DESCRIPTOR_SET_LAYOUT_CREATE_INFO{ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, .pNext = nullptr, @@ -323,18 +333,19 @@ void BindBlitState(vk::CommandBuffer cmdbuf, VkPipelineLayout layout, const Regi cmdbuf.SetScissor(0, scissor); cmdbuf.PushConstants(layout, VK_SHADER_STAGE_VERTEX_BIT, push_constants); } - } // Anonymous namespace BlitImageHelper::BlitImageHelper(const Device& device_, VKScheduler& scheduler_, - StateTracker& state_tracker_, VKDescriptorPool& descriptor_pool) + StateTracker& state_tracker_, DescriptorPool& descriptor_pool) : device{device_}, scheduler{scheduler_}, state_tracker{state_tracker_}, one_texture_set_layout(device.GetLogical().CreateDescriptorSetLayout( ONE_TEXTURE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO)), two_textures_set_layout(device.GetLogical().CreateDescriptorSetLayout( TWO_TEXTURES_DESCRIPTOR_SET_LAYOUT_CREATE_INFO)), - one_texture_descriptor_allocator(descriptor_pool, *one_texture_set_layout), - two_textures_descriptor_allocator(descriptor_pool, *two_textures_set_layout), + one_texture_descriptor_allocator{ + descriptor_pool.Allocator(*one_texture_set_layout, TEXTURE_DESCRIPTOR_BANK_INFO<1>)}, + two_textures_descriptor_allocator{ + descriptor_pool.Allocator(*two_textures_set_layout, TEXTURE_DESCRIPTOR_BANK_INFO<2>)}, one_texture_pipeline_layout(device.GetLogical().CreatePipelineLayout( PipelineLayoutCreateInfo(one_texture_set_layout.address()))), two_textures_pipeline_layout(device.GetLogical().CreatePipelineLayout( @@ -362,14 +373,14 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, const ImageV .operation = operation, }; const VkPipelineLayout layout = *one_texture_pipeline_layout; - const VkImageView src_view = src_image_view.Handle(ImageViewType::e2D); + const VkImageView src_view = src_image_view.Handle(Shader::TextureType::Color2D); const VkSampler sampler = is_linear ? *linear_sampler : *nearest_sampler; const VkPipeline pipeline = FindOrEmplacePipeline(key); - const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit(); scheduler.RequestRenderpass(dst_framebuffer); - scheduler.Record([dst_region, src_region, pipeline, layout, sampler, src_view, descriptor_set, - &device = device](vk::CommandBuffer cmdbuf) { + scheduler.Record([this, dst_region, src_region, pipeline, layout, sampler, + src_view](vk::CommandBuffer cmdbuf) { // TODO: Barriers + const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit(); UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_view); cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline); cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set, @@ -391,12 +402,11 @@ void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer, const VkPipelineLayout layout = *two_textures_pipeline_layout; const VkSampler sampler = *nearest_sampler; const VkPipeline pipeline = BlitDepthStencilPipeline(dst_framebuffer->RenderPass()); - const VkDescriptorSet descriptor_set = two_textures_descriptor_allocator.Commit(); scheduler.RequestRenderpass(dst_framebuffer); scheduler.Record([dst_region, src_region, pipeline, layout, sampler, src_depth_view, - src_stencil_view, descriptor_set, - &device = device](vk::CommandBuffer cmdbuf) { + src_stencil_view, this](vk::CommandBuffer cmdbuf) { // TODO: Barriers + const VkDescriptorSet descriptor_set = two_textures_descriptor_allocator.Commit(); UpdateTwoTexturesDescriptorSet(device, descriptor_set, sampler, src_depth_view, src_stencil_view); cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline); @@ -416,7 +426,6 @@ void BlitImageHelper::ConvertD32ToR32(const Framebuffer* dst_framebuffer, void BlitImageHelper::ConvertR32ToD32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view) { - ConvertColorToDepthPipeline(convert_r32_to_d32_pipeline, dst_framebuffer->RenderPass()); Convert(*convert_r32_to_d32_pipeline, dst_framebuffer, src_image_view); } @@ -436,16 +445,14 @@ void BlitImageHelper::ConvertR16ToD16(const Framebuffer* dst_framebuffer, void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer, const ImageView& src_image_view) { const VkPipelineLayout layout = *one_texture_pipeline_layout; - const VkImageView src_view = src_image_view.Handle(ImageViewType::e2D); + const VkImageView src_view = src_image_view.Handle(Shader::TextureType::Color2D); const VkSampler sampler = *nearest_sampler; - const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit(); const VkExtent2D extent{ .width = src_image_view.size.width, .height = src_image_view.size.height, }; scheduler.RequestRenderpass(dst_framebuffer); - scheduler.Record([pipeline, layout, sampler, src_view, descriptor_set, extent, - &device = device](vk::CommandBuffer cmdbuf) { + scheduler.Record([pipeline, layout, sampler, src_view, extent, this](vk::CommandBuffer cmdbuf) { const VkOffset2D offset{ .x = 0, .y = 0, @@ -466,6 +473,7 @@ void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_frameb .tex_scale = {viewport.width, viewport.height}, .tex_offset = {0.0f, 0.0f}, }; + const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit(); UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_view); // TODO: Barriers diff --git a/src/video_core/renderer_vulkan/blit_image.h b/src/video_core/renderer_vulkan/blit_image.h index 0d81a06ed..33ee095c1 100644 --- a/src/video_core/renderer_vulkan/blit_image.h +++ b/src/video_core/renderer_vulkan/blit_image.h @@ -31,7 +31,7 @@ struct BlitImagePipelineKey { class BlitImageHelper { public: explicit BlitImageHelper(const Device& device, VKScheduler& scheduler, - StateTracker& state_tracker, VKDescriptorPool& descriptor_pool); + StateTracker& state_tracker, DescriptorPool& descriptor_pool); ~BlitImageHelper(); void BlitColor(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp index 362278f01..d70153df3 100644 --- a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp +++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp @@ -15,9 +15,7 @@ #include "video_core/renderer_vulkan/vk_state_tracker.h" namespace Vulkan { - namespace { - constexpr size_t POINT = 0; constexpr size_t LINE = 1; constexpr size_t POLYGON = 2; @@ -39,10 +37,20 @@ constexpr std::array POLYGON_OFFSET_ENABLE_LUT = { POLYGON, // Patches }; +void RefreshXfbState(VideoCommon::TransformFeedbackState& state, const Maxwell& regs) { + std::ranges::transform(regs.tfb_layouts, state.layouts.begin(), [](const auto& layout) { + return VideoCommon::TransformFeedbackState::Layout{ + .stream = layout.stream, + .varying_count = layout.varying_count, + .stride = layout.stride, + }; + }); + state.varyings = regs.tfb_varying_locs; +} } // Anonymous namespace void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d, - bool has_extended_dynamic_state) { + bool has_extended_dynamic_state, bool has_dynamic_vertex_input) { const Maxwell& regs = maxwell3d.regs; const std::array enabled_lut{ regs.polygon_offset_point_enable, @@ -52,6 +60,9 @@ void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d, const u32 topology_index = static_cast<u32>(regs.draw.topology.Value()); raw1 = 0; + extended_dynamic_state.Assign(has_extended_dynamic_state ? 1 : 0); + dynamic_vertex_input.Assign(has_dynamic_vertex_input ? 1 : 0); + xfb_enabled.Assign(regs.tfb_enabled != 0); primitive_restart_enable.Assign(regs.primitive_restart.enabled != 0 ? 1 : 0); depth_bias_enable.Assign(enabled_lut[POLYGON_OFFSET_ENABLE_LUT[topology_index]] != 0 ? 1 : 0); depth_clamp_disabled.Assign(regs.view_volume_clip_control.depth_clamp_disabled.Value()); @@ -63,37 +74,66 @@ void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d, tessellation_clockwise.Assign(regs.tess_mode.cw.Value()); logic_op_enable.Assign(regs.logic_op.enable != 0 ? 1 : 0); logic_op.Assign(PackLogicOp(regs.logic_op.operation)); - rasterize_enable.Assign(regs.rasterize_enable != 0 ? 1 : 0); topology.Assign(regs.draw.topology); msaa_mode.Assign(regs.multisample_mode); raw2 = 0; + rasterize_enable.Assign(regs.rasterize_enable != 0 ? 1 : 0); const auto test_func = regs.alpha_test_enabled != 0 ? regs.alpha_test_func : Maxwell::ComparisonOp::Always; alpha_test_func.Assign(PackComparisonOp(test_func)); early_z.Assign(regs.force_early_fragment_tests != 0 ? 1 : 0); - + depth_enabled.Assign(regs.zeta_enable != 0 ? 1 : 0); + depth_format.Assign(static_cast<u32>(regs.zeta.format)); + y_negate.Assign(regs.screen_y_control.y_negate != 0 ? 1 : 0); + provoking_vertex_last.Assign(regs.provoking_vertex_last != 0 ? 1 : 0); + conservative_raster_enable.Assign(regs.conservative_raster_enable != 0 ? 1 : 0); + smooth_lines.Assign(regs.line_smooth_enable != 0 ? 1 : 0); + + for (size_t i = 0; i < regs.rt.size(); ++i) { + color_formats[i] = static_cast<u8>(regs.rt[i].format); + } alpha_test_ref = Common::BitCast<u32>(regs.alpha_test_ref); point_size = Common::BitCast<u32>(regs.point_size); - if (maxwell3d.dirty.flags[Dirty::InstanceDivisors]) { - maxwell3d.dirty.flags[Dirty::InstanceDivisors] = false; - for (size_t index = 0; index < Maxwell::NumVertexArrays; ++index) { - const bool is_enabled = regs.instanced_arrays.IsInstancingEnabled(index); - binding_divisors[index] = is_enabled ? regs.vertex_array[index].divisor : 0; - } - } - if (maxwell3d.dirty.flags[Dirty::VertexAttributes]) { - maxwell3d.dirty.flags[Dirty::VertexAttributes] = false; - for (size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) { - const auto& input = regs.vertex_attrib_format[index]; - auto& attribute = attributes[index]; - attribute.raw = 0; - attribute.enabled.Assign(input.IsConstant() ? 0 : 1); - attribute.buffer.Assign(input.buffer); - attribute.offset.Assign(input.offset); - attribute.type.Assign(static_cast<u32>(input.type.Value())); - attribute.size.Assign(static_cast<u32>(input.size.Value())); + if (maxwell3d.dirty.flags[Dirty::VertexInput]) { + if (has_dynamic_vertex_input) { + // Dirty flag will be reset by the command buffer update + static constexpr std::array LUT{ + 0u, // Invalid + 1u, // SignedNorm + 1u, // UnsignedNorm + 2u, // SignedInt + 3u, // UnsignedInt + 1u, // UnsignedScaled + 1u, // SignedScaled + 1u, // Float + }; + const auto& attrs = regs.vertex_attrib_format; + attribute_types = 0; + for (size_t i = 0; i < Maxwell::NumVertexAttributes; ++i) { + const u32 mask = attrs[i].constant != 0 ? 0 : 3; + const u32 type = LUT[static_cast<size_t>(attrs[i].type.Value())]; + attribute_types |= static_cast<u64>(type & mask) << (i * 2); + } + } else { + maxwell3d.dirty.flags[Dirty::VertexInput] = false; + enabled_divisors = 0; + for (size_t index = 0; index < Maxwell::NumVertexArrays; ++index) { + const bool is_enabled = regs.instanced_arrays.IsInstancingEnabled(index); + binding_divisors[index] = is_enabled ? regs.vertex_array[index].divisor : 0; + enabled_divisors |= (is_enabled ? u64{1} : 0) << index; + } + for (size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) { + const auto& input = regs.vertex_attrib_format[index]; + auto& attribute = attributes[index]; + attribute.raw = 0; + attribute.enabled.Assign(input.constant ? 0 : 1); + attribute.buffer.Assign(input.buffer); + attribute.offset.Assign(input.offset); + attribute.type.Assign(static_cast<u32>(input.type.Value())); + attribute.size.Assign(static_cast<u32>(input.size.Value())); + } } } if (maxwell3d.dirty.flags[Dirty::Blending]) { @@ -109,10 +149,12 @@ void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d, return static_cast<u16>(viewport.swizzle.raw); }); } - if (!has_extended_dynamic_state) { - no_extended_dynamic_state.Assign(1); + if (!extended_dynamic_state) { dynamic_state.Refresh(regs); } + if (xfb_enabled) { + RefreshXfbState(xfb_state, regs); + } } void FixedPipelineState::BlendingAttachment::Refresh(const Maxwell& regs, size_t index) { diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.h b/src/video_core/renderer_vulkan/fixed_pipeline_state.h index a0eb83a68..c9be37935 100644 --- a/src/video_core/renderer_vulkan/fixed_pipeline_state.h +++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.h @@ -12,6 +12,7 @@ #include "video_core/engines/maxwell_3d.h" #include "video_core/surface.h" +#include "video_core/transform_feedback.h" namespace Vulkan { @@ -60,7 +61,7 @@ struct FixedPipelineState { void Refresh(const Maxwell& regs, size_t index); - constexpr std::array<bool, 4> Mask() const noexcept { + std::array<bool, 4> Mask() const noexcept { return {mask_r != 0, mask_g != 0, mask_b != 0, mask_a != 0}; } @@ -97,11 +98,11 @@ struct FixedPipelineState { BitField<20, 3, u32> type; BitField<23, 6, u32> size; - constexpr Maxwell::VertexAttribute::Type Type() const noexcept { + Maxwell::VertexAttribute::Type Type() const noexcept { return static_cast<Maxwell::VertexAttribute::Type>(type.Value()); } - constexpr Maxwell::VertexAttribute::Size Size() const noexcept { + Maxwell::VertexAttribute::Size Size() const noexcept { return static_cast<Maxwell::VertexAttribute::Size>(size.Value()); } }; @@ -167,37 +168,53 @@ struct FixedPipelineState { union { u32 raw1; - BitField<0, 1, u32> no_extended_dynamic_state; - BitField<2, 1, u32> primitive_restart_enable; - BitField<3, 1, u32> depth_bias_enable; - BitField<4, 1, u32> depth_clamp_disabled; - BitField<5, 1, u32> ndc_minus_one_to_one; - BitField<6, 2, u32> polygon_mode; - BitField<8, 5, u32> patch_control_points_minus_one; - BitField<13, 2, u32> tessellation_primitive; - BitField<15, 2, u32> tessellation_spacing; - BitField<17, 1, u32> tessellation_clockwise; - BitField<18, 1, u32> logic_op_enable; - BitField<19, 4, u32> logic_op; - BitField<23, 1, u32> rasterize_enable; + BitField<0, 1, u32> extended_dynamic_state; + BitField<1, 1, u32> dynamic_vertex_input; + BitField<2, 1, u32> xfb_enabled; + BitField<3, 1, u32> primitive_restart_enable; + BitField<4, 1, u32> depth_bias_enable; + BitField<5, 1, u32> depth_clamp_disabled; + BitField<6, 1, u32> ndc_minus_one_to_one; + BitField<7, 2, u32> polygon_mode; + BitField<9, 5, u32> patch_control_points_minus_one; + BitField<14, 2, u32> tessellation_primitive; + BitField<16, 2, u32> tessellation_spacing; + BitField<18, 1, u32> tessellation_clockwise; + BitField<19, 1, u32> logic_op_enable; + BitField<20, 4, u32> logic_op; BitField<24, 4, Maxwell::PrimitiveTopology> topology; BitField<28, 4, Tegra::Texture::MsaaMode> msaa_mode; }; union { u32 raw2; - BitField<0, 3, u32> alpha_test_func; - BitField<3, 1, u32> early_z; + BitField<0, 1, u32> rasterize_enable; + BitField<1, 3, u32> alpha_test_func; + BitField<4, 1, u32> early_z; + BitField<5, 1, u32> depth_enabled; + BitField<6, 5, u32> depth_format; + BitField<11, 1, u32> y_negate; + BitField<12, 1, u32> provoking_vertex_last; + BitField<13, 1, u32> conservative_raster_enable; + BitField<14, 1, u32> smooth_lines; }; + std::array<u8, Maxwell::NumRenderTargets> color_formats; u32 alpha_test_ref; u32 point_size; - std::array<u32, Maxwell::NumVertexArrays> binding_divisors; - std::array<VertexAttribute, Maxwell::NumVertexAttributes> attributes; std::array<BlendingAttachment, Maxwell::NumRenderTargets> attachments; std::array<u16, Maxwell::NumViewports> viewport_swizzles; + union { + u64 attribute_types; // Used with VK_EXT_vertex_input_dynamic_state + u64 enabled_divisors; + }; + std::array<VertexAttribute, Maxwell::NumVertexAttributes> attributes; + std::array<u32, Maxwell::NumVertexArrays> binding_divisors; + DynamicState dynamic_state; + VideoCommon::TransformFeedbackState xfb_state; - void Refresh(Tegra::Engines::Maxwell3D& maxwell3d, bool has_extended_dynamic_state); + void Refresh(Tegra::Engines::Maxwell3D& maxwell3d, bool has_extended_dynamic_state, + bool has_dynamic_vertex_input); size_t Hash() const noexcept; @@ -208,8 +225,24 @@ struct FixedPipelineState { } size_t Size() const noexcept { - const size_t total_size = sizeof *this; - return total_size - (no_extended_dynamic_state != 0 ? 0 : sizeof(DynamicState)); + if (xfb_enabled) { + // When transform feedback is enabled, use the whole struct + return sizeof(*this); + } + if (dynamic_vertex_input) { + // Exclude dynamic state and attributes + return offsetof(FixedPipelineState, attributes); + } + if (extended_dynamic_state) { + // Exclude dynamic state + return offsetof(FixedPipelineState, dynamic_state); + } + // Default + return offsetof(FixedPipelineState, xfb_state); + } + + u32 DynamicAttributeType(size_t index) const noexcept { + return (attribute_types >> (index * 2)) & 0b11; } }; static_assert(std::has_unique_object_representations_v<FixedPipelineState>); diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp index f088447e9..68a23b602 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp @@ -157,7 +157,7 @@ struct FormatTuple { {VK_FORMAT_R32_SFLOAT, Attachable | Storage}, // R32_FLOAT {VK_FORMAT_R16_SFLOAT, Attachable | Storage}, // R16_FLOAT {VK_FORMAT_R16_UNORM, Attachable | Storage}, // R16_UNORM - {VK_FORMAT_UNDEFINED}, // R16_SNORM + {VK_FORMAT_R16_SNORM, Attachable | Storage}, // R16_SNORM {VK_FORMAT_R16_UINT, Attachable | Storage}, // R16_UINT {VK_FORMAT_UNDEFINED}, // R16_SINT {VK_FORMAT_R16G16_UNORM, Attachable | Storage}, // R16G16_UNORM @@ -266,19 +266,20 @@ FormatInfo SurfaceFormat(const Device& device, FormatType format_type, bool with return {device.GetSupportedFormat(tuple.format, usage, format_type), attachable, storage}; } -VkShaderStageFlagBits ShaderStage(Tegra::Engines::ShaderType stage) { +VkShaderStageFlagBits ShaderStage(Shader::Stage stage) { switch (stage) { - case Tegra::Engines::ShaderType::Vertex: + case Shader::Stage::VertexA: + case Shader::Stage::VertexB: return VK_SHADER_STAGE_VERTEX_BIT; - case Tegra::Engines::ShaderType::TesselationControl: + case Shader::Stage::TessellationControl: return VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT; - case Tegra::Engines::ShaderType::TesselationEval: + case Shader::Stage::TessellationEval: return VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT; - case Tegra::Engines::ShaderType::Geometry: + case Shader::Stage::Geometry: return VK_SHADER_STAGE_GEOMETRY_BIT; - case Tegra::Engines::ShaderType::Fragment: + case Shader::Stage::Fragment: return VK_SHADER_STAGE_FRAGMENT_BIT; - case Tegra::Engines::ShaderType::Compute: + case Shader::Stage::Compute: return VK_SHADER_STAGE_COMPUTE_BIT; } UNIMPLEMENTED_MSG("Unimplemented shader stage={}", stage); @@ -685,6 +686,19 @@ VkCullModeFlagBits CullFace(Maxwell::CullFace cull_face) { return {}; } +VkPolygonMode PolygonMode(Maxwell::PolygonMode polygon_mode) { + switch (polygon_mode) { + case Maxwell::PolygonMode::Point: + return VK_POLYGON_MODE_POINT; + case Maxwell::PolygonMode::Line: + return VK_POLYGON_MODE_LINE; + case Maxwell::PolygonMode::Fill: + return VK_POLYGON_MODE_FILL; + } + UNIMPLEMENTED_MSG("Unimplemented polygon mode={}", polygon_mode); + return {}; +} + VkComponentSwizzle SwizzleSource(Tegra::Texture::SwizzleSource swizzle) { switch (swizzle) { case Tegra::Texture::SwizzleSource::Zero: @@ -741,4 +755,28 @@ VkSamplerReductionMode SamplerReduction(Tegra::Texture::SamplerReduction reducti return VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT; } +VkSampleCountFlagBits MsaaMode(Tegra::Texture::MsaaMode msaa_mode) { + switch (msaa_mode) { + case Tegra::Texture::MsaaMode::Msaa1x1: + return VK_SAMPLE_COUNT_1_BIT; + case Tegra::Texture::MsaaMode::Msaa2x1: + case Tegra::Texture::MsaaMode::Msaa2x1_D3D: + return VK_SAMPLE_COUNT_2_BIT; + case Tegra::Texture::MsaaMode::Msaa2x2: + case Tegra::Texture::MsaaMode::Msaa2x2_VC4: + case Tegra::Texture::MsaaMode::Msaa2x2_VC12: + return VK_SAMPLE_COUNT_4_BIT; + case Tegra::Texture::MsaaMode::Msaa4x2: + case Tegra::Texture::MsaaMode::Msaa4x2_D3D: + case Tegra::Texture::MsaaMode::Msaa4x2_VC8: + case Tegra::Texture::MsaaMode::Msaa4x2_VC24: + return VK_SAMPLE_COUNT_8_BIT; + case Tegra::Texture::MsaaMode::Msaa4x4: + return VK_SAMPLE_COUNT_16_BIT; + default: + UNREACHABLE_MSG("Invalid msaa_mode={}", static_cast<int>(msaa_mode)); + return VK_SAMPLE_COUNT_1_BIT; + } +} + } // namespace Vulkan::MaxwellToVK diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.h b/src/video_core/renderer_vulkan/maxwell_to_vk.h index e3e06ba38..8a9616039 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.h +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.h @@ -5,6 +5,7 @@ #pragma once #include "common/common_types.h" +#include "shader_recompiler/stage.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/surface.h" #include "video_core/textures/texture.h" @@ -45,7 +46,7 @@ struct FormatInfo { [[nodiscard]] FormatInfo SurfaceFormat(const Device& device, FormatType format_type, bool with_srgb, PixelFormat pixel_format); -VkShaderStageFlagBits ShaderStage(Tegra::Engines::ShaderType stage); +VkShaderStageFlagBits ShaderStage(Shader::Stage stage); VkPrimitiveTopology PrimitiveTopology(const Device& device, Maxwell::PrimitiveTopology topology); @@ -65,10 +66,14 @@ VkFrontFace FrontFace(Maxwell::FrontFace front_face); VkCullModeFlagBits CullFace(Maxwell::CullFace cull_face); +VkPolygonMode PolygonMode(Maxwell::PolygonMode polygon_mode); + VkComponentSwizzle SwizzleSource(Tegra::Texture::SwizzleSource swizzle); VkViewportCoordinateSwizzleNV ViewportSwizzle(Maxwell::ViewportSwizzle swizzle); VkSamplerReductionMode SamplerReduction(Tegra::Texture::SamplerReduction reduction); +VkSampleCountFlagBits MsaaMode(Tegra::Texture::MsaaMode msaa_mode); + } // namespace Vulkan::MaxwellToVK diff --git a/src/video_core/renderer_vulkan/pipeline_helper.h b/src/video_core/renderer_vulkan/pipeline_helper.h new file mode 100644 index 000000000..4847db6b6 --- /dev/null +++ b/src/video_core/renderer_vulkan/pipeline_helper.h @@ -0,0 +1,154 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <cstddef> + +#include <boost/container/small_vector.hpp> + +#include "common/assert.h" +#include "common/common_types.h" +#include "shader_recompiler/shader_info.h" +#include "video_core/renderer_vulkan/vk_texture_cache.h" +#include "video_core/renderer_vulkan/vk_update_descriptor.h" +#include "video_core/texture_cache/texture_cache.h" +#include "video_core/texture_cache/types.h" +#include "video_core/textures/texture.h" +#include "video_core/vulkan_common/vulkan_device.h" + +namespace Vulkan { + +class DescriptorLayoutBuilder { +public: + DescriptorLayoutBuilder(const Device& device_) : device{&device_} {} + + bool CanUsePushDescriptor() const noexcept { + return device->IsKhrPushDescriptorSupported() && + num_descriptors <= device->MaxPushDescriptors(); + } + + vk::DescriptorSetLayout CreateDescriptorSetLayout(bool use_push_descriptor) const { + if (bindings.empty()) { + return nullptr; + } + const VkDescriptorSetLayoutCreateFlags flags = + use_push_descriptor ? VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR : 0; + return device->GetLogical().CreateDescriptorSetLayout({ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, + .pNext = nullptr, + .flags = flags, + .bindingCount = static_cast<u32>(bindings.size()), + .pBindings = bindings.data(), + }); + } + + vk::DescriptorUpdateTemplateKHR CreateTemplate(VkDescriptorSetLayout descriptor_set_layout, + VkPipelineLayout pipeline_layout, + bool use_push_descriptor) const { + if (entries.empty()) { + return nullptr; + } + const VkDescriptorUpdateTemplateType type = + use_push_descriptor ? VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_PUSH_DESCRIPTORS_KHR + : VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR; + return device->GetLogical().CreateDescriptorUpdateTemplateKHR({ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_UPDATE_TEMPLATE_CREATE_INFO_KHR, + .pNext = nullptr, + .flags = 0, + .descriptorUpdateEntryCount = static_cast<u32>(entries.size()), + .pDescriptorUpdateEntries = entries.data(), + .templateType = type, + .descriptorSetLayout = descriptor_set_layout, + .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS, + .pipelineLayout = pipeline_layout, + .set = 0, + }); + } + + vk::PipelineLayout CreatePipelineLayout(VkDescriptorSetLayout descriptor_set_layout) const { + return device->GetLogical().CreatePipelineLayout({ + .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .setLayoutCount = descriptor_set_layout ? 1U : 0U, + .pSetLayouts = bindings.empty() ? nullptr : &descriptor_set_layout, + .pushConstantRangeCount = 0, + .pPushConstantRanges = nullptr, + }); + } + + void Add(const Shader::Info& info, VkShaderStageFlags stage) { + Add(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, stage, info.constant_buffer_descriptors); + Add(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, stage, info.storage_buffers_descriptors); + Add(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, stage, info.texture_buffer_descriptors); + Add(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, stage, info.image_buffer_descriptors); + Add(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, stage, info.texture_descriptors); + Add(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, stage, info.image_descriptors); + } + +private: + template <typename Descriptors> + void Add(VkDescriptorType type, VkShaderStageFlags stage, const Descriptors& descriptors) { + const size_t num{descriptors.size()}; + for (size_t i = 0; i < num; ++i) { + bindings.push_back({ + .binding = binding, + .descriptorType = type, + .descriptorCount = descriptors[i].count, + .stageFlags = stage, + .pImmutableSamplers = nullptr, + }); + entries.push_back({ + .dstBinding = binding, + .dstArrayElement = 0, + .descriptorCount = descriptors[i].count, + .descriptorType = type, + .offset = offset, + .stride = sizeof(DescriptorUpdateEntry), + }); + ++binding; + num_descriptors += descriptors[i].count; + offset += sizeof(DescriptorUpdateEntry); + } + } + + const Device* device{}; + boost::container::small_vector<VkDescriptorSetLayoutBinding, 32> bindings; + boost::container::small_vector<VkDescriptorUpdateTemplateEntryKHR, 32> entries; + u32 binding{}; + u32 num_descriptors{}; + size_t offset{}; +}; + +inline void PushImageDescriptors(const Shader::Info& info, const VkSampler*& samplers, + const ImageId*& image_view_ids, TextureCache& texture_cache, + VKUpdateDescriptorQueue& update_descriptor_queue) { + for (const auto& desc : info.texture_buffer_descriptors) { + image_view_ids += desc.count; + } + for (const auto& desc : info.image_buffer_descriptors) { + image_view_ids += desc.count; + } + for (const auto& desc : info.texture_descriptors) { + for (u32 index = 0; index < desc.count; ++index) { + const VkSampler sampler{*(samplers++)}; + ImageView& image_view{texture_cache.GetImageView(*(image_view_ids++))}; + const VkImageView vk_image_view{image_view.Handle(desc.type)}; + update_descriptor_queue.AddSampledImage(vk_image_view, sampler); + } + } + for (const auto& desc : info.image_descriptors) { + for (u32 index = 0; index < desc.count; ++index) { + ImageView& image_view{texture_cache.GetImageView(*(image_view_ids++))}; + if (desc.is_written) { + texture_cache.MarkModification(image_view.image_id); + } + const VkImageView vk_image_view{image_view.StorageView(desc.type, desc.format)}; + update_descriptor_queue.AddImage(vk_image_view); + } + } +} + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp index bec3a81d9..a8d04dc61 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp +++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp @@ -130,35 +130,45 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { if (!framebuffer) { return; } - const auto& layout = render_window.GetFramebufferLayout(); - if (layout.width > 0 && layout.height > 0 && render_window.IsShown()) { - const VAddr framebuffer_addr = framebuffer->address + framebuffer->offset; - const bool use_accelerated = - rasterizer.AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride); - const bool is_srgb = use_accelerated && screen_info.is_srgb; - if (swapchain.HasFramebufferChanged(layout) || swapchain.GetSrgbState() != is_srgb) { - swapchain.Create(layout.width, layout.height, is_srgb); - blit_screen.Recreate(); - } - - scheduler.WaitWorker(); - - while (!swapchain.AcquireNextImage()) { - swapchain.Create(layout.width, layout.height, is_srgb); - blit_screen.Recreate(); + SCOPE_EXIT({ render_window.OnFrameDisplayed(); }); + if (!render_window.IsShown()) { + return; + } + const VAddr framebuffer_addr = framebuffer->address + framebuffer->offset; + const bool use_accelerated = + rasterizer.AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride); + const bool is_srgb = use_accelerated && screen_info.is_srgb; + + bool has_been_recreated = false; + const auto recreate_swapchain = [&] { + if (!has_been_recreated) { + has_been_recreated = true; + scheduler.WaitWorker(); } - const VkSemaphore render_semaphore = blit_screen.Draw(*framebuffer, use_accelerated); - - scheduler.Flush(render_semaphore); - - if (swapchain.Present(render_semaphore)) { - blit_screen.Recreate(); + const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout(); + swapchain.Create(layout.width, layout.height, is_srgb); + }; + if (swapchain.IsSubOptimal() || swapchain.HasColorSpaceChanged(is_srgb)) { + recreate_swapchain(); + } + bool is_outdated; + do { + swapchain.AcquireNextImage(); + is_outdated = swapchain.IsOutDated(); + if (is_outdated) { + recreate_swapchain(); } - gpu.RendererFrameEndNotify(); - rasterizer.TickFrame(); + } while (is_outdated); + if (has_been_recreated) { + blit_screen.Recreate(); } + const VkSemaphore render_semaphore = blit_screen.Draw(*framebuffer, use_accelerated); + scheduler.Flush(render_semaphore); + scheduler.WaitWorker(); + swapchain.Present(render_semaphore); - render_window.OnFrameDisplayed(); + gpu.RendererFrameEndNotify(); + rasterizer.TickFrame(); } void RendererVulkan::Report() const { diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp index a1a32aabe..516f428e7 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp +++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp @@ -184,55 +184,54 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool .depth = 1, }, }; - scheduler.Record( - [buffer = *buffer, image = *raw_images[image_index], copy](vk::CommandBuffer cmdbuf) { - const VkImageMemoryBarrier base_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = 0, - .dstAccessMask = 0, - .oldLayout = VK_IMAGE_LAYOUT_GENERAL, - .newLayout = VK_IMAGE_LAYOUT_GENERAL, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .image = image, - .subresourceRange = - { - .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, - .baseMipLevel = 0, - .levelCount = 1, - .baseArrayLayer = 0, - .layerCount = 1, - }, - }; - VkImageMemoryBarrier read_barrier = base_barrier; - read_barrier.srcAccessMask = VK_ACCESS_HOST_WRITE_BIT; - read_barrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; - read_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED; - - VkImageMemoryBarrier write_barrier = base_barrier; - write_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; - write_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; - - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, read_barrier); - cmdbuf.CopyBufferToImage(buffer, image, VK_IMAGE_LAYOUT_GENERAL, copy); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier); - }); + scheduler.Record([this, copy, image_index](vk::CommandBuffer cmdbuf) { + const VkImage image = *raw_images[image_index]; + const VkImageMemoryBarrier base_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = 0, + .dstAccessMask = 0, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = image, + .subresourceRange{ + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .baseMipLevel = 0, + .levelCount = 1, + .baseArrayLayer = 0, + .layerCount = 1, + }, + }; + VkImageMemoryBarrier read_barrier = base_barrier; + read_barrier.srcAccessMask = VK_ACCESS_HOST_WRITE_BIT; + read_barrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; + read_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED; + + VkImageMemoryBarrier write_barrier = base_barrier; + write_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; + write_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; + + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, + read_barrier); + cmdbuf.CopyBufferToImage(*buffer, image, VK_IMAGE_LAYOUT_GENERAL, copy); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier); + }); } - scheduler.Record([renderpass = *renderpass, framebuffer = *framebuffers[image_index], - descriptor_set = descriptor_sets[image_index], buffer = *buffer, - size = swapchain.GetSize(), pipeline = *pipeline, - layout = *pipeline_layout](vk::CommandBuffer cmdbuf) { + scheduler.Record([this, image_index, size = swapchain.GetSize()](vk::CommandBuffer cmdbuf) { + const f32 bg_red = Settings::values.bg_red.GetValue() / 255.0f; + const f32 bg_green = Settings::values.bg_green.GetValue() / 255.0f; + const f32 bg_blue = Settings::values.bg_blue.GetValue() / 255.0f; const VkClearValue clear_color{ - .color = {.float32 = {0.0f, 0.0f, 0.0f, 0.0f}}, + .color = {.float32 = {bg_red, bg_green, bg_blue, 1.0f}}, }; const VkRenderPassBeginInfo renderpass_bi{ .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO, .pNext = nullptr, - .renderPass = renderpass, - .framebuffer = framebuffer, + .renderPass = *renderpass, + .framebuffer = *framebuffers[image_index], .renderArea = { .offset = {0, 0}, @@ -254,12 +253,13 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool .extent = size, }; cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE); - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline); + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline); cmdbuf.SetViewport(0, viewport); cmdbuf.SetScissor(0, scissor); - cmdbuf.BindVertexBuffer(0, buffer, offsetof(BufferData, vertices)); - cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set, {}); + cmdbuf.BindVertexBuffer(0, *buffer, offsetof(BufferData, vertices)); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline_layout, 0, + descriptor_sets[image_index], {}); cmdbuf.Draw(4, 1, 0, 0); cmdbuf.EndRenderPass(); }); @@ -301,8 +301,7 @@ void VKBlitScreen::CreateShaders() { void VKBlitScreen::CreateSemaphores() { semaphores.resize(image_count); - std::generate(semaphores.begin(), semaphores.end(), - [this] { return device.GetLogical().CreateSemaphore(); }); + std::ranges::generate(semaphores, [this] { return device.GetLogical().CreateSemaphore(); }); } void VKBlitScreen::CreateDescriptorPool() { @@ -630,8 +629,8 @@ void VKBlitScreen::CreateFramebuffers() { } void VKBlitScreen::ReleaseRawImages() { - for (std::size_t i = 0; i < raw_images.size(); ++i) { - scheduler.Wait(resource_ticks.at(i)); + for (const u64 tick : resource_ticks) { + scheduler.Wait(tick); } raw_images.clear(); raw_buffer_commits.clear(); diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 0df4e1a1c..f4b3ee95c 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -60,38 +60,74 @@ std::array<T, 6> MakeQuadIndices(u32 quad, u32 first) { } return indices; } -} // Anonymous namespace - -Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params) - : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params) {} -Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_, - VAddr cpu_addr_, u64 size_bytes_) - : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_) { - buffer = runtime.device.GetLogical().CreateBuffer(VkBufferCreateInfo{ +vk::Buffer CreateBuffer(const Device& device, u64 size) { + VkBufferUsageFlags flags = + VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | + VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT | + VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | + VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_VERTEX_BUFFER_BIT; + if (device.IsExtTransformFeedbackSupported()) { + flags |= VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_BUFFER_BIT_EXT; + } + return device.GetLogical().CreateBuffer({ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, .pNext = nullptr, .flags = 0, - .size = SizeBytes(), - .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | - VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | - VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | - VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT | - VK_BUFFER_USAGE_VERTEX_BUFFER_BIT, + .size = size, + .usage = flags, .sharingMode = VK_SHARING_MODE_EXCLUSIVE, .queueFamilyIndexCount = 0, .pQueueFamilyIndices = nullptr, }); +} +} // Anonymous namespace + +Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params) + : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params) {} + +Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_, + VAddr cpu_addr_, u64 size_bytes_) + : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_), + device{&runtime.device}, buffer{CreateBuffer(*device, SizeBytes())}, + commit{runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal)} { if (runtime.device.HasDebuggingToolAttached()) { buffer.SetObjectNameEXT(fmt::format("Buffer 0x{:x}", CpuAddr()).c_str()); } - commit = runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal); +} + +VkBufferView Buffer::View(u32 offset, u32 size, VideoCore::Surface::PixelFormat format) { + if (!device) { + // Null buffer, return a null descriptor + return VK_NULL_HANDLE; + } + const auto it{std::ranges::find_if(views, [offset, size, format](const BufferView& view) { + return offset == view.offset && size == view.size && format == view.format; + })}; + if (it != views.end()) { + return *it->handle; + } + views.push_back({ + .offset = offset, + .size = size, + .format = format, + .handle = device->GetLogical().CreateBufferView({ + .sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .buffer = *buffer, + .format = MaxwellToVK::SurfaceFormat(*device, FormatType::Buffer, false, format).format, + .offset = offset, + .range = size, + }), + }); + return *views.back().handle; } BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_allocator_, VKScheduler& scheduler_, StagingBufferPool& staging_pool_, VKUpdateDescriptorQueue& update_descriptor_queue_, - VKDescriptorPool& descriptor_pool) + DescriptorPool& descriptor_pool) : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_}, staging_pool{staging_pool_}, update_descriptor_queue{update_descriptor_queue_}, uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue), @@ -136,6 +172,30 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer, }); } +void BufferCacheRuntime::ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t size, u32 value) { + static constexpr VkMemoryBarrier READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + }; + static constexpr VkMemoryBarrier WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + }; + + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([dest_buffer, offset, size, value](vk::CommandBuffer cmdbuf) { + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, + 0, READ_BARRIER); + cmdbuf.FillBuffer(dest_buffer, offset, size, value); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + 0, WRITE_BARRIER); + }); +} + void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format, u32 base_vertex, u32 num_indices, VkBuffer buffer, u32 offset, [[maybe_unused]] u32 size) { @@ -152,8 +212,8 @@ void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat } if (vk_buffer == VK_NULL_HANDLE) { // Vulkan doesn't support null index buffers. Replace it with our own null buffer. - ReserveNullIndexBuffer(); - vk_buffer = *null_index_buffer; + ReserveNullBuffer(); + vk_buffer = *null_buffer; } scheduler.Record([vk_buffer, vk_offset, vk_index_type](vk::CommandBuffer cmdbuf) { cmdbuf.BindIndexBuffer(vk_buffer, vk_offset, vk_index_type); @@ -161,6 +221,13 @@ void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat } void BufferCacheRuntime::BindQuadArrayIndexBuffer(u32 first, u32 count) { + if (count == 0) { + ReserveNullBuffer(); + scheduler.Record([this](vk::CommandBuffer cmdbuf) { + cmdbuf.BindIndexBuffer(*null_buffer, 0, VK_INDEX_TYPE_UINT32); + }); + return; + } ReserveQuadArrayLUT(first + count, true); // The LUT has the indices 0, 1, 2, and 3 copied as an array @@ -195,6 +262,14 @@ void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, VkBuffer buffer, // Already logged in the rasterizer return; } + if (buffer == VK_NULL_HANDLE) { + // Vulkan doesn't support null transform feedback buffers. + // Replace it with our own null buffer. + ReserveNullBuffer(); + buffer = *null_buffer; + offset = 0; + size = 0; + } scheduler.Record([index, buffer, offset, size](vk::CommandBuffer cmdbuf) { const VkDeviceSize vk_offset = offset; const VkDeviceSize vk_size = size; @@ -279,11 +354,11 @@ void BufferCacheRuntime::ReserveQuadArrayLUT(u32 num_indices, bool wait_for_idle }); } -void BufferCacheRuntime::ReserveNullIndexBuffer() { - if (null_index_buffer) { +void BufferCacheRuntime::ReserveNullBuffer() { + if (null_buffer) { return; } - null_index_buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{ + null_buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, .pNext = nullptr, .flags = 0, @@ -294,12 +369,12 @@ void BufferCacheRuntime::ReserveNullIndexBuffer() { .pQueueFamilyIndices = nullptr, }); if (device.HasDebuggingToolAttached()) { - null_index_buffer.SetObjectNameEXT("Null index buffer"); + null_buffer.SetObjectNameEXT("Null index buffer"); } - null_index_buffer_commit = memory_allocator.Commit(null_index_buffer, MemoryUsage::DeviceLocal); + null_buffer_commit = memory_allocator.Commit(null_buffer, MemoryUsage::DeviceLocal); scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([buffer = *null_index_buffer](vk::CommandBuffer cmdbuf) { + scheduler.Record([buffer = *null_buffer](vk::CommandBuffer cmdbuf) { cmdbuf.FillBuffer(buffer, 0, VK_WHOLE_SIZE, 0); }); } diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index 982e92191..c27402ff0 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -9,13 +9,14 @@ #include "video_core/renderer_vulkan/vk_compute_pass.h" #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" #include "video_core/renderer_vulkan/vk_update_descriptor.h" +#include "video_core/surface.h" #include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { class Device; -class VKDescriptorPool; +class DescriptorPool; class VKScheduler; class BufferCacheRuntime; @@ -26,6 +27,8 @@ public: explicit Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_, VAddr cpu_addr_, u64 size_bytes_); + [[nodiscard]] VkBufferView View(u32 offset, u32 size, VideoCore::Surface::PixelFormat format); + [[nodiscard]] VkBuffer Handle() const noexcept { return *buffer; } @@ -35,8 +38,17 @@ public: } private: + struct BufferView { + u32 offset; + u32 size; + VideoCore::Surface::PixelFormat format; + vk::BufferView handle; + }; + + const Device* device{}; vk::Buffer buffer; MemoryCommit commit; + std::vector<BufferView> views; }; class BufferCacheRuntime { @@ -49,7 +61,7 @@ public: explicit BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_manager_, VKScheduler& scheduler_, StagingBufferPool& staging_pool_, VKUpdateDescriptorQueue& update_descriptor_queue_, - VKDescriptorPool& descriptor_pool); + DescriptorPool& descriptor_pool); void Finish(); @@ -60,6 +72,8 @@ public: void CopyBuffer(VkBuffer src_buffer, VkBuffer dst_buffer, std::span<const VideoCommon::BufferCopy> copies); + void ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t size, u32 value); + void BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format, u32 num_indices, u32 base_vertex, VkBuffer buffer, u32 offset, u32 size); @@ -85,6 +99,11 @@ public: BindBuffer(buffer, offset, size); } + void BindTextureBuffer(Buffer& buffer, u32 offset, u32 size, + VideoCore::Surface::PixelFormat format) { + update_descriptor_queue.AddTexelBuffer(buffer.View(offset, size, format)); + } + private: void BindBuffer(VkBuffer buffer, u32 offset, u32 size) { update_descriptor_queue.AddBuffer(buffer, offset, size); @@ -92,7 +111,7 @@ private: void ReserveQuadArrayLUT(u32 num_indices, bool wait_for_idle); - void ReserveNullIndexBuffer(); + void ReserveNullBuffer(); const Device& device; MemoryAllocator& memory_allocator; @@ -105,8 +124,8 @@ private: VkIndexType quad_array_lut_index_type{}; u32 current_num_indices = 0; - vk::Buffer null_index_buffer; - MemoryCommit null_index_buffer_commit; + vk::Buffer null_buffer; + MemoryCommit null_buffer_commit; Uint8Pass uint8_pass; QuadIndexedPass quad_index_pass; @@ -122,6 +141,7 @@ struct BufferCacheParams { static constexpr bool NEEDS_BIND_UNIFORM_INDEX = false; static constexpr bool NEEDS_BIND_STORAGE_INDEX = false; static constexpr bool USE_MEMORY_MAPS = true; + static constexpr bool SEPARATE_IMAGE_BUFFER_BINDINGS = false; }; using BufferCache = VideoCommon::BufferCache<BufferCacheParams>; diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 205cd3b05..8e426ce2c 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -41,80 +41,92 @@ constexpr u32 ASTC_BINDING_SWIZZLE_BUFFER = 2; constexpr u32 ASTC_BINDING_OUTPUT_IMAGE = 3; constexpr size_t ASTC_NUM_BINDINGS = 4; -VkPushConstantRange BuildComputePushConstantRange(std::size_t size) { - return { - .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, - .offset = 0, - .size = static_cast<u32>(size), - }; -} - -std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBindings() { - return {{ - { - .binding = 0, - .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, - .descriptorCount = 1, - .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, - .pImmutableSamplers = nullptr, - }, - { - .binding = 1, - .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, - .descriptorCount = 1, - .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, - .pImmutableSamplers = nullptr, - }, - }}; -} +template <size_t size> +inline constexpr VkPushConstantRange COMPUTE_PUSH_CONSTANT_RANGE{ + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .offset = 0, + .size = static_cast<u32>(size), +}; -std::array<VkDescriptorSetLayoutBinding, ASTC_NUM_BINDINGS> BuildASTCDescriptorSetBindings() { - return {{ - { - .binding = ASTC_BINDING_INPUT_BUFFER, - .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, - .descriptorCount = 1, - .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, - .pImmutableSamplers = nullptr, - }, - { - .binding = ASTC_BINDING_ENC_BUFFER, - .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, - .descriptorCount = 1, - .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, - .pImmutableSamplers = nullptr, - }, - { - .binding = ASTC_BINDING_SWIZZLE_BUFFER, - .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, - .descriptorCount = 1, - .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, - .pImmutableSamplers = nullptr, - }, - { - .binding = ASTC_BINDING_OUTPUT_IMAGE, - .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, - .descriptorCount = 1, - .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, - .pImmutableSamplers = nullptr, - }, - }}; -} +constexpr std::array<VkDescriptorSetLayoutBinding, 2> INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS{{ + { + .binding = 0, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .pImmutableSamplers = nullptr, + }, + { + .binding = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .pImmutableSamplers = nullptr, + }, +}}; + +constexpr DescriptorBankInfo INPUT_OUTPUT_BANK_INFO{ + .uniform_buffers = 0, + .storage_buffers = 2, + .texture_buffers = 0, + .image_buffers = 0, + .textures = 0, + .images = 0, + .score = 2, +}; -VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() { - return { - .dstBinding = 0, - .dstArrayElement = 0, - .descriptorCount = 2, +constexpr std::array<VkDescriptorSetLayoutBinding, 4> ASTC_DESCRIPTOR_SET_BINDINGS{{ + { + .binding = ASTC_BINDING_INPUT_BUFFER, .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, - .offset = 0, - .stride = sizeof(DescriptorUpdateEntry), - }; -} + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .pImmutableSamplers = nullptr, + }, + { + .binding = ASTC_BINDING_ENC_BUFFER, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .pImmutableSamplers = nullptr, + }, + { + .binding = ASTC_BINDING_SWIZZLE_BUFFER, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .pImmutableSamplers = nullptr, + }, + { + .binding = ASTC_BINDING_OUTPUT_IMAGE, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .pImmutableSamplers = nullptr, + }, +}}; + +constexpr DescriptorBankInfo ASTC_BANK_INFO{ + .uniform_buffers = 0, + .storage_buffers = 3, + .texture_buffers = 0, + .image_buffers = 0, + .textures = 0, + .images = 1, + .score = 4, +}; -std::array<VkDescriptorUpdateTemplateEntryKHR, ASTC_NUM_BINDINGS> -BuildASTCPassDescriptorUpdateTemplateEntry() { - return {{ +constexpr VkDescriptorUpdateTemplateEntryKHR INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE{ + .dstBinding = 0, + .dstArrayElement = 0, + .descriptorCount = 2, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, + .offset = 0, + .stride = sizeof(DescriptorUpdateEntry), +}; + +constexpr std::array<VkDescriptorUpdateTemplateEntryKHR, ASTC_NUM_BINDINGS> + ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY{{ { .dstBinding = ASTC_BINDING_INPUT_BUFFER, .dstArrayElement = 0, @@ -148,7 +160,6 @@ BuildASTCPassDescriptorUpdateTemplateEntry() { .stride = sizeof(DescriptorUpdateEntry), }, }}; -} struct AstcPushConstants { std::array<u32, 2> blocks_dims; @@ -159,14 +170,14 @@ struct AstcPushConstants { u32 block_height; u32 block_height_mask; }; - } // Anonymous namespace -VKComputePass::VKComputePass(const Device& device, VKDescriptorPool& descriptor_pool, - vk::Span<VkDescriptorSetLayoutBinding> bindings, - vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates, - vk::Span<VkPushConstantRange> push_constants, - std::span<const u32> code) { +ComputePass::ComputePass(const Device& device_, DescriptorPool& descriptor_pool, + vk::Span<VkDescriptorSetLayoutBinding> bindings, + vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates, + const DescriptorBankInfo& bank_info, + vk::Span<VkPushConstantRange> push_constants, std::span<const u32> code) + : device{device_} { descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout({ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, .pNext = nullptr, @@ -196,8 +207,7 @@ VKComputePass::VKComputePass(const Device& device, VKDescriptorPool& descriptor_ .pipelineLayout = *layout, .set = 0, }); - - descriptor_allocator.emplace(descriptor_pool, *descriptor_set_layout); + descriptor_allocator = descriptor_pool.Allocator(*descriptor_set_layout, bank_info); } module = device.GetLogical().CreateShaderModule({ .sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO, @@ -206,43 +216,34 @@ VKComputePass::VKComputePass(const Device& device, VKDescriptorPool& descriptor_ .codeSize = static_cast<u32>(code.size_bytes()), .pCode = code.data(), }); + device.SaveShader(code); pipeline = device.GetLogical().CreateComputePipeline({ .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO, .pNext = nullptr, .flags = 0, - .stage = - { - .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .stage = VK_SHADER_STAGE_COMPUTE_BIT, - .module = *module, - .pName = "main", - .pSpecializationInfo = nullptr, - }, + .stage{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_COMPUTE_BIT, + .module = *module, + .pName = "main", + .pSpecializationInfo = nullptr, + }, .layout = *layout, .basePipelineHandle = nullptr, .basePipelineIndex = 0, }); } -VKComputePass::~VKComputePass() = default; - -VkDescriptorSet VKComputePass::CommitDescriptorSet( - VKUpdateDescriptorQueue& update_descriptor_queue) { - if (!descriptor_template) { - return nullptr; - } - const VkDescriptorSet set = descriptor_allocator->Commit(); - update_descriptor_queue.Send(*descriptor_template, set); - return set; -} +ComputePass::~ComputePass() = default; -Uint8Pass::Uint8Pass(const Device& device, VKScheduler& scheduler_, - VKDescriptorPool& descriptor_pool, StagingBufferPool& staging_buffer_pool_, +Uint8Pass::Uint8Pass(const Device& device_, VKScheduler& scheduler_, + DescriptorPool& descriptor_pool, StagingBufferPool& staging_buffer_pool_, VKUpdateDescriptorQueue& update_descriptor_queue_) - : VKComputePass(device, descriptor_pool, BuildInputOutputDescriptorSetBindings(), - BuildInputOutputDescriptorUpdateTemplate(), {}, VULKAN_UINT8_COMP_SPV), + : ComputePass(device_, descriptor_pool, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS, + INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, {}, + VULKAN_UINT8_COMP_SPV), scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, update_descriptor_queue{update_descriptor_queue_} {} @@ -256,11 +257,11 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer update_descriptor_queue.Acquire(); update_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices); update_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size); - const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue); + const void* const descriptor_data{update_descriptor_queue.UpdateData()}; + const VkBuffer buffer{staging.buffer}; scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging.buffer, set, - num_vertices](vk::CommandBuffer cmdbuf) { + scheduler.Record([this, buffer, descriptor_data, num_vertices](vk::CommandBuffer cmdbuf) { static constexpr u32 DISPATCH_SIZE = 1024; static constexpr VkMemoryBarrier WRITE_BARRIER{ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, @@ -268,8 +269,10 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, .dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT, }; - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline); - cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {}); + const VkDescriptorSet set = descriptor_allocator.Commit(); + device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); cmdbuf.Dispatch(Common::DivCeil(num_vertices, DISPATCH_SIZE), 1, 1); cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, WRITE_BARRIER); @@ -278,12 +281,12 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer } QuadIndexedPass::QuadIndexedPass(const Device& device_, VKScheduler& scheduler_, - VKDescriptorPool& descriptor_pool_, + DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, VKUpdateDescriptorQueue& update_descriptor_queue_) - : VKComputePass(device_, descriptor_pool_, BuildInputOutputDescriptorSetBindings(), - BuildInputOutputDescriptorUpdateTemplate(), - BuildComputePushConstantRange(sizeof(u32) * 2), VULKAN_QUAD_INDEXED_COMP_SPV), + : ComputePass(device_, descriptor_pool_, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS, + INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, + COMPUTE_PUSH_CONSTANT_RANGE<sizeof(u32) * 2>, VULKAN_QUAD_INDEXED_COMP_SPV), scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, update_descriptor_queue{update_descriptor_queue_} {} @@ -313,11 +316,11 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble( update_descriptor_queue.Acquire(); update_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size); update_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size); - const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue); + const void* const descriptor_data{update_descriptor_queue.UpdateData()}; scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging.buffer, set, - num_tri_vertices, base_vertex, index_shift](vk::CommandBuffer cmdbuf) { + scheduler.Record([this, buffer = staging.buffer, descriptor_data, num_tri_vertices, base_vertex, + index_shift](vk::CommandBuffer cmdbuf) { static constexpr u32 DISPATCH_SIZE = 1024; static constexpr VkMemoryBarrier WRITE_BARRIER{ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, @@ -325,10 +328,12 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble( .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, .dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT, }; - const std::array push_constants = {base_vertex, index_shift}; - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline); - cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {}); - cmdbuf.PushConstants(layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(push_constants), + const std::array push_constants{base_vertex, index_shift}; + const VkDescriptorSet set = descriptor_allocator.Commit(); + device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); + cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(push_constants), &push_constants); cmdbuf.Dispatch(Common::DivCeil(num_tri_vertices, DISPATCH_SIZE), 1, 1); cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, @@ -338,15 +343,14 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble( } ASTCDecoderPass::ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_, - VKDescriptorPool& descriptor_pool_, + DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, VKUpdateDescriptorQueue& update_descriptor_queue_, MemoryAllocator& memory_allocator_) - : VKComputePass(device_, descriptor_pool_, BuildASTCDescriptorSetBindings(), - BuildASTCPassDescriptorUpdateTemplateEntry(), - BuildComputePushConstantRange(sizeof(AstcPushConstants)), - ASTC_DECODER_COMP_SPV), - device{device_}, scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, + : ComputePass(device_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, + ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, + COMPUTE_PUSH_CONSTANT_RANGE<sizeof(AstcPushConstants)>, ASTC_DECODER_COMP_SPV), + scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, update_descriptor_queue{update_descriptor_queue_}, memory_allocator{memory_allocator_} {} ASTCDecoderPass::~ASTCDecoderPass() = default; @@ -374,20 +378,20 @@ void ASTCDecoderPass::MakeDataBuffer() { scheduler.Record([src = staging_ref.buffer, offset = staging_ref.offset, dst = *data_buffer, TOTAL_BUFFER_SIZE](vk::CommandBuffer cmdbuf) { - cmdbuf.CopyBuffer(src, dst, - VkBufferCopy{ - .srcOffset = offset, - .dstOffset = 0, - .size = TOTAL_BUFFER_SIZE, - }); - cmdbuf.PipelineBarrier( - VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, - VkMemoryBarrier{ - .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = 0, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, - }); + static constexpr VkMemoryBarrier write_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_SHADER_READ_BIT, + }; + const VkBufferCopy copy{ + .srcOffset = offset, + .dstOffset = 0, + .size = TOTAL_BUFFER_SIZE, + }; + cmdbuf.CopyBuffer(src, dst, copy); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + 0, write_barrier); }); } @@ -411,7 +415,7 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, const VkImageMemoryBarrier image_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .srcAccessMask = is_initialized ? VK_ACCESS_SHADER_WRITE_BIT : VkAccessFlags{}, .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED, .newLayout = VK_IMAGE_LAYOUT_GENERAL, @@ -426,7 +430,8 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, .layerCount = VK_REMAINING_ARRAY_LAYERS, }, }; - cmdbuf.PipelineBarrier(is_initialized ? VK_PIPELINE_STAGE_ALL_COMMANDS_BIT : 0, + cmdbuf.PipelineBarrier(is_initialized ? VK_PIPELINE_STAGE_ALL_COMMANDS_BIT + : VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier); cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline); }); @@ -443,16 +448,14 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, update_descriptor_queue.AddBuffer(*data_buffer, sizeof(ASTC_ENCODINGS_VALUES), sizeof(SWIZZLE_TABLE)); update_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); - - const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue); - const VkPipelineLayout vk_layout = *layout; + const void* const descriptor_data{update_descriptor_queue.UpdateData()}; // To unswizzle the ASTC data const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info); ASSERT(params.origin == (std::array<u32, 3>{0, 0, 0})); ASSERT(params.destination == (std::array<s32, 3>{0, 0, 0})); - scheduler.Record([vk_layout, num_dispatches_x, num_dispatches_y, num_dispatches_z, - block_dims, params, set](vk::CommandBuffer cmdbuf) { + scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, block_dims, + params, descriptor_data](vk::CommandBuffer cmdbuf) { const AstcPushConstants uniforms{ .blocks_dims = block_dims, .bytes_per_block_log2 = params.bytes_per_block_log2, @@ -462,8 +465,10 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, .block_height = params.block_height, .block_height_mask = params.block_height_mask, }; - cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, vk_layout, 0, set, {}); - cmdbuf.PushConstants(vk_layout, VK_SHADER_STAGE_COMPUTE_BIT, uniforms); + const VkDescriptorSet set = descriptor_allocator.Commit(); + device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); + cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, uniforms); cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z); }); } diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h index 5ea187c30..114aef2bd 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.h +++ b/src/video_core/renderer_vulkan/vk_compute_pass.h @@ -4,7 +4,6 @@ #pragma once -#include <optional> #include <span> #include <utility> @@ -27,31 +26,31 @@ class VKUpdateDescriptorQueue; class Image; struct StagingBufferRef; -class VKComputePass { +class ComputePass { public: - explicit VKComputePass(const Device& device, VKDescriptorPool& descriptor_pool, - vk::Span<VkDescriptorSetLayoutBinding> bindings, - vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates, - vk::Span<VkPushConstantRange> push_constants, std::span<const u32> code); - ~VKComputePass(); + explicit ComputePass(const Device& device, DescriptorPool& descriptor_pool, + vk::Span<VkDescriptorSetLayoutBinding> bindings, + vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates, + const DescriptorBankInfo& bank_info, + vk::Span<VkPushConstantRange> push_constants, std::span<const u32> code); + ~ComputePass(); protected: - VkDescriptorSet CommitDescriptorSet(VKUpdateDescriptorQueue& update_descriptor_queue); - + const Device& device; vk::DescriptorUpdateTemplateKHR descriptor_template; vk::PipelineLayout layout; vk::Pipeline pipeline; + vk::DescriptorSetLayout descriptor_set_layout; + DescriptorAllocator descriptor_allocator; private: - vk::DescriptorSetLayout descriptor_set_layout; - std::optional<DescriptorAllocator> descriptor_allocator; vk::ShaderModule module; }; -class Uint8Pass final : public VKComputePass { +class Uint8Pass final : public ComputePass { public: explicit Uint8Pass(const Device& device_, VKScheduler& scheduler_, - VKDescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, + DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, VKUpdateDescriptorQueue& update_descriptor_queue_); ~Uint8Pass(); @@ -66,10 +65,10 @@ private: VKUpdateDescriptorQueue& update_descriptor_queue; }; -class QuadIndexedPass final : public VKComputePass { +class QuadIndexedPass final : public ComputePass { public: explicit QuadIndexedPass(const Device& device_, VKScheduler& scheduler_, - VKDescriptorPool& descriptor_pool_, + DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, VKUpdateDescriptorQueue& update_descriptor_queue_); ~QuadIndexedPass(); @@ -84,10 +83,10 @@ private: VKUpdateDescriptorQueue& update_descriptor_queue; }; -class ASTCDecoderPass final : public VKComputePass { +class ASTCDecoderPass final : public ComputePass { public: explicit ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_, - VKDescriptorPool& descriptor_pool_, + DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, VKUpdateDescriptorQueue& update_descriptor_queue_, MemoryAllocator& memory_allocator_); @@ -99,7 +98,6 @@ public: private: void MakeDataBuffer(); - const Device& device; VKScheduler& scheduler; StagingBufferPool& staging_buffer_pool; VKUpdateDescriptorQueue& update_descriptor_queue; diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp index 3a48219b7..70b84c7a6 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp @@ -2,152 +2,198 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <algorithm> #include <vector> +#include <boost/container/small_vector.hpp> + +#include "video_core/renderer_vulkan/pipeline_helper.h" +#include "video_core/renderer_vulkan/vk_buffer_cache.h" #include "video_core/renderer_vulkan/vk_compute_pipeline.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/renderer_vulkan/vk_pipeline_cache.h" #include "video_core/renderer_vulkan/vk_scheduler.h" -#include "video_core/renderer_vulkan/vk_shader_decompiler.h" #include "video_core/renderer_vulkan/vk_update_descriptor.h" +#include "video_core/shader_notify.h" #include "video_core/vulkan_common/vulkan_device.h" #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { -VKComputePipeline::VKComputePipeline(const Device& device_, VKScheduler& scheduler_, - VKDescriptorPool& descriptor_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_, - const SPIRVShader& shader_) - : device{device_}, scheduler{scheduler_}, entries{shader_.entries}, - descriptor_set_layout{CreateDescriptorSetLayout()}, - descriptor_allocator{descriptor_pool_, *descriptor_set_layout}, - update_descriptor_queue{update_descriptor_queue_}, layout{CreatePipelineLayout()}, - descriptor_template{CreateDescriptorUpdateTemplate()}, - shader_module{CreateShaderModule(shader_.code)}, pipeline{CreatePipeline()} {} - -VKComputePipeline::~VKComputePipeline() = default; - -VkDescriptorSet VKComputePipeline::CommitDescriptorSet() { - if (!descriptor_template) { - return {}; - } - const VkDescriptorSet set = descriptor_allocator.Commit(); - update_descriptor_queue.Send(*descriptor_template, set); - return set; -} - -vk::DescriptorSetLayout VKComputePipeline::CreateDescriptorSetLayout() const { - std::vector<VkDescriptorSetLayoutBinding> bindings; - u32 binding = 0; - const auto add_bindings = [&](VkDescriptorType descriptor_type, std::size_t num_entries) { - // TODO(Rodrigo): Maybe make individual bindings here? - for (u32 bindpoint = 0; bindpoint < static_cast<u32>(num_entries); ++bindpoint) { - bindings.push_back({ - .binding = binding++, - .descriptorType = descriptor_type, - .descriptorCount = 1, - .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, - .pImmutableSamplers = nullptr, - }); - } - }; - add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, entries.const_buffers.size()); - add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, entries.global_buffers.size()); - add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, entries.uniform_texels.size()); - add_bindings(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, entries.samplers.size()); - add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, entries.storage_texels.size()); - add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, entries.images.size()); - - return device.GetLogical().CreateDescriptorSetLayout({ - .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .bindingCount = static_cast<u32>(bindings.size()), - .pBindings = bindings.data(), - }); -} - -vk::PipelineLayout VKComputePipeline::CreatePipelineLayout() const { - return device.GetLogical().CreatePipelineLayout({ - .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .setLayoutCount = 1, - .pSetLayouts = descriptor_set_layout.address(), - .pushConstantRangeCount = 0, - .pPushConstantRanges = nullptr, - }); -} - -vk::DescriptorUpdateTemplateKHR VKComputePipeline::CreateDescriptorUpdateTemplate() const { - std::vector<VkDescriptorUpdateTemplateEntryKHR> template_entries; - u32 binding = 0; - u32 offset = 0; - FillDescriptorUpdateTemplateEntries(entries, binding, offset, template_entries); - if (template_entries.empty()) { - // If the shader doesn't use descriptor sets, skip template creation. - return {}; +using Shader::ImageBufferDescriptor; +using Tegra::Texture::TexturePair; + +ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descriptor_pool, + VKUpdateDescriptorQueue& update_descriptor_queue_, + Common::ThreadWorker* thread_worker, + VideoCore::ShaderNotify* shader_notify, const Shader::Info& info_, + vk::ShaderModule spv_module_) + : device{device_}, update_descriptor_queue{update_descriptor_queue_}, info{info_}, + spv_module(std::move(spv_module_)) { + if (shader_notify) { + shader_notify->MarkShaderBuilding(); } - - return device.GetLogical().CreateDescriptorUpdateTemplateKHR({ - .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_UPDATE_TEMPLATE_CREATE_INFO_KHR, - .pNext = nullptr, - .flags = 0, - .descriptorUpdateEntryCount = static_cast<u32>(template_entries.size()), - .pDescriptorUpdateEntries = template_entries.data(), - .templateType = VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR, - .descriptorSetLayout = *descriptor_set_layout, - .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS, - .pipelineLayout = *layout, - .set = DESCRIPTOR_SET, - }); -} - -vk::ShaderModule VKComputePipeline::CreateShaderModule(const std::vector<u32>& code) const { - device.SaveShader(code); - - return device.GetLogical().CreateShaderModule({ - .sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .codeSize = code.size() * sizeof(u32), - .pCode = code.data(), - }); -} - -vk::Pipeline VKComputePipeline::CreatePipeline() const { - - VkComputePipelineCreateInfo ci{ - .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .stage = - { + std::copy_n(info.constant_buffer_used_sizes.begin(), uniform_buffer_sizes.size(), + uniform_buffer_sizes.begin()); + + auto func{[this, &descriptor_pool, shader_notify] { + DescriptorLayoutBuilder builder{device}; + builder.Add(info, VK_SHADER_STAGE_COMPUTE_BIT); + + descriptor_set_layout = builder.CreateDescriptorSetLayout(false); + pipeline_layout = builder.CreatePipelineLayout(*descriptor_set_layout); + descriptor_update_template = + builder.CreateTemplate(*descriptor_set_layout, *pipeline_layout, false); + descriptor_allocator = descriptor_pool.Allocator(*descriptor_set_layout, info); + const VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_REQUIRED_SUBGROUP_SIZE_CREATE_INFO_EXT, + .pNext = nullptr, + .requiredSubgroupSize = GuestWarpSize, + }; + pipeline = device.GetLogical().CreateComputePipeline({ + .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage{ .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, - .pNext = nullptr, + .pNext = device.IsExtSubgroupSizeControlSupported() ? &subgroup_size_ci : nullptr, .flags = 0, .stage = VK_SHADER_STAGE_COMPUTE_BIT, - .module = *shader_module, + .module = *spv_module, .pName = "main", .pSpecializationInfo = nullptr, }, - .layout = *layout, - .basePipelineHandle = nullptr, - .basePipelineIndex = 0, - }; - - const VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci{ - .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_REQUIRED_SUBGROUP_SIZE_CREATE_INFO_EXT, - .pNext = nullptr, - .requiredSubgroupSize = GuestWarpSize, - }; - - if (entries.uses_warps && device.IsGuestWarpSizeSupported(VK_SHADER_STAGE_COMPUTE_BIT)) { - ci.stage.pNext = &subgroup_size_ci; + .layout = *pipeline_layout, + .basePipelineHandle = 0, + .basePipelineIndex = 0, + }); + std::lock_guard lock{build_mutex}; + is_built = true; + build_condvar.notify_one(); + if (shader_notify) { + shader_notify->MarkShaderComplete(); + } + }}; + if (thread_worker) { + thread_worker->QueueWork(std::move(func)); + } else { + func(); + } +} + +void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, + Tegra::MemoryManager& gpu_memory, VKScheduler& scheduler, + BufferCache& buffer_cache, TextureCache& texture_cache) { + update_descriptor_queue.Acquire(); + + buffer_cache.SetComputeUniformBufferState(info.constant_buffer_mask, &uniform_buffer_sizes); + buffer_cache.UnbindComputeStorageBuffers(); + size_t ssbo_index{}; + for (const auto& desc : info.storage_buffers_descriptors) { + ASSERT(desc.count == 1); + buffer_cache.BindComputeStorageBuffer(ssbo_index, desc.cbuf_index, desc.cbuf_offset, + desc.is_written); + ++ssbo_index; } - return device.GetLogical().CreateComputePipeline(ci); + texture_cache.SynchronizeComputeDescriptors(); + + static constexpr size_t max_elements = 64; + std::array<ImageId, max_elements> image_view_ids; + boost::container::static_vector<u32, max_elements> image_view_indices; + boost::container::static_vector<VkSampler, max_elements> samplers; + + const auto& qmd{kepler_compute.launch_description}; + const auto& cbufs{qmd.const_buffer_config}; + const bool via_header_index{qmd.linked_tsc != 0}; + const auto read_handle{[&](const auto& desc, u32 index) { + ASSERT(((qmd.const_buffer_enable_mask >> desc.cbuf_index) & 1) != 0); + const u32 index_offset{index << desc.size_shift}; + const u32 offset{desc.cbuf_offset + index_offset}; + const GPUVAddr addr{cbufs[desc.cbuf_index].Address() + offset}; + if constexpr (std::is_same_v<decltype(desc), const Shader::TextureDescriptor&> || + std::is_same_v<decltype(desc), const Shader::TextureBufferDescriptor&>) { + if (desc.has_secondary) { + ASSERT(((qmd.const_buffer_enable_mask >> desc.secondary_cbuf_index) & 1) != 0); + const u32 secondary_offset{desc.secondary_cbuf_offset + index_offset}; + const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].Address() + + secondary_offset}; + const u32 lhs_raw{gpu_memory.Read<u32>(addr)}; + const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)}; + return TexturePair(lhs_raw | rhs_raw, via_header_index); + } + } + return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); + }}; + const auto add_image{[&](const auto& desc) { + for (u32 index = 0; index < desc.count; ++index) { + const auto handle{read_handle(desc, index)}; + image_view_indices.push_back(handle.first); + } + }}; + std::ranges::for_each(info.texture_buffer_descriptors, add_image); + std::ranges::for_each(info.image_buffer_descriptors, add_image); + for (const auto& desc : info.texture_descriptors) { + for (u32 index = 0; index < desc.count; ++index) { + const auto handle{read_handle(desc, index)}; + image_view_indices.push_back(handle.first); + + Sampler* const sampler = texture_cache.GetComputeSampler(handle.second); + samplers.push_back(sampler->Handle()); + } + } + std::ranges::for_each(info.image_descriptors, add_image); + + const std::span indices_span(image_view_indices.data(), image_view_indices.size()); + texture_cache.FillComputeImageViews(indices_span, image_view_ids); + + buffer_cache.UnbindComputeTextureBuffers(); + ImageId* texture_buffer_ids{image_view_ids.data()}; + size_t index{}; + const auto add_buffer{[&](const auto& desc) { + constexpr bool is_image = std::is_same_v<decltype(desc), const ImageBufferDescriptor&>; + for (u32 i = 0; i < desc.count; ++i) { + bool is_written{false}; + if constexpr (is_image) { + is_written = desc.is_written; + } + ImageView& image_view = texture_cache.GetImageView(*texture_buffer_ids); + buffer_cache.BindComputeTextureBuffer(index, image_view.GpuAddr(), + image_view.BufferSize(), image_view.format, + is_written, is_image); + ++texture_buffer_ids; + ++index; + } + }}; + std::ranges::for_each(info.texture_buffer_descriptors, add_buffer); + std::ranges::for_each(info.image_buffer_descriptors, add_buffer); + + buffer_cache.UpdateComputeBuffers(); + buffer_cache.BindHostComputeBuffers(); + + const VkSampler* samplers_it{samplers.data()}; + const ImageId* views_it{image_view_ids.data()}; + PushImageDescriptors(info, samplers_it, views_it, texture_cache, update_descriptor_queue); + + if (!is_built.load(std::memory_order::relaxed)) { + // Wait for the pipeline to be built + scheduler.Record([this](vk::CommandBuffer) { + std::unique_lock lock{build_mutex}; + build_condvar.wait(lock, [this] { return is_built.load(std::memory_order::relaxed); }); + }); + } + const void* const descriptor_data{update_descriptor_queue.UpdateData()}; + scheduler.Record([this, descriptor_data](vk::CommandBuffer cmdbuf) { + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); + if (!descriptor_set_layout) { + return; + } + const VkDescriptorSet descriptor_set{descriptor_allocator.Commit()}; + const vk::Device& dev{device.GetLogical()}; + dev.UpdateDescriptorSet(descriptor_set, *descriptor_update_template, descriptor_data); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline_layout, 0, + descriptor_set, nullptr); + }); } } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.h b/src/video_core/renderer_vulkan/vk_compute_pipeline.h index 7e16575ac..52fec04d3 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.h +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.h @@ -4,61 +4,63 @@ #pragma once +#include <atomic> +#include <condition_variable> +#include <mutex> + #include "common/common_types.h" +#include "common/thread_worker.h" +#include "shader_recompiler/shader_info.h" +#include "video_core/memory_manager.h" +#include "video_core/renderer_vulkan/vk_buffer_cache.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" -#include "video_core/renderer_vulkan/vk_shader_decompiler.h" +#include "video_core/renderer_vulkan/vk_texture_cache.h" +#include "video_core/renderer_vulkan/vk_update_descriptor.h" #include "video_core/vulkan_common/vulkan_wrapper.h" +namespace VideoCore { +class ShaderNotify; +} + namespace Vulkan { class Device; class VKScheduler; -class VKUpdateDescriptorQueue; -class VKComputePipeline final { +class ComputePipeline { public: - explicit VKComputePipeline(const Device& device_, VKScheduler& scheduler_, - VKDescriptorPool& descriptor_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_, - const SPIRVShader& shader_); - ~VKComputePipeline(); - - VkDescriptorSet CommitDescriptorSet(); + explicit ComputePipeline(const Device& device, DescriptorPool& descriptor_pool, + VKUpdateDescriptorQueue& update_descriptor_queue, + Common::ThreadWorker* thread_worker, + VideoCore::ShaderNotify* shader_notify, const Shader::Info& info, + vk::ShaderModule spv_module); - VkPipeline GetHandle() const { - return *pipeline; - } + ComputePipeline& operator=(ComputePipeline&&) noexcept = delete; + ComputePipeline(ComputePipeline&&) noexcept = delete; - VkPipelineLayout GetLayout() const { - return *layout; - } + ComputePipeline& operator=(const ComputePipeline&) = delete; + ComputePipeline(const ComputePipeline&) = delete; - const ShaderEntries& GetEntries() const { - return entries; - } + void Configure(Tegra::Engines::KeplerCompute& kepler_compute, Tegra::MemoryManager& gpu_memory, + VKScheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache); private: - vk::DescriptorSetLayout CreateDescriptorSetLayout() const; - - vk::PipelineLayout CreatePipelineLayout() const; - - vk::DescriptorUpdateTemplateKHR CreateDescriptorUpdateTemplate() const; - - vk::ShaderModule CreateShaderModule(const std::vector<u32>& code) const; - - vk::Pipeline CreatePipeline() const; - const Device& device; - VKScheduler& scheduler; - ShaderEntries entries; + VKUpdateDescriptorQueue& update_descriptor_queue; + Shader::Info info; + VideoCommon::ComputeUniformBufferSizes uniform_buffer_sizes{}; + + vk::ShaderModule spv_module; vk::DescriptorSetLayout descriptor_set_layout; DescriptorAllocator descriptor_allocator; - VKUpdateDescriptorQueue& update_descriptor_queue; - vk::PipelineLayout layout; - vk::DescriptorUpdateTemplateKHR descriptor_template; - vk::ShaderModule shader_module; + vk::PipelineLayout pipeline_layout; + vk::DescriptorUpdateTemplateKHR descriptor_update_template; vk::Pipeline pipeline; + + std::condition_variable build_condvar; + std::mutex build_mutex; + std::atomic_bool is_built{false}; }; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp index ef9fb5910..8e77e4796 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp @@ -2,6 +2,8 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <mutex> +#include <span> #include <vector> #include "common/common_types.h" @@ -13,79 +15,149 @@ namespace Vulkan { -// Prefer small grow rates to avoid saturating the descriptor pool with barely used pipelines. -constexpr std::size_t SETS_GROW_RATE = 0x20; +// Prefer small grow rates to avoid saturating the descriptor pool with barely used pipelines +constexpr size_t SETS_GROW_RATE = 16; +constexpr s32 SCORE_THRESHOLD = 3; +constexpr u32 SETS_PER_POOL = 64; -DescriptorAllocator::DescriptorAllocator(VKDescriptorPool& descriptor_pool_, - VkDescriptorSetLayout layout_) - : ResourcePool(descriptor_pool_.master_semaphore, SETS_GROW_RATE), - descriptor_pool{descriptor_pool_}, layout{layout_} {} +struct DescriptorBank { + DescriptorBankInfo info; + std::vector<vk::DescriptorPool> pools; +}; -DescriptorAllocator::~DescriptorAllocator() = default; +bool DescriptorBankInfo::IsSuperset(const DescriptorBankInfo& subset) const noexcept { + return uniform_buffers >= subset.uniform_buffers && storage_buffers >= subset.storage_buffers && + texture_buffers >= subset.texture_buffers && image_buffers >= subset.image_buffers && + textures >= subset.textures && images >= subset.image_buffers; +} -VkDescriptorSet DescriptorAllocator::Commit() { - const std::size_t index = CommitResource(); - return descriptors_allocations[index / SETS_GROW_RATE][index % SETS_GROW_RATE]; +template <typename Descriptors> +static u32 Accumulate(const Descriptors& descriptors) { + u32 count = 0; + for (const auto& descriptor : descriptors) { + count += descriptor.count; + } + return count; } -void DescriptorAllocator::Allocate(std::size_t begin, std::size_t end) { - descriptors_allocations.push_back(descriptor_pool.AllocateDescriptors(layout, end - begin)); +static DescriptorBankInfo MakeBankInfo(std::span<const Shader::Info> infos) { + DescriptorBankInfo bank; + for (const Shader::Info& info : infos) { + bank.uniform_buffers += Accumulate(info.constant_buffer_descriptors); + bank.storage_buffers += Accumulate(info.storage_buffers_descriptors); + bank.texture_buffers += Accumulate(info.texture_buffer_descriptors); + bank.image_buffers += Accumulate(info.image_buffer_descriptors); + bank.textures += Accumulate(info.texture_descriptors); + bank.images += Accumulate(info.image_descriptors); + } + bank.score = bank.uniform_buffers + bank.storage_buffers + bank.texture_buffers + + bank.image_buffers + bank.textures + bank.images; + return bank; } -VKDescriptorPool::VKDescriptorPool(const Device& device_, VKScheduler& scheduler) - : device{device_}, master_semaphore{scheduler.GetMasterSemaphore()}, active_pool{ - AllocateNewPool()} {} - -VKDescriptorPool::~VKDescriptorPool() = default; - -vk::DescriptorPool* VKDescriptorPool::AllocateNewPool() { - static constexpr u32 num_sets = 0x20000; - static constexpr VkDescriptorPoolSize pool_sizes[] = { - {VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, num_sets * 90}, - {VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, num_sets * 60}, - {VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, num_sets * 64}, - {VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, num_sets * 64}, - {VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, num_sets * 64}, - {VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, num_sets * 40}, +static void AllocatePool(const Device& device, DescriptorBank& bank) { + std::array<VkDescriptorPoolSize, 6> pool_sizes; + size_t pool_cursor{}; + const auto add = [&](VkDescriptorType type, u32 count) { + if (count > 0) { + pool_sizes[pool_cursor++] = { + .type = type, + .descriptorCount = count * SETS_PER_POOL, + }; + } }; - - const VkDescriptorPoolCreateInfo ci{ + const auto& info{bank.info}; + add(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, info.uniform_buffers); + add(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, info.storage_buffers); + add(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, info.texture_buffers); + add(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, info.image_buffers); + add(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, info.textures); + add(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, info.images); + bank.pools.push_back(device.GetLogical().CreateDescriptorPool({ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO, .pNext = nullptr, .flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, - .maxSets = num_sets, - .poolSizeCount = static_cast<u32>(std::size(pool_sizes)), + .maxSets = SETS_PER_POOL, + .poolSizeCount = static_cast<u32>(pool_cursor), .pPoolSizes = std::data(pool_sizes), - }; - return &pools.emplace_back(device.GetLogical().CreateDescriptorPool(ci)); + })); +} + +DescriptorAllocator::DescriptorAllocator(const Device& device_, MasterSemaphore& master_semaphore_, + DescriptorBank& bank_, VkDescriptorSetLayout layout_) + : ResourcePool(master_semaphore_, SETS_GROW_RATE), device{&device_}, bank{&bank_}, + layout{layout_} {} + +VkDescriptorSet DescriptorAllocator::Commit() { + const size_t index = CommitResource(); + return sets[index / SETS_GROW_RATE][index % SETS_GROW_RATE]; } -vk::DescriptorSets VKDescriptorPool::AllocateDescriptors(VkDescriptorSetLayout layout, - std::size_t count) { - const std::vector layout_copies(count, layout); - VkDescriptorSetAllocateInfo ai{ +void DescriptorAllocator::Allocate(size_t begin, size_t end) { + sets.push_back(AllocateDescriptors(end - begin)); +} + +vk::DescriptorSets DescriptorAllocator::AllocateDescriptors(size_t count) { + const std::vector<VkDescriptorSetLayout> layouts(count, layout); + VkDescriptorSetAllocateInfo allocate_info{ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO, .pNext = nullptr, - .descriptorPool = **active_pool, + .descriptorPool = *bank->pools.back(), .descriptorSetCount = static_cast<u32>(count), - .pSetLayouts = layout_copies.data(), + .pSetLayouts = layouts.data(), }; - - vk::DescriptorSets sets = active_pool->Allocate(ai); - if (!sets.IsOutOfPoolMemory()) { - return sets; + vk::DescriptorSets new_sets = bank->pools.back().Allocate(allocate_info); + if (!new_sets.IsOutOfPoolMemory()) { + return new_sets; } - // Our current pool is out of memory. Allocate a new one and retry - active_pool = AllocateNewPool(); - ai.descriptorPool = **active_pool; - sets = active_pool->Allocate(ai); - if (!sets.IsOutOfPoolMemory()) { - return sets; + AllocatePool(*device, *bank); + allocate_info.descriptorPool = *bank->pools.back(); + new_sets = bank->pools.back().Allocate(allocate_info); + if (!new_sets.IsOutOfPoolMemory()) { + return new_sets; } - // After allocating a new pool, we are out of memory again. We can't handle this from here. throw vk::Exception(VK_ERROR_OUT_OF_POOL_MEMORY); } +DescriptorPool::DescriptorPool(const Device& device_, VKScheduler& scheduler) + : device{device_}, master_semaphore{scheduler.GetMasterSemaphore()} {} + +DescriptorPool::~DescriptorPool() = default; + +DescriptorAllocator DescriptorPool::Allocator(VkDescriptorSetLayout layout, + std::span<const Shader::Info> infos) { + return Allocator(layout, MakeBankInfo(infos)); +} + +DescriptorAllocator DescriptorPool::Allocator(VkDescriptorSetLayout layout, + const Shader::Info& info) { + return Allocator(layout, MakeBankInfo(std::array{info})); +} + +DescriptorAllocator DescriptorPool::Allocator(VkDescriptorSetLayout layout, + const DescriptorBankInfo& info) { + return DescriptorAllocator(device, master_semaphore, Bank(info), layout); +} + +DescriptorBank& DescriptorPool::Bank(const DescriptorBankInfo& reqs) { + std::shared_lock read_lock{banks_mutex}; + const auto it = std::ranges::find_if(bank_infos, [&reqs](const DescriptorBankInfo& bank) { + return std::abs(bank.score - reqs.score) < SCORE_THRESHOLD && bank.IsSuperset(reqs); + }); + if (it != bank_infos.end()) { + return *banks[std::distance(bank_infos.begin(), it)].get(); + } + read_lock.unlock(); + + std::unique_lock write_lock{banks_mutex}; + bank_infos.push_back(reqs); + + auto& bank = *banks.emplace_back(std::make_unique<DescriptorBank>()); + bank.info = reqs; + AllocatePool(device, bank); + return bank; +} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.h b/src/video_core/renderer_vulkan/vk_descriptor_pool.h index f892be7be..59466aac5 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.h +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.h @@ -4,57 +4,85 @@ #pragma once +#include <shared_mutex> +#include <span> #include <vector> +#include "shader_recompiler/shader_info.h" #include "video_core/renderer_vulkan/vk_resource_pool.h" #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { class Device; -class VKDescriptorPool; class VKScheduler; +struct DescriptorBank; + +struct DescriptorBankInfo { + [[nodiscard]] bool IsSuperset(const DescriptorBankInfo& subset) const noexcept; + + u32 uniform_buffers{}; ///< Number of uniform buffer descriptors + u32 storage_buffers{}; ///< Number of storage buffer descriptors + u32 texture_buffers{}; ///< Number of texture buffer descriptors + u32 image_buffers{}; ///< Number of image buffer descriptors + u32 textures{}; ///< Number of texture descriptors + u32 images{}; ///< Number of image descriptors + s32 score{}; ///< Number of descriptors in total +}; + class DescriptorAllocator final : public ResourcePool { + friend class DescriptorPool; + public: - explicit DescriptorAllocator(VKDescriptorPool& descriptor_pool, VkDescriptorSetLayout layout); - ~DescriptorAllocator() override; + explicit DescriptorAllocator() = default; + ~DescriptorAllocator() override = default; + + DescriptorAllocator& operator=(DescriptorAllocator&&) noexcept = default; + DescriptorAllocator(DescriptorAllocator&&) noexcept = default; DescriptorAllocator& operator=(const DescriptorAllocator&) = delete; DescriptorAllocator(const DescriptorAllocator&) = delete; VkDescriptorSet Commit(); -protected: - void Allocate(std::size_t begin, std::size_t end) override; - private: - VKDescriptorPool& descriptor_pool; - const VkDescriptorSetLayout layout; + explicit DescriptorAllocator(const Device& device_, MasterSemaphore& master_semaphore_, + DescriptorBank& bank_, VkDescriptorSetLayout layout_); - std::vector<vk::DescriptorSets> descriptors_allocations; -}; + void Allocate(size_t begin, size_t end) override; + + vk::DescriptorSets AllocateDescriptors(size_t count); + + const Device* device{}; + DescriptorBank* bank{}; + VkDescriptorSetLayout layout{}; -class VKDescriptorPool final { - friend DescriptorAllocator; + std::vector<vk::DescriptorSets> sets; +}; +class DescriptorPool { public: - explicit VKDescriptorPool(const Device& device, VKScheduler& scheduler); - ~VKDescriptorPool(); + explicit DescriptorPool(const Device& device, VKScheduler& scheduler); + ~DescriptorPool(); - VKDescriptorPool(const VKDescriptorPool&) = delete; - VKDescriptorPool& operator=(const VKDescriptorPool&) = delete; + DescriptorPool& operator=(const DescriptorPool&) = delete; + DescriptorPool(const DescriptorPool&) = delete; -private: - vk::DescriptorPool* AllocateNewPool(); + DescriptorAllocator Allocator(VkDescriptorSetLayout layout, + std::span<const Shader::Info> infos); + DescriptorAllocator Allocator(VkDescriptorSetLayout layout, const Shader::Info& info); + DescriptorAllocator Allocator(VkDescriptorSetLayout layout, const DescriptorBankInfo& info); - vk::DescriptorSets AllocateDescriptors(VkDescriptorSetLayout layout, std::size_t count); +private: + DescriptorBank& Bank(const DescriptorBankInfo& reqs); const Device& device; MasterSemaphore& master_semaphore; - std::vector<vk::DescriptorPool> pools; - vk::DescriptorPool* active_pool; + std::shared_mutex banks_mutex; + std::vector<DescriptorBankInfo> bank_infos; + std::vector<std::unique_ptr<DescriptorBank>> banks; }; } // namespace Vulkan
\ No newline at end of file diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index fc6dd83eb..18482e1d0 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -1,29 +1,58 @@ -// Copyright 2019 yuzu Emulator Project +// Copyright 2021 yuzu Emulator Project // Licensed under GPLv2 or any later version // Refer to the license.txt file included. #include <algorithm> -#include <array> -#include <cstring> -#include <vector> +#include <span> -#include "common/common_types.h" -#include "common/microprofile.h" -#include "video_core/renderer_vulkan/fixed_pipeline_state.h" +#include <boost/container/small_vector.hpp> +#include <boost/container/static_vector.hpp> + +#include "common/bit_field.h" #include "video_core/renderer_vulkan/maxwell_to_vk.h" -#include "video_core/renderer_vulkan/vk_descriptor_pool.h" +#include "video_core/renderer_vulkan/pipeline_helper.h" +#include "video_core/renderer_vulkan/vk_buffer_cache.h" #include "video_core/renderer_vulkan/vk_graphics_pipeline.h" -#include "video_core/renderer_vulkan/vk_pipeline_cache.h" +#include "video_core/renderer_vulkan/vk_render_pass_cache.h" #include "video_core/renderer_vulkan/vk_scheduler.h" +#include "video_core/renderer_vulkan/vk_texture_cache.h" #include "video_core/renderer_vulkan/vk_update_descriptor.h" +#include "video_core/shader_notify.h" #include "video_core/vulkan_common/vulkan_device.h" -#include "video_core/vulkan_common/vulkan_wrapper.h" - -namespace Vulkan { -MICROPROFILE_DECLARE(Vulkan_PipelineCache); +#if defined(_MSC_VER) && defined(NDEBUG) +#define LAMBDA_FORCEINLINE [[msvc::forceinline]] +#else +#define LAMBDA_FORCEINLINE +#endif +namespace Vulkan { namespace { +using boost::container::small_vector; +using boost::container::static_vector; +using Shader::ImageBufferDescriptor; +using Tegra::Texture::TexturePair; +using VideoCore::Surface::PixelFormat; +using VideoCore::Surface::PixelFormatFromDepthFormat; +using VideoCore::Surface::PixelFormatFromRenderTargetFormat; + +constexpr size_t NUM_STAGES = Maxwell::MaxShaderStage; +constexpr size_t MAX_IMAGE_ELEMENTS = 64; + +DescriptorLayoutBuilder MakeBuilder(const Device& device, std::span<const Shader::Info> infos) { + DescriptorLayoutBuilder builder{device}; + for (size_t index = 0; index < infos.size(); ++index) { + static constexpr std::array stages{ + VK_SHADER_STAGE_VERTEX_BIT, + VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT, + VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT, + VK_SHADER_STAGE_GEOMETRY_BIT, + VK_SHADER_STAGE_FRAGMENT_BIT, + }; + builder.Add(infos[index], stages.at(index)); + } + return builder; +} template <class StencilFace> VkStencilOpState GetStencilFaceState(const StencilFace& face) { @@ -39,15 +68,24 @@ VkStencilOpState GetStencilFaceState(const StencilFace& face) { } bool SupportsPrimitiveRestart(VkPrimitiveTopology topology) { - static constexpr std::array unsupported_topologies = { + static constexpr std::array unsupported_topologies{ VK_PRIMITIVE_TOPOLOGY_POINT_LIST, VK_PRIMITIVE_TOPOLOGY_LINE_LIST, VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST, VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY, VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY, - VK_PRIMITIVE_TOPOLOGY_PATCH_LIST}; - return std::find(std::begin(unsupported_topologies), std::end(unsupported_topologies), - topology) == std::end(unsupported_topologies); + VK_PRIMITIVE_TOPOLOGY_PATCH_LIST, + // VK_PRIMITIVE_TOPOLOGY_QUAD_LIST_EXT, + }; + return std::ranges::find(unsupported_topologies, topology) == unsupported_topologies.end(); +} + +bool IsLine(VkPrimitiveTopology topology) { + static constexpr std::array line_topologies{ + VK_PRIMITIVE_TOPOLOGY_LINE_LIST, VK_PRIMITIVE_TOPOLOGY_LINE_STRIP, + // VK_PRIMITIVE_TOPOLOGY_LINE_LOOP_EXT, + }; + return std::ranges::find(line_topologies, topology) == line_topologies.end(); } VkViewportSwizzleNV UnpackViewportSwizzle(u16 swizzle) { @@ -59,8 +97,7 @@ VkViewportSwizzleNV UnpackViewportSwizzle(u16 swizzle) { BitField<12, 3, Maxwell::ViewportSwizzle> w; }; const Swizzle unpacked{swizzle}; - - return { + return VkViewportSwizzleNV{ .x = MaxwellToVK::ViewportSwizzle(unpacked.x), .y = MaxwellToVK::ViewportSwizzle(unpacked.y), .z = MaxwellToVK::ViewportSwizzle(unpacked.z), @@ -68,193 +105,446 @@ VkViewportSwizzleNV UnpackViewportSwizzle(u16 swizzle) { }; } -VkSampleCountFlagBits ConvertMsaaMode(Tegra::Texture::MsaaMode msaa_mode) { - switch (msaa_mode) { - case Tegra::Texture::MsaaMode::Msaa1x1: - return VK_SAMPLE_COUNT_1_BIT; - case Tegra::Texture::MsaaMode::Msaa2x1: - case Tegra::Texture::MsaaMode::Msaa2x1_D3D: - return VK_SAMPLE_COUNT_2_BIT; - case Tegra::Texture::MsaaMode::Msaa2x2: - case Tegra::Texture::MsaaMode::Msaa2x2_VC4: - case Tegra::Texture::MsaaMode::Msaa2x2_VC12: - return VK_SAMPLE_COUNT_4_BIT; - case Tegra::Texture::MsaaMode::Msaa4x2: - case Tegra::Texture::MsaaMode::Msaa4x2_D3D: - case Tegra::Texture::MsaaMode::Msaa4x2_VC8: - case Tegra::Texture::MsaaMode::Msaa4x2_VC24: - return VK_SAMPLE_COUNT_8_BIT; - case Tegra::Texture::MsaaMode::Msaa4x4: - return VK_SAMPLE_COUNT_16_BIT; - default: - UNREACHABLE_MSG("Invalid msaa_mode={}", static_cast<int>(msaa_mode)); - return VK_SAMPLE_COUNT_1_BIT; +PixelFormat DecodeFormat(u8 encoded_format) { + const auto format{static_cast<Tegra::RenderTargetFormat>(encoded_format)}; + if (format == Tegra::RenderTargetFormat::NONE) { + return PixelFormat::Invalid; } + return PixelFormatFromRenderTargetFormat(format); } -} // Anonymous namespace +RenderPassKey MakeRenderPassKey(const FixedPipelineState& state) { + RenderPassKey key; + std::ranges::transform(state.color_formats, key.color_formats.begin(), DecodeFormat); + if (state.depth_enabled != 0) { + const auto depth_format{static_cast<Tegra::DepthFormat>(state.depth_format.Value())}; + key.depth_format = PixelFormatFromDepthFormat(depth_format); + } else { + key.depth_format = PixelFormat::Invalid; + } + key.samples = MaxwellToVK::MsaaMode(state.msaa_mode); + return key; +} -VKGraphicsPipeline::VKGraphicsPipeline(const Device& device_, VKScheduler& scheduler_, - VKDescriptorPool& descriptor_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_, - const GraphicsPipelineCacheKey& key, - vk::Span<VkDescriptorSetLayoutBinding> bindings, - const SPIRVProgram& program, u32 num_color_buffers) - : device{device_}, scheduler{scheduler_}, cache_key{key}, hash{cache_key.Hash()}, - descriptor_set_layout{CreateDescriptorSetLayout(bindings)}, - descriptor_allocator{descriptor_pool_, *descriptor_set_layout}, - update_descriptor_queue{update_descriptor_queue_}, layout{CreatePipelineLayout()}, - descriptor_template{CreateDescriptorUpdateTemplate(program)}, - modules(CreateShaderModules(program)), - pipeline(CreatePipeline(program, cache_key.renderpass, num_color_buffers)) {} - -VKGraphicsPipeline::~VKGraphicsPipeline() = default; - -VkDescriptorSet VKGraphicsPipeline::CommitDescriptorSet() { - if (!descriptor_template) { - return {}; - } - const VkDescriptorSet set = descriptor_allocator.Commit(); - update_descriptor_queue.Send(*descriptor_template, set); - return set; +size_t NumAttachments(const FixedPipelineState& state) { + size_t num{}; + for (size_t index = 0; index < Maxwell::NumRenderTargets; ++index) { + const auto format{static_cast<Tegra::RenderTargetFormat>(state.color_formats[index])}; + if (format != Tegra::RenderTargetFormat::NONE) { + num = index + 1; + } + } + return num; } -vk::DescriptorSetLayout VKGraphicsPipeline::CreateDescriptorSetLayout( - vk::Span<VkDescriptorSetLayoutBinding> bindings) const { - const VkDescriptorSetLayoutCreateInfo ci{ - .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .bindingCount = bindings.size(), - .pBindings = bindings.data(), - }; - return device.GetLogical().CreateDescriptorSetLayout(ci); +template <typename Spec> +bool Passes(const std::array<vk::ShaderModule, NUM_STAGES>& modules, + const std::array<Shader::Info, NUM_STAGES>& stage_infos) { + for (size_t stage = 0; stage < NUM_STAGES; ++stage) { + if (!Spec::enabled_stages[stage] && modules[stage]) { + return false; + } + const auto& info{stage_infos[stage]}; + if constexpr (!Spec::has_storage_buffers) { + if (!info.storage_buffers_descriptors.empty()) { + return false; + } + } + if constexpr (!Spec::has_texture_buffers) { + if (!info.texture_buffer_descriptors.empty()) { + return false; + } + } + if constexpr (!Spec::has_image_buffers) { + if (!info.image_buffer_descriptors.empty()) { + return false; + } + } + if constexpr (!Spec::has_images) { + if (!info.image_descriptors.empty()) { + return false; + } + } + } + return true; } -vk::PipelineLayout VKGraphicsPipeline::CreatePipelineLayout() const { - const VkPipelineLayoutCreateInfo ci{ - .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .setLayoutCount = 1, - .pSetLayouts = descriptor_set_layout.address(), - .pushConstantRangeCount = 0, - .pPushConstantRanges = nullptr, - }; - return device.GetLogical().CreatePipelineLayout(ci); +using ConfigureFuncPtr = void (*)(GraphicsPipeline*, bool); + +template <typename Spec, typename... Specs> +ConfigureFuncPtr FindSpec(const std::array<vk::ShaderModule, NUM_STAGES>& modules, + const std::array<Shader::Info, NUM_STAGES>& stage_infos) { + if constexpr (sizeof...(Specs) > 0) { + if (!Passes<Spec>(modules, stage_infos)) { + return FindSpec<Specs...>(modules, stage_infos); + } + } + return GraphicsPipeline::MakeConfigureSpecFunc<Spec>(); } -vk::DescriptorUpdateTemplateKHR VKGraphicsPipeline::CreateDescriptorUpdateTemplate( - const SPIRVProgram& program) const { - std::vector<VkDescriptorUpdateTemplateEntry> template_entries; - u32 binding = 0; - u32 offset = 0; - for (const auto& stage : program) { - if (stage) { - FillDescriptorUpdateTemplateEntries(stage->entries, binding, offset, template_entries); +struct SimpleVertexFragmentSpec { + static constexpr std::array<bool, 5> enabled_stages{true, false, false, false, true}; + static constexpr bool has_storage_buffers = false; + static constexpr bool has_texture_buffers = false; + static constexpr bool has_image_buffers = false; + static constexpr bool has_images = false; +}; + +struct SimpleVertexSpec { + static constexpr std::array<bool, 5> enabled_stages{true, false, false, false, false}; + static constexpr bool has_storage_buffers = false; + static constexpr bool has_texture_buffers = false; + static constexpr bool has_image_buffers = false; + static constexpr bool has_images = false; +}; + +struct DefaultSpec { + static constexpr std::array<bool, 5> enabled_stages{true, true, true, true, true}; + static constexpr bool has_storage_buffers = true; + static constexpr bool has_texture_buffers = true; + static constexpr bool has_image_buffers = true; + static constexpr bool has_images = true; +}; + +ConfigureFuncPtr ConfigureFunc(const std::array<vk::ShaderModule, NUM_STAGES>& modules, + const std::array<Shader::Info, NUM_STAGES>& infos) { + return FindSpec<SimpleVertexSpec, SimpleVertexFragmentSpec, DefaultSpec>(modules, infos); +} +} // Anonymous namespace + +GraphicsPipeline::GraphicsPipeline( + Tegra::Engines::Maxwell3D& maxwell3d_, Tegra::MemoryManager& gpu_memory_, + VKScheduler& scheduler_, BufferCache& buffer_cache_, TextureCache& texture_cache_, + VideoCore::ShaderNotify* shader_notify, const Device& device_, DescriptorPool& descriptor_pool, + VKUpdateDescriptorQueue& update_descriptor_queue_, Common::ThreadWorker* worker_thread, + RenderPassCache& render_pass_cache, const GraphicsPipelineCacheKey& key_, + std::array<vk::ShaderModule, NUM_STAGES> stages, + const std::array<const Shader::Info*, NUM_STAGES>& infos) + : key{key_}, maxwell3d{maxwell3d_}, gpu_memory{gpu_memory_}, device{device_}, + texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, scheduler{scheduler_}, + update_descriptor_queue{update_descriptor_queue_}, spv_modules{std::move(stages)} { + if (shader_notify) { + shader_notify->MarkShaderBuilding(); + } + for (size_t stage = 0; stage < NUM_STAGES; ++stage) { + const Shader::Info* const info{infos[stage]}; + if (!info) { + continue; } + stage_infos[stage] = *info; + enabled_uniform_buffer_masks[stage] = info->constant_buffer_mask; + std::ranges::copy(info->constant_buffer_used_sizes, uniform_buffer_sizes[stage].begin()); } - if (template_entries.empty()) { - // If the shader doesn't use descriptor sets, skip template creation. - return {}; + auto func{[this, shader_notify, &render_pass_cache, &descriptor_pool] { + DescriptorLayoutBuilder builder{MakeBuilder(device, stage_infos)}; + uses_push_descriptor = builder.CanUsePushDescriptor(); + descriptor_set_layout = builder.CreateDescriptorSetLayout(uses_push_descriptor); + if (!uses_push_descriptor) { + descriptor_allocator = descriptor_pool.Allocator(*descriptor_set_layout, stage_infos); + } + const VkDescriptorSetLayout set_layout{*descriptor_set_layout}; + pipeline_layout = builder.CreatePipelineLayout(set_layout); + descriptor_update_template = + builder.CreateTemplate(set_layout, *pipeline_layout, uses_push_descriptor); + + const VkRenderPass render_pass{render_pass_cache.Get(MakeRenderPassKey(key.state))}; + Validate(); + MakePipeline(render_pass); + + std::lock_guard lock{build_mutex}; + is_built = true; + build_condvar.notify_one(); + if (shader_notify) { + shader_notify->MarkShaderComplete(); + } + }}; + if (worker_thread) { + worker_thread->QueueWork(std::move(func)); + } else { + func(); } + configure_func = ConfigureFunc(spv_modules, stage_infos); +} - const VkDescriptorUpdateTemplateCreateInfoKHR ci{ - .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_UPDATE_TEMPLATE_CREATE_INFO_KHR, - .pNext = nullptr, - .flags = 0, - .descriptorUpdateEntryCount = static_cast<u32>(template_entries.size()), - .pDescriptorUpdateEntries = template_entries.data(), - .templateType = VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR, - .descriptorSetLayout = *descriptor_set_layout, - .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS, - .pipelineLayout = *layout, - .set = DESCRIPTOR_SET, - }; - return device.GetLogical().CreateDescriptorUpdateTemplateKHR(ci); +void GraphicsPipeline::AddTransition(GraphicsPipeline* transition) { + transition_keys.push_back(transition->key); + transitions.push_back(transition); } -std::vector<vk::ShaderModule> VKGraphicsPipeline::CreateShaderModules( - const SPIRVProgram& program) const { - VkShaderModuleCreateInfo ci{ - .sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .codeSize = 0, - .pCode = nullptr, - }; +template <typename Spec> +void GraphicsPipeline::ConfigureImpl(bool is_indexed) { + std::array<ImageId, MAX_IMAGE_ELEMENTS> image_view_ids; + std::array<u32, MAX_IMAGE_ELEMENTS> image_view_indices; + std::array<VkSampler, MAX_IMAGE_ELEMENTS> samplers; + size_t sampler_index{}; + size_t image_index{}; + + texture_cache.SynchronizeGraphicsDescriptors(); + + buffer_cache.SetUniformBuffersState(enabled_uniform_buffer_masks, &uniform_buffer_sizes); + + const auto& regs{maxwell3d.regs}; + const bool via_header_index{regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex}; + const auto config_stage{[&](size_t stage) LAMBDA_FORCEINLINE { + const Shader::Info& info{stage_infos[stage]}; + buffer_cache.UnbindGraphicsStorageBuffers(stage); + if constexpr (Spec::has_storage_buffers) { + size_t ssbo_index{}; + for (const auto& desc : info.storage_buffers_descriptors) { + ASSERT(desc.count == 1); + buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, desc.cbuf_index, + desc.cbuf_offset, desc.is_written); + ++ssbo_index; + } + } + const auto& cbufs{maxwell3d.state.shader_stages[stage].const_buffers}; + const auto read_handle{[&](const auto& desc, u32 index) { + ASSERT(cbufs[desc.cbuf_index].enabled); + const u32 index_offset{index << desc.size_shift}; + const u32 offset{desc.cbuf_offset + index_offset}; + const GPUVAddr addr{cbufs[desc.cbuf_index].address + offset}; + if constexpr (std::is_same_v<decltype(desc), const Shader::TextureDescriptor&> || + std::is_same_v<decltype(desc), const Shader::TextureBufferDescriptor&>) { + if (desc.has_secondary) { + ASSERT(cbufs[desc.secondary_cbuf_index].enabled); + const u32 second_offset{desc.secondary_cbuf_offset + index_offset}; + const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].address + + second_offset}; + const u32 lhs_raw{gpu_memory.Read<u32>(addr)}; + const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)}; + const u32 raw{lhs_raw | rhs_raw}; + return TexturePair(raw, via_header_index); + } + } + return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); + }}; + const auto add_image{[&](const auto& desc) { + for (u32 index = 0; index < desc.count; ++index) { + const auto handle{read_handle(desc, index)}; + image_view_indices[image_index++] = handle.first; + } + }}; + if constexpr (Spec::has_texture_buffers) { + for (const auto& desc : info.texture_buffer_descriptors) { + add_image(desc); + } + } + if constexpr (Spec::has_image_buffers) { + for (const auto& desc : info.image_buffer_descriptors) { + add_image(desc); + } + } + for (const auto& desc : info.texture_descriptors) { + for (u32 index = 0; index < desc.count; ++index) { + const auto handle{read_handle(desc, index)}; + image_view_indices[image_index++] = handle.first; - std::vector<vk::ShaderModule> shader_modules; - shader_modules.reserve(Maxwell::MaxShaderStage); - for (std::size_t i = 0; i < Maxwell::MaxShaderStage; ++i) { - const auto& stage = program[i]; - if (!stage) { - continue; + Sampler* const sampler{texture_cache.GetGraphicsSampler(handle.second)}; + samplers[sampler_index++] = sampler->Handle(); + } + } + if constexpr (Spec::has_images) { + for (const auto& desc : info.image_descriptors) { + add_image(desc); + } } + }}; + if constexpr (Spec::enabled_stages[0]) { + config_stage(0); + } + if constexpr (Spec::enabled_stages[1]) { + config_stage(1); + } + if constexpr (Spec::enabled_stages[2]) { + config_stage(2); + } + if constexpr (Spec::enabled_stages[3]) { + config_stage(3); + } + if constexpr (Spec::enabled_stages[4]) { + config_stage(4); + } + const std::span indices_span(image_view_indices.data(), image_index); + texture_cache.FillGraphicsImageViews(indices_span, image_view_ids); + + ImageId* texture_buffer_index{image_view_ids.data()}; + const auto bind_stage_info{[&](size_t stage) LAMBDA_FORCEINLINE { + size_t index{}; + const auto add_buffer{[&](const auto& desc) { + constexpr bool is_image = std::is_same_v<decltype(desc), const ImageBufferDescriptor&>; + for (u32 i = 0; i < desc.count; ++i) { + bool is_written{false}; + if constexpr (is_image) { + is_written = desc.is_written; + } + ImageView& image_view{texture_cache.GetImageView(*texture_buffer_index)}; + buffer_cache.BindGraphicsTextureBuffer(stage, index, image_view.GpuAddr(), + image_view.BufferSize(), image_view.format, + is_written, is_image); + ++index; + ++texture_buffer_index; + } + }}; + buffer_cache.UnbindGraphicsTextureBuffers(stage); - device.SaveShader(stage->code); + const Shader::Info& info{stage_infos[stage]}; + if constexpr (Spec::has_texture_buffers) { + for (const auto& desc : info.texture_buffer_descriptors) { + add_buffer(desc); + } + } + if constexpr (Spec::has_image_buffers) { + for (const auto& desc : info.image_buffer_descriptors) { + add_buffer(desc); + } + } + for (const auto& desc : info.texture_descriptors) { + texture_buffer_index += desc.count; + } + if constexpr (Spec::has_images) { + for (const auto& desc : info.image_descriptors) { + texture_buffer_index += desc.count; + } + } + }}; + if constexpr (Spec::enabled_stages[0]) { + bind_stage_info(0); + } + if constexpr (Spec::enabled_stages[1]) { + bind_stage_info(1); + } + if constexpr (Spec::enabled_stages[2]) { + bind_stage_info(2); + } + if constexpr (Spec::enabled_stages[3]) { + bind_stage_info(3); + } + if constexpr (Spec::enabled_stages[4]) { + bind_stage_info(4); + } + + buffer_cache.UpdateGraphicsBuffers(is_indexed); + buffer_cache.BindHostGeometryBuffers(is_indexed); - ci.codeSize = stage->code.size() * sizeof(u32); - ci.pCode = stage->code.data(); - shader_modules.push_back(device.GetLogical().CreateShaderModule(ci)); + update_descriptor_queue.Acquire(); + + const VkSampler* samplers_it{samplers.data()}; + const ImageId* views_it{image_view_ids.data()}; + const auto prepare_stage{[&](size_t stage) LAMBDA_FORCEINLINE { + buffer_cache.BindHostStageBuffers(stage); + PushImageDescriptors(stage_infos[stage], samplers_it, views_it, texture_cache, + update_descriptor_queue); + }}; + if constexpr (Spec::enabled_stages[0]) { + prepare_stage(0); + } + if constexpr (Spec::enabled_stages[1]) { + prepare_stage(1); } - return shader_modules; + if constexpr (Spec::enabled_stages[2]) { + prepare_stage(2); + } + if constexpr (Spec::enabled_stages[3]) { + prepare_stage(3); + } + if constexpr (Spec::enabled_stages[4]) { + prepare_stage(4); + } + ConfigureDraw(); } -vk::Pipeline VKGraphicsPipeline::CreatePipeline(const SPIRVProgram& program, - VkRenderPass renderpass, - u32 num_color_buffers) const { - const auto& state = cache_key.fixed_state; - const auto& viewport_swizzles = state.viewport_swizzles; - - FixedPipelineState::DynamicState dynamic; - if (device.IsExtExtendedDynamicStateSupported()) { - // Insert dummy values, as long as they are valid they don't matter as extended dynamic - // state is ignored - dynamic.raw1 = 0; - dynamic.raw2 = 0; - dynamic.vertex_strides.fill(0); - } else { - dynamic = state.dynamic_state; - } - - std::vector<VkVertexInputBindingDescription> vertex_bindings; - std::vector<VkVertexInputBindingDivisorDescriptionEXT> vertex_binding_divisors; - for (std::size_t index = 0; index < Maxwell::NumVertexArrays; ++index) { - const bool instanced = state.binding_divisors[index] != 0; - const auto rate = instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX; - vertex_bindings.push_back({ - .binding = static_cast<u32>(index), - .stride = dynamic.vertex_strides[index], - .inputRate = rate, +void GraphicsPipeline::ConfigureDraw() { + texture_cache.UpdateRenderTargets(false); + scheduler.RequestRenderpass(texture_cache.GetFramebuffer()); + + if (!is_built.load(std::memory_order::relaxed)) { + // Wait for the pipeline to be built + scheduler.Record([this](vk::CommandBuffer) { + std::unique_lock lock{build_mutex}; + build_condvar.wait(lock, [this] { return is_built.load(std::memory_order::relaxed); }); }); - if (instanced) { - vertex_binding_divisors.push_back({ - .binding = static_cast<u32>(index), - .divisor = state.binding_divisors[index], - }); - } } + const bool bind_pipeline{scheduler.UpdateGraphicsPipeline(this)}; + const void* const descriptor_data{update_descriptor_queue.UpdateData()}; + scheduler.Record([this, descriptor_data, bind_pipeline](vk::CommandBuffer cmdbuf) { + if (bind_pipeline) { + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline); + } + if (!descriptor_set_layout) { + return; + } + if (uses_push_descriptor) { + cmdbuf.PushDescriptorSetWithTemplateKHR(*descriptor_update_template, *pipeline_layout, + 0, descriptor_data); + } else { + const VkDescriptorSet descriptor_set{descriptor_allocator.Commit()}; + const vk::Device& dev{device.GetLogical()}; + dev.UpdateDescriptorSet(descriptor_set, *descriptor_update_template, descriptor_data); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline_layout, 0, + descriptor_set, nullptr); + } + }); +} - std::vector<VkVertexInputAttributeDescription> vertex_attributes; - const auto& input_attributes = program[0]->entries.attributes; - for (std::size_t index = 0; index < state.attributes.size(); ++index) { - const auto& attribute = state.attributes[index]; - if (!attribute.enabled) { - continue; +void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) { + FixedPipelineState::DynamicState dynamic{}; + if (!key.state.extended_dynamic_state) { + dynamic = key.state.dynamic_state; + } + static_vector<VkVertexInputBindingDescription, 32> vertex_bindings; + static_vector<VkVertexInputBindingDivisorDescriptionEXT, 32> vertex_binding_divisors; + static_vector<VkVertexInputAttributeDescription, 32> vertex_attributes; + if (key.state.dynamic_vertex_input) { + for (size_t index = 0; index < key.state.attributes.size(); ++index) { + const u32 type = key.state.DynamicAttributeType(index); + if (!stage_infos[0].loads.Generic(index) || type == 0) { + continue; + } + vertex_attributes.push_back({ + .location = static_cast<u32>(index), + .binding = 0, + .format = type == 1 ? VK_FORMAT_R32_SFLOAT + : type == 2 ? VK_FORMAT_R32_SINT : VK_FORMAT_R32_UINT, + .offset = 0, + }); } - if (!input_attributes.contains(static_cast<u32>(index))) { - // Skip attributes not used by the vertex shaders. - continue; + if (!vertex_attributes.empty()) { + vertex_bindings.push_back({ + .binding = 0, + .stride = 4, + .inputRate = VK_VERTEX_INPUT_RATE_VERTEX, + }); + } + } else { + for (size_t index = 0; index < Maxwell::NumVertexArrays; ++index) { + const bool instanced = key.state.binding_divisors[index] != 0; + const auto rate = + instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX; + vertex_bindings.push_back({ + .binding = static_cast<u32>(index), + .stride = dynamic.vertex_strides[index], + .inputRate = rate, + }); + if (instanced) { + vertex_binding_divisors.push_back({ + .binding = static_cast<u32>(index), + .divisor = key.state.binding_divisors[index], + }); + } + } + for (size_t index = 0; index < key.state.attributes.size(); ++index) { + const auto& attribute = key.state.attributes[index]; + if (!attribute.enabled || !stage_infos[0].loads.Generic(index)) { + continue; + } + vertex_attributes.push_back({ + .location = static_cast<u32>(index), + .binding = attribute.buffer, + .format = MaxwellToVK::VertexFormat(attribute.Type(), attribute.Size()), + .offset = attribute.offset, + }); } - vertex_attributes.push_back({ - .location = static_cast<u32>(index), - .binding = attribute.buffer, - .format = MaxwellToVK::VertexFormat(attribute.Type(), attribute.Size()), - .offset = attribute.offset, - }); } - VkPipelineVertexInputStateCreateInfo vertex_input_ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO, .pNext = nullptr, @@ -264,7 +554,6 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const SPIRVProgram& program, .vertexAttributeDescriptionCount = static_cast<u32>(vertex_attributes.size()), .pVertexAttributeDescriptions = vertex_attributes.data(), }; - const VkPipelineVertexInputDivisorStateCreateInfoEXT input_divisor_ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_DIVISOR_STATE_CREATE_INFO_EXT, .pNext = nullptr, @@ -274,78 +563,113 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const SPIRVProgram& program, if (!vertex_binding_divisors.empty()) { vertex_input_ci.pNext = &input_divisor_ci; } - - const auto input_assembly_topology = MaxwellToVK::PrimitiveTopology(device, state.topology); + auto input_assembly_topology = MaxwellToVK::PrimitiveTopology(device, key.state.topology); + if (input_assembly_topology == VK_PRIMITIVE_TOPOLOGY_PATCH_LIST) { + if (!spv_modules[1] && !spv_modules[2]) { + LOG_WARNING(Render_Vulkan, "Patch topology used without tessellation, using points"); + input_assembly_topology = VK_PRIMITIVE_TOPOLOGY_POINT_LIST; + } + } const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO, .pNext = nullptr, .flags = 0, - .topology = MaxwellToVK::PrimitiveTopology(device, state.topology), - .primitiveRestartEnable = state.primitive_restart_enable != 0 && + .topology = input_assembly_topology, + .primitiveRestartEnable = key.state.primitive_restart_enable != 0 && SupportsPrimitiveRestart(input_assembly_topology), }; - const VkPipelineTessellationStateCreateInfo tessellation_ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO, .pNext = nullptr, .flags = 0, - .patchControlPoints = state.patch_control_points_minus_one.Value() + 1, - }; - - VkPipelineViewportStateCreateInfo viewport_ci{ - .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .viewportCount = Maxwell::NumViewports, - .pViewports = nullptr, - .scissorCount = Maxwell::NumViewports, - .pScissors = nullptr, + .patchControlPoints = key.state.patch_control_points_minus_one.Value() + 1, }; std::array<VkViewportSwizzleNV, Maxwell::NumViewports> swizzles; - std::ranges::transform(viewport_swizzles, swizzles.begin(), UnpackViewportSwizzle); - VkPipelineViewportSwizzleStateCreateInfoNV swizzle_ci{ + std::ranges::transform(key.state.viewport_swizzles, swizzles.begin(), UnpackViewportSwizzle); + const VkPipelineViewportSwizzleStateCreateInfoNV swizzle_ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_SWIZZLE_STATE_CREATE_INFO_NV, .pNext = nullptr, .flags = 0, .viewportCount = Maxwell::NumViewports, .pViewportSwizzles = swizzles.data(), }; - if (device.IsNvViewportSwizzleSupported()) { - viewport_ci.pNext = &swizzle_ci; - } + const VkPipelineViewportStateCreateInfo viewport_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO, + .pNext = device.IsNvViewportSwizzleSupported() ? &swizzle_ci : nullptr, + .flags = 0, + .viewportCount = Maxwell::NumViewports, + .pViewports = nullptr, + .scissorCount = Maxwell::NumViewports, + .pScissors = nullptr, + }; - const VkPipelineRasterizationStateCreateInfo rasterization_ci{ + VkPipelineRasterizationStateCreateInfo rasterization_ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO, .pNext = nullptr, .flags = 0, .depthClampEnable = - static_cast<VkBool32>(state.depth_clamp_disabled == 0 ? VK_TRUE : VK_FALSE), + static_cast<VkBool32>(key.state.depth_clamp_disabled == 0 ? VK_TRUE : VK_FALSE), .rasterizerDiscardEnable = - static_cast<VkBool32>(state.rasterize_enable == 0 ? VK_TRUE : VK_FALSE), - .polygonMode = VK_POLYGON_MODE_FILL, + static_cast<VkBool32>(key.state.rasterize_enable == 0 ? VK_TRUE : VK_FALSE), + .polygonMode = + MaxwellToVK::PolygonMode(FixedPipelineState::UnpackPolygonMode(key.state.polygon_mode)), .cullMode = static_cast<VkCullModeFlags>( dynamic.cull_enable ? MaxwellToVK::CullFace(dynamic.CullFace()) : VK_CULL_MODE_NONE), .frontFace = MaxwellToVK::FrontFace(dynamic.FrontFace()), - .depthBiasEnable = state.depth_bias_enable, + .depthBiasEnable = key.state.depth_bias_enable, .depthBiasConstantFactor = 0.0f, .depthBiasClamp = 0.0f, .depthBiasSlopeFactor = 0.0f, .lineWidth = 1.0f, }; + VkPipelineRasterizationLineStateCreateInfoEXT line_state{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_LINE_STATE_CREATE_INFO_EXT, + .pNext = nullptr, + .lineRasterizationMode = key.state.smooth_lines != 0 + ? VK_LINE_RASTERIZATION_MODE_RECTANGULAR_SMOOTH_EXT + : VK_LINE_RASTERIZATION_MODE_RECTANGULAR_EXT, + .stippledLineEnable = VK_FALSE, // TODO + .lineStippleFactor = 0, + .lineStipplePattern = 0, + }; + VkPipelineRasterizationConservativeStateCreateInfoEXT conservative_raster{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_CONSERVATIVE_STATE_CREATE_INFO_EXT, + .pNext = nullptr, + .flags = 0, + .conservativeRasterizationMode = key.state.conservative_raster_enable != 0 + ? VK_CONSERVATIVE_RASTERIZATION_MODE_OVERESTIMATE_EXT + : VK_CONSERVATIVE_RASTERIZATION_MODE_DISABLED_EXT, + .extraPrimitiveOverestimationSize = 0.0f, + }; + VkPipelineRasterizationProvokingVertexStateCreateInfoEXT provoking_vertex{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_PROVOKING_VERTEX_STATE_CREATE_INFO_EXT, + .pNext = nullptr, + .provokingVertexMode = key.state.provoking_vertex_last != 0 + ? VK_PROVOKING_VERTEX_MODE_LAST_VERTEX_EXT + : VK_PROVOKING_VERTEX_MODE_FIRST_VERTEX_EXT, + }; + if (IsLine(input_assembly_topology) && device.IsExtLineRasterizationSupported()) { + line_state.pNext = std::exchange(rasterization_ci.pNext, &line_state); + } + if (device.IsExtConservativeRasterizationSupported()) { + conservative_raster.pNext = std::exchange(rasterization_ci.pNext, &conservative_raster); + } + if (device.IsExtProvokingVertexSupported()) { + provoking_vertex.pNext = std::exchange(rasterization_ci.pNext, &provoking_vertex); + } const VkPipelineMultisampleStateCreateInfo multisample_ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO, .pNext = nullptr, .flags = 0, - .rasterizationSamples = ConvertMsaaMode(state.msaa_mode), + .rasterizationSamples = MaxwellToVK::MsaaMode(key.state.msaa_mode), .sampleShadingEnable = VK_FALSE, .minSampleShading = 0.0f, .pSampleMask = nullptr, .alphaToCoverageEnable = VK_FALSE, .alphaToOneEnable = VK_FALSE, }; - const VkPipelineDepthStencilStateCreateInfo depth_stencil_ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO, .pNext = nullptr, @@ -355,32 +679,32 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const SPIRVProgram& program, .depthCompareOp = dynamic.depth_test_enable ? MaxwellToVK::ComparisonOp(dynamic.DepthTestFunc()) : VK_COMPARE_OP_ALWAYS, - .depthBoundsTestEnable = dynamic.depth_bounds_enable, + .depthBoundsTestEnable = dynamic.depth_bounds_enable && device.IsDepthBoundsSupported(), .stencilTestEnable = dynamic.stencil_enable, .front = GetStencilFaceState(dynamic.front), .back = GetStencilFaceState(dynamic.back), .minDepthBounds = 0.0f, .maxDepthBounds = 0.0f, }; - - std::array<VkPipelineColorBlendAttachmentState, Maxwell::NumRenderTargets> cb_attachments; - for (std::size_t index = 0; index < num_color_buffers; ++index) { - static constexpr std::array COMPONENT_TABLE{ + if (dynamic.depth_bounds_enable && !device.IsDepthBoundsSupported()) { + LOG_WARNING(Render_Vulkan, "Depth bounds is enabled but not supported"); + } + static_vector<VkPipelineColorBlendAttachmentState, Maxwell::NumRenderTargets> cb_attachments; + const size_t num_attachments{NumAttachments(key.state)}; + for (size_t index = 0; index < num_attachments; ++index) { + static constexpr std::array mask_table{ VK_COLOR_COMPONENT_R_BIT, VK_COLOR_COMPONENT_G_BIT, VK_COLOR_COMPONENT_B_BIT, VK_COLOR_COMPONENT_A_BIT, }; - const auto& blend = state.attachments[index]; - - VkColorComponentFlags color_components = 0; - for (std::size_t i = 0; i < COMPONENT_TABLE.size(); ++i) { - if (blend.Mask()[i]) { - color_components |= COMPONENT_TABLE[i]; - } + const auto& blend{key.state.attachments[index]}; + const std::array mask{blend.Mask()}; + VkColorComponentFlags write_mask{}; + for (size_t i = 0; i < mask_table.size(); ++i) { + write_mask |= mask[i] ? mask_table[i] : 0; } - - cb_attachments[index] = { + cb_attachments.push_back({ .blendEnable = blend.enable != 0, .srcColorBlendFactor = MaxwellToVK::BlendFactor(blend.SourceRGBFactor()), .dstColorBlendFactor = MaxwellToVK::BlendFactor(blend.DestRGBFactor()), @@ -388,28 +712,27 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const SPIRVProgram& program, .srcAlphaBlendFactor = MaxwellToVK::BlendFactor(blend.SourceAlphaFactor()), .dstAlphaBlendFactor = MaxwellToVK::BlendFactor(blend.DestAlphaFactor()), .alphaBlendOp = MaxwellToVK::BlendEquation(blend.EquationAlpha()), - .colorWriteMask = color_components, - }; + .colorWriteMask = write_mask, + }); } - const VkPipelineColorBlendStateCreateInfo color_blend_ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO, .pNext = nullptr, .flags = 0, .logicOpEnable = VK_FALSE, .logicOp = VK_LOGIC_OP_COPY, - .attachmentCount = num_color_buffers, + .attachmentCount = static_cast<u32>(cb_attachments.size()), .pAttachments = cb_attachments.data(), .blendConstants = {}, }; - - std::vector dynamic_states{ + static_vector<VkDynamicState, 19> dynamic_states{ VK_DYNAMIC_STATE_VIEWPORT, VK_DYNAMIC_STATE_SCISSOR, VK_DYNAMIC_STATE_DEPTH_BIAS, VK_DYNAMIC_STATE_BLEND_CONSTANTS, VK_DYNAMIC_STATE_DEPTH_BOUNDS, VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK, VK_DYNAMIC_STATE_STENCIL_WRITE_MASK, VK_DYNAMIC_STATE_STENCIL_REFERENCE, + VK_DYNAMIC_STATE_LINE_WIDTH, }; - if (device.IsExtExtendedDynamicStateSupported()) { + if (key.state.extended_dynamic_state) { static constexpr std::array extended{ VK_DYNAMIC_STATE_CULL_MODE_EXT, VK_DYNAMIC_STATE_FRONT_FACE_EXT, @@ -421,9 +744,11 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const SPIRVProgram& program, VK_DYNAMIC_STATE_STENCIL_TEST_ENABLE_EXT, VK_DYNAMIC_STATE_STENCIL_OP_EXT, }; + if (key.state.dynamic_vertex_input) { + dynamic_states.push_back(VK_DYNAMIC_STATE_VERTEX_INPUT_EXT); + } dynamic_states.insert(dynamic_states.end(), extended.begin(), extended.end()); } - const VkPipelineDynamicStateCreateInfo dynamic_state_ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO, .pNext = nullptr, @@ -431,34 +756,33 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const SPIRVProgram& program, .dynamicStateCount = static_cast<u32>(dynamic_states.size()), .pDynamicStates = dynamic_states.data(), }; - - const VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci{ + [[maybe_unused]] const VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_REQUIRED_SUBGROUP_SIZE_CREATE_INFO_EXT, .pNext = nullptr, .requiredSubgroupSize = GuestWarpSize, }; - - std::vector<VkPipelineShaderStageCreateInfo> shader_stages; - std::size_t module_index = 0; - for (std::size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) { - if (!program[stage]) { + static_vector<VkPipelineShaderStageCreateInfo, 5> shader_stages; + for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) { + if (!spv_modules[stage]) { continue; } - - VkPipelineShaderStageCreateInfo& stage_ci = shader_stages.emplace_back(); - stage_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO; - stage_ci.pNext = nullptr; - stage_ci.flags = 0; - stage_ci.stage = MaxwellToVK::ShaderStage(static_cast<Tegra::Engines::ShaderType>(stage)); - stage_ci.module = *modules[module_index++]; - stage_ci.pName = "main"; - stage_ci.pSpecializationInfo = nullptr; - + [[maybe_unused]] auto& stage_ci = + shader_stages.emplace_back(VkPipelineShaderStageCreateInfo{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = MaxwellToVK::ShaderStage(Shader::StageFromIndex(stage)), + .module = *spv_modules[stage], + .pName = "main", + .pSpecializationInfo = nullptr, + }); + /* if (program[stage]->entries.uses_warps && device.IsGuestWarpSizeSupported(stage_ci.stage)) { stage_ci.pNext = &subgroup_size_ci; } + */ } - return device.GetLogical().CreateGraphicsPipeline(VkGraphicsPipelineCreateInfo{ + pipeline = device.GetLogical().CreateGraphicsPipeline({ .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, .pNext = nullptr, .flags = 0, @@ -473,12 +797,31 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const SPIRVProgram& program, .pDepthStencilState = &depth_stencil_ci, .pColorBlendState = &color_blend_ci, .pDynamicState = &dynamic_state_ci, - .layout = *layout, - .renderPass = renderpass, + .layout = *pipeline_layout, + .renderPass = render_pass, .subpass = 0, .basePipelineHandle = nullptr, .basePipelineIndex = 0, }); } +void GraphicsPipeline::Validate() { + size_t num_images{}; + for (const auto& info : stage_infos) { + for (const auto& desc : info.texture_buffer_descriptors) { + num_images += desc.count; + } + for (const auto& desc : info.image_buffer_descriptors) { + num_images += desc.count; + } + for (const auto& desc : info.texture_descriptors) { + num_images += desc.count; + } + for (const auto& desc : info.image_descriptors) { + num_images += desc.count; + } + } + ASSERT(num_images <= MAX_IMAGE_ELEMENTS); +} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h index 8b6a98fe0..2bd48d697 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h @@ -1,30 +1,36 @@ -// Copyright 2019 yuzu Emulator Project +// Copyright 2021 yuzu Emulator Project // Licensed under GPLv2 or any later version // Refer to the license.txt file included. #pragma once +#include <algorithm> #include <array> -#include <optional> -#include <vector> +#include <atomic> +#include <condition_variable> +#include <mutex> +#include <type_traits> -#include "common/common_types.h" +#include "common/thread_worker.h" +#include "shader_recompiler/shader_info.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/renderer_vulkan/fixed_pipeline_state.h" +#include "video_core/renderer_vulkan/vk_buffer_cache.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" -#include "video_core/renderer_vulkan/vk_shader_decompiler.h" +#include "video_core/renderer_vulkan/vk_texture_cache.h" #include "video_core/vulkan_common/vulkan_wrapper.h" -namespace Vulkan { +namespace VideoCore { +class ShaderNotify; +} -using Maxwell = Tegra::Engines::Maxwell3D::Regs; +namespace Vulkan { struct GraphicsPipelineCacheKey { - VkRenderPass renderpass; - std::array<GPUVAddr, Maxwell::MaxShaderProgram> shaders; - FixedPipelineState fixed_state; + std::array<u64, 6> unique_hashes; + FixedPipelineState state; - std::size_t Hash() const noexcept; + size_t Hash() const noexcept; bool operator==(const GraphicsPipelineCacheKey& rhs) const noexcept; @@ -32,72 +38,115 @@ struct GraphicsPipelineCacheKey { return !operator==(rhs); } - std::size_t Size() const noexcept { - return sizeof(renderpass) + sizeof(shaders) + fixed_state.Size(); + size_t Size() const noexcept { + return sizeof(unique_hashes) + state.Size(); } }; static_assert(std::has_unique_object_representations_v<GraphicsPipelineCacheKey>); static_assert(std::is_trivially_copyable_v<GraphicsPipelineCacheKey>); static_assert(std::is_trivially_constructible_v<GraphicsPipelineCacheKey>); +} // namespace Vulkan + +namespace std { +template <> +struct hash<Vulkan::GraphicsPipelineCacheKey> { + size_t operator()(const Vulkan::GraphicsPipelineCacheKey& k) const noexcept { + return k.Hash(); + } +}; +} // namespace std + +namespace Vulkan { + class Device; -class VKDescriptorPool; +class RenderPassCache; class VKScheduler; class VKUpdateDescriptorQueue; -using SPIRVProgram = std::array<std::optional<SPIRVShader>, Maxwell::MaxShaderStage>; +class GraphicsPipeline { + static constexpr size_t NUM_STAGES = Tegra::Engines::Maxwell3D::Regs::MaxShaderStage; -class VKGraphicsPipeline final { public: - explicit VKGraphicsPipeline(const Device& device_, VKScheduler& scheduler_, - VKDescriptorPool& descriptor_pool, - VKUpdateDescriptorQueue& update_descriptor_queue_, - const GraphicsPipelineCacheKey& key, - vk::Span<VkDescriptorSetLayoutBinding> bindings, - const SPIRVProgram& program, u32 num_color_buffers); - ~VKGraphicsPipeline(); - - VkDescriptorSet CommitDescriptorSet(); - - VkPipeline GetHandle() const { - return *pipeline; + explicit GraphicsPipeline( + Tegra::Engines::Maxwell3D& maxwell3d, Tegra::MemoryManager& gpu_memory, + VKScheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache, + VideoCore::ShaderNotify* shader_notify, const Device& device, + DescriptorPool& descriptor_pool, VKUpdateDescriptorQueue& update_descriptor_queue, + Common::ThreadWorker* worker_thread, RenderPassCache& render_pass_cache, + const GraphicsPipelineCacheKey& key, std::array<vk::ShaderModule, NUM_STAGES> stages, + const std::array<const Shader::Info*, NUM_STAGES>& infos); + + GraphicsPipeline& operator=(GraphicsPipeline&&) noexcept = delete; + GraphicsPipeline(GraphicsPipeline&&) noexcept = delete; + + GraphicsPipeline& operator=(const GraphicsPipeline&) = delete; + GraphicsPipeline(const GraphicsPipeline&) = delete; + + void AddTransition(GraphicsPipeline* transition); + + void Configure(bool is_indexed) { + configure_func(this, is_indexed); } - VkPipelineLayout GetLayout() const { - return *layout; + [[nodiscard]] GraphicsPipeline* Next(const GraphicsPipelineCacheKey& current_key) noexcept { + if (key == current_key) { + return this; + } + const auto it{std::find(transition_keys.begin(), transition_keys.end(), current_key)}; + return it != transition_keys.end() ? transitions[std::distance(transition_keys.begin(), it)] + : nullptr; } - GraphicsPipelineCacheKey GetCacheKey() const { - return cache_key; + [[nodiscard]] bool IsBuilt() const noexcept { + return is_built.load(std::memory_order::relaxed); } -private: - vk::DescriptorSetLayout CreateDescriptorSetLayout( - vk::Span<VkDescriptorSetLayoutBinding> bindings) const; + template <typename Spec> + static auto MakeConfigureSpecFunc() { + return [](GraphicsPipeline* pl, bool is_indexed) { pl->ConfigureImpl<Spec>(is_indexed); }; + } - vk::PipelineLayout CreatePipelineLayout() const; +private: + template <typename Spec> + void ConfigureImpl(bool is_indexed); - vk::DescriptorUpdateTemplateKHR CreateDescriptorUpdateTemplate( - const SPIRVProgram& program) const; + void ConfigureDraw(); - std::vector<vk::ShaderModule> CreateShaderModules(const SPIRVProgram& program) const; + void MakePipeline(VkRenderPass render_pass); - vk::Pipeline CreatePipeline(const SPIRVProgram& program, VkRenderPass renderpass, - u32 num_color_buffers) const; + void Validate(); + const GraphicsPipelineCacheKey key; + Tegra::Engines::Maxwell3D& maxwell3d; + Tegra::MemoryManager& gpu_memory; const Device& device; + TextureCache& texture_cache; + BufferCache& buffer_cache; VKScheduler& scheduler; - const GraphicsPipelineCacheKey cache_key; - const u64 hash; + VKUpdateDescriptorQueue& update_descriptor_queue; + + void (*configure_func)(GraphicsPipeline*, bool){}; + + std::vector<GraphicsPipelineCacheKey> transition_keys; + std::vector<GraphicsPipeline*> transitions; + + std::array<vk::ShaderModule, NUM_STAGES> spv_modules; + + std::array<Shader::Info, NUM_STAGES> stage_infos; + std::array<u32, 5> enabled_uniform_buffer_masks{}; + VideoCommon::UniformBufferSizes uniform_buffer_sizes{}; vk::DescriptorSetLayout descriptor_set_layout; DescriptorAllocator descriptor_allocator; - VKUpdateDescriptorQueue& update_descriptor_queue; - vk::PipelineLayout layout; - vk::DescriptorUpdateTemplateKHR descriptor_template; - std::vector<vk::ShaderModule> modules; - + vk::PipelineLayout pipeline_layout; + vk::DescriptorUpdateTemplateKHR descriptor_update_template; vk::Pipeline pipeline; + + std::condition_variable build_condvar; + std::mutex build_mutex; + std::atomic_bool is_built{false}; + bool uses_push_descriptor{false}; }; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_master_semaphore.h b/src/video_core/renderer_vulkan/vk_master_semaphore.h index ee3cd35d0..4f8688118 100644 --- a/src/video_core/renderer_vulkan/vk_master_semaphore.h +++ b/src/video_core/renderer_vulkan/vk_master_semaphore.h @@ -39,9 +39,9 @@ public: return KnownGpuTick() >= tick; } - /// Advance to the logical tick. - void NextTick() noexcept { - ++current_tick; + /// Advance to the logical tick and return the old one + [[nodiscard]] u64 NextTick() noexcept { + return current_tick.fetch_add(1, std::memory_order::relaxed); } /// Refresh the known GPU tick diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index 8991505ca..57b163247 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -4,444 +4,613 @@ #include <algorithm> #include <cstddef> +#include <fstream> #include <memory> +#include <thread> #include <vector> #include "common/bit_cast.h" #include "common/cityhash.h" +#include "common/fs/fs.h" +#include "common/fs/path_util.h" #include "common/microprofile.h" +#include "common/thread_worker.h" #include "core/core.h" #include "core/memory.h" +#include "shader_recompiler/backend/spirv/emit_spirv.h" +#include "shader_recompiler/environment.h" +#include "shader_recompiler/frontend/maxwell/control_flow.h" +#include "shader_recompiler/frontend/maxwell/translate_program.h" +#include "shader_recompiler/program_header.h" +#include "video_core/dirty_flags.h" #include "video_core/engines/kepler_compute.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/memory_manager.h" #include "video_core/renderer_vulkan/fixed_pipeline_state.h" #include "video_core/renderer_vulkan/maxwell_to_vk.h" +#include "video_core/renderer_vulkan/pipeline_helper.h" #include "video_core/renderer_vulkan/vk_compute_pipeline.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" -#include "video_core/renderer_vulkan/vk_graphics_pipeline.h" #include "video_core/renderer_vulkan/vk_pipeline_cache.h" #include "video_core/renderer_vulkan/vk_rasterizer.h" #include "video_core/renderer_vulkan/vk_scheduler.h" +#include "video_core/renderer_vulkan/vk_shader_util.h" #include "video_core/renderer_vulkan/vk_update_descriptor.h" -#include "video_core/shader/compiler_settings.h" -#include "video_core/shader/memory_util.h" #include "video_core/shader_cache.h" +#include "video_core/shader_environment.h" #include "video_core/shader_notify.h" #include "video_core/vulkan_common/vulkan_device.h" #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { - MICROPROFILE_DECLARE(Vulkan_PipelineCache); -using Tegra::Engines::ShaderType; -using VideoCommon::Shader::GetShaderAddress; -using VideoCommon::Shader::GetShaderCode; -using VideoCommon::Shader::KERNEL_MAIN_OFFSET; -using VideoCommon::Shader::ProgramCode; -using VideoCommon::Shader::STAGE_MAIN_OFFSET; - namespace { +using Shader::Backend::SPIRV::EmitSPIRV; +using Shader::Maxwell::MergeDualVertexPrograms; +using Shader::Maxwell::TranslateProgram; +using VideoCommon::ComputeEnvironment; +using VideoCommon::FileEnvironment; +using VideoCommon::GenericEnvironment; +using VideoCommon::GraphicsEnvironment; + +constexpr u32 CACHE_VERSION = 5; + +template <typename Container> +auto MakeSpan(Container& container) { + return std::span(container.data(), container.size()); +} -constexpr VkDescriptorType UNIFORM_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER; -constexpr VkDescriptorType STORAGE_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER; -constexpr VkDescriptorType UNIFORM_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER; -constexpr VkDescriptorType COMBINED_IMAGE_SAMPLER = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; -constexpr VkDescriptorType STORAGE_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER; -constexpr VkDescriptorType STORAGE_IMAGE = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE; - -constexpr VideoCommon::Shader::CompilerSettings compiler_settings{ - .depth = VideoCommon::Shader::CompileDepth::FullDecompile, - .disable_else_derivation = true, -}; - -constexpr std::size_t GetStageFromProgram(std::size_t program) { - return program == 0 ? 0 : program - 1; +Shader::CompareFunction MaxwellToCompareFunction(Maxwell::ComparisonOp comparison) { + switch (comparison) { + case Maxwell::ComparisonOp::Never: + case Maxwell::ComparisonOp::NeverOld: + return Shader::CompareFunction::Never; + case Maxwell::ComparisonOp::Less: + case Maxwell::ComparisonOp::LessOld: + return Shader::CompareFunction::Less; + case Maxwell::ComparisonOp::Equal: + case Maxwell::ComparisonOp::EqualOld: + return Shader::CompareFunction::Equal; + case Maxwell::ComparisonOp::LessEqual: + case Maxwell::ComparisonOp::LessEqualOld: + return Shader::CompareFunction::LessThanEqual; + case Maxwell::ComparisonOp::Greater: + case Maxwell::ComparisonOp::GreaterOld: + return Shader::CompareFunction::Greater; + case Maxwell::ComparisonOp::NotEqual: + case Maxwell::ComparisonOp::NotEqualOld: + return Shader::CompareFunction::NotEqual; + case Maxwell::ComparisonOp::GreaterEqual: + case Maxwell::ComparisonOp::GreaterEqualOld: + return Shader::CompareFunction::GreaterThanEqual; + case Maxwell::ComparisonOp::Always: + case Maxwell::ComparisonOp::AlwaysOld: + return Shader::CompareFunction::Always; + } + UNIMPLEMENTED_MSG("Unimplemented comparison op={}", comparison); + return {}; } -constexpr ShaderType GetStageFromProgram(Maxwell::ShaderProgram program) { - return static_cast<ShaderType>(GetStageFromProgram(static_cast<std::size_t>(program))); +Shader::AttributeType CastAttributeType(const FixedPipelineState::VertexAttribute& attr) { + if (attr.enabled == 0) { + return Shader::AttributeType::Disabled; + } + switch (attr.Type()) { + case Maxwell::VertexAttribute::Type::SignedNorm: + case Maxwell::VertexAttribute::Type::UnsignedNorm: + case Maxwell::VertexAttribute::Type::UnsignedScaled: + case Maxwell::VertexAttribute::Type::SignedScaled: + case Maxwell::VertexAttribute::Type::Float: + return Shader::AttributeType::Float; + case Maxwell::VertexAttribute::Type::SignedInt: + return Shader::AttributeType::SignedInt; + case Maxwell::VertexAttribute::Type::UnsignedInt: + return Shader::AttributeType::UnsignedInt; + } + return Shader::AttributeType::Float; } -ShaderType GetShaderType(Maxwell::ShaderProgram program) { - switch (program) { - case Maxwell::ShaderProgram::VertexB: - return ShaderType::Vertex; - case Maxwell::ShaderProgram::TesselationControl: - return ShaderType::TesselationControl; - case Maxwell::ShaderProgram::TesselationEval: - return ShaderType::TesselationEval; - case Maxwell::ShaderProgram::Geometry: - return ShaderType::Geometry; - case Maxwell::ShaderProgram::Fragment: - return ShaderType::Fragment; - default: - UNIMPLEMENTED_MSG("program={}", program); - return ShaderType::Vertex; +Shader::AttributeType AttributeType(const FixedPipelineState& state, size_t index) { + switch (state.DynamicAttributeType(index)) { + case 0: + return Shader::AttributeType::Disabled; + case 1: + return Shader::AttributeType::Float; + case 2: + return Shader::AttributeType::SignedInt; + case 3: + return Shader::AttributeType::UnsignedInt; } + return Shader::AttributeType::Disabled; } -template <VkDescriptorType descriptor_type, class Container> -void AddBindings(std::vector<VkDescriptorSetLayoutBinding>& bindings, u32& binding, - VkShaderStageFlags stage_flags, const Container& container) { - const u32 num_entries = static_cast<u32>(std::size(container)); - for (std::size_t i = 0; i < num_entries; ++i) { - u32 count = 1; - if constexpr (descriptor_type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) { - // Combined image samplers can be arrayed. - count = container[i].size; +Shader::RuntimeInfo MakeRuntimeInfo(std::span<const Shader::IR::Program> programs, + const GraphicsPipelineCacheKey& key, + const Shader::IR::Program& program, + const Shader::IR::Program* previous_program) { + Shader::RuntimeInfo info; + if (previous_program) { + info.previous_stage_stores = previous_program->info.stores; + if (previous_program->is_geometry_passthrough) { + info.previous_stage_stores.mask |= previous_program->info.passthrough.mask; } - bindings.push_back({ - .binding = binding++, - .descriptorType = descriptor_type, - .descriptorCount = count, - .stageFlags = stage_flags, - .pImmutableSamplers = nullptr, - }); + } else { + info.previous_stage_stores.mask.set(); + } + const Shader::Stage stage{program.stage}; + const bool has_geometry{key.unique_hashes[4] != 0 && !programs[4].is_geometry_passthrough}; + const bool gl_ndc{key.state.ndc_minus_one_to_one != 0}; + const float point_size{Common::BitCast<float>(key.state.point_size)}; + switch (stage) { + case Shader::Stage::VertexB: + if (!has_geometry) { + if (key.state.topology == Maxwell::PrimitiveTopology::Points) { + info.fixed_state_point_size = point_size; + } + if (key.state.xfb_enabled) { + info.xfb_varyings = VideoCommon::MakeTransformFeedbackVaryings(key.state.xfb_state); + } + info.convert_depth_mode = gl_ndc; + } + if (key.state.dynamic_vertex_input) { + for (size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) { + info.generic_input_types[index] = AttributeType(key.state, index); + } + } else { + std::ranges::transform(key.state.attributes, info.generic_input_types.begin(), + &CastAttributeType); + } + break; + case Shader::Stage::TessellationEval: + // We have to flip tessellation clockwise for some reason... + info.tess_clockwise = key.state.tessellation_clockwise == 0; + info.tess_primitive = [&key] { + const u32 raw{key.state.tessellation_primitive.Value()}; + switch (static_cast<Maxwell::TessellationPrimitive>(raw)) { + case Maxwell::TessellationPrimitive::Isolines: + return Shader::TessPrimitive::Isolines; + case Maxwell::TessellationPrimitive::Triangles: + return Shader::TessPrimitive::Triangles; + case Maxwell::TessellationPrimitive::Quads: + return Shader::TessPrimitive::Quads; + } + UNREACHABLE(); + return Shader::TessPrimitive::Triangles; + }(); + info.tess_spacing = [&] { + const u32 raw{key.state.tessellation_spacing}; + switch (static_cast<Maxwell::TessellationSpacing>(raw)) { + case Maxwell::TessellationSpacing::Equal: + return Shader::TessSpacing::Equal; + case Maxwell::TessellationSpacing::FractionalOdd: + return Shader::TessSpacing::FractionalOdd; + case Maxwell::TessellationSpacing::FractionalEven: + return Shader::TessSpacing::FractionalEven; + } + UNREACHABLE(); + return Shader::TessSpacing::Equal; + }(); + break; + case Shader::Stage::Geometry: + if (program.output_topology == Shader::OutputTopology::PointList) { + info.fixed_state_point_size = point_size; + } + if (key.state.xfb_enabled != 0) { + info.xfb_varyings = VideoCommon::MakeTransformFeedbackVaryings(key.state.xfb_state); + } + info.convert_depth_mode = gl_ndc; + break; + case Shader::Stage::Fragment: + info.alpha_test_func = MaxwellToCompareFunction( + key.state.UnpackComparisonOp(key.state.alpha_test_func.Value())); + info.alpha_test_reference = Common::BitCast<float>(key.state.alpha_test_ref); + break; + default: + break; + } + switch (key.state.topology) { + case Maxwell::PrimitiveTopology::Points: + info.input_topology = Shader::InputTopology::Points; + break; + case Maxwell::PrimitiveTopology::Lines: + case Maxwell::PrimitiveTopology::LineLoop: + case Maxwell::PrimitiveTopology::LineStrip: + info.input_topology = Shader::InputTopology::Lines; + break; + case Maxwell::PrimitiveTopology::Triangles: + case Maxwell::PrimitiveTopology::TriangleStrip: + case Maxwell::PrimitiveTopology::TriangleFan: + case Maxwell::PrimitiveTopology::Quads: + case Maxwell::PrimitiveTopology::QuadStrip: + case Maxwell::PrimitiveTopology::Polygon: + case Maxwell::PrimitiveTopology::Patches: + info.input_topology = Shader::InputTopology::Triangles; + break; + case Maxwell::PrimitiveTopology::LinesAdjacency: + case Maxwell::PrimitiveTopology::LineStripAdjacency: + info.input_topology = Shader::InputTopology::LinesAdjacency; + break; + case Maxwell::PrimitiveTopology::TrianglesAdjacency: + case Maxwell::PrimitiveTopology::TriangleStripAdjacency: + info.input_topology = Shader::InputTopology::TrianglesAdjacency; + break; } + info.force_early_z = key.state.early_z != 0; + info.y_negate = key.state.y_negate != 0; + return info; } +} // Anonymous namespace -u32 FillDescriptorLayout(const ShaderEntries& entries, - std::vector<VkDescriptorSetLayoutBinding>& bindings, - Maxwell::ShaderProgram program_type, u32 base_binding) { - const ShaderType stage = GetStageFromProgram(program_type); - const VkShaderStageFlags flags = MaxwellToVK::ShaderStage(stage); - - u32 binding = base_binding; - AddBindings<UNIFORM_BUFFER>(bindings, binding, flags, entries.const_buffers); - AddBindings<STORAGE_BUFFER>(bindings, binding, flags, entries.global_buffers); - AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.uniform_texels); - AddBindings<COMBINED_IMAGE_SAMPLER>(bindings, binding, flags, entries.samplers); - AddBindings<STORAGE_TEXEL_BUFFER>(bindings, binding, flags, entries.storage_texels); - AddBindings<STORAGE_IMAGE>(bindings, binding, flags, entries.images); - return binding; +size_t ComputePipelineCacheKey::Hash() const noexcept { + const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), sizeof *this); + return static_cast<size_t>(hash); } -} // Anonymous namespace +bool ComputePipelineCacheKey::operator==(const ComputePipelineCacheKey& rhs) const noexcept { + return std::memcmp(&rhs, this, sizeof *this) == 0; +} -std::size_t GraphicsPipelineCacheKey::Hash() const noexcept { +size_t GraphicsPipelineCacheKey::Hash() const noexcept { const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), Size()); - return static_cast<std::size_t>(hash); + return static_cast<size_t>(hash); } bool GraphicsPipelineCacheKey::operator==(const GraphicsPipelineCacheKey& rhs) const noexcept { return std::memcmp(&rhs, this, Size()) == 0; } -std::size_t ComputePipelineCacheKey::Hash() const noexcept { - const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), sizeof *this); - return static_cast<std::size_t>(hash); -} - -bool ComputePipelineCacheKey::operator==(const ComputePipelineCacheKey& rhs) const noexcept { - return std::memcmp(&rhs, this, sizeof *this) == 0; +PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxwell3D& maxwell3d_, + Tegra::Engines::KeplerCompute& kepler_compute_, + Tegra::MemoryManager& gpu_memory_, const Device& device_, + VKScheduler& scheduler_, DescriptorPool& descriptor_pool_, + VKUpdateDescriptorQueue& update_descriptor_queue_, + RenderPassCache& render_pass_cache_, BufferCache& buffer_cache_, + TextureCache& texture_cache_, VideoCore::ShaderNotify& shader_notify_) + : VideoCommon::ShaderCache{rasterizer_, gpu_memory_, maxwell3d_, kepler_compute_}, + device{device_}, scheduler{scheduler_}, descriptor_pool{descriptor_pool_}, + update_descriptor_queue{update_descriptor_queue_}, render_pass_cache{render_pass_cache_}, + buffer_cache{buffer_cache_}, texture_cache{texture_cache_}, shader_notify{shader_notify_}, + use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()}, + workers(std::max(std::thread::hardware_concurrency(), 2U) - 1, "yuzu:PipelineBuilder"), + serialization_thread(1, "yuzu:PipelineSerialization") { + const auto& float_control{device.FloatControlProperties()}; + const VkDriverIdKHR driver_id{device.GetDriverID()}; + profile = Shader::Profile{ + .supported_spirv = device.IsKhrSpirv1_4Supported() ? 0x00010400U : 0x00010000U, + .unified_descriptor_binding = true, + .support_descriptor_aliasing = true, + .support_int8 = true, + .support_int16 = device.IsShaderInt16Supported(), + .support_int64 = device.IsShaderInt64Supported(), + .support_vertex_instance_id = false, + .support_float_controls = true, + .support_separate_denorm_behavior = float_control.denormBehaviorIndependence == + VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_ALL_KHR, + .support_separate_rounding_mode = + float_control.roundingModeIndependence == VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_ALL_KHR, + .support_fp16_denorm_preserve = float_control.shaderDenormPreserveFloat16 != VK_FALSE, + .support_fp32_denorm_preserve = float_control.shaderDenormPreserveFloat32 != VK_FALSE, + .support_fp16_denorm_flush = float_control.shaderDenormFlushToZeroFloat16 != VK_FALSE, + .support_fp32_denorm_flush = float_control.shaderDenormFlushToZeroFloat32 != VK_FALSE, + .support_fp16_signed_zero_nan_preserve = + float_control.shaderSignedZeroInfNanPreserveFloat16 != VK_FALSE, + .support_fp32_signed_zero_nan_preserve = + float_control.shaderSignedZeroInfNanPreserveFloat32 != VK_FALSE, + .support_fp64_signed_zero_nan_preserve = + float_control.shaderSignedZeroInfNanPreserveFloat64 != VK_FALSE, + .support_explicit_workgroup_layout = device.IsKhrWorkgroupMemoryExplicitLayoutSupported(), + .support_vote = true, + .support_viewport_index_layer_non_geometry = + device.IsExtShaderViewportIndexLayerSupported(), + .support_viewport_mask = device.IsNvViewportArray2Supported(), + .support_typeless_image_loads = device.IsFormatlessImageLoadSupported(), + .support_demote_to_helper_invocation = true, + .support_int64_atomics = device.IsExtShaderAtomicInt64Supported(), + .support_derivative_control = true, + .support_geometry_shader_passthrough = device.IsNvGeometryShaderPassthroughSupported(), + + .warp_size_potentially_larger_than_guest = device.IsWarpSizePotentiallyBiggerThanGuest(), + + .lower_left_origin_mode = false, + .need_declared_frag_colors = false, + + .has_broken_spirv_clamp = driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS_KHR, + .has_broken_unsigned_image_offsets = false, + .has_broken_signed_operations = false, + .has_broken_fp16_float_controls = driver_id == VK_DRIVER_ID_NVIDIA_PROPRIETARY_KHR, + .ignore_nan_fp_comparisons = false, + }; + host_info = Shader::HostTranslateInfo{ + .support_float16 = device.IsFloat16Supported(), + .support_int64 = device.IsShaderInt64Supported(), + }; } -Shader::Shader(Tegra::Engines::ConstBufferEngineInterface& engine_, ShaderType stage_, - GPUVAddr gpu_addr_, VAddr cpu_addr_, ProgramCode program_code_, u32 main_offset_) - : gpu_addr(gpu_addr_), program_code(std::move(program_code_)), registry(stage_, engine_), - shader_ir(program_code, main_offset_, compiler_settings, registry), - entries(GenerateShaderEntries(shader_ir)) {} - -Shader::~Shader() = default; - -VKPipelineCache::VKPipelineCache(RasterizerVulkan& rasterizer_, Tegra::GPU& gpu_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, const Device& device_, - VKScheduler& scheduler_, VKDescriptorPool& descriptor_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_) - : VideoCommon::ShaderCache<Shader>{rasterizer_}, gpu{gpu_}, maxwell3d{maxwell3d_}, - kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_}, device{device_}, - scheduler{scheduler_}, descriptor_pool{descriptor_pool_}, update_descriptor_queue{ - update_descriptor_queue_} {} - -VKPipelineCache::~VKPipelineCache() = default; +PipelineCache::~PipelineCache() = default; -std::array<Shader*, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() { - std::array<Shader*, Maxwell::MaxShaderProgram> shaders{}; - - for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { - const auto program{static_cast<Maxwell::ShaderProgram>(index)}; - - // Skip stages that are not enabled - if (!maxwell3d.regs.IsShaderConfigEnabled(index)) { - continue; - } - - const GPUVAddr gpu_addr{GetShaderAddress(maxwell3d, program)}; - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); - ASSERT(cpu_addr); - - Shader* result = cpu_addr ? TryGet(*cpu_addr) : null_shader.get(); - if (!result) { - const u8* const host_ptr{gpu_memory.GetPointer(gpu_addr)}; - - // No shader found - create a new one - static constexpr u32 stage_offset = STAGE_MAIN_OFFSET; - const auto stage = static_cast<ShaderType>(index == 0 ? 0 : index - 1); - ProgramCode code = GetShaderCode(gpu_memory, gpu_addr, host_ptr, false); - const std::size_t size_in_bytes = code.size() * sizeof(u64); - - auto shader = std::make_unique<Shader>(maxwell3d, stage, gpu_addr, *cpu_addr, - std::move(code), stage_offset); - result = shader.get(); +GraphicsPipeline* PipelineCache::CurrentGraphicsPipeline() { + MICROPROFILE_SCOPE(Vulkan_PipelineCache); - if (cpu_addr) { - Register(std::move(shader), *cpu_addr, size_in_bytes); - } else { - null_shader = std::move(shader); - } + if (!RefreshStages(graphics_key.unique_hashes)) { + current_pipeline = nullptr; + return nullptr; + } + graphics_key.state.Refresh(maxwell3d, device.IsExtExtendedDynamicStateSupported(), + device.IsExtVertexInputDynamicStateSupported()); + + if (current_pipeline) { + GraphicsPipeline* const next{current_pipeline->Next(graphics_key)}; + if (next) { + current_pipeline = next; + return BuiltPipeline(current_pipeline); } - shaders[index] = result; } - return last_shaders = shaders; + return CurrentGraphicsPipelineSlowPath(); } -VKGraphicsPipeline* VKPipelineCache::GetGraphicsPipeline( - const GraphicsPipelineCacheKey& key, u32 num_color_buffers, - VideoCommon::Shader::AsyncShaders& async_shaders) { +ComputePipeline* PipelineCache::CurrentComputePipeline() { MICROPROFILE_SCOPE(Vulkan_PipelineCache); - if (last_graphics_pipeline && last_graphics_key == key) { - return last_graphics_pipeline; - } - last_graphics_key = key; - - if (device.UseAsynchronousShaders() && async_shaders.IsShaderAsync(gpu)) { - std::unique_lock lock{pipeline_cache}; - const auto [pair, is_cache_miss] = graphics_cache.try_emplace(key); - if (is_cache_miss) { - gpu.ShaderNotify().MarkSharderBuilding(); - LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash()); - const auto [program, bindings] = DecompileShaders(key.fixed_state); - async_shaders.QueueVulkanShader(this, device, scheduler, descriptor_pool, - update_descriptor_queue, bindings, program, key, - num_color_buffers); - } - last_graphics_pipeline = pair->second.get(); - return last_graphics_pipeline; + const ShaderInfo* const shader{ComputeShader()}; + if (!shader) { + return nullptr; } - - const auto [pair, is_cache_miss] = graphics_cache.try_emplace(key); - auto& entry = pair->second; - if (is_cache_miss) { - gpu.ShaderNotify().MarkSharderBuilding(); - LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash()); - const auto [program, bindings] = DecompileShaders(key.fixed_state); - entry = std::make_unique<VKGraphicsPipeline>(device, scheduler, descriptor_pool, - update_descriptor_queue, key, bindings, - program, num_color_buffers); - gpu.ShaderNotify().MarkShaderComplete(); + const auto& qmd{kepler_compute.launch_description}; + const ComputePipelineCacheKey key{ + .unique_hash = shader->unique_hash, + .shared_memory_size = qmd.shared_alloc, + .workgroup_size{qmd.block_dim_x, qmd.block_dim_y, qmd.block_dim_z}, + }; + const auto [pair, is_new]{compute_cache.try_emplace(key)}; + auto& pipeline{pair->second}; + if (!is_new) { + return pipeline.get(); } - last_graphics_pipeline = entry.get(); - return last_graphics_pipeline; + pipeline = CreateComputePipeline(key, shader); + return pipeline.get(); } -VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCacheKey& key) { - MICROPROFILE_SCOPE(Vulkan_PipelineCache); - - const auto [pair, is_cache_miss] = compute_cache.try_emplace(key); - auto& entry = pair->second; - if (!is_cache_miss) { - return *entry; +void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading, + const VideoCore::DiskResourceLoadCallback& callback) { + if (title_id == 0) { + return; } - LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash()); - - const GPUVAddr gpu_addr = key.shader; - - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); - ASSERT(cpu_addr); + const auto shader_dir{Common::FS::GetYuzuPath(Common::FS::YuzuPath::ShaderDir)}; + const auto base_dir{shader_dir / fmt::format("{:016x}", title_id)}; + if (!Common::FS::CreateDir(shader_dir) || !Common::FS::CreateDir(base_dir)) { + LOG_ERROR(Common_Filesystem, "Failed to create pipeline cache directories"); + return; + } + pipeline_cache_filename = base_dir / "vulkan.bin"; + + struct { + std::mutex mutex; + size_t total{}; + size_t built{}; + bool has_loaded{}; + } state; + + const auto load_compute{[&](std::ifstream& file, FileEnvironment env) { + ComputePipelineCacheKey key; + file.read(reinterpret_cast<char*>(&key), sizeof(key)); + + workers.QueueWork([this, key, env = std::move(env), &state, &callback]() mutable { + ShaderPools pools; + auto pipeline{CreateComputePipeline(pools, key, env, false)}; + std::lock_guard lock{state.mutex}; + if (pipeline) { + compute_cache.emplace(key, std::move(pipeline)); + } + ++state.built; + if (state.has_loaded) { + callback(VideoCore::LoadCallbackStage::Build, state.built, state.total); + } + }); + ++state.total; + }}; + const bool extended_dynamic_state = device.IsExtExtendedDynamicStateSupported(); + const bool dynamic_vertex_input = device.IsExtVertexInputDynamicStateSupported(); + const auto load_graphics{[&](std::ifstream& file, std::vector<FileEnvironment> envs) { + GraphicsPipelineCacheKey key; + file.read(reinterpret_cast<char*>(&key), sizeof(key)); + + if ((key.state.extended_dynamic_state != 0) != extended_dynamic_state || + (key.state.dynamic_vertex_input != 0) != dynamic_vertex_input) { + return; + } + workers.QueueWork([this, key, envs = std::move(envs), &state, &callback]() mutable { + ShaderPools pools; + boost::container::static_vector<Shader::Environment*, 5> env_ptrs; + for (auto& env : envs) { + env_ptrs.push_back(&env); + } + auto pipeline{CreateGraphicsPipeline(pools, key, MakeSpan(env_ptrs), false)}; - Shader* shader = cpu_addr ? TryGet(*cpu_addr) : null_kernel.get(); - if (!shader) { - // No shader found - create a new one - const auto host_ptr = gpu_memory.GetPointer(gpu_addr); + std::lock_guard lock{state.mutex}; + graphics_cache.emplace(key, std::move(pipeline)); + ++state.built; + if (state.has_loaded) { + callback(VideoCore::LoadCallbackStage::Build, state.built, state.total); + } + }); + ++state.total; + }}; + VideoCommon::LoadPipelines(stop_loading, pipeline_cache_filename, CACHE_VERSION, load_compute, + load_graphics); - ProgramCode code = GetShaderCode(gpu_memory, gpu_addr, host_ptr, true); - const std::size_t size_in_bytes = code.size() * sizeof(u64); + std::unique_lock lock{state.mutex}; + callback(VideoCore::LoadCallbackStage::Build, 0, state.total); + state.has_loaded = true; + lock.unlock(); - auto shader_info = std::make_unique<Shader>(kepler_compute, ShaderType::Compute, gpu_addr, - *cpu_addr, std::move(code), KERNEL_MAIN_OFFSET); - shader = shader_info.get(); + workers.WaitForRequests(); +} - if (cpu_addr) { - Register(std::move(shader_info), *cpu_addr, size_in_bytes); - } else { - null_kernel = std::move(shader_info); - } +GraphicsPipeline* PipelineCache::CurrentGraphicsPipelineSlowPath() { + const auto [pair, is_new]{graphics_cache.try_emplace(graphics_key)}; + auto& pipeline{pair->second}; + if (is_new) { + pipeline = CreateGraphicsPipeline(); } - - const Specialization specialization{ - .base_binding = 0, - .workgroup_size = key.workgroup_size, - .shared_memory_size = key.shared_memory_size, - .point_size = std::nullopt, - .enabled_attributes = {}, - .attribute_types = {}, - .ndc_minus_one_to_one = false, - }; - const SPIRVShader spirv_shader{Decompile(device, shader->GetIR(), ShaderType::Compute, - shader->GetRegistry(), specialization), - shader->GetEntries()}; - entry = std::make_unique<VKComputePipeline>(device, scheduler, descriptor_pool, - update_descriptor_queue, spirv_shader); - return *entry; + if (!pipeline) { + return nullptr; + } + if (current_pipeline) { + current_pipeline->AddTransition(pipeline.get()); + } + current_pipeline = pipeline.get(); + return BuiltPipeline(current_pipeline); } -void VKPipelineCache::EmplacePipeline(std::unique_ptr<VKGraphicsPipeline> pipeline) { - gpu.ShaderNotify().MarkShaderComplete(); - std::unique_lock lock{pipeline_cache}; - graphics_cache.at(pipeline->GetCacheKey()) = std::move(pipeline); +GraphicsPipeline* PipelineCache::BuiltPipeline(GraphicsPipeline* pipeline) const noexcept { + if (pipeline->IsBuilt()) { + return pipeline; + } + if (!use_asynchronous_shaders) { + return pipeline; + } + // If something is using depth, we can assume that games are not rendering anything which + // will be used one time. + if (maxwell3d.regs.zeta_enable) { + return nullptr; + } + // If games are using a small index count, we can assume these are full screen quads. + // Usually these shaders are only used once for building textures so we can assume they + // can't be built async + if (maxwell3d.regs.index_array.count <= 6 || maxwell3d.regs.vertex_buffer.count <= 6) { + return pipeline; + } + return nullptr; } -void VKPipelineCache::OnShaderRemoval(Shader* shader) { - bool finished = false; - const auto Finish = [&] { - // TODO(Rodrigo): Instead of finishing here, wait for the fences that use this pipeline and - // flush. - if (finished) { - return; - } - finished = true; - scheduler.Finish(); - }; - - const GPUVAddr invalidated_addr = shader->GetGpuAddr(); - for (auto it = graphics_cache.begin(); it != graphics_cache.end();) { - auto& entry = it->first; - if (std::find(entry.shaders.begin(), entry.shaders.end(), invalidated_addr) == - entry.shaders.end()) { - ++it; +std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline( + ShaderPools& pools, const GraphicsPipelineCacheKey& key, + std::span<Shader::Environment* const> envs, bool build_in_parallel) try { + LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash()); + size_t env_index{0}; + std::array<Shader::IR::Program, Maxwell::MaxShaderProgram> programs; + const bool uses_vertex_a{key.unique_hashes[0] != 0}; + const bool uses_vertex_b{key.unique_hashes[1] != 0}; + for (size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { + if (key.unique_hashes[index] == 0) { continue; } - Finish(); - it = graphics_cache.erase(it); + Shader::Environment& env{*envs[env_index]}; + ++env_index; + + const u32 cfg_offset{static_cast<u32>(env.StartAddress() + sizeof(Shader::ProgramHeader))}; + Shader::Maxwell::Flow::CFG cfg(env, pools.flow_block, cfg_offset, index == 0); + if (!uses_vertex_a || index != 1) { + // Normal path + programs[index] = TranslateProgram(pools.inst, pools.block, env, cfg, host_info); + } else { + // VertexB path when VertexA is present. + auto& program_va{programs[0]}; + auto program_vb{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)}; + programs[index] = MergeDualVertexPrograms(program_va, program_vb, env); + } } - for (auto it = compute_cache.begin(); it != compute_cache.end();) { - auto& entry = it->first; - if (entry.shader != invalidated_addr) { - ++it; + std::array<const Shader::Info*, Maxwell::MaxShaderStage> infos{}; + std::array<vk::ShaderModule, Maxwell::MaxShaderStage> modules; + + const Shader::IR::Program* previous_stage{}; + Shader::Backend::Bindings binding; + for (size_t index = uses_vertex_a && uses_vertex_b ? 1 : 0; index < Maxwell::MaxShaderProgram; + ++index) { + if (key.unique_hashes[index] == 0) { continue; } - Finish(); - it = compute_cache.erase(it); + UNIMPLEMENTED_IF(index == 0); + + Shader::IR::Program& program{programs[index]}; + const size_t stage_index{index - 1}; + infos[stage_index] = &program.info; + + const auto runtime_info{MakeRuntimeInfo(programs, key, program, previous_stage)}; + const std::vector<u32> code{EmitSPIRV(profile, runtime_info, program, binding)}; + device.SaveShader(code); + modules[stage_index] = BuildShader(device, code); + if (device.HasDebuggingToolAttached()) { + const std::string name{fmt::format("Shader {:016x}", key.unique_hashes[index])}; + modules[stage_index].SetObjectNameEXT(name.c_str()); + } + previous_stage = &program; } + Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr}; + return std::make_unique<GraphicsPipeline>( + maxwell3d, gpu_memory, scheduler, buffer_cache, texture_cache, &shader_notify, device, + descriptor_pool, update_descriptor_queue, thread_worker, render_pass_cache, key, + std::move(modules), infos); + +} catch (const Shader::Exception& exception) { + LOG_ERROR(Render_Vulkan, "{}", exception.what()); + return nullptr; } -std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>> -VKPipelineCache::DecompileShaders(const FixedPipelineState& fixed_state) { - Specialization specialization; - if (fixed_state.topology == Maxwell::PrimitiveTopology::Points) { - float point_size; - std::memcpy(&point_size, &fixed_state.point_size, sizeof(float)); - specialization.point_size = point_size; - ASSERT(point_size != 0.0f); - } - for (std::size_t i = 0; i < Maxwell::NumVertexAttributes; ++i) { - const auto& attribute = fixed_state.attributes[i]; - specialization.enabled_attributes[i] = attribute.enabled.Value() != 0; - specialization.attribute_types[i] = attribute.Type(); - } - specialization.ndc_minus_one_to_one = fixed_state.ndc_minus_one_to_one; - specialization.early_fragment_tests = fixed_state.early_z; - - // Alpha test - specialization.alpha_test_func = - FixedPipelineState::UnpackComparisonOp(fixed_state.alpha_test_func.Value()); - specialization.alpha_test_ref = Common::BitCast<float>(fixed_state.alpha_test_ref); - - SPIRVProgram program; - std::vector<VkDescriptorSetLayoutBinding> bindings; +std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline() { + GraphicsEnvironments environments; + GetGraphicsEnvironments(environments, graphics_key.unique_hashes); - for (std::size_t index = 1; index < Maxwell::MaxShaderProgram; ++index) { - const auto program_enum = static_cast<Maxwell::ShaderProgram>(index); - // Skip stages that are not enabled - if (!maxwell3d.regs.IsShaderConfigEnabled(index)) { - continue; - } - const GPUVAddr gpu_addr = GetShaderAddress(maxwell3d, program_enum); - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); - Shader* const shader = cpu_addr ? TryGet(*cpu_addr) : null_shader.get(); - - const std::size_t stage = index == 0 ? 0 : index - 1; // Stage indices are 0 - 5 - const ShaderType program_type = GetShaderType(program_enum); - const auto& entries = shader->GetEntries(); - program[stage] = { - Decompile(device, shader->GetIR(), program_type, shader->GetRegistry(), specialization), - entries, - }; - - const u32 old_binding = specialization.base_binding; - specialization.base_binding = - FillDescriptorLayout(entries, bindings, program_enum, specialization.base_binding); - ASSERT(old_binding + entries.NumBindings() == specialization.base_binding); + main_pools.ReleaseContents(); + auto pipeline{CreateGraphicsPipeline(main_pools, graphics_key, environments.Span(), true)}; + if (!pipeline || pipeline_cache_filename.empty()) { + return pipeline; } - return {std::move(program), std::move(bindings)}; -} - -template <VkDescriptorType descriptor_type, class Container> -void AddEntry(std::vector<VkDescriptorUpdateTemplateEntry>& template_entries, u32& binding, - u32& offset, const Container& container) { - static constexpr u32 entry_size = static_cast<u32>(sizeof(DescriptorUpdateEntry)); - const u32 count = static_cast<u32>(std::size(container)); - - if constexpr (descriptor_type == COMBINED_IMAGE_SAMPLER) { - for (u32 i = 0; i < count; ++i) { - const u32 num_samplers = container[i].size; - template_entries.push_back({ - .dstBinding = binding, - .dstArrayElement = 0, - .descriptorCount = num_samplers, - .descriptorType = descriptor_type, - .offset = offset, - .stride = entry_size, - }); - - ++binding; - offset += num_samplers * entry_size; + serialization_thread.QueueWork([this, key = graphics_key, envs = std::move(environments.envs)] { + boost::container::static_vector<const GenericEnvironment*, Maxwell::MaxShaderProgram> + env_ptrs; + for (size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { + if (key.unique_hashes[index] != 0) { + env_ptrs.push_back(&envs[index]); + } } - return; - } + SerializePipeline(key, env_ptrs, pipeline_cache_filename, CACHE_VERSION); + }); + return pipeline; +} - if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER || - descriptor_type == STORAGE_TEXEL_BUFFER) { - // Nvidia has a bug where updating multiple texels at once causes the driver to crash. - // Note: Fixed in driver Windows 443.24, Linux 440.66.15 - for (u32 i = 0; i < count; ++i) { - template_entries.push_back({ - .dstBinding = binding + i, - .dstArrayElement = 0, - .descriptorCount = 1, - .descriptorType = descriptor_type, - .offset = static_cast<std::size_t>(offset + i * entry_size), - .stride = entry_size, - }); - } - } else if (count > 0) { - template_entries.push_back({ - .dstBinding = binding, - .dstArrayElement = 0, - .descriptorCount = count, - .descriptorType = descriptor_type, - .offset = offset, - .stride = entry_size, - }); +std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline( + const ComputePipelineCacheKey& key, const ShaderInfo* shader) { + const GPUVAddr program_base{kepler_compute.regs.code_loc.Address()}; + const auto& qmd{kepler_compute.launch_description}; + ComputeEnvironment env{kepler_compute, gpu_memory, program_base, qmd.program_start}; + env.SetCachedSize(shader->size_bytes); + + main_pools.ReleaseContents(); + auto pipeline{CreateComputePipeline(main_pools, key, env, true)}; + if (!pipeline || pipeline_cache_filename.empty()) { + return pipeline; } - offset += count * entry_size; - binding += count; + serialization_thread.QueueWork([this, key, env = std::move(env)] { + SerializePipeline(key, std::array<const GenericEnvironment*, 1>{&env}, + pipeline_cache_filename, CACHE_VERSION); + }); + return pipeline; } -void FillDescriptorUpdateTemplateEntries( - const ShaderEntries& entries, u32& binding, u32& offset, - std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries) { - AddEntry<UNIFORM_BUFFER>(template_entries, offset, binding, entries.const_buffers); - AddEntry<STORAGE_BUFFER>(template_entries, offset, binding, entries.global_buffers); - AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.uniform_texels); - AddEntry<COMBINED_IMAGE_SAMPLER>(template_entries, offset, binding, entries.samplers); - AddEntry<STORAGE_TEXEL_BUFFER>(template_entries, offset, binding, entries.storage_texels); - AddEntry<STORAGE_IMAGE>(template_entries, offset, binding, entries.images); +std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline( + ShaderPools& pools, const ComputePipelineCacheKey& key, Shader::Environment& env, + bool build_in_parallel) try { + LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash()); + + Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()}; + auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)}; + const std::vector<u32> code{EmitSPIRV(profile, program)}; + device.SaveShader(code); + vk::ShaderModule spv_module{BuildShader(device, code)}; + if (device.HasDebuggingToolAttached()) { + const auto name{fmt::format("Shader {:016x}", key.unique_hash)}; + spv_module.SetObjectNameEXT(name.c_str()); + } + Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr}; + return std::make_unique<ComputePipeline>(device, descriptor_pool, update_descriptor_queue, + thread_worker, &shader_notify, program.info, + std::move(spv_module)); + +} catch (const Shader::Exception& exception) { + LOG_ERROR(Render_Vulkan, "{}", exception.what()); + return nullptr; } } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.h b/src/video_core/renderer_vulkan/vk_pipeline_cache.h index 89d635a3d..efe5a7ed8 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.h +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.h @@ -6,24 +6,28 @@ #include <array> #include <cstddef> +#include <filesystem> +#include <iosfwd> #include <memory> #include <type_traits> #include <unordered_map> #include <utility> #include <vector> -#include <boost/functional/hash.hpp> - #include "common/common_types.h" -#include "video_core/engines/const_buffer_engine_interface.h" +#include "common/thread_worker.h" +#include "shader_recompiler/frontend/ir/basic_block.h" +#include "shader_recompiler/frontend/ir/value.h" +#include "shader_recompiler/frontend/maxwell/control_flow.h" +#include "shader_recompiler/host_translate_info.h" +#include "shader_recompiler/object_pool.h" +#include "shader_recompiler/profile.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/renderer_vulkan/fixed_pipeline_state.h" +#include "video_core/renderer_vulkan/vk_buffer_cache.h" +#include "video_core/renderer_vulkan/vk_compute_pipeline.h" #include "video_core/renderer_vulkan/vk_graphics_pipeline.h" -#include "video_core/renderer_vulkan/vk_shader_decompiler.h" -#include "video_core/shader/async_shaders.h" -#include "video_core/shader/memory_util.h" -#include "video_core/shader/registry.h" -#include "video_core/shader/shader_ir.h" +#include "video_core/renderer_vulkan/vk_texture_cache.h" #include "video_core/shader_cache.h" #include "video_core/vulkan_common/vulkan_wrapper.h" @@ -31,23 +35,24 @@ namespace Core { class System; } -namespace Vulkan { +namespace Shader::IR { +struct Program; +} -class Device; -class RasterizerVulkan; -class VKComputePipeline; -class VKDescriptorPool; -class VKScheduler; -class VKUpdateDescriptorQueue; +namespace VideoCore { +class ShaderNotify; +} + +namespace Vulkan { using Maxwell = Tegra::Engines::Maxwell3D::Regs; struct ComputePipelineCacheKey { - GPUVAddr shader; + u64 unique_hash; u32 shared_memory_size; std::array<u32, 3> workgroup_size; - std::size_t Hash() const noexcept; + size_t Hash() const noexcept; bool operator==(const ComputePipelineCacheKey& rhs) const noexcept; @@ -64,15 +69,8 @@ static_assert(std::is_trivially_constructible_v<ComputePipelineCacheKey>); namespace std { template <> -struct hash<Vulkan::GraphicsPipelineCacheKey> { - std::size_t operator()(const Vulkan::GraphicsPipelineCacheKey& k) const noexcept { - return k.Hash(); - } -}; - -template <> struct hash<Vulkan::ComputePipelineCacheKey> { - std::size_t operator()(const Vulkan::ComputePipelineCacheKey& k) const noexcept { + size_t operator()(const Vulkan::ComputePipelineCacheKey& k) const noexcept { return k.Hash(); } }; @@ -81,94 +79,90 @@ struct hash<Vulkan::ComputePipelineCacheKey> { namespace Vulkan { -class Shader { -public: - explicit Shader(Tegra::Engines::ConstBufferEngineInterface& engine_, - Tegra::Engines::ShaderType stage_, GPUVAddr gpu_addr, VAddr cpu_addr_, - VideoCommon::Shader::ProgramCode program_code, u32 main_offset_); - ~Shader(); - - GPUVAddr GetGpuAddr() const { - return gpu_addr; - } - - VideoCommon::Shader::ShaderIR& GetIR() { - return shader_ir; - } - - const VideoCommon::Shader::ShaderIR& GetIR() const { - return shader_ir; - } +class ComputePipeline; +class Device; +class DescriptorPool; +class RasterizerVulkan; +class RenderPassCache; +class VKScheduler; +class VKUpdateDescriptorQueue; - const VideoCommon::Shader::Registry& GetRegistry() const { - return registry; - } +using VideoCommon::ShaderInfo; - const ShaderEntries& GetEntries() const { - return entries; +struct ShaderPools { + void ReleaseContents() { + flow_block.ReleaseContents(); + block.ReleaseContents(); + inst.ReleaseContents(); } -private: - GPUVAddr gpu_addr{}; - VideoCommon::Shader::ProgramCode program_code; - VideoCommon::Shader::Registry registry; - VideoCommon::Shader::ShaderIR shader_ir; - ShaderEntries entries; + Shader::ObjectPool<Shader::IR::Inst> inst; + Shader::ObjectPool<Shader::IR::Block> block; + Shader::ObjectPool<Shader::Maxwell::Flow::Block> flow_block; }; -class VKPipelineCache final : public VideoCommon::ShaderCache<Shader> { +class PipelineCache : public VideoCommon::ShaderCache { public: - explicit VKPipelineCache(RasterizerVulkan& rasterizer, Tegra::GPU& gpu, - Tegra::Engines::Maxwell3D& maxwell3d, - Tegra::Engines::KeplerCompute& kepler_compute, - Tegra::MemoryManager& gpu_memory, const Device& device, - VKScheduler& scheduler, VKDescriptorPool& descriptor_pool, - VKUpdateDescriptorQueue& update_descriptor_queue); - ~VKPipelineCache() override; + explicit PipelineCache(RasterizerVulkan& rasterizer, Tegra::Engines::Maxwell3D& maxwell3d, + Tegra::Engines::KeplerCompute& kepler_compute, + Tegra::MemoryManager& gpu_memory, const Device& device, + VKScheduler& scheduler, DescriptorPool& descriptor_pool, + VKUpdateDescriptorQueue& update_descriptor_queue, + RenderPassCache& render_pass_cache, BufferCache& buffer_cache, + TextureCache& texture_cache, VideoCore::ShaderNotify& shader_notify_); + ~PipelineCache(); + + [[nodiscard]] GraphicsPipeline* CurrentGraphicsPipeline(); - std::array<Shader*, Maxwell::MaxShaderProgram> GetShaders(); + [[nodiscard]] ComputePipeline* CurrentComputePipeline(); - VKGraphicsPipeline* GetGraphicsPipeline(const GraphicsPipelineCacheKey& key, - u32 num_color_buffers, - VideoCommon::Shader::AsyncShaders& async_shaders); + void LoadDiskResources(u64 title_id, std::stop_token stop_loading, + const VideoCore::DiskResourceLoadCallback& callback); - VKComputePipeline& GetComputePipeline(const ComputePipelineCacheKey& key); +private: + [[nodiscard]] GraphicsPipeline* CurrentGraphicsPipelineSlowPath(); - void EmplacePipeline(std::unique_ptr<VKGraphicsPipeline> pipeline); + [[nodiscard]] GraphicsPipeline* BuiltPipeline(GraphicsPipeline* pipeline) const noexcept; -protected: - void OnShaderRemoval(Shader* shader) final; + std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline(); -private: - std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>> DecompileShaders( - const FixedPipelineState& fixed_state); + std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline( + ShaderPools& pools, const GraphicsPipelineCacheKey& key, + std::span<Shader::Environment* const> envs, bool build_in_parallel); - Tegra::GPU& gpu; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::Engines::KeplerCompute& kepler_compute; - Tegra::MemoryManager& gpu_memory; + std::unique_ptr<ComputePipeline> CreateComputePipeline(const ComputePipelineCacheKey& key, + const ShaderInfo* shader); + + std::unique_ptr<ComputePipeline> CreateComputePipeline(ShaderPools& pools, + const ComputePipelineCacheKey& key, + Shader::Environment& env, + bool build_in_parallel); const Device& device; VKScheduler& scheduler; - VKDescriptorPool& descriptor_pool; + DescriptorPool& descriptor_pool; VKUpdateDescriptorQueue& update_descriptor_queue; + RenderPassCache& render_pass_cache; + BufferCache& buffer_cache; + TextureCache& texture_cache; + VideoCore::ShaderNotify& shader_notify; + bool use_asynchronous_shaders{}; - std::unique_ptr<Shader> null_shader; - std::unique_ptr<Shader> null_kernel; + GraphicsPipelineCacheKey graphics_key{}; + GraphicsPipeline* current_pipeline{}; - std::array<Shader*, Maxwell::MaxShaderProgram> last_shaders{}; + std::unordered_map<ComputePipelineCacheKey, std::unique_ptr<ComputePipeline>> compute_cache; + std::unordered_map<GraphicsPipelineCacheKey, std::unique_ptr<GraphicsPipeline>> graphics_cache; - GraphicsPipelineCacheKey last_graphics_key; - VKGraphicsPipeline* last_graphics_pipeline = nullptr; + ShaderPools main_pools; - std::mutex pipeline_cache; - std::unordered_map<GraphicsPipelineCacheKey, std::unique_ptr<VKGraphicsPipeline>> - graphics_cache; - std::unordered_map<ComputePipelineCacheKey, std::unique_ptr<VKComputePipeline>> compute_cache; -}; + Shader::Profile profile; + Shader::HostTranslateInfo host_info; -void FillDescriptorUpdateTemplateEntries( - const ShaderEntries& entries, u32& binding, u32& offset, - std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries); + std::filesystem::path pipeline_cache_filename; + + Common::ThreadWorker workers; + Common::ThreadWorker serialization_thread; +}; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_query_cache.cpp b/src/video_core/renderer_vulkan/vk_query_cache.cpp index 7cadd5147..c9cb32d71 100644 --- a/src/video_core/renderer_vulkan/vk_query_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_query_cache.cpp @@ -114,14 +114,10 @@ void HostCounter::EndQuery() { } u64 HostCounter::BlockingQuery() const { - if (tick >= cache.GetScheduler().CurrentTick()) { - cache.GetScheduler().Flush(); - } - + cache.GetScheduler().Wait(tick); u64 data; const VkResult query_result = cache.GetDevice().GetLogical().GetQueryResults( - query.first, query.second, 1, sizeof(data), &data, sizeof(data), - VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT); + query.first, query.second, 1, sizeof(data), &data, sizeof(data), VK_QUERY_RESULT_64_BIT); switch (query_result) { case VK_SUCCESS: diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index bd4d649cc..c7a07fdd8 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -24,7 +24,6 @@ #include "video_core/renderer_vulkan/vk_buffer_cache.h" #include "video_core/renderer_vulkan/vk_compute_pipeline.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" -#include "video_core/renderer_vulkan/vk_graphics_pipeline.h" #include "video_core/renderer_vulkan/vk_pipeline_cache.h" #include "video_core/renderer_vulkan/vk_rasterizer.h" #include "video_core/renderer_vulkan/vk_scheduler.h" @@ -55,11 +54,10 @@ struct DrawParams { u32 num_instances; u32 base_vertex; u32 num_vertices; + u32 first_index; bool is_indexed; }; -constexpr auto COMPUTE_SHADER_INDEX = static_cast<size_t>(Tegra::Engines::ShaderType::Compute); - VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t index) { const auto& src = regs.viewport_transform[index]; const float width = src.scale_x * 2.0f; @@ -97,118 +95,6 @@ VkRect2D GetScissorState(const Maxwell& regs, size_t index) { return scissor; } -std::array<GPUVAddr, Maxwell::MaxShaderProgram> GetShaderAddresses( - const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders) { - std::array<GPUVAddr, Maxwell::MaxShaderProgram> addresses; - for (size_t i = 0; i < std::size(addresses); ++i) { - addresses[i] = shaders[i] ? shaders[i]->GetGpuAddr() : 0; - } - return addresses; -} - -struct TextureHandle { - constexpr TextureHandle(u32 data, bool via_header_index) { - const Tegra::Texture::TextureHandle handle{data}; - image = handle.tic_id; - sampler = via_header_index ? image : handle.tsc_id.Value(); - } - - u32 image; - u32 sampler; -}; - -template <typename Engine, typename Entry> -TextureHandle GetTextureInfo(const Engine& engine, bool via_header_index, const Entry& entry, - size_t stage, size_t index = 0) { - const auto shader_type = static_cast<Tegra::Engines::ShaderType>(stage); - if constexpr (std::is_same_v<Entry, SamplerEntry>) { - if (entry.is_separated) { - const u32 buffer_1 = entry.buffer; - const u32 buffer_2 = entry.secondary_buffer; - const u32 offset_1 = entry.offset; - const u32 offset_2 = entry.secondary_offset; - const u32 handle_1 = engine.AccessConstBuffer32(shader_type, buffer_1, offset_1); - const u32 handle_2 = engine.AccessConstBuffer32(shader_type, buffer_2, offset_2); - return TextureHandle(handle_1 | handle_2, via_header_index); - } - } - if (entry.is_bindless) { - const u32 raw = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset); - return TextureHandle(raw, via_header_index); - } - const u32 buffer = engine.GetBoundBuffer(); - const u64 offset = (entry.offset + index) * sizeof(u32); - return TextureHandle(engine.AccessConstBuffer32(shader_type, buffer, offset), via_header_index); -} - -ImageViewType ImageViewTypeFromEntry(const SamplerEntry& entry) { - if (entry.is_buffer) { - return ImageViewType::e2D; - } - switch (entry.type) { - case Tegra::Shader::TextureType::Texture1D: - return entry.is_array ? ImageViewType::e1DArray : ImageViewType::e1D; - case Tegra::Shader::TextureType::Texture2D: - return entry.is_array ? ImageViewType::e2DArray : ImageViewType::e2D; - case Tegra::Shader::TextureType::Texture3D: - return ImageViewType::e3D; - case Tegra::Shader::TextureType::TextureCube: - return entry.is_array ? ImageViewType::CubeArray : ImageViewType::Cube; - } - UNREACHABLE(); - return ImageViewType::e2D; -} - -ImageViewType ImageViewTypeFromEntry(const ImageEntry& entry) { - switch (entry.type) { - case Tegra::Shader::ImageType::Texture1D: - return ImageViewType::e1D; - case Tegra::Shader::ImageType::Texture1DArray: - return ImageViewType::e1DArray; - case Tegra::Shader::ImageType::Texture2D: - return ImageViewType::e2D; - case Tegra::Shader::ImageType::Texture2DArray: - return ImageViewType::e2DArray; - case Tegra::Shader::ImageType::Texture3D: - return ImageViewType::e3D; - case Tegra::Shader::ImageType::TextureBuffer: - return ImageViewType::Buffer; - } - UNREACHABLE(); - return ImageViewType::e2D; -} - -void PushImageDescriptors(const ShaderEntries& entries, TextureCache& texture_cache, - VKUpdateDescriptorQueue& update_descriptor_queue, - ImageViewId*& image_view_id_ptr, VkSampler*& sampler_ptr) { - for ([[maybe_unused]] const auto& entry : entries.uniform_texels) { - const ImageViewId image_view_id = *image_view_id_ptr++; - const ImageView& image_view = texture_cache.GetImageView(image_view_id); - update_descriptor_queue.AddTexelBuffer(image_view.BufferView()); - } - for (const auto& entry : entries.samplers) { - for (size_t i = 0; i < entry.size; ++i) { - const VkSampler sampler = *sampler_ptr++; - const ImageViewId image_view_id = *image_view_id_ptr++; - const ImageView& image_view = texture_cache.GetImageView(image_view_id); - const VkImageView handle = image_view.Handle(ImageViewTypeFromEntry(entry)); - update_descriptor_queue.AddSampledImage(handle, sampler); - } - } - for ([[maybe_unused]] const auto& entry : entries.storage_texels) { - const ImageViewId image_view_id = *image_view_id_ptr++; - const ImageView& image_view = texture_cache.GetImageView(image_view_id); - update_descriptor_queue.AddTexelBuffer(image_view.BufferView()); - } - for (const auto& entry : entries.images) { - // TODO: Mark as modified - const ImageViewId image_view_id = *image_view_id_ptr++; - const ImageView& image_view = texture_cache.GetImageView(image_view_id); - const VkImageView handle = image_view.Handle(ImageViewTypeFromEntry(entry)); - update_descriptor_queue.AddImage(handle); - } -} - DrawParams MakeDrawParams(const Maxwell& regs, u32 num_instances, bool is_instanced, bool is_indexed) { DrawParams params{ @@ -216,6 +102,7 @@ DrawParams MakeDrawParams(const Maxwell& regs, u32 num_instances, bool is_instan .num_instances = is_instanced ? num_instances : 1, .base_vertex = is_indexed ? regs.vb_element_base : regs.vertex_buffer.first, .num_vertices = is_indexed ? regs.index_array.count : regs.vertex_buffer.count, + .first_index = is_indexed ? regs.index_array.first : 0, .is_indexed = is_indexed, }; if (regs.draw.topology == Maxwell::PrimitiveTopology::Quads) { @@ -243,21 +130,21 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra blit_image(device, scheduler, state_tracker, descriptor_pool), astc_decoder_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue, memory_allocator), - texture_cache_runtime{device, scheduler, memory_allocator, - staging_pool, blit_image, astc_decoder_pass}, + render_pass_cache(device), texture_cache_runtime{device, scheduler, + memory_allocator, staging_pool, + blit_image, astc_decoder_pass, + render_pass_cache}, texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory), buffer_cache_runtime(device, memory_allocator, scheduler, staging_pool, update_descriptor_queue, descriptor_pool), buffer_cache(*this, maxwell3d, kepler_compute, gpu_memory, cpu_memory_, buffer_cache_runtime), - pipeline_cache(*this, gpu, maxwell3d, kepler_compute, gpu_memory, device, scheduler, - descriptor_pool, update_descriptor_queue), - query_cache{*this, maxwell3d, gpu_memory, device, scheduler}, + pipeline_cache(*this, maxwell3d, kepler_compute, gpu_memory, device, scheduler, + descriptor_pool, update_descriptor_queue, render_pass_cache, buffer_cache, + texture_cache, gpu.ShaderNotify()), + query_cache{*this, maxwell3d, gpu_memory, device, scheduler}, accelerate_dma{buffer_cache}, fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler), - wfi_event(device.GetLogical().CreateEvent()), async_shaders(emu_window_) { + wfi_event(device.GetLogical().CreateEvent()) { scheduler.SetQueryCache(query_cache); - if (device.UseAsynchronousShaders()) { - async_shaders.AllocateWorkers(); - } } RasterizerVulkan::~RasterizerVulkan() = default; @@ -270,53 +157,30 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) { query_cache.UpdateCounters(); - graphics_key.fixed_state.Refresh(maxwell3d, device.IsExtExtendedDynamicStateSupported()); - - std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - - texture_cache.SynchronizeGraphicsDescriptors(); - texture_cache.UpdateRenderTargets(false); - - const auto shaders = pipeline_cache.GetShaders(); - graphics_key.shaders = GetShaderAddresses(shaders); - - SetupShaderDescriptors(shaders, is_indexed); - - const Framebuffer* const framebuffer = texture_cache.GetFramebuffer(); - graphics_key.renderpass = framebuffer->RenderPass(); - - VKGraphicsPipeline* const pipeline = pipeline_cache.GetGraphicsPipeline( - graphics_key, framebuffer->NumColorBuffers(), async_shaders); - if (pipeline == nullptr || pipeline->GetHandle() == VK_NULL_HANDLE) { - // Async graphics pipeline was not ready. + GraphicsPipeline* const pipeline{pipeline_cache.CurrentGraphicsPipeline()}; + if (!pipeline) { return; } + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + pipeline->Configure(is_indexed); BeginTransformFeedback(); - scheduler.RequestRenderpass(framebuffer); - scheduler.BindGraphicsPipeline(pipeline->GetHandle()); UpdateDynamicStates(); - const auto& regs = maxwell3d.regs; - const u32 num_instances = maxwell3d.mme_draw.instance_count; - const DrawParams draw_params = MakeDrawParams(regs, num_instances, is_instanced, is_indexed); - const VkPipelineLayout pipeline_layout = pipeline->GetLayout(); - const VkDescriptorSet descriptor_set = pipeline->CommitDescriptorSet(); - scheduler.Record([pipeline_layout, descriptor_set, draw_params](vk::CommandBuffer cmdbuf) { - if (descriptor_set) { - cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline_layout, - DESCRIPTOR_SET, descriptor_set, nullptr); - } + const auto& regs{maxwell3d.regs}; + const u32 num_instances{maxwell3d.mme_draw.instance_count}; + const DrawParams draw_params{MakeDrawParams(regs, num_instances, is_instanced, is_indexed)}; + scheduler.Record([draw_params](vk::CommandBuffer cmdbuf) { if (draw_params.is_indexed) { - cmdbuf.DrawIndexed(draw_params.num_vertices, draw_params.num_instances, 0, - draw_params.base_vertex, draw_params.base_instance); + cmdbuf.DrawIndexed(draw_params.num_vertices, draw_params.num_instances, + draw_params.first_index, draw_params.base_vertex, + draw_params.base_instance); } else { cmdbuf.Draw(draw_params.num_vertices, draw_params.num_instances, draw_params.base_vertex, draw_params.base_instance); } }); - EndTransformFeedback(); } @@ -326,6 +190,7 @@ void RasterizerVulkan::Clear() { if (!maxwell3d.ShouldExecute()) { return; } + FlushWork(); query_cache.UpdateCounters(); @@ -357,11 +222,13 @@ void RasterizerVulkan::Clear() { .height = std::min(clear_rect.rect.extent.height, render_area.height), }; - if (use_color) { + const u32 color_attachment = regs.clear_buffers.RT; + const auto attachment_aspect_mask = framebuffer->ImageRanges()[color_attachment].aspectMask; + const bool is_color_rt = (attachment_aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT) != 0; + if (use_color && is_color_rt) { VkClearValue clear_value; std::memcpy(clear_value.color.float32, regs.clear_color, sizeof(regs.clear_color)); - const u32 color_attachment = regs.clear_buffers.RT; scheduler.Record([color_attachment, clear_value, clear_rect](vk::CommandBuffer cmdbuf) { const VkClearAttachment attachment{ .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, @@ -393,73 +260,20 @@ void RasterizerVulkan::Clear() { }); } -void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) { - MICROPROFILE_SCOPE(Vulkan_Compute); - - query_cache.UpdateCounters(); +void RasterizerVulkan::DispatchCompute() { + FlushWork(); - const auto& launch_desc = kepler_compute.launch_description; - auto& pipeline = pipeline_cache.GetComputePipeline({ - .shader = code_addr, - .shared_memory_size = launch_desc.shared_alloc, - .workgroup_size{ - launch_desc.block_dim_x, - launch_desc.block_dim_y, - launch_desc.block_dim_z, - }, - }); + ComputePipeline* const pipeline{pipeline_cache.CurrentComputePipeline()}; + if (!pipeline) { + return; + } + std::scoped_lock lock{texture_cache.mutex, buffer_cache.mutex}; + pipeline->Configure(kepler_compute, gpu_memory, scheduler, buffer_cache, texture_cache); - // Compute dispatches can't be executed inside a renderpass + const auto& qmd{kepler_compute.launch_description}; + const std::array<u32, 3> dim{qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z}; scheduler.RequestOutsideRenderPassOperationContext(); - - image_view_indices.clear(); - sampler_handles.clear(); - - std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - - const auto& entries = pipeline.GetEntries(); - buffer_cache.SetEnabledComputeUniformBuffers(entries.enabled_uniform_buffers); - buffer_cache.UnbindComputeStorageBuffers(); - u32 ssbo_index = 0; - for (const auto& buffer : entries.global_buffers) { - buffer_cache.BindComputeStorageBuffer(ssbo_index, buffer.cbuf_index, buffer.cbuf_offset, - buffer.is_written); - ++ssbo_index; - } - buffer_cache.UpdateComputeBuffers(); - - texture_cache.SynchronizeComputeDescriptors(); - - SetupComputeUniformTexels(entries); - SetupComputeTextures(entries); - SetupComputeStorageTexels(entries); - SetupComputeImages(entries); - - const std::span indices_span(image_view_indices.data(), image_view_indices.size()); - texture_cache.FillComputeImageViews(indices_span, image_view_ids); - - update_descriptor_queue.Acquire(); - - buffer_cache.BindHostComputeBuffers(); - - ImageViewId* image_view_id_ptr = image_view_ids.data(); - VkSampler* sampler_ptr = sampler_handles.data(); - PushImageDescriptors(entries, texture_cache, update_descriptor_queue, image_view_id_ptr, - sampler_ptr); - - const VkPipeline pipeline_handle = pipeline.GetHandle(); - const VkPipelineLayout pipeline_layout = pipeline.GetLayout(); - const VkDescriptorSet descriptor_set = pipeline.CommitDescriptorSet(); - scheduler.Record([grid_x = launch_desc.grid_dim_x, grid_y = launch_desc.grid_dim_y, - grid_z = launch_desc.grid_dim_z, pipeline_handle, pipeline_layout, - descriptor_set](vk::CommandBuffer cmdbuf) { - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline_handle); - if (descriptor_set) { - cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline_layout, - DESCRIPTOR_SET, descriptor_set, nullptr); - } - cmdbuf.Dispatch(grid_x, grid_y, grid_z); - }); + scheduler.Record([dim](vk::CommandBuffer cmdbuf) { cmdbuf.Dispatch(dim[0], dim[1], dim[2]); }); } void RasterizerVulkan::ResetCounter(VideoCore::QueryType type) { @@ -580,6 +394,13 @@ void RasterizerVulkan::SignalSyncPoint(u32 value) { fence_manager.SignalSyncPoint(value); } +void RasterizerVulkan::SignalReference() { + if (!gpu.IsAsync()) { + return; + } + fence_manager.SignalOrdering(); +} + void RasterizerVulkan::ReleaseFences() { if (!gpu.IsAsync()) { return; @@ -612,10 +433,12 @@ void RasterizerVulkan::WaitForIdle() { cmdbuf.SetEvent(event, flags); cmdbuf.WaitEvents(event, flags, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, {}, {}, {}); }); + SignalReference(); } void RasterizerVulkan::FragmentBarrier() { // We already put barriers when a render pass finishes + scheduler.RequestOutsideRenderPassOperationContext(); } void RasterizerVulkan::TiledCacheBarrier() { @@ -623,10 +446,11 @@ void RasterizerVulkan::TiledCacheBarrier() { } void RasterizerVulkan::FlushCommands() { - if (draw_counter > 0) { - draw_counter = 0; - scheduler.Flush(); + if (draw_counter == 0) { + return; } + draw_counter = 0; + scheduler.Flush(); } void RasterizerVulkan::TickFrame() { @@ -652,6 +476,10 @@ bool RasterizerVulkan::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surf return true; } +Tegra::Engines::AccelerateDMAInterface& RasterizerVulkan::AccessAccelerateDMA() { + return accelerate_dma; +} + bool RasterizerVulkan::AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) { if (!framebuffer_addr) { @@ -662,13 +490,18 @@ bool RasterizerVulkan::AccelerateDisplay(const Tegra::FramebufferConfig& config, if (!image_view) { return false; } - screen_info.image_view = image_view->Handle(VideoCommon::ImageViewType::e2D); + screen_info.image_view = image_view->Handle(Shader::TextureType::Color2D); screen_info.width = image_view->size.width; screen_info.height = image_view->size.height; screen_info.is_srgb = VideoCore::Surface::IsPixelFormatSRGB(image_view->format); return true; } +void RasterizerVulkan::LoadDiskResources(u64 title_id, std::stop_token stop_loading, + const VideoCore::DiskResourceLoadCallback& callback) { + pipeline_cache.LoadDiskResources(title_id, stop_loading, callback); +} + void RasterizerVulkan::FlushWork() { static constexpr u32 DRAWS_TO_DISPATCH = 4096; @@ -677,63 +510,27 @@ void RasterizerVulkan::FlushWork() { if ((++draw_counter & 7) != 7) { return; } - if (draw_counter < DRAWS_TO_DISPATCH) { // Send recorded tasks to the worker thread scheduler.DispatchWork(); return; } - // Otherwise (every certain number of draws) flush execution. // This submits commands to the Vulkan driver. scheduler.Flush(); draw_counter = 0; } -void RasterizerVulkan::SetupShaderDescriptors( - const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders, bool is_indexed) { - image_view_indices.clear(); - sampler_handles.clear(); - for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) { - Shader* const shader = shaders[stage + 1]; - if (!shader) { - continue; - } - const ShaderEntries& entries = shader->GetEntries(); - SetupGraphicsUniformTexels(entries, stage); - SetupGraphicsTextures(entries, stage); - SetupGraphicsStorageTexels(entries, stage); - SetupGraphicsImages(entries, stage); - - buffer_cache.SetEnabledUniformBuffers(stage, entries.enabled_uniform_buffers); - buffer_cache.UnbindGraphicsStorageBuffers(stage); - u32 ssbo_index = 0; - for (const auto& buffer : entries.global_buffers) { - buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, buffer.cbuf_index, - buffer.cbuf_offset, buffer.is_written); - ++ssbo_index; - } - } - const std::span indices_span(image_view_indices.data(), image_view_indices.size()); - buffer_cache.UpdateGraphicsBuffers(is_indexed); - texture_cache.FillGraphicsImageViews(indices_span, image_view_ids); - - buffer_cache.BindHostGeometryBuffers(is_indexed); +AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_) : buffer_cache{buffer_cache_} {} - update_descriptor_queue.Acquire(); +bool AccelerateDMA::BufferClear(GPUVAddr src_address, u64 amount, u32 value) { + std::scoped_lock lock{buffer_cache.mutex}; + return buffer_cache.DMAClear(src_address, amount, value); +} - ImageViewId* image_view_id_ptr = image_view_ids.data(); - VkSampler* sampler_ptr = sampler_handles.data(); - for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) { - // Skip VertexA stage - Shader* const shader = shaders[stage + 1]; - if (!shader) { - continue; - } - buffer_cache.BindHostStageBuffers(stage); - PushImageDescriptors(shader->GetEntries(), texture_cache, update_descriptor_queue, - image_view_id_ptr, sampler_ptr); - } +bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) { + std::scoped_lock lock{buffer_cache.mutex}; + return buffer_cache.DMACopy(src_address, dest_address, amount); } void RasterizerVulkan::UpdateDynamicStates() { @@ -744,6 +541,7 @@ void RasterizerVulkan::UpdateDynamicStates() { UpdateBlendConstants(regs); UpdateDepthBounds(regs); UpdateStencilFaces(regs); + UpdateLineWidth(regs); if (device.IsExtExtendedDynamicStateSupported()) { UpdateCullMode(regs); UpdateDepthBoundsTestEnable(regs); @@ -753,6 +551,9 @@ void RasterizerVulkan::UpdateDynamicStates() { UpdateFrontFace(regs); UpdateStencilOp(regs); UpdateStencilTestEnable(regs); + if (device.IsExtVertexInputDynamicStateSupported()) { + UpdateVertexInput(regs); + } } } @@ -784,89 +585,6 @@ void RasterizerVulkan::EndTransformFeedback() { [](vk::CommandBuffer cmdbuf) { cmdbuf.EndTransformFeedbackEXT(0, 0, nullptr, nullptr); }); } -void RasterizerVulkan::SetupGraphicsUniformTexels(const ShaderEntries& entries, size_t stage) { - const auto& regs = maxwell3d.regs; - const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; - for (const auto& entry : entries.uniform_texels) { - const TextureHandle handle = GetTextureInfo(maxwell3d, via_header_index, entry, stage); - image_view_indices.push_back(handle.image); - } -} - -void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, size_t stage) { - const auto& regs = maxwell3d.regs; - const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; - for (const auto& entry : entries.samplers) { - for (size_t index = 0; index < entry.size; ++index) { - const TextureHandle handle = - GetTextureInfo(maxwell3d, via_header_index, entry, stage, index); - image_view_indices.push_back(handle.image); - - Sampler* const sampler = texture_cache.GetGraphicsSampler(handle.sampler); - sampler_handles.push_back(sampler->Handle()); - } - } -} - -void RasterizerVulkan::SetupGraphicsStorageTexels(const ShaderEntries& entries, size_t stage) { - const auto& regs = maxwell3d.regs; - const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; - for (const auto& entry : entries.storage_texels) { - const TextureHandle handle = GetTextureInfo(maxwell3d, via_header_index, entry, stage); - image_view_indices.push_back(handle.image); - } -} - -void RasterizerVulkan::SetupGraphicsImages(const ShaderEntries& entries, size_t stage) { - const auto& regs = maxwell3d.regs; - const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; - for (const auto& entry : entries.images) { - const TextureHandle handle = GetTextureInfo(maxwell3d, via_header_index, entry, stage); - image_view_indices.push_back(handle.image); - } -} - -void RasterizerVulkan::SetupComputeUniformTexels(const ShaderEntries& entries) { - const bool via_header_index = kepler_compute.launch_description.linked_tsc; - for (const auto& entry : entries.uniform_texels) { - const TextureHandle handle = - GetTextureInfo(kepler_compute, via_header_index, entry, COMPUTE_SHADER_INDEX); - image_view_indices.push_back(handle.image); - } -} - -void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) { - const bool via_header_index = kepler_compute.launch_description.linked_tsc; - for (const auto& entry : entries.samplers) { - for (size_t index = 0; index < entry.size; ++index) { - const TextureHandle handle = GetTextureInfo(kepler_compute, via_header_index, entry, - COMPUTE_SHADER_INDEX, index); - image_view_indices.push_back(handle.image); - - Sampler* const sampler = texture_cache.GetComputeSampler(handle.sampler); - sampler_handles.push_back(sampler->Handle()); - } - } -} - -void RasterizerVulkan::SetupComputeStorageTexels(const ShaderEntries& entries) { - const bool via_header_index = kepler_compute.launch_description.linked_tsc; - for (const auto& entry : entries.storage_texels) { - const TextureHandle handle = - GetTextureInfo(kepler_compute, via_header_index, entry, COMPUTE_SHADER_INDEX); - image_view_indices.push_back(handle.image); - } -} - -void RasterizerVulkan::SetupComputeImages(const ShaderEntries& entries) { - const bool via_header_index = kepler_compute.launch_description.linked_tsc; - for (const auto& entry : entries.images) { - const TextureHandle handle = - GetTextureInfo(kepler_compute, via_header_index, entry, COMPUTE_SHADER_INDEX); - image_view_indices.push_back(handle.image); - } -} - void RasterizerVulkan::UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs) { if (!state_tracker.TouchViewports()) { return; @@ -959,6 +677,14 @@ void RasterizerVulkan::UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs) } } +void RasterizerVulkan::UpdateLineWidth(Tegra::Engines::Maxwell3D::Regs& regs) { + if (!state_tracker.TouchLineWidth()) { + return; + } + const float width = regs.line_smooth_enable ? regs.line_width_smooth : regs.line_width_aliased; + scheduler.Record([width](vk::CommandBuffer cmdbuf) { cmdbuf.SetLineWidth(width); }); +} + void RasterizerVulkan::UpdateCullMode(Tegra::Engines::Maxwell3D::Regs& regs) { if (!state_tracker.TouchCullMode()) { return; @@ -973,6 +699,11 @@ void RasterizerVulkan::UpdateDepthBoundsTestEnable(Tegra::Engines::Maxwell3D::Re if (!state_tracker.TouchDepthBoundsTestEnable()) { return; } + bool enabled = regs.depth_bounds_enable; + if (enabled && !device.IsDepthBoundsSupported()) { + LOG_WARNING(Render_Vulkan, "Depth bounds is enabled but not supported"); + enabled = false; + } scheduler.Record([enable = regs.depth_bounds_enable](vk::CommandBuffer cmdbuf) { cmdbuf.SetDepthBoundsTestEnableEXT(enable); }); @@ -1060,4 +791,62 @@ void RasterizerVulkan::UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs& }); } +void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs) { + auto& dirty{maxwell3d.dirty.flags}; + if (!dirty[Dirty::VertexInput]) { + return; + } + dirty[Dirty::VertexInput] = false; + + boost::container::static_vector<VkVertexInputBindingDescription2EXT, 32> bindings; + boost::container::static_vector<VkVertexInputAttributeDescription2EXT, 32> attributes; + + // There seems to be a bug on Nvidia's driver where updating only higher attributes ends up + // generating dirty state. Track the highest dirty attribute and update all attributes until + // that one. + size_t highest_dirty_attr{}; + for (size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) { + if (dirty[Dirty::VertexAttribute0 + index]) { + highest_dirty_attr = index; + } + } + for (size_t index = 0; index < highest_dirty_attr; ++index) { + const Maxwell::VertexAttribute attribute{regs.vertex_attrib_format[index]}; + const u32 binding{attribute.buffer}; + dirty[Dirty::VertexAttribute0 + index] = false; + dirty[Dirty::VertexBinding0 + static_cast<size_t>(binding)] = true; + if (!attribute.constant) { + attributes.push_back({ + .sType = VK_STRUCTURE_TYPE_VERTEX_INPUT_ATTRIBUTE_DESCRIPTION_2_EXT, + .pNext = nullptr, + .location = static_cast<u32>(index), + .binding = binding, + .format = MaxwellToVK::VertexFormat(attribute.type, attribute.size), + .offset = attribute.offset, + }); + } + } + for (size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) { + if (!dirty[Dirty::VertexBinding0 + index]) { + continue; + } + dirty[Dirty::VertexBinding0 + index] = false; + + const u32 binding{static_cast<u32>(index)}; + const auto& input_binding{regs.vertex_array[binding]}; + const bool is_instanced{regs.instanced_arrays.IsInstancingEnabled(binding)}; + bindings.push_back({ + .sType = VK_STRUCTURE_TYPE_VERTEX_INPUT_BINDING_DESCRIPTION_2_EXT, + .pNext = nullptr, + .binding = binding, + .stride = input_binding.stride, + .inputRate = is_instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX, + .divisor = is_instanced ? input_binding.divisor : 1, + }); + } + scheduler.Record([bindings, attributes](vk::CommandBuffer cmdbuf) { + cmdbuf.SetVertexInputEXT(bindings, attributes); + }); +} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index 41459c5c5..866827247 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -13,6 +13,7 @@ #include <boost/container/static_vector.hpp> #include "common/common_types.h" +#include "video_core/engines/maxwell_dma.h" #include "video_core/rasterizer_accelerated.h" #include "video_core/rasterizer_interface.h" #include "video_core/renderer_vulkan/blit_image.h" @@ -20,14 +21,13 @@ #include "video_core/renderer_vulkan/vk_buffer_cache.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/renderer_vulkan/vk_fence_manager.h" -#include "video_core/renderer_vulkan/vk_graphics_pipeline.h" #include "video_core/renderer_vulkan/vk_pipeline_cache.h" #include "video_core/renderer_vulkan/vk_query_cache.h" +#include "video_core/renderer_vulkan/vk_render_pass_cache.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" #include "video_core/renderer_vulkan/vk_texture_cache.h" #include "video_core/renderer_vulkan/vk_update_descriptor.h" -#include "video_core/shader/async_shaders.h" #include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_wrapper.h" @@ -49,6 +49,18 @@ struct VKScreenInfo; class StateTracker; +class AccelerateDMA : public Tegra::Engines::AccelerateDMAInterface { +public: + explicit AccelerateDMA(BufferCache& buffer_cache); + + bool BufferCopy(GPUVAddr start_address, GPUVAddr end_address, u64 amount) override; + + bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) override; + +private: + BufferCache& buffer_cache; +}; + class RasterizerVulkan final : public VideoCore::RasterizerAccelerated { public: explicit RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, @@ -60,7 +72,7 @@ public: void Draw(bool is_indexed, bool is_instanced) override; void Clear() override; - void DispatchCompute(GPUVAddr code_addr) override; + void DispatchCompute() override; void ResetCounter(VideoCore::QueryType type) override; void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override; void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override; @@ -75,6 +87,7 @@ public: void ModifyGPUMemory(GPUVAddr addr, u64 size) override; void SignalSemaphore(GPUVAddr addr, u32 value) override; void SignalSyncPoint(u32 value) override; + void SignalReference() override; void ReleaseFences() override; void FlushAndInvalidateRegion(VAddr addr, u64 size) override; void WaitForIdle() override; @@ -85,21 +98,11 @@ public: bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surface& src, const Tegra::Engines::Fermi2D::Surface& dst, const Tegra::Engines::Fermi2D::Config& copy_config) override; + Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() override; bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) override; - - VideoCommon::Shader::AsyncShaders& GetAsyncShaders() { - return async_shaders; - } - - const VideoCommon::Shader::AsyncShaders& GetAsyncShaders() const { - return async_shaders; - } - - /// Maximum supported size that a constbuffer can have in bytes. - static constexpr size_t MaxConstbufferSize = 0x10000; - static_assert(MaxConstbufferSize % (4 * sizeof(float)) == 0, - "The maximum size of a constbuffer must be a multiple of the size of GLvec4"); + void LoadDiskResources(u64 title_id, std::stop_token stop_loading, + const VideoCore::DiskResourceLoadCallback& callback) override; private: static constexpr size_t MAX_TEXTURES = 192; @@ -110,46 +113,19 @@ private: void FlushWork(); - /// Setup descriptors in the graphics pipeline. - void SetupShaderDescriptors(const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders, - bool is_indexed); - void UpdateDynamicStates(); void BeginTransformFeedback(); void EndTransformFeedback(); - /// Setup uniform texels in the graphics pipeline. - void SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage); - - /// Setup textures in the graphics pipeline. - void SetupGraphicsTextures(const ShaderEntries& entries, std::size_t stage); - - /// Setup storage texels in the graphics pipeline. - void SetupGraphicsStorageTexels(const ShaderEntries& entries, std::size_t stage); - - /// Setup images in the graphics pipeline. - void SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage); - - /// Setup texel buffers in the compute pipeline. - void SetupComputeUniformTexels(const ShaderEntries& entries); - - /// Setup textures in the compute pipeline. - void SetupComputeTextures(const ShaderEntries& entries); - - /// Setup storage texels in the compute pipeline. - void SetupComputeStorageTexels(const ShaderEntries& entries); - - /// Setup images in the compute pipeline. - void SetupComputeImages(const ShaderEntries& entries); - void UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateDepthBias(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateBlendConstants(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateDepthBounds(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs); + void UpdateLineWidth(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateCullMode(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateDepthBoundsTestEnable(Tegra::Engines::Maxwell3D::Regs& regs); @@ -160,6 +136,8 @@ private: void UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs& regs); + void UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs); + Tegra::GPU& gpu; Tegra::MemoryManager& gpu_memory; Tegra::Engines::Maxwell3D& maxwell3d; @@ -172,23 +150,22 @@ private: VKScheduler& scheduler; StagingBufferPool staging_pool; - VKDescriptorPool descriptor_pool; + DescriptorPool descriptor_pool; VKUpdateDescriptorQueue update_descriptor_queue; BlitImageHelper blit_image; ASTCDecoderPass astc_decoder_pass; - - GraphicsPipelineCacheKey graphics_key; + RenderPassCache render_pass_cache; TextureCacheRuntime texture_cache_runtime; TextureCache texture_cache; BufferCacheRuntime buffer_cache_runtime; BufferCache buffer_cache; - VKPipelineCache pipeline_cache; + PipelineCache pipeline_cache; VKQueryCache query_cache; + AccelerateDMA accelerate_dma; VKFenceManager fence_manager; vk::Event wfi_event; - VideoCommon::Shader::AsyncShaders async_shaders; boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices; std::array<VideoCommon::ImageViewId, MAX_IMAGE_VIEWS> image_view_ids; diff --git a/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp b/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp new file mode 100644 index 000000000..451ffe019 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp @@ -0,0 +1,96 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <unordered_map> + +#include <boost/container/static_vector.hpp> + +#include "video_core/renderer_vulkan/maxwell_to_vk.h" +#include "video_core/renderer_vulkan/vk_render_pass_cache.h" +#include "video_core/surface.h" +#include "video_core/vulkan_common/vulkan_device.h" +#include "video_core/vulkan_common/vulkan_wrapper.h" + +namespace Vulkan { +namespace { +using VideoCore::Surface::PixelFormat; + +VkAttachmentDescription AttachmentDescription(const Device& device, PixelFormat format, + VkSampleCountFlagBits samples) { + using MaxwellToVK::SurfaceFormat; + return { + .flags = VK_ATTACHMENT_DESCRIPTION_MAY_ALIAS_BIT, + .format = SurfaceFormat(device, FormatType::Optimal, true, format).format, + .samples = samples, + .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD, + .storeOp = VK_ATTACHMENT_STORE_OP_STORE, + .stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD, + .stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE, + .initialLayout = VK_IMAGE_LAYOUT_GENERAL, + .finalLayout = VK_IMAGE_LAYOUT_GENERAL, + }; +} +} // Anonymous namespace + +RenderPassCache::RenderPassCache(const Device& device_) : device{&device_} {} + +VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { + std::lock_guard lock{mutex}; + const auto [pair, is_new] = cache.try_emplace(key); + if (!is_new) { + return *pair->second; + } + boost::container::static_vector<VkAttachmentDescription, 9> descriptions; + std::array<VkAttachmentReference, 8> references{}; + u32 num_attachments{}; + u32 num_colors{}; + for (size_t index = 0; index < key.color_formats.size(); ++index) { + const PixelFormat format{key.color_formats[index]}; + const bool is_valid{format != PixelFormat::Invalid}; + references[index] = VkAttachmentReference{ + .attachment = is_valid ? num_colors : VK_ATTACHMENT_UNUSED, + .layout = VK_IMAGE_LAYOUT_GENERAL, + }; + if (is_valid) { + descriptions.push_back(AttachmentDescription(*device, format, key.samples)); + num_attachments = static_cast<u32>(index + 1); + ++num_colors; + } + } + const bool has_depth{key.depth_format != PixelFormat::Invalid}; + VkAttachmentReference depth_reference{}; + if (key.depth_format != PixelFormat::Invalid) { + depth_reference = VkAttachmentReference{ + .attachment = num_colors, + .layout = VK_IMAGE_LAYOUT_GENERAL, + }; + descriptions.push_back(AttachmentDescription(*device, key.depth_format, key.samples)); + } + const VkSubpassDescription subpass{ + .flags = 0, + .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS, + .inputAttachmentCount = 0, + .pInputAttachments = nullptr, + .colorAttachmentCount = num_attachments, + .pColorAttachments = references.data(), + .pResolveAttachments = nullptr, + .pDepthStencilAttachment = has_depth ? &depth_reference : nullptr, + .preserveAttachmentCount = 0, + .pPreserveAttachments = nullptr, + }; + pair->second = device->GetLogical().CreateRenderPass({ + .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .attachmentCount = static_cast<u32>(descriptions.size()), + .pAttachments = descriptions.empty() ? nullptr : descriptions.data(), + .subpassCount = 1, + .pSubpasses = &subpass, + .dependencyCount = 0, + .pDependencies = nullptr, + }); + return *pair->second; +} + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_render_pass_cache.h b/src/video_core/renderer_vulkan/vk_render_pass_cache.h new file mode 100644 index 000000000..eaa0ed775 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_render_pass_cache.h @@ -0,0 +1,55 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <mutex> +#include <unordered_map> + +#include "video_core/surface.h" +#include "video_core/vulkan_common/vulkan_wrapper.h" + +namespace Vulkan { + +struct RenderPassKey { + auto operator<=>(const RenderPassKey&) const noexcept = default; + + std::array<VideoCore::Surface::PixelFormat, 8> color_formats; + VideoCore::Surface::PixelFormat depth_format; + VkSampleCountFlagBits samples; +}; + +} // namespace Vulkan + +namespace std { +template <> +struct hash<Vulkan::RenderPassKey> { + [[nodiscard]] size_t operator()(const Vulkan::RenderPassKey& key) const noexcept { + size_t value = static_cast<size_t>(key.depth_format) << 48; + value ^= static_cast<size_t>(key.samples) << 52; + for (size_t i = 0; i < key.color_formats.size(); ++i) { + value ^= static_cast<size_t>(key.color_formats[i]) << (i * 6); + } + return value; + } +}; +} // namespace std + +namespace Vulkan { + +class Device; + +class RenderPassCache { +public: + explicit RenderPassCache(const Device& device_); + + VkRenderPass Get(const RenderPassKey& key); + +private: + const Device* device{}; + std::unordered_map<RenderPassKey, vk::RenderPass> cache; + std::mutex mutex; +}; + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_resource_pool.cpp b/src/video_core/renderer_vulkan/vk_resource_pool.cpp index a8bf7bda8..2dd514968 100644 --- a/src/video_core/renderer_vulkan/vk_resource_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_resource_pool.cpp @@ -10,18 +10,16 @@ namespace Vulkan { ResourcePool::ResourcePool(MasterSemaphore& master_semaphore_, size_t grow_step_) - : master_semaphore{master_semaphore_}, grow_step{grow_step_} {} - -ResourcePool::~ResourcePool() = default; + : master_semaphore{&master_semaphore_}, grow_step{grow_step_} {} size_t ResourcePool::CommitResource() { // Refresh semaphore to query updated results - master_semaphore.Refresh(); - const u64 gpu_tick = master_semaphore.KnownGpuTick(); + master_semaphore->Refresh(); + const u64 gpu_tick = master_semaphore->KnownGpuTick(); const auto search = [this, gpu_tick](size_t begin, size_t end) -> std::optional<size_t> { for (size_t iterator = begin; iterator < end; ++iterator) { if (gpu_tick >= ticks[iterator]) { - ticks[iterator] = master_semaphore.CurrentTick(); + ticks[iterator] = master_semaphore->CurrentTick(); return iterator; } } @@ -36,7 +34,7 @@ size_t ResourcePool::CommitResource() { // Both searches failed, the pool is full; handle it. const size_t free_resource = ManageOverflow(); - ticks[free_resource] = master_semaphore.CurrentTick(); + ticks[free_resource] = master_semaphore->CurrentTick(); found = free_resource; } } diff --git a/src/video_core/renderer_vulkan/vk_resource_pool.h b/src/video_core/renderer_vulkan/vk_resource_pool.h index 9d0bb3b4d..f0b80ad59 100644 --- a/src/video_core/renderer_vulkan/vk_resource_pool.h +++ b/src/video_core/renderer_vulkan/vk_resource_pool.h @@ -18,8 +18,16 @@ class MasterSemaphore; */ class ResourcePool { public: + explicit ResourcePool() = default; explicit ResourcePool(MasterSemaphore& master_semaphore, size_t grow_step); - virtual ~ResourcePool(); + + virtual ~ResourcePool() = default; + + ResourcePool& operator=(ResourcePool&&) noexcept = default; + ResourcePool(ResourcePool&&) noexcept = default; + + ResourcePool& operator=(const ResourcePool&) = default; + ResourcePool(const ResourcePool&) = default; protected: size_t CommitResource(); @@ -34,7 +42,7 @@ private: /// Allocates a new page of resources. void Grow(); - MasterSemaphore& master_semaphore; + MasterSemaphore* master_semaphore{}; size_t grow_step = 0; ///< Number of new resources created after an overflow size_t hint_iterator = 0; ///< Hint to where the next free resources is likely to be found std::vector<u64> ticks; ///< Ticks for each resource diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index f35c120b0..4840962de 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -31,7 +31,7 @@ void VKScheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf) { command->~Command(); command = next; } - + submit = false; command_offset = 0; first = nullptr; last = nullptr; @@ -42,13 +42,16 @@ VKScheduler::VKScheduler(const Device& device_, StateTracker& state_tracker_) master_semaphore{std::make_unique<MasterSemaphore>(device)}, command_pool{std::make_unique<CommandPool>(*master_semaphore, device)} { AcquireNewChunk(); - AllocateNewContext(); + AllocateWorkerCommandBuffer(); worker_thread = std::thread(&VKScheduler::WorkerThread, this); } VKScheduler::~VKScheduler() { - quit = true; - cv.notify_all(); + { + std::lock_guard lock{work_mutex}; + quit = true; + } + work_cv.notify_all(); worker_thread.join(); } @@ -60,6 +63,7 @@ void VKScheduler::Flush(VkSemaphore semaphore) { void VKScheduler::Finish(VkSemaphore semaphore) { const u64 presubmit_tick = CurrentTick(); SubmitExecution(semaphore); + WaitWorker(); Wait(presubmit_tick); AllocateNewContext(); } @@ -68,20 +72,19 @@ void VKScheduler::WaitWorker() { MICROPROFILE_SCOPE(Vulkan_WaitForWorker); DispatchWork(); - bool finished = false; - do { - cv.notify_all(); - std::unique_lock lock{mutex}; - finished = chunk_queue.Empty(); - } while (!finished); + std::unique_lock lock{work_mutex}; + wait_cv.wait(lock, [this] { return work_queue.empty(); }); } void VKScheduler::DispatchWork() { if (chunk->Empty()) { return; } - chunk_queue.Push(std::move(chunk)); - cv.notify_all(); + { + std::lock_guard lock{work_mutex}; + work_queue.push(std::move(chunk)); + } + work_cv.notify_one(); AcquireNewChunk(); } @@ -124,93 +127,101 @@ void VKScheduler::RequestOutsideRenderPassOperationContext() { EndRenderPass(); } -void VKScheduler::BindGraphicsPipeline(VkPipeline pipeline) { +bool VKScheduler::UpdateGraphicsPipeline(GraphicsPipeline* pipeline) { if (state.graphics_pipeline == pipeline) { - return; + return false; } state.graphics_pipeline = pipeline; - Record([pipeline](vk::CommandBuffer cmdbuf) { - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline); - }); + return true; } void VKScheduler::WorkerThread() { - Common::SetCurrentThreadPriority(Common::ThreadPriority::High); - std::unique_lock lock{mutex}; + Common::SetCurrentThreadName("yuzu:VulkanWorker"); do { - cv.wait(lock, [this] { return !chunk_queue.Empty() || quit; }); - if (quit) { - continue; + if (work_queue.empty()) { + wait_cv.notify_all(); + } + std::unique_ptr<CommandChunk> work; + { + std::unique_lock lock{work_mutex}; + work_cv.wait(lock, [this] { return !work_queue.empty() || quit; }); + if (quit) { + continue; + } + work = std::move(work_queue.front()); + work_queue.pop(); + } + const bool has_submit = work->HasSubmit(); + work->ExecuteAll(current_cmdbuf); + if (has_submit) { + AllocateWorkerCommandBuffer(); } - auto extracted_chunk = std::move(chunk_queue.Front()); - chunk_queue.Pop(); - extracted_chunk->ExecuteAll(current_cmdbuf); - chunk_reserve.Push(std::move(extracted_chunk)); + std::lock_guard reserve_lock{reserve_mutex}; + chunk_reserve.push_back(std::move(work)); } while (!quit); } +void VKScheduler::AllocateWorkerCommandBuffer() { + current_cmdbuf = vk::CommandBuffer(command_pool->Commit(), device.GetDispatchLoader()); + current_cmdbuf.Begin({ + .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, + .pNext = nullptr, + .flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT, + .pInheritanceInfo = nullptr, + }); +} + void VKScheduler::SubmitExecution(VkSemaphore semaphore) { EndPendingOperations(); InvalidateState(); - WaitWorker(); - std::unique_lock lock{mutex}; + const u64 signal_value = master_semaphore->NextTick(); + Record([semaphore, signal_value, this](vk::CommandBuffer cmdbuf) { + cmdbuf.End(); - current_cmdbuf.End(); + const u32 num_signal_semaphores = semaphore ? 2U : 1U; - const VkSemaphore timeline_semaphore = master_semaphore->Handle(); - const u32 num_signal_semaphores = semaphore ? 2U : 1U; + const u64 wait_value = signal_value - 1; + const VkPipelineStageFlags wait_stage_mask = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT; - const u64 signal_value = master_semaphore->CurrentTick(); - const u64 wait_value = signal_value - 1; - const VkPipelineStageFlags wait_stage_mask = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT; + const VkSemaphore timeline_semaphore = master_semaphore->Handle(); + const std::array signal_values{signal_value, u64(0)}; + const std::array signal_semaphores{timeline_semaphore, semaphore}; - master_semaphore->NextTick(); - - const std::array signal_values{signal_value, u64(0)}; - const std::array signal_semaphores{timeline_semaphore, semaphore}; - - const VkTimelineSemaphoreSubmitInfoKHR timeline_si{ - .sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR, - .pNext = nullptr, - .waitSemaphoreValueCount = 1, - .pWaitSemaphoreValues = &wait_value, - .signalSemaphoreValueCount = num_signal_semaphores, - .pSignalSemaphoreValues = signal_values.data(), - }; - const VkSubmitInfo submit_info{ - .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO, - .pNext = &timeline_si, - .waitSemaphoreCount = 1, - .pWaitSemaphores = &timeline_semaphore, - .pWaitDstStageMask = &wait_stage_mask, - .commandBufferCount = 1, - .pCommandBuffers = current_cmdbuf.address(), - .signalSemaphoreCount = num_signal_semaphores, - .pSignalSemaphores = signal_semaphores.data(), - }; - switch (const VkResult result = device.GetGraphicsQueue().Submit(submit_info)) { - case VK_SUCCESS: - break; - case VK_ERROR_DEVICE_LOST: - device.ReportLoss(); - [[fallthrough]]; - default: - vk::Check(result); - } + const VkTimelineSemaphoreSubmitInfoKHR timeline_si{ + .sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR, + .pNext = nullptr, + .waitSemaphoreValueCount = 1, + .pWaitSemaphoreValues = &wait_value, + .signalSemaphoreValueCount = num_signal_semaphores, + .pSignalSemaphoreValues = signal_values.data(), + }; + const VkSubmitInfo submit_info{ + .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO, + .pNext = &timeline_si, + .waitSemaphoreCount = 1, + .pWaitSemaphores = &timeline_semaphore, + .pWaitDstStageMask = &wait_stage_mask, + .commandBufferCount = 1, + .pCommandBuffers = cmdbuf.address(), + .signalSemaphoreCount = num_signal_semaphores, + .pSignalSemaphores = signal_semaphores.data(), + }; + switch (const VkResult result = device.GetGraphicsQueue().Submit(submit_info)) { + case VK_SUCCESS: + break; + case VK_ERROR_DEVICE_LOST: + device.ReportLoss(); + [[fallthrough]]; + default: + vk::Check(result); + } + }); + chunk->MarkSubmit(); + DispatchWork(); } void VKScheduler::AllocateNewContext() { - std::unique_lock lock{mutex}; - - current_cmdbuf = vk::CommandBuffer(command_pool->Commit(), device.GetDispatchLoader()); - current_cmdbuf.Begin({ - .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, - .pNext = nullptr, - .flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT, - .pInheritanceInfo = nullptr, - }); - // Enable counters once again. These are disabled when a command buffer is finished. if (query_cache) { query_cache->UpdateCounters(); @@ -265,12 +276,13 @@ void VKScheduler::EndRenderPass() { } void VKScheduler::AcquireNewChunk() { - if (chunk_reserve.Empty()) { + std::lock_guard lock{reserve_mutex}; + if (chunk_reserve.empty()) { chunk = std::make_unique<CommandChunk>(); return; } - chunk = std::move(chunk_reserve.Front()); - chunk_reserve.Pop(); + chunk = std::move(chunk_reserve.back()); + chunk_reserve.pop_back(); } } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index 3ce48e9d2..cf39a2363 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -8,12 +8,12 @@ #include <condition_variable> #include <cstddef> #include <memory> -#include <stack> #include <thread> #include <utility> +#include <queue> + #include "common/alignment.h" #include "common/common_types.h" -#include "common/threadsafe_queue.h" #include "video_core/renderer_vulkan/vk_master_semaphore.h" #include "video_core/vulkan_common/vulkan_wrapper.h" @@ -22,6 +22,7 @@ namespace Vulkan { class CommandPool; class Device; class Framebuffer; +class GraphicsPipeline; class StateTracker; class VKQueryCache; @@ -52,8 +53,8 @@ public: /// of a renderpass. void RequestOutsideRenderPassOperationContext(); - /// Binds a pipeline to the current execution context. - void BindGraphicsPipeline(VkPipeline pipeline); + /// Update the pipeline to the current execution context. + bool UpdateGraphicsPipeline(GraphicsPipeline* pipeline); /// Invalidates current command buffer state except for render passes void InvalidateState(); @@ -85,6 +86,10 @@ public: /// Waits for the given tick to trigger on the GPU. void Wait(u64 tick) { + if (tick >= master_semaphore->CurrentTick()) { + // Make sure we are not waiting for the current tick without signalling + Flush(); + } master_semaphore->Wait(tick); } @@ -154,15 +159,24 @@ private: return true; } + void MarkSubmit() { + submit = true; + } + bool Empty() const { return command_offset == 0; } + bool HasSubmit() const { + return submit; + } + private: Command* first = nullptr; Command* last = nullptr; size_t command_offset = 0; + bool submit = false; alignas(std::max_align_t) std::array<u8, 0x8000> data{}; }; @@ -170,11 +184,13 @@ private: VkRenderPass renderpass = nullptr; VkFramebuffer framebuffer = nullptr; VkExtent2D render_area = {0, 0}; - VkPipeline graphics_pipeline = nullptr; + GraphicsPipeline* graphics_pipeline = nullptr; }; void WorkerThread(); + void AllocateWorkerCommandBuffer(); + void SubmitExecution(VkSemaphore semaphore); void AllocateNewContext(); @@ -204,11 +220,13 @@ private: std::array<VkImage, 9> renderpass_images{}; std::array<VkImageSubresourceRange, 9> renderpass_image_ranges{}; - Common::SPSCQueue<std::unique_ptr<CommandChunk>> chunk_queue; - Common::SPSCQueue<std::unique_ptr<CommandChunk>> chunk_reserve; - std::mutex mutex; - std::condition_variable cv; - bool quit = false; + std::queue<std::unique_ptr<CommandChunk>> work_queue; + std::vector<std::unique_ptr<CommandChunk>> chunk_reserve; + std::mutex reserve_mutex; + std::mutex work_mutex; + std::condition_variable work_cv; + std::condition_variable wait_cv; + std::atomic_bool quit{}; }; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp deleted file mode 100644 index c6846d886..000000000 --- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp +++ /dev/null @@ -1,3166 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <functional> -#include <limits> -#include <map> -#include <optional> -#include <type_traits> -#include <unordered_map> -#include <utility> - -#include <fmt/format.h> - -#include <sirit/sirit.h> - -#include "common/alignment.h" -#include "common/assert.h" -#include "common/common_types.h" -#include "common/logging/log.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/engines/shader_header.h" -#include "video_core/engines/shader_type.h" -#include "video_core/renderer_vulkan/vk_shader_decompiler.h" -#include "video_core/shader/node.h" -#include "video_core/shader/shader_ir.h" -#include "video_core/shader/transform_feedback.h" -#include "video_core/vulkan_common/vulkan_device.h" - -namespace Vulkan { - -namespace { - -using Sirit::Id; -using Tegra::Engines::ShaderType; -using Tegra::Shader::Attribute; -using Tegra::Shader::PixelImap; -using Tegra::Shader::Register; -using namespace VideoCommon::Shader; - -using Maxwell = Tegra::Engines::Maxwell3D::Regs; -using Operation = const OperationNode&; - -class ASTDecompiler; -class ExprDecompiler; - -// TODO(Rodrigo): Use rasterizer's value -constexpr u32 MaxConstBufferFloats = 0x4000; -constexpr u32 MaxConstBufferElements = MaxConstBufferFloats / 4; - -constexpr u32 NumInputPatches = 32; // This value seems to be the standard - -enum class Type { Void, Bool, Bool2, Float, Int, Uint, HalfFloat }; - -class Expression final { -public: - Expression(Id id_, Type type_) : id{id_}, type{type_} { - ASSERT(type_ != Type::Void); - } - Expression() : type{Type::Void} {} - - Id id{}; - Type type{}; -}; -static_assert(std::is_standard_layout_v<Expression>); - -struct TexelBuffer { - Id image_type{}; - Id image{}; -}; - -struct SampledImage { - Id image_type{}; - Id sampler_type{}; - Id sampler_pointer_type{}; - Id variable{}; -}; - -struct StorageImage { - Id image_type{}; - Id image{}; -}; - -struct AttributeType { - Type type; - Id scalar; - Id vector; -}; - -struct VertexIndices { - std::optional<u32> position; - std::optional<u32> layer; - std::optional<u32> viewport; - std::optional<u32> point_size; - std::optional<u32> clip_distances; -}; - -struct GenericVaryingDescription { - Id id = nullptr; - u32 first_element = 0; - bool is_scalar = false; -}; - -spv::Dim GetSamplerDim(const SamplerEntry& sampler) { - ASSERT(!sampler.is_buffer); - switch (sampler.type) { - case Tegra::Shader::TextureType::Texture1D: - return spv::Dim::Dim1D; - case Tegra::Shader::TextureType::Texture2D: - return spv::Dim::Dim2D; - case Tegra::Shader::TextureType::Texture3D: - return spv::Dim::Dim3D; - case Tegra::Shader::TextureType::TextureCube: - return spv::Dim::Cube; - default: - UNIMPLEMENTED_MSG("Unimplemented sampler type={}", sampler.type); - return spv::Dim::Dim2D; - } -} - -std::pair<spv::Dim, bool> GetImageDim(const ImageEntry& image) { - switch (image.type) { - case Tegra::Shader::ImageType::Texture1D: - return {spv::Dim::Dim1D, false}; - case Tegra::Shader::ImageType::TextureBuffer: - return {spv::Dim::Buffer, false}; - case Tegra::Shader::ImageType::Texture1DArray: - return {spv::Dim::Dim1D, true}; - case Tegra::Shader::ImageType::Texture2D: - return {spv::Dim::Dim2D, false}; - case Tegra::Shader::ImageType::Texture2DArray: - return {spv::Dim::Dim2D, true}; - case Tegra::Shader::ImageType::Texture3D: - return {spv::Dim::Dim3D, false}; - default: - UNIMPLEMENTED_MSG("Unimplemented image type={}", image.type); - return {spv::Dim::Dim2D, false}; - } -} - -/// Returns the number of vertices present in a primitive topology. -u32 GetNumPrimitiveTopologyVertices(Maxwell::PrimitiveTopology primitive_topology) { - switch (primitive_topology) { - case Maxwell::PrimitiveTopology::Points: - return 1; - case Maxwell::PrimitiveTopology::Lines: - case Maxwell::PrimitiveTopology::LineLoop: - case Maxwell::PrimitiveTopology::LineStrip: - return 2; - case Maxwell::PrimitiveTopology::Triangles: - case Maxwell::PrimitiveTopology::TriangleStrip: - case Maxwell::PrimitiveTopology::TriangleFan: - return 3; - case Maxwell::PrimitiveTopology::LinesAdjacency: - case Maxwell::PrimitiveTopology::LineStripAdjacency: - return 4; - case Maxwell::PrimitiveTopology::TrianglesAdjacency: - case Maxwell::PrimitiveTopology::TriangleStripAdjacency: - return 6; - case Maxwell::PrimitiveTopology::Quads: - UNIMPLEMENTED_MSG("Quads"); - return 3; - case Maxwell::PrimitiveTopology::QuadStrip: - UNIMPLEMENTED_MSG("QuadStrip"); - return 3; - case Maxwell::PrimitiveTopology::Polygon: - UNIMPLEMENTED_MSG("Polygon"); - return 3; - case Maxwell::PrimitiveTopology::Patches: - UNIMPLEMENTED_MSG("Patches"); - return 3; - default: - UNREACHABLE(); - return 3; - } -} - -spv::ExecutionMode GetExecutionMode(Maxwell::TessellationPrimitive primitive) { - switch (primitive) { - case Maxwell::TessellationPrimitive::Isolines: - return spv::ExecutionMode::Isolines; - case Maxwell::TessellationPrimitive::Triangles: - return spv::ExecutionMode::Triangles; - case Maxwell::TessellationPrimitive::Quads: - return spv::ExecutionMode::Quads; - } - UNREACHABLE(); - return spv::ExecutionMode::Triangles; -} - -spv::ExecutionMode GetExecutionMode(Maxwell::TessellationSpacing spacing) { - switch (spacing) { - case Maxwell::TessellationSpacing::Equal: - return spv::ExecutionMode::SpacingEqual; - case Maxwell::TessellationSpacing::FractionalOdd: - return spv::ExecutionMode::SpacingFractionalOdd; - case Maxwell::TessellationSpacing::FractionalEven: - return spv::ExecutionMode::SpacingFractionalEven; - } - UNREACHABLE(); - return spv::ExecutionMode::SpacingEqual; -} - -spv::ExecutionMode GetExecutionMode(Maxwell::PrimitiveTopology input_topology) { - switch (input_topology) { - case Maxwell::PrimitiveTopology::Points: - return spv::ExecutionMode::InputPoints; - case Maxwell::PrimitiveTopology::Lines: - case Maxwell::PrimitiveTopology::LineLoop: - case Maxwell::PrimitiveTopology::LineStrip: - return spv::ExecutionMode::InputLines; - case Maxwell::PrimitiveTopology::Triangles: - case Maxwell::PrimitiveTopology::TriangleStrip: - case Maxwell::PrimitiveTopology::TriangleFan: - return spv::ExecutionMode::Triangles; - case Maxwell::PrimitiveTopology::LinesAdjacency: - case Maxwell::PrimitiveTopology::LineStripAdjacency: - return spv::ExecutionMode::InputLinesAdjacency; - case Maxwell::PrimitiveTopology::TrianglesAdjacency: - case Maxwell::PrimitiveTopology::TriangleStripAdjacency: - return spv::ExecutionMode::InputTrianglesAdjacency; - case Maxwell::PrimitiveTopology::Quads: - UNIMPLEMENTED_MSG("Quads"); - return spv::ExecutionMode::Triangles; - case Maxwell::PrimitiveTopology::QuadStrip: - UNIMPLEMENTED_MSG("QuadStrip"); - return spv::ExecutionMode::Triangles; - case Maxwell::PrimitiveTopology::Polygon: - UNIMPLEMENTED_MSG("Polygon"); - return spv::ExecutionMode::Triangles; - case Maxwell::PrimitiveTopology::Patches: - UNIMPLEMENTED_MSG("Patches"); - return spv::ExecutionMode::Triangles; - } - UNREACHABLE(); - return spv::ExecutionMode::Triangles; -} - -spv::ExecutionMode GetExecutionMode(Tegra::Shader::OutputTopology output_topology) { - switch (output_topology) { - case Tegra::Shader::OutputTopology::PointList: - return spv::ExecutionMode::OutputPoints; - case Tegra::Shader::OutputTopology::LineStrip: - return spv::ExecutionMode::OutputLineStrip; - case Tegra::Shader::OutputTopology::TriangleStrip: - return spv::ExecutionMode::OutputTriangleStrip; - default: - UNREACHABLE(); - return spv::ExecutionMode::OutputPoints; - } -} - -/// Returns true if an attribute index is one of the 32 generic attributes -constexpr bool IsGenericAttribute(Attribute::Index attribute) { - return attribute >= Attribute::Index::Attribute_0 && - attribute <= Attribute::Index::Attribute_31; -} - -/// Returns the location of a generic attribute -u32 GetGenericAttributeLocation(Attribute::Index attribute) { - ASSERT(IsGenericAttribute(attribute)); - return static_cast<u32>(attribute) - static_cast<u32>(Attribute::Index::Attribute_0); -} - -/// Returns true if an object has to be treated as precise -bool IsPrecise(Operation operand) { - const auto& meta{operand.GetMeta()}; - if (std::holds_alternative<MetaArithmetic>(meta)) { - return std::get<MetaArithmetic>(meta).precise; - } - return false; -} - -class SPIRVDecompiler final : public Sirit::Module { -public: - explicit SPIRVDecompiler(const Device& device_, const ShaderIR& ir_, ShaderType stage_, - const Registry& registry_, const Specialization& specialization_) - : Module(0x00010300), device{device_}, ir{ir_}, stage{stage_}, header{ir_.GetHeader()}, - registry{registry_}, specialization{specialization_} { - if (stage_ != ShaderType::Compute) { - transform_feedback = BuildTransformFeedback(registry_.GetGraphicsInfo()); - } - - AddCapability(spv::Capability::Shader); - AddCapability(spv::Capability::UniformAndStorageBuffer16BitAccess); - AddCapability(spv::Capability::ImageQuery); - AddCapability(spv::Capability::Image1D); - AddCapability(spv::Capability::ImageBuffer); - AddCapability(spv::Capability::ImageGatherExtended); - AddCapability(spv::Capability::SampledBuffer); - AddCapability(spv::Capability::StorageImageWriteWithoutFormat); - AddCapability(spv::Capability::DrawParameters); - AddCapability(spv::Capability::SubgroupBallotKHR); - AddCapability(spv::Capability::SubgroupVoteKHR); - AddExtension("SPV_KHR_16bit_storage"); - AddExtension("SPV_KHR_shader_ballot"); - AddExtension("SPV_KHR_subgroup_vote"); - AddExtension("SPV_KHR_storage_buffer_storage_class"); - AddExtension("SPV_KHR_variable_pointers"); - AddExtension("SPV_KHR_shader_draw_parameters"); - - if (!transform_feedback.empty()) { - if (device.IsExtTransformFeedbackSupported()) { - AddCapability(spv::Capability::TransformFeedback); - } else { - LOG_ERROR(Render_Vulkan, "Shader requires transform feedbacks but these are not " - "supported on this device"); - } - } - if (ir.UsesLayer() || ir.UsesViewportIndex()) { - if (ir.UsesViewportIndex()) { - AddCapability(spv::Capability::MultiViewport); - } - if (stage != ShaderType::Geometry && device.IsExtShaderViewportIndexLayerSupported()) { - AddExtension("SPV_EXT_shader_viewport_index_layer"); - AddCapability(spv::Capability::ShaderViewportIndexLayerEXT); - } - } - if (device.IsFormatlessImageLoadSupported()) { - AddCapability(spv::Capability::StorageImageReadWithoutFormat); - } - if (device.IsFloat16Supported()) { - AddCapability(spv::Capability::Float16); - } - t_scalar_half = Name(TypeFloat(device_.IsFloat16Supported() ? 16 : 32), "scalar_half"); - t_half = Name(TypeVector(t_scalar_half, 2), "half"); - - const Id main = Decompile(); - - switch (stage) { - case ShaderType::Vertex: - AddEntryPoint(spv::ExecutionModel::Vertex, main, "main", interfaces); - break; - case ShaderType::TesselationControl: - AddCapability(spv::Capability::Tessellation); - AddEntryPoint(spv::ExecutionModel::TessellationControl, main, "main", interfaces); - AddExecutionMode(main, spv::ExecutionMode::OutputVertices, - header.common2.threads_per_input_primitive); - break; - case ShaderType::TesselationEval: { - const auto& info = registry.GetGraphicsInfo(); - AddCapability(spv::Capability::Tessellation); - AddEntryPoint(spv::ExecutionModel::TessellationEvaluation, main, "main", interfaces); - AddExecutionMode(main, GetExecutionMode(info.tessellation_primitive)); - AddExecutionMode(main, GetExecutionMode(info.tessellation_spacing)); - AddExecutionMode(main, info.tessellation_clockwise - ? spv::ExecutionMode::VertexOrderCw - : spv::ExecutionMode::VertexOrderCcw); - break; - } - case ShaderType::Geometry: { - const auto& info = registry.GetGraphicsInfo(); - AddCapability(spv::Capability::Geometry); - AddEntryPoint(spv::ExecutionModel::Geometry, main, "main", interfaces); - AddExecutionMode(main, GetExecutionMode(info.primitive_topology)); - AddExecutionMode(main, GetExecutionMode(header.common3.output_topology)); - AddExecutionMode(main, spv::ExecutionMode::OutputVertices, - header.common4.max_output_vertices); - // TODO(Rodrigo): Where can we get this info from? - AddExecutionMode(main, spv::ExecutionMode::Invocations, 1U); - break; - } - case ShaderType::Fragment: - AddEntryPoint(spv::ExecutionModel::Fragment, main, "main", interfaces); - AddExecutionMode(main, spv::ExecutionMode::OriginUpperLeft); - if (header.ps.omap.depth) { - AddExecutionMode(main, spv::ExecutionMode::DepthReplacing); - } - if (specialization.early_fragment_tests) { - AddExecutionMode(main, spv::ExecutionMode::EarlyFragmentTests); - } - break; - case ShaderType::Compute: - const auto workgroup_size = specialization.workgroup_size; - AddExecutionMode(main, spv::ExecutionMode::LocalSize, workgroup_size[0], - workgroup_size[1], workgroup_size[2]); - AddEntryPoint(spv::ExecutionModel::GLCompute, main, "main", interfaces); - break; - } - } - -private: - Id Decompile() { - DeclareCommon(); - DeclareVertex(); - DeclareTessControl(); - DeclareTessEval(); - DeclareGeometry(); - DeclareFragment(); - DeclareCompute(); - DeclareRegisters(); - DeclareCustomVariables(); - DeclarePredicates(); - DeclareLocalMemory(); - DeclareSharedMemory(); - DeclareInternalFlags(); - DeclareInputAttributes(); - DeclareOutputAttributes(); - - u32 binding = specialization.base_binding; - binding = DeclareConstantBuffers(binding); - binding = DeclareGlobalBuffers(binding); - binding = DeclareUniformTexels(binding); - binding = DeclareSamplers(binding); - binding = DeclareStorageTexels(binding); - binding = DeclareImages(binding); - - const Id main = OpFunction(t_void, {}, TypeFunction(t_void)); - AddLabel(); - - if (ir.IsDecompiled()) { - DeclareFlowVariables(); - DecompileAST(); - } else { - AllocateLabels(); - DecompileBranchMode(); - } - - OpReturn(); - OpFunctionEnd(); - - return main; - } - - void DefinePrologue() { - if (stage == ShaderType::Vertex) { - // Clear Position to avoid reading trash on the Z conversion. - const auto position_index = out_indices.position.value(); - const Id position = AccessElement(t_out_float4, out_vertex, position_index); - OpStore(position, v_varying_default); - - if (specialization.point_size) { - const u32 point_size_index = out_indices.point_size.value(); - const Id out_point_size = AccessElement(t_out_float, out_vertex, point_size_index); - OpStore(out_point_size, Constant(t_float, *specialization.point_size)); - } - } - } - - void DecompileAST(); - - void DecompileBranchMode() { - const u32 first_address = ir.GetBasicBlocks().begin()->first; - const Id loop_label = OpLabel("loop"); - const Id merge_label = OpLabel("merge"); - const Id dummy_label = OpLabel(); - const Id jump_label = OpLabel(); - continue_label = OpLabel("continue"); - - std::vector<Sirit::Literal> literals; - std::vector<Id> branch_labels; - for (const auto& [literal, label] : labels) { - literals.push_back(literal); - branch_labels.push_back(label); - } - - jmp_to = OpVariable(TypePointer(spv::StorageClass::Function, t_uint), - spv::StorageClass::Function, Constant(t_uint, first_address)); - AddLocalVariable(jmp_to); - - std::tie(ssy_flow_stack, ssy_flow_stack_top) = CreateFlowStack(); - std::tie(pbk_flow_stack, pbk_flow_stack_top) = CreateFlowStack(); - - Name(jmp_to, "jmp_to"); - Name(ssy_flow_stack, "ssy_flow_stack"); - Name(ssy_flow_stack_top, "ssy_flow_stack_top"); - Name(pbk_flow_stack, "pbk_flow_stack"); - Name(pbk_flow_stack_top, "pbk_flow_stack_top"); - - DefinePrologue(); - - OpBranch(loop_label); - AddLabel(loop_label); - OpLoopMerge(merge_label, continue_label, spv::LoopControlMask::MaskNone); - OpBranch(dummy_label); - - AddLabel(dummy_label); - const Id default_branch = OpLabel(); - const Id jmp_to_load = OpLoad(t_uint, jmp_to); - OpSelectionMerge(jump_label, spv::SelectionControlMask::MaskNone); - OpSwitch(jmp_to_load, default_branch, literals, branch_labels); - - AddLabel(default_branch); - OpReturn(); - - for (const auto& [address, bb] : ir.GetBasicBlocks()) { - AddLabel(labels.at(address)); - - VisitBasicBlock(bb); - - const auto next_it = labels.lower_bound(address + 1); - const Id next_label = next_it != labels.end() ? next_it->second : default_branch; - OpBranch(next_label); - } - - AddLabel(jump_label); - OpBranch(continue_label); - AddLabel(continue_label); - OpBranch(loop_label); - AddLabel(merge_label); - } - -private: - friend class ASTDecompiler; - friend class ExprDecompiler; - - static constexpr auto INTERNAL_FLAGS_COUNT = static_cast<std::size_t>(InternalFlag::Amount); - - void AllocateLabels() { - for (const auto& pair : ir.GetBasicBlocks()) { - const u32 address = pair.first; - labels.emplace(address, OpLabel(fmt::format("label_0x{:x}", address))); - } - } - - void DeclareCommon() { - thread_id = - DeclareInputBuiltIn(spv::BuiltIn::SubgroupLocalInvocationId, t_in_uint, "thread_id"); - thread_masks[0] = - DeclareInputBuiltIn(spv::BuiltIn::SubgroupEqMask, t_in_uint4, "thread_eq_mask"); - thread_masks[1] = - DeclareInputBuiltIn(spv::BuiltIn::SubgroupGeMask, t_in_uint4, "thread_ge_mask"); - thread_masks[2] = - DeclareInputBuiltIn(spv::BuiltIn::SubgroupGtMask, t_in_uint4, "thread_gt_mask"); - thread_masks[3] = - DeclareInputBuiltIn(spv::BuiltIn::SubgroupLeMask, t_in_uint4, "thread_le_mask"); - thread_masks[4] = - DeclareInputBuiltIn(spv::BuiltIn::SubgroupLtMask, t_in_uint4, "thread_lt_mask"); - } - - void DeclareVertex() { - if (stage != ShaderType::Vertex) { - return; - } - Id out_vertex_struct; - std::tie(out_vertex_struct, out_indices) = DeclareVertexStruct(); - const Id vertex_ptr = TypePointer(spv::StorageClass::Output, out_vertex_struct); - out_vertex = OpVariable(vertex_ptr, spv::StorageClass::Output); - interfaces.push_back(AddGlobalVariable(Name(out_vertex, "out_vertex"))); - - // Declare input attributes - vertex_index = DeclareInputBuiltIn(spv::BuiltIn::VertexIndex, t_in_int, "vertex_index"); - instance_index = - DeclareInputBuiltIn(spv::BuiltIn::InstanceIndex, t_in_int, "instance_index"); - base_vertex = DeclareInputBuiltIn(spv::BuiltIn::BaseVertex, t_in_int, "base_vertex"); - base_instance = DeclareInputBuiltIn(spv::BuiltIn::BaseInstance, t_in_int, "base_instance"); - } - - void DeclareTessControl() { - if (stage != ShaderType::TesselationControl) { - return; - } - DeclareInputVertexArray(NumInputPatches); - DeclareOutputVertexArray(header.common2.threads_per_input_primitive); - - tess_level_outer = DeclareBuiltIn( - spv::BuiltIn::TessLevelOuter, spv::StorageClass::Output, - TypePointer(spv::StorageClass::Output, TypeArray(t_float, Constant(t_uint, 4U))), - "tess_level_outer"); - Decorate(tess_level_outer, spv::Decoration::Patch); - - tess_level_inner = DeclareBuiltIn( - spv::BuiltIn::TessLevelInner, spv::StorageClass::Output, - TypePointer(spv::StorageClass::Output, TypeArray(t_float, Constant(t_uint, 2U))), - "tess_level_inner"); - Decorate(tess_level_inner, spv::Decoration::Patch); - - invocation_id = DeclareInputBuiltIn(spv::BuiltIn::InvocationId, t_in_int, "invocation_id"); - } - - void DeclareTessEval() { - if (stage != ShaderType::TesselationEval) { - return; - } - DeclareInputVertexArray(NumInputPatches); - DeclareOutputVertex(); - - tess_coord = DeclareInputBuiltIn(spv::BuiltIn::TessCoord, t_in_float3, "tess_coord"); - } - - void DeclareGeometry() { - if (stage != ShaderType::Geometry) { - return; - } - const auto& info = registry.GetGraphicsInfo(); - const u32 num_input = GetNumPrimitiveTopologyVertices(info.primitive_topology); - DeclareInputVertexArray(num_input); - DeclareOutputVertex(); - } - - void DeclareFragment() { - if (stage != ShaderType::Fragment) { - return; - } - - for (u32 rt = 0; rt < static_cast<u32>(std::size(frag_colors)); ++rt) { - if (!IsRenderTargetEnabled(rt)) { - continue; - } - const Id id = AddGlobalVariable(OpVariable(t_out_float4, spv::StorageClass::Output)); - Name(id, fmt::format("frag_color{}", rt)); - Decorate(id, spv::Decoration::Location, rt); - - frag_colors[rt] = id; - interfaces.push_back(id); - } - - if (header.ps.omap.depth) { - frag_depth = AddGlobalVariable(OpVariable(t_out_float, spv::StorageClass::Output)); - Name(frag_depth, "frag_depth"); - Decorate(frag_depth, spv::Decoration::BuiltIn, - static_cast<u32>(spv::BuiltIn::FragDepth)); - - interfaces.push_back(frag_depth); - } - - frag_coord = DeclareInputBuiltIn(spv::BuiltIn::FragCoord, t_in_float4, "frag_coord"); - front_facing = DeclareInputBuiltIn(spv::BuiltIn::FrontFacing, t_in_bool, "front_facing"); - point_coord = DeclareInputBuiltIn(spv::BuiltIn::PointCoord, t_in_float2, "point_coord"); - } - - void DeclareCompute() { - if (stage != ShaderType::Compute) { - return; - } - - workgroup_id = DeclareInputBuiltIn(spv::BuiltIn::WorkgroupId, t_in_uint3, "workgroup_id"); - local_invocation_id = - DeclareInputBuiltIn(spv::BuiltIn::LocalInvocationId, t_in_uint3, "local_invocation_id"); - } - - void DeclareRegisters() { - for (const u32 gpr : ir.GetRegisters()) { - const Id id = OpVariable(t_prv_float, spv::StorageClass::Private, v_float_zero); - Name(id, fmt::format("gpr_{}", gpr)); - registers.emplace(gpr, AddGlobalVariable(id)); - } - } - - void DeclareCustomVariables() { - const u32 num_custom_variables = ir.GetNumCustomVariables(); - for (u32 i = 0; i < num_custom_variables; ++i) { - const Id id = OpVariable(t_prv_float, spv::StorageClass::Private, v_float_zero); - Name(id, fmt::format("custom_var_{}", i)); - custom_variables.emplace(i, AddGlobalVariable(id)); - } - } - - void DeclarePredicates() { - for (const auto pred : ir.GetPredicates()) { - const Id id = OpVariable(t_prv_bool, spv::StorageClass::Private, v_false); - Name(id, fmt::format("pred_{}", static_cast<u32>(pred))); - predicates.emplace(pred, AddGlobalVariable(id)); - } - } - - void DeclareFlowVariables() { - for (u32 i = 0; i < ir.GetASTNumVariables(); i++) { - const Id id = OpVariable(t_prv_bool, spv::StorageClass::Private, v_false); - Name(id, fmt::format("flow_var_{}", static_cast<u32>(i))); - flow_variables.emplace(i, AddGlobalVariable(id)); - } - } - - void DeclareLocalMemory() { - // TODO(Rodrigo): Unstub kernel local memory size and pass it from a register at - // specialization time. - const u64 lmem_size = stage == ShaderType::Compute ? 0x400 : header.GetLocalMemorySize(); - if (lmem_size == 0) { - return; - } - const auto element_count = static_cast<u32>(Common::AlignUp(lmem_size, 4) / 4); - const Id type_array = TypeArray(t_float, Constant(t_uint, element_count)); - const Id type_pointer = TypePointer(spv::StorageClass::Private, type_array); - Name(type_pointer, "LocalMemory"); - - local_memory = - OpVariable(type_pointer, spv::StorageClass::Private, ConstantNull(type_array)); - AddGlobalVariable(Name(local_memory, "local_memory")); - } - - void DeclareSharedMemory() { - if (stage != ShaderType::Compute) { - return; - } - t_smem_uint = TypePointer(spv::StorageClass::Workgroup, t_uint); - - u32 smem_size = specialization.shared_memory_size * 4; - if (smem_size == 0) { - // Avoid declaring an empty array. - return; - } - const u32 limit = device.GetMaxComputeSharedMemorySize(); - if (smem_size > limit) { - LOG_ERROR(Render_Vulkan, "Shared memory size {} is clamped to host's limit {}", - smem_size, limit); - smem_size = limit; - } - - const Id type_array = TypeArray(t_uint, Constant(t_uint, smem_size / 4)); - const Id type_pointer = TypePointer(spv::StorageClass::Workgroup, type_array); - Name(type_pointer, "SharedMemory"); - - shared_memory = OpVariable(type_pointer, spv::StorageClass::Workgroup); - AddGlobalVariable(Name(shared_memory, "shared_memory")); - } - - void DeclareInternalFlags() { - static constexpr std::array names{"zero", "sign", "carry", "overflow"}; - - for (std::size_t flag = 0; flag < INTERNAL_FLAGS_COUNT; ++flag) { - const Id id = OpVariable(t_prv_bool, spv::StorageClass::Private, v_false); - internal_flags[flag] = AddGlobalVariable(Name(id, names[flag])); - } - } - - void DeclareInputVertexArray(u32 length) { - constexpr auto storage = spv::StorageClass::Input; - std::tie(in_indices, in_vertex) = DeclareVertexArray(storage, "in_indices", length); - } - - void DeclareOutputVertexArray(u32 length) { - constexpr auto storage = spv::StorageClass::Output; - std::tie(out_indices, out_vertex) = DeclareVertexArray(storage, "out_indices", length); - } - - std::tuple<VertexIndices, Id> DeclareVertexArray(spv::StorageClass storage_class, - std::string name, u32 length) { - const auto [struct_id, indices] = DeclareVertexStruct(); - const Id vertex_array = TypeArray(struct_id, Constant(t_uint, length)); - const Id vertex_ptr = TypePointer(storage_class, vertex_array); - const Id vertex = OpVariable(vertex_ptr, storage_class); - AddGlobalVariable(Name(vertex, std::move(name))); - interfaces.push_back(vertex); - return {indices, vertex}; - } - - void DeclareOutputVertex() { - Id out_vertex_struct; - std::tie(out_vertex_struct, out_indices) = DeclareVertexStruct(); - const Id out_vertex_ptr = TypePointer(spv::StorageClass::Output, out_vertex_struct); - out_vertex = OpVariable(out_vertex_ptr, spv::StorageClass::Output); - interfaces.push_back(AddGlobalVariable(Name(out_vertex, "out_vertex"))); - } - - void DeclareInputAttributes() { - for (const auto index : ir.GetInputAttributes()) { - if (!IsGenericAttribute(index)) { - continue; - } - const u32 location = GetGenericAttributeLocation(index); - if (!IsAttributeEnabled(location)) { - continue; - } - const auto type_descriptor = GetAttributeType(location); - Id type; - if (IsInputAttributeArray()) { - type = GetTypeVectorDefinitionLut(type_descriptor.type).at(3); - type = TypeArray(type, Constant(t_uint, GetNumInputVertices())); - type = TypePointer(spv::StorageClass::Input, type); - } else { - type = type_descriptor.vector; - } - const Id id = OpVariable(type, spv::StorageClass::Input); - AddGlobalVariable(Name(id, fmt::format("in_attr{}", location))); - input_attributes.emplace(index, id); - interfaces.push_back(id); - - Decorate(id, spv::Decoration::Location, location); - - if (stage != ShaderType::Fragment) { - continue; - } - switch (header.ps.GetPixelImap(location)) { - case PixelImap::Constant: - Decorate(id, spv::Decoration::Flat); - break; - case PixelImap::Perspective: - // Default - break; - case PixelImap::ScreenLinear: - Decorate(id, spv::Decoration::NoPerspective); - break; - default: - UNREACHABLE_MSG("Unused attribute being fetched"); - } - } - } - - void DeclareOutputAttributes() { - if (stage == ShaderType::Compute || stage == ShaderType::Fragment) { - return; - } - - UNIMPLEMENTED_IF(registry.GetGraphicsInfo().tfb_enabled && stage != ShaderType::Vertex); - for (const auto index : ir.GetOutputAttributes()) { - if (!IsGenericAttribute(index)) { - continue; - } - DeclareOutputAttribute(index); - } - } - - void DeclareOutputAttribute(Attribute::Index index) { - static constexpr std::string_view swizzle = "xyzw"; - - const u32 location = GetGenericAttributeLocation(index); - u8 element = 0; - while (element < 4) { - const std::size_t remainder = 4 - element; - - std::size_t num_components = remainder; - const std::optional tfb = GetTransformFeedbackInfo(index, element); - if (tfb) { - num_components = tfb->components; - } - - Id type = GetTypeVectorDefinitionLut(Type::Float).at(num_components - 1); - Id varying_default = v_varying_default; - if (IsOutputAttributeArray()) { - const u32 num = GetNumOutputVertices(); - type = TypeArray(type, Constant(t_uint, num)); - if (device.GetDriverID() != VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS_KHR) { - // Intel's proprietary driver fails to setup defaults for arrayed output - // attributes. - varying_default = ConstantComposite(type, std::vector(num, varying_default)); - } - } - type = TypePointer(spv::StorageClass::Output, type); - - std::string name = fmt::format("out_attr{}", location); - if (num_components < 4 || element > 0) { - name = fmt::format("{}_{}", name, swizzle.substr(element, num_components)); - } - - const Id id = OpVariable(type, spv::StorageClass::Output, varying_default); - Name(AddGlobalVariable(id), name); - - GenericVaryingDescription description; - description.id = id; - description.first_element = element; - description.is_scalar = num_components == 1; - for (u32 i = 0; i < num_components; ++i) { - const u8 offset = static_cast<u8>(static_cast<u32>(index) * 4 + element + i); - output_attributes.emplace(offset, description); - } - interfaces.push_back(id); - - Decorate(id, spv::Decoration::Location, location); - if (element > 0) { - Decorate(id, spv::Decoration::Component, static_cast<u32>(element)); - } - if (tfb && device.IsExtTransformFeedbackSupported()) { - Decorate(id, spv::Decoration::XfbBuffer, static_cast<u32>(tfb->buffer)); - Decorate(id, spv::Decoration::XfbStride, static_cast<u32>(tfb->stride)); - Decorate(id, spv::Decoration::Offset, static_cast<u32>(tfb->offset)); - } - - element = static_cast<u8>(static_cast<std::size_t>(element) + num_components); - } - } - - std::optional<VaryingTFB> GetTransformFeedbackInfo(Attribute::Index index, u8 element = 0) { - const u8 location = static_cast<u8>(static_cast<u32>(index) * 4 + element); - const auto it = transform_feedback.find(location); - if (it == transform_feedback.end()) { - return {}; - } - return it->second; - } - - u32 DeclareConstantBuffers(u32 binding) { - for (const auto& [index, size] : ir.GetConstantBuffers()) { - const Id type = device.IsKhrUniformBufferStandardLayoutSupported() ? t_cbuf_scalar_ubo - : t_cbuf_std140_ubo; - const Id id = OpVariable(type, spv::StorageClass::Uniform); - AddGlobalVariable(Name(id, fmt::format("cbuf_{}", index))); - - Decorate(id, spv::Decoration::Binding, binding++); - Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET); - constant_buffers.emplace(index, id); - } - return binding; - } - - u32 DeclareGlobalBuffers(u32 binding) { - for (const auto& [base, usage] : ir.GetGlobalMemory()) { - const Id id = OpVariable(t_gmem_ssbo, spv::StorageClass::StorageBuffer); - AddGlobalVariable( - Name(id, fmt::format("gmem_{}_{}", base.cbuf_index, base.cbuf_offset))); - - Decorate(id, spv::Decoration::Binding, binding++); - Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET); - global_buffers.emplace(base, id); - } - return binding; - } - - u32 DeclareUniformTexels(u32 binding) { - for (const auto& sampler : ir.GetSamplers()) { - if (!sampler.is_buffer) { - continue; - } - ASSERT(!sampler.is_array); - ASSERT(!sampler.is_shadow); - - constexpr auto dim = spv::Dim::Buffer; - constexpr int depth = 0; - constexpr int arrayed = 0; - constexpr bool ms = false; - constexpr int sampled = 1; - constexpr auto format = spv::ImageFormat::Unknown; - const Id image_type = TypeImage(t_float, dim, depth, arrayed, ms, sampled, format); - const Id pointer_type = TypePointer(spv::StorageClass::UniformConstant, image_type); - const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant); - AddGlobalVariable(Name(id, fmt::format("sampler_{}", sampler.index))); - Decorate(id, spv::Decoration::Binding, binding++); - Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET); - - uniform_texels.emplace(sampler.index, TexelBuffer{image_type, id}); - } - return binding; - } - - u32 DeclareSamplers(u32 binding) { - for (const auto& sampler : ir.GetSamplers()) { - if (sampler.is_buffer) { - continue; - } - const auto dim = GetSamplerDim(sampler); - const int depth = sampler.is_shadow ? 1 : 0; - const int arrayed = sampler.is_array ? 1 : 0; - constexpr bool ms = false; - constexpr int sampled = 1; - constexpr auto format = spv::ImageFormat::Unknown; - const Id image_type = TypeImage(t_float, dim, depth, arrayed, ms, sampled, format); - const Id sampler_type = TypeSampledImage(image_type); - const Id sampler_pointer_type = - TypePointer(spv::StorageClass::UniformConstant, sampler_type); - const Id type = sampler.is_indexed - ? TypeArray(sampler_type, Constant(t_uint, sampler.size)) - : sampler_type; - const Id pointer_type = TypePointer(spv::StorageClass::UniformConstant, type); - const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant); - AddGlobalVariable(Name(id, fmt::format("sampler_{}", sampler.index))); - Decorate(id, spv::Decoration::Binding, binding++); - Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET); - - sampled_images.emplace( - sampler.index, SampledImage{image_type, sampler_type, sampler_pointer_type, id}); - } - return binding; - } - - u32 DeclareStorageTexels(u32 binding) { - for (const auto& image : ir.GetImages()) { - if (image.type != Tegra::Shader::ImageType::TextureBuffer) { - continue; - } - DeclareImage(image, binding); - } - return binding; - } - - u32 DeclareImages(u32 binding) { - for (const auto& image : ir.GetImages()) { - if (image.type == Tegra::Shader::ImageType::TextureBuffer) { - continue; - } - DeclareImage(image, binding); - } - return binding; - } - - void DeclareImage(const ImageEntry& image, u32& binding) { - const auto [dim, arrayed] = GetImageDim(image); - constexpr int depth = 0; - constexpr bool ms = false; - constexpr int sampled = 2; // This won't be accessed with a sampler - const auto format = image.is_atomic ? spv::ImageFormat::R32ui : spv::ImageFormat::Unknown; - const Id image_type = TypeImage(t_uint, dim, depth, arrayed, ms, sampled, format, {}); - const Id pointer_type = TypePointer(spv::StorageClass::UniformConstant, image_type); - const Id id = OpVariable(pointer_type, spv::StorageClass::UniformConstant); - AddGlobalVariable(Name(id, fmt::format("image_{}", image.index))); - - Decorate(id, spv::Decoration::Binding, binding++); - Decorate(id, spv::Decoration::DescriptorSet, DESCRIPTOR_SET); - if (image.is_read && !image.is_written) { - Decorate(id, spv::Decoration::NonWritable); - } else if (image.is_written && !image.is_read) { - Decorate(id, spv::Decoration::NonReadable); - } - - images.emplace(image.index, StorageImage{image_type, id}); - } - - bool IsRenderTargetEnabled(u32 rt) const { - for (u32 component = 0; component < 4; ++component) { - if (header.ps.IsColorComponentOutputEnabled(rt, component)) { - return true; - } - } - return false; - } - - bool IsInputAttributeArray() const { - return stage == ShaderType::TesselationControl || stage == ShaderType::TesselationEval || - stage == ShaderType::Geometry; - } - - bool IsOutputAttributeArray() const { - return stage == ShaderType::TesselationControl; - } - - bool IsAttributeEnabled(u32 location) const { - return stage != ShaderType::Vertex || specialization.enabled_attributes[location]; - } - - u32 GetNumInputVertices() const { - switch (stage) { - case ShaderType::Geometry: - return GetNumPrimitiveTopologyVertices(registry.GetGraphicsInfo().primitive_topology); - case ShaderType::TesselationControl: - case ShaderType::TesselationEval: - return NumInputPatches; - default: - UNREACHABLE(); - return 1; - } - } - - u32 GetNumOutputVertices() const { - switch (stage) { - case ShaderType::TesselationControl: - return header.common2.threads_per_input_primitive; - default: - UNREACHABLE(); - return 1; - } - } - - std::tuple<Id, VertexIndices> DeclareVertexStruct() { - struct BuiltIn { - Id type; - spv::BuiltIn builtin; - const char* name; - }; - std::vector<BuiltIn> members; - members.reserve(4); - - const auto AddBuiltIn = [&](Id type, spv::BuiltIn builtin, const char* name) { - const auto index = static_cast<u32>(members.size()); - members.push_back(BuiltIn{type, builtin, name}); - return index; - }; - - VertexIndices indices; - indices.position = AddBuiltIn(t_float4, spv::BuiltIn::Position, "position"); - - if (ir.UsesLayer()) { - if (stage != ShaderType::Vertex || device.IsExtShaderViewportIndexLayerSupported()) { - indices.layer = AddBuiltIn(t_int, spv::BuiltIn::Layer, "layer"); - } else { - LOG_ERROR( - Render_Vulkan, - "Shader requires Layer but it's not supported on this stage with this device."); - } - } - - if (ir.UsesViewportIndex()) { - if (stage != ShaderType::Vertex || device.IsExtShaderViewportIndexLayerSupported()) { - indices.viewport = AddBuiltIn(t_int, spv::BuiltIn::ViewportIndex, "viewport_index"); - } else { - LOG_ERROR(Render_Vulkan, "Shader requires ViewportIndex but it's not supported on " - "this stage with this device."); - } - } - - if (ir.UsesPointSize() || specialization.point_size) { - indices.point_size = AddBuiltIn(t_float, spv::BuiltIn::PointSize, "point_size"); - } - - const auto& ir_output_attributes = ir.GetOutputAttributes(); - const bool declare_clip_distances = std::any_of( - ir_output_attributes.begin(), ir_output_attributes.end(), [](const auto& index) { - return index == Attribute::Index::ClipDistances0123 || - index == Attribute::Index::ClipDistances4567; - }); - if (declare_clip_distances) { - indices.clip_distances = AddBuiltIn(TypeArray(t_float, Constant(t_uint, 8)), - spv::BuiltIn::ClipDistance, "clip_distances"); - } - - std::vector<Id> member_types; - member_types.reserve(members.size()); - for (std::size_t i = 0; i < members.size(); ++i) { - member_types.push_back(members[i].type); - } - const Id per_vertex_struct = Name(TypeStruct(member_types), "PerVertex"); - Decorate(per_vertex_struct, spv::Decoration::Block); - - for (std::size_t index = 0; index < members.size(); ++index) { - const auto& member = members[index]; - MemberName(per_vertex_struct, static_cast<u32>(index), member.name); - MemberDecorate(per_vertex_struct, static_cast<u32>(index), spv::Decoration::BuiltIn, - static_cast<u32>(member.builtin)); - } - - return {per_vertex_struct, indices}; - } - - void VisitBasicBlock(const NodeBlock& bb) { - for (const auto& node : bb) { - Visit(node); - } - } - - Expression Visit(const Node& node) { - if (const auto operation = std::get_if<OperationNode>(&*node)) { - if (const auto amend_index = operation->GetAmendIndex()) { - [[maybe_unused]] const Type type = Visit(ir.GetAmendNode(*amend_index)).type; - ASSERT(type == Type::Void); - } - const auto operation_index = static_cast<std::size_t>(operation->GetCode()); - const auto decompiler = operation_decompilers[operation_index]; - if (decompiler == nullptr) { - UNREACHABLE_MSG("Operation decompiler {} not defined", operation_index); - } - return (this->*decompiler)(*operation); - } - - if (const auto gpr = std::get_if<GprNode>(&*node)) { - const u32 index = gpr->GetIndex(); - if (index == Register::ZeroIndex) { - return {v_float_zero, Type::Float}; - } - return {OpLoad(t_float, registers.at(index)), Type::Float}; - } - - if (const auto cv = std::get_if<CustomVarNode>(&*node)) { - const u32 index = cv->GetIndex(); - return {OpLoad(t_float, custom_variables.at(index)), Type::Float}; - } - - if (const auto immediate = std::get_if<ImmediateNode>(&*node)) { - return {Constant(t_uint, immediate->GetValue()), Type::Uint}; - } - - if (const auto predicate = std::get_if<PredicateNode>(&*node)) { - const auto value = [&]() -> Id { - switch (const auto index = predicate->GetIndex(); index) { - case Tegra::Shader::Pred::UnusedIndex: - return v_true; - case Tegra::Shader::Pred::NeverExecute: - return v_false; - default: - return OpLoad(t_bool, predicates.at(index)); - } - }(); - if (predicate->IsNegated()) { - return {OpLogicalNot(t_bool, value), Type::Bool}; - } - return {value, Type::Bool}; - } - - if (const auto abuf = std::get_if<AbufNode>(&*node)) { - const auto attribute = abuf->GetIndex(); - const u32 element = abuf->GetElement(); - const auto& buffer = abuf->GetBuffer(); - - const auto ArrayPass = [&](Id pointer_type, Id composite, std::vector<u32> indices) { - std::vector<Id> members; - members.reserve(std::size(indices) + 1); - - if (buffer && IsInputAttributeArray()) { - members.push_back(AsUint(Visit(buffer))); - } - for (const u32 index : indices) { - members.push_back(Constant(t_uint, index)); - } - return OpAccessChain(pointer_type, composite, members); - }; - - switch (attribute) { - case Attribute::Index::Position: { - if (stage == ShaderType::Fragment) { - return {OpLoad(t_float, AccessElement(t_in_float, frag_coord, element)), - Type::Float}; - } - const std::vector elements = {in_indices.position.value(), element}; - return {OpLoad(t_float, ArrayPass(t_in_float, in_vertex, elements)), Type::Float}; - } - case Attribute::Index::PointCoord: { - switch (element) { - case 0: - case 1: - return {OpCompositeExtract(t_float, OpLoad(t_float2, point_coord), element), - Type::Float}; - } - UNIMPLEMENTED_MSG("Unimplemented point coord element={}", element); - return {v_float_zero, Type::Float}; - } - case Attribute::Index::TessCoordInstanceIDVertexID: - // TODO(Subv): Find out what the values are for the first two elements when inside a - // vertex shader, and what's the value of the fourth element when inside a Tess Eval - // shader. - switch (element) { - case 0: - case 1: - return {OpLoad(t_float, AccessElement(t_in_float, tess_coord, element)), - Type::Float}; - case 2: - return { - OpISub(t_int, OpLoad(t_int, instance_index), OpLoad(t_int, base_instance)), - Type::Int}; - case 3: - return {OpISub(t_int, OpLoad(t_int, vertex_index), OpLoad(t_int, base_vertex)), - Type::Int}; - } - UNIMPLEMENTED_MSG("Unmanaged TessCoordInstanceIDVertexID element={}", element); - return {Constant(t_uint, 0U), Type::Uint}; - case Attribute::Index::FrontFacing: - // TODO(Subv): Find out what the values are for the other elements. - ASSERT(stage == ShaderType::Fragment); - if (element == 3) { - const Id is_front_facing = OpLoad(t_bool, front_facing); - const Id true_value = Constant(t_int, static_cast<s32>(-1)); - const Id false_value = Constant(t_int, 0); - return {OpSelect(t_int, is_front_facing, true_value, false_value), Type::Int}; - } - UNIMPLEMENTED_MSG("Unmanaged FrontFacing element={}", element); - return {v_float_zero, Type::Float}; - default: - if (!IsGenericAttribute(attribute)) { - break; - } - const u32 location = GetGenericAttributeLocation(attribute); - if (!IsAttributeEnabled(location)) { - // Disabled attributes (also known as constant attributes) always return zero. - return {v_float_zero, Type::Float}; - } - const auto type_descriptor = GetAttributeType(location); - const Type type = type_descriptor.type; - const Id attribute_id = input_attributes.at(attribute); - const std::vector elements = {element}; - const Id pointer = ArrayPass(type_descriptor.scalar, attribute_id, elements); - return {OpLoad(GetTypeDefinition(type), pointer), type}; - } - UNIMPLEMENTED_MSG("Unhandled input attribute: {}", attribute); - return {v_float_zero, Type::Float}; - } - - if (const auto cbuf = std::get_if<CbufNode>(&*node)) { - const Node& offset = cbuf->GetOffset(); - const Id buffer_id = constant_buffers.at(cbuf->GetIndex()); - - Id pointer{}; - if (device.IsKhrUniformBufferStandardLayoutSupported()) { - const Id buffer_offset = - OpShiftRightLogical(t_uint, AsUint(Visit(offset)), Constant(t_uint, 2U)); - pointer = - OpAccessChain(t_cbuf_float, buffer_id, Constant(t_uint, 0U), buffer_offset); - } else { - Id buffer_index{}; - Id buffer_element{}; - if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) { - // Direct access - const u32 offset_imm = immediate->GetValue(); - ASSERT(offset_imm % 4 == 0); - buffer_index = Constant(t_uint, offset_imm / 16); - buffer_element = Constant(t_uint, (offset_imm / 4) % 4); - } else if (std::holds_alternative<OperationNode>(*offset)) { - // Indirect access - const Id offset_id = AsUint(Visit(offset)); - const Id unsafe_offset = OpUDiv(t_uint, offset_id, Constant(t_uint, 4)); - const Id final_offset = - OpUMod(t_uint, unsafe_offset, Constant(t_uint, MaxConstBufferElements - 1)); - buffer_index = OpUDiv(t_uint, final_offset, Constant(t_uint, 4)); - buffer_element = OpUMod(t_uint, final_offset, Constant(t_uint, 4)); - } else { - UNREACHABLE_MSG("Unmanaged offset node type"); - } - pointer = OpAccessChain(t_cbuf_float, buffer_id, v_uint_zero, buffer_index, - buffer_element); - } - return {OpLoad(t_float, pointer), Type::Float}; - } - - if (const auto gmem = std::get_if<GmemNode>(&*node)) { - return {OpLoad(t_uint, GetGlobalMemoryPointer(*gmem)), Type::Uint}; - } - - if (const auto lmem = std::get_if<LmemNode>(&*node)) { - Id address = AsUint(Visit(lmem->GetAddress())); - address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U)); - const Id pointer = OpAccessChain(t_prv_float, local_memory, address); - return {OpLoad(t_float, pointer), Type::Float}; - } - - if (const auto smem = std::get_if<SmemNode>(&*node)) { - return {OpLoad(t_uint, GetSharedMemoryPointer(*smem)), Type::Uint}; - } - - if (const auto internal_flag = std::get_if<InternalFlagNode>(&*node)) { - const Id flag = internal_flags.at(static_cast<std::size_t>(internal_flag->GetFlag())); - return {OpLoad(t_bool, flag), Type::Bool}; - } - - if (const auto conditional = std::get_if<ConditionalNode>(&*node)) { - if (const auto amend_index = conditional->GetAmendIndex()) { - [[maybe_unused]] const Type type = Visit(ir.GetAmendNode(*amend_index)).type; - ASSERT(type == Type::Void); - } - // It's invalid to call conditional on nested nodes, use an operation instead - const Id true_label = OpLabel(); - const Id skip_label = OpLabel(); - const Id condition = AsBool(Visit(conditional->GetCondition())); - OpSelectionMerge(skip_label, spv::SelectionControlMask::MaskNone); - OpBranchConditional(condition, true_label, skip_label); - AddLabel(true_label); - - conditional_branch_set = true; - inside_branch = false; - VisitBasicBlock(conditional->GetCode()); - conditional_branch_set = false; - if (!inside_branch) { - OpBranch(skip_label); - } else { - inside_branch = false; - } - AddLabel(skip_label); - return {}; - } - - if (const auto comment = std::get_if<CommentNode>(&*node)) { - if (device.HasDebuggingToolAttached()) { - // We should insert comments with OpString instead of using named variables - Name(OpUndef(t_int), comment->GetText()); - } - return {}; - } - - UNREACHABLE(); - return {}; - } - - template <Id (Module::*func)(Id, Id), Type result_type, Type type_a = result_type> - Expression Unary(Operation operation) { - const Id type_def = GetTypeDefinition(result_type); - const Id op_a = As(Visit(operation[0]), type_a); - - const Id value = (this->*func)(type_def, op_a); - if (IsPrecise(operation)) { - Decorate(value, spv::Decoration::NoContraction); - } - return {value, result_type}; - } - - template <Id (Module::*func)(Id, Id, Id), Type result_type, Type type_a = result_type, - Type type_b = type_a> - Expression Binary(Operation operation) { - const Id type_def = GetTypeDefinition(result_type); - const Id op_a = As(Visit(operation[0]), type_a); - const Id op_b = As(Visit(operation[1]), type_b); - - const Id value = (this->*func)(type_def, op_a, op_b); - if (IsPrecise(operation)) { - Decorate(value, spv::Decoration::NoContraction); - } - return {value, result_type}; - } - - template <Id (Module::*func)(Id, Id, Id, Id), Type result_type, Type type_a = result_type, - Type type_b = type_a, Type type_c = type_b> - Expression Ternary(Operation operation) { - const Id type_def = GetTypeDefinition(result_type); - const Id op_a = As(Visit(operation[0]), type_a); - const Id op_b = As(Visit(operation[1]), type_b); - const Id op_c = As(Visit(operation[2]), type_c); - - const Id value = (this->*func)(type_def, op_a, op_b, op_c); - if (IsPrecise(operation)) { - Decorate(value, spv::Decoration::NoContraction); - } - return {value, result_type}; - } - - template <Id (Module::*func)(Id, Id, Id, Id, Id), Type result_type, Type type_a = result_type, - Type type_b = type_a, Type type_c = type_b, Type type_d = type_c> - Expression Quaternary(Operation operation) { - const Id type_def = GetTypeDefinition(result_type); - const Id op_a = As(Visit(operation[0]), type_a); - const Id op_b = As(Visit(operation[1]), type_b); - const Id op_c = As(Visit(operation[2]), type_c); - const Id op_d = As(Visit(operation[3]), type_d); - - const Id value = (this->*func)(type_def, op_a, op_b, op_c, op_d); - if (IsPrecise(operation)) { - Decorate(value, spv::Decoration::NoContraction); - } - return {value, result_type}; - } - - Expression Assign(Operation operation) { - const Node& dest = operation[0]; - const Node& src = operation[1]; - - Expression target{}; - if (const auto gpr = std::get_if<GprNode>(&*dest)) { - if (gpr->GetIndex() == Register::ZeroIndex) { - // Writing to Register::ZeroIndex is a no op but we still have to visit its source - // because it might have side effects. - Visit(src); - return {}; - } - target = {registers.at(gpr->GetIndex()), Type::Float}; - - } else if (const auto abuf = std::get_if<AbufNode>(&*dest)) { - const auto& buffer = abuf->GetBuffer(); - const auto ArrayPass = [&](Id pointer_type, Id composite, std::vector<u32> indices) { - std::vector<Id> members; - members.reserve(std::size(indices) + 1); - - if (buffer && IsOutputAttributeArray()) { - members.push_back(AsUint(Visit(buffer))); - } - for (const u32 index : indices) { - members.push_back(Constant(t_uint, index)); - } - return OpAccessChain(pointer_type, composite, members); - }; - - target = [&]() -> Expression { - const u32 element = abuf->GetElement(); - switch (const auto attribute = abuf->GetIndex(); attribute) { - case Attribute::Index::Position: { - const u32 index = out_indices.position.value(); - return {ArrayPass(t_out_float, out_vertex, {index, element}), Type::Float}; - } - case Attribute::Index::LayerViewportPointSize: - switch (element) { - case 1: { - if (!out_indices.layer) { - return {}; - } - const u32 index = out_indices.layer.value(); - return {AccessElement(t_out_int, out_vertex, index), Type::Int}; - } - case 2: { - if (!out_indices.viewport) { - return {}; - } - const u32 index = out_indices.viewport.value(); - return {AccessElement(t_out_int, out_vertex, index), Type::Int}; - } - case 3: { - const auto index = out_indices.point_size.value(); - return {AccessElement(t_out_float, out_vertex, index), Type::Float}; - } - default: - UNIMPLEMENTED_MSG("LayerViewportPoint element={}", abuf->GetElement()); - return {}; - } - case Attribute::Index::ClipDistances0123: { - const u32 index = out_indices.clip_distances.value(); - return {AccessElement(t_out_float, out_vertex, index, element), Type::Float}; - } - case Attribute::Index::ClipDistances4567: { - const u32 index = out_indices.clip_distances.value(); - return {AccessElement(t_out_float, out_vertex, index, element + 4), - Type::Float}; - } - default: - if (IsGenericAttribute(attribute)) { - const u8 offset = static_cast<u8>(static_cast<u8>(attribute) * 4 + element); - const GenericVaryingDescription description = output_attributes.at(offset); - const Id composite = description.id; - std::vector<u32> indices; - if (!description.is_scalar) { - indices.push_back(element - description.first_element); - } - return {ArrayPass(t_out_float, composite, indices), Type::Float}; - } - UNIMPLEMENTED_MSG("Unhandled output attribute: {}", - static_cast<u32>(attribute)); - return {}; - } - }(); - - } else if (const auto patch = std::get_if<PatchNode>(&*dest)) { - target = [&]() -> Expression { - const u32 offset = patch->GetOffset(); - switch (offset) { - case 0: - case 1: - case 2: - case 3: - return {AccessElement(t_out_float, tess_level_outer, offset % 4), Type::Float}; - case 4: - case 5: - return {AccessElement(t_out_float, tess_level_inner, offset % 4), Type::Float}; - } - UNIMPLEMENTED_MSG("Unhandled patch output offset: {}", offset); - return {}; - }(); - - } else if (const auto lmem = std::get_if<LmemNode>(&*dest)) { - Id address = AsUint(Visit(lmem->GetAddress())); - address = OpUDiv(t_uint, address, Constant(t_uint, 4)); - target = {OpAccessChain(t_prv_float, local_memory, address), Type::Float}; - - } else if (const auto smem = std::get_if<SmemNode>(&*dest)) { - target = {GetSharedMemoryPointer(*smem), Type::Uint}; - - } else if (const auto gmem = std::get_if<GmemNode>(&*dest)) { - target = {GetGlobalMemoryPointer(*gmem), Type::Uint}; - - } else if (const auto cv = std::get_if<CustomVarNode>(&*dest)) { - target = {custom_variables.at(cv->GetIndex()), Type::Float}; - - } else { - UNIMPLEMENTED(); - } - - if (!target.id) { - // On failure we return a nullptr target.id, skip these stores. - return {}; - } - - OpStore(target.id, As(Visit(src), target.type)); - return {}; - } - - template <u32 offset> - Expression FCastHalf(Operation operation) { - const Id value = AsHalfFloat(Visit(operation[0])); - return {GetFloatFromHalfScalar(OpCompositeExtract(t_scalar_half, value, offset)), - Type::Float}; - } - - Expression FSwizzleAdd(Operation operation) { - const Id minus = Constant(t_float, -1.0f); - const Id plus = v_float_one; - const Id zero = v_float_zero; - const Id lut_a = ConstantComposite(t_float4, minus, plus, minus, zero); - const Id lut_b = ConstantComposite(t_float4, minus, minus, plus, minus); - - Id mask = OpLoad(t_uint, thread_id); - mask = OpBitwiseAnd(t_uint, mask, Constant(t_uint, 3)); - mask = OpShiftLeftLogical(t_uint, mask, Constant(t_uint, 1)); - mask = OpShiftRightLogical(t_uint, AsUint(Visit(operation[2])), mask); - mask = OpBitwiseAnd(t_uint, mask, Constant(t_uint, 3)); - - const Id modifier_a = OpVectorExtractDynamic(t_float, lut_a, mask); - const Id modifier_b = OpVectorExtractDynamic(t_float, lut_b, mask); - - const Id op_a = OpFMul(t_float, AsFloat(Visit(operation[0])), modifier_a); - const Id op_b = OpFMul(t_float, AsFloat(Visit(operation[1])), modifier_b); - return {OpFAdd(t_float, op_a, op_b), Type::Float}; - } - - Expression HNegate(Operation operation) { - const bool is_f16 = device.IsFloat16Supported(); - const Id minus_one = Constant(t_scalar_half, is_f16 ? 0xbc00 : 0xbf800000); - const Id one = Constant(t_scalar_half, is_f16 ? 0x3c00 : 0x3f800000); - const auto GetNegate = [&](std::size_t index) { - return OpSelect(t_scalar_half, AsBool(Visit(operation[index])), minus_one, one); - }; - const Id negation = OpCompositeConstruct(t_half, GetNegate(1), GetNegate(2)); - return {OpFMul(t_half, AsHalfFloat(Visit(operation[0])), negation), Type::HalfFloat}; - } - - Expression HClamp(Operation operation) { - const auto Pack = [&](std::size_t index) { - const Id scalar = GetHalfScalarFromFloat(AsFloat(Visit(operation[index]))); - return OpCompositeConstruct(t_half, scalar, scalar); - }; - const Id value = AsHalfFloat(Visit(operation[0])); - const Id min = Pack(1); - const Id max = Pack(2); - - const Id clamped = OpFClamp(t_half, value, min, max); - if (IsPrecise(operation)) { - Decorate(clamped, spv::Decoration::NoContraction); - } - return {clamped, Type::HalfFloat}; - } - - Expression HCastFloat(Operation operation) { - const Id value = GetHalfScalarFromFloat(AsFloat(Visit(operation[0]))); - return {OpCompositeConstruct(t_half, value, Constant(t_scalar_half, 0)), Type::HalfFloat}; - } - - Expression HUnpack(Operation operation) { - Expression operand = Visit(operation[0]); - const auto type = std::get<Tegra::Shader::HalfType>(operation.GetMeta()); - if (type == Tegra::Shader::HalfType::H0_H1) { - return operand; - } - const auto value = [&] { - switch (std::get<Tegra::Shader::HalfType>(operation.GetMeta())) { - case Tegra::Shader::HalfType::F32: - return GetHalfScalarFromFloat(AsFloat(operand)); - case Tegra::Shader::HalfType::H0_H0: - return OpCompositeExtract(t_scalar_half, AsHalfFloat(operand), 0); - case Tegra::Shader::HalfType::H1_H1: - return OpCompositeExtract(t_scalar_half, AsHalfFloat(operand), 1); - default: - UNREACHABLE(); - return ConstantNull(t_half); - } - }(); - return {OpCompositeConstruct(t_half, value, value), Type::HalfFloat}; - } - - Expression HMergeF32(Operation operation) { - const Id value = AsHalfFloat(Visit(operation[0])); - return {GetFloatFromHalfScalar(OpCompositeExtract(t_scalar_half, value, 0)), Type::Float}; - } - - template <u32 offset> - Expression HMergeHN(Operation operation) { - const Id target = AsHalfFloat(Visit(operation[0])); - const Id source = AsHalfFloat(Visit(operation[1])); - const Id object = OpCompositeExtract(t_scalar_half, source, offset); - return {OpCompositeInsert(t_half, object, target, offset), Type::HalfFloat}; - } - - Expression HPack2(Operation operation) { - const Id low = GetHalfScalarFromFloat(AsFloat(Visit(operation[0]))); - const Id high = GetHalfScalarFromFloat(AsFloat(Visit(operation[1]))); - return {OpCompositeConstruct(t_half, low, high), Type::HalfFloat}; - } - - Expression LogicalAddCarry(Operation operation) { - const Id op_a = AsUint(Visit(operation[0])); - const Id op_b = AsUint(Visit(operation[1])); - - const Id result = OpIAddCarry(TypeStruct({t_uint, t_uint}), op_a, op_b); - const Id carry = OpCompositeExtract(t_uint, result, 1); - return {OpINotEqual(t_bool, carry, v_uint_zero), Type::Bool}; - } - - Expression LogicalAssign(Operation operation) { - const Node& dest = operation[0]; - const Node& src = operation[1]; - - Id target{}; - if (const auto pred = std::get_if<PredicateNode>(&*dest)) { - ASSERT_MSG(!pred->IsNegated(), "Negating logical assignment"); - - const auto index = pred->GetIndex(); - switch (index) { - case Tegra::Shader::Pred::NeverExecute: - case Tegra::Shader::Pred::UnusedIndex: - // Writing to these predicates is a no-op - return {}; - } - target = predicates.at(index); - - } else if (const auto flag = std::get_if<InternalFlagNode>(&*dest)) { - target = internal_flags.at(static_cast<u32>(flag->GetFlag())); - } - - OpStore(target, AsBool(Visit(src))); - return {}; - } - - Expression LogicalFOrdered(Operation operation) { - // Emulate SPIR-V's OpOrdered - const Id op_a = AsFloat(Visit(operation[0])); - const Id op_b = AsFloat(Visit(operation[1])); - const Id is_num_a = OpFOrdEqual(t_bool, op_a, op_a); - const Id is_num_b = OpFOrdEqual(t_bool, op_b, op_b); - return {OpLogicalAnd(t_bool, is_num_a, is_num_b), Type::Bool}; - } - - Expression LogicalFUnordered(Operation operation) { - // Emulate SPIR-V's OpUnordered - const Id op_a = AsFloat(Visit(operation[0])); - const Id op_b = AsFloat(Visit(operation[1])); - const Id is_nan_a = OpIsNan(t_bool, op_a); - const Id is_nan_b = OpIsNan(t_bool, op_b); - return {OpLogicalOr(t_bool, is_nan_a, is_nan_b), Type::Bool}; - } - - Id GetTextureSampler(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - ASSERT(!meta.sampler.is_buffer); - - const auto& entry = sampled_images.at(meta.sampler.index); - Id sampler = entry.variable; - if (meta.sampler.is_indexed) { - const Id index = AsInt(Visit(meta.index)); - sampler = OpAccessChain(entry.sampler_pointer_type, sampler, index); - } - return OpLoad(entry.sampler_type, sampler); - } - - Id GetTextureImage(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - const u32 index = meta.sampler.index; - if (meta.sampler.is_buffer) { - const auto& entry = uniform_texels.at(index); - return OpLoad(entry.image_type, entry.image); - } else { - const auto& entry = sampled_images.at(index); - return OpImage(entry.image_type, GetTextureSampler(operation)); - } - } - - Id GetImage(Operation operation) { - const auto& meta = std::get<MetaImage>(operation.GetMeta()); - const auto entry = images.at(meta.image.index); - return OpLoad(entry.image_type, entry.image); - } - - Id AssembleVector(const std::vector<Id>& coords, Type type) { - const Id coords_type = GetTypeVectorDefinitionLut(type).at(coords.size() - 1); - return coords.size() == 1 ? coords[0] : OpCompositeConstruct(coords_type, coords); - } - - Id GetCoordinates(Operation operation, Type type) { - std::vector<Id> coords; - for (std::size_t i = 0; i < operation.GetOperandsCount(); ++i) { - coords.push_back(As(Visit(operation[i]), type)); - } - if (const auto meta = std::get_if<MetaTexture>(&operation.GetMeta())) { - // Add array coordinate for textures - if (meta->sampler.is_array) { - Id array = AsInt(Visit(meta->array)); - if (type == Type::Float) { - array = OpConvertSToF(t_float, array); - } - coords.push_back(array); - } - } - return AssembleVector(coords, type); - } - - Id GetOffsetCoordinates(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - std::vector<Id> coords; - coords.reserve(meta.aoffi.size()); - for (const auto& coord : meta.aoffi) { - coords.push_back(AsInt(Visit(coord))); - } - return AssembleVector(coords, Type::Int); - } - - std::pair<Id, Id> GetDerivatives(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - const auto& derivatives = meta.derivates; - ASSERT(derivatives.size() % 2 == 0); - - const std::size_t components = derivatives.size() / 2; - std::vector<Id> dx, dy; - dx.reserve(components); - dy.reserve(components); - for (std::size_t index = 0; index < components; ++index) { - dx.push_back(AsFloat(Visit(derivatives.at(index * 2 + 0)))); - dy.push_back(AsFloat(Visit(derivatives.at(index * 2 + 1)))); - } - return {AssembleVector(dx, Type::Float), AssembleVector(dy, Type::Float)}; - } - - Expression GetTextureElement(Operation operation, Id sample_value, Type type) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - const auto type_def = GetTypeDefinition(type); - return {OpCompositeExtract(type_def, sample_value, meta.element), type}; - } - - Expression Texture(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - - const bool can_implicit = stage == ShaderType::Fragment; - const Id sampler = GetTextureSampler(operation); - const Id coords = GetCoordinates(operation, Type::Float); - - std::vector<Id> operands; - spv::ImageOperandsMask mask{}; - if (meta.bias) { - mask = mask | spv::ImageOperandsMask::Bias; - operands.push_back(AsFloat(Visit(meta.bias))); - } - - if (!can_implicit) { - mask = mask | spv::ImageOperandsMask::Lod; - operands.push_back(v_float_zero); - } - - if (!meta.aoffi.empty()) { - mask = mask | spv::ImageOperandsMask::Offset; - operands.push_back(GetOffsetCoordinates(operation)); - } - - if (meta.depth_compare) { - // Depth sampling - UNIMPLEMENTED_IF(meta.bias); - const Id dref = AsFloat(Visit(meta.depth_compare)); - if (can_implicit) { - return { - OpImageSampleDrefImplicitLod(t_float, sampler, coords, dref, mask, operands), - Type::Float}; - } else { - return { - OpImageSampleDrefExplicitLod(t_float, sampler, coords, dref, mask, operands), - Type::Float}; - } - } - - Id texture; - if (can_implicit) { - texture = OpImageSampleImplicitLod(t_float4, sampler, coords, mask, operands); - } else { - texture = OpImageSampleExplicitLod(t_float4, sampler, coords, mask, operands); - } - return GetTextureElement(operation, texture, Type::Float); - } - - Expression TextureLod(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - - const Id sampler = GetTextureSampler(operation); - const Id coords = GetCoordinates(operation, Type::Float); - const Id lod = AsFloat(Visit(meta.lod)); - - spv::ImageOperandsMask mask = spv::ImageOperandsMask::Lod; - std::vector<Id> operands{lod}; - - if (!meta.aoffi.empty()) { - mask = mask | spv::ImageOperandsMask::Offset; - operands.push_back(GetOffsetCoordinates(operation)); - } - - if (meta.sampler.is_shadow) { - const Id dref = AsFloat(Visit(meta.depth_compare)); - return {OpImageSampleDrefExplicitLod(t_float, sampler, coords, dref, mask, operands), - Type::Float}; - } - const Id texture = OpImageSampleExplicitLod(t_float4, sampler, coords, mask, operands); - return GetTextureElement(operation, texture, Type::Float); - } - - Expression TextureGather(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - - const Id coords = GetCoordinates(operation, Type::Float); - - spv::ImageOperandsMask mask = spv::ImageOperandsMask::MaskNone; - std::vector<Id> operands; - Id texture{}; - - if (!meta.aoffi.empty()) { - mask = mask | spv::ImageOperandsMask::Offset; - operands.push_back(GetOffsetCoordinates(operation)); - } - - if (meta.sampler.is_shadow) { - texture = OpImageDrefGather(t_float4, GetTextureSampler(operation), coords, - AsFloat(Visit(meta.depth_compare)), mask, operands); - } else { - u32 component_value = 0; - if (meta.component) { - const auto component = std::get_if<ImmediateNode>(&*meta.component); - ASSERT_MSG(component, "Component is not an immediate value"); - component_value = component->GetValue(); - } - texture = OpImageGather(t_float4, GetTextureSampler(operation), coords, - Constant(t_uint, component_value), mask, operands); - } - return GetTextureElement(operation, texture, Type::Float); - } - - Expression TextureQueryDimensions(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - UNIMPLEMENTED_IF(!meta.aoffi.empty()); - UNIMPLEMENTED_IF(meta.depth_compare); - - const auto image_id = GetTextureImage(operation); - if (meta.element == 3) { - return {OpImageQueryLevels(t_int, image_id), Type::Int}; - } - - const Id lod = AsUint(Visit(operation[0])); - const std::size_t coords_count = [&meta] { - switch (const auto type = meta.sampler.type) { - case Tegra::Shader::TextureType::Texture1D: - return 1; - case Tegra::Shader::TextureType::Texture2D: - case Tegra::Shader::TextureType::TextureCube: - return 2; - case Tegra::Shader::TextureType::Texture3D: - return 3; - default: - UNREACHABLE_MSG("Invalid texture type={}", type); - return 2; - } - }(); - - if (meta.element >= coords_count) { - return {v_float_zero, Type::Float}; - } - - const std::array<Id, 3> types = {t_int, t_int2, t_int3}; - const Id sizes = OpImageQuerySizeLod(types.at(coords_count - 1), image_id, lod); - const Id size = OpCompositeExtract(t_int, sizes, meta.element); - return {size, Type::Int}; - } - - Expression TextureQueryLod(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - UNIMPLEMENTED_IF(!meta.aoffi.empty()); - UNIMPLEMENTED_IF(meta.depth_compare); - - if (meta.element >= 2) { - UNREACHABLE_MSG("Invalid element"); - return {v_float_zero, Type::Float}; - } - const auto sampler_id = GetTextureSampler(operation); - - const Id multiplier = Constant(t_float, 256.0f); - const Id multipliers = ConstantComposite(t_float2, multiplier, multiplier); - - const Id coords = GetCoordinates(operation, Type::Float); - Id size = OpImageQueryLod(t_float2, sampler_id, coords); - size = OpFMul(t_float2, size, multipliers); - size = OpConvertFToS(t_int2, size); - return GetTextureElement(operation, size, Type::Int); - } - - Expression TexelFetch(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - UNIMPLEMENTED_IF(meta.depth_compare); - - const Id image = GetTextureImage(operation); - const Id coords = GetCoordinates(operation, Type::Int); - - spv::ImageOperandsMask mask = spv::ImageOperandsMask::MaskNone; - std::vector<Id> operands; - Id fetch; - - if (meta.lod && !meta.sampler.is_buffer) { - mask = mask | spv::ImageOperandsMask::Lod; - operands.push_back(AsInt(Visit(meta.lod))); - } - - if (!meta.aoffi.empty()) { - mask = mask | spv::ImageOperandsMask::Offset; - operands.push_back(GetOffsetCoordinates(operation)); - } - - fetch = OpImageFetch(t_float4, image, coords, mask, operands); - return GetTextureElement(operation, fetch, Type::Float); - } - - Expression TextureGradient(Operation operation) { - const auto& meta = std::get<MetaTexture>(operation.GetMeta()); - UNIMPLEMENTED_IF(!meta.aoffi.empty()); - - const Id sampler = GetTextureSampler(operation); - const Id coords = GetCoordinates(operation, Type::Float); - const auto [dx, dy] = GetDerivatives(operation); - const std::vector grad = {dx, dy}; - - static constexpr auto mask = spv::ImageOperandsMask::Grad; - const Id texture = OpImageSampleExplicitLod(t_float4, sampler, coords, mask, grad); - return GetTextureElement(operation, texture, Type::Float); - } - - Expression ImageLoad(Operation operation) { - if (!device.IsFormatlessImageLoadSupported()) { - return {v_float_zero, Type::Float}; - } - - const auto& meta{std::get<MetaImage>(operation.GetMeta())}; - - const Id coords = GetCoordinates(operation, Type::Int); - const Id texel = OpImageRead(t_uint4, GetImage(operation), coords); - - return {OpCompositeExtract(t_uint, texel, meta.element), Type::Uint}; - } - - Expression ImageStore(Operation operation) { - const auto meta{std::get<MetaImage>(operation.GetMeta())}; - std::vector<Id> colors; - for (const auto& value : meta.values) { - colors.push_back(AsUint(Visit(value))); - } - - const Id coords = GetCoordinates(operation, Type::Int); - const Id texel = OpCompositeConstruct(t_uint4, colors); - - OpImageWrite(GetImage(operation), coords, texel, {}); - return {}; - } - - template <Id (Module::*func)(Id, Id, Id, Id, Id)> - Expression AtomicImage(Operation operation) { - const auto& meta{std::get<MetaImage>(operation.GetMeta())}; - ASSERT(meta.values.size() == 1); - - const Id coordinate = GetCoordinates(operation, Type::Int); - const Id image = images.at(meta.image.index).image; - const Id sample = v_uint_zero; - const Id pointer = OpImageTexelPointer(t_image_uint, image, coordinate, sample); - - const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device)); - const Id semantics = v_uint_zero; - const Id value = AsUint(Visit(meta.values[0])); - return {(this->*func)(t_uint, pointer, scope, semantics, value), Type::Uint}; - } - - template <Id (Module::*func)(Id, Id, Id, Id, Id)> - Expression Atomic(Operation operation) { - Id pointer; - if (const auto smem = std::get_if<SmemNode>(&*operation[0])) { - pointer = GetSharedMemoryPointer(*smem); - } else if (const auto gmem = std::get_if<GmemNode>(&*operation[0])) { - pointer = GetGlobalMemoryPointer(*gmem); - } else { - UNREACHABLE(); - return {v_float_zero, Type::Float}; - } - const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device)); - const Id semantics = v_uint_zero; - const Id value = AsUint(Visit(operation[1])); - - return {(this->*func)(t_uint, pointer, scope, semantics, value), Type::Uint}; - } - - template <Id (Module::*func)(Id, Id, Id, Id, Id)> - Expression Reduce(Operation operation) { - Atomic<func>(operation); - return {}; - } - - Expression Branch(Operation operation) { - const auto& target = std::get<ImmediateNode>(*operation[0]); - OpStore(jmp_to, Constant(t_uint, target.GetValue())); - OpBranch(continue_label); - inside_branch = true; - if (!conditional_branch_set) { - AddLabel(); - } - return {}; - } - - Expression BranchIndirect(Operation operation) { - const Id op_a = AsUint(Visit(operation[0])); - - OpStore(jmp_to, op_a); - OpBranch(continue_label); - inside_branch = true; - if (!conditional_branch_set) { - AddLabel(); - } - return {}; - } - - Expression PushFlowStack(Operation operation) { - const auto& target = std::get<ImmediateNode>(*operation[0]); - const auto [flow_stack, flow_stack_top] = GetFlowStack(operation); - const Id current = OpLoad(t_uint, flow_stack_top); - const Id next = OpIAdd(t_uint, current, Constant(t_uint, 1)); - const Id access = OpAccessChain(t_func_uint, flow_stack, current); - - OpStore(access, Constant(t_uint, target.GetValue())); - OpStore(flow_stack_top, next); - return {}; - } - - Expression PopFlowStack(Operation operation) { - const auto [flow_stack, flow_stack_top] = GetFlowStack(operation); - const Id current = OpLoad(t_uint, flow_stack_top); - const Id previous = OpISub(t_uint, current, Constant(t_uint, 1)); - const Id access = OpAccessChain(t_func_uint, flow_stack, previous); - const Id target = OpLoad(t_uint, access); - - OpStore(flow_stack_top, previous); - OpStore(jmp_to, target); - OpBranch(continue_label); - inside_branch = true; - if (!conditional_branch_set) { - AddLabel(); - } - return {}; - } - - Id MaxwellToSpirvComparison(Maxwell::ComparisonOp compare_op, Id operand_1, Id operand_2) { - using Compare = Maxwell::ComparisonOp; - switch (compare_op) { - case Compare::NeverOld: - return v_false; // Never let the test pass - case Compare::LessOld: - return OpFOrdLessThan(t_bool, operand_1, operand_2); - case Compare::EqualOld: - return OpFOrdEqual(t_bool, operand_1, operand_2); - case Compare::LessEqualOld: - return OpFOrdLessThanEqual(t_bool, operand_1, operand_2); - case Compare::GreaterOld: - return OpFOrdGreaterThan(t_bool, operand_1, operand_2); - case Compare::NotEqualOld: - return OpFOrdNotEqual(t_bool, operand_1, operand_2); - case Compare::GreaterEqualOld: - return OpFOrdGreaterThanEqual(t_bool, operand_1, operand_2); - default: - UNREACHABLE(); - return v_true; - } - } - - void AlphaTest(Id pointer) { - if (specialization.alpha_test_func == Maxwell::ComparisonOp::AlwaysOld) { - return; - } - const Id true_label = OpLabel(); - const Id discard_label = OpLabel(); - const Id alpha_reference = Constant(t_float, specialization.alpha_test_ref); - const Id alpha_value = OpLoad(t_float, pointer); - const Id condition = - MaxwellToSpirvComparison(specialization.alpha_test_func, alpha_value, alpha_reference); - - OpBranchConditional(condition, true_label, discard_label); - AddLabel(discard_label); - OpKill(); - AddLabel(true_label); - } - - void PreExit() { - if (stage == ShaderType::Vertex && specialization.ndc_minus_one_to_one) { - const u32 position_index = out_indices.position.value(); - const Id z_pointer = AccessElement(t_out_float, out_vertex, position_index, 2U); - const Id w_pointer = AccessElement(t_out_float, out_vertex, position_index, 3U); - Id depth = OpLoad(t_float, z_pointer); - depth = OpFAdd(t_float, depth, OpLoad(t_float, w_pointer)); - depth = OpFMul(t_float, depth, Constant(t_float, 0.5f)); - OpStore(z_pointer, depth); - } - if (stage == ShaderType::Fragment) { - const auto SafeGetRegister = [this](u32 reg) { - if (const auto it = registers.find(reg); it != registers.end()) { - return OpLoad(t_float, it->second); - } - return v_float_zero; - }; - - UNIMPLEMENTED_IF_MSG(header.ps.omap.sample_mask != 0, - "Sample mask write is unimplemented"); - - // Write the color outputs using the data in the shader registers, disabled - // rendertargets/components are skipped in the register assignment. - u32 current_reg = 0; - for (u32 rt = 0; rt < Maxwell::NumRenderTargets; ++rt) { - // TODO(Subv): Figure out how dual-source blending is configured in the Switch. - for (u32 component = 0; component < 4; ++component) { - if (!header.ps.IsColorComponentOutputEnabled(rt, component)) { - continue; - } - const Id pointer = AccessElement(t_out_float, frag_colors[rt], component); - OpStore(pointer, SafeGetRegister(current_reg)); - if (rt == 0 && component == 3) { - AlphaTest(pointer); - } - ++current_reg; - } - } - if (header.ps.omap.depth) { - // The depth output is always 2 registers after the last color output, and - // current_reg already contains one past the last color register. - OpStore(frag_depth, SafeGetRegister(current_reg + 1)); - } - } - } - - Expression Exit(Operation operation) { - PreExit(); - inside_branch = true; - if (conditional_branch_set) { - OpReturn(); - } else { - const Id dummy = OpLabel(); - OpBranch(dummy); - AddLabel(dummy); - OpReturn(); - AddLabel(); - } - return {}; - } - - Expression Discard(Operation operation) { - inside_branch = true; - if (conditional_branch_set) { - OpKill(); - } else { - const Id dummy = OpLabel(); - OpBranch(dummy); - AddLabel(dummy); - OpKill(); - AddLabel(); - } - return {}; - } - - Expression EmitVertex(Operation) { - OpEmitVertex(); - return {}; - } - - Expression EndPrimitive(Operation operation) { - OpEndPrimitive(); - return {}; - } - - Expression InvocationId(Operation) { - return {OpLoad(t_int, invocation_id), Type::Int}; - } - - Expression YNegate(Operation) { - LOG_WARNING(Render_Vulkan, "(STUBBED)"); - return {Constant(t_float, 1.0f), Type::Float}; - } - - template <u32 element> - Expression LocalInvocationId(Operation) { - const Id id = OpLoad(t_uint3, local_invocation_id); - return {OpCompositeExtract(t_uint, id, element), Type::Uint}; - } - - template <u32 element> - Expression WorkGroupId(Operation operation) { - const Id id = OpLoad(t_uint3, workgroup_id); - return {OpCompositeExtract(t_uint, id, element), Type::Uint}; - } - - Expression BallotThread(Operation operation) { - const Id predicate = AsBool(Visit(operation[0])); - const Id ballot = OpSubgroupBallotKHR(t_uint4, predicate); - - if (!device.IsWarpSizePotentiallyBiggerThanGuest()) { - // Guest-like devices can just return the first index. - return {OpCompositeExtract(t_uint, ballot, 0U), Type::Uint}; - } - - // The others will have to return what is local to the current thread. - // For instance a device with a warp size of 64 will return the upper uint when the current - // thread is 38. - const Id tid = OpLoad(t_uint, thread_id); - const Id thread_index = OpShiftRightLogical(t_uint, tid, Constant(t_uint, 5)); - return {OpVectorExtractDynamic(t_uint, ballot, thread_index), Type::Uint}; - } - - template <Id (Module::*func)(Id, Id)> - Expression Vote(Operation operation) { - // TODO(Rodrigo): Handle devices with different warp sizes - const Id predicate = AsBool(Visit(operation[0])); - return {(this->*func)(t_bool, predicate), Type::Bool}; - } - - Expression ThreadId(Operation) { - return {OpLoad(t_uint, thread_id), Type::Uint}; - } - - template <std::size_t index> - Expression ThreadMask(Operation) { - // TODO(Rodrigo): Handle devices with different warp sizes - const Id mask = thread_masks[index]; - return {OpLoad(t_uint, AccessElement(t_in_uint, mask, 0)), Type::Uint}; - } - - Expression ShuffleIndexed(Operation operation) { - const Id value = AsFloat(Visit(operation[0])); - const Id index = AsUint(Visit(operation[1])); - return {OpSubgroupReadInvocationKHR(t_float, value, index), Type::Float}; - } - - Expression Barrier(Operation) { - if (!ir.IsDecompiled()) { - LOG_ERROR(Render_Vulkan, "OpBarrier used by shader is not decompiled"); - return {}; - } - - const auto scope = spv::Scope::Workgroup; - const auto memory = spv::Scope::Workgroup; - const auto semantics = - spv::MemorySemanticsMask::WorkgroupMemory | spv::MemorySemanticsMask::AcquireRelease; - OpControlBarrier(Constant(t_uint, static_cast<u32>(scope)), - Constant(t_uint, static_cast<u32>(memory)), - Constant(t_uint, static_cast<u32>(semantics))); - return {}; - } - - template <spv::Scope scope> - Expression MemoryBarrier(Operation) { - const auto semantics = - spv::MemorySemanticsMask::AcquireRelease | spv::MemorySemanticsMask::UniformMemory | - spv::MemorySemanticsMask::WorkgroupMemory | - spv::MemorySemanticsMask::AtomicCounterMemory | spv::MemorySemanticsMask::ImageMemory; - - OpMemoryBarrier(Constant(t_uint, static_cast<u32>(scope)), - Constant(t_uint, static_cast<u32>(semantics))); - return {}; - } - - Id DeclareBuiltIn(spv::BuiltIn builtin, spv::StorageClass storage, Id type, std::string name) { - const Id id = OpVariable(type, storage); - Decorate(id, spv::Decoration::BuiltIn, static_cast<u32>(builtin)); - AddGlobalVariable(Name(id, std::move(name))); - interfaces.push_back(id); - return id; - } - - Id DeclareInputBuiltIn(spv::BuiltIn builtin, Id type, std::string name) { - return DeclareBuiltIn(builtin, spv::StorageClass::Input, type, std::move(name)); - } - - template <typename... Args> - Id AccessElement(Id pointer_type, Id composite, Args... elements_) { - std::vector<Id> members; - auto elements = {elements_...}; - for (const auto element : elements) { - members.push_back(Constant(t_uint, element)); - } - - return OpAccessChain(pointer_type, composite, members); - } - - Id As(Expression expr, Type wanted_type) { - switch (wanted_type) { - case Type::Bool: - return AsBool(expr); - case Type::Bool2: - return AsBool2(expr); - case Type::Float: - return AsFloat(expr); - case Type::Int: - return AsInt(expr); - case Type::Uint: - return AsUint(expr); - case Type::HalfFloat: - return AsHalfFloat(expr); - default: - UNREACHABLE(); - return expr.id; - } - } - - Id AsBool(Expression expr) { - ASSERT(expr.type == Type::Bool); - return expr.id; - } - - Id AsBool2(Expression expr) { - ASSERT(expr.type == Type::Bool2); - return expr.id; - } - - Id AsFloat(Expression expr) { - switch (expr.type) { - case Type::Float: - return expr.id; - case Type::Int: - case Type::Uint: - return OpBitcast(t_float, expr.id); - case Type::HalfFloat: - if (device.IsFloat16Supported()) { - return OpBitcast(t_float, expr.id); - } - return OpBitcast(t_float, OpPackHalf2x16(t_uint, expr.id)); - default: - UNREACHABLE(); - return expr.id; - } - } - - Id AsInt(Expression expr) { - switch (expr.type) { - case Type::Int: - return expr.id; - case Type::Float: - case Type::Uint: - return OpBitcast(t_int, expr.id); - case Type::HalfFloat: - if (device.IsFloat16Supported()) { - return OpBitcast(t_int, expr.id); - } - return OpPackHalf2x16(t_int, expr.id); - default: - UNREACHABLE(); - return expr.id; - } - } - - Id AsUint(Expression expr) { - switch (expr.type) { - case Type::Uint: - return expr.id; - case Type::Float: - case Type::Int: - return OpBitcast(t_uint, expr.id); - case Type::HalfFloat: - if (device.IsFloat16Supported()) { - return OpBitcast(t_uint, expr.id); - } - return OpPackHalf2x16(t_uint, expr.id); - default: - UNREACHABLE(); - return expr.id; - } - } - - Id AsHalfFloat(Expression expr) { - switch (expr.type) { - case Type::HalfFloat: - return expr.id; - case Type::Float: - case Type::Int: - case Type::Uint: - if (device.IsFloat16Supported()) { - return OpBitcast(t_half, expr.id); - } - return OpUnpackHalf2x16(t_half, AsUint(expr)); - default: - UNREACHABLE(); - return expr.id; - } - } - - Id GetHalfScalarFromFloat(Id value) { - if (device.IsFloat16Supported()) { - return OpFConvert(t_scalar_half, value); - } - return value; - } - - Id GetFloatFromHalfScalar(Id value) { - if (device.IsFloat16Supported()) { - return OpFConvert(t_float, value); - } - return value; - } - - AttributeType GetAttributeType(u32 location) const { - if (stage != ShaderType::Vertex) { - return {Type::Float, t_in_float, t_in_float4}; - } - switch (specialization.attribute_types.at(location)) { - case Maxwell::VertexAttribute::Type::SignedNorm: - case Maxwell::VertexAttribute::Type::UnsignedNorm: - case Maxwell::VertexAttribute::Type::UnsignedScaled: - case Maxwell::VertexAttribute::Type::SignedScaled: - case Maxwell::VertexAttribute::Type::Float: - return {Type::Float, t_in_float, t_in_float4}; - case Maxwell::VertexAttribute::Type::SignedInt: - return {Type::Int, t_in_int, t_in_int4}; - case Maxwell::VertexAttribute::Type::UnsignedInt: - return {Type::Uint, t_in_uint, t_in_uint4}; - default: - UNREACHABLE(); - return {Type::Float, t_in_float, t_in_float4}; - } - } - - Id GetTypeDefinition(Type type) const { - switch (type) { - case Type::Bool: - return t_bool; - case Type::Bool2: - return t_bool2; - case Type::Float: - return t_float; - case Type::Int: - return t_int; - case Type::Uint: - return t_uint; - case Type::HalfFloat: - return t_half; - default: - UNREACHABLE(); - return {}; - } - } - - std::array<Id, 4> GetTypeVectorDefinitionLut(Type type) const { - switch (type) { - case Type::Float: - return {t_float, t_float2, t_float3, t_float4}; - case Type::Int: - return {t_int, t_int2, t_int3, t_int4}; - case Type::Uint: - return {t_uint, t_uint2, t_uint3, t_uint4}; - default: - UNIMPLEMENTED(); - return {}; - } - } - - std::tuple<Id, Id> CreateFlowStack() { - // TODO(Rodrigo): Figure out the actual depth of the flow stack, for now it seems unlikely - // that shaders will use 20 nested SSYs and PBKs. - constexpr u32 FLOW_STACK_SIZE = 20; - constexpr auto storage_class = spv::StorageClass::Function; - - const Id flow_stack_type = TypeArray(t_uint, Constant(t_uint, FLOW_STACK_SIZE)); - const Id stack = OpVariable(TypePointer(storage_class, flow_stack_type), storage_class, - ConstantNull(flow_stack_type)); - const Id top = OpVariable(t_func_uint, storage_class, Constant(t_uint, 0)); - AddLocalVariable(stack); - AddLocalVariable(top); - return std::tie(stack, top); - } - - std::pair<Id, Id> GetFlowStack(Operation operation) { - const auto stack_class = std::get<MetaStackClass>(operation.GetMeta()); - switch (stack_class) { - case MetaStackClass::Ssy: - return {ssy_flow_stack, ssy_flow_stack_top}; - case MetaStackClass::Pbk: - return {pbk_flow_stack, pbk_flow_stack_top}; - } - UNREACHABLE(); - return {}; - } - - Id GetGlobalMemoryPointer(const GmemNode& gmem) { - const Id real = AsUint(Visit(gmem.GetRealAddress())); - const Id base = AsUint(Visit(gmem.GetBaseAddress())); - const Id diff = OpISub(t_uint, real, base); - const Id offset = OpShiftRightLogical(t_uint, diff, Constant(t_uint, 2)); - const Id buffer = global_buffers.at(gmem.GetDescriptor()); - return OpAccessChain(t_gmem_uint, buffer, Constant(t_uint, 0), offset); - } - - Id GetSharedMemoryPointer(const SmemNode& smem) { - ASSERT(stage == ShaderType::Compute); - Id address = AsUint(Visit(smem.GetAddress())); - address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U)); - return OpAccessChain(t_smem_uint, shared_memory, address); - } - - static constexpr std::array operation_decompilers = { - &SPIRVDecompiler::Assign, - - &SPIRVDecompiler::Ternary<&Module::OpSelect, Type::Float, Type::Bool, Type::Float, - Type::Float>, - - &SPIRVDecompiler::Binary<&Module::OpFAdd, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFMul, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFDiv, Type::Float>, - &SPIRVDecompiler::Ternary<&Module::OpFma, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpFNegate, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpFAbs, Type::Float>, - &SPIRVDecompiler::Ternary<&Module::OpFClamp, Type::Float>, - &SPIRVDecompiler::FCastHalf<0>, - &SPIRVDecompiler::FCastHalf<1>, - &SPIRVDecompiler::Binary<&Module::OpFMin, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFMax, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpCos, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpSin, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpExp2, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpLog2, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpInverseSqrt, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpSqrt, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpRoundEven, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpFloor, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpCeil, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpTrunc, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpConvertSToF, Type::Float, Type::Int>, - &SPIRVDecompiler::Unary<&Module::OpConvertUToF, Type::Float, Type::Uint>, - &SPIRVDecompiler::FSwizzleAdd, - - &SPIRVDecompiler::Binary<&Module::OpIAdd, Type::Int>, - &SPIRVDecompiler::Binary<&Module::OpIMul, Type::Int>, - &SPIRVDecompiler::Binary<&Module::OpSDiv, Type::Int>, - &SPIRVDecompiler::Unary<&Module::OpSNegate, Type::Int>, - &SPIRVDecompiler::Unary<&Module::OpSAbs, Type::Int>, - &SPIRVDecompiler::Binary<&Module::OpSMin, Type::Int>, - &SPIRVDecompiler::Binary<&Module::OpSMax, Type::Int>, - - &SPIRVDecompiler::Unary<&Module::OpConvertFToS, Type::Int, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpBitcast, Type::Int, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpShiftLeftLogical, Type::Int, Type::Int, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpShiftRightLogical, Type::Int, Type::Int, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpShiftRightArithmetic, Type::Int, Type::Int, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpBitwiseAnd, Type::Int>, - &SPIRVDecompiler::Binary<&Module::OpBitwiseOr, Type::Int>, - &SPIRVDecompiler::Binary<&Module::OpBitwiseXor, Type::Int>, - &SPIRVDecompiler::Unary<&Module::OpNot, Type::Int>, - &SPIRVDecompiler::Quaternary<&Module::OpBitFieldInsert, Type::Int>, - &SPIRVDecompiler::Ternary<&Module::OpBitFieldSExtract, Type::Int>, - &SPIRVDecompiler::Unary<&Module::OpBitCount, Type::Int>, - &SPIRVDecompiler::Unary<&Module::OpFindSMsb, Type::Int>, - - &SPIRVDecompiler::Binary<&Module::OpIAdd, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpIMul, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpUDiv, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpUMin, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpUMax, Type::Uint>, - &SPIRVDecompiler::Unary<&Module::OpConvertFToU, Type::Uint, Type::Float>, - &SPIRVDecompiler::Unary<&Module::OpBitcast, Type::Uint, Type::Int>, - &SPIRVDecompiler::Binary<&Module::OpShiftLeftLogical, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpShiftRightLogical, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpShiftRightLogical, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpBitwiseAnd, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpBitwiseOr, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpBitwiseXor, Type::Uint>, - &SPIRVDecompiler::Unary<&Module::OpNot, Type::Uint>, - &SPIRVDecompiler::Quaternary<&Module::OpBitFieldInsert, Type::Uint>, - &SPIRVDecompiler::Ternary<&Module::OpBitFieldUExtract, Type::Uint>, - &SPIRVDecompiler::Unary<&Module::OpBitCount, Type::Uint>, - &SPIRVDecompiler::Unary<&Module::OpFindUMsb, Type::Uint>, - - &SPIRVDecompiler::Binary<&Module::OpFAdd, Type::HalfFloat>, - &SPIRVDecompiler::Binary<&Module::OpFMul, Type::HalfFloat>, - &SPIRVDecompiler::Ternary<&Module::OpFma, Type::HalfFloat>, - &SPIRVDecompiler::Unary<&Module::OpFAbs, Type::HalfFloat>, - &SPIRVDecompiler::HNegate, - &SPIRVDecompiler::HClamp, - &SPIRVDecompiler::HCastFloat, - &SPIRVDecompiler::HUnpack, - &SPIRVDecompiler::HMergeF32, - &SPIRVDecompiler::HMergeHN<0>, - &SPIRVDecompiler::HMergeHN<1>, - &SPIRVDecompiler::HPack2, - - &SPIRVDecompiler::LogicalAssign, - &SPIRVDecompiler::Binary<&Module::OpLogicalAnd, Type::Bool>, - &SPIRVDecompiler::Binary<&Module::OpLogicalOr, Type::Bool>, - &SPIRVDecompiler::Binary<&Module::OpLogicalNotEqual, Type::Bool>, - &SPIRVDecompiler::Unary<&Module::OpLogicalNot, Type::Bool>, - &SPIRVDecompiler::Binary<&Module::OpVectorExtractDynamic, Type::Bool, Type::Bool2, - Type::Uint>, - &SPIRVDecompiler::Unary<&Module::OpAll, Type::Bool, Type::Bool2>, - - &SPIRVDecompiler::Binary<&Module::OpFOrdLessThan, Type::Bool, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFOrdEqual, Type::Bool, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFOrdLessThanEqual, Type::Bool, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFOrdGreaterThan, Type::Bool, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFOrdNotEqual, Type::Bool, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFOrdGreaterThanEqual, Type::Bool, Type::Float>, - &SPIRVDecompiler::LogicalFOrdered, - &SPIRVDecompiler::LogicalFUnordered, - &SPIRVDecompiler::Binary<&Module::OpFUnordLessThan, Type::Bool, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFUnordEqual, Type::Bool, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFUnordLessThanEqual, Type::Bool, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFUnordGreaterThan, Type::Bool, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFUnordNotEqual, Type::Bool, Type::Float>, - &SPIRVDecompiler::Binary<&Module::OpFUnordGreaterThanEqual, Type::Bool, Type::Float>, - - &SPIRVDecompiler::Binary<&Module::OpSLessThan, Type::Bool, Type::Int>, - &SPIRVDecompiler::Binary<&Module::OpIEqual, Type::Bool, Type::Int>, - &SPIRVDecompiler::Binary<&Module::OpSLessThanEqual, Type::Bool, Type::Int>, - &SPIRVDecompiler::Binary<&Module::OpSGreaterThan, Type::Bool, Type::Int>, - &SPIRVDecompiler::Binary<&Module::OpINotEqual, Type::Bool, Type::Int>, - &SPIRVDecompiler::Binary<&Module::OpSGreaterThanEqual, Type::Bool, Type::Int>, - - &SPIRVDecompiler::Binary<&Module::OpULessThan, Type::Bool, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpIEqual, Type::Bool, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpULessThanEqual, Type::Bool, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpUGreaterThan, Type::Bool, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpINotEqual, Type::Bool, Type::Uint>, - &SPIRVDecompiler::Binary<&Module::OpUGreaterThanEqual, Type::Bool, Type::Uint>, - - &SPIRVDecompiler::LogicalAddCarry, - - &SPIRVDecompiler::Binary<&Module::OpFOrdLessThan, Type::Bool2, Type::HalfFloat>, - &SPIRVDecompiler::Binary<&Module::OpFOrdEqual, Type::Bool2, Type::HalfFloat>, - &SPIRVDecompiler::Binary<&Module::OpFOrdLessThanEqual, Type::Bool2, Type::HalfFloat>, - &SPIRVDecompiler::Binary<&Module::OpFOrdGreaterThan, Type::Bool2, Type::HalfFloat>, - &SPIRVDecompiler::Binary<&Module::OpFOrdNotEqual, Type::Bool2, Type::HalfFloat>, - &SPIRVDecompiler::Binary<&Module::OpFOrdGreaterThanEqual, Type::Bool2, Type::HalfFloat>, - // TODO(Rodrigo): Should these use the OpFUnord* variants? - &SPIRVDecompiler::Binary<&Module::OpFOrdLessThan, Type::Bool2, Type::HalfFloat>, - &SPIRVDecompiler::Binary<&Module::OpFOrdEqual, Type::Bool2, Type::HalfFloat>, - &SPIRVDecompiler::Binary<&Module::OpFOrdLessThanEqual, Type::Bool2, Type::HalfFloat>, - &SPIRVDecompiler::Binary<&Module::OpFOrdGreaterThan, Type::Bool2, Type::HalfFloat>, - &SPIRVDecompiler::Binary<&Module::OpFOrdNotEqual, Type::Bool2, Type::HalfFloat>, - &SPIRVDecompiler::Binary<&Module::OpFOrdGreaterThanEqual, Type::Bool2, Type::HalfFloat>, - - &SPIRVDecompiler::Texture, - &SPIRVDecompiler::TextureLod, - &SPIRVDecompiler::TextureGather, - &SPIRVDecompiler::TextureQueryDimensions, - &SPIRVDecompiler::TextureQueryLod, - &SPIRVDecompiler::TexelFetch, - &SPIRVDecompiler::TextureGradient, - - &SPIRVDecompiler::ImageLoad, - &SPIRVDecompiler::ImageStore, - &SPIRVDecompiler::AtomicImage<&Module::OpAtomicIAdd>, - &SPIRVDecompiler::AtomicImage<&Module::OpAtomicAnd>, - &SPIRVDecompiler::AtomicImage<&Module::OpAtomicOr>, - &SPIRVDecompiler::AtomicImage<&Module::OpAtomicXor>, - &SPIRVDecompiler::AtomicImage<&Module::OpAtomicExchange>, - - &SPIRVDecompiler::Atomic<&Module::OpAtomicExchange>, - &SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd>, - &SPIRVDecompiler::Atomic<&Module::OpAtomicUMin>, - &SPIRVDecompiler::Atomic<&Module::OpAtomicUMax>, - &SPIRVDecompiler::Atomic<&Module::OpAtomicAnd>, - &SPIRVDecompiler::Atomic<&Module::OpAtomicOr>, - &SPIRVDecompiler::Atomic<&Module::OpAtomicXor>, - - &SPIRVDecompiler::Atomic<&Module::OpAtomicExchange>, - &SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd>, - &SPIRVDecompiler::Atomic<&Module::OpAtomicSMin>, - &SPIRVDecompiler::Atomic<&Module::OpAtomicSMax>, - &SPIRVDecompiler::Atomic<&Module::OpAtomicAnd>, - &SPIRVDecompiler::Atomic<&Module::OpAtomicOr>, - &SPIRVDecompiler::Atomic<&Module::OpAtomicXor>, - - &SPIRVDecompiler::Reduce<&Module::OpAtomicIAdd>, - &SPIRVDecompiler::Reduce<&Module::OpAtomicUMin>, - &SPIRVDecompiler::Reduce<&Module::OpAtomicUMax>, - &SPIRVDecompiler::Reduce<&Module::OpAtomicAnd>, - &SPIRVDecompiler::Reduce<&Module::OpAtomicOr>, - &SPIRVDecompiler::Reduce<&Module::OpAtomicXor>, - - &SPIRVDecompiler::Reduce<&Module::OpAtomicIAdd>, - &SPIRVDecompiler::Reduce<&Module::OpAtomicSMin>, - &SPIRVDecompiler::Reduce<&Module::OpAtomicSMax>, - &SPIRVDecompiler::Reduce<&Module::OpAtomicAnd>, - &SPIRVDecompiler::Reduce<&Module::OpAtomicOr>, - &SPIRVDecompiler::Reduce<&Module::OpAtomicXor>, - - &SPIRVDecompiler::Branch, - &SPIRVDecompiler::BranchIndirect, - &SPIRVDecompiler::PushFlowStack, - &SPIRVDecompiler::PopFlowStack, - &SPIRVDecompiler::Exit, - &SPIRVDecompiler::Discard, - - &SPIRVDecompiler::EmitVertex, - &SPIRVDecompiler::EndPrimitive, - - &SPIRVDecompiler::InvocationId, - &SPIRVDecompiler::YNegate, - &SPIRVDecompiler::LocalInvocationId<0>, - &SPIRVDecompiler::LocalInvocationId<1>, - &SPIRVDecompiler::LocalInvocationId<2>, - &SPIRVDecompiler::WorkGroupId<0>, - &SPIRVDecompiler::WorkGroupId<1>, - &SPIRVDecompiler::WorkGroupId<2>, - - &SPIRVDecompiler::BallotThread, - &SPIRVDecompiler::Vote<&Module::OpSubgroupAllKHR>, - &SPIRVDecompiler::Vote<&Module::OpSubgroupAnyKHR>, - &SPIRVDecompiler::Vote<&Module::OpSubgroupAllEqualKHR>, - - &SPIRVDecompiler::ThreadId, - &SPIRVDecompiler::ThreadMask<0>, // Eq - &SPIRVDecompiler::ThreadMask<1>, // Ge - &SPIRVDecompiler::ThreadMask<2>, // Gt - &SPIRVDecompiler::ThreadMask<3>, // Le - &SPIRVDecompiler::ThreadMask<4>, // Lt - &SPIRVDecompiler::ShuffleIndexed, - - &SPIRVDecompiler::Barrier, - &SPIRVDecompiler::MemoryBarrier<spv::Scope::Workgroup>, - &SPIRVDecompiler::MemoryBarrier<spv::Scope::Device>, - }; - static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount)); - - const Device& device; - const ShaderIR& ir; - const ShaderType stage; - const Tegra::Shader::Header header; - const Registry& registry; - const Specialization& specialization; - std::unordered_map<u8, VaryingTFB> transform_feedback; - - const Id t_void = Name(TypeVoid(), "void"); - - const Id t_bool = Name(TypeBool(), "bool"); - const Id t_bool2 = Name(TypeVector(t_bool, 2), "bool2"); - - const Id t_int = Name(TypeInt(32, true), "int"); - const Id t_int2 = Name(TypeVector(t_int, 2), "int2"); - const Id t_int3 = Name(TypeVector(t_int, 3), "int3"); - const Id t_int4 = Name(TypeVector(t_int, 4), "int4"); - - const Id t_uint = Name(TypeInt(32, false), "uint"); - const Id t_uint2 = Name(TypeVector(t_uint, 2), "uint2"); - const Id t_uint3 = Name(TypeVector(t_uint, 3), "uint3"); - const Id t_uint4 = Name(TypeVector(t_uint, 4), "uint4"); - - const Id t_float = Name(TypeFloat(32), "float"); - const Id t_float2 = Name(TypeVector(t_float, 2), "float2"); - const Id t_float3 = Name(TypeVector(t_float, 3), "float3"); - const Id t_float4 = Name(TypeVector(t_float, 4), "float4"); - - const Id t_prv_bool = Name(TypePointer(spv::StorageClass::Private, t_bool), "prv_bool"); - const Id t_prv_float = Name(TypePointer(spv::StorageClass::Private, t_float), "prv_float"); - - const Id t_func_uint = Name(TypePointer(spv::StorageClass::Function, t_uint), "func_uint"); - - const Id t_in_bool = Name(TypePointer(spv::StorageClass::Input, t_bool), "in_bool"); - const Id t_in_int = Name(TypePointer(spv::StorageClass::Input, t_int), "in_int"); - const Id t_in_int4 = Name(TypePointer(spv::StorageClass::Input, t_int4), "in_int4"); - const Id t_in_uint = Name(TypePointer(spv::StorageClass::Input, t_uint), "in_uint"); - const Id t_in_uint3 = Name(TypePointer(spv::StorageClass::Input, t_uint3), "in_uint3"); - const Id t_in_uint4 = Name(TypePointer(spv::StorageClass::Input, t_uint4), "in_uint4"); - const Id t_in_float = Name(TypePointer(spv::StorageClass::Input, t_float), "in_float"); - const Id t_in_float2 = Name(TypePointer(spv::StorageClass::Input, t_float2), "in_float2"); - const Id t_in_float3 = Name(TypePointer(spv::StorageClass::Input, t_float3), "in_float3"); - const Id t_in_float4 = Name(TypePointer(spv::StorageClass::Input, t_float4), "in_float4"); - - const Id t_out_int = Name(TypePointer(spv::StorageClass::Output, t_int), "out_int"); - - const Id t_out_float = Name(TypePointer(spv::StorageClass::Output, t_float), "out_float"); - const Id t_out_float4 = Name(TypePointer(spv::StorageClass::Output, t_float4), "out_float4"); - - const Id t_cbuf_float = TypePointer(spv::StorageClass::Uniform, t_float); - const Id t_cbuf_std140 = Decorate( - Name(TypeArray(t_float4, Constant(t_uint, MaxConstBufferElements)), "CbufStd140Array"), - spv::Decoration::ArrayStride, 16U); - const Id t_cbuf_scalar = Decorate( - Name(TypeArray(t_float, Constant(t_uint, MaxConstBufferFloats)), "CbufScalarArray"), - spv::Decoration::ArrayStride, 4U); - const Id t_cbuf_std140_struct = MemberDecorate( - Decorate(TypeStruct(t_cbuf_std140), spv::Decoration::Block), 0, spv::Decoration::Offset, 0); - const Id t_cbuf_scalar_struct = MemberDecorate( - Decorate(TypeStruct(t_cbuf_scalar), spv::Decoration::Block), 0, spv::Decoration::Offset, 0); - const Id t_cbuf_std140_ubo = TypePointer(spv::StorageClass::Uniform, t_cbuf_std140_struct); - const Id t_cbuf_scalar_ubo = TypePointer(spv::StorageClass::Uniform, t_cbuf_scalar_struct); - - Id t_smem_uint{}; - - const Id t_gmem_uint = TypePointer(spv::StorageClass::StorageBuffer, t_uint); - const Id t_gmem_array = - Name(Decorate(TypeRuntimeArray(t_uint), spv::Decoration::ArrayStride, 4U), "GmemArray"); - const Id t_gmem_struct = MemberDecorate( - Decorate(TypeStruct(t_gmem_array), spv::Decoration::Block), 0, spv::Decoration::Offset, 0); - const Id t_gmem_ssbo = TypePointer(spv::StorageClass::StorageBuffer, t_gmem_struct); - - const Id t_image_uint = TypePointer(spv::StorageClass::Image, t_uint); - - const Id v_float_zero = Constant(t_float, 0.0f); - const Id v_float_one = Constant(t_float, 1.0f); - const Id v_uint_zero = Constant(t_uint, 0); - - // Nvidia uses these defaults for varyings (e.g. position and generic attributes) - const Id v_varying_default = - ConstantComposite(t_float4, v_float_zero, v_float_zero, v_float_zero, v_float_one); - - const Id v_true = ConstantTrue(t_bool); - const Id v_false = ConstantFalse(t_bool); - - Id t_scalar_half{}; - Id t_half{}; - - Id out_vertex{}; - Id in_vertex{}; - std::map<u32, Id> registers; - std::map<u32, Id> custom_variables; - std::map<Tegra::Shader::Pred, Id> predicates; - std::map<u32, Id> flow_variables; - Id local_memory{}; - Id shared_memory{}; - std::array<Id, INTERNAL_FLAGS_COUNT> internal_flags{}; - std::map<Attribute::Index, Id> input_attributes; - std::unordered_map<u8, GenericVaryingDescription> output_attributes; - std::map<u32, Id> constant_buffers; - std::map<GlobalMemoryBase, Id> global_buffers; - std::map<u32, TexelBuffer> uniform_texels; - std::map<u32, SampledImage> sampled_images; - std::map<u32, StorageImage> images; - - std::array<Id, Maxwell::NumRenderTargets> frag_colors{}; - Id instance_index{}; - Id vertex_index{}; - Id base_instance{}; - Id base_vertex{}; - Id frag_depth{}; - Id frag_coord{}; - Id front_facing{}; - Id point_coord{}; - Id tess_level_outer{}; - Id tess_level_inner{}; - Id tess_coord{}; - Id invocation_id{}; - Id workgroup_id{}; - Id local_invocation_id{}; - Id thread_id{}; - std::array<Id, 5> thread_masks{}; // eq, ge, gt, le, lt - - VertexIndices in_indices; - VertexIndices out_indices; - - std::vector<Id> interfaces; - - Id jmp_to{}; - Id ssy_flow_stack_top{}; - Id pbk_flow_stack_top{}; - Id ssy_flow_stack{}; - Id pbk_flow_stack{}; - Id continue_label{}; - std::map<u32, Id> labels; - - bool conditional_branch_set{}; - bool inside_branch{}; -}; - -class ExprDecompiler { -public: - explicit ExprDecompiler(SPIRVDecompiler& decomp_) : decomp{decomp_} {} - - Id operator()(const ExprAnd& expr) { - const Id type_def = decomp.GetTypeDefinition(Type::Bool); - const Id op1 = Visit(expr.operand1); - const Id op2 = Visit(expr.operand2); - return decomp.OpLogicalAnd(type_def, op1, op2); - } - - Id operator()(const ExprOr& expr) { - const Id type_def = decomp.GetTypeDefinition(Type::Bool); - const Id op1 = Visit(expr.operand1); - const Id op2 = Visit(expr.operand2); - return decomp.OpLogicalOr(type_def, op1, op2); - } - - Id operator()(const ExprNot& expr) { - const Id type_def = decomp.GetTypeDefinition(Type::Bool); - const Id op1 = Visit(expr.operand1); - return decomp.OpLogicalNot(type_def, op1); - } - - Id operator()(const ExprPredicate& expr) { - const auto pred = static_cast<Tegra::Shader::Pred>(expr.predicate); - return decomp.OpLoad(decomp.t_bool, decomp.predicates.at(pred)); - } - - Id operator()(const ExprCondCode& expr) { - return decomp.AsBool(decomp.Visit(decomp.ir.GetConditionCode(expr.cc))); - } - - Id operator()(const ExprVar& expr) { - return decomp.OpLoad(decomp.t_bool, decomp.flow_variables.at(expr.var_index)); - } - - Id operator()(const ExprBoolean& expr) { - return expr.value ? decomp.v_true : decomp.v_false; - } - - Id operator()(const ExprGprEqual& expr) { - const Id target = decomp.Constant(decomp.t_uint, expr.value); - Id gpr = decomp.OpLoad(decomp.t_float, decomp.registers.at(expr.gpr)); - gpr = decomp.OpBitcast(decomp.t_uint, gpr); - return decomp.OpIEqual(decomp.t_bool, gpr, target); - } - - Id Visit(const Expr& node) { - return std::visit(*this, *node); - } - -private: - SPIRVDecompiler& decomp; -}; - -class ASTDecompiler { -public: - explicit ASTDecompiler(SPIRVDecompiler& decomp_) : decomp{decomp_} {} - - void operator()(const ASTProgram& ast) { - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - } - - void operator()(const ASTIfThen& ast) { - ExprDecompiler expr_parser{decomp}; - const Id condition = expr_parser.Visit(ast.condition); - const Id then_label = decomp.OpLabel(); - const Id endif_label = decomp.OpLabel(); - decomp.OpSelectionMerge(endif_label, spv::SelectionControlMask::MaskNone); - decomp.OpBranchConditional(condition, then_label, endif_label); - decomp.AddLabel(then_label); - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - decomp.OpBranch(endif_label); - decomp.AddLabel(endif_label); - } - - void operator()([[maybe_unused]] const ASTIfElse& ast) { - UNREACHABLE(); - } - - void operator()([[maybe_unused]] const ASTBlockEncoded& ast) { - UNREACHABLE(); - } - - void operator()(const ASTBlockDecoded& ast) { - decomp.VisitBasicBlock(ast.nodes); - } - - void operator()(const ASTVarSet& ast) { - ExprDecompiler expr_parser{decomp}; - const Id condition = expr_parser.Visit(ast.condition); - decomp.OpStore(decomp.flow_variables.at(ast.index), condition); - } - - void operator()([[maybe_unused]] const ASTLabel& ast) { - // Do nothing - } - - void operator()([[maybe_unused]] const ASTGoto& ast) { - UNREACHABLE(); - } - - void operator()(const ASTDoWhile& ast) { - const Id loop_label = decomp.OpLabel(); - const Id endloop_label = decomp.OpLabel(); - const Id loop_start_block = decomp.OpLabel(); - const Id loop_continue_block = decomp.OpLabel(); - current_loop_exit = endloop_label; - decomp.OpBranch(loop_label); - decomp.AddLabel(loop_label); - decomp.OpLoopMerge(endloop_label, loop_continue_block, spv::LoopControlMask::MaskNone); - decomp.OpBranch(loop_start_block); - decomp.AddLabel(loop_start_block); - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - decomp.OpBranch(loop_continue_block); - decomp.AddLabel(loop_continue_block); - ExprDecompiler expr_parser{decomp}; - const Id condition = expr_parser.Visit(ast.condition); - decomp.OpBranchConditional(condition, loop_label, endloop_label); - decomp.AddLabel(endloop_label); - } - - void operator()(const ASTReturn& ast) { - if (!VideoCommon::Shader::ExprIsTrue(ast.condition)) { - ExprDecompiler expr_parser{decomp}; - const Id condition = expr_parser.Visit(ast.condition); - const Id then_label = decomp.OpLabel(); - const Id endif_label = decomp.OpLabel(); - decomp.OpSelectionMerge(endif_label, spv::SelectionControlMask::MaskNone); - decomp.OpBranchConditional(condition, then_label, endif_label); - decomp.AddLabel(then_label); - if (ast.kills) { - decomp.OpKill(); - } else { - decomp.PreExit(); - decomp.OpReturn(); - } - decomp.AddLabel(endif_label); - } else { - const Id next_block = decomp.OpLabel(); - decomp.OpBranch(next_block); - decomp.AddLabel(next_block); - if (ast.kills) { - decomp.OpKill(); - } else { - decomp.PreExit(); - decomp.OpReturn(); - } - decomp.AddLabel(decomp.OpLabel()); - } - } - - void operator()(const ASTBreak& ast) { - if (!VideoCommon::Shader::ExprIsTrue(ast.condition)) { - ExprDecompiler expr_parser{decomp}; - const Id condition = expr_parser.Visit(ast.condition); - const Id then_label = decomp.OpLabel(); - const Id endif_label = decomp.OpLabel(); - decomp.OpSelectionMerge(endif_label, spv::SelectionControlMask::MaskNone); - decomp.OpBranchConditional(condition, then_label, endif_label); - decomp.AddLabel(then_label); - decomp.OpBranch(current_loop_exit); - decomp.AddLabel(endif_label); - } else { - const Id next_block = decomp.OpLabel(); - decomp.OpBranch(next_block); - decomp.AddLabel(next_block); - decomp.OpBranch(current_loop_exit); - decomp.AddLabel(decomp.OpLabel()); - } - } - - void Visit(const ASTNode& node) { - std::visit(*this, *node->GetInnerData()); - } - -private: - SPIRVDecompiler& decomp; - Id current_loop_exit{}; -}; - -void SPIRVDecompiler::DecompileAST() { - const u32 num_flow_variables = ir.GetASTNumVariables(); - for (u32 i = 0; i < num_flow_variables; i++) { - const Id id = OpVariable(t_prv_bool, spv::StorageClass::Private, v_false); - Name(id, fmt::format("flow_var_{}", i)); - flow_variables.emplace(i, AddGlobalVariable(id)); - } - - DefinePrologue(); - - const ASTNode program = ir.GetASTProgram(); - ASTDecompiler decompiler{*this}; - decompiler.Visit(program); - - const Id next_block = OpLabel(); - OpBranch(next_block); - AddLabel(next_block); -} - -} // Anonymous namespace - -ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir) { - ShaderEntries entries; - for (const auto& cbuf : ir.GetConstantBuffers()) { - entries.const_buffers.emplace_back(cbuf.second, cbuf.first); - } - for (const auto& [base, usage] : ir.GetGlobalMemory()) { - entries.global_buffers.emplace_back(GlobalBufferEntry{ - .cbuf_index = base.cbuf_index, - .cbuf_offset = base.cbuf_offset, - .is_written = usage.is_written, - }); - } - for (const auto& sampler : ir.GetSamplers()) { - if (sampler.is_buffer) { - entries.uniform_texels.emplace_back(sampler); - } else { - entries.samplers.emplace_back(sampler); - } - } - for (const auto& image : ir.GetImages()) { - if (image.type == Tegra::Shader::ImageType::TextureBuffer) { - entries.storage_texels.emplace_back(image); - } else { - entries.images.emplace_back(image); - } - } - for (const auto& attribute : ir.GetInputAttributes()) { - if (IsGenericAttribute(attribute)) { - entries.attributes.insert(GetGenericAttributeLocation(attribute)); - } - } - for (const auto& buffer : entries.const_buffers) { - entries.enabled_uniform_buffers |= 1U << buffer.GetIndex(); - } - entries.clip_distances = ir.GetClipDistances(); - entries.shader_length = ir.GetLength(); - entries.uses_warps = ir.UsesWarps(); - return entries; -} - -std::vector<u32> Decompile(const Device& device, const VideoCommon::Shader::ShaderIR& ir, - ShaderType stage, const VideoCommon::Shader::Registry& registry, - const Specialization& specialization) { - return SPIRVDecompiler(device, ir, stage, registry, specialization).Assemble(); -} - -} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.h b/src/video_core/renderer_vulkan/vk_shader_decompiler.h deleted file mode 100644 index 5d94132a5..000000000 --- a/src/video_core/renderer_vulkan/vk_shader_decompiler.h +++ /dev/null @@ -1,99 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <array> -#include <set> -#include <vector> - -#include "common/common_types.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/shader_type.h" -#include "video_core/shader/registry.h" -#include "video_core/shader/shader_ir.h" - -namespace Vulkan { - -class Device; - -using Maxwell = Tegra::Engines::Maxwell3D::Regs; -using UniformTexelEntry = VideoCommon::Shader::SamplerEntry; -using SamplerEntry = VideoCommon::Shader::SamplerEntry; -using StorageTexelEntry = VideoCommon::Shader::ImageEntry; -using ImageEntry = VideoCommon::Shader::ImageEntry; - -constexpr u32 DESCRIPTOR_SET = 0; - -class ConstBufferEntry : public VideoCommon::Shader::ConstBuffer { -public: - explicit constexpr ConstBufferEntry(const ConstBuffer& entry_, u32 index_) - : ConstBuffer{entry_}, index{index_} {} - - constexpr u32 GetIndex() const { - return index; - } - -private: - u32 index{}; -}; - -struct GlobalBufferEntry { - u32 cbuf_index{}; - u32 cbuf_offset{}; - bool is_written{}; -}; - -struct ShaderEntries { - u32 NumBindings() const { - return static_cast<u32>(const_buffers.size() + global_buffers.size() + - uniform_texels.size() + samplers.size() + storage_texels.size() + - images.size()); - } - - std::vector<ConstBufferEntry> const_buffers; - std::vector<GlobalBufferEntry> global_buffers; - std::vector<UniformTexelEntry> uniform_texels; - std::vector<SamplerEntry> samplers; - std::vector<StorageTexelEntry> storage_texels; - std::vector<ImageEntry> images; - std::set<u32> attributes; - std::array<bool, Maxwell::NumClipDistances> clip_distances{}; - std::size_t shader_length{}; - u32 enabled_uniform_buffers{}; - bool uses_warps{}; -}; - -struct Specialization final { - u32 base_binding{}; - - // Compute specific - std::array<u32, 3> workgroup_size{}; - u32 shared_memory_size{}; - - // Graphics specific - std::optional<float> point_size; - std::bitset<Maxwell::NumVertexAttributes> enabled_attributes; - std::array<Maxwell::VertexAttribute::Type, Maxwell::NumVertexAttributes> attribute_types{}; - bool ndc_minus_one_to_one{}; - bool early_fragment_tests{}; - float alpha_test_ref{}; - Maxwell::ComparisonOp alpha_test_func{}; -}; -// Old gcc versions don't consider this trivially copyable. -// static_assert(std::is_trivially_copyable_v<Specialization>); - -struct SPIRVShader { - std::vector<u32> code; - ShaderEntries entries; -}; - -ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir); - -std::vector<u32> Decompile(const Device& device, const VideoCommon::Shader::ShaderIR& ir, - Tegra::Engines::ShaderType stage, - const VideoCommon::Shader::Registry& registry, - const Specialization& specialization); - -} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp index 0412b5234..555b12ed7 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp @@ -91,7 +91,7 @@ StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& mem .flags = 0, .size = STREAM_BUFFER_SIZE, .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | - VK_BUFFER_USAGE_INDEX_BUFFER_BIT, + VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, .sharingMode = VK_SHARING_MODE_EXCLUSIVE, .queueFamilyIndexCount = 0, .pQueueFamilyIndices = nullptr, diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.cpp b/src/video_core/renderer_vulkan/vk_state_tracker.cpp index 956f86845..e3b7dd61c 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.cpp +++ b/src/video_core/renderer_vulkan/vk_state_tracker.cpp @@ -29,9 +29,10 @@ using Flags = Maxwell3D::DirtyState::Flags; Flags MakeInvalidationFlags() { static constexpr int INVALIDATION_FLAGS[]{ - Viewports, Scissors, DepthBias, BlendConstants, DepthBounds, - StencilProperties, CullMode, DepthBoundsEnable, DepthTestEnable, DepthWriteEnable, - DepthCompareOp, FrontFace, StencilOp, StencilTestEnable, VertexBuffers, + Viewports, Scissors, DepthBias, BlendConstants, DepthBounds, + StencilProperties, LineWidth, CullMode, DepthBoundsEnable, DepthTestEnable, + DepthWriteEnable, DepthCompareOp, FrontFace, StencilOp, StencilTestEnable, + VertexBuffers, VertexInput, }; Flags flags{}; for (const int flag : INVALIDATION_FLAGS) { @@ -40,6 +41,12 @@ Flags MakeInvalidationFlags() { for (int index = VertexBuffer0; index <= VertexBuffer31; ++index) { flags[index] = true; } + for (int index = VertexAttribute0; index <= VertexAttribute31; ++index) { + flags[index] = true; + } + for (int index = VertexBinding0; index <= VertexBinding31; ++index) { + flags[index] = true; + } return flags; } @@ -79,6 +86,11 @@ void SetupDirtyStencilProperties(Tables& tables) { table[OFF(stencil_back_func_mask)] = StencilProperties; } +void SetupDirtyLineWidth(Tables& tables) { + tables[0][OFF(line_width_smooth)] = LineWidth; + tables[0][OFF(line_width_aliased)] = LineWidth; +} + void SetupDirtyCullMode(Tables& tables) { auto& table = tables[0]; table[OFF(cull_face)] = CullMode; @@ -134,31 +146,38 @@ void SetupDirtyBlending(Tables& tables) { FillBlock(tables[0], OFF(independent_blend), NUM(independent_blend), Blending); } -void SetupDirtyInstanceDivisors(Tables& tables) { - static constexpr size_t divisor_offset = 3; - for (size_t index = 0; index < Regs::NumVertexArrays; ++index) { - tables[0][OFF(instanced_arrays) + index] = InstanceDivisors; - tables[0][OFF(vertex_array) + index * NUM(vertex_array[0]) + divisor_offset] = - InstanceDivisors; +void SetupDirtyViewportSwizzles(Tables& tables) { + static constexpr size_t swizzle_offset = 6; + for (size_t index = 0; index < Regs::NumViewports; ++index) { + tables[0][OFF(viewport_transform) + index * NUM(viewport_transform[0]) + swizzle_offset] = + ViewportSwizzles; } } void SetupDirtyVertexAttributes(Tables& tables) { - FillBlock(tables[0], OFF(vertex_attrib_format), NUM(vertex_attrib_format), VertexAttributes); + for (size_t i = 0; i < Regs::NumVertexAttributes; ++i) { + const size_t offset = OFF(vertex_attrib_format) + i * NUM(vertex_attrib_format[0]); + FillBlock(tables[0], offset, NUM(vertex_attrib_format[0]), VertexAttribute0 + i); + } + FillBlock(tables[1], OFF(vertex_attrib_format), Regs::NumVertexAttributes, VertexInput); } -void SetupDirtyViewportSwizzles(Tables& tables) { - static constexpr size_t swizzle_offset = 6; - for (size_t index = 0; index < Regs::NumViewports; ++index) { - tables[0][OFF(viewport_transform) + index * NUM(viewport_transform[0]) + swizzle_offset] = - ViewportSwizzles; +void SetupDirtyVertexBindings(Tables& tables) { + // Do NOT include stride here, it's implicit in VertexBuffer + static constexpr size_t divisor_offset = 3; + for (size_t i = 0; i < Regs::NumVertexArrays; ++i) { + const u8 flag = static_cast<u8>(VertexBinding0 + i); + tables[0][OFF(instanced_arrays) + i] = VertexInput; + tables[1][OFF(instanced_arrays) + i] = flag; + tables[0][OFF(vertex_array) + i * NUM(vertex_array[0]) + divisor_offset] = VertexInput; + tables[1][OFF(vertex_array) + i * NUM(vertex_array[0]) + divisor_offset] = flag; } } } // Anonymous namespace StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags}, invalidation_flags{MakeInvalidationFlags()} { - auto& tables = gpu.Maxwell3D().dirty.tables; + auto& tables{gpu.Maxwell3D().dirty.tables}; SetupDirtyFlags(tables); SetupDirtyViewports(tables); SetupDirtyScissors(tables); @@ -166,6 +185,7 @@ StateTracker::StateTracker(Tegra::GPU& gpu) SetupDirtyBlendConstants(tables); SetupDirtyDepthBounds(tables); SetupDirtyStencilProperties(tables); + SetupDirtyLineWidth(tables); SetupDirtyCullMode(tables); SetupDirtyDepthBoundsEnable(tables); SetupDirtyDepthTestEnable(tables); @@ -175,9 +195,9 @@ StateTracker::StateTracker(Tegra::GPU& gpu) SetupDirtyStencilOp(tables); SetupDirtyStencilTestEnable(tables); SetupDirtyBlending(tables); - SetupDirtyInstanceDivisors(tables); - SetupDirtyVertexAttributes(tables); SetupDirtyViewportSwizzles(tables); + SetupDirtyVertexAttributes(tables); + SetupDirtyVertexBindings(tables); } } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.h b/src/video_core/renderer_vulkan/vk_state_tracker.h index 84e918a71..5f78f6950 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.h +++ b/src/video_core/renderer_vulkan/vk_state_tracker.h @@ -19,12 +19,19 @@ namespace Dirty { enum : u8 { First = VideoCommon::Dirty::LastCommonEntry, + VertexInput, + VertexAttribute0, + VertexAttribute31 = VertexAttribute0 + 31, + VertexBinding0, + VertexBinding31 = VertexBinding0 + 31, + Viewports, Scissors, DepthBias, BlendConstants, DepthBounds, StencilProperties, + LineWidth, CullMode, DepthBoundsEnable, @@ -36,11 +43,9 @@ enum : u8 { StencilTestEnable, Blending, - InstanceDivisors, - VertexAttributes, ViewportSwizzles, - Last + Last, }; static_assert(Last <= std::numeric_limits<u8>::max()); @@ -89,6 +94,10 @@ public: return Exchange(Dirty::StencilProperties, false); } + bool TouchLineWidth() const { + return Exchange(Dirty::LineWidth, false); + } + bool TouchCullMode() { return Exchange(Dirty::CullMode, false); } diff --git a/src/video_core/renderer_vulkan/vk_swapchain.cpp b/src/video_core/renderer_vulkan/vk_swapchain.cpp index dfd5c65ba..d990eefba 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.cpp +++ b/src/video_core/renderer_vulkan/vk_swapchain.cpp @@ -65,6 +65,9 @@ VKSwapchain::VKSwapchain(VkSurfaceKHR surface_, const Device& device_, VKSchedul VKSwapchain::~VKSwapchain() = default; void VKSwapchain::Create(u32 width, u32 height, bool srgb) { + is_outdated = false; + is_suboptimal = false; + const auto physical_device = device.GetPhysical(); const auto capabilities{physical_device.GetSurfaceCapabilitiesKHR(surface)}; if (capabilities.maxImageExtent.width == 0 || capabilities.maxImageExtent.height == 0) { @@ -82,21 +85,31 @@ void VKSwapchain::Create(u32 width, u32 height, bool srgb) { resource_ticks.resize(image_count); } -bool VKSwapchain::AcquireNextImage() { - const VkResult result = - device.GetLogical().AcquireNextImageKHR(*swapchain, std::numeric_limits<u64>::max(), - *present_semaphores[frame_index], {}, &image_index); - +void VKSwapchain::AcquireNextImage() { + const VkResult result = device.GetLogical().AcquireNextImageKHR( + *swapchain, std::numeric_limits<u64>::max(), *present_semaphores[frame_index], + VK_NULL_HANDLE, &image_index); + switch (result) { + case VK_SUCCESS: + break; + case VK_SUBOPTIMAL_KHR: + is_suboptimal = true; + break; + case VK_ERROR_OUT_OF_DATE_KHR: + is_outdated = true; + break; + default: + LOG_ERROR(Render_Vulkan, "vkAcquireNextImageKHR returned {}", vk::ToString(result)); + break; + } scheduler.Wait(resource_ticks[image_index]); - return result == VK_SUCCESS || result == VK_SUBOPTIMAL_KHR; + resource_ticks[image_index] = scheduler.CurrentTick(); } -bool VKSwapchain::Present(VkSemaphore render_semaphore) { +void VKSwapchain::Present(VkSemaphore render_semaphore) { const VkSemaphore present_semaphore{*present_semaphores[frame_index]}; const std::array<VkSemaphore, 2> semaphores{present_semaphore, render_semaphore}; const auto present_queue{device.GetPresentQueue()}; - bool recreated = false; - const VkPresentInfoKHR present_info{ .sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR, .pNext = nullptr, @@ -107,7 +120,6 @@ bool VKSwapchain::Present(VkSemaphore render_semaphore) { .pImageIndices = &image_index, .pResults = nullptr, }; - switch (const VkResult result = present_queue.Present(present_info)) { case VK_SUCCESS: break; @@ -115,24 +127,16 @@ bool VKSwapchain::Present(VkSemaphore render_semaphore) { LOG_DEBUG(Render_Vulkan, "Suboptimal swapchain"); break; case VK_ERROR_OUT_OF_DATE_KHR: - if (current_width > 0 && current_height > 0) { - Create(current_width, current_height, current_srgb); - recreated = true; - } + is_outdated = true; break; default: LOG_CRITICAL(Render_Vulkan, "Failed to present with error {}", vk::ToString(result)); break; } - - resource_ticks[image_index] = scheduler.CurrentTick(); - frame_index = (frame_index + 1) % static_cast<u32>(image_count); - return recreated; -} - -bool VKSwapchain::HasFramebufferChanged(const Layout::FramebufferLayout& framebuffer) const { - // TODO(Rodrigo): Handle framebuffer pixel format changes - return framebuffer.width != current_width || framebuffer.height != current_height; + ++frame_index; + if (frame_index >= image_count) { + frame_index = 0; + } } void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, u32 width, @@ -148,7 +152,6 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, if (capabilities.maxImageCount > 0 && requested_image_count > capabilities.maxImageCount) { requested_image_count = capabilities.maxImageCount; } - VkSwapchainCreateInfoKHR swapchain_ci{ .sType = VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR, .pNext = nullptr, @@ -169,7 +172,6 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, .clipped = VK_FALSE, .oldSwapchain = nullptr, }; - const u32 graphics_family{device.GetGraphicsFamily()}; const u32 present_family{device.GetPresentFamily()}; const std::array<u32, 2> queue_indices{graphics_family, present_family}; @@ -178,7 +180,6 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, swapchain_ci.queueFamilyIndexCount = static_cast<u32>(queue_indices.size()); swapchain_ci.pQueueFamilyIndices = queue_indices.data(); } - // Request the size again to reduce the possibility of a TOCTOU race condition. const auto updated_capabilities = physical_device.GetSurfaceCapabilitiesKHR(surface); swapchain_ci.imageExtent = ChooseSwapExtent(updated_capabilities, width, height); @@ -186,8 +187,6 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, swapchain = device.GetLogical().CreateSwapchainKHR(swapchain_ci); extent = swapchain_ci.imageExtent; - current_width = extent.width; - current_height = extent.height; current_srgb = srgb; images = swapchain.GetImages(); @@ -197,8 +196,8 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, void VKSwapchain::CreateSemaphores() { present_semaphores.resize(image_count); - std::generate(present_semaphores.begin(), present_semaphores.end(), - [this] { return device.GetLogical().CreateSemaphore(); }); + std::ranges::generate(present_semaphores, + [this] { return device.GetLogical().CreateSemaphore(); }); } void VKSwapchain::CreateImageViews() { diff --git a/src/video_core/renderer_vulkan/vk_swapchain.h b/src/video_core/renderer_vulkan/vk_swapchain.h index adc8d27cf..35c2cdc14 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.h +++ b/src/video_core/renderer_vulkan/vk_swapchain.h @@ -28,14 +28,25 @@ public: void Create(u32 width, u32 height, bool srgb); /// Acquires the next image in the swapchain, waits as needed. - bool AcquireNextImage(); + void AcquireNextImage(); - /// Presents the rendered image to the swapchain. Returns true when the swapchains had to be - /// recreated. Takes responsability for the ownership of fence. - bool Present(VkSemaphore render_semaphore); + /// Presents the rendered image to the swapchain. + void Present(VkSemaphore render_semaphore); - /// Returns true when the framebuffer layout has changed. - bool HasFramebufferChanged(const Layout::FramebufferLayout& framebuffer) const; + /// Returns true when the color space has changed. + bool HasColorSpaceChanged(bool is_srgb) const { + return current_srgb != is_srgb; + } + + /// Returns true when the swapchain is outdated. + bool IsOutDated() const { + return is_outdated; + } + + /// Returns true when the swapchain is suboptimal. + bool IsSubOptimal() const { + return is_suboptimal; + } VkExtent2D GetSize() const { return extent; @@ -61,10 +72,6 @@ public: return image_format; } - bool GetSrgbState() const { - return current_srgb; - } - private: void CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, u32 width, u32 height, bool srgb); @@ -92,9 +99,9 @@ private: VkFormat image_format{}; VkExtent2D extent{}; - u32 current_width{}; - u32 current_height{}; bool current_srgb{}; + bool is_outdated{}; + bool is_suboptimal{}; }; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index a2ab4d1ee..8e029bcb3 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -15,6 +15,7 @@ #include "video_core/renderer_vulkan/maxwell_to_vk.h" #include "video_core/renderer_vulkan/vk_compute_pass.h" #include "video_core/renderer_vulkan/vk_rasterizer.h" +#include "video_core/renderer_vulkan/vk_render_pass_cache.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" #include "video_core/renderer_vulkan/vk_texture_cache.h" @@ -34,19 +35,6 @@ using VideoCommon::SubresourceRange; using VideoCore::Surface::IsPixelFormatASTC; namespace { - -constexpr std::array ATTACHMENT_REFERENCES{ - VkAttachmentReference{0, VK_IMAGE_LAYOUT_GENERAL}, - VkAttachmentReference{1, VK_IMAGE_LAYOUT_GENERAL}, - VkAttachmentReference{2, VK_IMAGE_LAYOUT_GENERAL}, - VkAttachmentReference{3, VK_IMAGE_LAYOUT_GENERAL}, - VkAttachmentReference{4, VK_IMAGE_LAYOUT_GENERAL}, - VkAttachmentReference{5, VK_IMAGE_LAYOUT_GENERAL}, - VkAttachmentReference{6, VK_IMAGE_LAYOUT_GENERAL}, - VkAttachmentReference{7, VK_IMAGE_LAYOUT_GENERAL}, - VkAttachmentReference{8, VK_IMAGE_LAYOUT_GENERAL}, -}; - constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { if (color == std::array<float, 4>{0, 0, 0, 0}) { return VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK; @@ -174,25 +162,6 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { return device.GetLogical().CreateImage(MakeImageCreateInfo(device, info)); } -[[nodiscard]] vk::Buffer MakeBuffer(const Device& device, const ImageInfo& info) { - if (info.type != ImageType::Buffer) { - return vk::Buffer{}; - } - const size_t bytes_per_block = VideoCore::Surface::BytesPerBlock(info.format); - return device.GetLogical().CreateBuffer(VkBufferCreateInfo{ - .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .size = info.size.width * bytes_per_block, - .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | - VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | - VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT, - .sharingMode = VK_SHARING_MODE_EXCLUSIVE, - .queueFamilyIndexCount = 0, - .pQueueFamilyIndices = nullptr, - }); -} - [[nodiscard]] VkImageAspectFlags ImageAspectMask(PixelFormat format) { switch (VideoCore::Surface::GetFormatType(format)) { case VideoCore::Surface::SurfaceType::ColorTexture: @@ -226,23 +195,6 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { } } -[[nodiscard]] VkAttachmentDescription AttachmentDescription(const Device& device, - const ImageView* image_view) { - using MaxwellToVK::SurfaceFormat; - const PixelFormat pixel_format = image_view->format; - return VkAttachmentDescription{ - .flags = VK_ATTACHMENT_DESCRIPTION_MAY_ALIAS_BIT, - .format = SurfaceFormat(device, FormatType::Optimal, true, pixel_format).format, - .samples = image_view->Samples(), - .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD, - .storeOp = VK_ATTACHMENT_STORE_OP_STORE, - .stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD, - .stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE, - .initialLayout = VK_IMAGE_LAYOUT_GENERAL, - .finalLayout = VK_IMAGE_LAYOUT_GENERAL, - }; -} - [[nodiscard]] VkComponentSwizzle ComponentSwizzle(SwizzleSource swizzle) { switch (swizzle) { case SwizzleSource::Zero: @@ -263,6 +215,30 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { return VK_COMPONENT_SWIZZLE_ZERO; } +[[nodiscard]] VkImageViewType ImageViewType(Shader::TextureType type) { + switch (type) { + case Shader::TextureType::Color1D: + return VK_IMAGE_VIEW_TYPE_1D; + case Shader::TextureType::Color2D: + return VK_IMAGE_VIEW_TYPE_2D; + case Shader::TextureType::ColorCube: + return VK_IMAGE_VIEW_TYPE_CUBE; + case Shader::TextureType::Color3D: + return VK_IMAGE_VIEW_TYPE_3D; + case Shader::TextureType::ColorArray1D: + return VK_IMAGE_VIEW_TYPE_1D_ARRAY; + case Shader::TextureType::ColorArray2D: + return VK_IMAGE_VIEW_TYPE_2D_ARRAY; + case Shader::TextureType::ColorArrayCube: + return VK_IMAGE_VIEW_TYPE_CUBE_ARRAY; + case Shader::TextureType::Buffer: + UNREACHABLE_MSG("Texture buffers can't be image views"); + return VK_IMAGE_VIEW_TYPE_1D; + } + UNREACHABLE_MSG("Invalid image view type={}", type); + return VK_IMAGE_VIEW_TYPE_2D; +} + [[nodiscard]] VkImageViewType ImageViewType(VideoCommon::ImageViewType type) { switch (type) { case VideoCommon::ImageViewType::e1D: @@ -280,7 +256,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { case VideoCommon::ImageViewType::CubeArray: return VK_IMAGE_VIEW_TYPE_CUBE_ARRAY; case VideoCommon::ImageViewType::Rect: - LOG_WARNING(Render_Vulkan, "Unnormalized image view type not supported"); + UNIMPLEMENTED_MSG("Rect image view"); return VK_IMAGE_VIEW_TYPE_2D; case VideoCommon::ImageViewType::Buffer: UNREACHABLE_MSG("Texture buffers can't be image views"); @@ -327,7 +303,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { }; } -[[nodiscard]] std::vector<VkBufferCopy> TransformBufferCopies( +[[maybe_unused]] [[nodiscard]] std::vector<VkBufferCopy> TransformBufferCopies( std::span<const VideoCommon::BufferCopy> copies, size_t buffer_offset) { std::vector<VkBufferCopy> result(copies.size()); std::ranges::transform( @@ -587,6 +563,28 @@ struct RangedBarrierRange { } }; +[[nodiscard]] VkFormat Format(Shader::ImageFormat format) { + switch (format) { + case Shader::ImageFormat::Typeless: + break; + case Shader::ImageFormat::R8_SINT: + return VK_FORMAT_R8_SINT; + case Shader::ImageFormat::R8_UINT: + return VK_FORMAT_R8_UINT; + case Shader::ImageFormat::R16_UINT: + return VK_FORMAT_R16_UINT; + case Shader::ImageFormat::R16_SINT: + return VK_FORMAT_R16_SINT; + case Shader::ImageFormat::R32_UINT: + return VK_FORMAT_R32_UINT; + case Shader::ImageFormat::R32G32_UINT: + return VK_FORMAT_R32G32_UINT; + case Shader::ImageFormat::R32G32B32A32_UINT: + return VK_FORMAT_R32G32B32A32_UINT; + } + UNREACHABLE_MSG("Invalid image format={}", format); + return VK_FORMAT_R32_UINT; +} } // Anonymous namespace void TextureCacheRuntime::Finish() { @@ -608,7 +606,10 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst const VkImageAspectFlags aspect_mask = ImageAspectMask(src.format); const bool is_dst_msaa = dst.Samples() != VK_SAMPLE_COUNT_1_BIT; const bool is_src_msaa = src.Samples() != VK_SAMPLE_COUNT_1_BIT; - ASSERT(aspect_mask == ImageAspectMask(dst.format)); + if (aspect_mask != ImageAspectMask(dst.format)) { + UNIMPLEMENTED_MSG("Incompatible blit from format {} to {}", src.format, dst.format); + return; + } if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT && !is_src_msaa && !is_dst_msaa) { blit_image_helper.BlitColor(dst_framebuffer, src, dst_region, src_region, filter, operation); @@ -622,7 +623,7 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst return; } } - ASSERT(src.ImageFormat() == dst.ImageFormat()); + ASSERT(src.format == dst.format); ASSERT(!(is_dst_msaa && !is_src_msaa)); ASSERT(operation == Fermi2D::Operation::SrcCopy); @@ -762,7 +763,7 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, dst_range.AddLayers(copy.dstSubresource); src_range.AddLayers(copy.srcSubresource); } - const std::array read_barriers{ + const std::array pre_barriers{ VkImageMemoryBarrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, .pNext = nullptr, @@ -771,7 +772,7 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, VK_ACCESS_TRANSFER_WRITE_BIT, .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, - .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .image = src_image, @@ -792,29 +793,43 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, .subresourceRange = dst_range.SubresourceRange(aspect_mask), }, }; - const VkImageMemoryBarrier write_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, - .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, - .newLayout = VK_IMAGE_LAYOUT_GENERAL, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .image = dst_image, - .subresourceRange = dst_range.SubresourceRange(aspect_mask), + const std::array post_barriers{ + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = 0, + .dstAccessMask = 0, + .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = src_image, + .subresourceRange = src_range.SubresourceRange(aspect_mask), + }, + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | + VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | + VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | + VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = dst_image, + .subresourceRange = dst_range.SubresourceRange(aspect_mask), + }, }; cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, {}, {}, read_barriers); - cmdbuf.CopyImage(src_image, VK_IMAGE_LAYOUT_GENERAL, dst_image, + 0, {}, {}, pre_barriers); + cmdbuf.CopyImage(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, vk_copies); cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, - 0, write_barrier); + 0, {}, {}, post_barriers); }); } @@ -825,13 +840,9 @@ u64 TextureCacheRuntime::GetDeviceLocalMemory() const { Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_addr_, VAddr cpu_addr_) : VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), scheduler{&runtime.scheduler}, - image(MakeImage(runtime.device, info)), buffer(MakeBuffer(runtime.device, info)), + image(MakeImage(runtime.device, info)), + commit(runtime.memory_allocator.Commit(image, MemoryUsage::DeviceLocal)), aspect_mask(ImageAspectMask(info.format)) { - if (image) { - commit = runtime.memory_allocator.Commit(image, MemoryUsage::DeviceLocal); - } else { - commit = runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal); - } if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) { if (Settings::values.accelerate_astc.GetValue()) { flags |= VideoCommon::ImageFlagBits::AcceleratedUpload; @@ -840,11 +851,7 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_ } } if (runtime.device.HasDebuggingToolAttached()) { - if (image) { - image.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); - } else { - buffer.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); - } + image.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); } static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO, @@ -896,21 +903,9 @@ void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImag }); } -void Image::UploadMemory(const StagingBufferRef& map, - std::span<const VideoCommon::BufferCopy> copies) { - // TODO: Move this to another API - scheduler->RequestOutsideRenderPassOperationContext(); - std::vector vk_copies = TransformBufferCopies(copies, map.offset); - const VkBuffer src_buffer = map.buffer; - const VkBuffer dst_buffer = *buffer; - scheduler->Record([src_buffer, dst_buffer, vk_copies](vk::CommandBuffer cmdbuf) { - // TODO: Barriers - cmdbuf.CopyBuffer(src_buffer, dst_buffer, vk_copies); - }); -} - void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) { std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask); + scheduler->RequestOutsideRenderPassOperationContext(); scheduler->Record([buffer = map.buffer, image = *image, aspect_mask = aspect_mask, vk_copies](vk::CommandBuffer cmdbuf) { const VkImageMemoryBarrier read_barrier{ @@ -966,8 +961,9 @@ void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferIm ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewInfo& info, ImageId image_id_, Image& image) : VideoCommon::ImageViewBase{info, image.info, image_id_}, device{&runtime.device}, - image_handle{image.Handle()}, image_format{image.info.format}, samples{ConvertSampleCount( - image.info.num_samples)} { + image_handle{image.Handle()}, samples{ConvertSampleCount(image.info.num_samples)} { + using Shader::TextureType; + const VkImageAspectFlags aspect_mask = ImageViewAspectMask(info); std::array<SwizzleSource, 4> swizzle{ SwizzleSource::R, @@ -1005,57 +1001,54 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI }, .subresourceRange = MakeSubresourceRange(aspect_mask, info.range), }; - const auto create = [&](VideoCommon::ImageViewType view_type, std::optional<u32> num_layers) { + const auto create = [&](TextureType tex_type, std::optional<u32> num_layers) { VkImageViewCreateInfo ci{create_info}; - ci.viewType = ImageViewType(view_type); + ci.viewType = ImageViewType(tex_type); if (num_layers) { ci.subresourceRange.layerCount = *num_layers; } vk::ImageView handle = device->GetLogical().CreateImageView(ci); if (device->HasDebuggingToolAttached()) { - handle.SetObjectNameEXT(VideoCommon::Name(*this, view_type).c_str()); + handle.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); } - image_views[static_cast<size_t>(view_type)] = std::move(handle); + image_views[static_cast<size_t>(tex_type)] = std::move(handle); }; switch (info.type) { case VideoCommon::ImageViewType::e1D: case VideoCommon::ImageViewType::e1DArray: - create(VideoCommon::ImageViewType::e1D, 1); - create(VideoCommon::ImageViewType::e1DArray, std::nullopt); - render_target = Handle(VideoCommon::ImageViewType::e1DArray); + create(TextureType::Color1D, 1); + create(TextureType::ColorArray1D, std::nullopt); + render_target = Handle(TextureType::ColorArray1D); break; case VideoCommon::ImageViewType::e2D: case VideoCommon::ImageViewType::e2DArray: - create(VideoCommon::ImageViewType::e2D, 1); - create(VideoCommon::ImageViewType::e2DArray, std::nullopt); - render_target = Handle(VideoCommon::ImageViewType::e2DArray); + create(TextureType::Color2D, 1); + create(TextureType::ColorArray2D, std::nullopt); + render_target = Handle(Shader::TextureType::ColorArray2D); break; case VideoCommon::ImageViewType::e3D: - create(VideoCommon::ImageViewType::e3D, std::nullopt); - render_target = Handle(VideoCommon::ImageViewType::e3D); + create(TextureType::Color3D, std::nullopt); + render_target = Handle(Shader::TextureType::Color3D); break; case VideoCommon::ImageViewType::Cube: case VideoCommon::ImageViewType::CubeArray: - create(VideoCommon::ImageViewType::Cube, 6); - create(VideoCommon::ImageViewType::CubeArray, std::nullopt); + create(TextureType::ColorCube, 6); + create(TextureType::ColorArrayCube, std::nullopt); break; case VideoCommon::ImageViewType::Rect: UNIMPLEMENTED(); break; case VideoCommon::ImageViewType::Buffer: - buffer_view = device->GetLogical().CreateBufferView(VkBufferViewCreateInfo{ - .sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .buffer = image.Buffer(), - .format = format_info.format, - .offset = 0, // TODO: Redesign buffer cache to support this - .range = image.guest_size_bytes, - }); + UNREACHABLE(); break; } } +ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, + const VideoCommon::ImageViewInfo& view_info, GPUVAddr gpu_addr_) + : VideoCommon::ImageViewBase{info, view_info}, gpu_addr{gpu_addr_}, + buffer_size{VideoCommon::CalculateGuestSizeInBytes(info)} {} + ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams& params) : VideoCommon::ImageViewBase{params} {} @@ -1063,7 +1056,8 @@ VkImageView ImageView::DepthView() { if (depth_view) { return *depth_view; } - depth_view = MakeDepthStencilView(VK_IMAGE_ASPECT_DEPTH_BIT); + const auto& info = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format); + depth_view = MakeView(info.format, VK_IMAGE_ASPECT_DEPTH_BIT); return *depth_view; } @@ -1071,18 +1065,38 @@ VkImageView ImageView::StencilView() { if (stencil_view) { return *stencil_view; } - stencil_view = MakeDepthStencilView(VK_IMAGE_ASPECT_STENCIL_BIT); + const auto& info = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format); + stencil_view = MakeView(info.format, VK_IMAGE_ASPECT_STENCIL_BIT); return *stencil_view; } -vk::ImageView ImageView::MakeDepthStencilView(VkImageAspectFlags aspect_mask) { +VkImageView ImageView::StorageView(Shader::TextureType texture_type, + Shader::ImageFormat image_format) { + if (image_format == Shader::ImageFormat::Typeless) { + return Handle(texture_type); + } + const bool is_signed{image_format == Shader::ImageFormat::R8_SINT || + image_format == Shader::ImageFormat::R16_SINT}; + if (!storage_views) { + storage_views = std::make_unique<StorageViews>(); + } + auto& views{is_signed ? storage_views->signeds : storage_views->unsigneds}; + auto& view{views[static_cast<size_t>(texture_type)]}; + if (view) { + return *view; + } + view = MakeView(Format(image_format), VK_IMAGE_ASPECT_COLOR_BIT); + return *view; +} + +vk::ImageView ImageView::MakeView(VkFormat vk_format, VkImageAspectFlags aspect_mask) { return device->GetLogical().CreateImageView({ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, .pNext = nullptr, .flags = 0, .image = image_handle, .viewType = ImageViewType(type), - .format = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format).format, + .format = vk_format, .components{ .r = VK_COMPONENT_SWIZZLE_IDENTITY, .g = VK_COMPONENT_SWIZZLE_IDENTITY, @@ -1146,7 +1160,6 @@ Sampler::Sampler(TextureCacheRuntime& runtime, const Tegra::Texture::TSCEntry& t Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM_RT> color_buffers, ImageView* depth_buffer, const VideoCommon::RenderTargets& key) { - std::vector<VkAttachmentDescription> descriptions; std::vector<VkImageView> attachments; RenderPassKey renderpass_key{}; s32 num_layers = 1; @@ -1157,7 +1170,6 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM renderpass_key.color_formats[index] = PixelFormat::Invalid; continue; } - descriptions.push_back(AttachmentDescription(runtime.device, color_buffer)); attachments.push_back(color_buffer->RenderTarget()); renderpass_key.color_formats[index] = color_buffer->format; num_layers = std::max(num_layers, color_buffer->range.extent.layers); @@ -1167,10 +1179,7 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM ++num_images; } const size_t num_colors = attachments.size(); - const VkAttachmentReference* depth_attachment = - depth_buffer ? &ATTACHMENT_REFERENCES[num_colors] : nullptr; if (depth_buffer) { - descriptions.push_back(AttachmentDescription(runtime.device, depth_buffer)); attachments.push_back(depth_buffer->RenderTarget()); renderpass_key.depth_format = depth_buffer->format; num_layers = std::max(num_layers, depth_buffer->range.extent.layers); @@ -1183,40 +1192,14 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM } renderpass_key.samples = samples; - const auto& device = runtime.device.GetLogical(); - const auto [cache_pair, is_new] = runtime.renderpass_cache.try_emplace(renderpass_key); - if (is_new) { - const VkSubpassDescription subpass{ - .flags = 0, - .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS, - .inputAttachmentCount = 0, - .pInputAttachments = nullptr, - .colorAttachmentCount = static_cast<u32>(num_colors), - .pColorAttachments = num_colors != 0 ? ATTACHMENT_REFERENCES.data() : nullptr, - .pResolveAttachments = nullptr, - .pDepthStencilAttachment = depth_attachment, - .preserveAttachmentCount = 0, - .pPreserveAttachments = nullptr, - }; - cache_pair->second = device.CreateRenderPass(VkRenderPassCreateInfo{ - .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .attachmentCount = static_cast<u32>(descriptions.size()), - .pAttachments = descriptions.data(), - .subpassCount = 1, - .pSubpasses = &subpass, - .dependencyCount = 0, - .pDependencies = nullptr, - }); - } - renderpass = *cache_pair->second; + renderpass = runtime.render_pass_cache.Get(renderpass_key); + render_area = VkExtent2D{ .width = key.size.width, .height = key.size.height, }; num_color_buffers = static_cast<u32>(num_colors); - framebuffer = device.CreateFramebuffer(VkFramebufferCreateInfo{ + framebuffer = runtime.device.GetLogical().CreateFramebuffer({ .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO, .pNext = nullptr, .flags = 0, diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 172bcdf98..0b73d55f8 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -7,6 +7,7 @@ #include <compare> #include <span> +#include "shader_recompiler/shader_info.h" #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" #include "video_core/texture_cache/texture_cache.h" #include "video_core/vulkan_common/vulkan_memory_allocator.h" @@ -26,35 +27,10 @@ class Device; class Image; class ImageView; class Framebuffer; +class RenderPassCache; class StagingBufferPool; class VKScheduler; -struct RenderPassKey { - constexpr auto operator<=>(const RenderPassKey&) const noexcept = default; - - std::array<PixelFormat, NUM_RT> color_formats; - PixelFormat depth_format; - VkSampleCountFlagBits samples; -}; - -} // namespace Vulkan - -namespace std { -template <> -struct hash<Vulkan::RenderPassKey> { - [[nodiscard]] constexpr size_t operator()(const Vulkan::RenderPassKey& key) const noexcept { - size_t value = static_cast<size_t>(key.depth_format) << 48; - value ^= static_cast<size_t>(key.samples) << 52; - for (size_t i = 0; i < key.color_formats.size(); ++i) { - value ^= static_cast<size_t>(key.color_formats[i]) << (i * 6); - } - return value; - } -}; -} // namespace std - -namespace Vulkan { - struct TextureCacheRuntime { const Device& device; VKScheduler& scheduler; @@ -62,13 +38,13 @@ struct TextureCacheRuntime { StagingBufferPool& staging_buffer_pool; BlitImageHelper& blit_image_helper; ASTCDecoderPass& astc_decoder_pass; - std::unordered_map<RenderPassKey, vk::RenderPass> renderpass_cache{}; + RenderPassCache& render_pass_cache; void Finish(); - [[nodiscard]] StagingBufferRef UploadStagingBuffer(size_t size); + StagingBufferRef UploadStagingBuffer(size_t size); - [[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size); + StagingBufferRef DownloadStagingBuffer(size_t size); void BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src, const Region2D& dst_region, const Region2D& src_region, @@ -79,7 +55,7 @@ struct TextureCacheRuntime { void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view); - [[nodiscard]] bool CanAccelerateImageUpload(Image&) const noexcept { + bool CanAccelerateImageUpload(Image&) const noexcept { return false; } @@ -117,8 +93,6 @@ public: void UploadMemory(const StagingBufferRef& map, std::span<const VideoCommon::BufferImageCopy> copies); - void UploadMemory(const StagingBufferRef& map, std::span<const VideoCommon::BufferCopy> copies); - void DownloadMemory(const StagingBufferRef& map, std::span<const VideoCommon::BufferImageCopy> copies); @@ -126,10 +100,6 @@ public: return *image; } - [[nodiscard]] VkBuffer Buffer() const noexcept { - return *buffer; - } - [[nodiscard]] VkImageAspectFlags AspectMask() const noexcept { return aspect_mask; } @@ -146,7 +116,6 @@ public: private: VKScheduler* scheduler; vk::Image image; - vk::Buffer buffer; MemoryCommit commit; vk::ImageView image_view; std::vector<vk::ImageView> storage_image_views; @@ -157,18 +126,19 @@ private: class ImageView : public VideoCommon::ImageViewBase { public: explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageViewInfo&, ImageId, Image&); + explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo&, + const VideoCommon::ImageViewInfo&, GPUVAddr); explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams&); [[nodiscard]] VkImageView DepthView(); [[nodiscard]] VkImageView StencilView(); - [[nodiscard]] VkImageView Handle(VideoCommon::ImageViewType query_type) const noexcept { - return *image_views[static_cast<size_t>(query_type)]; - } + [[nodiscard]] VkImageView StorageView(Shader::TextureType texture_type, + Shader::ImageFormat image_format); - [[nodiscard]] VkBufferView BufferView() const noexcept { - return *buffer_view; + [[nodiscard]] VkImageView Handle(Shader::TextureType texture_type) const noexcept { + return *image_views[static_cast<size_t>(texture_type)]; } [[nodiscard]] VkImage ImageHandle() const noexcept { @@ -179,26 +149,36 @@ public: return render_target; } - [[nodiscard]] PixelFormat ImageFormat() const noexcept { - return image_format; - } - [[nodiscard]] VkSampleCountFlagBits Samples() const noexcept { return samples; } + [[nodiscard]] GPUVAddr GpuAddr() const noexcept { + return gpu_addr; + } + + [[nodiscard]] u32 BufferSize() const noexcept { + return buffer_size; + } + private: - [[nodiscard]] vk::ImageView MakeDepthStencilView(VkImageAspectFlags aspect_mask); + struct StorageViews { + std::array<vk::ImageView, Shader::NUM_TEXTURE_TYPES> signeds; + std::array<vk::ImageView, Shader::NUM_TEXTURE_TYPES> unsigneds; + }; + + [[nodiscard]] vk::ImageView MakeView(VkFormat vk_format, VkImageAspectFlags aspect_mask); const Device* device = nullptr; - std::array<vk::ImageView, VideoCommon::NUM_IMAGE_VIEW_TYPES> image_views; + std::array<vk::ImageView, Shader::NUM_TEXTURE_TYPES> image_views; + std::unique_ptr<StorageViews> storage_views; vk::ImageView depth_view; vk::ImageView stencil_view; - vk::BufferView buffer_view; VkImage image_handle = VK_NULL_HANDLE; VkImageView render_target = VK_NULL_HANDLE; - PixelFormat image_format = PixelFormat::Invalid; VkSampleCountFlagBits samples = VK_SAMPLE_COUNT_1_BIT; + GPUVAddr gpu_addr = 0; + u32 buffer_size = 0; }; class ImageAlloc : public VideoCommon::ImageAllocBase {}; diff --git a/src/video_core/renderer_vulkan/vk_update_descriptor.cpp b/src/video_core/renderer_vulkan/vk_update_descriptor.cpp index dc45fdcb1..0df3a7fe9 100644 --- a/src/video_core/renderer_vulkan/vk_update_descriptor.cpp +++ b/src/video_core/renderer_vulkan/vk_update_descriptor.cpp @@ -15,7 +15,9 @@ namespace Vulkan { VKUpdateDescriptorQueue::VKUpdateDescriptorQueue(const Device& device_, VKScheduler& scheduler_) - : device{device_}, scheduler{scheduler_} {} + : device{device_}, scheduler{scheduler_} { + payload_cursor = payload.data(); +} VKUpdateDescriptorQueue::~VKUpdateDescriptorQueue() = default; @@ -36,13 +38,4 @@ void VKUpdateDescriptorQueue::Acquire() { upload_start = payload_cursor; } -void VKUpdateDescriptorQueue::Send(VkDescriptorUpdateTemplateKHR update_template, - VkDescriptorSet set) { - const void* const data = upload_start; - const vk::Device* const logical = &device.GetLogical(); - scheduler.Record([data, logical, set, update_template](vk::CommandBuffer) { - logical->UpdateDescriptorSet(set, update_template, data); - }); -} - } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_update_descriptor.h b/src/video_core/renderer_vulkan/vk_update_descriptor.h index d35e77c44..d7de4c490 100644 --- a/src/video_core/renderer_vulkan/vk_update_descriptor.h +++ b/src/video_core/renderer_vulkan/vk_update_descriptor.h @@ -39,7 +39,9 @@ public: void Acquire(); - void Send(VkDescriptorUpdateTemplateKHR update_template, VkDescriptorSet set); + const DescriptorUpdateEntry* UpdateData() const noexcept { + return upload_start; + } void AddSampledImage(VkImageView image_view, VkSampler sampler) { *(payload_cursor++) = VkDescriptorImageInfo{ diff --git a/src/video_core/shader/ast.cpp b/src/video_core/shader/ast.cpp deleted file mode 100644 index db11144c7..000000000 --- a/src/video_core/shader/ast.cpp +++ /dev/null @@ -1,752 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <string> -#include <string_view> - -#include <fmt/format.h> - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/shader/ast.h" -#include "video_core/shader/expr.h" - -namespace VideoCommon::Shader { - -ASTZipper::ASTZipper() = default; - -void ASTZipper::Init(const ASTNode new_first, const ASTNode parent) { - ASSERT(new_first->manager == nullptr); - first = new_first; - last = new_first; - - ASTNode current = first; - while (current) { - current->manager = this; - current->parent = parent; - last = current; - current = current->next; - } -} - -void ASTZipper::PushBack(const ASTNode new_node) { - ASSERT(new_node->manager == nullptr); - new_node->previous = last; - if (last) { - last->next = new_node; - } - new_node->next.reset(); - last = new_node; - if (!first) { - first = new_node; - } - new_node->manager = this; -} - -void ASTZipper::PushFront(const ASTNode new_node) { - ASSERT(new_node->manager == nullptr); - new_node->previous.reset(); - new_node->next = first; - if (first) { - first->previous = new_node; - } - if (last == first) { - last = new_node; - } - first = new_node; - new_node->manager = this; -} - -void ASTZipper::InsertAfter(const ASTNode new_node, const ASTNode at_node) { - ASSERT(new_node->manager == nullptr); - if (!at_node) { - PushFront(new_node); - return; - } - const ASTNode next = at_node->next; - if (next) { - next->previous = new_node; - } - new_node->previous = at_node; - if (at_node == last) { - last = new_node; - } - new_node->next = next; - at_node->next = new_node; - new_node->manager = this; -} - -void ASTZipper::InsertBefore(const ASTNode new_node, const ASTNode at_node) { - ASSERT(new_node->manager == nullptr); - if (!at_node) { - PushBack(new_node); - return; - } - const ASTNode previous = at_node->previous; - if (previous) { - previous->next = new_node; - } - new_node->next = at_node; - if (at_node == first) { - first = new_node; - } - new_node->previous = previous; - at_node->previous = new_node; - new_node->manager = this; -} - -void ASTZipper::DetachTail(ASTNode node) { - ASSERT(node->manager == this); - if (node == first) { - first.reset(); - last.reset(); - return; - } - - last = node->previous; - last->next.reset(); - node->previous.reset(); - - ASTNode current = std::move(node); - while (current) { - current->manager = nullptr; - current->parent.reset(); - current = current->next; - } -} - -void ASTZipper::DetachSegment(const ASTNode start, const ASTNode end) { - ASSERT(start->manager == this && end->manager == this); - if (start == end) { - DetachSingle(start); - return; - } - const ASTNode prev = start->previous; - const ASTNode post = end->next; - if (!prev) { - first = post; - } else { - prev->next = post; - } - if (!post) { - last = prev; - } else { - post->previous = prev; - } - start->previous.reset(); - end->next.reset(); - ASTNode current = start; - bool found = false; - while (current) { - current->manager = nullptr; - current->parent.reset(); - found |= current == end; - current = current->next; - } - ASSERT(found); -} - -void ASTZipper::DetachSingle(const ASTNode node) { - ASSERT(node->manager == this); - const ASTNode prev = node->previous; - const ASTNode post = node->next; - node->previous.reset(); - node->next.reset(); - if (!prev) { - first = post; - } else { - prev->next = post; - } - if (!post) { - last = prev; - } else { - post->previous = prev; - } - - node->manager = nullptr; - node->parent.reset(); -} - -void ASTZipper::Remove(const ASTNode node) { - ASSERT(node->manager == this); - const ASTNode next = node->next; - const ASTNode previous = node->previous; - if (previous) { - previous->next = next; - } - if (next) { - next->previous = previous; - } - node->parent.reset(); - node->manager = nullptr; - if (node == last) { - last = previous; - } - if (node == first) { - first = next; - } -} - -class ExprPrinter final { -public: - void operator()(const ExprAnd& expr) { - inner += "( "; - std::visit(*this, *expr.operand1); - inner += " && "; - std::visit(*this, *expr.operand2); - inner += ')'; - } - - void operator()(const ExprOr& expr) { - inner += "( "; - std::visit(*this, *expr.operand1); - inner += " || "; - std::visit(*this, *expr.operand2); - inner += ')'; - } - - void operator()(const ExprNot& expr) { - inner += "!"; - std::visit(*this, *expr.operand1); - } - - void operator()(const ExprPredicate& expr) { - inner += fmt::format("P{}", expr.predicate); - } - - void operator()(const ExprCondCode& expr) { - inner += fmt::format("CC{}", expr.cc); - } - - void operator()(const ExprVar& expr) { - inner += fmt::format("V{}", expr.var_index); - } - - void operator()(const ExprBoolean& expr) { - inner += expr.value ? "true" : "false"; - } - - void operator()(const ExprGprEqual& expr) { - inner += fmt::format("(gpr_{} == {})", expr.gpr, expr.value); - } - - const std::string& GetResult() const { - return inner; - } - -private: - std::string inner; -}; - -class ASTPrinter { -public: - void operator()(const ASTProgram& ast) { - scope++; - inner += "program {\n"; - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - inner += "}\n"; - scope--; - } - - void operator()(const ASTIfThen& ast) { - ExprPrinter expr_parser{}; - std::visit(expr_parser, *ast.condition); - inner += fmt::format("{}if ({}) {{\n", Indent(), expr_parser.GetResult()); - scope++; - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - scope--; - inner += fmt::format("{}}}\n", Indent()); - } - - void operator()(const ASTIfElse& ast) { - inner += Indent(); - inner += "else {\n"; - - scope++; - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - scope--; - - inner += Indent(); - inner += "}\n"; - } - - void operator()(const ASTBlockEncoded& ast) { - inner += fmt::format("{}Block({}, {});\n", Indent(), ast.start, ast.end); - } - - void operator()([[maybe_unused]] const ASTBlockDecoded& ast) { - inner += Indent(); - inner += "Block;\n"; - } - - void operator()(const ASTVarSet& ast) { - ExprPrinter expr_parser{}; - std::visit(expr_parser, *ast.condition); - inner += fmt::format("{}V{} := {};\n", Indent(), ast.index, expr_parser.GetResult()); - } - - void operator()(const ASTLabel& ast) { - inner += fmt::format("Label_{}:\n", ast.index); - } - - void operator()(const ASTGoto& ast) { - ExprPrinter expr_parser{}; - std::visit(expr_parser, *ast.condition); - inner += - fmt::format("{}({}) -> goto Label_{};\n", Indent(), expr_parser.GetResult(), ast.label); - } - - void operator()(const ASTDoWhile& ast) { - ExprPrinter expr_parser{}; - std::visit(expr_parser, *ast.condition); - inner += fmt::format("{}do {{\n", Indent()); - scope++; - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - scope--; - inner += fmt::format("{}}} while ({});\n", Indent(), expr_parser.GetResult()); - } - - void operator()(const ASTReturn& ast) { - ExprPrinter expr_parser{}; - std::visit(expr_parser, *ast.condition); - inner += fmt::format("{}({}) -> {};\n", Indent(), expr_parser.GetResult(), - ast.kills ? "discard" : "exit"); - } - - void operator()(const ASTBreak& ast) { - ExprPrinter expr_parser{}; - std::visit(expr_parser, *ast.condition); - inner += fmt::format("{}({}) -> break;\n", Indent(), expr_parser.GetResult()); - } - - void Visit(const ASTNode& node) { - std::visit(*this, *node->GetInnerData()); - } - - const std::string& GetResult() const { - return inner; - } - -private: - std::string_view Indent() { - if (space_segment_scope == scope) { - return space_segment; - } - - // Ensure that we don't exceed our view. - ASSERT(scope * 2 < spaces.size()); - - space_segment = spaces.substr(0, scope * 2); - space_segment_scope = scope; - return space_segment; - } - - std::string inner{}; - std::string_view space_segment; - - u32 scope{}; - u32 space_segment_scope{}; - - static constexpr std::string_view spaces{" "}; -}; - -std::string ASTManager::Print() const { - ASTPrinter printer{}; - printer.Visit(main_node); - return printer.GetResult(); -} - -ASTManager::ASTManager(bool do_full_decompile, bool disable_else_derivation_) - : full_decompile{do_full_decompile}, disable_else_derivation{disable_else_derivation_} {} - -ASTManager::~ASTManager() { - Clear(); -} - -void ASTManager::Init() { - main_node = ASTBase::Make<ASTProgram>(ASTNode{}); - program = std::get_if<ASTProgram>(main_node->GetInnerData()); - false_condition = MakeExpr<ExprBoolean>(false); -} - -void ASTManager::DeclareLabel(u32 address) { - const auto pair = labels_map.emplace(address, labels_count); - if (pair.second) { - labels_count++; - labels.resize(labels_count); - } -} - -void ASTManager::InsertLabel(u32 address) { - const u32 index = labels_map[address]; - const ASTNode label = ASTBase::Make<ASTLabel>(main_node, index); - labels[index] = label; - program->nodes.PushBack(label); -} - -void ASTManager::InsertGoto(Expr condition, u32 address) { - const u32 index = labels_map[address]; - const ASTNode goto_node = ASTBase::Make<ASTGoto>(main_node, std::move(condition), index); - gotos.push_back(goto_node); - program->nodes.PushBack(goto_node); -} - -void ASTManager::InsertBlock(u32 start_address, u32 end_address) { - ASTNode block = ASTBase::Make<ASTBlockEncoded>(main_node, start_address, end_address); - program->nodes.PushBack(std::move(block)); -} - -void ASTManager::InsertReturn(Expr condition, bool kills) { - ASTNode node = ASTBase::Make<ASTReturn>(main_node, std::move(condition), kills); - program->nodes.PushBack(std::move(node)); -} - -// The decompile algorithm is based on -// "Taming control flow: A structured approach to eliminating goto statements" -// by AM Erosa, LJ Hendren 1994. In general, the idea is to get gotos to be -// on the same structured level as the label which they jump to. This is done, -// through outward/inward movements and lifting. Once they are at the same -// level, you can enclose them in an "if" structure or a "do-while" structure. -void ASTManager::Decompile() { - auto it = gotos.begin(); - while (it != gotos.end()) { - const ASTNode goto_node = *it; - const auto label_index = goto_node->GetGotoLabel(); - if (!label_index) { - return; - } - const ASTNode label = labels[*label_index]; - if (!full_decompile) { - // We only decompile backward jumps - if (!IsBackwardsJump(goto_node, label)) { - it++; - continue; - } - } - if (IndirectlyRelated(goto_node, label)) { - while (!DirectlyRelated(goto_node, label)) { - MoveOutward(goto_node); - } - } - if (DirectlyRelated(goto_node, label)) { - u32 goto_level = goto_node->GetLevel(); - const u32 label_level = label->GetLevel(); - while (label_level < goto_level) { - MoveOutward(goto_node); - goto_level--; - } - // TODO(Blinkhawk): Implement Lifting and Inward Movements - } - if (label->GetParent() == goto_node->GetParent()) { - bool is_loop = false; - ASTNode current = goto_node->GetPrevious(); - while (current) { - if (current == label) { - is_loop = true; - break; - } - current = current->GetPrevious(); - } - - if (is_loop) { - EncloseDoWhile(goto_node, label); - } else { - EncloseIfThen(goto_node, label); - } - it = gotos.erase(it); - continue; - } - it++; - } - if (full_decompile) { - for (const ASTNode& label : labels) { - auto& manager = label->GetManager(); - manager.Remove(label); - } - labels.clear(); - } else { - auto label_it = labels.begin(); - while (label_it != labels.end()) { - bool can_remove = true; - ASTNode label = *label_it; - for (const ASTNode& goto_node : gotos) { - const auto label_index = goto_node->GetGotoLabel(); - if (!label_index) { - return; - } - ASTNode& glabel = labels[*label_index]; - if (glabel == label) { - can_remove = false; - break; - } - } - if (can_remove) { - label->MarkLabelUnused(); - } - } - } -} - -bool ASTManager::IsBackwardsJump(ASTNode goto_node, ASTNode label_node) const { - u32 goto_level = goto_node->GetLevel(); - u32 label_level = label_node->GetLevel(); - while (goto_level > label_level) { - goto_level--; - goto_node = goto_node->GetParent(); - } - while (label_level > goto_level) { - label_level--; - label_node = label_node->GetParent(); - } - while (goto_node->GetParent() != label_node->GetParent()) { - goto_node = goto_node->GetParent(); - label_node = label_node->GetParent(); - } - ASTNode current = goto_node->GetPrevious(); - while (current) { - if (current == label_node) { - return true; - } - current = current->GetPrevious(); - } - return false; -} - -bool ASTManager::IndirectlyRelated(const ASTNode& first, const ASTNode& second) const { - return !(first->GetParent() == second->GetParent() || DirectlyRelated(first, second)); -} - -bool ASTManager::DirectlyRelated(const ASTNode& first, const ASTNode& second) const { - if (first->GetParent() == second->GetParent()) { - return false; - } - const u32 first_level = first->GetLevel(); - const u32 second_level = second->GetLevel(); - u32 min_level; - u32 max_level; - ASTNode max; - ASTNode min; - if (first_level > second_level) { - min_level = second_level; - min = second; - max_level = first_level; - max = first; - } else { - min_level = first_level; - min = first; - max_level = second_level; - max = second; - } - - while (max_level > min_level) { - max_level--; - max = max->GetParent(); - } - - return min->GetParent() == max->GetParent(); -} - -void ASTManager::ShowCurrentState(std::string_view state) const { - LOG_CRITICAL(HW_GPU, "\nState {}:\n\n{}\n", state, Print()); - SanityCheck(); -} - -void ASTManager::SanityCheck() const { - for (const auto& label : labels) { - if (!label->GetParent()) { - LOG_CRITICAL(HW_GPU, "Sanity Check Failed"); - } - } -} - -void ASTManager::EncloseDoWhile(ASTNode goto_node, ASTNode label) { - ASTZipper& zipper = goto_node->GetManager(); - const ASTNode loop_start = label->GetNext(); - if (loop_start == goto_node) { - zipper.Remove(goto_node); - return; - } - const ASTNode parent = label->GetParent(); - const Expr condition = goto_node->GetGotoCondition(); - zipper.DetachSegment(loop_start, goto_node); - const ASTNode do_while_node = ASTBase::Make<ASTDoWhile>(parent, condition); - ASTZipper* sub_zipper = do_while_node->GetSubNodes(); - sub_zipper->Init(loop_start, do_while_node); - zipper.InsertAfter(do_while_node, label); - sub_zipper->Remove(goto_node); -} - -void ASTManager::EncloseIfThen(ASTNode goto_node, ASTNode label) { - ASTZipper& zipper = goto_node->GetManager(); - const ASTNode if_end = label->GetPrevious(); - if (if_end == goto_node) { - zipper.Remove(goto_node); - return; - } - const ASTNode prev = goto_node->GetPrevious(); - const Expr condition = goto_node->GetGotoCondition(); - bool do_else = false; - if (!disable_else_derivation && prev->IsIfThen()) { - const Expr if_condition = prev->GetIfCondition(); - do_else = ExprAreEqual(if_condition, condition); - } - const ASTNode parent = label->GetParent(); - zipper.DetachSegment(goto_node, if_end); - ASTNode if_node; - if (do_else) { - if_node = ASTBase::Make<ASTIfElse>(parent); - } else { - Expr neg_condition = MakeExprNot(condition); - if_node = ASTBase::Make<ASTIfThen>(parent, neg_condition); - } - ASTZipper* sub_zipper = if_node->GetSubNodes(); - sub_zipper->Init(goto_node, if_node); - zipper.InsertAfter(if_node, prev); - sub_zipper->Remove(goto_node); -} - -void ASTManager::MoveOutward(ASTNode goto_node) { - ASTZipper& zipper = goto_node->GetManager(); - const ASTNode parent = goto_node->GetParent(); - ASTZipper& zipper2 = parent->GetManager(); - const ASTNode grandpa = parent->GetParent(); - const bool is_loop = parent->IsLoop(); - const bool is_else = parent->IsIfElse(); - const bool is_if = parent->IsIfThen(); - - const ASTNode prev = goto_node->GetPrevious(); - const ASTNode post = goto_node->GetNext(); - - const Expr condition = goto_node->GetGotoCondition(); - zipper.DetachSingle(goto_node); - if (is_loop) { - const u32 var_index = NewVariable(); - const Expr var_condition = MakeExpr<ExprVar>(var_index); - const ASTNode var_node = ASTBase::Make<ASTVarSet>(parent, var_index, condition); - const ASTNode var_node_init = ASTBase::Make<ASTVarSet>(parent, var_index, false_condition); - zipper2.InsertBefore(var_node_init, parent); - zipper.InsertAfter(var_node, prev); - goto_node->SetGotoCondition(var_condition); - const ASTNode break_node = ASTBase::Make<ASTBreak>(parent, var_condition); - zipper.InsertAfter(break_node, var_node); - } else if (is_if || is_else) { - const u32 var_index = NewVariable(); - const Expr var_condition = MakeExpr<ExprVar>(var_index); - const ASTNode var_node = ASTBase::Make<ASTVarSet>(parent, var_index, condition); - const ASTNode var_node_init = ASTBase::Make<ASTVarSet>(parent, var_index, false_condition); - if (is_if) { - zipper2.InsertBefore(var_node_init, parent); - } else { - zipper2.InsertBefore(var_node_init, parent->GetPrevious()); - } - zipper.InsertAfter(var_node, prev); - goto_node->SetGotoCondition(var_condition); - if (post) { - zipper.DetachTail(post); - const ASTNode if_node = ASTBase::Make<ASTIfThen>(parent, MakeExprNot(var_condition)); - ASTZipper* sub_zipper = if_node->GetSubNodes(); - sub_zipper->Init(post, if_node); - zipper.InsertAfter(if_node, var_node); - } - } else { - UNREACHABLE(); - } - const ASTNode next = parent->GetNext(); - if (is_if && next && next->IsIfElse()) { - zipper2.InsertAfter(goto_node, next); - goto_node->SetParent(grandpa); - return; - } - zipper2.InsertAfter(goto_node, parent); - goto_node->SetParent(grandpa); -} - -class ASTClearer { -public: - ASTClearer() = default; - - void operator()(const ASTProgram& ast) { - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - } - - void operator()(const ASTIfThen& ast) { - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - } - - void operator()(const ASTIfElse& ast) { - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - } - - void operator()([[maybe_unused]] const ASTBlockEncoded& ast) {} - - void operator()(ASTBlockDecoded& ast) { - ast.nodes.clear(); - } - - void operator()([[maybe_unused]] const ASTVarSet& ast) {} - - void operator()([[maybe_unused]] const ASTLabel& ast) {} - - void operator()([[maybe_unused]] const ASTGoto& ast) {} - - void operator()(const ASTDoWhile& ast) { - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - } - - void operator()([[maybe_unused]] const ASTReturn& ast) {} - - void operator()([[maybe_unused]] const ASTBreak& ast) {} - - void Visit(const ASTNode& node) { - std::visit(*this, *node->GetInnerData()); - node->Clear(); - } -}; - -void ASTManager::Clear() { - if (!main_node) { - return; - } - ASTClearer clearer{}; - clearer.Visit(main_node); - main_node.reset(); - program = nullptr; - labels_map.clear(); - labels.clear(); - gotos.clear(); -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/ast.h b/src/video_core/shader/ast.h deleted file mode 100644 index dc49b369e..000000000 --- a/src/video_core/shader/ast.h +++ /dev/null @@ -1,398 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <functional> -#include <list> -#include <memory> -#include <optional> -#include <string> -#include <unordered_map> -#include <vector> - -#include "video_core/shader/expr.h" -#include "video_core/shader/node.h" - -namespace VideoCommon::Shader { - -class ASTBase; -class ASTBlockDecoded; -class ASTBlockEncoded; -class ASTBreak; -class ASTDoWhile; -class ASTGoto; -class ASTIfElse; -class ASTIfThen; -class ASTLabel; -class ASTProgram; -class ASTReturn; -class ASTVarSet; - -using ASTData = std::variant<ASTProgram, ASTIfThen, ASTIfElse, ASTBlockEncoded, ASTBlockDecoded, - ASTVarSet, ASTGoto, ASTLabel, ASTDoWhile, ASTReturn, ASTBreak>; - -using ASTNode = std::shared_ptr<ASTBase>; - -enum class ASTZipperType : u32 { - Program, - IfThen, - IfElse, - Loop, -}; - -class ASTZipper final { -public: - explicit ASTZipper(); - - void Init(ASTNode first, ASTNode parent); - - ASTNode GetFirst() const { - return first; - } - - ASTNode GetLast() const { - return last; - } - - void PushBack(ASTNode new_node); - void PushFront(ASTNode new_node); - void InsertAfter(ASTNode new_node, ASTNode at_node); - void InsertBefore(ASTNode new_node, ASTNode at_node); - void DetachTail(ASTNode node); - void DetachSingle(ASTNode node); - void DetachSegment(ASTNode start, ASTNode end); - void Remove(ASTNode node); - - ASTNode first; - ASTNode last; -}; - -class ASTProgram { -public: - ASTZipper nodes{}; -}; - -class ASTIfThen { -public: - explicit ASTIfThen(Expr condition_) : condition{std::move(condition_)} {} - Expr condition; - ASTZipper nodes{}; -}; - -class ASTIfElse { -public: - ASTZipper nodes{}; -}; - -class ASTBlockEncoded { -public: - explicit ASTBlockEncoded(u32 start_, u32 _) : start{start_}, end{_} {} - u32 start; - u32 end; -}; - -class ASTBlockDecoded { -public: - explicit ASTBlockDecoded(NodeBlock&& new_nodes_) : nodes(std::move(new_nodes_)) {} - NodeBlock nodes; -}; - -class ASTVarSet { -public: - explicit ASTVarSet(u32 index_, Expr condition_) - : index{index_}, condition{std::move(condition_)} {} - - u32 index; - Expr condition; -}; - -class ASTLabel { -public: - explicit ASTLabel(u32 index_) : index{index_} {} - u32 index; - bool unused{}; -}; - -class ASTGoto { -public: - explicit ASTGoto(Expr condition_, u32 label_) - : condition{std::move(condition_)}, label{label_} {} - - Expr condition; - u32 label; -}; - -class ASTDoWhile { -public: - explicit ASTDoWhile(Expr condition_) : condition{std::move(condition_)} {} - Expr condition; - ASTZipper nodes{}; -}; - -class ASTReturn { -public: - explicit ASTReturn(Expr condition_, bool kills_) - : condition{std::move(condition_)}, kills{kills_} {} - - Expr condition; - bool kills; -}; - -class ASTBreak { -public: - explicit ASTBreak(Expr condition_) : condition{std::move(condition_)} {} - Expr condition; -}; - -class ASTBase { -public: - explicit ASTBase(ASTNode parent_, ASTData data_) - : data{std::move(data_)}, parent{std::move(parent_)} {} - - template <class U, class... Args> - static ASTNode Make(ASTNode parent, Args&&... args) { - return std::make_shared<ASTBase>(std::move(parent), - ASTData(U(std::forward<Args>(args)...))); - } - - void SetParent(ASTNode new_parent) { - parent = std::move(new_parent); - } - - ASTNode& GetParent() { - return parent; - } - - const ASTNode& GetParent() const { - return parent; - } - - u32 GetLevel() const { - u32 level = 0; - auto next_parent = parent; - while (next_parent) { - next_parent = next_parent->GetParent(); - level++; - } - return level; - } - - ASTData* GetInnerData() { - return &data; - } - - const ASTData* GetInnerData() const { - return &data; - } - - ASTNode GetNext() const { - return next; - } - - ASTNode GetPrevious() const { - return previous; - } - - ASTZipper& GetManager() { - return *manager; - } - - const ASTZipper& GetManager() const { - return *manager; - } - - std::optional<u32> GetGotoLabel() const { - if (const auto* inner = std::get_if<ASTGoto>(&data)) { - return {inner->label}; - } - return std::nullopt; - } - - Expr GetGotoCondition() const { - if (const auto* inner = std::get_if<ASTGoto>(&data)) { - return inner->condition; - } - return nullptr; - } - - void MarkLabelUnused() { - if (auto* inner = std::get_if<ASTLabel>(&data)) { - inner->unused = true; - } - } - - bool IsLabelUnused() const { - if (const auto* inner = std::get_if<ASTLabel>(&data)) { - return inner->unused; - } - return true; - } - - std::optional<u32> GetLabelIndex() const { - if (const auto* inner = std::get_if<ASTLabel>(&data)) { - return {inner->index}; - } - return std::nullopt; - } - - Expr GetIfCondition() const { - if (const auto* inner = std::get_if<ASTIfThen>(&data)) { - return inner->condition; - } - return nullptr; - } - - void SetGotoCondition(Expr new_condition) { - if (auto* inner = std::get_if<ASTGoto>(&data)) { - inner->condition = std::move(new_condition); - } - } - - bool IsIfThen() const { - return std::holds_alternative<ASTIfThen>(data); - } - - bool IsIfElse() const { - return std::holds_alternative<ASTIfElse>(data); - } - - bool IsBlockEncoded() const { - return std::holds_alternative<ASTBlockEncoded>(data); - } - - void TransformBlockEncoded(NodeBlock&& nodes) { - data = ASTBlockDecoded(std::move(nodes)); - } - - bool IsLoop() const { - return std::holds_alternative<ASTDoWhile>(data); - } - - ASTZipper* GetSubNodes() { - if (std::holds_alternative<ASTProgram>(data)) { - return &std::get_if<ASTProgram>(&data)->nodes; - } - if (std::holds_alternative<ASTIfThen>(data)) { - return &std::get_if<ASTIfThen>(&data)->nodes; - } - if (std::holds_alternative<ASTIfElse>(data)) { - return &std::get_if<ASTIfElse>(&data)->nodes; - } - if (std::holds_alternative<ASTDoWhile>(data)) { - return &std::get_if<ASTDoWhile>(&data)->nodes; - } - return nullptr; - } - - void Clear() { - next.reset(); - previous.reset(); - parent.reset(); - manager = nullptr; - } - -private: - friend class ASTZipper; - - ASTData data; - ASTNode parent; - ASTNode next; - ASTNode previous; - ASTZipper* manager{}; -}; - -class ASTManager final { -public: - explicit ASTManager(bool do_full_decompile, bool disable_else_derivation_); - ~ASTManager(); - - ASTManager(const ASTManager& o) = delete; - ASTManager& operator=(const ASTManager& other) = delete; - - ASTManager(ASTManager&& other) noexcept = default; - ASTManager& operator=(ASTManager&& other) noexcept = default; - - void Init(); - - void DeclareLabel(u32 address); - - void InsertLabel(u32 address); - - void InsertGoto(Expr condition, u32 address); - - void InsertBlock(u32 start_address, u32 end_address); - - void InsertReturn(Expr condition, bool kills); - - std::string Print() const; - - void Decompile(); - - void ShowCurrentState(std::string_view state) const; - - void SanityCheck() const; - - void Clear(); - - bool IsFullyDecompiled() const { - if (full_decompile) { - return gotos.empty(); - } - - for (ASTNode goto_node : gotos) { - auto label_index = goto_node->GetGotoLabel(); - if (!label_index) { - return false; - } - ASTNode glabel = labels[*label_index]; - if (IsBackwardsJump(goto_node, glabel)) { - return false; - } - } - return true; - } - - ASTNode GetProgram() const { - return main_node; - } - - u32 GetVariables() const { - return variables; - } - - const std::vector<ASTNode>& GetLabels() const { - return labels; - } - -private: - bool IsBackwardsJump(ASTNode goto_node, ASTNode label_node) const; - - bool IndirectlyRelated(const ASTNode& first, const ASTNode& second) const; - - bool DirectlyRelated(const ASTNode& first, const ASTNode& second) const; - - void EncloseDoWhile(ASTNode goto_node, ASTNode label); - - void EncloseIfThen(ASTNode goto_node, ASTNode label); - - void MoveOutward(ASTNode goto_node); - - u32 NewVariable() { - return variables++; - } - - bool full_decompile{}; - bool disable_else_derivation{}; - std::unordered_map<u32, u32> labels_map{}; - u32 labels_count{}; - std::vector<ASTNode> labels{}; - std::list<ASTNode> gotos{}; - u32 variables{}; - ASTProgram* program{}; - ASTNode main_node{}; - Expr false_condition{}; -}; - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/async_shaders.cpp b/src/video_core/shader/async_shaders.cpp deleted file mode 100644 index 02adcf9c7..000000000 --- a/src/video_core/shader/async_shaders.cpp +++ /dev/null @@ -1,234 +0,0 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <condition_variable> -#include <mutex> -#include <thread> -#include <vector> -#include "video_core/engines/maxwell_3d.h" -#include "video_core/renderer_base.h" -#include "video_core/renderer_opengl/gl_shader_cache.h" -#include "video_core/shader/async_shaders.h" - -namespace VideoCommon::Shader { - -AsyncShaders::AsyncShaders(Core::Frontend::EmuWindow& emu_window_) : emu_window(emu_window_) {} - -AsyncShaders::~AsyncShaders() { - KillWorkers(); -} - -void AsyncShaders::AllocateWorkers() { - // Use at least one thread - u32 num_workers = 1; - - // Deduce how many more threads we can use - const u32 thread_count = std::thread::hardware_concurrency(); - if (thread_count >= 8) { - // Increase async workers by 1 for every 2 threads >= 8 - num_workers += 1 + (thread_count - 8) / 2; - } - - // If we already have workers queued, ignore - if (num_workers == worker_threads.size()) { - return; - } - - // If workers already exist, clear them - if (!worker_threads.empty()) { - FreeWorkers(); - } - - // Create workers - for (std::size_t i = 0; i < num_workers; i++) { - context_list.push_back(emu_window.CreateSharedContext()); - worker_threads.emplace_back(&AsyncShaders::ShaderCompilerThread, this, - context_list[i].get()); - } -} - -void AsyncShaders::FreeWorkers() { - // Mark all threads to quit - is_thread_exiting.store(true); - cv.notify_all(); - for (auto& thread : worker_threads) { - thread.join(); - } - // Clear our shared contexts - context_list.clear(); - - // Clear our worker threads - worker_threads.clear(); -} - -void AsyncShaders::KillWorkers() { - is_thread_exiting.store(true); - cv.notify_all(); - for (auto& thread : worker_threads) { - thread.detach(); - } - // Clear our shared contexts - context_list.clear(); - - // Clear our worker threads - worker_threads.clear(); -} - -bool AsyncShaders::HasWorkQueued() const { - return !pending_queue.empty(); -} - -bool AsyncShaders::HasCompletedWork() const { - std::shared_lock lock{completed_mutex}; - return !finished_work.empty(); -} - -bool AsyncShaders::IsShaderAsync(const Tegra::GPU& gpu) const { - const auto& regs = gpu.Maxwell3D().regs; - - // If something is using depth, we can assume that games are not rendering anything which will - // be used one time. - if (regs.zeta_enable) { - return true; - } - - // If games are using a small index count, we can assume these are full screen quads. Usually - // these shaders are only used once for building textures so we can assume they can't be built - // async - if (regs.index_array.count <= 6 || regs.vertex_buffer.count <= 6) { - return false; - } - - return true; -} - -std::vector<AsyncShaders::Result> AsyncShaders::GetCompletedWork() { - std::vector<Result> results; - { - std::unique_lock lock{completed_mutex}; - results = std::move(finished_work); - finished_work.clear(); - } - return results; -} - -void AsyncShaders::QueueOpenGLShader(const OpenGL::Device& device, - Tegra::Engines::ShaderType shader_type, u64 uid, - std::vector<u64> code, std::vector<u64> code_b, - u32 main_offset, CompilerSettings compiler_settings, - const Registry& registry, VAddr cpu_addr) { - std::unique_lock lock(queue_mutex); - pending_queue.push({ - .backend = device.UseAssemblyShaders() ? Backend::GLASM : Backend::OpenGL, - .device = &device, - .shader_type = shader_type, - .uid = uid, - .code = std::move(code), - .code_b = std::move(code_b), - .main_offset = main_offset, - .compiler_settings = compiler_settings, - .registry = registry, - .cpu_address = cpu_addr, - .pp_cache = nullptr, - .vk_device = nullptr, - .scheduler = nullptr, - .descriptor_pool = nullptr, - .update_descriptor_queue = nullptr, - .bindings{}, - .program{}, - .key{}, - .num_color_buffers = 0, - }); - cv.notify_one(); -} - -void AsyncShaders::QueueVulkanShader(Vulkan::VKPipelineCache* pp_cache, - const Vulkan::Device& device, Vulkan::VKScheduler& scheduler, - Vulkan::VKDescriptorPool& descriptor_pool, - Vulkan::VKUpdateDescriptorQueue& update_descriptor_queue, - std::vector<VkDescriptorSetLayoutBinding> bindings, - Vulkan::SPIRVProgram program, - Vulkan::GraphicsPipelineCacheKey key, u32 num_color_buffers) { - std::unique_lock lock(queue_mutex); - pending_queue.push({ - .backend = Backend::Vulkan, - .device = nullptr, - .shader_type{}, - .uid = 0, - .code{}, - .code_b{}, - .main_offset = 0, - .compiler_settings{}, - .registry{}, - .cpu_address = 0, - .pp_cache = pp_cache, - .vk_device = &device, - .scheduler = &scheduler, - .descriptor_pool = &descriptor_pool, - .update_descriptor_queue = &update_descriptor_queue, - .bindings = std::move(bindings), - .program = std::move(program), - .key = key, - .num_color_buffers = num_color_buffers, - }); - cv.notify_one(); -} - -void AsyncShaders::ShaderCompilerThread(Core::Frontend::GraphicsContext* context) { - while (!is_thread_exiting.load(std::memory_order_relaxed)) { - std::unique_lock lock{queue_mutex}; - cv.wait(lock, [this] { return HasWorkQueued() || is_thread_exiting; }); - if (is_thread_exiting) { - return; - } - - // Partial lock to allow all threads to read at the same time - if (!HasWorkQueued()) { - continue; - } - // Another thread beat us, just unlock and wait for the next load - if (pending_queue.empty()) { - continue; - } - - // Pull work from queue - WorkerParams work = std::move(pending_queue.front()); - pending_queue.pop(); - lock.unlock(); - - if (work.backend == Backend::OpenGL || work.backend == Backend::GLASM) { - const ShaderIR ir(work.code, work.main_offset, work.compiler_settings, *work.registry); - const auto scope = context->Acquire(); - auto program = - OpenGL::BuildShader(*work.device, work.shader_type, work.uid, ir, *work.registry); - Result result{}; - result.backend = work.backend; - result.cpu_address = work.cpu_address; - result.uid = work.uid; - result.code = std::move(work.code); - result.code_b = std::move(work.code_b); - result.shader_type = work.shader_type; - - if (work.backend == Backend::OpenGL) { - result.program.opengl = std::move(program->source_program); - } else if (work.backend == Backend::GLASM) { - result.program.glasm = std::move(program->assembly_program); - } - - { - std::unique_lock complete_lock(completed_mutex); - finished_work.push_back(std::move(result)); - } - } else if (work.backend == Backend::Vulkan) { - auto pipeline = std::make_unique<Vulkan::VKGraphicsPipeline>( - *work.vk_device, *work.scheduler, *work.descriptor_pool, - *work.update_descriptor_queue, work.key, work.bindings, work.program, - work.num_color_buffers); - - work.pp_cache->EmplacePipeline(std::move(pipeline)); - } - } -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/async_shaders.h b/src/video_core/shader/async_shaders.h deleted file mode 100644 index 7fdff6e56..000000000 --- a/src/video_core/shader/async_shaders.h +++ /dev/null @@ -1,138 +0,0 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <condition_variable> -#include <memory> -#include <shared_mutex> -#include <thread> - -#include <glad/glad.h> - -#include "common/common_types.h" -#include "video_core/renderer_opengl/gl_device.h" -#include "video_core/renderer_opengl/gl_resource_manager.h" -#include "video_core/renderer_opengl/gl_shader_decompiler.h" -#include "video_core/renderer_vulkan/vk_pipeline_cache.h" -#include "video_core/renderer_vulkan/vk_scheduler.h" -#include "video_core/vulkan_common/vulkan_device.h" - -namespace Core::Frontend { -class EmuWindow; -class GraphicsContext; -} // namespace Core::Frontend - -namespace Tegra { -class GPU; -} - -namespace Vulkan { -class VKPipelineCache; -} - -namespace VideoCommon::Shader { - -class AsyncShaders { -public: - enum class Backend { - OpenGL, - GLASM, - Vulkan, - }; - - struct ResultPrograms { - OpenGL::OGLProgram opengl; - OpenGL::OGLAssemblyProgram glasm; - }; - - struct Result { - u64 uid; - VAddr cpu_address; - Backend backend; - ResultPrograms program; - std::vector<u64> code; - std::vector<u64> code_b; - Tegra::Engines::ShaderType shader_type; - }; - - explicit AsyncShaders(Core::Frontend::EmuWindow& emu_window_); - ~AsyncShaders(); - - /// Start up shader worker threads - void AllocateWorkers(); - - /// Clear the shader queue and kill all worker threads - void FreeWorkers(); - - // Force end all threads - void KillWorkers(); - - /// Check to see if any shaders have actually been compiled - [[nodiscard]] bool HasCompletedWork() const; - - /// Deduce if a shader can be build on another thread of MUST be built in sync. We cannot build - /// every shader async as some shaders are only built and executed once. We try to "guess" which - /// shader would be used only once - [[nodiscard]] bool IsShaderAsync(const Tegra::GPU& gpu) const; - - /// Pulls completed compiled shaders - [[nodiscard]] std::vector<Result> GetCompletedWork(); - - void QueueOpenGLShader(const OpenGL::Device& device, Tegra::Engines::ShaderType shader_type, - u64 uid, std::vector<u64> code, std::vector<u64> code_b, u32 main_offset, - CompilerSettings compiler_settings, const Registry& registry, - VAddr cpu_addr); - - void QueueVulkanShader(Vulkan::VKPipelineCache* pp_cache, const Vulkan::Device& device, - Vulkan::VKScheduler& scheduler, - Vulkan::VKDescriptorPool& descriptor_pool, - Vulkan::VKUpdateDescriptorQueue& update_descriptor_queue, - std::vector<VkDescriptorSetLayoutBinding> bindings, - Vulkan::SPIRVProgram program, Vulkan::GraphicsPipelineCacheKey key, - u32 num_color_buffers); - -private: - void ShaderCompilerThread(Core::Frontend::GraphicsContext* context); - - /// Check our worker queue to see if we have any work queued already - [[nodiscard]] bool HasWorkQueued() const; - - struct WorkerParams { - Backend backend; - // For OGL - const OpenGL::Device* device; - Tegra::Engines::ShaderType shader_type; - u64 uid; - std::vector<u64> code; - std::vector<u64> code_b; - u32 main_offset; - CompilerSettings compiler_settings; - std::optional<Registry> registry; - VAddr cpu_address; - - // For Vulkan - Vulkan::VKPipelineCache* pp_cache; - const Vulkan::Device* vk_device; - Vulkan::VKScheduler* scheduler; - Vulkan::VKDescriptorPool* descriptor_pool; - Vulkan::VKUpdateDescriptorQueue* update_descriptor_queue; - std::vector<VkDescriptorSetLayoutBinding> bindings; - Vulkan::SPIRVProgram program; - Vulkan::GraphicsPipelineCacheKey key; - u32 num_color_buffers; - }; - - std::condition_variable cv; - mutable std::mutex queue_mutex; - mutable std::shared_mutex completed_mutex; - std::atomic<bool> is_thread_exiting{}; - std::vector<std::unique_ptr<Core::Frontend::GraphicsContext>> context_list; - std::vector<std::thread> worker_threads; - std::queue<WorkerParams> pending_queue; - std::vector<Result> finished_work; - Core::Frontend::EmuWindow& emu_window; -}; - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/compiler_settings.cpp b/src/video_core/shader/compiler_settings.cpp deleted file mode 100644 index cddcbd4f0..000000000 --- a/src/video_core/shader/compiler_settings.cpp +++ /dev/null @@ -1,26 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "video_core/shader/compiler_settings.h" - -namespace VideoCommon::Shader { - -std::string CompileDepthAsString(const CompileDepth cd) { - switch (cd) { - case CompileDepth::BruteForce: - return "Brute Force Compile"; - case CompileDepth::FlowStack: - return "Simple Flow Stack Mode"; - case CompileDepth::NoFlowStack: - return "Remove Flow Stack"; - case CompileDepth::DecompileBackwards: - return "Decompile Backward Jumps"; - case CompileDepth::FullDecompile: - return "Full Decompilation"; - default: - return "Unknown Compiler Process"; - } -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/compiler_settings.h b/src/video_core/shader/compiler_settings.h deleted file mode 100644 index 916018c01..000000000 --- a/src/video_core/shader/compiler_settings.h +++ /dev/null @@ -1,26 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include "video_core/engines/shader_bytecode.h" - -namespace VideoCommon::Shader { - -enum class CompileDepth : u32 { - BruteForce = 0, - FlowStack = 1, - NoFlowStack = 2, - DecompileBackwards = 3, - FullDecompile = 4, -}; - -std::string CompileDepthAsString(CompileDepth cd); - -struct CompilerSettings { - CompileDepth depth{CompileDepth::NoFlowStack}; - bool disable_else_derivation{true}; -}; - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/control_flow.cpp b/src/video_core/shader/control_flow.cpp deleted file mode 100644 index 43d965f2f..000000000 --- a/src/video_core/shader/control_flow.cpp +++ /dev/null @@ -1,751 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <list> -#include <map> -#include <set> -#include <stack> -#include <unordered_map> -#include <vector> - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/shader/ast.h" -#include "video_core/shader/control_flow.h" -#include "video_core/shader/memory_util.h" -#include "video_core/shader/registry.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -namespace { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -constexpr s32 unassigned_branch = -2; - -struct Query { - u32 address{}; - std::stack<u32> ssy_stack{}; - std::stack<u32> pbk_stack{}; -}; - -struct BlockStack { - BlockStack() = default; - explicit BlockStack(const Query& q) : ssy_stack{q.ssy_stack}, pbk_stack{q.pbk_stack} {} - std::stack<u32> ssy_stack{}; - std::stack<u32> pbk_stack{}; -}; - -template <typename T, typename... Args> -BlockBranchInfo MakeBranchInfo(Args&&... args) { - static_assert(std::is_convertible_v<T, BranchData>); - return std::make_shared<BranchData>(T(std::forward<Args>(args)...)); -} - -bool BlockBranchIsIgnored(BlockBranchInfo first) { - bool ignore = false; - if (std::holds_alternative<SingleBranch>(*first)) { - const auto branch = std::get_if<SingleBranch>(first.get()); - ignore = branch->ignore; - } - return ignore; -} - -struct BlockInfo { - u32 start{}; - u32 end{}; - bool visited{}; - BlockBranchInfo branch{}; - - bool IsInside(const u32 address) const { - return start <= address && address <= end; - } -}; - -struct CFGRebuildState { - explicit CFGRebuildState(const ProgramCode& program_code_, u32 start_, Registry& registry_) - : program_code{program_code_}, registry{registry_}, start{start_} {} - - const ProgramCode& program_code; - Registry& registry; - u32 start{}; - std::vector<BlockInfo> block_info; - std::list<u32> inspect_queries; - std::list<Query> queries; - std::unordered_map<u32, u32> registered; - std::set<u32> labels; - std::map<u32, u32> ssy_labels; - std::map<u32, u32> pbk_labels; - std::unordered_map<u32, BlockStack> stacks; - ASTManager* manager{}; -}; - -enum class BlockCollision : u32 { None, Found, Inside }; - -std::pair<BlockCollision, u32> TryGetBlock(CFGRebuildState& state, u32 address) { - const auto& blocks = state.block_info; - for (u32 index = 0; index < blocks.size(); index++) { - if (blocks[index].start == address) { - return {BlockCollision::Found, index}; - } - if (blocks[index].IsInside(address)) { - return {BlockCollision::Inside, index}; - } - } - return {BlockCollision::None, 0xFFFFFFFF}; -} - -struct ParseInfo { - BlockBranchInfo branch_info{}; - u32 end_address{}; -}; - -BlockInfo& CreateBlockInfo(CFGRebuildState& state, u32 start, u32 end) { - auto& it = state.block_info.emplace_back(); - it.start = start; - it.end = end; - const u32 index = static_cast<u32>(state.block_info.size() - 1); - state.registered.insert({start, index}); - return it; -} - -Pred GetPredicate(u32 index, bool negated) { - return static_cast<Pred>(static_cast<u64>(index) + (negated ? 8ULL : 0ULL)); -} - -enum class ParseResult : u32 { - ControlCaught, - BlockEnd, - AbnormalFlow, -}; - -struct BranchIndirectInfo { - u32 buffer{}; - u32 offset{}; - u32 entries{}; - s32 relative_position{}; -}; - -struct BufferInfo { - u32 index; - u32 offset; -}; - -std::optional<std::pair<s32, u64>> GetBRXInfo(const CFGRebuildState& state, u32& pos) { - const Instruction instr = state.program_code[pos]; - const auto opcode = OpCode::Decode(instr); - if (opcode->get().GetId() != OpCode::Id::BRX) { - return std::nullopt; - } - if (instr.brx.constant_buffer != 0) { - return std::nullopt; - } - --pos; - return std::make_pair(instr.brx.GetBranchExtend(), instr.gpr8.Value()); -} - -template <typename Result, typename TestCallable, typename PackCallable> -// requires std::predicate<TestCallable, Instruction, const OpCode::Matcher&> -// requires std::invocable<PackCallable, Instruction, const OpCode::Matcher&> -std::optional<Result> TrackInstruction(const CFGRebuildState& state, u32& pos, TestCallable test, - PackCallable pack) { - for (; pos >= state.start; --pos) { - if (IsSchedInstruction(pos, state.start)) { - continue; - } - const Instruction instr = state.program_code[pos]; - const auto opcode = OpCode::Decode(instr); - if (!opcode) { - continue; - } - if (test(instr, opcode->get())) { - --pos; - return std::make_optional(pack(instr, opcode->get())); - } - } - return std::nullopt; -} - -std::optional<std::pair<BufferInfo, u64>> TrackLDC(const CFGRebuildState& state, u32& pos, - u64 brx_tracked_register) { - return TrackInstruction<std::pair<BufferInfo, u64>>( - state, pos, - [brx_tracked_register](auto instr, const auto& opcode) { - return opcode.GetId() == OpCode::Id::LD_C && - instr.gpr0.Value() == brx_tracked_register && - instr.ld_c.type.Value() == Tegra::Shader::UniformType::Single; - }, - [](auto instr, const auto& opcode) { - const BufferInfo info = {static_cast<u32>(instr.cbuf36.index.Value()), - static_cast<u32>(instr.cbuf36.GetOffset())}; - return std::make_pair(info, instr.gpr8.Value()); - }); -} - -std::optional<u64> TrackSHLRegister(const CFGRebuildState& state, u32& pos, - u64 ldc_tracked_register) { - return TrackInstruction<u64>( - state, pos, - [ldc_tracked_register](auto instr, const auto& opcode) { - return opcode.GetId() == OpCode::Id::SHL_IMM && - instr.gpr0.Value() == ldc_tracked_register; - }, - [](auto instr, const auto&) { return instr.gpr8.Value(); }); -} - -std::optional<u32> TrackIMNMXValue(const CFGRebuildState& state, u32& pos, - u64 shl_tracked_register) { - return TrackInstruction<u32>( - state, pos, - [shl_tracked_register](auto instr, const auto& opcode) { - return opcode.GetId() == OpCode::Id::IMNMX_IMM && - instr.gpr0.Value() == shl_tracked_register; - }, - [](auto instr, const auto&) { - return static_cast<u32>(instr.alu.GetSignedImm20_20() + 1); - }); -} - -std::optional<BranchIndirectInfo> TrackBranchIndirectInfo(const CFGRebuildState& state, u32 pos) { - const auto brx_info = GetBRXInfo(state, pos); - if (!brx_info) { - return std::nullopt; - } - const auto [relative_position, brx_tracked_register] = *brx_info; - - const auto ldc_info = TrackLDC(state, pos, brx_tracked_register); - if (!ldc_info) { - return std::nullopt; - } - const auto [buffer_info, ldc_tracked_register] = *ldc_info; - - const auto shl_tracked_register = TrackSHLRegister(state, pos, ldc_tracked_register); - if (!shl_tracked_register) { - return std::nullopt; - } - - const auto entries = TrackIMNMXValue(state, pos, *shl_tracked_register); - if (!entries) { - return std::nullopt; - } - - return BranchIndirectInfo{buffer_info.index, buffer_info.offset, *entries, relative_position}; -} - -std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address) { - u32 offset = static_cast<u32>(address); - const u32 end_address = static_cast<u32>(state.program_code.size()); - ParseInfo parse_info{}; - SingleBranch single_branch{}; - - const auto insert_label = [](CFGRebuildState& rebuild_state, u32 label_address) { - const auto pair = rebuild_state.labels.emplace(label_address); - if (pair.second) { - rebuild_state.inspect_queries.push_back(label_address); - } - }; - - while (true) { - if (offset >= end_address) { - // ASSERT_OR_EXECUTE can't be used, as it ignores the break - ASSERT_MSG(false, "Shader passed the current limit!"); - - single_branch.address = exit_branch; - single_branch.ignore = false; - break; - } - if (state.registered.contains(offset)) { - single_branch.address = offset; - single_branch.ignore = true; - break; - } - if (IsSchedInstruction(offset, state.start)) { - offset++; - continue; - } - const Instruction instr = {state.program_code[offset]}; - const auto opcode = OpCode::Decode(instr); - if (!opcode || opcode->get().GetType() != OpCode::Type::Flow) { - offset++; - continue; - } - - switch (opcode->get().GetId()) { - case OpCode::Id::EXIT: { - const auto pred_index = static_cast<u32>(instr.pred.pred_index); - single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0); - if (single_branch.condition.predicate == Pred::NeverExecute) { - offset++; - continue; - } - const ConditionCode cc = instr.flow_condition_code; - single_branch.condition.cc = cc; - if (cc == ConditionCode::F) { - offset++; - continue; - } - single_branch.address = exit_branch; - single_branch.kill = false; - single_branch.is_sync = false; - single_branch.is_brk = false; - single_branch.ignore = false; - parse_info.end_address = offset; - parse_info.branch_info = MakeBranchInfo<SingleBranch>( - single_branch.condition, single_branch.address, single_branch.kill, - single_branch.is_sync, single_branch.is_brk, single_branch.ignore); - - return {ParseResult::ControlCaught, parse_info}; - } - case OpCode::Id::BRA: { - if (instr.bra.constant_buffer != 0) { - return {ParseResult::AbnormalFlow, parse_info}; - } - const auto pred_index = static_cast<u32>(instr.pred.pred_index); - single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0); - if (single_branch.condition.predicate == Pred::NeverExecute) { - offset++; - continue; - } - const ConditionCode cc = instr.flow_condition_code; - single_branch.condition.cc = cc; - if (cc == ConditionCode::F) { - offset++; - continue; - } - const u32 branch_offset = offset + instr.bra.GetBranchTarget(); - if (branch_offset == 0) { - single_branch.address = exit_branch; - } else { - single_branch.address = branch_offset; - } - insert_label(state, branch_offset); - single_branch.kill = false; - single_branch.is_sync = false; - single_branch.is_brk = false; - single_branch.ignore = false; - parse_info.end_address = offset; - parse_info.branch_info = MakeBranchInfo<SingleBranch>( - single_branch.condition, single_branch.address, single_branch.kill, - single_branch.is_sync, single_branch.is_brk, single_branch.ignore); - - return {ParseResult::ControlCaught, parse_info}; - } - case OpCode::Id::SYNC: { - const auto pred_index = static_cast<u32>(instr.pred.pred_index); - single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0); - if (single_branch.condition.predicate == Pred::NeverExecute) { - offset++; - continue; - } - const ConditionCode cc = instr.flow_condition_code; - single_branch.condition.cc = cc; - if (cc == ConditionCode::F) { - offset++; - continue; - } - single_branch.address = unassigned_branch; - single_branch.kill = false; - single_branch.is_sync = true; - single_branch.is_brk = false; - single_branch.ignore = false; - parse_info.end_address = offset; - parse_info.branch_info = MakeBranchInfo<SingleBranch>( - single_branch.condition, single_branch.address, single_branch.kill, - single_branch.is_sync, single_branch.is_brk, single_branch.ignore); - - return {ParseResult::ControlCaught, parse_info}; - } - case OpCode::Id::BRK: { - const auto pred_index = static_cast<u32>(instr.pred.pred_index); - single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0); - if (single_branch.condition.predicate == Pred::NeverExecute) { - offset++; - continue; - } - const ConditionCode cc = instr.flow_condition_code; - single_branch.condition.cc = cc; - if (cc == ConditionCode::F) { - offset++; - continue; - } - single_branch.address = unassigned_branch; - single_branch.kill = false; - single_branch.is_sync = false; - single_branch.is_brk = true; - single_branch.ignore = false; - parse_info.end_address = offset; - parse_info.branch_info = MakeBranchInfo<SingleBranch>( - single_branch.condition, single_branch.address, single_branch.kill, - single_branch.is_sync, single_branch.is_brk, single_branch.ignore); - - return {ParseResult::ControlCaught, parse_info}; - } - case OpCode::Id::KIL: { - const auto pred_index = static_cast<u32>(instr.pred.pred_index); - single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0); - if (single_branch.condition.predicate == Pred::NeverExecute) { - offset++; - continue; - } - const ConditionCode cc = instr.flow_condition_code; - single_branch.condition.cc = cc; - if (cc == ConditionCode::F) { - offset++; - continue; - } - single_branch.address = exit_branch; - single_branch.kill = true; - single_branch.is_sync = false; - single_branch.is_brk = false; - single_branch.ignore = false; - parse_info.end_address = offset; - parse_info.branch_info = MakeBranchInfo<SingleBranch>( - single_branch.condition, single_branch.address, single_branch.kill, - single_branch.is_sync, single_branch.is_brk, single_branch.ignore); - - return {ParseResult::ControlCaught, parse_info}; - } - case OpCode::Id::SSY: { - const u32 target = offset + instr.bra.GetBranchTarget(); - insert_label(state, target); - state.ssy_labels.emplace(offset, target); - break; - } - case OpCode::Id::PBK: { - const u32 target = offset + instr.bra.GetBranchTarget(); - insert_label(state, target); - state.pbk_labels.emplace(offset, target); - break; - } - case OpCode::Id::BRX: { - const auto tmp = TrackBranchIndirectInfo(state, offset); - if (!tmp) { - LOG_WARNING(HW_GPU, "BRX Track Unsuccesful"); - return {ParseResult::AbnormalFlow, parse_info}; - } - - const auto result = *tmp; - const s32 pc_target = offset + result.relative_position; - std::vector<CaseBranch> branches; - for (u32 i = 0; i < result.entries; i++) { - auto key = state.registry.ObtainKey(result.buffer, result.offset + i * 4); - if (!key) { - return {ParseResult::AbnormalFlow, parse_info}; - } - u32 value = *key; - u32 target = static_cast<u32>((value >> 3) + pc_target); - insert_label(state, target); - branches.emplace_back(value, target); - } - parse_info.end_address = offset; - parse_info.branch_info = MakeBranchInfo<MultiBranch>( - static_cast<u32>(instr.gpr8.Value()), std::move(branches)); - - return {ParseResult::ControlCaught, parse_info}; - } - default: - break; - } - - offset++; - } - single_branch.kill = false; - single_branch.is_sync = false; - single_branch.is_brk = false; - parse_info.end_address = offset - 1; - parse_info.branch_info = MakeBranchInfo<SingleBranch>( - single_branch.condition, single_branch.address, single_branch.kill, single_branch.is_sync, - single_branch.is_brk, single_branch.ignore); - return {ParseResult::BlockEnd, parse_info}; -} - -bool TryInspectAddress(CFGRebuildState& state) { - if (state.inspect_queries.empty()) { - return false; - } - - const u32 address = state.inspect_queries.front(); - state.inspect_queries.pop_front(); - const auto [result, block_index] = TryGetBlock(state, address); - switch (result) { - case BlockCollision::Found: { - return true; - } - case BlockCollision::Inside: { - // This case is the tricky one: - // We need to split the block into 2 separate blocks - const u32 end = state.block_info[block_index].end; - BlockInfo& new_block = CreateBlockInfo(state, address, end); - BlockInfo& current_block = state.block_info[block_index]; - current_block.end = address - 1; - new_block.branch = std::move(current_block.branch); - BlockBranchInfo forward_branch = MakeBranchInfo<SingleBranch>(); - const auto branch = std::get_if<SingleBranch>(forward_branch.get()); - branch->address = address; - branch->ignore = true; - current_block.branch = std::move(forward_branch); - return true; - } - default: - break; - } - const auto [parse_result, parse_info] = ParseCode(state, address); - if (parse_result == ParseResult::AbnormalFlow) { - // if it's AbnormalFlow, we end it as false, ending the CFG reconstruction - return false; - } - - BlockInfo& block_info = CreateBlockInfo(state, address, parse_info.end_address); - block_info.branch = parse_info.branch_info; - if (std::holds_alternative<SingleBranch>(*block_info.branch)) { - const auto branch = std::get_if<SingleBranch>(block_info.branch.get()); - if (branch->condition.IsUnconditional()) { - return true; - } - const u32 fallthrough_address = parse_info.end_address + 1; - state.inspect_queries.push_front(fallthrough_address); - return true; - } - return true; -} - -bool TryQuery(CFGRebuildState& state) { - const auto gather_labels = [](std::stack<u32>& cc, std::map<u32, u32>& labels, - BlockInfo& block) { - auto gather_start = labels.lower_bound(block.start); - const auto gather_end = labels.upper_bound(block.end); - while (gather_start != gather_end) { - cc.push(gather_start->second); - ++gather_start; - } - }; - if (state.queries.empty()) { - return false; - } - - Query& q = state.queries.front(); - const u32 block_index = state.registered[q.address]; - BlockInfo& block = state.block_info[block_index]; - // If the block is visited, check if the stacks match, else gather the ssy/pbk - // labels into the current stack and look if the branch at the end of the block - // consumes a label. Schedule new queries accordingly - if (block.visited) { - BlockStack& stack = state.stacks[q.address]; - const bool all_okay = (stack.ssy_stack.empty() || q.ssy_stack == stack.ssy_stack) && - (stack.pbk_stack.empty() || q.pbk_stack == stack.pbk_stack); - state.queries.pop_front(); - return all_okay; - } - block.visited = true; - state.stacks.insert_or_assign(q.address, BlockStack{q}); - - Query q2(q); - state.queries.pop_front(); - gather_labels(q2.ssy_stack, state.ssy_labels, block); - gather_labels(q2.pbk_stack, state.pbk_labels, block); - if (std::holds_alternative<SingleBranch>(*block.branch)) { - auto* branch = std::get_if<SingleBranch>(block.branch.get()); - if (!branch->condition.IsUnconditional()) { - q2.address = block.end + 1; - state.queries.push_back(q2); - } - - auto& conditional_query = state.queries.emplace_back(q2); - if (branch->is_sync) { - if (branch->address == unassigned_branch) { - branch->address = conditional_query.ssy_stack.top(); - } - conditional_query.ssy_stack.pop(); - } - if (branch->is_brk) { - if (branch->address == unassigned_branch) { - branch->address = conditional_query.pbk_stack.top(); - } - conditional_query.pbk_stack.pop(); - } - conditional_query.address = branch->address; - return true; - } - - const auto* multi_branch = std::get_if<MultiBranch>(block.branch.get()); - for (const auto& branch_case : multi_branch->branches) { - auto& conditional_query = state.queries.emplace_back(q2); - conditional_query.address = branch_case.address; - } - - return true; -} - -void InsertBranch(ASTManager& mm, const BlockBranchInfo& branch_info) { - const auto get_expr = [](const Condition& cond) -> Expr { - Expr result; - if (cond.cc != ConditionCode::T) { - result = MakeExpr<ExprCondCode>(cond.cc); - } - if (cond.predicate != Pred::UnusedIndex) { - u32 pred = static_cast<u32>(cond.predicate); - bool negate = false; - if (pred > 7) { - negate = true; - pred -= 8; - } - Expr extra = MakeExpr<ExprPredicate>(pred); - if (negate) { - extra = MakeExpr<ExprNot>(std::move(extra)); - } - if (result) { - return MakeExpr<ExprAnd>(std::move(extra), std::move(result)); - } - return extra; - } - if (result) { - return result; - } - return MakeExpr<ExprBoolean>(true); - }; - - if (std::holds_alternative<SingleBranch>(*branch_info)) { - const auto* branch = std::get_if<SingleBranch>(branch_info.get()); - if (branch->address < 0) { - if (branch->kill) { - mm.InsertReturn(get_expr(branch->condition), true); - return; - } - mm.InsertReturn(get_expr(branch->condition), false); - return; - } - mm.InsertGoto(get_expr(branch->condition), branch->address); - return; - } - const auto* multi_branch = std::get_if<MultiBranch>(branch_info.get()); - for (const auto& branch_case : multi_branch->branches) { - mm.InsertGoto(MakeExpr<ExprGprEqual>(multi_branch->gpr, branch_case.cmp_value), - branch_case.address); - } -} - -void DecompileShader(CFGRebuildState& state) { - state.manager->Init(); - for (auto label : state.labels) { - state.manager->DeclareLabel(label); - } - for (const auto& block : state.block_info) { - if (state.labels.contains(block.start)) { - state.manager->InsertLabel(block.start); - } - const bool ignore = BlockBranchIsIgnored(block.branch); - const u32 end = ignore ? block.end + 1 : block.end; - state.manager->InsertBlock(block.start, end); - if (!ignore) { - InsertBranch(*state.manager, block.branch); - } - } - state.manager->Decompile(); -} - -} // Anonymous namespace - -std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address, - const CompilerSettings& settings, - Registry& registry) { - auto result_out = std::make_unique<ShaderCharacteristics>(); - if (settings.depth == CompileDepth::BruteForce) { - result_out->settings.depth = CompileDepth::BruteForce; - return result_out; - } - - CFGRebuildState state{program_code, start_address, registry}; - // Inspect Code and generate blocks - state.labels.clear(); - state.labels.emplace(start_address); - state.inspect_queries.push_back(state.start); - while (!state.inspect_queries.empty()) { - if (!TryInspectAddress(state)) { - result_out->settings.depth = CompileDepth::BruteForce; - return result_out; - } - } - - bool use_flow_stack = true; - - bool decompiled = false; - - if (settings.depth != CompileDepth::FlowStack) { - // Decompile Stacks - state.queries.push_back(Query{state.start, {}, {}}); - decompiled = true; - while (!state.queries.empty()) { - if (!TryQuery(state)) { - decompiled = false; - break; - } - } - } - - use_flow_stack = !decompiled; - - // Sort and organize results - std::sort(state.block_info.begin(), state.block_info.end(), - [](const BlockInfo& a, const BlockInfo& b) -> bool { return a.start < b.start; }); - if (decompiled && settings.depth != CompileDepth::NoFlowStack) { - ASTManager manager{settings.depth != CompileDepth::DecompileBackwards, - settings.disable_else_derivation}; - state.manager = &manager; - DecompileShader(state); - decompiled = state.manager->IsFullyDecompiled(); - if (!decompiled) { - if (settings.depth == CompileDepth::FullDecompile) { - LOG_CRITICAL(HW_GPU, "Failed to remove all the gotos!:"); - } else { - LOG_CRITICAL(HW_GPU, "Failed to remove all backward gotos!:"); - } - state.manager->ShowCurrentState("Of Shader"); - state.manager->Clear(); - } else { - auto characteristics = std::make_unique<ShaderCharacteristics>(); - characteristics->start = start_address; - characteristics->settings.depth = settings.depth; - characteristics->manager = std::move(manager); - characteristics->end = state.block_info.back().end + 1; - return characteristics; - } - } - - result_out->start = start_address; - result_out->settings.depth = - use_flow_stack ? CompileDepth::FlowStack : CompileDepth::NoFlowStack; - result_out->blocks.clear(); - for (auto& block : state.block_info) { - ShaderBlock new_block{}; - new_block.start = block.start; - new_block.end = block.end; - new_block.ignore_branch = BlockBranchIsIgnored(block.branch); - if (!new_block.ignore_branch) { - new_block.branch = block.branch; - } - result_out->end = std::max(result_out->end, block.end); - result_out->blocks.push_back(new_block); - } - if (!use_flow_stack) { - result_out->labels = std::move(state.labels); - return result_out; - } - - auto back = result_out->blocks.begin(); - auto next = std::next(back); - while (next != result_out->blocks.end()) { - if (!state.labels.contains(next->start) && next->start == back->end + 1) { - back->end = next->end; - next = result_out->blocks.erase(next); - continue; - } - back = next; - ++next; - } - - return result_out; -} -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/control_flow.h b/src/video_core/shader/control_flow.h deleted file mode 100644 index 37bf96492..000000000 --- a/src/video_core/shader/control_flow.h +++ /dev/null @@ -1,117 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <list> -#include <optional> -#include <set> -#include <variant> - -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/ast.h" -#include "video_core/shader/compiler_settings.h" -#include "video_core/shader/registry.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::ConditionCode; -using Tegra::Shader::Pred; - -constexpr s32 exit_branch = -1; - -struct Condition { - Pred predicate{Pred::UnusedIndex}; - ConditionCode cc{ConditionCode::T}; - - bool IsUnconditional() const { - return predicate == Pred::UnusedIndex && cc == ConditionCode::T; - } - - bool operator==(const Condition& other) const { - return std::tie(predicate, cc) == std::tie(other.predicate, other.cc); - } - - bool operator!=(const Condition& other) const { - return !operator==(other); - } -}; - -class SingleBranch { -public: - SingleBranch() = default; - explicit SingleBranch(Condition condition_, s32 address_, bool kill_, bool is_sync_, - bool is_brk_, bool ignore_) - : condition{condition_}, address{address_}, kill{kill_}, is_sync{is_sync_}, is_brk{is_brk_}, - ignore{ignore_} {} - - bool operator==(const SingleBranch& b) const { - return std::tie(condition, address, kill, is_sync, is_brk, ignore) == - std::tie(b.condition, b.address, b.kill, b.is_sync, b.is_brk, b.ignore); - } - - bool operator!=(const SingleBranch& b) const { - return !operator==(b); - } - - Condition condition{}; - s32 address{exit_branch}; - bool kill{}; - bool is_sync{}; - bool is_brk{}; - bool ignore{}; -}; - -struct CaseBranch { - explicit CaseBranch(u32 cmp_value_, u32 address_) : cmp_value{cmp_value_}, address{address_} {} - u32 cmp_value; - u32 address; -}; - -class MultiBranch { -public: - explicit MultiBranch(u32 gpr_, std::vector<CaseBranch>&& branches_) - : gpr{gpr_}, branches{std::move(branches_)} {} - - u32 gpr{}; - std::vector<CaseBranch> branches{}; -}; - -using BranchData = std::variant<SingleBranch, MultiBranch>; -using BlockBranchInfo = std::shared_ptr<BranchData>; - -bool BlockBranchInfoAreEqual(BlockBranchInfo first, BlockBranchInfo second); - -struct ShaderBlock { - u32 start{}; - u32 end{}; - bool ignore_branch{}; - BlockBranchInfo branch{}; - - bool operator==(const ShaderBlock& sb) const { - return std::tie(start, end, ignore_branch) == - std::tie(sb.start, sb.end, sb.ignore_branch) && - BlockBranchInfoAreEqual(branch, sb.branch); - } - - bool operator!=(const ShaderBlock& sb) const { - return !operator==(sb); - } -}; - -struct ShaderCharacteristics { - std::list<ShaderBlock> blocks{}; - std::set<u32> labels{}; - u32 start{}; - u32 end{}; - ASTManager manager{true, true}; - CompilerSettings settings{}; -}; - -std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address, - const CompilerSettings& settings, - Registry& registry); - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode.cpp b/src/video_core/shader/decode.cpp deleted file mode 100644 index 6576d1208..000000000 --- a/src/video_core/shader/decode.cpp +++ /dev/null @@ -1,368 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <cstring> -#include <limits> -#include <set> - -#include <fmt/format.h> - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/engines/shader_header.h" -#include "video_core/shader/control_flow.h" -#include "video_core/shader/memory_util.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -namespace { - -void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile& gpu_driver, - const std::list<SamplerEntry>& used_samplers) { - if (gpu_driver.IsTextureHandlerSizeKnown() || used_samplers.size() <= 1) { - return; - } - u32 count{}; - std::vector<u32> bound_offsets; - for (const auto& sampler : used_samplers) { - if (sampler.is_bindless) { - continue; - } - ++count; - bound_offsets.emplace_back(sampler.offset); - } - if (count > 1) { - gpu_driver.DeduceTextureHandlerSize(std::move(bound_offsets)); - } -} - -std::optional<u32> TryDeduceSamplerSize(const SamplerEntry& sampler_to_deduce, - VideoCore::GuestDriverProfile& gpu_driver, - const std::list<SamplerEntry>& used_samplers) { - const u32 base_offset = sampler_to_deduce.offset; - u32 max_offset{std::numeric_limits<u32>::max()}; - for (const auto& sampler : used_samplers) { - if (sampler.is_bindless) { - continue; - } - if (sampler.offset > base_offset) { - max_offset = std::min(sampler.offset, max_offset); - } - } - if (max_offset == std::numeric_limits<u32>::max()) { - return std::nullopt; - } - return ((max_offset - base_offset) * 4) / gpu_driver.GetTextureHandlerSize(); -} - -} // Anonymous namespace - -class ASTDecoder { -public: - explicit ASTDecoder(ShaderIR& ir_) : ir(ir_) {} - - void operator()(ASTProgram& ast) { - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - } - - void operator()(ASTIfThen& ast) { - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - } - - void operator()(ASTIfElse& ast) { - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - } - - void operator()(ASTBlockEncoded& ast) {} - - void operator()(ASTBlockDecoded& ast) {} - - void operator()(ASTVarSet& ast) {} - - void operator()(ASTLabel& ast) {} - - void operator()(ASTGoto& ast) {} - - void operator()(ASTDoWhile& ast) { - ASTNode current = ast.nodes.GetFirst(); - while (current) { - Visit(current); - current = current->GetNext(); - } - } - - void operator()(ASTReturn& ast) {} - - void operator()(ASTBreak& ast) {} - - void Visit(ASTNode& node) { - std::visit(*this, *node->GetInnerData()); - if (node->IsBlockEncoded()) { - auto block = std::get_if<ASTBlockEncoded>(node->GetInnerData()); - NodeBlock bb = ir.DecodeRange(block->start, block->end); - node->TransformBlockEncoded(std::move(bb)); - } - } - -private: - ShaderIR& ir; -}; - -void ShaderIR::Decode() { - std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header)); - - decompiled = false; - auto info = ScanFlow(program_code, main_offset, settings, registry); - auto& shader_info = *info; - coverage_begin = shader_info.start; - coverage_end = shader_info.end; - switch (shader_info.settings.depth) { - case CompileDepth::FlowStack: { - for (const auto& block : shader_info.blocks) { - basic_blocks.insert({block.start, DecodeRange(block.start, block.end + 1)}); - } - break; - } - case CompileDepth::NoFlowStack: { - disable_flow_stack = true; - const auto insert_block = [this](NodeBlock& nodes, u32 label) { - if (label == static_cast<u32>(exit_branch)) { - return; - } - basic_blocks.insert({label, nodes}); - }; - const auto& blocks = shader_info.blocks; - NodeBlock current_block; - u32 current_label = static_cast<u32>(exit_branch); - for (const auto& block : blocks) { - if (shader_info.labels.contains(block.start)) { - insert_block(current_block, current_label); - current_block.clear(); - current_label = block.start; - } - if (!block.ignore_branch) { - DecodeRangeInner(current_block, block.start, block.end); - InsertControlFlow(current_block, block); - } else { - DecodeRangeInner(current_block, block.start, block.end + 1); - } - } - insert_block(current_block, current_label); - break; - } - case CompileDepth::DecompileBackwards: - case CompileDepth::FullDecompile: { - program_manager = std::move(shader_info.manager); - disable_flow_stack = true; - decompiled = true; - ASTDecoder decoder{*this}; - ASTNode program = GetASTProgram(); - decoder.Visit(program); - break; - } - default: - LOG_CRITICAL(HW_GPU, "Unknown decompilation mode!"); - [[fallthrough]]; - case CompileDepth::BruteForce: { - const auto shader_end = static_cast<u32>(program_code.size()); - coverage_begin = main_offset; - coverage_end = shader_end; - for (u32 label = main_offset; label < shader_end; ++label) { - basic_blocks.insert({label, DecodeRange(label, label + 1)}); - } - break; - } - } - if (settings.depth != shader_info.settings.depth) { - LOG_WARNING( - HW_GPU, "Decompiling to this setting \"{}\" failed, downgrading to this setting \"{}\"", - CompileDepthAsString(settings.depth), CompileDepthAsString(shader_info.settings.depth)); - } -} - -NodeBlock ShaderIR::DecodeRange(u32 begin, u32 end) { - NodeBlock basic_block; - DecodeRangeInner(basic_block, begin, end); - return basic_block; -} - -void ShaderIR::DecodeRangeInner(NodeBlock& bb, u32 begin, u32 end) { - for (u32 pc = begin; pc < (begin > end ? MAX_PROGRAM_LENGTH : end);) { - pc = DecodeInstr(bb, pc); - } -} - -void ShaderIR::InsertControlFlow(NodeBlock& bb, const ShaderBlock& block) { - const auto apply_conditions = [&](const Condition& cond, Node n) -> Node { - Node result = n; - if (cond.cc != ConditionCode::T) { - result = Conditional(GetConditionCode(cond.cc), {result}); - } - if (cond.predicate != Pred::UnusedIndex) { - u32 pred = static_cast<u32>(cond.predicate); - const bool is_neg = pred > 7; - if (is_neg) { - pred -= 8; - } - result = Conditional(GetPredicate(pred, is_neg), {result}); - } - return result; - }; - if (std::holds_alternative<SingleBranch>(*block.branch)) { - auto branch = std::get_if<SingleBranch>(block.branch.get()); - if (branch->address < 0) { - if (branch->kill) { - Node n = Operation(OperationCode::Discard); - n = apply_conditions(branch->condition, n); - bb.push_back(n); - global_code.push_back(n); - return; - } - Node n = Operation(OperationCode::Exit); - n = apply_conditions(branch->condition, n); - bb.push_back(n); - global_code.push_back(n); - return; - } - Node n = Operation(OperationCode::Branch, Immediate(branch->address)); - n = apply_conditions(branch->condition, n); - bb.push_back(n); - global_code.push_back(n); - return; - } - auto multi_branch = std::get_if<MultiBranch>(block.branch.get()); - Node op_a = GetRegister(multi_branch->gpr); - for (auto& branch_case : multi_branch->branches) { - Node n = Operation(OperationCode::Branch, Immediate(branch_case.address)); - Node op_b = Immediate(branch_case.cmp_value); - Node condition = - GetPredicateComparisonInteger(Tegra::Shader::PredCondition::EQ, false, op_a, op_b); - auto result = Conditional(condition, {n}); - bb.push_back(result); - global_code.push_back(result); - } -} - -u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) { - // Ignore sched instructions when generating code. - if (IsSchedInstruction(pc, main_offset)) { - return pc + 1; - } - - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - const u32 nv_address = ConvertAddressToNvidiaSpace(pc); - - // Decoding failure - if (!opcode) { - UNIMPLEMENTED_MSG("Unhandled instruction: {0:x}", instr.value); - bb.push_back(Comment(fmt::format("{:05x} Unimplemented Shader instruction (0x{:016x})", - nv_address, instr.value))); - return pc + 1; - } - - bb.push_back(Comment( - fmt::format("{:05x} {} (0x{:016x})", nv_address, opcode->get().GetName(), instr.value))); - - using Tegra::Shader::Pred; - UNIMPLEMENTED_IF_MSG(instr.pred.full_pred == Pred::NeverExecute, - "NeverExecute predicate not implemented"); - - static const std::map<OpCode::Type, u32 (ShaderIR::*)(NodeBlock&, u32)> decoders = { - {OpCode::Type::Arithmetic, &ShaderIR::DecodeArithmetic}, - {OpCode::Type::ArithmeticImmediate, &ShaderIR::DecodeArithmeticImmediate}, - {OpCode::Type::Bfe, &ShaderIR::DecodeBfe}, - {OpCode::Type::Bfi, &ShaderIR::DecodeBfi}, - {OpCode::Type::Shift, &ShaderIR::DecodeShift}, - {OpCode::Type::ArithmeticInteger, &ShaderIR::DecodeArithmeticInteger}, - {OpCode::Type::ArithmeticIntegerImmediate, &ShaderIR::DecodeArithmeticIntegerImmediate}, - {OpCode::Type::ArithmeticHalf, &ShaderIR::DecodeArithmeticHalf}, - {OpCode::Type::ArithmeticHalfImmediate, &ShaderIR::DecodeArithmeticHalfImmediate}, - {OpCode::Type::Ffma, &ShaderIR::DecodeFfma}, - {OpCode::Type::Hfma2, &ShaderIR::DecodeHfma2}, - {OpCode::Type::Conversion, &ShaderIR::DecodeConversion}, - {OpCode::Type::Warp, &ShaderIR::DecodeWarp}, - {OpCode::Type::Memory, &ShaderIR::DecodeMemory}, - {OpCode::Type::Texture, &ShaderIR::DecodeTexture}, - {OpCode::Type::Image, &ShaderIR::DecodeImage}, - {OpCode::Type::FloatSetPredicate, &ShaderIR::DecodeFloatSetPredicate}, - {OpCode::Type::IntegerSetPredicate, &ShaderIR::DecodeIntegerSetPredicate}, - {OpCode::Type::HalfSetPredicate, &ShaderIR::DecodeHalfSetPredicate}, - {OpCode::Type::PredicateSetRegister, &ShaderIR::DecodePredicateSetRegister}, - {OpCode::Type::PredicateSetPredicate, &ShaderIR::DecodePredicateSetPredicate}, - {OpCode::Type::RegisterSetPredicate, &ShaderIR::DecodeRegisterSetPredicate}, - {OpCode::Type::FloatSet, &ShaderIR::DecodeFloatSet}, - {OpCode::Type::IntegerSet, &ShaderIR::DecodeIntegerSet}, - {OpCode::Type::HalfSet, &ShaderIR::DecodeHalfSet}, - {OpCode::Type::Video, &ShaderIR::DecodeVideo}, - {OpCode::Type::Xmad, &ShaderIR::DecodeXmad}, - }; - - std::vector<Node> tmp_block; - if (const auto decoder = decoders.find(opcode->get().GetType()); decoder != decoders.end()) { - pc = (this->*decoder->second)(tmp_block, pc); - } else { - pc = DecodeOther(tmp_block, pc); - } - - // Some instructions (like SSY) don't have a predicate field, they are always unconditionally - // executed. - const bool can_be_predicated = OpCode::IsPredicatedInstruction(opcode->get().GetId()); - const auto pred_index = static_cast<u32>(instr.pred.pred_index); - - if (can_be_predicated && pred_index != static_cast<u32>(Pred::UnusedIndex)) { - const Node conditional = - Conditional(GetPredicate(pred_index, instr.negate_pred != 0), std::move(tmp_block)); - global_code.push_back(conditional); - bb.push_back(conditional); - } else { - for (auto& node : tmp_block) { - global_code.push_back(node); - bb.push_back(node); - } - } - - return pc + 1; -} - -void ShaderIR::PostDecode() { - // Deduce texture handler size if needed - auto gpu_driver = registry.AccessGuestDriverProfile(); - DeduceTextureHandlerSize(gpu_driver, used_samplers); - // Deduce Indexed Samplers - if (!uses_indexed_samplers) { - return; - } - for (auto& sampler : used_samplers) { - if (!sampler.is_indexed) { - continue; - } - if (const auto size = TryDeduceSamplerSize(sampler, gpu_driver, used_samplers)) { - sampler.size = *size; - } else { - LOG_CRITICAL(HW_GPU, "Failed to deduce size of indexed sampler"); - sampler.size = 1; - } - } -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/arithmetic.cpp b/src/video_core/shader/decode/arithmetic.cpp deleted file mode 100644 index 15eb700e7..000000000 --- a/src/video_core/shader/decode/arithmetic.cpp +++ /dev/null @@ -1,166 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "common/logging/log.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::SubOp; - -u32 ShaderIR::DecodeArithmetic(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - Node op_a = GetRegister(instr.gpr8); - - Node op_b = [&] { - if (instr.is_b_imm) { - return GetImmediate19(instr); - } else if (instr.is_b_gpr) { - return GetRegister(instr.gpr20); - } else { - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - } - }(); - - switch (opcode->get().GetId()) { - case OpCode::Id::MOV_C: - case OpCode::Id::MOV_R: { - // MOV does not have neither 'abs' nor 'neg' bits. - SetRegister(bb, instr.gpr0, op_b); - break; - } - case OpCode::Id::FMUL_C: - case OpCode::Id::FMUL_R: - case OpCode::Id::FMUL_IMM: { - // FMUL does not have 'abs' bits and only the second operand has a 'neg' bit. - if (instr.fmul.tab5cb8_2 != 0) { - LOG_DEBUG(HW_GPU, "FMUL tab5cb8_2({}) is not implemented", - instr.fmul.tab5cb8_2.Value()); - } - if (instr.fmul.tab5c68_0 != 1) { - LOG_DEBUG(HW_GPU, "FMUL tab5cb8_0({}) is not implemented", - instr.fmul.tab5c68_0.Value()); - } - - op_b = GetOperandAbsNegFloat(op_b, false, instr.fmul.negate_b); - - static constexpr std::array FmulPostFactor = { - 1.000f, // None - 0.500f, // Divide 2 - 0.250f, // Divide 4 - 0.125f, // Divide 8 - 8.000f, // Mul 8 - 4.000f, // Mul 4 - 2.000f, // Mul 2 - }; - - if (instr.fmul.postfactor != 0) { - op_a = Operation(OperationCode::FMul, NO_PRECISE, op_a, - Immediate(FmulPostFactor[instr.fmul.postfactor])); - } - - // TODO(Rodrigo): Should precise be used when there's a postfactor? - Node value = Operation(OperationCode::FMul, PRECISE, op_a, op_b); - - value = GetSaturatedFloat(value, instr.alu.saturate_d); - - SetInternalFlagsFromFloat(bb, value, instr.generates_cc); - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::FADD_C: - case OpCode::Id::FADD_R: - case OpCode::Id::FADD_IMM: { - op_a = GetOperandAbsNegFloat(op_a, instr.alu.abs_a, instr.alu.negate_a); - op_b = GetOperandAbsNegFloat(op_b, instr.alu.abs_b, instr.alu.negate_b); - - Node value = Operation(OperationCode::FAdd, PRECISE, op_a, op_b); - value = GetSaturatedFloat(value, instr.alu.saturate_d); - - SetInternalFlagsFromFloat(bb, value, instr.generates_cc); - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::MUFU: { - op_a = GetOperandAbsNegFloat(op_a, instr.alu.abs_a, instr.alu.negate_a); - - Node value = [&]() { - switch (instr.sub_op) { - case SubOp::Cos: - return Operation(OperationCode::FCos, PRECISE, op_a); - case SubOp::Sin: - return Operation(OperationCode::FSin, PRECISE, op_a); - case SubOp::Ex2: - return Operation(OperationCode::FExp2, PRECISE, op_a); - case SubOp::Lg2: - return Operation(OperationCode::FLog2, PRECISE, op_a); - case SubOp::Rcp: - return Operation(OperationCode::FDiv, PRECISE, Immediate(1.0f), op_a); - case SubOp::Rsq: - return Operation(OperationCode::FInverseSqrt, PRECISE, op_a); - case SubOp::Sqrt: - return Operation(OperationCode::FSqrt, PRECISE, op_a); - default: - UNIMPLEMENTED_MSG("Unhandled MUFU sub op={0:x}", instr.sub_op.Value()); - return Immediate(0); - } - }(); - value = GetSaturatedFloat(value, instr.alu.saturate_d); - - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::FMNMX_C: - case OpCode::Id::FMNMX_R: - case OpCode::Id::FMNMX_IMM: { - op_a = GetOperandAbsNegFloat(op_a, instr.alu.abs_a, instr.alu.negate_a); - op_b = GetOperandAbsNegFloat(op_b, instr.alu.abs_b, instr.alu.negate_b); - - const Node condition = GetPredicate(instr.alu.fmnmx.pred, instr.alu.fmnmx.negate_pred != 0); - - const Node min = Operation(OperationCode::FMin, NO_PRECISE, op_a, op_b); - const Node max = Operation(OperationCode::FMax, NO_PRECISE, op_a, op_b); - const Node value = Operation(OperationCode::Select, NO_PRECISE, condition, min, max); - - SetInternalFlagsFromFloat(bb, value, instr.generates_cc); - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::FCMP_RR: - case OpCode::Id::FCMP_RC: - case OpCode::Id::FCMP_IMMR: { - UNIMPLEMENTED_IF(instr.fcmp.ftz == 0); - Node op_c = GetRegister(instr.gpr39); - Node comp = GetPredicateComparisonFloat(instr.fcmp.cond, std::move(op_c), Immediate(0.0f)); - SetRegister( - bb, instr.gpr0, - Operation(OperationCode::Select, std::move(comp), std::move(op_a), std::move(op_b))); - break; - } - case OpCode::Id::RRO_C: - case OpCode::Id::RRO_R: - case OpCode::Id::RRO_IMM: { - LOG_DEBUG(HW_GPU, "(STUBBED) RRO used"); - - // Currently RRO is only implemented as a register move. - op_b = GetOperandAbsNegFloat(op_b, instr.alu.abs_b, instr.alu.negate_b); - SetRegister(bb, instr.gpr0, op_b); - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled arithmetic instruction: {}", opcode->get().GetName()); - } - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/arithmetic_half.cpp b/src/video_core/shader/decode/arithmetic_half.cpp deleted file mode 100644 index 88103fede..000000000 --- a/src/video_core/shader/decode/arithmetic_half.cpp +++ /dev/null @@ -1,101 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "common/logging/log.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::HalfType; -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -u32 ShaderIR::DecodeArithmeticHalf(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - bool negate_a = false; - bool negate_b = false; - bool absolute_a = false; - bool absolute_b = false; - - switch (opcode->get().GetId()) { - case OpCode::Id::HADD2_R: - if (instr.alu_half.ftz == 0) { - LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName()); - } - negate_a = ((instr.value >> 43) & 1) != 0; - negate_b = ((instr.value >> 31) & 1) != 0; - absolute_a = ((instr.value >> 44) & 1) != 0; - absolute_b = ((instr.value >> 30) & 1) != 0; - break; - case OpCode::Id::HADD2_C: - if (instr.alu_half.ftz == 0) { - LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName()); - } - negate_a = ((instr.value >> 43) & 1) != 0; - negate_b = ((instr.value >> 56) & 1) != 0; - absolute_a = ((instr.value >> 44) & 1) != 0; - absolute_b = ((instr.value >> 54) & 1) != 0; - break; - case OpCode::Id::HMUL2_R: - negate_a = ((instr.value >> 43) & 1) != 0; - absolute_a = ((instr.value >> 44) & 1) != 0; - absolute_b = ((instr.value >> 30) & 1) != 0; - break; - case OpCode::Id::HMUL2_C: - negate_b = ((instr.value >> 31) & 1) != 0; - absolute_a = ((instr.value >> 44) & 1) != 0; - absolute_b = ((instr.value >> 54) & 1) != 0; - break; - default: - UNREACHABLE(); - break; - } - - Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.alu_half.type_a); - op_a = GetOperandAbsNegHalf(op_a, absolute_a, negate_a); - - auto [type_b, op_b] = [this, instr, opcode]() -> std::pair<HalfType, Node> { - switch (opcode->get().GetId()) { - case OpCode::Id::HADD2_C: - case OpCode::Id::HMUL2_C: - return {HalfType::F32, GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())}; - case OpCode::Id::HADD2_R: - case OpCode::Id::HMUL2_R: - return {instr.alu_half.type_b, GetRegister(instr.gpr20)}; - default: - UNREACHABLE(); - return {HalfType::F32, Immediate(0)}; - } - }(); - op_b = UnpackHalfFloat(op_b, type_b); - op_b = GetOperandAbsNegHalf(op_b, absolute_b, negate_b); - - Node value = [this, opcode, op_a, op_b = op_b] { - switch (opcode->get().GetId()) { - case OpCode::Id::HADD2_C: - case OpCode::Id::HADD2_R: - return Operation(OperationCode::HAdd, PRECISE, op_a, op_b); - case OpCode::Id::HMUL2_C: - case OpCode::Id::HMUL2_R: - return Operation(OperationCode::HMul, PRECISE, op_a, op_b); - default: - UNIMPLEMENTED_MSG("Unhandled half float instruction: {}", opcode->get().GetName()); - return Immediate(0); - } - }(); - value = GetSaturatedHalfFloat(value, instr.alu_half.saturate); - value = HalfMerge(GetRegister(instr.gpr0), value, instr.alu_half.merge); - - SetRegister(bb, instr.gpr0, value); - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/arithmetic_half_immediate.cpp b/src/video_core/shader/decode/arithmetic_half_immediate.cpp deleted file mode 100644 index d179b9873..000000000 --- a/src/video_core/shader/decode/arithmetic_half_immediate.cpp +++ /dev/null @@ -1,54 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "common/logging/log.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -u32 ShaderIR::DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - if (opcode->get().GetId() == OpCode::Id::HADD2_IMM) { - if (instr.alu_half_imm.ftz == 0) { - LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName()); - } - } else { - if (instr.alu_half_imm.precision != Tegra::Shader::HalfPrecision::FTZ) { - LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName()); - } - } - - Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.alu_half_imm.type_a); - op_a = GetOperandAbsNegHalf(op_a, instr.alu_half_imm.abs_a, instr.alu_half_imm.negate_a); - - const Node op_b = UnpackHalfImmediate(instr, true); - - Node value = [&]() { - switch (opcode->get().GetId()) { - case OpCode::Id::HADD2_IMM: - return Operation(OperationCode::HAdd, PRECISE, op_a, op_b); - case OpCode::Id::HMUL2_IMM: - return Operation(OperationCode::HMul, PRECISE, op_a, op_b); - default: - UNREACHABLE(); - return Immediate(0); - } - }(); - - value = GetSaturatedHalfFloat(value, instr.alu_half_imm.saturate); - value = HalfMerge(GetRegister(instr.gpr0), value, instr.alu_half_imm.merge); - SetRegister(bb, instr.gpr0, value); - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/arithmetic_immediate.cpp b/src/video_core/shader/decode/arithmetic_immediate.cpp deleted file mode 100644 index f1875967c..000000000 --- a/src/video_core/shader/decode/arithmetic_immediate.cpp +++ /dev/null @@ -1,53 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -u32 ShaderIR::DecodeArithmeticImmediate(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - switch (opcode->get().GetId()) { - case OpCode::Id::MOV32_IMM: { - SetRegister(bb, instr.gpr0, GetImmediate32(instr)); - break; - } - case OpCode::Id::FMUL32_IMM: { - Node value = - Operation(OperationCode::FMul, PRECISE, GetRegister(instr.gpr8), GetImmediate32(instr)); - value = GetSaturatedFloat(value, instr.fmul32.saturate); - - SetInternalFlagsFromFloat(bb, value, instr.op_32.generates_cc); - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::FADD32I: { - const Node op_a = GetOperandAbsNegFloat(GetRegister(instr.gpr8), instr.fadd32i.abs_a, - instr.fadd32i.negate_a); - const Node op_b = GetOperandAbsNegFloat(GetImmediate32(instr), instr.fadd32i.abs_b, - instr.fadd32i.negate_b); - - const Node value = Operation(OperationCode::FAdd, PRECISE, op_a, op_b); - SetInternalFlagsFromFloat(bb, value, instr.op_32.generates_cc); - SetRegister(bb, instr.gpr0, value); - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled arithmetic immediate instruction: {}", - opcode->get().GetName()); - } - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/arithmetic_integer.cpp b/src/video_core/shader/decode/arithmetic_integer.cpp deleted file mode 100644 index 7b5bb7003..000000000 --- a/src/video_core/shader/decode/arithmetic_integer.cpp +++ /dev/null @@ -1,375 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::IAdd3Height; -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::Pred; -using Tegra::Shader::Register; - -u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - Node op_a = GetRegister(instr.gpr8); - Node op_b = [&]() { - if (instr.is_b_imm) { - return Immediate(instr.alu.GetSignedImm20_20()); - } else if (instr.is_b_gpr) { - return GetRegister(instr.gpr20); - } else { - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - } - }(); - - switch (opcode->get().GetId()) { - case OpCode::Id::IADD_C: - case OpCode::Id::IADD_R: - case OpCode::Id::IADD_IMM: { - UNIMPLEMENTED_IF_MSG(instr.alu.saturate_d, "IADD.SAT"); - UNIMPLEMENTED_IF_MSG(instr.iadd.x && instr.generates_cc, "IADD.X Rd.CC"); - - op_a = GetOperandAbsNegInteger(op_a, false, instr.alu_integer.negate_a, true); - op_b = GetOperandAbsNegInteger(op_b, false, instr.alu_integer.negate_b, true); - - Node value = Operation(OperationCode::UAdd, op_a, op_b); - - if (instr.iadd.x) { - Node carry = GetInternalFlag(InternalFlag::Carry); - Node x = Operation(OperationCode::Select, std::move(carry), Immediate(1), Immediate(0)); - value = Operation(OperationCode::UAdd, std::move(value), std::move(x)); - } - - if (instr.generates_cc) { - const Node i0 = Immediate(0); - - Node zero = Operation(OperationCode::LogicalIEqual, value, i0); - Node sign = Operation(OperationCode::LogicalILessThan, value, i0); - Node carry = Operation(OperationCode::LogicalAddCarry, op_a, op_b); - - Node pos_a = Operation(OperationCode::LogicalIGreaterThan, op_a, i0); - Node pos_b = Operation(OperationCode::LogicalIGreaterThan, op_b, i0); - Node pos = Operation(OperationCode::LogicalAnd, std::move(pos_a), std::move(pos_b)); - Node overflow = Operation(OperationCode::LogicalAnd, pos, sign); - - SetInternalFlag(bb, InternalFlag::Zero, std::move(zero)); - SetInternalFlag(bb, InternalFlag::Sign, std::move(sign)); - SetInternalFlag(bb, InternalFlag::Carry, std::move(carry)); - SetInternalFlag(bb, InternalFlag::Overflow, std::move(overflow)); - } - SetRegister(bb, instr.gpr0, std::move(value)); - break; - } - case OpCode::Id::IADD3_C: - case OpCode::Id::IADD3_R: - case OpCode::Id::IADD3_IMM: { - Node op_c = GetRegister(instr.gpr39); - - const auto ApplyHeight = [&](IAdd3Height height, Node value) { - switch (height) { - case IAdd3Height::None: - return value; - case IAdd3Height::LowerHalfWord: - return BitfieldExtract(value, 0, 16); - case IAdd3Height::UpperHalfWord: - return BitfieldExtract(value, 16, 16); - default: - UNIMPLEMENTED_MSG("Unhandled IADD3 height: {}", height); - return Immediate(0); - } - }; - - if (opcode->get().GetId() == OpCode::Id::IADD3_R) { - op_a = ApplyHeight(instr.iadd3.height_a, op_a); - op_b = ApplyHeight(instr.iadd3.height_b, op_b); - op_c = ApplyHeight(instr.iadd3.height_c, op_c); - } - - op_a = GetOperandAbsNegInteger(op_a, false, instr.iadd3.neg_a, true); - op_b = GetOperandAbsNegInteger(op_b, false, instr.iadd3.neg_b, true); - op_c = GetOperandAbsNegInteger(op_c, false, instr.iadd3.neg_c, true); - - const Node value = [&] { - Node add_ab = Operation(OperationCode::IAdd, NO_PRECISE, op_a, op_b); - if (opcode->get().GetId() != OpCode::Id::IADD3_R) { - return Operation(OperationCode::IAdd, NO_PRECISE, add_ab, op_c); - } - const Node shifted = [&] { - switch (instr.iadd3.mode) { - case Tegra::Shader::IAdd3Mode::RightShift: - // TODO(tech4me): According to - // https://envytools.readthedocs.io/en/latest/hw/graph/maxwell/cuda/int.html?highlight=iadd3 - // The addition between op_a and op_b should be done in uint33, more - // investigation required - return Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, add_ab, - Immediate(16)); - case Tegra::Shader::IAdd3Mode::LeftShift: - return Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, add_ab, - Immediate(16)); - default: - return add_ab; - } - }(); - return Operation(OperationCode::IAdd, NO_PRECISE, shifted, op_c); - }(); - - SetInternalFlagsFromInteger(bb, value, instr.generates_cc); - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::ISCADD_C: - case OpCode::Id::ISCADD_R: - case OpCode::Id::ISCADD_IMM: { - UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "Condition codes generation in ISCADD is not implemented"); - - op_a = GetOperandAbsNegInteger(op_a, false, instr.alu_integer.negate_a, true); - op_b = GetOperandAbsNegInteger(op_b, false, instr.alu_integer.negate_b, true); - - const Node shift = Immediate(static_cast<u32>(instr.alu_integer.shift_amount)); - const Node shifted_a = Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, op_a, shift); - const Node value = Operation(OperationCode::IAdd, NO_PRECISE, shifted_a, op_b); - - SetInternalFlagsFromInteger(bb, value, instr.generates_cc); - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::POPC_C: - case OpCode::Id::POPC_R: - case OpCode::Id::POPC_IMM: { - if (instr.popc.invert) { - op_b = Operation(OperationCode::IBitwiseNot, NO_PRECISE, op_b); - } - const Node value = Operation(OperationCode::IBitCount, PRECISE, op_b); - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::FLO_R: - case OpCode::Id::FLO_C: - case OpCode::Id::FLO_IMM: { - Node value; - if (instr.flo.invert) { - op_b = Operation(OperationCode::IBitwiseNot, NO_PRECISE, std::move(op_b)); - } - if (instr.flo.is_signed) { - value = Operation(OperationCode::IBitMSB, NO_PRECISE, std::move(op_b)); - } else { - value = Operation(OperationCode::UBitMSB, NO_PRECISE, std::move(op_b)); - } - if (instr.flo.sh) { - value = - Operation(OperationCode::UBitwiseXor, NO_PRECISE, std::move(value), Immediate(31)); - } - SetRegister(bb, instr.gpr0, std::move(value)); - break; - } - case OpCode::Id::SEL_C: - case OpCode::Id::SEL_R: - case OpCode::Id::SEL_IMM: { - const Node condition = GetPredicate(instr.sel.pred, instr.sel.neg_pred != 0); - const Node value = Operation(OperationCode::Select, PRECISE, condition, op_a, op_b); - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::ICMP_CR: - case OpCode::Id::ICMP_R: - case OpCode::Id::ICMP_RC: - case OpCode::Id::ICMP_IMM: { - const Node zero = Immediate(0); - - const auto [op_rhs, test] = [&]() -> std::pair<Node, Node> { - switch (opcode->get().GetId()) { - case OpCode::Id::ICMP_CR: - return {GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()), - GetRegister(instr.gpr39)}; - case OpCode::Id::ICMP_R: - return {GetRegister(instr.gpr20), GetRegister(instr.gpr39)}; - case OpCode::Id::ICMP_RC: - return {GetRegister(instr.gpr39), - GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())}; - case OpCode::Id::ICMP_IMM: - return {Immediate(instr.alu.GetSignedImm20_20()), GetRegister(instr.gpr39)}; - default: - UNREACHABLE(); - return {zero, zero}; - } - }(); - const Node op_lhs = GetRegister(instr.gpr8); - const Node comparison = - GetPredicateComparisonInteger(instr.icmp.cond, instr.icmp.is_signed != 0, test, zero); - SetRegister(bb, instr.gpr0, Operation(OperationCode::Select, comparison, op_lhs, op_rhs)); - break; - } - case OpCode::Id::LOP_C: - case OpCode::Id::LOP_R: - case OpCode::Id::LOP_IMM: { - if (instr.alu.lop.invert_a) - op_a = Operation(OperationCode::IBitwiseNot, NO_PRECISE, op_a); - if (instr.alu.lop.invert_b) - op_b = Operation(OperationCode::IBitwiseNot, NO_PRECISE, op_b); - - WriteLogicOperation(bb, instr.gpr0, instr.alu.lop.operation, op_a, op_b, - instr.alu.lop.pred_result_mode, instr.alu.lop.pred48, - instr.generates_cc); - break; - } - case OpCode::Id::LOP3_C: - case OpCode::Id::LOP3_R: - case OpCode::Id::LOP3_IMM: { - const Node op_c = GetRegister(instr.gpr39); - const Node lut = [&]() { - if (opcode->get().GetId() == OpCode::Id::LOP3_R) { - return Immediate(instr.alu.lop3.GetImmLut28()); - } else { - return Immediate(instr.alu.lop3.GetImmLut48()); - } - }(); - - WriteLop3Instruction(bb, instr.gpr0, op_a, op_b, op_c, lut, instr.generates_cc); - break; - } - case OpCode::Id::IMNMX_C: - case OpCode::Id::IMNMX_R: - case OpCode::Id::IMNMX_IMM: { - UNIMPLEMENTED_IF(instr.imnmx.exchange != Tegra::Shader::IMinMaxExchange::None); - - const bool is_signed = instr.imnmx.is_signed; - - const Node condition = GetPredicate(instr.imnmx.pred, instr.imnmx.negate_pred != 0); - const Node min = SignedOperation(OperationCode::IMin, is_signed, NO_PRECISE, op_a, op_b); - const Node max = SignedOperation(OperationCode::IMax, is_signed, NO_PRECISE, op_a, op_b); - const Node value = Operation(OperationCode::Select, NO_PRECISE, condition, min, max); - - SetInternalFlagsFromInteger(bb, value, instr.generates_cc); - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::LEA_R2: - case OpCode::Id::LEA_R1: - case OpCode::Id::LEA_IMM: - case OpCode::Id::LEA_RZ: - case OpCode::Id::LEA_HI: { - auto [op_a_, op_b_, op_c_] = [&]() -> std::tuple<Node, Node, Node> { - switch (opcode->get().GetId()) { - case OpCode::Id::LEA_R2: { - return {GetRegister(instr.gpr20), GetRegister(instr.gpr39), - Immediate(static_cast<u32>(instr.lea.r2.entry_a))}; - } - case OpCode::Id::LEA_R1: { - const bool neg = instr.lea.r1.neg != 0; - return {GetOperandAbsNegInteger(GetRegister(instr.gpr8), false, neg, true), - GetRegister(instr.gpr20), - Immediate(static_cast<u32>(instr.lea.r1.entry_a))}; - } - case OpCode::Id::LEA_IMM: { - const bool neg = instr.lea.imm.neg != 0; - return {GetOperandAbsNegInteger(GetRegister(instr.gpr8), false, neg, true), - Immediate(static_cast<u32>(instr.lea.imm.entry_a)), - Immediate(static_cast<u32>(instr.lea.imm.entry_b))}; - } - case OpCode::Id::LEA_RZ: { - const bool neg = instr.lea.rz.neg != 0; - return {GetConstBuffer(instr.lea.rz.cb_index, instr.lea.rz.cb_offset), - GetOperandAbsNegInteger(GetRegister(instr.gpr8), false, neg, true), - Immediate(static_cast<u32>(instr.lea.rz.entry_a))}; - } - case OpCode::Id::LEA_HI: - default: - UNIMPLEMENTED_MSG("Unhandled LEA subinstruction: {}", opcode->get().GetName()); - - return {Immediate(static_cast<u32>(instr.lea.imm.entry_a)), GetRegister(instr.gpr8), - Immediate(static_cast<u32>(instr.lea.imm.entry_b))}; - } - }(); - - UNIMPLEMENTED_IF_MSG(instr.lea.pred48 != static_cast<u64>(Pred::UnusedIndex), - "Unhandled LEA Predicate"); - - Node value = - Operation(OperationCode::ILogicalShiftLeft, std::move(op_a_), std::move(op_c_)); - value = Operation(OperationCode::IAdd, std::move(op_b_), std::move(value)); - SetRegister(bb, instr.gpr0, std::move(value)); - - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled ArithmeticInteger instruction: {}", opcode->get().GetName()); - } - - return pc; -} - -void ShaderIR::WriteLop3Instruction(NodeBlock& bb, Register dest, Node op_a, Node op_b, Node op_c, - Node imm_lut, bool sets_cc) { - const Node lop3_fast = [&](const Node na, const Node nb, const Node nc, const Node ttbl) { - Node value = Immediate(0); - const ImmediateNode imm = std::get<ImmediateNode>(*ttbl); - if (imm.GetValue() & 0x01) { - const Node a = Operation(OperationCode::IBitwiseNot, na); - const Node b = Operation(OperationCode::IBitwiseNot, nb); - const Node c = Operation(OperationCode::IBitwiseNot, nc); - Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, b); - r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, c); - value = Operation(OperationCode::IBitwiseOr, value, r); - } - if (imm.GetValue() & 0x02) { - const Node a = Operation(OperationCode::IBitwiseNot, na); - const Node b = Operation(OperationCode::IBitwiseNot, nb); - Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, b); - r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, nc); - value = Operation(OperationCode::IBitwiseOr, value, r); - } - if (imm.GetValue() & 0x04) { - const Node a = Operation(OperationCode::IBitwiseNot, na); - const Node c = Operation(OperationCode::IBitwiseNot, nc); - Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, nb); - r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, c); - value = Operation(OperationCode::IBitwiseOr, value, r); - } - if (imm.GetValue() & 0x08) { - const Node a = Operation(OperationCode::IBitwiseNot, na); - Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, a, nb); - r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, nc); - value = Operation(OperationCode::IBitwiseOr, value, r); - } - if (imm.GetValue() & 0x10) { - const Node b = Operation(OperationCode::IBitwiseNot, nb); - const Node c = Operation(OperationCode::IBitwiseNot, nc); - Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, na, b); - r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, c); - value = Operation(OperationCode::IBitwiseOr, value, r); - } - if (imm.GetValue() & 0x20) { - const Node b = Operation(OperationCode::IBitwiseNot, nb); - Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, na, b); - r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, nc); - value = Operation(OperationCode::IBitwiseOr, value, r); - } - if (imm.GetValue() & 0x40) { - const Node c = Operation(OperationCode::IBitwiseNot, nc); - Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, na, nb); - r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, c); - value = Operation(OperationCode::IBitwiseOr, value, r); - } - if (imm.GetValue() & 0x80) { - Node r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, na, nb); - r = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, r, nc); - value = Operation(OperationCode::IBitwiseOr, value, r); - } - return value; - }(op_a, op_b, op_c, imm_lut); - - SetInternalFlagsFromInteger(bb, lop3_fast, sets_cc); - SetRegister(bb, dest, lop3_fast); -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/arithmetic_integer_immediate.cpp b/src/video_core/shader/decode/arithmetic_integer_immediate.cpp deleted file mode 100644 index 73580277a..000000000 --- a/src/video_core/shader/decode/arithmetic_integer_immediate.cpp +++ /dev/null @@ -1,99 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::LogicOperation; -using Tegra::Shader::OpCode; -using Tegra::Shader::Pred; -using Tegra::Shader::PredicateResultMode; -using Tegra::Shader::Register; - -u32 ShaderIR::DecodeArithmeticIntegerImmediate(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - Node op_a = GetRegister(instr.gpr8); - Node op_b = Immediate(static_cast<s32>(instr.alu.imm20_32)); - - switch (opcode->get().GetId()) { - case OpCode::Id::IADD32I: { - UNIMPLEMENTED_IF_MSG(instr.iadd32i.saturate, "IADD32I saturation is not implemented"); - - op_a = GetOperandAbsNegInteger(std::move(op_a), false, instr.iadd32i.negate_a != 0, true); - - Node value = Operation(OperationCode::IAdd, PRECISE, std::move(op_a), std::move(op_b)); - - SetInternalFlagsFromInteger(bb, value, instr.op_32.generates_cc != 0); - SetRegister(bb, instr.gpr0, std::move(value)); - break; - } - case OpCode::Id::LOP32I: { - if (instr.alu.lop32i.invert_a) { - op_a = Operation(OperationCode::IBitwiseNot, NO_PRECISE, std::move(op_a)); - } - - if (instr.alu.lop32i.invert_b) { - op_b = Operation(OperationCode::IBitwiseNot, NO_PRECISE, std::move(op_b)); - } - - WriteLogicOperation(bb, instr.gpr0, instr.alu.lop32i.operation, std::move(op_a), - std::move(op_b), PredicateResultMode::None, Pred::UnusedIndex, - instr.op_32.generates_cc != 0); - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled ArithmeticIntegerImmediate instruction: {}", - opcode->get().GetName()); - } - - return pc; -} - -void ShaderIR::WriteLogicOperation(NodeBlock& bb, Register dest, LogicOperation logic_op, Node op_a, - Node op_b, PredicateResultMode predicate_mode, Pred predicate, - bool sets_cc) { - Node result = [&] { - switch (logic_op) { - case LogicOperation::And: - return Operation(OperationCode::IBitwiseAnd, PRECISE, std::move(op_a), std::move(op_b)); - case LogicOperation::Or: - return Operation(OperationCode::IBitwiseOr, PRECISE, std::move(op_a), std::move(op_b)); - case LogicOperation::Xor: - return Operation(OperationCode::IBitwiseXor, PRECISE, std::move(op_a), std::move(op_b)); - case LogicOperation::PassB: - return op_b; - default: - UNIMPLEMENTED_MSG("Unimplemented logic operation={}", logic_op); - return Immediate(0); - } - }(); - - SetInternalFlagsFromInteger(bb, result, sets_cc); - SetRegister(bb, dest, result); - - // Write the predicate value depending on the predicate mode. - switch (predicate_mode) { - case PredicateResultMode::None: - // Do nothing. - return; - case PredicateResultMode::NotZero: { - // Set the predicate to true if the result is not zero. - Node compare = Operation(OperationCode::LogicalINotEqual, std::move(result), Immediate(0)); - SetPredicate(bb, static_cast<u64>(predicate), std::move(compare)); - break; - } - default: - UNIMPLEMENTED_MSG("Unimplemented predicate result mode: {}", predicate_mode); - } -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/bfe.cpp b/src/video_core/shader/decode/bfe.cpp deleted file mode 100644 index 8e3b46e8e..000000000 --- a/src/video_core/shader/decode/bfe.cpp +++ /dev/null @@ -1,77 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -u32 ShaderIR::DecodeBfe(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - Node op_a = GetRegister(instr.gpr8); - Node op_b = [&] { - switch (opcode->get().GetId()) { - case OpCode::Id::BFE_R: - return GetRegister(instr.gpr20); - case OpCode::Id::BFE_C: - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - case OpCode::Id::BFE_IMM: - return Immediate(instr.alu.GetSignedImm20_20()); - default: - UNREACHABLE(); - return Immediate(0); - } - }(); - - UNIMPLEMENTED_IF_MSG(instr.bfe.rd_cc, "Condition codes in BFE is not implemented"); - - const bool is_signed = instr.bfe.is_signed; - - // using reverse parallel method in - // https://graphics.stanford.edu/~seander/bithacks.html#ReverseParallel - // note for later if possible to implement faster method. - if (instr.bfe.brev) { - const auto swap = [&](u32 s, u32 mask) { - Node v1 = - SignedOperation(OperationCode::ILogicalShiftRight, is_signed, op_a, Immediate(s)); - if (mask != 0) { - v1 = SignedOperation(OperationCode::IBitwiseAnd, is_signed, std::move(v1), - Immediate(mask)); - } - Node v2 = op_a; - if (mask != 0) { - v2 = SignedOperation(OperationCode::IBitwiseAnd, is_signed, std::move(v2), - Immediate(mask)); - } - v2 = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, std::move(v2), - Immediate(s)); - return SignedOperation(OperationCode::IBitwiseOr, is_signed, std::move(v1), - std::move(v2)); - }; - op_a = swap(1, 0x55555555U); - op_a = swap(2, 0x33333333U); - op_a = swap(4, 0x0F0F0F0FU); - op_a = swap(8, 0x00FF00FFU); - op_a = swap(16, 0); - } - - const auto offset = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_b, - Immediate(0), Immediate(8)); - const auto bits = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_b, - Immediate(8), Immediate(8)); - auto result = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_a, offset, bits); - SetRegister(bb, instr.gpr0, std::move(result)); - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/bfi.cpp b/src/video_core/shader/decode/bfi.cpp deleted file mode 100644 index 70d1c055b..000000000 --- a/src/video_core/shader/decode/bfi.cpp +++ /dev/null @@ -1,45 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -u32 ShaderIR::DecodeBfi(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - const auto [packed_shift, base] = [&]() -> std::pair<Node, Node> { - switch (opcode->get().GetId()) { - case OpCode::Id::BFI_RC: - return {GetRegister(instr.gpr39), - GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())}; - case OpCode::Id::BFI_IMM_R: - return {Immediate(instr.alu.GetSignedImm20_20()), GetRegister(instr.gpr39)}; - default: - UNREACHABLE(); - return {Immediate(0), Immediate(0)}; - } - }(); - const Node insert = GetRegister(instr.gpr8); - const Node offset = BitfieldExtract(packed_shift, 0, 8); - const Node bits = BitfieldExtract(packed_shift, 8, 8); - - const Node value = - Operation(OperationCode::UBitfieldInsert, PRECISE, base, insert, offset, bits); - - SetInternalFlagsFromInteger(bb, value, instr.generates_cc); - SetRegister(bb, instr.gpr0, value); - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/conversion.cpp b/src/video_core/shader/decode/conversion.cpp deleted file mode 100644 index fea7a54df..000000000 --- a/src/video_core/shader/decode/conversion.cpp +++ /dev/null @@ -1,321 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <limits> -#include <optional> -#include <utility> - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::Register; - -namespace { - -constexpr OperationCode GetFloatSelector(u64 selector) { - return selector == 0 ? OperationCode::FCastHalf0 : OperationCode::FCastHalf1; -} - -constexpr u32 SizeInBits(Register::Size size) { - switch (size) { - case Register::Size::Byte: - return 8; - case Register::Size::Short: - return 16; - case Register::Size::Word: - return 32; - case Register::Size::Long: - return 64; - } - return 0; -} - -constexpr std::optional<std::pair<s32, s32>> IntegerSaturateBounds(Register::Size src_size, - Register::Size dst_size, - bool src_signed, - bool dst_signed) { - const u32 dst_bits = SizeInBits(dst_size); - if (src_size == Register::Size::Word && dst_size == Register::Size::Word) { - if (src_signed == dst_signed) { - return std::nullopt; - } - return std::make_pair(0, std::numeric_limits<s32>::max()); - } - if (dst_signed) { - // Signed destination, clamp to [-128, 127] for instance - return std::make_pair(-(1 << (dst_bits - 1)), (1 << (dst_bits - 1)) - 1); - } else { - // Unsigned destination - if (dst_bits == 32) { - // Avoid shifting by 32, that is undefined behavior - return std::make_pair(0, s32(std::numeric_limits<u32>::max())); - } - return std::make_pair(0, (1 << dst_bits) - 1); - } -} - -} // Anonymous namespace - -u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - switch (opcode->get().GetId()) { - case OpCode::Id::I2I_R: - case OpCode::Id::I2I_C: - case OpCode::Id::I2I_IMM: { - const bool src_signed = instr.conversion.is_input_signed; - const bool dst_signed = instr.conversion.is_output_signed; - const Register::Size src_size = instr.conversion.src_size; - const Register::Size dst_size = instr.conversion.dst_size; - const u32 selector = static_cast<u32>(instr.conversion.int_src.selector); - - Node value = [this, instr, opcode] { - switch (opcode->get().GetId()) { - case OpCode::Id::I2I_R: - return GetRegister(instr.gpr20); - case OpCode::Id::I2I_C: - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - case OpCode::Id::I2I_IMM: - return Immediate(instr.alu.GetSignedImm20_20()); - default: - UNREACHABLE(); - return Immediate(0); - } - }(); - - // Ensure the source selector is valid - switch (instr.conversion.src_size) { - case Register::Size::Byte: - break; - case Register::Size::Short: - ASSERT(selector == 0 || selector == 2); - break; - default: - ASSERT(selector == 0); - break; - } - - if (src_size != Register::Size::Word || selector != 0) { - value = SignedOperation(OperationCode::IBitfieldExtract, src_signed, std::move(value), - Immediate(selector * 8), Immediate(SizeInBits(src_size))); - } - - value = GetOperandAbsNegInteger(std::move(value), instr.conversion.abs_a, - instr.conversion.negate_a, src_signed); - - if (instr.alu.saturate_d) { - if (src_signed && !dst_signed) { - Node is_negative = Operation(OperationCode::LogicalUGreaterEqual, value, - Immediate(1 << (SizeInBits(src_size) - 1))); - value = Operation(OperationCode::Select, std::move(is_negative), Immediate(0), - std::move(value)); - - // Simplify generated expressions, this can be removed without semantic impact - SetTemporary(bb, 0, std::move(value)); - value = GetTemporary(0); - - if (dst_size != Register::Size::Word) { - const Node limit = Immediate((1 << SizeInBits(dst_size)) - 1); - Node is_large = - Operation(OperationCode::LogicalUGreaterThan, std::move(value), limit); - value = Operation(OperationCode::Select, std::move(is_large), limit, - std::move(value)); - } - } else if (const std::optional bounds = - IntegerSaturateBounds(src_size, dst_size, src_signed, dst_signed)) { - value = SignedOperation(OperationCode::IMax, src_signed, std::move(value), - Immediate(bounds->first)); - value = SignedOperation(OperationCode::IMin, src_signed, std::move(value), - Immediate(bounds->second)); - } - } else if (dst_size != Register::Size::Word) { - // No saturation, we only have to mask the result - Node mask = Immediate((1 << SizeInBits(dst_size)) - 1); - value = Operation(OperationCode::UBitwiseAnd, std::move(value), std::move(mask)); - } - - SetInternalFlagsFromInteger(bb, value, instr.generates_cc); - SetRegister(bb, instr.gpr0, std::move(value)); - break; - } - case OpCode::Id::I2F_R: - case OpCode::Id::I2F_C: - case OpCode::Id::I2F_IMM: { - UNIMPLEMENTED_IF(instr.conversion.dst_size == Register::Size::Long); - UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "Condition codes generation in I2F is not implemented"); - - Node value = [&] { - switch (opcode->get().GetId()) { - case OpCode::Id::I2F_R: - return GetRegister(instr.gpr20); - case OpCode::Id::I2F_C: - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - case OpCode::Id::I2F_IMM: - return Immediate(instr.alu.GetSignedImm20_20()); - default: - UNREACHABLE(); - return Immediate(0); - } - }(); - - const bool input_signed = instr.conversion.is_input_signed; - - if (const u32 offset = static_cast<u32>(instr.conversion.int_src.selector); offset > 0) { - ASSERT(instr.conversion.src_size == Register::Size::Byte || - instr.conversion.src_size == Register::Size::Short); - if (instr.conversion.src_size == Register::Size::Short) { - ASSERT(offset == 0 || offset == 2); - } - value = SignedOperation(OperationCode::ILogicalShiftRight, input_signed, - std::move(value), Immediate(offset * 8)); - } - - value = ConvertIntegerSize(value, instr.conversion.src_size, input_signed); - value = GetOperandAbsNegInteger(value, instr.conversion.abs_a, false, input_signed); - value = SignedOperation(OperationCode::FCastInteger, input_signed, PRECISE, value); - value = GetOperandAbsNegFloat(value, false, instr.conversion.negate_a); - - SetInternalFlagsFromFloat(bb, value, instr.generates_cc); - - if (instr.conversion.dst_size == Register::Size::Short) { - value = Operation(OperationCode::HCastFloat, PRECISE, value); - } - - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::F2F_R: - case OpCode::Id::F2F_C: - case OpCode::Id::F2F_IMM: { - UNIMPLEMENTED_IF(instr.conversion.dst_size == Register::Size::Long); - UNIMPLEMENTED_IF(instr.conversion.src_size == Register::Size::Long); - UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "Condition codes generation in F2F is not implemented"); - - Node value = [&]() { - switch (opcode->get().GetId()) { - case OpCode::Id::F2F_R: - return GetRegister(instr.gpr20); - case OpCode::Id::F2F_C: - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - case OpCode::Id::F2F_IMM: - return GetImmediate19(instr); - default: - UNREACHABLE(); - return Immediate(0); - } - }(); - - if (instr.conversion.src_size == Register::Size::Short) { - value = Operation(GetFloatSelector(instr.conversion.float_src.selector), NO_PRECISE, - std::move(value)); - } else { - ASSERT(instr.conversion.float_src.selector == 0); - } - - value = GetOperandAbsNegFloat(value, instr.conversion.abs_a, instr.conversion.negate_a); - - value = [&] { - if (instr.conversion.src_size != instr.conversion.dst_size) { - // Rounding operations only matter when the source and destination conversion size - // is the same. - return value; - } - switch (instr.conversion.f2f.GetRoundingMode()) { - case Tegra::Shader::F2fRoundingOp::None: - return value; - case Tegra::Shader::F2fRoundingOp::Round: - return Operation(OperationCode::FRoundEven, value); - case Tegra::Shader::F2fRoundingOp::Floor: - return Operation(OperationCode::FFloor, value); - case Tegra::Shader::F2fRoundingOp::Ceil: - return Operation(OperationCode::FCeil, value); - case Tegra::Shader::F2fRoundingOp::Trunc: - return Operation(OperationCode::FTrunc, value); - default: - UNIMPLEMENTED_MSG("Unimplemented F2F rounding mode {}", - instr.conversion.f2f.rounding.Value()); - return value; - } - }(); - value = GetSaturatedFloat(value, instr.alu.saturate_d); - - SetInternalFlagsFromFloat(bb, value, instr.generates_cc); - - if (instr.conversion.dst_size == Register::Size::Short) { - value = Operation(OperationCode::HCastFloat, PRECISE, value); - } - - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::F2I_R: - case OpCode::Id::F2I_C: - case OpCode::Id::F2I_IMM: { - UNIMPLEMENTED_IF(instr.conversion.src_size == Register::Size::Long); - UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "Condition codes generation in F2I is not implemented"); - Node value = [&]() { - switch (opcode->get().GetId()) { - case OpCode::Id::F2I_R: - return GetRegister(instr.gpr20); - case OpCode::Id::F2I_C: - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - case OpCode::Id::F2I_IMM: - return GetImmediate19(instr); - default: - UNREACHABLE(); - return Immediate(0); - } - }(); - - if (instr.conversion.src_size == Register::Size::Short) { - value = Operation(GetFloatSelector(instr.conversion.float_src.selector), NO_PRECISE, - std::move(value)); - } else { - ASSERT(instr.conversion.float_src.selector == 0); - } - - value = GetOperandAbsNegFloat(value, instr.conversion.abs_a, instr.conversion.negate_a); - - value = [&]() { - switch (instr.conversion.f2i.rounding) { - case Tegra::Shader::F2iRoundingOp::RoundEven: - return Operation(OperationCode::FRoundEven, PRECISE, value); - case Tegra::Shader::F2iRoundingOp::Floor: - return Operation(OperationCode::FFloor, PRECISE, value); - case Tegra::Shader::F2iRoundingOp::Ceil: - return Operation(OperationCode::FCeil, PRECISE, value); - case Tegra::Shader::F2iRoundingOp::Trunc: - return Operation(OperationCode::FTrunc, PRECISE, value); - default: - UNIMPLEMENTED_MSG("Unimplemented F2I rounding mode {}", - instr.conversion.f2i.rounding.Value()); - return Immediate(0); - } - }(); - const bool is_signed = instr.conversion.is_output_signed; - value = SignedOperation(OperationCode::ICastFloat, is_signed, PRECISE, value); - value = ConvertIntegerSize(value, instr.conversion.dst_size, is_signed); - - SetRegister(bb, instr.gpr0, value); - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled conversion instruction: {}", opcode->get().GetName()); - } - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/ffma.cpp b/src/video_core/shader/decode/ffma.cpp deleted file mode 100644 index 5973588d6..000000000 --- a/src/video_core/shader/decode/ffma.cpp +++ /dev/null @@ -1,62 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -u32 ShaderIR::DecodeFfma(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - UNIMPLEMENTED_IF_MSG(instr.ffma.cc != 0, "FFMA cc not implemented"); - if (instr.ffma.tab5980_0 != 1) { - LOG_DEBUG(HW_GPU, "FFMA tab5980_0({}) not implemented", instr.ffma.tab5980_0.Value()); - } - if (instr.ffma.tab5980_1 != 0) { - LOG_DEBUG(HW_GPU, "FFMA tab5980_1({}) not implemented", instr.ffma.tab5980_1.Value()); - } - - const Node op_a = GetRegister(instr.gpr8); - - auto [op_b, op_c] = [&]() -> std::tuple<Node, Node> { - switch (opcode->get().GetId()) { - case OpCode::Id::FFMA_CR: { - return {GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()), - GetRegister(instr.gpr39)}; - } - case OpCode::Id::FFMA_RR: - return {GetRegister(instr.gpr20), GetRegister(instr.gpr39)}; - case OpCode::Id::FFMA_RC: { - return {GetRegister(instr.gpr39), - GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())}; - } - case OpCode::Id::FFMA_IMM: - return {GetImmediate19(instr), GetRegister(instr.gpr39)}; - default: - UNIMPLEMENTED_MSG("Unhandled FFMA instruction: {}", opcode->get().GetName()); - return {Immediate(0), Immediate(0)}; - } - }(); - - op_b = GetOperandAbsNegFloat(op_b, false, instr.ffma.negate_b); - op_c = GetOperandAbsNegFloat(op_c, false, instr.ffma.negate_c); - - Node value = Operation(OperationCode::FFma, PRECISE, op_a, op_b, op_c); - value = GetSaturatedFloat(value, instr.alu.saturate_d); - - SetInternalFlagsFromFloat(bb, value, instr.generates_cc); - SetRegister(bb, instr.gpr0, value); - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/float_set.cpp b/src/video_core/shader/decode/float_set.cpp deleted file mode 100644 index 5614e8a0d..000000000 --- a/src/video_core/shader/decode/float_set.cpp +++ /dev/null @@ -1,58 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -u32 ShaderIR::DecodeFloatSet(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - - const Node op_a = GetOperandAbsNegFloat(GetRegister(instr.gpr8), instr.fset.abs_a != 0, - instr.fset.neg_a != 0); - - Node op_b = [&]() { - if (instr.is_b_imm) { - return GetImmediate19(instr); - } else if (instr.is_b_gpr) { - return GetRegister(instr.gpr20); - } else { - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - } - }(); - - op_b = GetOperandAbsNegFloat(op_b, instr.fset.abs_b != 0, instr.fset.neg_b != 0); - - // The fset instruction sets a register to 1.0 or -1 (depending on the bf bit) if the - // condition is true, and to 0 otherwise. - const Node second_pred = GetPredicate(instr.fset.pred39, instr.fset.neg_pred != 0); - - const OperationCode combiner = GetPredicateCombiner(instr.fset.op); - const Node first_pred = GetPredicateComparisonFloat(instr.fset.cond, op_a, op_b); - - const Node predicate = Operation(combiner, first_pred, second_pred); - - const Node true_value = instr.fset.bf ? Immediate(1.0f) : Immediate(-1); - const Node false_value = instr.fset.bf ? Immediate(0.0f) : Immediate(0); - const Node value = - Operation(OperationCode::Select, PRECISE, predicate, true_value, false_value); - - if (instr.fset.bf) { - SetInternalFlagsFromFloat(bb, value, instr.generates_cc); - } else { - SetInternalFlagsFromInteger(bb, value, instr.generates_cc); - } - SetRegister(bb, instr.gpr0, value); - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/float_set_predicate.cpp b/src/video_core/shader/decode/float_set_predicate.cpp deleted file mode 100644 index 200c2c983..000000000 --- a/src/video_core/shader/decode/float_set_predicate.cpp +++ /dev/null @@ -1,57 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::Pred; - -u32 ShaderIR::DecodeFloatSetPredicate(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - - Node op_a = GetOperandAbsNegFloat(GetRegister(instr.gpr8), instr.fsetp.abs_a != 0, - instr.fsetp.neg_a != 0); - Node op_b = [&]() { - if (instr.is_b_imm) { - return GetImmediate19(instr); - } else if (instr.is_b_gpr) { - return GetRegister(instr.gpr20); - } else { - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - } - }(); - op_b = GetOperandAbsNegFloat(std::move(op_b), instr.fsetp.abs_b, instr.fsetp.neg_b); - - // We can't use the constant predicate as destination. - ASSERT(instr.fsetp.pred3 != static_cast<u64>(Pred::UnusedIndex)); - - const Node predicate = - GetPredicateComparisonFloat(instr.fsetp.cond, std::move(op_a), std::move(op_b)); - const Node second_pred = GetPredicate(instr.fsetp.pred39, instr.fsetp.neg_pred != 0); - - const OperationCode combiner = GetPredicateCombiner(instr.fsetp.op); - const Node value = Operation(combiner, predicate, second_pred); - - // Set the primary predicate to the result of Predicate OP SecondPredicate - SetPredicate(bb, instr.fsetp.pred3, value); - - if (instr.fsetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) { - // Set the secondary predicate to the result of !Predicate OP SecondPredicate, - // if enabled - const Node negated_pred = Operation(OperationCode::LogicalNegate, predicate); - const Node second_value = Operation(combiner, negated_pred, second_pred); - SetPredicate(bb, instr.fsetp.pred0, second_value); - } - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/half_set.cpp b/src/video_core/shader/decode/half_set.cpp deleted file mode 100644 index fa83108cd..000000000 --- a/src/video_core/shader/decode/half_set.cpp +++ /dev/null @@ -1,115 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <array> - -#include "common/assert.h" -#include "common/common_types.h" -#include "common/logging/log.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using std::move; -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::PredCondition; - -u32 ShaderIR::DecodeHalfSet(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - PredCondition cond{}; - bool bf = false; - bool ftz = false; - bool neg_a = false; - bool abs_a = false; - bool neg_b = false; - bool abs_b = false; - switch (opcode->get().GetId()) { - case OpCode::Id::HSET2_C: - case OpCode::Id::HSET2_IMM: - cond = instr.hsetp2.cbuf_and_imm.cond; - bf = instr.Bit(53); - ftz = instr.Bit(54); - neg_a = instr.Bit(43); - abs_a = instr.Bit(44); - neg_b = instr.Bit(56); - abs_b = instr.Bit(54); - break; - case OpCode::Id::HSET2_R: - cond = instr.hsetp2.reg.cond; - bf = instr.Bit(49); - ftz = instr.Bit(50); - neg_a = instr.Bit(43); - abs_a = instr.Bit(44); - neg_b = instr.Bit(31); - abs_b = instr.Bit(30); - break; - default: - UNREACHABLE(); - } - - Node op_b = [this, instr, opcode] { - switch (opcode->get().GetId()) { - case OpCode::Id::HSET2_C: - // Inform as unimplemented as this is not tested. - UNIMPLEMENTED_MSG("HSET2_C is not implemented"); - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - case OpCode::Id::HSET2_R: - return GetRegister(instr.gpr20); - case OpCode::Id::HSET2_IMM: - return UnpackHalfImmediate(instr, true); - default: - UNREACHABLE(); - return Node{}; - } - }(); - - if (!ftz) { - LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName()); - } - - Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hset2.type_a); - op_a = GetOperandAbsNegHalf(op_a, abs_a, neg_a); - - switch (opcode->get().GetId()) { - case OpCode::Id::HSET2_R: - op_b = GetOperandAbsNegHalf(move(op_b), abs_b, neg_b); - [[fallthrough]]; - case OpCode::Id::HSET2_C: - op_b = UnpackHalfFloat(move(op_b), instr.hset2.type_b); - break; - default: - break; - } - - Node second_pred = GetPredicate(instr.hset2.pred39, instr.hset2.neg_pred); - - Node comparison_pair = GetPredicateComparisonHalf(cond, op_a, op_b); - - const OperationCode combiner = GetPredicateCombiner(instr.hset2.op); - - // HSET2 operates on each half float in the pack. - std::array<Node, 2> values; - for (u32 i = 0; i < 2; ++i) { - const u32 raw_value = bf ? 0x3c00 : 0xffff; - Node true_value = Immediate(raw_value << (i * 16)); - Node false_value = Immediate(0); - - Node comparison = Operation(OperationCode::LogicalPick2, comparison_pair, Immediate(i)); - Node predicate = Operation(combiner, comparison, second_pred); - values[i] = - Operation(OperationCode::Select, predicate, move(true_value), move(false_value)); - } - - Node value = Operation(OperationCode::UBitwiseOr, values[0], values[1]); - SetRegister(bb, instr.gpr0, move(value)); - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/half_set_predicate.cpp b/src/video_core/shader/decode/half_set_predicate.cpp deleted file mode 100644 index 310655619..000000000 --- a/src/video_core/shader/decode/half_set_predicate.cpp +++ /dev/null @@ -1,80 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "common/logging/log.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::Pred; - -u32 ShaderIR::DecodeHalfSetPredicate(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - if (instr.hsetp2.ftz != 0) { - LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName()); - } - - Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hsetp2.type_a); - op_a = GetOperandAbsNegHalf(op_a, instr.hsetp2.abs_a, instr.hsetp2.negate_a); - - Tegra::Shader::PredCondition cond{}; - bool h_and{}; - Node op_b{}; - switch (opcode->get().GetId()) { - case OpCode::Id::HSETP2_C: - cond = instr.hsetp2.cbuf_and_imm.cond; - h_and = instr.hsetp2.cbuf_and_imm.h_and; - op_b = GetOperandAbsNegHalf(GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()), - instr.hsetp2.cbuf.abs_b, instr.hsetp2.cbuf.negate_b); - // F32 is hardcoded in hardware - op_b = UnpackHalfFloat(std::move(op_b), Tegra::Shader::HalfType::F32); - break; - case OpCode::Id::HSETP2_IMM: - cond = instr.hsetp2.cbuf_and_imm.cond; - h_and = instr.hsetp2.cbuf_and_imm.h_and; - op_b = UnpackHalfImmediate(instr, true); - break; - case OpCode::Id::HSETP2_R: - cond = instr.hsetp2.reg.cond; - h_and = instr.hsetp2.reg.h_and; - op_b = - GetOperandAbsNegHalf(UnpackHalfFloat(GetRegister(instr.gpr20), instr.hsetp2.reg.type_b), - instr.hsetp2.reg.abs_b, instr.hsetp2.reg.negate_b); - break; - default: - UNREACHABLE(); - op_b = Immediate(0); - } - - const OperationCode combiner = GetPredicateCombiner(instr.hsetp2.op); - const Node combined_pred = GetPredicate(instr.hsetp2.pred39, instr.hsetp2.neg_pred); - - const auto Write = [&](u64 dest, Node src) { - SetPredicate(bb, dest, Operation(combiner, std::move(src), combined_pred)); - }; - - const Node comparison = GetPredicateComparisonHalf(cond, op_a, op_b); - const u64 first = instr.hsetp2.pred3; - const u64 second = instr.hsetp2.pred0; - if (h_and) { - Node joined = Operation(OperationCode::LogicalAnd2, comparison); - Write(first, joined); - Write(second, Operation(OperationCode::LogicalNegate, std::move(joined))); - } else { - Write(first, Operation(OperationCode::LogicalPick2, comparison, Immediate(0U))); - Write(second, Operation(OperationCode::LogicalPick2, comparison, Immediate(1U))); - } - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/hfma2.cpp b/src/video_core/shader/decode/hfma2.cpp deleted file mode 100644 index 5b44cb79c..000000000 --- a/src/video_core/shader/decode/hfma2.cpp +++ /dev/null @@ -1,73 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <tuple> - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::HalfPrecision; -using Tegra::Shader::HalfType; -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -u32 ShaderIR::DecodeHfma2(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - if (opcode->get().GetId() == OpCode::Id::HFMA2_RR) { - DEBUG_ASSERT(instr.hfma2.rr.precision == HalfPrecision::None); - } else { - DEBUG_ASSERT(instr.hfma2.precision == HalfPrecision::None); - } - - constexpr auto identity = HalfType::H0_H1; - bool neg_b{}, neg_c{}; - auto [saturate, type_b, op_b, type_c, - op_c] = [&]() -> std::tuple<bool, HalfType, Node, HalfType, Node> { - switch (opcode->get().GetId()) { - case OpCode::Id::HFMA2_CR: - neg_b = instr.hfma2.negate_b; - neg_c = instr.hfma2.negate_c; - return {instr.hfma2.saturate, HalfType::F32, - GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()), - instr.hfma2.type_reg39, GetRegister(instr.gpr39)}; - case OpCode::Id::HFMA2_RC: - neg_b = instr.hfma2.negate_b; - neg_c = instr.hfma2.negate_c; - return {instr.hfma2.saturate, instr.hfma2.type_reg39, GetRegister(instr.gpr39), - HalfType::F32, GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())}; - case OpCode::Id::HFMA2_RR: - neg_b = instr.hfma2.rr.negate_b; - neg_c = instr.hfma2.rr.negate_c; - return {instr.hfma2.rr.saturate, instr.hfma2.type_b, GetRegister(instr.gpr20), - instr.hfma2.rr.type_c, GetRegister(instr.gpr39)}; - case OpCode::Id::HFMA2_IMM_R: - neg_c = instr.hfma2.negate_c; - return {instr.hfma2.saturate, identity, UnpackHalfImmediate(instr, true), - instr.hfma2.type_reg39, GetRegister(instr.gpr39)}; - default: - return {false, identity, Immediate(0), identity, Immediate(0)}; - } - }(); - - const Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hfma2.type_a); - op_b = GetOperandAbsNegHalf(UnpackHalfFloat(op_b, type_b), false, neg_b); - op_c = GetOperandAbsNegHalf(UnpackHalfFloat(op_c, type_c), false, neg_c); - - Node value = Operation(OperationCode::HFma, PRECISE, op_a, op_b, op_c); - value = GetSaturatedHalfFloat(value, saturate); - value = HalfMerge(GetRegister(instr.gpr0), value, instr.hfma2.merge); - - SetRegister(bb, instr.gpr0, value); - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/image.cpp b/src/video_core/shader/decode/image.cpp deleted file mode 100644 index 5470e8cf4..000000000 --- a/src/video_core/shader/decode/image.cpp +++ /dev/null @@ -1,536 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <vector> -#include <fmt/format.h> - -#include "common/assert.h" -#include "common/bit_field.h" -#include "common/common_types.h" -#include "common/logging/log.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" -#include "video_core/textures/texture.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::PredCondition; -using Tegra::Shader::StoreType; -using Tegra::Texture::ComponentType; -using Tegra::Texture::TextureFormat; -using Tegra::Texture::TICEntry; - -namespace { - -ComponentType GetComponentType(Tegra::Engines::SamplerDescriptor descriptor, - std::size_t component) { - const TextureFormat format{descriptor.format}; - switch (format) { - case TextureFormat::R16G16B16A16: - case TextureFormat::R32G32B32A32: - case TextureFormat::R32G32B32: - case TextureFormat::R32G32: - case TextureFormat::R16G16: - case TextureFormat::R32: - case TextureFormat::R16: - case TextureFormat::R8: - case TextureFormat::R1: - if (component == 0) { - return descriptor.r_type; - } - if (component == 1) { - return descriptor.g_type; - } - if (component == 2) { - return descriptor.b_type; - } - if (component == 3) { - return descriptor.a_type; - } - break; - case TextureFormat::A8R8G8B8: - if (component == 0) { - return descriptor.a_type; - } - if (component == 1) { - return descriptor.r_type; - } - if (component == 2) { - return descriptor.g_type; - } - if (component == 3) { - return descriptor.b_type; - } - break; - case TextureFormat::A2B10G10R10: - case TextureFormat::A4B4G4R4: - case TextureFormat::A5B5G5R1: - case TextureFormat::A1B5G5R5: - if (component == 0) { - return descriptor.a_type; - } - if (component == 1) { - return descriptor.b_type; - } - if (component == 2) { - return descriptor.g_type; - } - if (component == 3) { - return descriptor.r_type; - } - break; - case TextureFormat::R32_B24G8: - if (component == 0) { - return descriptor.r_type; - } - if (component == 1) { - return descriptor.b_type; - } - if (component == 2) { - return descriptor.g_type; - } - break; - case TextureFormat::B5G6R5: - case TextureFormat::B6G5R5: - case TextureFormat::B10G11R11: - if (component == 0) { - return descriptor.b_type; - } - if (component == 1) { - return descriptor.g_type; - } - if (component == 2) { - return descriptor.r_type; - } - break; - case TextureFormat::R24G8: - case TextureFormat::R8G24: - case TextureFormat::R8G8: - case TextureFormat::G4R4: - if (component == 0) { - return descriptor.g_type; - } - if (component == 1) { - return descriptor.r_type; - } - break; - default: - break; - } - UNIMPLEMENTED_MSG("Texture format not implemented={}", format); - return ComponentType::FLOAT; -} - -bool IsComponentEnabled(std::size_t component_mask, std::size_t component) { - constexpr u8 R = 0b0001; - constexpr u8 G = 0b0010; - constexpr u8 B = 0b0100; - constexpr u8 A = 0b1000; - constexpr std::array<u8, 16> mask = { - 0, (R), (G), (R | G), (B), (R | B), (G | B), (R | G | B), - (A), (R | A), (G | A), (R | G | A), (B | A), (R | B | A), (G | B | A), (R | G | B | A)}; - return std::bitset<4>{mask.at(component_mask)}.test(component); -} - -u32 GetComponentSize(TextureFormat format, std::size_t component) { - switch (format) { - case TextureFormat::R32G32B32A32: - return 32; - case TextureFormat::R16G16B16A16: - return 16; - case TextureFormat::R32G32B32: - return component <= 2 ? 32 : 0; - case TextureFormat::R32G32: - return component <= 1 ? 32 : 0; - case TextureFormat::R16G16: - return component <= 1 ? 16 : 0; - case TextureFormat::R32: - return component == 0 ? 32 : 0; - case TextureFormat::R16: - return component == 0 ? 16 : 0; - case TextureFormat::R8: - return component == 0 ? 8 : 0; - case TextureFormat::R1: - return component == 0 ? 1 : 0; - case TextureFormat::A8R8G8B8: - return 8; - case TextureFormat::A2B10G10R10: - return (component == 3 || component == 2 || component == 1) ? 10 : 2; - case TextureFormat::A4B4G4R4: - return 4; - case TextureFormat::A5B5G5R1: - return (component == 0 || component == 1 || component == 2) ? 5 : 1; - case TextureFormat::A1B5G5R5: - return (component == 1 || component == 2 || component == 3) ? 5 : 1; - case TextureFormat::R32_B24G8: - if (component == 0) { - return 32; - } - if (component == 1) { - return 24; - } - if (component == 2) { - return 8; - } - return 0; - case TextureFormat::B5G6R5: - if (component == 0 || component == 2) { - return 5; - } - if (component == 1) { - return 6; - } - return 0; - case TextureFormat::B6G5R5: - if (component == 1 || component == 2) { - return 5; - } - if (component == 0) { - return 6; - } - return 0; - case TextureFormat::B10G11R11: - if (component == 1 || component == 2) { - return 11; - } - if (component == 0) { - return 10; - } - return 0; - case TextureFormat::R24G8: - if (component == 0) { - return 8; - } - if (component == 1) { - return 24; - } - return 0; - case TextureFormat::R8G24: - if (component == 0) { - return 24; - } - if (component == 1) { - return 8; - } - return 0; - case TextureFormat::R8G8: - return (component == 0 || component == 1) ? 8 : 0; - case TextureFormat::G4R4: - return (component == 0 || component == 1) ? 4 : 0; - default: - UNIMPLEMENTED_MSG("Texture format not implemented={}", format); - return 0; - } -} - -std::size_t GetImageComponentMask(TextureFormat format) { - constexpr u8 R = 0b0001; - constexpr u8 G = 0b0010; - constexpr u8 B = 0b0100; - constexpr u8 A = 0b1000; - switch (format) { - case TextureFormat::R32G32B32A32: - case TextureFormat::R16G16B16A16: - case TextureFormat::A8R8G8B8: - case TextureFormat::A2B10G10R10: - case TextureFormat::A4B4G4R4: - case TextureFormat::A5B5G5R1: - case TextureFormat::A1B5G5R5: - return std::size_t{R | G | B | A}; - case TextureFormat::R32G32B32: - case TextureFormat::R32_B24G8: - case TextureFormat::B5G6R5: - case TextureFormat::B6G5R5: - case TextureFormat::B10G11R11: - return std::size_t{R | G | B}; - case TextureFormat::R32G32: - case TextureFormat::R16G16: - case TextureFormat::R24G8: - case TextureFormat::R8G24: - case TextureFormat::R8G8: - case TextureFormat::G4R4: - return std::size_t{R | G}; - case TextureFormat::R32: - case TextureFormat::R16: - case TextureFormat::R8: - case TextureFormat::R1: - return std::size_t{R}; - default: - UNIMPLEMENTED_MSG("Texture format not implemented={}", format); - return std::size_t{R | G | B | A}; - } -} - -std::size_t GetImageTypeNumCoordinates(Tegra::Shader::ImageType image_type) { - switch (image_type) { - case Tegra::Shader::ImageType::Texture1D: - case Tegra::Shader::ImageType::TextureBuffer: - return 1; - case Tegra::Shader::ImageType::Texture1DArray: - case Tegra::Shader::ImageType::Texture2D: - return 2; - case Tegra::Shader::ImageType::Texture2DArray: - case Tegra::Shader::ImageType::Texture3D: - return 3; - } - UNREACHABLE(); - return 1; -} -} // Anonymous namespace - -std::pair<Node, bool> ShaderIR::GetComponentValue(ComponentType component_type, u32 component_size, - Node original_value) { - switch (component_type) { - case ComponentType::SNORM: { - // range [-1.0, 1.0] - auto cnv_value = Operation(OperationCode::FMul, original_value, - Immediate(static_cast<float>(1 << component_size) / 2.f - 1.f)); - cnv_value = Operation(OperationCode::ICastFloat, std::move(cnv_value)); - return {BitfieldExtract(std::move(cnv_value), 0, component_size), true}; - } - case ComponentType::SINT: - case ComponentType::UNORM: { - bool is_signed = component_type == ComponentType::SINT; - // range [0.0, 1.0] - auto cnv_value = Operation(OperationCode::FMul, original_value, - Immediate(static_cast<float>(1 << component_size) - 1.f)); - return {SignedOperation(OperationCode::ICastFloat, is_signed, std::move(cnv_value)), - is_signed}; - } - case ComponentType::UINT: // range [0, (1 << component_size) - 1] - return {std::move(original_value), false}; - case ComponentType::FLOAT: - if (component_size == 16) { - return {Operation(OperationCode::HCastFloat, original_value), true}; - } else { - return {std::move(original_value), true}; - } - default: - UNIMPLEMENTED_MSG("Unimplemented component type={}", component_type); - return {std::move(original_value), true}; - } -} - -u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - const auto GetCoordinates = [this, instr](Tegra::Shader::ImageType image_type) { - std::vector<Node> coords; - const std::size_t num_coords{GetImageTypeNumCoordinates(image_type)}; - coords.reserve(num_coords); - for (std::size_t i = 0; i < num_coords; ++i) { - coords.push_back(GetRegister(instr.gpr8.Value() + i)); - } - return coords; - }; - - switch (opcode->get().GetId()) { - case OpCode::Id::SULD: { - UNIMPLEMENTED_IF(instr.suldst.out_of_bounds_store != - Tegra::Shader::OutOfBoundsStore::Ignore); - - const auto type{instr.suldst.image_type}; - auto& image{instr.suldst.is_immediate ? GetImage(instr.image, type) - : GetBindlessImage(instr.gpr39, type)}; - image.MarkRead(); - - if (instr.suldst.mode == Tegra::Shader::SurfaceDataMode::P) { - u32 indexer = 0; - for (u32 element = 0; element < 4; ++element) { - if (!instr.suldst.IsComponentEnabled(element)) { - continue; - } - MetaImage meta{image, {}, element}; - Node value = Operation(OperationCode::ImageLoad, meta, GetCoordinates(type)); - SetTemporary(bb, indexer++, std::move(value)); - } - for (u32 i = 0; i < indexer; ++i) { - SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i)); - } - } else if (instr.suldst.mode == Tegra::Shader::SurfaceDataMode::D_BA) { - UNIMPLEMENTED_IF(instr.suldst.GetStoreDataLayout() != StoreType::Bits32 && - instr.suldst.GetStoreDataLayout() != StoreType::Bits64); - - auto descriptor = [this, instr] { - std::optional<Tegra::Engines::SamplerDescriptor> sampler_descriptor; - if (instr.suldst.is_immediate) { - sampler_descriptor = - registry.ObtainBoundSampler(static_cast<u32>(instr.image.index.Value())); - } else { - const Node image_register = GetRegister(instr.gpr39); - const auto result = TrackCbuf(image_register, global_code, - static_cast<s64>(global_code.size())); - const auto buffer = std::get<1>(result); - const auto offset = std::get<2>(result); - sampler_descriptor = registry.ObtainBindlessSampler(buffer, offset); - } - if (!sampler_descriptor) { - UNREACHABLE_MSG("Failed to obtain image descriptor"); - } - return *sampler_descriptor; - }(); - - const auto comp_mask = GetImageComponentMask(descriptor.format); - - switch (instr.suldst.GetStoreDataLayout()) { - case StoreType::Bits32: - case StoreType::Bits64: { - u32 indexer = 0; - u32 shifted_counter = 0; - Node value = Immediate(0); - for (u32 element = 0; element < 4; ++element) { - if (!IsComponentEnabled(comp_mask, element)) { - continue; - } - const auto component_type = GetComponentType(descriptor, element); - const auto component_size = GetComponentSize(descriptor.format, element); - MetaImage meta{image, {}, element}; - - auto [converted_value, is_signed] = GetComponentValue( - component_type, component_size, - Operation(OperationCode::ImageLoad, meta, GetCoordinates(type))); - - // shift element to correct position - const auto shifted = shifted_counter; - if (shifted > 0) { - converted_value = - SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, - std::move(converted_value), Immediate(shifted)); - } - shifted_counter += component_size; - - // add value into result - value = Operation(OperationCode::UBitwiseOr, value, std::move(converted_value)); - - // if we shifted enough for 1 byte -> we save it into temp - if (shifted_counter >= 32) { - SetTemporary(bb, indexer++, std::move(value)); - // reset counter and value to prepare pack next byte - value = Immediate(0); - shifted_counter = 0; - } - } - for (u32 i = 0; i < indexer; ++i) { - SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i)); - } - break; - } - default: - UNREACHABLE(); - break; - } - } - break; - } - case OpCode::Id::SUST: { - UNIMPLEMENTED_IF(instr.suldst.mode != Tegra::Shader::SurfaceDataMode::P); - UNIMPLEMENTED_IF(instr.suldst.out_of_bounds_store != - Tegra::Shader::OutOfBoundsStore::Ignore); - UNIMPLEMENTED_IF(instr.suldst.component_mask_selector != 0xf); // Ensure we have RGBA - - std::vector<Node> values; - constexpr std::size_t hardcoded_size{4}; - for (std::size_t i = 0; i < hardcoded_size; ++i) { - values.push_back(GetRegister(instr.gpr0.Value() + i)); - } - - const auto type{instr.suldst.image_type}; - auto& image{instr.suldst.is_immediate ? GetImage(instr.image, type) - : GetBindlessImage(instr.gpr39, type)}; - image.MarkWrite(); - - MetaImage meta{image, std::move(values)}; - bb.push_back(Operation(OperationCode::ImageStore, meta, GetCoordinates(type))); - break; - } - case OpCode::Id::SUATOM: { - UNIMPLEMENTED_IF(instr.suatom_d.is_ba != 0); - - const OperationCode operation_code = [instr] { - switch (instr.suatom_d.operation_type) { - case Tegra::Shader::ImageAtomicOperationType::S32: - case Tegra::Shader::ImageAtomicOperationType::U32: - switch (instr.suatom_d.operation) { - case Tegra::Shader::ImageAtomicOperation::Add: - return OperationCode::AtomicImageAdd; - case Tegra::Shader::ImageAtomicOperation::And: - return OperationCode::AtomicImageAnd; - case Tegra::Shader::ImageAtomicOperation::Or: - return OperationCode::AtomicImageOr; - case Tegra::Shader::ImageAtomicOperation::Xor: - return OperationCode::AtomicImageXor; - case Tegra::Shader::ImageAtomicOperation::Exch: - return OperationCode::AtomicImageExchange; - default: - break; - } - break; - default: - break; - } - UNIMPLEMENTED_MSG("Unimplemented operation={}, type={}", - static_cast<u64>(instr.suatom_d.operation.Value()), - static_cast<u64>(instr.suatom_d.operation_type.Value())); - return OperationCode::AtomicImageAdd; - }(); - - Node value = GetRegister(instr.gpr0); - - const auto type = instr.suatom_d.image_type; - auto& image = GetImage(instr.image, type); - image.MarkAtomic(); - - MetaImage meta{image, {std::move(value)}}; - SetRegister(bb, instr.gpr0, Operation(operation_code, meta, GetCoordinates(type))); - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled image instruction: {}", opcode->get().GetName()); - } - - return pc; -} - -ImageEntry& ShaderIR::GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type) { - const auto offset = static_cast<u32>(image.index.Value()); - - const auto it = - std::find_if(std::begin(used_images), std::end(used_images), - [offset](const ImageEntry& entry) { return entry.offset == offset; }); - if (it != std::end(used_images)) { - ASSERT(!it->is_bindless && it->type == type); - return *it; - } - - const auto next_index = static_cast<u32>(used_images.size()); - return used_images.emplace_back(next_index, offset, type); -} - -ImageEntry& ShaderIR::GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type) { - const Node image_register = GetRegister(reg); - const auto result = - TrackCbuf(image_register, global_code, static_cast<s64>(global_code.size())); - - const auto buffer = std::get<1>(result); - const auto offset = std::get<2>(result); - - const auto it = std::find_if(std::begin(used_images), std::end(used_images), - [buffer, offset](const ImageEntry& entry) { - return entry.buffer == buffer && entry.offset == offset; - }); - if (it != std::end(used_images)) { - ASSERT(it->is_bindless && it->type == type); - return *it; - } - - const auto next_index = static_cast<u32>(used_images.size()); - return used_images.emplace_back(next_index, offset, buffer, type); -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/integer_set.cpp b/src/video_core/shader/decode/integer_set.cpp deleted file mode 100644 index 59809bcd8..000000000 --- a/src/video_core/shader/decode/integer_set.cpp +++ /dev/null @@ -1,49 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -u32 ShaderIR::DecodeIntegerSet(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - - const Node op_a = GetRegister(instr.gpr8); - const Node op_b = [&]() { - if (instr.is_b_imm) { - return Immediate(instr.alu.GetSignedImm20_20()); - } else if (instr.is_b_gpr) { - return GetRegister(instr.gpr20); - } else { - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - } - }(); - - // The iset instruction sets a register to 1.0 or -1 (depending on the bf bit) if the condition - // is true, and to 0 otherwise. - const Node second_pred = GetPredicate(instr.iset.pred39, instr.iset.neg_pred != 0); - const Node first_pred = - GetPredicateComparisonInteger(instr.iset.cond, instr.iset.is_signed, op_a, op_b); - - const OperationCode combiner = GetPredicateCombiner(instr.iset.op); - - const Node predicate = Operation(combiner, first_pred, second_pred); - - const Node true_value = instr.iset.bf ? Immediate(1.0f) : Immediate(-1); - const Node false_value = instr.iset.bf ? Immediate(0.0f) : Immediate(0); - const Node value = - Operation(OperationCode::Select, PRECISE, predicate, true_value, false_value); - - SetRegister(bb, instr.gpr0, value); - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/integer_set_predicate.cpp b/src/video_core/shader/decode/integer_set_predicate.cpp deleted file mode 100644 index 25e48fef8..000000000 --- a/src/video_core/shader/decode/integer_set_predicate.cpp +++ /dev/null @@ -1,53 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::Pred; - -u32 ShaderIR::DecodeIntegerSetPredicate(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - - const Node op_a = GetRegister(instr.gpr8); - - const Node op_b = [&]() { - if (instr.is_b_imm) { - return Immediate(instr.alu.GetSignedImm20_20()); - } else if (instr.is_b_gpr) { - return GetRegister(instr.gpr20); - } else { - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - } - }(); - - // We can't use the constant predicate as destination. - ASSERT(instr.isetp.pred3 != static_cast<u64>(Pred::UnusedIndex)); - - const Node second_pred = GetPredicate(instr.isetp.pred39, instr.isetp.neg_pred != 0); - const Node predicate = - GetPredicateComparisonInteger(instr.isetp.cond, instr.isetp.is_signed, op_a, op_b); - - // Set the primary predicate to the result of Predicate OP SecondPredicate - const OperationCode combiner = GetPredicateCombiner(instr.isetp.op); - const Node value = Operation(combiner, predicate, second_pred); - SetPredicate(bb, instr.isetp.pred3, value); - - if (instr.isetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) { - // Set the secondary predicate to the result of !Predicate OP SecondPredicate, if enabled - const Node negated_pred = Operation(OperationCode::LogicalNegate, predicate); - SetPredicate(bb, instr.isetp.pred0, Operation(combiner, negated_pred, second_pred)); - } - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp deleted file mode 100644 index 7728f600e..000000000 --- a/src/video_core/shader/decode/memory.cpp +++ /dev/null @@ -1,493 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <utility> -#include <vector> - -#include <fmt/format.h> - -#include "common/alignment.h" -#include "common/assert.h" -#include "common/common_types.h" -#include "common/logging/log.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using std::move; -using Tegra::Shader::AtomicOp; -using Tegra::Shader::AtomicType; -using Tegra::Shader::Attribute; -using Tegra::Shader::GlobalAtomicType; -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::Register; -using Tegra::Shader::StoreType; - -namespace { - -OperationCode GetAtomOperation(AtomicOp op) { - switch (op) { - case AtomicOp::Add: - return OperationCode::AtomicIAdd; - case AtomicOp::Min: - return OperationCode::AtomicIMin; - case AtomicOp::Max: - return OperationCode::AtomicIMax; - case AtomicOp::And: - return OperationCode::AtomicIAnd; - case AtomicOp::Or: - return OperationCode::AtomicIOr; - case AtomicOp::Xor: - return OperationCode::AtomicIXor; - case AtomicOp::Exch: - return OperationCode::AtomicIExchange; - default: - UNIMPLEMENTED_MSG("op={}", op); - return OperationCode::AtomicIAdd; - } -} - -bool IsUnaligned(Tegra::Shader::UniformType uniform_type) { - return uniform_type == Tegra::Shader::UniformType::UnsignedByte || - uniform_type == Tegra::Shader::UniformType::UnsignedShort; -} - -u32 GetUnalignedMask(Tegra::Shader::UniformType uniform_type) { - switch (uniform_type) { - case Tegra::Shader::UniformType::UnsignedByte: - return 0b11; - case Tegra::Shader::UniformType::UnsignedShort: - return 0b10; - default: - UNREACHABLE(); - return 0; - } -} - -u32 GetMemorySize(Tegra::Shader::UniformType uniform_type) { - switch (uniform_type) { - case Tegra::Shader::UniformType::UnsignedByte: - return 8; - case Tegra::Shader::UniformType::UnsignedShort: - return 16; - case Tegra::Shader::UniformType::Single: - return 32; - case Tegra::Shader::UniformType::Double: - return 64; - case Tegra::Shader::UniformType::Quad: - case Tegra::Shader::UniformType::UnsignedQuad: - return 128; - default: - UNIMPLEMENTED_MSG("Unimplemented size={}!", uniform_type); - return 32; - } -} - -Node ExtractUnaligned(Node value, Node address, u32 mask, u32 size) { - Node offset = Operation(OperationCode::UBitwiseAnd, address, Immediate(mask)); - offset = Operation(OperationCode::ULogicalShiftLeft, move(offset), Immediate(3)); - return Operation(OperationCode::UBitfieldExtract, move(value), move(offset), Immediate(size)); -} - -Node InsertUnaligned(Node dest, Node value, Node address, u32 mask, u32 size) { - Node offset = Operation(OperationCode::UBitwiseAnd, move(address), Immediate(mask)); - offset = Operation(OperationCode::ULogicalShiftLeft, move(offset), Immediate(3)); - return Operation(OperationCode::UBitfieldInsert, move(dest), move(value), move(offset), - Immediate(size)); -} - -Node Sign16Extend(Node value) { - Node sign = Operation(OperationCode::UBitwiseAnd, value, Immediate(1U << 15)); - Node is_sign = Operation(OperationCode::LogicalUEqual, move(sign), Immediate(1U << 15)); - Node extend = Operation(OperationCode::Select, is_sign, Immediate(0xFFFF0000), Immediate(0)); - return Operation(OperationCode::UBitwiseOr, move(value), move(extend)); -} - -} // Anonymous namespace - -u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - switch (opcode->get().GetId()) { - case OpCode::Id::LD_A: { - // Note: Shouldn't this be interp mode flat? As in no interpolation made. - UNIMPLEMENTED_IF_MSG(instr.gpr8.Value() != Register::ZeroIndex, - "Indirect attribute loads are not supported"); - UNIMPLEMENTED_IF_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) != 0, - "Unaligned attribute loads are not supported"); - UNIMPLEMENTED_IF_MSG(instr.attribute.fmt20.IsPhysical() && - instr.attribute.fmt20.size != Tegra::Shader::AttributeSize::Word, - "Non-32 bits PHYS reads are not implemented"); - - const Node buffer{GetRegister(instr.gpr39)}; - - u64 next_element = instr.attribute.fmt20.element; - auto next_index = static_cast<u64>(instr.attribute.fmt20.index.Value()); - - const auto LoadNextElement = [&](u32 reg_offset) { - const Node attribute{instr.attribute.fmt20.IsPhysical() - ? GetPhysicalInputAttribute(instr.gpr8, buffer) - : GetInputAttribute(static_cast<Attribute::Index>(next_index), - next_element, buffer)}; - - SetRegister(bb, instr.gpr0.Value() + reg_offset, attribute); - - // Load the next attribute element into the following register. If the element - // to load goes beyond the vec4 size, load the first element of the next - // attribute. - next_element = (next_element + 1) % 4; - next_index = next_index + (next_element == 0 ? 1 : 0); - }; - - const u32 num_words = static_cast<u32>(instr.attribute.fmt20.size.Value()) + 1; - for (u32 reg_offset = 0; reg_offset < num_words; ++reg_offset) { - LoadNextElement(reg_offset); - } - break; - } - case OpCode::Id::LD_C: { - UNIMPLEMENTED_IF(instr.ld_c.unknown != 0); - - Node index = GetRegister(instr.gpr8); - - const Node op_a = - GetConstBufferIndirect(instr.cbuf36.index, instr.cbuf36.GetOffset() + 0, index); - - switch (instr.ld_c.type.Value()) { - case Tegra::Shader::UniformType::Single: - SetRegister(bb, instr.gpr0, op_a); - break; - - case Tegra::Shader::UniformType::Double: { - const Node op_b = - GetConstBufferIndirect(instr.cbuf36.index, instr.cbuf36.GetOffset() + 4, index); - - SetTemporary(bb, 0, op_a); - SetTemporary(bb, 1, op_b); - SetRegister(bb, instr.gpr0, GetTemporary(0)); - SetRegister(bb, instr.gpr0.Value() + 1, GetTemporary(1)); - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled type: {}", instr.ld_c.type.Value()); - } - break; - } - case OpCode::Id::LD_L: - LOG_DEBUG(HW_GPU, "LD_L cache management mode: {}", instr.ld_l.unknown); - [[fallthrough]]; - case OpCode::Id::LD_S: { - const auto GetAddress = [&](s32 offset) { - ASSERT(offset % 4 == 0); - const Node immediate_offset = Immediate(static_cast<s32>(instr.smem_imm) + offset); - return Operation(OperationCode::IAdd, GetRegister(instr.gpr8), immediate_offset); - }; - const auto GetMemory = [&](s32 offset) { - return opcode->get().GetId() == OpCode::Id::LD_S ? GetSharedMemory(GetAddress(offset)) - : GetLocalMemory(GetAddress(offset)); - }; - - switch (instr.ldst_sl.type.Value()) { - case StoreType::Signed16: - SetRegister(bb, instr.gpr0, - Sign16Extend(ExtractUnaligned(GetMemory(0), GetAddress(0), 0b10, 16))); - break; - case StoreType::Bits32: - case StoreType::Bits64: - case StoreType::Bits128: { - const u32 count = [&] { - switch (instr.ldst_sl.type.Value()) { - case StoreType::Bits32: - return 1; - case StoreType::Bits64: - return 2; - case StoreType::Bits128: - return 4; - default: - UNREACHABLE(); - return 0; - } - }(); - for (u32 i = 0; i < count; ++i) { - SetTemporary(bb, i, GetMemory(i * 4)); - } - for (u32 i = 0; i < count; ++i) { - SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i)); - } - break; - } - default: - UNIMPLEMENTED_MSG("{} Unhandled type: {}", opcode->get().GetName(), - instr.ldst_sl.type.Value()); - } - break; - } - case OpCode::Id::LD: - case OpCode::Id::LDG: { - const auto type = [instr, &opcode]() -> Tegra::Shader::UniformType { - switch (opcode->get().GetId()) { - case OpCode::Id::LD: - UNIMPLEMENTED_IF_MSG(!instr.generic.extended, "Unextended LD is not implemented"); - return instr.generic.type; - case OpCode::Id::LDG: - return instr.ldg.type; - default: - UNREACHABLE(); - return {}; - } - }(); - - const auto [real_address_base, base_address, descriptor] = - TrackGlobalMemory(bb, instr, true, false); - - const u32 size = GetMemorySize(type); - const u32 count = Common::AlignUp(size, 32) / 32; - if (!real_address_base || !base_address) { - // Tracking failed, load zeroes. - for (u32 i = 0; i < count; ++i) { - SetRegister(bb, instr.gpr0.Value() + i, Immediate(0.0f)); - } - break; - } - - for (u32 i = 0; i < count; ++i) { - const Node it_offset = Immediate(i * 4); - const Node real_address = Operation(OperationCode::UAdd, real_address_base, it_offset); - Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor); - - // To handle unaligned loads get the bytes used to dereference global memory and extract - // those bytes from the loaded u32. - if (IsUnaligned(type)) { - gmem = ExtractUnaligned(gmem, real_address, GetUnalignedMask(type), size); - } - - SetTemporary(bb, i, gmem); - } - - for (u32 i = 0; i < count; ++i) { - SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i)); - } - break; - } - case OpCode::Id::ST_A: { - UNIMPLEMENTED_IF_MSG(instr.gpr8.Value() != Register::ZeroIndex, - "Indirect attribute loads are not supported"); - UNIMPLEMENTED_IF_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) != 0, - "Unaligned attribute loads are not supported"); - - u64 element = instr.attribute.fmt20.element; - auto index = static_cast<u64>(instr.attribute.fmt20.index.Value()); - - const u32 num_words = static_cast<u32>(instr.attribute.fmt20.size.Value()) + 1; - for (u32 reg_offset = 0; reg_offset < num_words; ++reg_offset) { - Node dest; - if (instr.attribute.fmt20.patch) { - const u32 offset = static_cast<u32>(index) * 4 + static_cast<u32>(element); - dest = MakeNode<PatchNode>(offset); - } else { - dest = GetOutputAttribute(static_cast<Attribute::Index>(index), element, - GetRegister(instr.gpr39)); - } - const auto src = GetRegister(instr.gpr0.Value() + reg_offset); - - bb.push_back(Operation(OperationCode::Assign, dest, src)); - - // Load the next attribute element into the following register. If the element to load - // goes beyond the vec4 size, load the first element of the next attribute. - element = (element + 1) % 4; - index = index + (element == 0 ? 1 : 0); - } - break; - } - case OpCode::Id::ST_L: - LOG_DEBUG(HW_GPU, "ST_L cache management mode: {}", instr.st_l.cache_management.Value()); - [[fallthrough]]; - case OpCode::Id::ST_S: { - const auto GetAddress = [&](s32 offset) { - ASSERT(offset % 4 == 0); - const Node immediate = Immediate(static_cast<s32>(instr.smem_imm) + offset); - return Operation(OperationCode::IAdd, NO_PRECISE, GetRegister(instr.gpr8), immediate); - }; - - const bool is_local = opcode->get().GetId() == OpCode::Id::ST_L; - const auto set_memory = is_local ? &ShaderIR::SetLocalMemory : &ShaderIR::SetSharedMemory; - const auto get_memory = is_local ? &ShaderIR::GetLocalMemory : &ShaderIR::GetSharedMemory; - - switch (instr.ldst_sl.type.Value()) { - case StoreType::Bits128: - (this->*set_memory)(bb, GetAddress(12), GetRegister(instr.gpr0.Value() + 3)); - (this->*set_memory)(bb, GetAddress(8), GetRegister(instr.gpr0.Value() + 2)); - [[fallthrough]]; - case StoreType::Bits64: - (this->*set_memory)(bb, GetAddress(4), GetRegister(instr.gpr0.Value() + 1)); - [[fallthrough]]; - case StoreType::Bits32: - (this->*set_memory)(bb, GetAddress(0), GetRegister(instr.gpr0)); - break; - case StoreType::Unsigned16: - case StoreType::Signed16: { - Node address = GetAddress(0); - Node memory = (this->*get_memory)(address); - (this->*set_memory)( - bb, address, InsertUnaligned(memory, GetRegister(instr.gpr0), address, 0b10, 16)); - break; - } - default: - UNIMPLEMENTED_MSG("{} unhandled type: {}", opcode->get().GetName(), - instr.ldst_sl.type.Value()); - } - break; - } - case OpCode::Id::ST: - case OpCode::Id::STG: { - const auto type = [instr, &opcode]() -> Tegra::Shader::UniformType { - switch (opcode->get().GetId()) { - case OpCode::Id::ST: - UNIMPLEMENTED_IF_MSG(!instr.generic.extended, "Unextended ST is not implemented"); - return instr.generic.type; - case OpCode::Id::STG: - return instr.stg.type; - default: - UNREACHABLE(); - return {}; - } - }(); - - // For unaligned reads we have to read memory too. - const bool is_read = IsUnaligned(type); - const auto [real_address_base, base_address, descriptor] = - TrackGlobalMemory(bb, instr, is_read, true); - if (!real_address_base || !base_address) { - // Tracking failed, skip the store. - break; - } - - const u32 size = GetMemorySize(type); - const u32 count = Common::AlignUp(size, 32) / 32; - for (u32 i = 0; i < count; ++i) { - const Node it_offset = Immediate(i * 4); - const Node real_address = Operation(OperationCode::UAdd, real_address_base, it_offset); - const Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor); - Node value = GetRegister(instr.gpr0.Value() + i); - - if (IsUnaligned(type)) { - const u32 mask = GetUnalignedMask(type); - value = InsertUnaligned(gmem, move(value), real_address, mask, size); - } - - bb.push_back(Operation(OperationCode::Assign, gmem, value)); - } - break; - } - case OpCode::Id::RED: { - UNIMPLEMENTED_IF_MSG(instr.red.type != GlobalAtomicType::U32, "type={}", - instr.red.type.Value()); - const auto [real_address, base_address, descriptor] = - TrackGlobalMemory(bb, instr, true, true); - if (!real_address || !base_address) { - // Tracking failed, skip atomic. - break; - } - Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor); - Node value = GetRegister(instr.gpr0); - bb.push_back(Operation(GetAtomOperation(instr.red.operation), move(gmem), move(value))); - break; - } - case OpCode::Id::ATOM: { - UNIMPLEMENTED_IF_MSG(instr.atom.operation == AtomicOp::Inc || - instr.atom.operation == AtomicOp::Dec || - instr.atom.operation == AtomicOp::SafeAdd, - "operation={}", instr.atom.operation.Value()); - UNIMPLEMENTED_IF_MSG(instr.atom.type == GlobalAtomicType::S64 || - instr.atom.type == GlobalAtomicType::U64 || - instr.atom.type == GlobalAtomicType::F16x2_FTZ_RN || - instr.atom.type == GlobalAtomicType::F32_FTZ_RN, - "type={}", instr.atom.type.Value()); - - const auto [real_address, base_address, descriptor] = - TrackGlobalMemory(bb, instr, true, true); - if (!real_address || !base_address) { - // Tracking failed, skip atomic. - break; - } - - const bool is_signed = - instr.atom.type == GlobalAtomicType::S32 || instr.atom.type == GlobalAtomicType::S64; - Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor); - SetRegister(bb, instr.gpr0, - SignedOperation(GetAtomOperation(instr.atom.operation), is_signed, gmem, - GetRegister(instr.gpr20))); - break; - } - case OpCode::Id::ATOMS: { - UNIMPLEMENTED_IF_MSG(instr.atoms.operation == AtomicOp::Inc || - instr.atoms.operation == AtomicOp::Dec, - "operation={}", instr.atoms.operation.Value()); - UNIMPLEMENTED_IF_MSG(instr.atoms.type == AtomicType::S64 || - instr.atoms.type == AtomicType::U64, - "type={}", instr.atoms.type.Value()); - const bool is_signed = - instr.atoms.type == AtomicType::S32 || instr.atoms.type == AtomicType::S64; - const s32 offset = instr.atoms.GetImmediateOffset(); - Node address = GetRegister(instr.gpr8); - address = Operation(OperationCode::IAdd, move(address), Immediate(offset)); - SetRegister(bb, instr.gpr0, - SignedOperation(GetAtomOperation(instr.atoms.operation), is_signed, - GetSharedMemory(move(address)), GetRegister(instr.gpr20))); - break; - } - case OpCode::Id::AL2P: { - // Ignore al2p.direction since we don't care about it. - - // Calculate emulation fake physical address. - const Node fixed_address{Immediate(static_cast<u32>(instr.al2p.address))}; - const Node reg{GetRegister(instr.gpr8)}; - const Node fake_address{Operation(OperationCode::IAdd, NO_PRECISE, reg, fixed_address)}; - - // Set the fake address to target register. - SetRegister(bb, instr.gpr0, fake_address); - - // Signal the shader IR to declare all possible attributes and varyings - uses_physical_attributes = true; - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled memory instruction: {}", opcode->get().GetName()); - } - - return pc; -} - -std::tuple<Node, Node, GlobalMemoryBase> ShaderIR::TrackGlobalMemory(NodeBlock& bb, - Instruction instr, - bool is_read, bool is_write) { - const auto addr_register{GetRegister(instr.gmem.gpr)}; - const auto immediate_offset{static_cast<u32>(instr.gmem.offset)}; - - const auto [base_address, index, offset] = - TrackCbuf(addr_register, global_code, static_cast<s64>(global_code.size())); - ASSERT_OR_EXECUTE_MSG( - base_address != nullptr, { return std::make_tuple(nullptr, nullptr, GlobalMemoryBase{}); }, - "Global memory tracking failed"); - - bb.push_back(Comment(fmt::format("Base address is c[0x{:x}][0x{:x}]", index, offset))); - - const GlobalMemoryBase descriptor{index, offset}; - const auto& entry = used_global_memory.try_emplace(descriptor).first; - auto& usage = entry->second; - usage.is_written |= is_write; - usage.is_read |= is_read; - - const auto real_address = - Operation(OperationCode::UAdd, NO_PRECISE, Immediate(immediate_offset), addr_register); - - return {real_address, base_address, descriptor}; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/other.cpp b/src/video_core/shader/decode/other.cpp deleted file mode 100644 index 5f88537bc..000000000 --- a/src/video_core/shader/decode/other.cpp +++ /dev/null @@ -1,322 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "common/logging/log.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using std::move; -using Tegra::Shader::ConditionCode; -using Tegra::Shader::Instruction; -using Tegra::Shader::IpaInterpMode; -using Tegra::Shader::OpCode; -using Tegra::Shader::PixelImap; -using Tegra::Shader::Register; -using Tegra::Shader::SystemVariable; - -using Index = Tegra::Shader::Attribute::Index; - -u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - switch (opcode->get().GetId()) { - case OpCode::Id::NOP: { - UNIMPLEMENTED_IF(instr.nop.cc != Tegra::Shader::ConditionCode::T); - UNIMPLEMENTED_IF(instr.nop.trigger != 0); - // With the previous preconditions, this instruction is a no-operation. - break; - } - case OpCode::Id::EXIT: { - const ConditionCode cc = instr.flow_condition_code; - UNIMPLEMENTED_IF_MSG(cc != ConditionCode::T, "EXIT condition code used: {}", cc); - - switch (instr.flow.cond) { - case Tegra::Shader::FlowCondition::Always: - bb.push_back(Operation(OperationCode::Exit)); - if (instr.pred.pred_index == static_cast<u64>(Pred::UnusedIndex)) { - // If this is an unconditional exit then just end processing here, - // otherwise we have to account for the possibility of the condition - // not being met, so continue processing the next instruction. - pc = MAX_PROGRAM_LENGTH - 1; - } - break; - - case Tegra::Shader::FlowCondition::Fcsm_Tr: - // TODO(bunnei): What is this used for? If we assume this conditon is not - // satisifed, dual vertex shaders in Farming Simulator make more sense - UNIMPLEMENTED_MSG("Skipping unknown FlowCondition::Fcsm_Tr"); - break; - - default: - UNIMPLEMENTED_MSG("Unhandled flow condition: {}", instr.flow.cond.Value()); - } - break; - } - case OpCode::Id::KIL: { - UNIMPLEMENTED_IF(instr.flow.cond != Tegra::Shader::FlowCondition::Always); - - const ConditionCode cc = instr.flow_condition_code; - UNIMPLEMENTED_IF_MSG(cc != ConditionCode::T, "KIL condition code used: {}", cc); - - bb.push_back(Operation(OperationCode::Discard)); - break; - } - case OpCode::Id::S2R: { - const Node value = [this, instr] { - switch (instr.sys20) { - case SystemVariable::LaneId: - return Operation(OperationCode::ThreadId); - case SystemVariable::InvocationId: - return Operation(OperationCode::InvocationId); - case SystemVariable::Ydirection: - uses_y_negate = true; - return Operation(OperationCode::YNegate); - case SystemVariable::InvocationInfo: - LOG_WARNING(HW_GPU, "S2R instruction with InvocationInfo is incomplete"); - return Immediate(0x00ff'0000U); - case SystemVariable::WscaleFactorXY: - UNIMPLEMENTED_MSG("S2R WscaleFactorXY is not implemented"); - return Immediate(0U); - case SystemVariable::WscaleFactorZ: - UNIMPLEMENTED_MSG("S2R WscaleFactorZ is not implemented"); - return Immediate(0U); - case SystemVariable::Tid: { - Node val = Immediate(0); - val = BitfieldInsert(val, Operation(OperationCode::LocalInvocationIdX), 0, 9); - val = BitfieldInsert(val, Operation(OperationCode::LocalInvocationIdY), 16, 9); - val = BitfieldInsert(val, Operation(OperationCode::LocalInvocationIdZ), 26, 5); - return val; - } - case SystemVariable::TidX: - return Operation(OperationCode::LocalInvocationIdX); - case SystemVariable::TidY: - return Operation(OperationCode::LocalInvocationIdY); - case SystemVariable::TidZ: - return Operation(OperationCode::LocalInvocationIdZ); - case SystemVariable::CtaIdX: - return Operation(OperationCode::WorkGroupIdX); - case SystemVariable::CtaIdY: - return Operation(OperationCode::WorkGroupIdY); - case SystemVariable::CtaIdZ: - return Operation(OperationCode::WorkGroupIdZ); - case SystemVariable::EqMask: - case SystemVariable::LtMask: - case SystemVariable::LeMask: - case SystemVariable::GtMask: - case SystemVariable::GeMask: - uses_warps = true; - switch (instr.sys20) { - case SystemVariable::EqMask: - return Operation(OperationCode::ThreadEqMask); - case SystemVariable::LtMask: - return Operation(OperationCode::ThreadLtMask); - case SystemVariable::LeMask: - return Operation(OperationCode::ThreadLeMask); - case SystemVariable::GtMask: - return Operation(OperationCode::ThreadGtMask); - case SystemVariable::GeMask: - return Operation(OperationCode::ThreadGeMask); - default: - UNREACHABLE(); - return Immediate(0u); - } - default: - UNIMPLEMENTED_MSG("Unhandled system move: {}", instr.sys20.Value()); - return Immediate(0u); - } - }(); - SetRegister(bb, instr.gpr0, value); - - break; - } - case OpCode::Id::BRA: { - Node branch; - if (instr.bra.constant_buffer == 0) { - const u32 target = pc + instr.bra.GetBranchTarget(); - branch = Operation(OperationCode::Branch, Immediate(target)); - } else { - const u32 target = pc + 1; - const Node op_a = GetConstBuffer(instr.cbuf36.index, instr.cbuf36.GetOffset()); - const Node convert = SignedOperation(OperationCode::IArithmeticShiftRight, true, - PRECISE, op_a, Immediate(3)); - const Node operand = - Operation(OperationCode::IAdd, PRECISE, convert, Immediate(target)); - branch = Operation(OperationCode::BranchIndirect, operand); - } - - const Tegra::Shader::ConditionCode cc = instr.flow_condition_code; - if (cc != Tegra::Shader::ConditionCode::T) { - bb.push_back(Conditional(GetConditionCode(cc), {branch})); - } else { - bb.push_back(branch); - } - break; - } - case OpCode::Id::BRX: { - Node operand; - if (instr.brx.constant_buffer != 0) { - const s32 target = pc + 1; - const Node index = GetRegister(instr.gpr8); - const Node op_a = - GetConstBufferIndirect(instr.cbuf36.index, instr.cbuf36.GetOffset() + 0, index); - const Node convert = SignedOperation(OperationCode::IArithmeticShiftRight, true, - PRECISE, op_a, Immediate(3)); - operand = Operation(OperationCode::IAdd, PRECISE, convert, Immediate(target)); - } else { - const s32 target = pc + instr.brx.GetBranchExtend(); - const Node op_a = GetRegister(instr.gpr8); - const Node convert = SignedOperation(OperationCode::IArithmeticShiftRight, true, - PRECISE, op_a, Immediate(3)); - operand = Operation(OperationCode::IAdd, PRECISE, convert, Immediate(target)); - } - const Node branch = Operation(OperationCode::BranchIndirect, operand); - - const ConditionCode cc = instr.flow_condition_code; - if (cc != ConditionCode::T) { - bb.push_back(Conditional(GetConditionCode(cc), {branch})); - } else { - bb.push_back(branch); - } - break; - } - case OpCode::Id::SSY: { - UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0, - "Constant buffer flow is not supported"); - - if (disable_flow_stack) { - break; - } - - // The SSY opcode tells the GPU where to re-converge divergent execution paths with SYNC. - const u32 target = pc + instr.bra.GetBranchTarget(); - bb.push_back( - Operation(OperationCode::PushFlowStack, MetaStackClass::Ssy, Immediate(target))); - break; - } - case OpCode::Id::PBK: { - UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0, - "Constant buffer PBK is not supported"); - - if (disable_flow_stack) { - break; - } - - // PBK pushes to a stack the address where BRK will jump to. - const u32 target = pc + instr.bra.GetBranchTarget(); - bb.push_back( - Operation(OperationCode::PushFlowStack, MetaStackClass::Pbk, Immediate(target))); - break; - } - case OpCode::Id::SYNC: { - const ConditionCode cc = instr.flow_condition_code; - UNIMPLEMENTED_IF_MSG(cc != ConditionCode::T, "SYNC condition code used: {}", cc); - - if (decompiled) { - break; - } - - // The SYNC opcode jumps to the address previously set by the SSY opcode - bb.push_back(Operation(OperationCode::PopFlowStack, MetaStackClass::Ssy)); - break; - } - case OpCode::Id::BRK: { - const ConditionCode cc = instr.flow_condition_code; - UNIMPLEMENTED_IF_MSG(cc != ConditionCode::T, "BRK condition code used: {}", cc); - if (decompiled) { - break; - } - - // The BRK opcode jumps to the address previously set by the PBK opcode - bb.push_back(Operation(OperationCode::PopFlowStack, MetaStackClass::Pbk)); - break; - } - case OpCode::Id::IPA: { - const bool is_physical = instr.ipa.idx && instr.gpr8.Value() != 0xff; - const auto attribute = instr.attribute.fmt28; - const Index index = attribute.index; - - Node value = is_physical ? GetPhysicalInputAttribute(instr.gpr8) - : GetInputAttribute(index, attribute.element); - - // Code taken from Ryujinx. - if (index >= Index::Attribute_0 && index <= Index::Attribute_31) { - const u32 location = static_cast<u32>(index) - static_cast<u32>(Index::Attribute_0); - if (header.ps.GetPixelImap(location) == PixelImap::Perspective) { - Node position_w = GetInputAttribute(Index::Position, 3); - value = Operation(OperationCode::FMul, move(value), move(position_w)); - } - } - - if (instr.ipa.interp_mode == IpaInterpMode::Multiply) { - value = Operation(OperationCode::FMul, move(value), GetRegister(instr.gpr20)); - } - - value = GetSaturatedFloat(move(value), instr.ipa.saturate); - - SetRegister(bb, instr.gpr0, move(value)); - break; - } - case OpCode::Id::OUT_R: { - UNIMPLEMENTED_IF_MSG(instr.gpr20.Value() != Register::ZeroIndex, - "Stream buffer is not supported"); - - if (instr.out.emit) { - // gpr0 is used to store the next address and gpr8 contains the address to emit. - // Hardware uses pointers here but we just ignore it - bb.push_back(Operation(OperationCode::EmitVertex)); - SetRegister(bb, instr.gpr0, Immediate(0)); - } - if (instr.out.cut) { - bb.push_back(Operation(OperationCode::EndPrimitive)); - } - break; - } - case OpCode::Id::ISBERD: { - UNIMPLEMENTED_IF(instr.isberd.o != 0); - UNIMPLEMENTED_IF(instr.isberd.skew != 0); - UNIMPLEMENTED_IF(instr.isberd.shift != Tegra::Shader::IsberdShift::None); - UNIMPLEMENTED_IF(instr.isberd.mode != Tegra::Shader::IsberdMode::None); - LOG_WARNING(HW_GPU, "ISBERD instruction is incomplete"); - SetRegister(bb, instr.gpr0, GetRegister(instr.gpr8)); - break; - } - case OpCode::Id::BAR: { - UNIMPLEMENTED_IF_MSG(instr.value != 0xF0A81B8000070000ULL, "BAR is not BAR.SYNC 0x0"); - bb.push_back(Operation(OperationCode::Barrier)); - break; - } - case OpCode::Id::MEMBAR: { - UNIMPLEMENTED_IF(instr.membar.unknown != Tegra::Shader::MembarUnknown::Default); - const OperationCode type = [instr] { - switch (instr.membar.type) { - case Tegra::Shader::MembarType::CTA: - return OperationCode::MemoryBarrierGroup; - case Tegra::Shader::MembarType::GL: - return OperationCode::MemoryBarrierGlobal; - default: - UNIMPLEMENTED_MSG("MEMBAR type={}", instr.membar.type.Value()); - return OperationCode::MemoryBarrierGlobal; - } - }(); - bb.push_back(Operation(type)); - break; - } - case OpCode::Id::DEPBAR: { - LOG_DEBUG(HW_GPU, "DEPBAR instruction is stubbed"); - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled instruction: {}", opcode->get().GetName()); - } - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/predicate_set_predicate.cpp b/src/video_core/shader/decode/predicate_set_predicate.cpp deleted file mode 100644 index 9290d22eb..000000000 --- a/src/video_core/shader/decode/predicate_set_predicate.cpp +++ /dev/null @@ -1,68 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::Pred; - -u32 ShaderIR::DecodePredicateSetPredicate(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - switch (opcode->get().GetId()) { - case OpCode::Id::PSETP: { - const Node op_a = GetPredicate(instr.psetp.pred12, instr.psetp.neg_pred12 != 0); - const Node op_b = GetPredicate(instr.psetp.pred29, instr.psetp.neg_pred29 != 0); - - // We can't use the constant predicate as destination. - ASSERT(instr.psetp.pred3 != static_cast<u64>(Pred::UnusedIndex)); - - const Node second_pred = GetPredicate(instr.psetp.pred39, instr.psetp.neg_pred39 != 0); - - const OperationCode combiner = GetPredicateCombiner(instr.psetp.op); - const Node predicate = Operation(combiner, op_a, op_b); - - // Set the primary predicate to the result of Predicate OP SecondPredicate - SetPredicate(bb, instr.psetp.pred3, Operation(combiner, predicate, second_pred)); - - if (instr.psetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) { - // Set the secondary predicate to the result of !Predicate OP SecondPredicate, if - // enabled - SetPredicate(bb, instr.psetp.pred0, - Operation(combiner, Operation(OperationCode::LogicalNegate, predicate), - second_pred)); - } - break; - } - case OpCode::Id::CSETP: { - const Node pred = GetPredicate(instr.csetp.pred39, instr.csetp.neg_pred39 != 0); - const Node condition_code = GetConditionCode(instr.csetp.cc); - - const OperationCode combiner = GetPredicateCombiner(instr.csetp.op); - - if (instr.csetp.pred3 != static_cast<u64>(Pred::UnusedIndex)) { - SetPredicate(bb, instr.csetp.pred3, Operation(combiner, condition_code, pred)); - } - if (instr.csetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) { - const Node neg_cc = Operation(OperationCode::LogicalNegate, condition_code); - SetPredicate(bb, instr.csetp.pred0, Operation(combiner, neg_cc, pred)); - } - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled predicate instruction: {}", opcode->get().GetName()); - } - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/predicate_set_register.cpp b/src/video_core/shader/decode/predicate_set_register.cpp deleted file mode 100644 index 84dbc50fe..000000000 --- a/src/video_core/shader/decode/predicate_set_register.cpp +++ /dev/null @@ -1,46 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -u32 ShaderIR::DecodePredicateSetRegister(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - - UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "Condition codes generation in PSET is not implemented"); - - const Node op_a = GetPredicate(instr.pset.pred12, instr.pset.neg_pred12 != 0); - const Node op_b = GetPredicate(instr.pset.pred29, instr.pset.neg_pred29 != 0); - const Node first_pred = Operation(GetPredicateCombiner(instr.pset.cond), op_a, op_b); - - const Node second_pred = GetPredicate(instr.pset.pred39, instr.pset.neg_pred39 != 0); - - const OperationCode combiner = GetPredicateCombiner(instr.pset.op); - const Node predicate = Operation(combiner, first_pred, second_pred); - - const Node true_value = instr.pset.bf ? Immediate(1.0f) : Immediate(0xffffffff); - const Node false_value = instr.pset.bf ? Immediate(0.0f) : Immediate(0); - const Node value = - Operation(OperationCode::Select, PRECISE, predicate, true_value, false_value); - - if (instr.pset.bf) { - SetInternalFlagsFromFloat(bb, value, instr.generates_cc); - } else { - SetInternalFlagsFromInteger(bb, value, instr.generates_cc); - } - SetRegister(bb, instr.gpr0, value); - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/register_set_predicate.cpp b/src/video_core/shader/decode/register_set_predicate.cpp deleted file mode 100644 index 6116c31aa..000000000 --- a/src/video_core/shader/decode/register_set_predicate.cpp +++ /dev/null @@ -1,86 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <utility> - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using std::move; -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; - -namespace { -constexpr u64 NUM_CONDITION_CODES = 4; -constexpr u64 NUM_PREDICATES = 7; -} // namespace - -u32 ShaderIR::DecodeRegisterSetPredicate(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - Node apply_mask = [this, opcode, instr] { - switch (opcode->get().GetId()) { - case OpCode::Id::R2P_IMM: - case OpCode::Id::P2R_IMM: - return Immediate(static_cast<u32>(instr.p2r_r2p.immediate_mask)); - default: - UNREACHABLE(); - return Immediate(0); - } - }(); - - const u32 offset = static_cast<u32>(instr.p2r_r2p.byte) * 8; - - const bool cc = instr.p2r_r2p.mode == Tegra::Shader::R2pMode::Cc; - const u64 num_entries = cc ? NUM_CONDITION_CODES : NUM_PREDICATES; - const auto get_entry = [this, cc](u64 entry) { - return cc ? GetInternalFlag(static_cast<InternalFlag>(entry)) : GetPredicate(entry); - }; - - switch (opcode->get().GetId()) { - case OpCode::Id::R2P_IMM: { - Node mask = GetRegister(instr.gpr8); - - for (u64 entry = 0; entry < num_entries; ++entry) { - const u32 shift = static_cast<u32>(entry); - - Node apply = BitfieldExtract(apply_mask, shift, 1); - Node condition = Operation(OperationCode::LogicalUNotEqual, apply, Immediate(0)); - - Node compare = BitfieldExtract(mask, offset + shift, 1); - Node value = Operation(OperationCode::LogicalUNotEqual, move(compare), Immediate(0)); - - Node code = Operation(OperationCode::LogicalAssign, get_entry(entry), move(value)); - bb.push_back(Conditional(condition, {move(code)})); - } - break; - } - case OpCode::Id::P2R_IMM: { - Node value = Immediate(0); - for (u64 entry = 0; entry < num_entries; ++entry) { - Node bit = Operation(OperationCode::Select, get_entry(entry), Immediate(1U << entry), - Immediate(0)); - value = Operation(OperationCode::UBitwiseOr, move(value), move(bit)); - } - value = Operation(OperationCode::UBitwiseAnd, move(value), apply_mask); - value = BitfieldInsert(GetRegister(instr.gpr8), move(value), offset, 8); - - SetRegister(bb, instr.gpr0, move(value)); - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled P2R/R2R instruction: {}", opcode->get().GetName()); - break; - } - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/shift.cpp b/src/video_core/shader/decode/shift.cpp deleted file mode 100644 index a53819c15..000000000 --- a/src/video_core/shader/decode/shift.cpp +++ /dev/null @@ -1,153 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using std::move; -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::ShfType; -using Tegra::Shader::ShfXmode; - -namespace { - -Node IsFull(Node shift) { - return Operation(OperationCode::LogicalIEqual, move(shift), Immediate(32)); -} - -Node Shift(OperationCode opcode, Node value, Node shift) { - Node shifted = Operation(opcode, move(value), shift); - return Operation(OperationCode::Select, IsFull(move(shift)), Immediate(0), move(shifted)); -} - -Node ClampShift(Node shift, s32 size = 32) { - shift = Operation(OperationCode::IMax, move(shift), Immediate(0)); - return Operation(OperationCode::IMin, move(shift), Immediate(size)); -} - -Node WrapShift(Node shift, s32 size = 32) { - return Operation(OperationCode::UBitwiseAnd, move(shift), Immediate(size - 1)); -} - -Node ShiftRight(Node low, Node high, Node shift, Node low_shift, ShfType type) { - // These values are used when the shift value is less than 32 - Node less_low = Shift(OperationCode::ILogicalShiftRight, low, shift); - Node less_high = Shift(OperationCode::ILogicalShiftLeft, high, low_shift); - Node less = Operation(OperationCode::IBitwiseOr, move(less_high), move(less_low)); - - if (type == ShfType::Bits32) { - // On 32 bit shifts we are either full (shifting 32) or shifting less than 32 bits - return Operation(OperationCode::Select, IsFull(move(shift)), move(high), move(less)); - } - - // And these when it's larger than or 32 - const bool is_signed = type == ShfType::S64; - const auto opcode = SignedToUnsignedCode(OperationCode::IArithmeticShiftRight, is_signed); - Node reduced = Operation(OperationCode::IAdd, shift, Immediate(-32)); - Node greater = Shift(opcode, high, move(reduced)); - - Node is_less = Operation(OperationCode::LogicalILessThan, shift, Immediate(32)); - Node is_zero = Operation(OperationCode::LogicalIEqual, move(shift), Immediate(0)); - - Node value = Operation(OperationCode::Select, move(is_less), move(less), move(greater)); - return Operation(OperationCode::Select, move(is_zero), move(high), move(value)); -} - -Node ShiftLeft(Node low, Node high, Node shift, Node low_shift, ShfType type) { - // These values are used when the shift value is less than 32 - Node less_low = Operation(OperationCode::ILogicalShiftRight, low, low_shift); - Node less_high = Operation(OperationCode::ILogicalShiftLeft, high, shift); - Node less = Operation(OperationCode::IBitwiseOr, move(less_low), move(less_high)); - - if (type == ShfType::Bits32) { - // On 32 bit shifts we are either full (shifting 32) or shifting less than 32 bits - return Operation(OperationCode::Select, IsFull(move(shift)), move(low), move(less)); - } - - // And these when it's larger than or 32 - Node reduced = Operation(OperationCode::IAdd, shift, Immediate(-32)); - Node greater = Shift(OperationCode::ILogicalShiftLeft, move(low), move(reduced)); - - Node is_less = Operation(OperationCode::LogicalILessThan, shift, Immediate(32)); - Node is_zero = Operation(OperationCode::LogicalIEqual, move(shift), Immediate(0)); - - Node value = Operation(OperationCode::Select, move(is_less), move(less), move(greater)); - return Operation(OperationCode::Select, move(is_zero), move(high), move(value)); -} - -} // Anonymous namespace - -u32 ShaderIR::DecodeShift(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - Node op_a = GetRegister(instr.gpr8); - Node op_b = [this, instr] { - if (instr.is_b_imm) { - return Immediate(instr.alu.GetSignedImm20_20()); - } else if (instr.is_b_gpr) { - return GetRegister(instr.gpr20); - } else { - return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); - } - }(); - - switch (const auto opid = opcode->get().GetId(); opid) { - case OpCode::Id::SHR_C: - case OpCode::Id::SHR_R: - case OpCode::Id::SHR_IMM: { - op_b = instr.shr.wrap ? WrapShift(move(op_b)) : ClampShift(move(op_b)); - - Node value = SignedOperation(OperationCode::IArithmeticShiftRight, instr.shift.is_signed, - move(op_a), move(op_b)); - SetInternalFlagsFromInteger(bb, value, instr.generates_cc); - SetRegister(bb, instr.gpr0, move(value)); - break; - } - case OpCode::Id::SHL_C: - case OpCode::Id::SHL_R: - case OpCode::Id::SHL_IMM: { - Node value = Operation(OperationCode::ILogicalShiftLeft, op_a, op_b); - SetInternalFlagsFromInteger(bb, value, instr.generates_cc); - SetRegister(bb, instr.gpr0, move(value)); - break; - } - case OpCode::Id::SHF_RIGHT_R: - case OpCode::Id::SHF_RIGHT_IMM: - case OpCode::Id::SHF_LEFT_R: - case OpCode::Id::SHF_LEFT_IMM: { - UNIMPLEMENTED_IF(instr.generates_cc); - UNIMPLEMENTED_IF_MSG(instr.shf.xmode != ShfXmode::None, "xmode={}", - instr.shf.xmode.Value()); - - if (instr.is_b_imm) { - op_b = Immediate(static_cast<u32>(instr.shf.immediate)); - } - const s32 size = instr.shf.type == ShfType::Bits32 ? 32 : 64; - Node shift = instr.shf.wrap ? WrapShift(move(op_b), size) : ClampShift(move(op_b), size); - - Node negated_shift = Operation(OperationCode::INegate, shift); - Node low_shift = Operation(OperationCode::IAdd, move(negated_shift), Immediate(32)); - - const bool is_right = opid == OpCode::Id::SHF_RIGHT_R || opid == OpCode::Id::SHF_RIGHT_IMM; - Node value = (is_right ? ShiftRight : ShiftLeft)( - move(op_a), GetRegister(instr.gpr39), move(shift), move(low_shift), instr.shf.type); - - SetRegister(bb, instr.gpr0, move(value)); - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled shift instruction: {}", opcode->get().GetName()); - } - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/texture.cpp b/src/video_core/shader/decode/texture.cpp deleted file mode 100644 index c69681e8d..000000000 --- a/src/video_core/shader/decode/texture.cpp +++ /dev/null @@ -1,935 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <vector> -#include <fmt/format.h> - -#include "common/assert.h" -#include "common/bit_field.h" -#include "common/common_types.h" -#include "common/logging/log.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/registry.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::Register; -using Tegra::Shader::TextureMiscMode; -using Tegra::Shader::TextureProcessMode; -using Tegra::Shader::TextureType; - -static std::size_t GetCoordCount(TextureType texture_type) { - switch (texture_type) { - case TextureType::Texture1D: - return 1; - case TextureType::Texture2D: - return 2; - case TextureType::Texture3D: - case TextureType::TextureCube: - return 3; - default: - UNIMPLEMENTED_MSG("Unhandled texture type: {}", texture_type); - return 0; - } -} - -u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - bool is_bindless = false; - switch (opcode->get().GetId()) { - case OpCode::Id::TEX: { - const TextureType texture_type{instr.tex.texture_type}; - const bool is_array = instr.tex.array != 0; - const bool is_aoffi = instr.tex.UsesMiscMode(TextureMiscMode::AOFFI); - const bool depth_compare = instr.tex.UsesMiscMode(TextureMiscMode::DC); - const auto process_mode = instr.tex.GetTextureProcessMode(); - WriteTexInstructionFloat( - bb, instr, - GetTexCode(instr, texture_type, process_mode, depth_compare, is_array, is_aoffi, {})); - break; - } - case OpCode::Id::TEX_B: { - UNIMPLEMENTED_IF_MSG(instr.tex.UsesMiscMode(TextureMiscMode::AOFFI), - "AOFFI is not implemented"); - - const TextureType texture_type{instr.tex_b.texture_type}; - const bool is_array = instr.tex_b.array != 0; - const bool is_aoffi = instr.tex.UsesMiscMode(TextureMiscMode::AOFFI); - const bool depth_compare = instr.tex_b.UsesMiscMode(TextureMiscMode::DC); - const auto process_mode = instr.tex_b.GetTextureProcessMode(); - WriteTexInstructionFloat(bb, instr, - GetTexCode(instr, texture_type, process_mode, depth_compare, - is_array, is_aoffi, {instr.gpr20})); - break; - } - case OpCode::Id::TEXS: { - const TextureType texture_type{instr.texs.GetTextureType()}; - const bool is_array{instr.texs.IsArrayTexture()}; - const bool depth_compare = instr.texs.UsesMiscMode(TextureMiscMode::DC); - const auto process_mode = instr.texs.GetTextureProcessMode(); - - const Node4 components = - GetTexsCode(instr, texture_type, process_mode, depth_compare, is_array); - - if (instr.texs.fp32_flag) { - WriteTexsInstructionFloat(bb, instr, components); - } else { - WriteTexsInstructionHalfFloat(bb, instr, components); - } - break; - } - case OpCode::Id::TLD4_B: { - is_bindless = true; - [[fallthrough]]; - } - case OpCode::Id::TLD4: { - UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(TextureMiscMode::NDV), - "NDV is not implemented"); - const auto texture_type = instr.tld4.texture_type.Value(); - const bool depth_compare = is_bindless ? instr.tld4_b.UsesMiscMode(TextureMiscMode::DC) - : instr.tld4.UsesMiscMode(TextureMiscMode::DC); - const bool is_array = instr.tld4.array != 0; - const bool is_aoffi = is_bindless ? instr.tld4_b.UsesMiscMode(TextureMiscMode::AOFFI) - : instr.tld4.UsesMiscMode(TextureMiscMode::AOFFI); - const bool is_ptp = is_bindless ? instr.tld4_b.UsesMiscMode(TextureMiscMode::PTP) - : instr.tld4.UsesMiscMode(TextureMiscMode::PTP); - WriteTexInstructionFloat(bb, instr, - GetTld4Code(instr, texture_type, depth_compare, is_array, is_aoffi, - is_ptp, is_bindless)); - break; - } - case OpCode::Id::TLD4S: { - constexpr std::size_t num_coords = 2; - const bool is_aoffi = instr.tld4s.UsesMiscMode(TextureMiscMode::AOFFI); - const bool is_depth_compare = instr.tld4s.UsesMiscMode(TextureMiscMode::DC); - const Node op_a = GetRegister(instr.gpr8); - const Node op_b = GetRegister(instr.gpr20); - - // TODO(Subv): Figure out how the sampler type is encoded in the TLD4S instruction. - std::vector<Node> coords; - std::vector<Node> aoffi; - Node depth_compare; - if (is_depth_compare) { - // Note: TLD4S coordinate encoding works just like TEXS's - const Node op_y = GetRegister(instr.gpr8.Value() + 1); - coords.push_back(op_a); - coords.push_back(op_y); - if (is_aoffi) { - aoffi = GetAoffiCoordinates(op_b, num_coords, true); - depth_compare = GetRegister(instr.gpr20.Value() + 1); - } else { - depth_compare = op_b; - } - } else { - // There's no depth compare - coords.push_back(op_a); - if (is_aoffi) { - coords.push_back(GetRegister(instr.gpr8.Value() + 1)); - aoffi = GetAoffiCoordinates(op_b, num_coords, true); - } else { - coords.push_back(op_b); - } - } - const Node component = Immediate(static_cast<u32>(instr.tld4s.component)); - - SamplerInfo info; - info.is_shadow = is_depth_compare; - const std::optional<SamplerEntry> sampler = GetSampler(instr.sampler, info); - - Node4 values; - for (u32 element = 0; element < values.size(); ++element) { - MetaTexture meta{*sampler, {}, depth_compare, aoffi, {}, {}, - {}, {}, component, element, {}}; - values[element] = Operation(OperationCode::TextureGather, meta, coords); - } - - if (instr.tld4s.fp16_flag) { - WriteTexsInstructionHalfFloat(bb, instr, values, true); - } else { - WriteTexsInstructionFloat(bb, instr, values, true); - } - break; - } - case OpCode::Id::TXD_B: - is_bindless = true; - [[fallthrough]]; - case OpCode::Id::TXD: { - UNIMPLEMENTED_IF_MSG(instr.txd.UsesMiscMode(TextureMiscMode::AOFFI), - "AOFFI is not implemented"); - - const bool is_array = instr.txd.is_array != 0; - const auto derivate_reg = instr.gpr20.Value(); - const auto texture_type = instr.txd.texture_type.Value(); - const auto coord_count = GetCoordCount(texture_type); - u64 base_reg = instr.gpr8.Value(); - Node index_var; - SamplerInfo info; - info.type = texture_type; - info.is_array = is_array; - const std::optional<SamplerEntry> sampler = - is_bindless ? GetBindlessSampler(base_reg, info, index_var) - : GetSampler(instr.sampler, info); - Node4 values; - if (!sampler) { - std::generate(values.begin(), values.end(), [this] { return Immediate(0); }); - WriteTexInstructionFloat(bb, instr, values); - break; - } - - if (is_bindless) { - base_reg++; - } - - std::vector<Node> coords; - std::vector<Node> derivates; - for (std::size_t i = 0; i < coord_count; ++i) { - coords.push_back(GetRegister(base_reg + i)); - const std::size_t derivate = i * 2; - derivates.push_back(GetRegister(derivate_reg + derivate)); - derivates.push_back(GetRegister(derivate_reg + derivate + 1)); - } - - Node array_node = {}; - if (is_array) { - const Node info_reg = GetRegister(base_reg + coord_count); - array_node = BitfieldExtract(info_reg, 0, 16); - } - - for (u32 element = 0; element < values.size(); ++element) { - MetaTexture meta{*sampler, array_node, {}, {}, {}, derivates, - {}, {}, {}, element, index_var}; - values[element] = Operation(OperationCode::TextureGradient, std::move(meta), coords); - } - - WriteTexInstructionFloat(bb, instr, values); - - break; - } - case OpCode::Id::TXQ_B: - is_bindless = true; - [[fallthrough]]; - case OpCode::Id::TXQ: { - Node index_var; - const std::optional<SamplerEntry> sampler = - is_bindless ? GetBindlessSampler(instr.gpr8, {}, index_var) - : GetSampler(instr.sampler, {}); - - if (!sampler) { - u32 indexer = 0; - for (u32 element = 0; element < 4; ++element) { - if (!instr.txq.IsComponentEnabled(element)) { - continue; - } - const Node value = Immediate(0); - SetTemporary(bb, indexer++, value); - } - for (u32 i = 0; i < indexer; ++i) { - SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i)); - } - break; - } - - u32 indexer = 0; - switch (instr.txq.query_type) { - case Tegra::Shader::TextureQueryType::Dimension: { - for (u32 element = 0; element < 4; ++element) { - if (!instr.txq.IsComponentEnabled(element)) { - continue; - } - MetaTexture meta{*sampler, {}, {}, {}, {}, {}, {}, {}, {}, element, index_var}; - const Node value = - Operation(OperationCode::TextureQueryDimensions, meta, - GetRegister(instr.gpr8.Value() + (is_bindless ? 1 : 0))); - SetTemporary(bb, indexer++, value); - } - for (u32 i = 0; i < indexer; ++i) { - SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i)); - } - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled texture query type: {}", instr.txq.query_type.Value()); - } - break; - } - case OpCode::Id::TMML_B: - is_bindless = true; - [[fallthrough]]; - case OpCode::Id::TMML: { - UNIMPLEMENTED_IF_MSG(instr.tmml.UsesMiscMode(Tegra::Shader::TextureMiscMode::NDV), - "NDV is not implemented"); - - const auto texture_type = instr.tmml.texture_type.Value(); - const bool is_array = instr.tmml.array != 0; - SamplerInfo info; - info.type = texture_type; - info.is_array = is_array; - Node index_var; - const std::optional<SamplerEntry> sampler = - is_bindless ? GetBindlessSampler(instr.gpr20, info, index_var) - : GetSampler(instr.sampler, info); - - if (!sampler) { - u32 indexer = 0; - for (u32 element = 0; element < 2; ++element) { - if (!instr.tmml.IsComponentEnabled(element)) { - continue; - } - const Node value = Immediate(0); - SetTemporary(bb, indexer++, value); - } - for (u32 i = 0; i < indexer; ++i) { - SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i)); - } - break; - } - - const u64 base_index = is_array ? 1 : 0; - const u64 num_components = [texture_type] { - switch (texture_type) { - case TextureType::Texture1D: - return 1; - case TextureType::Texture2D: - return 2; - case TextureType::TextureCube: - return 3; - default: - UNIMPLEMENTED_MSG("Unhandled texture type {}", texture_type); - return 2; - } - }(); - // TODO: What's the array component used for? - - std::vector<Node> coords; - coords.reserve(num_components); - for (u64 component = 0; component < num_components; ++component) { - coords.push_back(GetRegister(instr.gpr8.Value() + base_index + component)); - } - - u32 indexer = 0; - for (u32 element = 0; element < 2; ++element) { - if (!instr.tmml.IsComponentEnabled(element)) { - continue; - } - MetaTexture meta{*sampler, {}, {}, {}, {}, {}, {}, {}, {}, element, index_var}; - Node value = Operation(OperationCode::TextureQueryLod, meta, coords); - SetTemporary(bb, indexer++, std::move(value)); - } - for (u32 i = 0; i < indexer; ++i) { - SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i)); - } - break; - } - case OpCode::Id::TLD: { - UNIMPLEMENTED_IF_MSG(instr.tld.aoffi, "AOFFI is not implemented"); - UNIMPLEMENTED_IF_MSG(instr.tld.ms, "MS is not implemented"); - UNIMPLEMENTED_IF_MSG(instr.tld.cl, "CL is not implemented"); - - WriteTexInstructionFloat(bb, instr, GetTldCode(instr)); - break; - } - case OpCode::Id::TLDS: { - const TextureType texture_type{instr.tlds.GetTextureType()}; - const bool is_array{instr.tlds.IsArrayTexture()}; - - UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(TextureMiscMode::AOFFI), - "AOFFI is not implemented"); - UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(TextureMiscMode::MZ), "MZ is not implemented"); - - const Node4 components = GetTldsCode(instr, texture_type, is_array); - - if (instr.tlds.fp32_flag) { - WriteTexsInstructionFloat(bb, instr, components); - } else { - WriteTexsInstructionHalfFloat(bb, instr, components); - } - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled memory instruction: {}", opcode->get().GetName()); - } - - return pc; -} - -ShaderIR::SamplerInfo ShaderIR::GetSamplerInfo( - SamplerInfo info, std::optional<Tegra::Engines::SamplerDescriptor> sampler) { - if (info.IsComplete()) { - return info; - } - if (!sampler) { - LOG_WARNING(HW_GPU, "Unknown sampler info"); - info.type = info.type.value_or(Tegra::Shader::TextureType::Texture2D); - info.is_array = info.is_array.value_or(false); - info.is_shadow = info.is_shadow.value_or(false); - info.is_buffer = info.is_buffer.value_or(false); - return info; - } - info.type = info.type.value_or(sampler->texture_type); - info.is_array = info.is_array.value_or(sampler->is_array != 0); - info.is_shadow = info.is_shadow.value_or(sampler->is_shadow != 0); - info.is_buffer = info.is_buffer.value_or(sampler->is_buffer != 0); - return info; -} - -std::optional<SamplerEntry> ShaderIR::GetSampler(Tegra::Shader::Sampler sampler, - SamplerInfo sampler_info) { - const u32 offset = static_cast<u32>(sampler.index.Value()); - const auto info = GetSamplerInfo(sampler_info, registry.ObtainBoundSampler(offset)); - - // If this sampler has already been used, return the existing mapping. - const auto it = - std::find_if(used_samplers.begin(), used_samplers.end(), - [offset](const SamplerEntry& entry) { return entry.offset == offset; }); - if (it != used_samplers.end()) { - ASSERT(!it->is_bindless && it->type == info.type && it->is_array == info.is_array && - it->is_shadow == info.is_shadow && it->is_buffer == info.is_buffer); - return *it; - } - - // Otherwise create a new mapping for this sampler - const auto next_index = static_cast<u32>(used_samplers.size()); - return used_samplers.emplace_back(next_index, offset, *info.type, *info.is_array, - *info.is_shadow, *info.is_buffer, false); -} - -std::optional<SamplerEntry> ShaderIR::GetBindlessSampler(Tegra::Shader::Register reg, - SamplerInfo info, Node& index_var) { - const Node sampler_register = GetRegister(reg); - const auto [base_node, tracked_sampler_info] = - TrackBindlessSampler(sampler_register, global_code, static_cast<s64>(global_code.size())); - if (!base_node) { - UNREACHABLE(); - return std::nullopt; - } - - if (const auto sampler_info = std::get_if<BindlessSamplerNode>(&*tracked_sampler_info)) { - const u32 buffer = sampler_info->index; - const u32 offset = sampler_info->offset; - info = GetSamplerInfo(info, registry.ObtainBindlessSampler(buffer, offset)); - - // If this sampler has already been used, return the existing mapping. - const auto it = std::find_if(used_samplers.begin(), used_samplers.end(), - [buffer, offset](const SamplerEntry& entry) { - return entry.buffer == buffer && entry.offset == offset; - }); - if (it != used_samplers.end()) { - ASSERT(it->is_bindless && it->type == info.type && it->is_array == info.is_array && - it->is_shadow == info.is_shadow); - return *it; - } - - // Otherwise create a new mapping for this sampler - const auto next_index = static_cast<u32>(used_samplers.size()); - return used_samplers.emplace_back(next_index, offset, buffer, *info.type, *info.is_array, - *info.is_shadow, *info.is_buffer, false); - } - if (const auto sampler_info = std::get_if<SeparateSamplerNode>(&*tracked_sampler_info)) { - const std::pair indices = sampler_info->indices; - const std::pair offsets = sampler_info->offsets; - info = GetSamplerInfo(info, registry.ObtainSeparateSampler(indices, offsets)); - - // Try to use an already created sampler if it exists - const auto it = - std::find_if(used_samplers.begin(), used_samplers.end(), - [indices, offsets](const SamplerEntry& entry) { - return offsets == std::pair{entry.offset, entry.secondary_offset} && - indices == std::pair{entry.buffer, entry.secondary_buffer}; - }); - if (it != used_samplers.end()) { - ASSERT(it->is_separated && it->type == info.type && it->is_array == info.is_array && - it->is_shadow == info.is_shadow && it->is_buffer == info.is_buffer); - return *it; - } - - // Otherwise create a new mapping for this sampler - const u32 next_index = static_cast<u32>(used_samplers.size()); - return used_samplers.emplace_back(next_index, offsets, indices, *info.type, *info.is_array, - *info.is_shadow, *info.is_buffer); - } - if (const auto sampler_info = std::get_if<ArraySamplerNode>(&*tracked_sampler_info)) { - const u32 base_offset = sampler_info->base_offset / 4; - index_var = GetCustomVariable(sampler_info->bindless_var); - info = GetSamplerInfo(info, registry.ObtainBoundSampler(base_offset)); - - // If this sampler has already been used, return the existing mapping. - const auto it = std::find_if( - used_samplers.begin(), used_samplers.end(), - [base_offset](const SamplerEntry& entry) { return entry.offset == base_offset; }); - if (it != used_samplers.end()) { - ASSERT(!it->is_bindless && it->type == info.type && it->is_array == info.is_array && - it->is_shadow == info.is_shadow && it->is_buffer == info.is_buffer && - it->is_indexed); - return *it; - } - - uses_indexed_samplers = true; - // Otherwise create a new mapping for this sampler - const auto next_index = static_cast<u32>(used_samplers.size()); - return used_samplers.emplace_back(next_index, base_offset, *info.type, *info.is_array, - *info.is_shadow, *info.is_buffer, true); - } - return std::nullopt; -} - -void ShaderIR::WriteTexInstructionFloat(NodeBlock& bb, Instruction instr, const Node4& components) { - u32 dest_elem = 0; - for (u32 elem = 0; elem < 4; ++elem) { - if (!instr.tex.IsComponentEnabled(elem)) { - // Skip disabled components - continue; - } - SetTemporary(bb, dest_elem++, components[elem]); - } - // After writing values in temporals, move them to the real registers - for (u32 i = 0; i < dest_elem; ++i) { - SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i)); - } -} - -void ShaderIR::WriteTexsInstructionFloat(NodeBlock& bb, Instruction instr, const Node4& components, - bool ignore_mask) { - // TEXS has two destination registers and a swizzle. The first two elements in the swizzle - // go into gpr0+0 and gpr0+1, and the rest goes into gpr28+0 and gpr28+1 - - u32 dest_elem = 0; - for (u32 component = 0; component < 4; ++component) { - if (!instr.texs.IsComponentEnabled(component) && !ignore_mask) - continue; - SetTemporary(bb, dest_elem++, components[component]); - } - - for (u32 i = 0; i < dest_elem; ++i) { - if (i < 2) { - // Write the first two swizzle components to gpr0 and gpr0+1 - SetRegister(bb, instr.gpr0.Value() + i % 2, GetTemporary(i)); - } else { - ASSERT(instr.texs.HasTwoDestinations()); - // Write the rest of the swizzle components to gpr28 and gpr28+1 - SetRegister(bb, instr.gpr28.Value() + i % 2, GetTemporary(i)); - } - } -} - -void ShaderIR::WriteTexsInstructionHalfFloat(NodeBlock& bb, Instruction instr, - const Node4& components, bool ignore_mask) { - // TEXS.F16 destionation registers are packed in two registers in pairs (just like any half - // float instruction). - - Node4 values; - u32 dest_elem = 0; - for (u32 component = 0; component < 4; ++component) { - if (!instr.texs.IsComponentEnabled(component) && !ignore_mask) - continue; - values[dest_elem++] = components[component]; - } - if (dest_elem == 0) - return; - - std::generate(values.begin() + dest_elem, values.end(), [&]() { return Immediate(0); }); - - const Node first_value = Operation(OperationCode::HPack2, values[0], values[1]); - if (dest_elem <= 2) { - SetRegister(bb, instr.gpr0, first_value); - return; - } - - SetTemporary(bb, 0, first_value); - SetTemporary(bb, 1, Operation(OperationCode::HPack2, values[2], values[3])); - - SetRegister(bb, instr.gpr0, GetTemporary(0)); - SetRegister(bb, instr.gpr28, GetTemporary(1)); -} - -Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type, - TextureProcessMode process_mode, std::vector<Node> coords, - Node array, Node depth_compare, u32 bias_offset, - std::vector<Node> aoffi, - std::optional<Tegra::Shader::Register> bindless_reg) { - const bool is_array = array != nullptr; - const bool is_shadow = depth_compare != nullptr; - const bool is_bindless = bindless_reg.has_value(); - - ASSERT_MSG(texture_type != TextureType::Texture3D || !is_array || !is_shadow, - "Illegal texture type"); - - SamplerInfo info; - info.type = texture_type; - info.is_array = is_array; - info.is_shadow = is_shadow; - info.is_buffer = false; - - Node index_var; - const std::optional<SamplerEntry> sampler = - is_bindless ? GetBindlessSampler(*bindless_reg, info, index_var) - : GetSampler(instr.sampler, info); - if (!sampler) { - return {Immediate(0), Immediate(0), Immediate(0), Immediate(0)}; - } - - const bool lod_needed = process_mode == TextureProcessMode::LZ || - process_mode == TextureProcessMode::LL || - process_mode == TextureProcessMode::LLA; - const OperationCode opcode = lod_needed ? OperationCode::TextureLod : OperationCode::Texture; - - Node bias; - Node lod; - switch (process_mode) { - case TextureProcessMode::None: - break; - case TextureProcessMode::LZ: - lod = Immediate(0.0f); - break; - case TextureProcessMode::LB: - // If present, lod or bias are always stored in the register indexed by the gpr20 field with - // an offset depending on the usage of the other registers. - bias = GetRegister(instr.gpr20.Value() + bias_offset); - break; - case TextureProcessMode::LL: - lod = GetRegister(instr.gpr20.Value() + bias_offset); - break; - default: - UNIMPLEMENTED_MSG("Unimplemented process mode={}", process_mode); - break; - } - - Node4 values; - for (u32 element = 0; element < values.size(); ++element) { - MetaTexture meta{*sampler, array, depth_compare, aoffi, {}, {}, bias, - lod, {}, element, index_var}; - values[element] = Operation(opcode, meta, coords); - } - - return values; -} - -Node4 ShaderIR::GetTexCode(Instruction instr, TextureType texture_type, - TextureProcessMode process_mode, bool depth_compare, bool is_array, - bool is_aoffi, std::optional<Tegra::Shader::Register> bindless_reg) { - const bool lod_bias_enabled{ - (process_mode != TextureProcessMode::None && process_mode != TextureProcessMode::LZ)}; - - const bool is_bindless = bindless_reg.has_value(); - - u64 parameter_register = instr.gpr20.Value(); - if (is_bindless) { - ++parameter_register; - } - - const u32 bias_lod_offset = (is_bindless ? 1 : 0); - if (lod_bias_enabled) { - ++parameter_register; - } - - const auto coord_counts = ValidateAndGetCoordinateElement(texture_type, depth_compare, is_array, - lod_bias_enabled, 4, 5); - const auto coord_count = std::get<0>(coord_counts); - // If enabled arrays index is always stored in the gpr8 field - const u64 array_register = instr.gpr8.Value(); - // First coordinate index is the gpr8 or gpr8 + 1 when arrays are used - const u64 coord_register = array_register + (is_array ? 1 : 0); - - std::vector<Node> coords; - for (std::size_t i = 0; i < coord_count; ++i) { - coords.push_back(GetRegister(coord_register + i)); - } - // 1D.DC in OpenGL the 2nd component is ignored. - if (depth_compare && !is_array && texture_type == TextureType::Texture1D) { - coords.push_back(Immediate(0.0f)); - } - - const Node array = is_array ? GetRegister(array_register) : nullptr; - - std::vector<Node> aoffi; - if (is_aoffi) { - aoffi = GetAoffiCoordinates(GetRegister(parameter_register++), coord_count, false); - } - - Node dc; - if (depth_compare) { - // Depth is always stored in the register signaled by gpr20 or in the next register if lod - // or bias are used - dc = GetRegister(parameter_register++); - } - - return GetTextureCode(instr, texture_type, process_mode, coords, array, dc, bias_lod_offset, - aoffi, bindless_reg); -} - -Node4 ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type, - TextureProcessMode process_mode, bool depth_compare, bool is_array) { - const bool lod_bias_enabled = - (process_mode != TextureProcessMode::None && process_mode != TextureProcessMode::LZ); - - const auto coord_counts = ValidateAndGetCoordinateElement(texture_type, depth_compare, is_array, - lod_bias_enabled, 4, 4); - const auto coord_count = std::get<0>(coord_counts); - - // If enabled arrays index is always stored in the gpr8 field - const u64 array_register = instr.gpr8.Value(); - // First coordinate index is stored in gpr8 field or (gpr8 + 1) when arrays are used - const u64 coord_register = array_register + (is_array ? 1 : 0); - const u64 last_coord_register = - (is_array || !(lod_bias_enabled || depth_compare) || (coord_count > 2)) - ? static_cast<u64>(instr.gpr20.Value()) - : coord_register + 1; - const u32 bias_offset = coord_count > 2 ? 1 : 0; - - std::vector<Node> coords; - for (std::size_t i = 0; i < coord_count; ++i) { - const bool last = (i == (coord_count - 1)) && (coord_count > 1); - coords.push_back(GetRegister(last ? last_coord_register : coord_register + i)); - } - - const Node array = is_array ? GetRegister(array_register) : nullptr; - - Node dc; - if (depth_compare) { - // Depth is always stored in the register signaled by gpr20 or in the next register if lod - // or bias are used - const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0); - dc = GetRegister(depth_register); - } - - return GetTextureCode(instr, texture_type, process_mode, coords, array, dc, bias_offset, {}, - {}); -} - -Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool depth_compare, - bool is_array, bool is_aoffi, bool is_ptp, bool is_bindless) { - ASSERT_MSG(!(is_aoffi && is_ptp), "AOFFI and PTP can't be enabled at the same time"); - - const std::size_t coord_count = GetCoordCount(texture_type); - - // If enabled arrays index is always stored in the gpr8 field - const u64 array_register = instr.gpr8.Value(); - // First coordinate index is the gpr8 or gpr8 + 1 when arrays are used - const u64 coord_register = array_register + (is_array ? 1 : 0); - - std::vector<Node> coords; - for (std::size_t i = 0; i < coord_count; ++i) { - coords.push_back(GetRegister(coord_register + i)); - } - - u64 parameter_register = instr.gpr20.Value(); - - SamplerInfo info; - info.type = texture_type; - info.is_array = is_array; - info.is_shadow = depth_compare; - - Node index_var; - const std::optional<SamplerEntry> sampler = - is_bindless ? GetBindlessSampler(parameter_register++, info, index_var) - : GetSampler(instr.sampler, info); - Node4 values; - if (!sampler) { - for (u32 element = 0; element < values.size(); ++element) { - values[element] = Immediate(0); - } - return values; - } - - std::vector<Node> aoffi, ptp; - if (is_aoffi) { - aoffi = GetAoffiCoordinates(GetRegister(parameter_register++), coord_count, true); - } else if (is_ptp) { - ptp = GetPtpCoordinates( - {GetRegister(parameter_register++), GetRegister(parameter_register++)}); - } - - Node dc; - if (depth_compare) { - dc = GetRegister(parameter_register++); - } - - const Node component = is_bindless ? Immediate(static_cast<u32>(instr.tld4_b.component)) - : Immediate(static_cast<u32>(instr.tld4.component)); - - for (u32 element = 0; element < values.size(); ++element) { - auto coords_copy = coords; - MetaTexture meta{ - *sampler, GetRegister(array_register), dc, aoffi, ptp, {}, {}, {}, component, element, - index_var}; - values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy)); - } - - return values; -} - -Node4 ShaderIR::GetTldCode(Tegra::Shader::Instruction instr) { - const auto texture_type{instr.tld.texture_type}; - const bool is_array{instr.tld.is_array != 0}; - const bool lod_enabled{instr.tld.GetTextureProcessMode() == TextureProcessMode::LL}; - const std::size_t coord_count{GetCoordCount(texture_type)}; - - u64 gpr8_cursor{instr.gpr8.Value()}; - const Node array_register{is_array ? GetRegister(gpr8_cursor++) : nullptr}; - - std::vector<Node> coords; - coords.reserve(coord_count); - for (std::size_t i = 0; i < coord_count; ++i) { - coords.push_back(GetRegister(gpr8_cursor++)); - } - - u64 gpr20_cursor{instr.gpr20.Value()}; - // const Node bindless_register{is_bindless ? GetRegister(gpr20_cursor++) : nullptr}; - const Node lod{lod_enabled ? GetRegister(gpr20_cursor++) : Immediate(0u)}; - // const Node aoffi_register{is_aoffi ? GetRegister(gpr20_cursor++) : nullptr}; - // const Node multisample{is_multisample ? GetRegister(gpr20_cursor++) : nullptr}; - - const std::optional<SamplerEntry> sampler = GetSampler(instr.sampler, {}); - - Node4 values; - for (u32 element = 0; element < values.size(); ++element) { - auto coords_copy = coords; - MetaTexture meta{*sampler, array_register, {}, {}, {}, {}, {}, lod, {}, element, {}}; - values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy)); - } - - return values; -} - -Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is_array) { - SamplerInfo info; - info.type = texture_type; - info.is_array = is_array; - info.is_shadow = false; - const std::optional<SamplerEntry> sampler = GetSampler(instr.sampler, info); - - const std::size_t type_coord_count = GetCoordCount(texture_type); - const bool lod_enabled = instr.tlds.GetTextureProcessMode() == TextureProcessMode::LL; - const bool aoffi_enabled = instr.tlds.UsesMiscMode(TextureMiscMode::AOFFI); - - // If enabled arrays index is always stored in the gpr8 field - const u64 array_register = instr.gpr8.Value(); - // if is array gpr20 is used - const u64 coord_register = is_array ? instr.gpr20.Value() : instr.gpr8.Value(); - - const u64 last_coord_register = - ((type_coord_count > 2) || (type_coord_count == 2 && !lod_enabled)) && !is_array - ? static_cast<u64>(instr.gpr20.Value()) - : coord_register + 1; - - std::vector<Node> coords; - for (std::size_t i = 0; i < type_coord_count; ++i) { - const bool last = (i == (type_coord_count - 1)) && (type_coord_count > 1); - coords.push_back( - GetRegister(last && !aoffi_enabled ? last_coord_register : coord_register + i)); - } - - const Node array = is_array ? GetRegister(array_register) : nullptr; - // When lod is used always is in gpr20 - const Node lod = lod_enabled ? GetRegister(instr.gpr20) : Immediate(0); - - std::vector<Node> aoffi; - if (aoffi_enabled) { - aoffi = GetAoffiCoordinates(GetRegister(instr.gpr20), type_coord_count, false); - } - - Node4 values; - for (u32 element = 0; element < values.size(); ++element) { - auto coords_copy = coords; - MetaTexture meta{*sampler, array, {}, aoffi, {}, {}, {}, lod, {}, element, {}}; - values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy)); - } - return values; -} - -std::tuple<std::size_t, std::size_t> ShaderIR::ValidateAndGetCoordinateElement( - TextureType texture_type, bool depth_compare, bool is_array, bool lod_bias_enabled, - std::size_t max_coords, std::size_t max_inputs) { - const std::size_t coord_count = GetCoordCount(texture_type); - - std::size_t total_coord_count = coord_count + (is_array ? 1 : 0) + (depth_compare ? 1 : 0); - const std::size_t total_reg_count = total_coord_count + (lod_bias_enabled ? 1 : 0); - if (total_coord_count > max_coords || total_reg_count > max_inputs) { - UNIMPLEMENTED_MSG("Unsupported Texture operation"); - total_coord_count = std::min(total_coord_count, max_coords); - } - // 1D.DC OpenGL is using a vec3 but 2nd component is ignored later. - total_coord_count += - (depth_compare && !is_array && texture_type == TextureType::Texture1D) ? 1 : 0; - - return {coord_count, total_coord_count}; -} - -std::vector<Node> ShaderIR::GetAoffiCoordinates(Node aoffi_reg, std::size_t coord_count, - bool is_tld4) { - const std::array coord_offsets = is_tld4 ? std::array{0U, 8U, 16U} : std::array{0U, 4U, 8U}; - const u32 size = is_tld4 ? 6 : 4; - const s32 wrap_value = is_tld4 ? 32 : 8; - const s32 diff_value = is_tld4 ? 64 : 16; - const u32 mask = (1U << size) - 1; - - std::vector<Node> aoffi; - aoffi.reserve(coord_count); - - const auto aoffi_immediate{ - TrackImmediate(aoffi_reg, global_code, static_cast<s64>(global_code.size()))}; - if (!aoffi_immediate) { - // Variable access, not supported on AMD. - LOG_WARNING(HW_GPU, - "AOFFI constant folding failed, some hardware might have graphical issues"); - for (std::size_t coord = 0; coord < coord_count; ++coord) { - const Node value = BitfieldExtract(aoffi_reg, coord_offsets[coord], size); - const Node condition = - Operation(OperationCode::LogicalIGreaterEqual, value, Immediate(wrap_value)); - const Node negative = Operation(OperationCode::IAdd, value, Immediate(-diff_value)); - aoffi.push_back(Operation(OperationCode::Select, condition, negative, value)); - } - return aoffi; - } - - for (std::size_t coord = 0; coord < coord_count; ++coord) { - s32 value = (*aoffi_immediate >> coord_offsets[coord]) & mask; - if (value >= wrap_value) { - value -= diff_value; - } - aoffi.push_back(Immediate(value)); - } - return aoffi; -} - -std::vector<Node> ShaderIR::GetPtpCoordinates(std::array<Node, 2> ptp_regs) { - static constexpr u32 num_entries = 8; - - std::vector<Node> ptp; - ptp.reserve(num_entries); - - const auto global_size = static_cast<s64>(global_code.size()); - const std::optional low = TrackImmediate(ptp_regs[0], global_code, global_size); - const std::optional high = TrackImmediate(ptp_regs[1], global_code, global_size); - if (!low || !high) { - for (u32 entry = 0; entry < num_entries; ++entry) { - const u32 reg = entry / 4; - const u32 offset = entry % 4; - const Node value = BitfieldExtract(ptp_regs[reg], offset * 8, 6); - const Node condition = - Operation(OperationCode::LogicalIGreaterEqual, value, Immediate(32)); - const Node negative = Operation(OperationCode::IAdd, value, Immediate(-64)); - ptp.push_back(Operation(OperationCode::Select, condition, negative, value)); - } - return ptp; - } - - const u64 immediate = (static_cast<u64>(*high) << 32) | static_cast<u64>(*low); - for (u32 entry = 0; entry < num_entries; ++entry) { - s32 value = (immediate >> (entry * 8)) & 0b111111; - if (value >= 32) { - value -= 64; - } - ptp.push_back(Immediate(value)); - } - - return ptp; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/video.cpp b/src/video_core/shader/decode/video.cpp deleted file mode 100644 index 1c0957277..000000000 --- a/src/video_core/shader/decode/video.cpp +++ /dev/null @@ -1,169 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using std::move; -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::Pred; -using Tegra::Shader::VideoType; -using Tegra::Shader::VmadShr; -using Tegra::Shader::VmnmxOperation; -using Tegra::Shader::VmnmxType; - -u32 ShaderIR::DecodeVideo(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - if (opcode->get().GetId() == OpCode::Id::VMNMX) { - DecodeVMNMX(bb, instr); - return pc; - } - - const Node op_a = - GetVideoOperand(GetRegister(instr.gpr8), instr.video.is_byte_chunk_a, instr.video.signed_a, - instr.video.type_a, instr.video.byte_height_a); - const Node op_b = [this, instr] { - if (instr.video.use_register_b) { - return GetVideoOperand(GetRegister(instr.gpr20), instr.video.is_byte_chunk_b, - instr.video.signed_b, instr.video.type_b, - instr.video.byte_height_b); - } - if (instr.video.signed_b) { - const auto imm = static_cast<s16>(instr.alu.GetImm20_16()); - return Immediate(static_cast<u32>(imm)); - } else { - return Immediate(instr.alu.GetImm20_16()); - } - }(); - - switch (opcode->get().GetId()) { - case OpCode::Id::VMAD: { - const bool result_signed = instr.video.signed_a == 1 || instr.video.signed_b == 1; - const Node op_c = GetRegister(instr.gpr39); - - Node value = SignedOperation(OperationCode::IMul, result_signed, NO_PRECISE, op_a, op_b); - value = SignedOperation(OperationCode::IAdd, result_signed, NO_PRECISE, value, op_c); - - if (instr.vmad.shr == VmadShr::Shr7 || instr.vmad.shr == VmadShr::Shr15) { - const Node shift = Immediate(instr.vmad.shr == VmadShr::Shr7 ? 7 : 15); - value = - SignedOperation(OperationCode::IArithmeticShiftRight, result_signed, value, shift); - } - - SetInternalFlagsFromInteger(bb, value, instr.generates_cc); - SetRegister(bb, instr.gpr0, value); - break; - } - case OpCode::Id::VSETP: { - // We can't use the constant predicate as destination. - ASSERT(instr.vsetp.pred3 != static_cast<u64>(Pred::UnusedIndex)); - - const bool sign = instr.video.signed_a == 1 || instr.video.signed_b == 1; - const Node first_pred = GetPredicateComparisonInteger(instr.vsetp.cond, sign, op_a, op_b); - const Node second_pred = GetPredicate(instr.vsetp.pred39, false); - - const OperationCode combiner = GetPredicateCombiner(instr.vsetp.op); - - // Set the primary predicate to the result of Predicate OP SecondPredicate - SetPredicate(bb, instr.vsetp.pred3, Operation(combiner, first_pred, second_pred)); - - if (instr.vsetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) { - // Set the secondary predicate to the result of !Predicate OP SecondPredicate, - // if enabled - const Node negate_pred = Operation(OperationCode::LogicalNegate, first_pred); - SetPredicate(bb, instr.vsetp.pred0, Operation(combiner, negate_pred, second_pred)); - } - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled video instruction: {}", opcode->get().GetName()); - } - - return pc; -} - -Node ShaderIR::GetVideoOperand(Node op, bool is_chunk, bool is_signed, VideoType type, - u64 byte_height) { - if (!is_chunk) { - return BitfieldExtract(op, static_cast<u32>(byte_height * 8), 8); - } - - switch (type) { - case VideoType::Size16_Low: - return BitfieldExtract(op, 0, 16); - case VideoType::Size16_High: - return BitfieldExtract(op, 16, 16); - case VideoType::Size32: - // TODO(Rodrigo): From my hardware tests it becomes a bit "mad" when this type is used - // (1 * 1 + 0 == 0x5b800000). Until a better explanation is found: abort. - UNIMPLEMENTED(); - return Immediate(0); - case VideoType::Invalid: - UNREACHABLE_MSG("Invalid instruction encoding"); - return Immediate(0); - default: - UNREACHABLE(); - return Immediate(0); - } -} - -void ShaderIR::DecodeVMNMX(NodeBlock& bb, Tegra::Shader::Instruction instr) { - UNIMPLEMENTED_IF(!instr.vmnmx.is_op_b_register); - UNIMPLEMENTED_IF(instr.vmnmx.SourceFormatA() != VmnmxType::Bits32); - UNIMPLEMENTED_IF(instr.vmnmx.SourceFormatB() != VmnmxType::Bits32); - UNIMPLEMENTED_IF(instr.vmnmx.is_src_a_signed != instr.vmnmx.is_src_b_signed); - UNIMPLEMENTED_IF(instr.vmnmx.sat); - UNIMPLEMENTED_IF(instr.generates_cc); - - Node op_a = GetRegister(instr.gpr8); - Node op_b = GetRegister(instr.gpr20); - Node op_c = GetRegister(instr.gpr39); - - const bool is_oper1_signed = instr.vmnmx.is_src_a_signed; // Stubbed - const bool is_oper2_signed = instr.vmnmx.is_dest_signed; - - const auto operation_a = instr.vmnmx.mx ? OperationCode::IMax : OperationCode::IMin; - Node value = SignedOperation(operation_a, is_oper1_signed, move(op_a), move(op_b)); - - switch (instr.vmnmx.operation) { - case VmnmxOperation::Mrg_16H: - value = BitfieldInsert(move(op_c), move(value), 16, 16); - break; - case VmnmxOperation::Mrg_16L: - value = BitfieldInsert(move(op_c), move(value), 0, 16); - break; - case VmnmxOperation::Mrg_8B0: - value = BitfieldInsert(move(op_c), move(value), 0, 8); - break; - case VmnmxOperation::Mrg_8B2: - value = BitfieldInsert(move(op_c), move(value), 16, 8); - break; - case VmnmxOperation::Acc: - value = Operation(OperationCode::IAdd, move(value), move(op_c)); - break; - case VmnmxOperation::Min: - value = SignedOperation(OperationCode::IMin, is_oper2_signed, move(value), move(op_c)); - break; - case VmnmxOperation::Max: - value = SignedOperation(OperationCode::IMax, is_oper2_signed, move(value), move(op_c)); - break; - case VmnmxOperation::Nop: - break; - default: - UNREACHABLE(); - break; - } - - SetRegister(bb, instr.gpr0, move(value)); -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/warp.cpp b/src/video_core/shader/decode/warp.cpp deleted file mode 100644 index 37433d783..000000000 --- a/src/video_core/shader/decode/warp.cpp +++ /dev/null @@ -1,117 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::Pred; -using Tegra::Shader::ShuffleOperation; -using Tegra::Shader::VoteOperation; - -namespace { - -OperationCode GetOperationCode(VoteOperation vote_op) { - switch (vote_op) { - case VoteOperation::All: - return OperationCode::VoteAll; - case VoteOperation::Any: - return OperationCode::VoteAny; - case VoteOperation::Eq: - return OperationCode::VoteEqual; - default: - UNREACHABLE_MSG("Invalid vote operation={}", vote_op); - return OperationCode::VoteAll; - } -} - -} // Anonymous namespace - -u32 ShaderIR::DecodeWarp(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - // Signal the backend that this shader uses warp instructions. - uses_warps = true; - - switch (opcode->get().GetId()) { - case OpCode::Id::VOTE: { - const Node value = GetPredicate(instr.vote.value, instr.vote.negate_value != 0); - const Node active = Operation(OperationCode::BallotThread, value); - const Node vote = Operation(GetOperationCode(instr.vote.operation), value); - SetRegister(bb, instr.gpr0, active); - SetPredicate(bb, instr.vote.dest_pred, vote); - break; - } - case OpCode::Id::SHFL: { - Node mask = instr.shfl.is_mask_imm ? Immediate(static_cast<u32>(instr.shfl.mask_imm)) - : GetRegister(instr.gpr39); - Node index = instr.shfl.is_index_imm ? Immediate(static_cast<u32>(instr.shfl.index_imm)) - : GetRegister(instr.gpr20); - - Node thread_id = Operation(OperationCode::ThreadId); - Node clamp = Operation(OperationCode::IBitwiseAnd, mask, Immediate(0x1FU)); - Node seg_mask = BitfieldExtract(mask, 8, 16); - - Node neg_seg_mask = Operation(OperationCode::IBitwiseNot, seg_mask); - Node min_thread_id = Operation(OperationCode::IBitwiseAnd, thread_id, seg_mask); - Node max_thread_id = Operation(OperationCode::IBitwiseOr, min_thread_id, - Operation(OperationCode::IBitwiseAnd, clamp, neg_seg_mask)); - - Node src_thread_id = [instr, index, neg_seg_mask, min_thread_id, thread_id] { - switch (instr.shfl.operation) { - case ShuffleOperation::Idx: - return Operation(OperationCode::IBitwiseOr, - Operation(OperationCode::IBitwiseAnd, index, neg_seg_mask), - min_thread_id); - case ShuffleOperation::Down: - return Operation(OperationCode::IAdd, thread_id, index); - case ShuffleOperation::Up: - return Operation(OperationCode::IAdd, thread_id, - Operation(OperationCode::INegate, index)); - case ShuffleOperation::Bfly: - return Operation(OperationCode::IBitwiseXor, thread_id, index); - } - UNREACHABLE(); - return Immediate(0U); - }(); - - Node in_bounds = [instr, src_thread_id, min_thread_id, max_thread_id] { - if (instr.shfl.operation == ShuffleOperation::Up) { - return Operation(OperationCode::LogicalIGreaterEqual, src_thread_id, min_thread_id); - } else { - return Operation(OperationCode::LogicalILessEqual, src_thread_id, max_thread_id); - } - }(); - - SetPredicate(bb, instr.shfl.pred48, in_bounds); - SetRegister( - bb, instr.gpr0, - Operation(OperationCode::ShuffleIndexed, GetRegister(instr.gpr8), src_thread_id)); - break; - } - case OpCode::Id::FSWZADD: { - UNIMPLEMENTED_IF(instr.fswzadd.ndv); - - Node op_a = GetRegister(instr.gpr8); - Node op_b = GetRegister(instr.gpr20); - Node mask = Immediate(static_cast<u32>(instr.fswzadd.swizzle)); - SetRegister(bb, instr.gpr0, Operation(OperationCode::FSwizzleAdd, op_a, op_b, mask)); - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled warp instruction: {}", opcode->get().GetName()); - break; - } - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/xmad.cpp b/src/video_core/shader/decode/xmad.cpp deleted file mode 100644 index 233b8fa42..000000000 --- a/src/video_core/shader/decode/xmad.cpp +++ /dev/null @@ -1,156 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Instruction; -using Tegra::Shader::OpCode; -using Tegra::Shader::PredCondition; - -u32 ShaderIR::DecodeXmad(NodeBlock& bb, u32 pc) { - const Instruction instr = {program_code[pc]}; - const auto opcode = OpCode::Decode(instr); - - UNIMPLEMENTED_IF(instr.xmad.sign_a); - UNIMPLEMENTED_IF(instr.xmad.sign_b); - UNIMPLEMENTED_IF_MSG(instr.generates_cc, - "Condition codes generation in XMAD is not implemented"); - - Node op_a = GetRegister(instr.gpr8); - - // TODO(bunnei): Needs to be fixed once op_a or op_b is signed - UNIMPLEMENTED_IF(instr.xmad.sign_a != instr.xmad.sign_b); - const bool is_signed_a = instr.xmad.sign_a == 1; - const bool is_signed_b = instr.xmad.sign_b == 1; - const bool is_signed_c = is_signed_a; - - auto [is_merge, is_psl, is_high_b, mode, op_b_binding, - op_c] = [&]() -> std::tuple<bool, bool, bool, Tegra::Shader::XmadMode, Node, Node> { - switch (opcode->get().GetId()) { - case OpCode::Id::XMAD_CR: - return {instr.xmad.merge_56, - instr.xmad.product_shift_left_second, - instr.xmad.high_b, - instr.xmad.mode_cbf, - GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()), - GetRegister(instr.gpr39)}; - case OpCode::Id::XMAD_RR: - return {instr.xmad.merge_37, instr.xmad.product_shift_left, instr.xmad.high_b_rr, - instr.xmad.mode, GetRegister(instr.gpr20), GetRegister(instr.gpr39)}; - case OpCode::Id::XMAD_RC: - return {false, - false, - instr.xmad.high_b, - instr.xmad.mode_cbf, - GetRegister(instr.gpr39), - GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())}; - case OpCode::Id::XMAD_IMM: - return {instr.xmad.merge_37, - instr.xmad.product_shift_left, - false, - instr.xmad.mode, - Immediate(static_cast<u32>(instr.xmad.imm20_16)), - GetRegister(instr.gpr39)}; - default: - UNIMPLEMENTED_MSG("Unhandled XMAD instruction: {}", opcode->get().GetName()); - return {false, false, false, Tegra::Shader::XmadMode::None, Immediate(0), Immediate(0)}; - } - }(); - - op_a = SignedOperation(OperationCode::IBitfieldExtract, is_signed_a, std::move(op_a), - instr.xmad.high_a ? Immediate(16) : Immediate(0), Immediate(16)); - - const Node original_b = op_b_binding; - const Node op_b = - SignedOperation(OperationCode::IBitfieldExtract, is_signed_b, std::move(op_b_binding), - is_high_b ? Immediate(16) : Immediate(0), Immediate(16)); - - // we already check sign_a and sign_b is difference or not before so just use one in here. - Node product = SignedOperation(OperationCode::IMul, is_signed_a, op_a, op_b); - if (is_psl) { - product = - SignedOperation(OperationCode::ILogicalShiftLeft, is_signed_a, product, Immediate(16)); - } - SetTemporary(bb, 0, product); - product = GetTemporary(0); - - Node original_c = op_c; - const Tegra::Shader::XmadMode set_mode = mode; // Workaround to clang compile error - op_c = [&] { - switch (set_mode) { - case Tegra::Shader::XmadMode::None: - return original_c; - case Tegra::Shader::XmadMode::CLo: - return BitfieldExtract(std::move(original_c), 0, 16); - case Tegra::Shader::XmadMode::CHi: - return BitfieldExtract(std::move(original_c), 16, 16); - case Tegra::Shader::XmadMode::CBcc: { - Node shifted_b = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed_b, - original_b, Immediate(16)); - return SignedOperation(OperationCode::IAdd, is_signed_c, std::move(original_c), - std::move(shifted_b)); - } - case Tegra::Shader::XmadMode::CSfu: { - const Node comp_a = - GetPredicateComparisonInteger(PredCondition::EQ, is_signed_a, op_a, Immediate(0)); - const Node comp_b = - GetPredicateComparisonInteger(PredCondition::EQ, is_signed_b, op_b, Immediate(0)); - const Node comp = Operation(OperationCode::LogicalOr, comp_a, comp_b); - - const Node comp_minus_a = GetPredicateComparisonInteger( - PredCondition::NE, is_signed_a, - SignedOperation(OperationCode::IBitwiseAnd, is_signed_a, op_a, - Immediate(0x80000000)), - Immediate(0)); - const Node comp_minus_b = GetPredicateComparisonInteger( - PredCondition::NE, is_signed_b, - SignedOperation(OperationCode::IBitwiseAnd, is_signed_b, op_b, - Immediate(0x80000000)), - Immediate(0)); - - Node new_c = Operation( - OperationCode::Select, comp_minus_a, - SignedOperation(OperationCode::IAdd, is_signed_c, original_c, Immediate(-65536)), - original_c); - new_c = Operation( - OperationCode::Select, comp_minus_b, - SignedOperation(OperationCode::IAdd, is_signed_c, new_c, Immediate(-65536)), - std::move(new_c)); - - return Operation(OperationCode::Select, comp, original_c, std::move(new_c)); - } - default: - UNREACHABLE(); - return Immediate(0); - } - }(); - - SetTemporary(bb, 1, op_c); - op_c = GetTemporary(1); - - // TODO(Rodrigo): Use an appropiate sign for this operation - Node sum = SignedOperation(OperationCode::IAdd, is_signed_a, product, std::move(op_c)); - SetTemporary(bb, 2, sum); - sum = GetTemporary(2); - if (is_merge) { - const Node a = SignedOperation(OperationCode::IBitfieldExtract, is_signed_a, std::move(sum), - Immediate(0), Immediate(16)); - const Node b = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed_b, original_b, - Immediate(16)); - sum = SignedOperation(OperationCode::IBitwiseOr, is_signed_a, a, b); - } - - SetInternalFlagsFromInteger(bb, sum, instr.generates_cc); - SetRegister(bb, instr.gpr0, std::move(sum)); - - return pc; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/expr.cpp b/src/video_core/shader/expr.cpp deleted file mode 100644 index 2647865d4..000000000 --- a/src/video_core/shader/expr.cpp +++ /dev/null @@ -1,93 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <memory> -#include <variant> - -#include "video_core/shader/expr.h" - -namespace VideoCommon::Shader { -namespace { -bool ExprIsBoolean(const Expr& expr) { - return std::holds_alternative<ExprBoolean>(*expr); -} - -bool ExprBooleanGet(const Expr& expr) { - return std::get_if<ExprBoolean>(expr.get())->value; -} -} // Anonymous namespace - -bool ExprAnd::operator==(const ExprAnd& b) const { - return (*operand1 == *b.operand1) && (*operand2 == *b.operand2); -} - -bool ExprAnd::operator!=(const ExprAnd& b) const { - return !operator==(b); -} - -bool ExprOr::operator==(const ExprOr& b) const { - return (*operand1 == *b.operand1) && (*operand2 == *b.operand2); -} - -bool ExprOr::operator!=(const ExprOr& b) const { - return !operator==(b); -} - -bool ExprNot::operator==(const ExprNot& b) const { - return *operand1 == *b.operand1; -} - -bool ExprNot::operator!=(const ExprNot& b) const { - return !operator==(b); -} - -Expr MakeExprNot(Expr first) { - if (std::holds_alternative<ExprNot>(*first)) { - return std::get_if<ExprNot>(first.get())->operand1; - } - return MakeExpr<ExprNot>(std::move(first)); -} - -Expr MakeExprAnd(Expr first, Expr second) { - if (ExprIsBoolean(first)) { - return ExprBooleanGet(first) ? second : first; - } - if (ExprIsBoolean(second)) { - return ExprBooleanGet(second) ? first : second; - } - return MakeExpr<ExprAnd>(std::move(first), std::move(second)); -} - -Expr MakeExprOr(Expr first, Expr second) { - if (ExprIsBoolean(first)) { - return ExprBooleanGet(first) ? first : second; - } - if (ExprIsBoolean(second)) { - return ExprBooleanGet(second) ? second : first; - } - return MakeExpr<ExprOr>(std::move(first), std::move(second)); -} - -bool ExprAreEqual(const Expr& first, const Expr& second) { - return (*first) == (*second); -} - -bool ExprAreOpposite(const Expr& first, const Expr& second) { - if (std::holds_alternative<ExprNot>(*first)) { - return ExprAreEqual(std::get_if<ExprNot>(first.get())->operand1, second); - } - if (std::holds_alternative<ExprNot>(*second)) { - return ExprAreEqual(std::get_if<ExprNot>(second.get())->operand1, first); - } - return false; -} - -bool ExprIsTrue(const Expr& first) { - if (ExprIsBoolean(first)) { - return ExprBooleanGet(first); - } - return false; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/expr.h b/src/video_core/shader/expr.h deleted file mode 100644 index cda284c72..000000000 --- a/src/video_core/shader/expr.h +++ /dev/null @@ -1,156 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <memory> -#include <variant> - -#include "video_core/engines/shader_bytecode.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::ConditionCode; -using Tegra::Shader::Pred; - -class ExprAnd; -class ExprBoolean; -class ExprCondCode; -class ExprGprEqual; -class ExprNot; -class ExprOr; -class ExprPredicate; -class ExprVar; - -using ExprData = std::variant<ExprVar, ExprCondCode, ExprPredicate, ExprNot, ExprOr, ExprAnd, - ExprBoolean, ExprGprEqual>; -using Expr = std::shared_ptr<ExprData>; - -class ExprAnd final { -public: - explicit ExprAnd(Expr a, Expr b) : operand1{std::move(a)}, operand2{std::move(b)} {} - - bool operator==(const ExprAnd& b) const; - bool operator!=(const ExprAnd& b) const; - - Expr operand1; - Expr operand2; -}; - -class ExprOr final { -public: - explicit ExprOr(Expr a, Expr b) : operand1{std::move(a)}, operand2{std::move(b)} {} - - bool operator==(const ExprOr& b) const; - bool operator!=(const ExprOr& b) const; - - Expr operand1; - Expr operand2; -}; - -class ExprNot final { -public: - explicit ExprNot(Expr a) : operand1{std::move(a)} {} - - bool operator==(const ExprNot& b) const; - bool operator!=(const ExprNot& b) const; - - Expr operand1; -}; - -class ExprVar final { -public: - explicit ExprVar(u32 index) : var_index{index} {} - - bool operator==(const ExprVar& b) const { - return var_index == b.var_index; - } - - bool operator!=(const ExprVar& b) const { - return !operator==(b); - } - - u32 var_index; -}; - -class ExprPredicate final { -public: - explicit ExprPredicate(u32 predicate_) : predicate{predicate_} {} - - bool operator==(const ExprPredicate& b) const { - return predicate == b.predicate; - } - - bool operator!=(const ExprPredicate& b) const { - return !operator==(b); - } - - u32 predicate; -}; - -class ExprCondCode final { -public: - explicit ExprCondCode(ConditionCode condition_code) : cc{condition_code} {} - - bool operator==(const ExprCondCode& b) const { - return cc == b.cc; - } - - bool operator!=(const ExprCondCode& b) const { - return !operator==(b); - } - - ConditionCode cc; -}; - -class ExprBoolean final { -public: - explicit ExprBoolean(bool val) : value{val} {} - - bool operator==(const ExprBoolean& b) const { - return value == b.value; - } - - bool operator!=(const ExprBoolean& b) const { - return !operator==(b); - } - - bool value; -}; - -class ExprGprEqual final { -public: - explicit ExprGprEqual(u32 gpr_, u32 value_) : gpr{gpr_}, value{value_} {} - - bool operator==(const ExprGprEqual& b) const { - return gpr == b.gpr && value == b.value; - } - - bool operator!=(const ExprGprEqual& b) const { - return !operator==(b); - } - - u32 gpr; - u32 value; -}; - -template <typename T, typename... Args> -Expr MakeExpr(Args&&... args) { - static_assert(std::is_convertible_v<T, ExprData>); - return std::make_shared<ExprData>(T(std::forward<Args>(args)...)); -} - -bool ExprAreEqual(const Expr& first, const Expr& second); - -bool ExprAreOpposite(const Expr& first, const Expr& second); - -Expr MakeExprNot(Expr first); - -Expr MakeExprAnd(Expr first, Expr second); - -Expr MakeExprOr(Expr first, Expr second); - -bool ExprIsTrue(const Expr& first); - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/memory_util.cpp b/src/video_core/shader/memory_util.cpp deleted file mode 100644 index e18ccba8e..000000000 --- a/src/video_core/shader/memory_util.cpp +++ /dev/null @@ -1,76 +0,0 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <cstddef> - -#include <boost/container_hash/hash.hpp> - -#include "common/common_types.h" -#include "core/core.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/memory_manager.h" -#include "video_core/shader/memory_util.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -GPUVAddr GetShaderAddress(Tegra::Engines::Maxwell3D& maxwell3d, - Tegra::Engines::Maxwell3D::Regs::ShaderProgram program) { - const auto& shader_config{maxwell3d.regs.shader_config[static_cast<std::size_t>(program)]}; - return maxwell3d.regs.code_address.CodeAddress() + shader_config.offset; -} - -bool IsSchedInstruction(std::size_t offset, std::size_t main_offset) { - // Sched instructions appear once every 4 instructions. - constexpr std::size_t SchedPeriod = 4; - const std::size_t absolute_offset = offset - main_offset; - return (absolute_offset % SchedPeriod) == 0; -} - -std::size_t CalculateProgramSize(const ProgramCode& program, bool is_compute) { - // This is the encoded version of BRA that jumps to itself. All Nvidia - // shaders end with one. - static constexpr u64 SELF_JUMPING_BRANCH = 0xE2400FFFFF07000FULL; - static constexpr u64 MASK = 0xFFFFFFFFFF7FFFFFULL; - - const std::size_t start_offset = is_compute ? KERNEL_MAIN_OFFSET : STAGE_MAIN_OFFSET; - std::size_t offset = start_offset; - while (offset < program.size()) { - const u64 instruction = program[offset]; - if (!IsSchedInstruction(offset, start_offset)) { - if ((instruction & MASK) == SELF_JUMPING_BRANCH) { - // End on Maxwell's "nop" instruction - break; - } - if (instruction == 0) { - break; - } - } - ++offset; - } - // The last instruction is included in the program size - return std::min(offset + 1, program.size()); -} - -ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, GPUVAddr gpu_addr, - const u8* host_ptr, bool is_compute) { - ProgramCode code(VideoCommon::Shader::MAX_PROGRAM_LENGTH); - ASSERT_OR_EXECUTE(host_ptr != nullptr, { return code; }); - memory_manager.ReadBlockUnsafe(gpu_addr, code.data(), code.size() * sizeof(u64)); - code.resize(CalculateProgramSize(code, is_compute)); - return code; -} - -u64 GetUniqueIdentifier(Tegra::Engines::ShaderType shader_type, bool is_a, const ProgramCode& code, - const ProgramCode& code_b) { - size_t unique_identifier = boost::hash_value(code); - if (is_a) { - // VertexA programs include two programs - boost::hash_combine(unique_identifier, boost::hash_value(code_b)); - } - return static_cast<u64>(unique_identifier); -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/memory_util.h b/src/video_core/shader/memory_util.h deleted file mode 100644 index 4624d38e6..000000000 --- a/src/video_core/shader/memory_util.h +++ /dev/null @@ -1,43 +0,0 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <cstddef> -#include <vector> - -#include "common/common_types.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/shader_type.h" - -namespace Tegra { -class MemoryManager; -} - -namespace VideoCommon::Shader { - -using ProgramCode = std::vector<u64>; - -constexpr u32 STAGE_MAIN_OFFSET = 10; -constexpr u32 KERNEL_MAIN_OFFSET = 0; - -/// Gets the address for the specified shader stage program -GPUVAddr GetShaderAddress(Tegra::Engines::Maxwell3D& maxwell3d, - Tegra::Engines::Maxwell3D::Regs::ShaderProgram program); - -/// Gets if the current instruction offset is a scheduler instruction -bool IsSchedInstruction(std::size_t offset, std::size_t main_offset); - -/// Calculates the size of a program stream -std::size_t CalculateProgramSize(const ProgramCode& program, bool is_compute); - -/// Gets the shader program code from memory for the specified address -ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, GPUVAddr gpu_addr, - const u8* host_ptr, bool is_compute); - -/// Hashes one (or two) program streams -u64 GetUniqueIdentifier(Tegra::Engines::ShaderType shader_type, bool is_a, const ProgramCode& code, - const ProgramCode& code_b = {}); - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h deleted file mode 100644 index b54d33763..000000000 --- a/src/video_core/shader/node.h +++ /dev/null @@ -1,701 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <array> -#include <cstddef> -#include <memory> -#include <optional> -#include <string> -#include <tuple> -#include <utility> -#include <variant> -#include <vector> - -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" - -namespace VideoCommon::Shader { - -enum class OperationCode { - Assign, /// (float& dest, float src) -> void - - Select, /// (MetaArithmetic, bool pred, float a, float b) -> float - - FAdd, /// (MetaArithmetic, float a, float b) -> float - FMul, /// (MetaArithmetic, float a, float b) -> float - FDiv, /// (MetaArithmetic, float a, float b) -> float - FFma, /// (MetaArithmetic, float a, float b, float c) -> float - FNegate, /// (MetaArithmetic, float a) -> float - FAbsolute, /// (MetaArithmetic, float a) -> float - FClamp, /// (MetaArithmetic, float value, float min, float max) -> float - FCastHalf0, /// (MetaArithmetic, f16vec2 a) -> float - FCastHalf1, /// (MetaArithmetic, f16vec2 a) -> float - FMin, /// (MetaArithmetic, float a, float b) -> float - FMax, /// (MetaArithmetic, float a, float b) -> float - FCos, /// (MetaArithmetic, float a) -> float - FSin, /// (MetaArithmetic, float a) -> float - FExp2, /// (MetaArithmetic, float a) -> float - FLog2, /// (MetaArithmetic, float a) -> float - FInverseSqrt, /// (MetaArithmetic, float a) -> float - FSqrt, /// (MetaArithmetic, float a) -> float - FRoundEven, /// (MetaArithmetic, float a) -> float - FFloor, /// (MetaArithmetic, float a) -> float - FCeil, /// (MetaArithmetic, float a) -> float - FTrunc, /// (MetaArithmetic, float a) -> float - FCastInteger, /// (MetaArithmetic, int a) -> float - FCastUInteger, /// (MetaArithmetic, uint a) -> float - FSwizzleAdd, /// (float a, float b, uint mask) -> float - - IAdd, /// (MetaArithmetic, int a, int b) -> int - IMul, /// (MetaArithmetic, int a, int b) -> int - IDiv, /// (MetaArithmetic, int a, int b) -> int - INegate, /// (MetaArithmetic, int a) -> int - IAbsolute, /// (MetaArithmetic, int a) -> int - IMin, /// (MetaArithmetic, int a, int b) -> int - IMax, /// (MetaArithmetic, int a, int b) -> int - ICastFloat, /// (MetaArithmetic, float a) -> int - ICastUnsigned, /// (MetaArithmetic, uint a) -> int - ILogicalShiftLeft, /// (MetaArithmetic, int a, uint b) -> int - ILogicalShiftRight, /// (MetaArithmetic, int a, uint b) -> int - IArithmeticShiftRight, /// (MetaArithmetic, int a, uint b) -> int - IBitwiseAnd, /// (MetaArithmetic, int a, int b) -> int - IBitwiseOr, /// (MetaArithmetic, int a, int b) -> int - IBitwiseXor, /// (MetaArithmetic, int a, int b) -> int - IBitwiseNot, /// (MetaArithmetic, int a) -> int - IBitfieldInsert, /// (MetaArithmetic, int base, int insert, int offset, int bits) -> int - IBitfieldExtract, /// (MetaArithmetic, int value, int offset, int offset) -> int - IBitCount, /// (MetaArithmetic, int) -> int - IBitMSB, /// (MetaArithmetic, int) -> int - - UAdd, /// (MetaArithmetic, uint a, uint b) -> uint - UMul, /// (MetaArithmetic, uint a, uint b) -> uint - UDiv, /// (MetaArithmetic, uint a, uint b) -> uint - UMin, /// (MetaArithmetic, uint a, uint b) -> uint - UMax, /// (MetaArithmetic, uint a, uint b) -> uint - UCastFloat, /// (MetaArithmetic, float a) -> uint - UCastSigned, /// (MetaArithmetic, int a) -> uint - ULogicalShiftLeft, /// (MetaArithmetic, uint a, uint b) -> uint - ULogicalShiftRight, /// (MetaArithmetic, uint a, uint b) -> uint - UArithmeticShiftRight, /// (MetaArithmetic, uint a, uint b) -> uint - UBitwiseAnd, /// (MetaArithmetic, uint a, uint b) -> uint - UBitwiseOr, /// (MetaArithmetic, uint a, uint b) -> uint - UBitwiseXor, /// (MetaArithmetic, uint a, uint b) -> uint - UBitwiseNot, /// (MetaArithmetic, uint a) -> uint - UBitfieldInsert, /// (MetaArithmetic, uint base, uint insert, int offset, int bits) -> uint - UBitfieldExtract, /// (MetaArithmetic, uint value, int offset, int offset) -> uint - UBitCount, /// (MetaArithmetic, uint) -> uint - UBitMSB, /// (MetaArithmetic, uint) -> uint - - HAdd, /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2 - HMul, /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2 - HFma, /// (MetaArithmetic, f16vec2 a, f16vec2 b, f16vec2 c) -> f16vec2 - HAbsolute, /// (f16vec2 a) -> f16vec2 - HNegate, /// (f16vec2 a, bool first, bool second) -> f16vec2 - HClamp, /// (f16vec2 src, float min, float max) -> f16vec2 - HCastFloat, /// (MetaArithmetic, float a) -> f16vec2 - HUnpack, /// (Tegra::Shader::HalfType, T value) -> f16vec2 - HMergeF32, /// (f16vec2 src) -> float - HMergeH0, /// (f16vec2 dest, f16vec2 src) -> f16vec2 - HMergeH1, /// (f16vec2 dest, f16vec2 src) -> f16vec2 - HPack2, /// (float a, float b) -> f16vec2 - - LogicalAssign, /// (bool& dst, bool src) -> void - LogicalAnd, /// (bool a, bool b) -> bool - LogicalOr, /// (bool a, bool b) -> bool - LogicalXor, /// (bool a, bool b) -> bool - LogicalNegate, /// (bool a) -> bool - LogicalPick2, /// (bool2 pair, uint index) -> bool - LogicalAnd2, /// (bool2 a) -> bool - - LogicalFOrdLessThan, /// (float a, float b) -> bool - LogicalFOrdEqual, /// (float a, float b) -> bool - LogicalFOrdLessEqual, /// (float a, float b) -> bool - LogicalFOrdGreaterThan, /// (float a, float b) -> bool - LogicalFOrdNotEqual, /// (float a, float b) -> bool - LogicalFOrdGreaterEqual, /// (float a, float b) -> bool - LogicalFOrdered, /// (float a, float b) -> bool - LogicalFUnordered, /// (float a, float b) -> bool - LogicalFUnordLessThan, /// (float a, float b) -> bool - LogicalFUnordEqual, /// (float a, float b) -> bool - LogicalFUnordLessEqual, /// (float a, float b) -> bool - LogicalFUnordGreaterThan, /// (float a, float b) -> bool - LogicalFUnordNotEqual, /// (float a, float b) -> bool - LogicalFUnordGreaterEqual, /// (float a, float b) -> bool - - LogicalILessThan, /// (int a, int b) -> bool - LogicalIEqual, /// (int a, int b) -> bool - LogicalILessEqual, /// (int a, int b) -> bool - LogicalIGreaterThan, /// (int a, int b) -> bool - LogicalINotEqual, /// (int a, int b) -> bool - LogicalIGreaterEqual, /// (int a, int b) -> bool - - LogicalULessThan, /// (uint a, uint b) -> bool - LogicalUEqual, /// (uint a, uint b) -> bool - LogicalULessEqual, /// (uint a, uint b) -> bool - LogicalUGreaterThan, /// (uint a, uint b) -> bool - LogicalUNotEqual, /// (uint a, uint b) -> bool - LogicalUGreaterEqual, /// (uint a, uint b) -> bool - - LogicalAddCarry, /// (uint a, uint b) -> bool - - Logical2HLessThan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HLessEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HGreaterThan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HNotEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HGreaterEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HLessThanWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HLessEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HGreaterThanWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HNotEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HGreaterEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - - Texture, /// (MetaTexture, float[N] coords) -> float4 - TextureLod, /// (MetaTexture, float[N] coords) -> float4 - TextureGather, /// (MetaTexture, float[N] coords) -> float4 - TextureQueryDimensions, /// (MetaTexture, float a) -> float4 - TextureQueryLod, /// (MetaTexture, float[N] coords) -> float4 - TexelFetch, /// (MetaTexture, int[N], int) -> float4 - TextureGradient, /// (MetaTexture, float[N] coords, float[N*2] derivates) -> float4 - - ImageLoad, /// (MetaImage, int[N] coords) -> void - ImageStore, /// (MetaImage, int[N] coords) -> void - - AtomicImageAdd, /// (MetaImage, int[N] coords) -> void - AtomicImageAnd, /// (MetaImage, int[N] coords) -> void - AtomicImageOr, /// (MetaImage, int[N] coords) -> void - AtomicImageXor, /// (MetaImage, int[N] coords) -> void - AtomicImageExchange, /// (MetaImage, int[N] coords) -> void - - AtomicUExchange, /// (memory, uint) -> uint - AtomicUAdd, /// (memory, uint) -> uint - AtomicUMin, /// (memory, uint) -> uint - AtomicUMax, /// (memory, uint) -> uint - AtomicUAnd, /// (memory, uint) -> uint - AtomicUOr, /// (memory, uint) -> uint - AtomicUXor, /// (memory, uint) -> uint - - AtomicIExchange, /// (memory, int) -> int - AtomicIAdd, /// (memory, int) -> int - AtomicIMin, /// (memory, int) -> int - AtomicIMax, /// (memory, int) -> int - AtomicIAnd, /// (memory, int) -> int - AtomicIOr, /// (memory, int) -> int - AtomicIXor, /// (memory, int) -> int - - ReduceUAdd, /// (memory, uint) -> void - ReduceUMin, /// (memory, uint) -> void - ReduceUMax, /// (memory, uint) -> void - ReduceUAnd, /// (memory, uint) -> void - ReduceUOr, /// (memory, uint) -> void - ReduceUXor, /// (memory, uint) -> void - - ReduceIAdd, /// (memory, int) -> void - ReduceIMin, /// (memory, int) -> void - ReduceIMax, /// (memory, int) -> void - ReduceIAnd, /// (memory, int) -> void - ReduceIOr, /// (memory, int) -> void - ReduceIXor, /// (memory, int) -> void - - Branch, /// (uint branch_target) -> void - BranchIndirect, /// (uint branch_target) -> void - PushFlowStack, /// (uint branch_target) -> void - PopFlowStack, /// () -> void - Exit, /// () -> void - Discard, /// () -> void - - EmitVertex, /// () -> void - EndPrimitive, /// () -> void - - InvocationId, /// () -> int - YNegate, /// () -> float - LocalInvocationIdX, /// () -> uint - LocalInvocationIdY, /// () -> uint - LocalInvocationIdZ, /// () -> uint - WorkGroupIdX, /// () -> uint - WorkGroupIdY, /// () -> uint - WorkGroupIdZ, /// () -> uint - - BallotThread, /// (bool) -> uint - VoteAll, /// (bool) -> bool - VoteAny, /// (bool) -> bool - VoteEqual, /// (bool) -> bool - - ThreadId, /// () -> uint - ThreadEqMask, /// () -> uint - ThreadGeMask, /// () -> uint - ThreadGtMask, /// () -> uint - ThreadLeMask, /// () -> uint - ThreadLtMask, /// () -> uint - ShuffleIndexed, /// (uint value, uint index) -> uint - - Barrier, /// () -> void - MemoryBarrierGroup, /// () -> void - MemoryBarrierGlobal, /// () -> void - - Amount, -}; - -enum class InternalFlag { - Zero = 0, - Sign = 1, - Carry = 2, - Overflow = 3, - Amount = 4, -}; - -enum class MetaStackClass { - Ssy, - Pbk, -}; - -class OperationNode; -class ConditionalNode; -class GprNode; -class CustomVarNode; -class ImmediateNode; -class InternalFlagNode; -class PredicateNode; -class AbufNode; -class CbufNode; -class LmemNode; -class PatchNode; -class SmemNode; -class GmemNode; -class CommentNode; - -using NodeData = std::variant<OperationNode, ConditionalNode, GprNode, CustomVarNode, ImmediateNode, - InternalFlagNode, PredicateNode, AbufNode, PatchNode, CbufNode, - LmemNode, SmemNode, GmemNode, CommentNode>; -using Node = std::shared_ptr<NodeData>; -using Node4 = std::array<Node, 4>; -using NodeBlock = std::vector<Node>; - -struct ArraySamplerNode; -struct BindlessSamplerNode; -struct SeparateSamplerNode; - -using TrackSamplerData = std::variant<BindlessSamplerNode, SeparateSamplerNode, ArraySamplerNode>; -using TrackSampler = std::shared_ptr<TrackSamplerData>; - -struct SamplerEntry { - /// Bound samplers constructor - explicit SamplerEntry(u32 index_, u32 offset_, Tegra::Shader::TextureType type_, bool is_array_, - bool is_shadow_, bool is_buffer_, bool is_indexed_) - : index{index_}, offset{offset_}, type{type_}, is_array{is_array_}, is_shadow{is_shadow_}, - is_buffer{is_buffer_}, is_indexed{is_indexed_} {} - - /// Separate sampler constructor - explicit SamplerEntry(u32 index_, std::pair<u32, u32> offsets, std::pair<u32, u32> buffers, - Tegra::Shader::TextureType type_, bool is_array_, bool is_shadow_, - bool is_buffer_) - : index{index_}, offset{offsets.first}, secondary_offset{offsets.second}, - buffer{buffers.first}, secondary_buffer{buffers.second}, type{type_}, is_array{is_array_}, - is_shadow{is_shadow_}, is_buffer{is_buffer_}, is_separated{true} {} - - /// Bindless samplers constructor - explicit SamplerEntry(u32 index_, u32 offset_, u32 buffer_, Tegra::Shader::TextureType type_, - bool is_array_, bool is_shadow_, bool is_buffer_, bool is_indexed_) - : index{index_}, offset{offset_}, buffer{buffer_}, type{type_}, is_array{is_array_}, - is_shadow{is_shadow_}, is_buffer{is_buffer_}, is_bindless{true}, is_indexed{is_indexed_} { - } - - u32 index = 0; ///< Emulated index given for the this sampler. - u32 offset = 0; ///< Offset in the const buffer from where the sampler is being read. - u32 secondary_offset = 0; ///< Secondary offset in the const buffer. - u32 buffer = 0; ///< Buffer where the bindless sampler is read. - u32 secondary_buffer = 0; ///< Secondary buffer where the bindless sampler is read. - u32 size = 1; ///< Size of the sampler. - - Tegra::Shader::TextureType type{}; ///< The type used to sample this texture (Texture2D, etc) - bool is_array = false; ///< Whether the texture is being sampled as an array texture or not. - bool is_shadow = false; ///< Whether the texture is being sampled as a depth texture or not. - bool is_buffer = false; ///< Whether the texture is a texture buffer without sampler. - bool is_bindless = false; ///< Whether this sampler belongs to a bindless texture or not. - bool is_indexed = false; ///< Whether this sampler is an indexed array of textures. - bool is_separated = false; ///< Whether the image and sampler is separated or not. -}; - -/// Represents a tracked bindless sampler into a direct const buffer -struct ArraySamplerNode { - u32 index; - u32 base_offset; - u32 bindless_var; -}; - -/// Represents a tracked separate sampler image pair that was folded statically -struct SeparateSamplerNode { - std::pair<u32, u32> indices; - std::pair<u32, u32> offsets; -}; - -/// Represents a tracked bindless sampler into a direct const buffer -struct BindlessSamplerNode { - u32 index; - u32 offset; -}; - -struct ImageEntry { -public: - /// Bound images constructor - explicit ImageEntry(u32 index_, u32 offset_, Tegra::Shader::ImageType type_) - : index{index_}, offset{offset_}, type{type_} {} - - /// Bindless samplers constructor - explicit ImageEntry(u32 index_, u32 offset_, u32 buffer_, Tegra::Shader::ImageType type_) - : index{index_}, offset{offset_}, buffer{buffer_}, type{type_}, is_bindless{true} {} - - void MarkWrite() { - is_written = true; - } - - void MarkRead() { - is_read = true; - } - - void MarkAtomic() { - MarkWrite(); - MarkRead(); - is_atomic = true; - } - - u32 index = 0; - u32 offset = 0; - u32 buffer = 0; - - Tegra::Shader::ImageType type{}; - bool is_bindless = false; - bool is_written = false; - bool is_read = false; - bool is_atomic = false; -}; - -struct GlobalMemoryBase { - u32 cbuf_index = 0; - u32 cbuf_offset = 0; - - [[nodiscard]] bool operator<(const GlobalMemoryBase& rhs) const { - return std::tie(cbuf_index, cbuf_offset) < std::tie(rhs.cbuf_index, rhs.cbuf_offset); - } -}; - -/// Parameters describing an arithmetic operation -struct MetaArithmetic { - bool precise{}; ///< Whether the operation can be constraint or not -}; - -/// Parameters describing a texture sampler -struct MetaTexture { - SamplerEntry sampler; - Node array; - Node depth_compare; - std::vector<Node> aoffi; - std::vector<Node> ptp; - std::vector<Node> derivates; - Node bias; - Node lod; - Node component; - u32 element{}; - Node index; -}; - -struct MetaImage { - const ImageEntry& image; - std::vector<Node> values; - u32 element{}; -}; - -/// Parameters that modify an operation but are not part of any particular operand -using Meta = - std::variant<MetaArithmetic, MetaTexture, MetaImage, MetaStackClass, Tegra::Shader::HalfType>; - -class AmendNode { -public: - [[nodiscard]] std::optional<std::size_t> GetAmendIndex() const { - if (amend_index == amend_null_index) { - return std::nullopt; - } - return {amend_index}; - } - - void SetAmendIndex(std::size_t index) { - amend_index = index; - } - - void ClearAmend() { - amend_index = amend_null_index; - } - -private: - static constexpr std::size_t amend_null_index = 0xFFFFFFFFFFFFFFFFULL; - std::size_t amend_index{amend_null_index}; -}; - -/// Holds any kind of operation that can be done in the IR -class OperationNode final : public AmendNode { -public: - explicit OperationNode(OperationCode code_) : OperationNode(code_, Meta{}) {} - - explicit OperationNode(OperationCode code_, Meta meta_) - : OperationNode(code_, std::move(meta_), std::vector<Node>{}) {} - - explicit OperationNode(OperationCode code_, std::vector<Node> operands_) - : OperationNode(code_, Meta{}, std::move(operands_)) {} - - explicit OperationNode(OperationCode code_, Meta meta_, std::vector<Node> operands_) - : code{code_}, meta{std::move(meta_)}, operands{std::move(operands_)} {} - - template <typename... Args> - explicit OperationNode(OperationCode code_, Meta meta_, Args&&... operands_) - : code{code_}, meta{std::move(meta_)}, operands{operands_...} {} - - [[nodiscard]] OperationCode GetCode() const { - return code; - } - - [[nodiscard]] const Meta& GetMeta() const { - return meta; - } - - [[nodiscard]] std::size_t GetOperandsCount() const { - return operands.size(); - } - - [[nodiscard]] const Node& operator[](std::size_t operand_index) const { - return operands.at(operand_index); - } - -private: - OperationCode code{}; - Meta meta{}; - std::vector<Node> operands; -}; - -/// Encloses inside any kind of node that returns a boolean conditionally-executed code -class ConditionalNode final : public AmendNode { -public: - explicit ConditionalNode(Node condition_, std::vector<Node>&& code_) - : condition{std::move(condition_)}, code{std::move(code_)} {} - - [[nodiscard]] const Node& GetCondition() const { - return condition; - } - - [[nodiscard]] const std::vector<Node>& GetCode() const { - return code; - } - -private: - Node condition; ///< Condition to be satisfied - std::vector<Node> code; ///< Code to execute -}; - -/// A general purpose register -class GprNode final { -public: - explicit constexpr GprNode(Tegra::Shader::Register index_) : index{index_} {} - - [[nodiscard]] constexpr u32 GetIndex() const { - return static_cast<u32>(index); - } - -private: - Tegra::Shader::Register index{}; -}; - -/// A custom variable -class CustomVarNode final { -public: - explicit constexpr CustomVarNode(u32 index_) : index{index_} {} - - [[nodiscard]] constexpr u32 GetIndex() const { - return index; - } - -private: - u32 index{}; -}; - -/// A 32-bits value that represents an immediate value -class ImmediateNode final { -public: - explicit constexpr ImmediateNode(u32 value_) : value{value_} {} - - [[nodiscard]] constexpr u32 GetValue() const { - return value; - } - -private: - u32 value{}; -}; - -/// One of Maxwell's internal flags -class InternalFlagNode final { -public: - explicit constexpr InternalFlagNode(InternalFlag flag_) : flag{flag_} {} - - [[nodiscard]] constexpr InternalFlag GetFlag() const { - return flag; - } - -private: - InternalFlag flag{}; -}; - -/// A predicate register, it can be negated without additional nodes -class PredicateNode final { -public: - explicit constexpr PredicateNode(Tegra::Shader::Pred index_, bool negated_) - : index{index_}, negated{negated_} {} - - [[nodiscard]] constexpr Tegra::Shader::Pred GetIndex() const { - return index; - } - - [[nodiscard]] constexpr bool IsNegated() const { - return negated; - } - -private: - Tegra::Shader::Pred index{}; - bool negated{}; -}; - -/// Attribute buffer memory (known as attributes or varyings in GLSL terms) -class AbufNode final { -public: - // Initialize for standard attributes (index is explicit). - explicit AbufNode(Tegra::Shader::Attribute::Index index_, u32 element_, Node buffer_ = {}) - : buffer{std::move(buffer_)}, index{index_}, element{element_} {} - - // Initialize for physical attributes (index is a variable value). - explicit AbufNode(Node physical_address_, Node buffer_ = {}) - : physical_address{std::move(physical_address_)}, buffer{std::move(buffer_)} {} - - [[nodiscard]] Tegra::Shader::Attribute::Index GetIndex() const { - return index; - } - - [[nodiscard]] u32 GetElement() const { - return element; - } - - [[nodiscard]] const Node& GetBuffer() const { - return buffer; - } - - [[nodiscard]] bool IsPhysicalBuffer() const { - return static_cast<bool>(physical_address); - } - - [[nodiscard]] const Node& GetPhysicalAddress() const { - return physical_address; - } - -private: - Node physical_address; - Node buffer; - Tegra::Shader::Attribute::Index index{}; - u32 element{}; -}; - -/// Patch memory (used to communicate tessellation stages). -class PatchNode final { -public: - explicit constexpr PatchNode(u32 offset_) : offset{offset_} {} - - [[nodiscard]] constexpr u32 GetOffset() const { - return offset; - } - -private: - u32 offset{}; -}; - -/// Constant buffer node, usually mapped to uniform buffers in GLSL -class CbufNode final { -public: - explicit CbufNode(u32 index_, Node offset_) : index{index_}, offset{std::move(offset_)} {} - - [[nodiscard]] u32 GetIndex() const { - return index; - } - - [[nodiscard]] const Node& GetOffset() const { - return offset; - } - -private: - u32 index{}; - Node offset; -}; - -/// Local memory node -class LmemNode final { -public: - explicit LmemNode(Node address_) : address{std::move(address_)} {} - - [[nodiscard]] const Node& GetAddress() const { - return address; - } - -private: - Node address; -}; - -/// Shared memory node -class SmemNode final { -public: - explicit SmemNode(Node address_) : address{std::move(address_)} {} - - [[nodiscard]] const Node& GetAddress() const { - return address; - } - -private: - Node address; -}; - -/// Global memory node -class GmemNode final { -public: - explicit GmemNode(Node real_address_, Node base_address_, const GlobalMemoryBase& descriptor_) - : real_address{std::move(real_address_)}, base_address{std::move(base_address_)}, - descriptor{descriptor_} {} - - [[nodiscard]] const Node& GetRealAddress() const { - return real_address; - } - - [[nodiscard]] const Node& GetBaseAddress() const { - return base_address; - } - - [[nodiscard]] const GlobalMemoryBase& GetDescriptor() const { - return descriptor; - } - -private: - Node real_address; - Node base_address; - GlobalMemoryBase descriptor; -}; - -/// Commentary, can be dropped -class CommentNode final { -public: - explicit CommentNode(std::string text_) : text{std::move(text_)} {} - - [[nodiscard]] const std::string& GetText() const { - return text; - } - -private: - std::string text; -}; - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/node_helper.cpp b/src/video_core/shader/node_helper.cpp deleted file mode 100644 index 6a5b6940d..000000000 --- a/src/video_core/shader/node_helper.cpp +++ /dev/null @@ -1,115 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <cstring> -#include <vector> - -#include "common/common_types.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -Node Conditional(Node condition, std::vector<Node> code) { - return MakeNode<ConditionalNode>(std::move(condition), std::move(code)); -} - -Node Comment(std::string text) { - return MakeNode<CommentNode>(std::move(text)); -} - -Node Immediate(u32 value) { - return MakeNode<ImmediateNode>(value); -} - -Node Immediate(s32 value) { - return Immediate(static_cast<u32>(value)); -} - -Node Immediate(f32 value) { - u32 integral; - std::memcpy(&integral, &value, sizeof(u32)); - return Immediate(integral); -} - -OperationCode SignedToUnsignedCode(OperationCode operation_code, bool is_signed) { - if (is_signed) { - return operation_code; - } - switch (operation_code) { - case OperationCode::FCastInteger: - return OperationCode::FCastUInteger; - case OperationCode::IAdd: - return OperationCode::UAdd; - case OperationCode::IMul: - return OperationCode::UMul; - case OperationCode::IDiv: - return OperationCode::UDiv; - case OperationCode::IMin: - return OperationCode::UMin; - case OperationCode::IMax: - return OperationCode::UMax; - case OperationCode::ICastFloat: - return OperationCode::UCastFloat; - case OperationCode::ICastUnsigned: - return OperationCode::UCastSigned; - case OperationCode::ILogicalShiftLeft: - return OperationCode::ULogicalShiftLeft; - case OperationCode::ILogicalShiftRight: - return OperationCode::ULogicalShiftRight; - case OperationCode::IArithmeticShiftRight: - return OperationCode::UArithmeticShiftRight; - case OperationCode::IBitwiseAnd: - return OperationCode::UBitwiseAnd; - case OperationCode::IBitwiseOr: - return OperationCode::UBitwiseOr; - case OperationCode::IBitwiseXor: - return OperationCode::UBitwiseXor; - case OperationCode::IBitwiseNot: - return OperationCode::UBitwiseNot; - case OperationCode::IBitfieldExtract: - return OperationCode::UBitfieldExtract; - case OperationCode::IBitfieldInsert: - return OperationCode::UBitfieldInsert; - case OperationCode::IBitCount: - return OperationCode::UBitCount; - case OperationCode::LogicalILessThan: - return OperationCode::LogicalULessThan; - case OperationCode::LogicalIEqual: - return OperationCode::LogicalUEqual; - case OperationCode::LogicalILessEqual: - return OperationCode::LogicalULessEqual; - case OperationCode::LogicalIGreaterThan: - return OperationCode::LogicalUGreaterThan; - case OperationCode::LogicalINotEqual: - return OperationCode::LogicalUNotEqual; - case OperationCode::LogicalIGreaterEqual: - return OperationCode::LogicalUGreaterEqual; - case OperationCode::AtomicIExchange: - return OperationCode::AtomicUExchange; - case OperationCode::AtomicIAdd: - return OperationCode::AtomicUAdd; - case OperationCode::AtomicIMin: - return OperationCode::AtomicUMin; - case OperationCode::AtomicIMax: - return OperationCode::AtomicUMax; - case OperationCode::AtomicIAnd: - return OperationCode::AtomicUAnd; - case OperationCode::AtomicIOr: - return OperationCode::AtomicUOr; - case OperationCode::AtomicIXor: - return OperationCode::AtomicUXor; - case OperationCode::INegate: - UNREACHABLE_MSG("Can't negate an unsigned integer"); - return {}; - case OperationCode::IAbsolute: - UNREACHABLE_MSG("Can't apply absolute to an unsigned integer"); - return {}; - default: - UNREACHABLE_MSG("Unknown signed operation with code={}", operation_code); - return {}; - } -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/node_helper.h b/src/video_core/shader/node_helper.h deleted file mode 100644 index 1e0886185..000000000 --- a/src/video_core/shader/node_helper.h +++ /dev/null @@ -1,71 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <memory> -#include <string> -#include <tuple> -#include <type_traits> -#include <utility> -#include <vector> - -#include "common/common_types.h" -#include "video_core/shader/node.h" - -namespace VideoCommon::Shader { - -/// This arithmetic operation cannot be constraint -inline constexpr MetaArithmetic PRECISE = {true}; -/// This arithmetic operation can be optimized away -inline constexpr MetaArithmetic NO_PRECISE = {false}; - -/// Creates a conditional node -Node Conditional(Node condition, std::vector<Node> code); - -/// Creates a commentary node -Node Comment(std::string text); - -/// Creates an u32 immediate -Node Immediate(u32 value); - -/// Creates a s32 immediate -Node Immediate(s32 value); - -/// Creates a f32 immediate -Node Immediate(f32 value); - -/// Converts an signed operation code to an unsigned operation code -OperationCode SignedToUnsignedCode(OperationCode operation_code, bool is_signed); - -template <typename T, typename... Args> -Node MakeNode(Args&&... args) { - static_assert(std::is_convertible_v<T, NodeData>); - return std::make_shared<NodeData>(T(std::forward<Args>(args)...)); -} - -template <typename T, typename... Args> -TrackSampler MakeTrackSampler(Args&&... args) { - static_assert(std::is_convertible_v<T, TrackSamplerData>); - return std::make_shared<TrackSamplerData>(T{std::forward<Args>(args)...}); -} - -template <typename... Args> -Node Operation(OperationCode code, Args&&... args) { - if constexpr (sizeof...(args) == 0) { - return MakeNode<OperationNode>(code); - } else if constexpr (std::is_convertible_v<std::tuple_element_t<0, std::tuple<Args...>>, - Meta>) { - return MakeNode<OperationNode>(code, std::forward<Args>(args)...); - } else { - return MakeNode<OperationNode>(code, Meta{}, std::forward<Args>(args)...); - } -} - -template <typename... Args> -Node SignedOperation(OperationCode code, bool is_signed, Args&&... args) { - return Operation(SignedToUnsignedCode(code, is_signed), std::forward<Args>(args)...); -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/registry.cpp b/src/video_core/shader/registry.cpp deleted file mode 100644 index 148d91fcb..000000000 --- a/src/video_core/shader/registry.cpp +++ /dev/null @@ -1,181 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <tuple> - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/kepler_compute.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/shader_type.h" -#include "video_core/shader/registry.h" - -namespace VideoCommon::Shader { - -using Tegra::Engines::ConstBufferEngineInterface; -using Tegra::Engines::SamplerDescriptor; -using Tegra::Engines::ShaderType; - -namespace { - -GraphicsInfo MakeGraphicsInfo(ShaderType shader_stage, ConstBufferEngineInterface& engine) { - if (shader_stage == ShaderType::Compute) { - return {}; - } - - auto& graphics = dynamic_cast<Tegra::Engines::Maxwell3D&>(engine); - - return { - .tfb_layouts = graphics.regs.tfb_layouts, - .tfb_varying_locs = graphics.regs.tfb_varying_locs, - .primitive_topology = graphics.regs.draw.topology, - .tessellation_primitive = graphics.regs.tess_mode.prim, - .tessellation_spacing = graphics.regs.tess_mode.spacing, - .tfb_enabled = graphics.regs.tfb_enabled != 0, - .tessellation_clockwise = graphics.regs.tess_mode.cw.Value() != 0, - }; -} - -ComputeInfo MakeComputeInfo(ShaderType shader_stage, ConstBufferEngineInterface& engine) { - if (shader_stage != ShaderType::Compute) { - return {}; - } - - auto& compute = dynamic_cast<Tegra::Engines::KeplerCompute&>(engine); - const auto& launch = compute.launch_description; - - return { - .workgroup_size = {launch.block_dim_x, launch.block_dim_y, launch.block_dim_z}, - .shared_memory_size_in_words = launch.shared_alloc, - .local_memory_size_in_words = launch.local_pos_alloc, - }; -} - -} // Anonymous namespace - -Registry::Registry(ShaderType shader_stage, const SerializedRegistryInfo& info) - : stage{shader_stage}, stored_guest_driver_profile{info.guest_driver_profile}, - bound_buffer{info.bound_buffer}, graphics_info{info.graphics}, compute_info{info.compute} {} - -Registry::Registry(ShaderType shader_stage, ConstBufferEngineInterface& engine_) - : stage{shader_stage}, engine{&engine_}, bound_buffer{engine_.GetBoundBuffer()}, - graphics_info{MakeGraphicsInfo(shader_stage, engine_)}, compute_info{MakeComputeInfo( - shader_stage, engine_)} {} - -Registry::~Registry() = default; - -std::optional<u32> Registry::ObtainKey(u32 buffer, u32 offset) { - const std::pair<u32, u32> key = {buffer, offset}; - const auto iter = keys.find(key); - if (iter != keys.end()) { - return iter->second; - } - if (!engine) { - return std::nullopt; - } - const u32 value = engine->AccessConstBuffer32(stage, buffer, offset); - keys.emplace(key, value); - return value; -} - -std::optional<SamplerDescriptor> Registry::ObtainBoundSampler(u32 offset) { - const u32 key = offset; - const auto iter = bound_samplers.find(key); - if (iter != bound_samplers.end()) { - return iter->second; - } - if (!engine) { - return std::nullopt; - } - const SamplerDescriptor value = engine->AccessBoundSampler(stage, offset); - bound_samplers.emplace(key, value); - return value; -} - -std::optional<Tegra::Engines::SamplerDescriptor> Registry::ObtainSeparateSampler( - std::pair<u32, u32> buffers, std::pair<u32, u32> offsets) { - SeparateSamplerKey key; - key.buffers = buffers; - key.offsets = offsets; - const auto iter = separate_samplers.find(key); - if (iter != separate_samplers.end()) { - return iter->second; - } - if (!engine) { - return std::nullopt; - } - - const u32 handle_1 = engine->AccessConstBuffer32(stage, key.buffers.first, key.offsets.first); - const u32 handle_2 = engine->AccessConstBuffer32(stage, key.buffers.second, key.offsets.second); - const SamplerDescriptor value = engine->AccessSampler(handle_1 | handle_2); - separate_samplers.emplace(key, value); - return value; -} - -std::optional<SamplerDescriptor> Registry::ObtainBindlessSampler(u32 buffer, u32 offset) { - const std::pair key = {buffer, offset}; - const auto iter = bindless_samplers.find(key); - if (iter != bindless_samplers.end()) { - return iter->second; - } - if (!engine) { - return std::nullopt; - } - const SamplerDescriptor value = engine->AccessBindlessSampler(stage, buffer, offset); - bindless_samplers.emplace(key, value); - return value; -} - -void Registry::InsertKey(u32 buffer, u32 offset, u32 value) { - keys.insert_or_assign({buffer, offset}, value); -} - -void Registry::InsertBoundSampler(u32 offset, SamplerDescriptor sampler) { - bound_samplers.insert_or_assign(offset, sampler); -} - -void Registry::InsertBindlessSampler(u32 buffer, u32 offset, SamplerDescriptor sampler) { - bindless_samplers.insert_or_assign({buffer, offset}, sampler); -} - -bool Registry::IsConsistent() const { - if (!engine) { - return true; - } - return std::all_of(keys.begin(), keys.end(), - [this](const auto& pair) { - const auto [cbuf, offset] = pair.first; - const auto value = pair.second; - return value == engine->AccessConstBuffer32(stage, cbuf, offset); - }) && - std::all_of(bound_samplers.begin(), bound_samplers.end(), - [this](const auto& sampler) { - const auto [key, value] = sampler; - return value == engine->AccessBoundSampler(stage, key); - }) && - std::all_of(bindless_samplers.begin(), bindless_samplers.end(), - [this](const auto& sampler) { - const auto [cbuf, offset] = sampler.first; - const auto value = sampler.second; - return value == engine->AccessBindlessSampler(stage, cbuf, offset); - }); -} - -bool Registry::HasEqualKeys(const Registry& rhs) const { - return std::tie(keys, bound_samplers, bindless_samplers) == - std::tie(rhs.keys, rhs.bound_samplers, rhs.bindless_samplers); -} - -const GraphicsInfo& Registry::GetGraphicsInfo() const { - ASSERT(stage != Tegra::Engines::ShaderType::Compute); - return graphics_info; -} - -const ComputeInfo& Registry::GetComputeInfo() const { - ASSERT(stage == Tegra::Engines::ShaderType::Compute); - return compute_info; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/registry.h b/src/video_core/shader/registry.h deleted file mode 100644 index 4bebefdde..000000000 --- a/src/video_core/shader/registry.h +++ /dev/null @@ -1,172 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <array> -#include <optional> -#include <type_traits> -#include <unordered_map> -#include <utility> - -#include "common/common_types.h" -#include "common/hash.h" -#include "video_core/engines/const_buffer_engine_interface.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/shader_type.h" -#include "video_core/guest_driver.h" - -namespace VideoCommon::Shader { - -struct SeparateSamplerKey { - std::pair<u32, u32> buffers; - std::pair<u32, u32> offsets; -}; - -} // namespace VideoCommon::Shader - -namespace std { - -template <> -struct hash<VideoCommon::Shader::SeparateSamplerKey> { - std::size_t operator()(const VideoCommon::Shader::SeparateSamplerKey& key) const noexcept { - return std::hash<u32>{}(key.buffers.first ^ key.buffers.second ^ key.offsets.first ^ - key.offsets.second); - } -}; - -template <> -struct equal_to<VideoCommon::Shader::SeparateSamplerKey> { - bool operator()(const VideoCommon::Shader::SeparateSamplerKey& lhs, - const VideoCommon::Shader::SeparateSamplerKey& rhs) const noexcept { - return lhs.buffers == rhs.buffers && lhs.offsets == rhs.offsets; - } -}; - -} // namespace std - -namespace VideoCommon::Shader { - -using KeyMap = std::unordered_map<std::pair<u32, u32>, u32, Common::PairHash>; -using BoundSamplerMap = std::unordered_map<u32, Tegra::Engines::SamplerDescriptor>; -using SeparateSamplerMap = - std::unordered_map<SeparateSamplerKey, Tegra::Engines::SamplerDescriptor>; -using BindlessSamplerMap = - std::unordered_map<std::pair<u32, u32>, Tegra::Engines::SamplerDescriptor, Common::PairHash>; - -struct GraphicsInfo { - using Maxwell = Tegra::Engines::Maxwell3D::Regs; - - std::array<Maxwell::TransformFeedbackLayout, Maxwell::NumTransformFeedbackBuffers> - tfb_layouts{}; - std::array<std::array<u8, 128>, Maxwell::NumTransformFeedbackBuffers> tfb_varying_locs{}; - Maxwell::PrimitiveTopology primitive_topology{}; - Maxwell::TessellationPrimitive tessellation_primitive{}; - Maxwell::TessellationSpacing tessellation_spacing{}; - bool tfb_enabled = false; - bool tessellation_clockwise = false; -}; -static_assert(std::is_trivially_copyable_v<GraphicsInfo> && - std::is_standard_layout_v<GraphicsInfo>); - -struct ComputeInfo { - std::array<u32, 3> workgroup_size{}; - u32 shared_memory_size_in_words = 0; - u32 local_memory_size_in_words = 0; -}; -static_assert(std::is_trivially_copyable_v<ComputeInfo> && std::is_standard_layout_v<ComputeInfo>); - -struct SerializedRegistryInfo { - VideoCore::GuestDriverProfile guest_driver_profile; - u32 bound_buffer = 0; - GraphicsInfo graphics; - ComputeInfo compute; -}; - -/** - * The Registry is a class use to interface the 3D and compute engines with the shader compiler. - * With it, the shader can obtain required data from GPU state and store it for disk shader - * compilation. - */ -class Registry { -public: - explicit Registry(Tegra::Engines::ShaderType shader_stage, const SerializedRegistryInfo& info); - - explicit Registry(Tegra::Engines::ShaderType shader_stage, - Tegra::Engines::ConstBufferEngineInterface& engine_); - - ~Registry(); - - /// Retrieves a key from the registry, if it's registered, it will give the registered value, if - /// not it will obtain it from maxwell3d and register it. - std::optional<u32> ObtainKey(u32 buffer, u32 offset); - - std::optional<Tegra::Engines::SamplerDescriptor> ObtainBoundSampler(u32 offset); - - std::optional<Tegra::Engines::SamplerDescriptor> ObtainSeparateSampler( - std::pair<u32, u32> buffers, std::pair<u32, u32> offsets); - - std::optional<Tegra::Engines::SamplerDescriptor> ObtainBindlessSampler(u32 buffer, u32 offset); - - /// Inserts a key. - void InsertKey(u32 buffer, u32 offset, u32 value); - - /// Inserts a bound sampler key. - void InsertBoundSampler(u32 offset, Tegra::Engines::SamplerDescriptor sampler); - - /// Inserts a bindless sampler key. - void InsertBindlessSampler(u32 buffer, u32 offset, Tegra::Engines::SamplerDescriptor sampler); - - /// Checks keys and samplers against engine's current const buffers. - /// Returns true if they are the same value, false otherwise. - bool IsConsistent() const; - - /// Returns true if the keys are equal to the other ones in the registry. - bool HasEqualKeys(const Registry& rhs) const; - - /// Returns graphics information from this shader - const GraphicsInfo& GetGraphicsInfo() const; - - /// Returns compute information from this shader - const ComputeInfo& GetComputeInfo() const; - - /// Gives an getter to the const buffer keys in the database. - const KeyMap& GetKeys() const { - return keys; - } - - /// Gets samplers database. - const BoundSamplerMap& GetBoundSamplers() const { - return bound_samplers; - } - - /// Gets bindless samplers database. - const BindlessSamplerMap& GetBindlessSamplers() const { - return bindless_samplers; - } - - /// Gets bound buffer used on this shader - u32 GetBoundBuffer() const { - return bound_buffer; - } - - /// Obtains access to the guest driver's profile. - VideoCore::GuestDriverProfile& AccessGuestDriverProfile() { - return engine ? engine->AccessGuestDriverProfile() : stored_guest_driver_profile; - } - -private: - const Tegra::Engines::ShaderType stage; - VideoCore::GuestDriverProfile stored_guest_driver_profile; - Tegra::Engines::ConstBufferEngineInterface* engine = nullptr; - KeyMap keys; - BoundSamplerMap bound_samplers; - SeparateSamplerMap separate_samplers; - BindlessSamplerMap bindless_samplers; - u32 bound_buffer; - GraphicsInfo graphics_info; - ComputeInfo compute_info; -}; - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/shader_ir.cpp b/src/video_core/shader/shader_ir.cpp deleted file mode 100644 index a4987ffc6..000000000 --- a/src/video_core/shader/shader_ir.cpp +++ /dev/null @@ -1,464 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <array> -#include <cmath> - -#include "common/assert.h" -#include "common/common_types.h" -#include "common/logging/log.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/shader/node.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/registry.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -using Tegra::Shader::Attribute; -using Tegra::Shader::Instruction; -using Tegra::Shader::IpaMode; -using Tegra::Shader::Pred; -using Tegra::Shader::PredCondition; -using Tegra::Shader::PredOperation; -using Tegra::Shader::Register; - -ShaderIR::ShaderIR(const ProgramCode& program_code_, u32 main_offset_, CompilerSettings settings_, - Registry& registry_) - : program_code{program_code_}, main_offset{main_offset_}, settings{settings_}, registry{ - registry_} { - Decode(); - PostDecode(); -} - -ShaderIR::~ShaderIR() = default; - -Node ShaderIR::GetRegister(Register reg) { - if (reg != Register::ZeroIndex) { - used_registers.insert(static_cast<u32>(reg)); - } - return MakeNode<GprNode>(reg); -} - -Node ShaderIR::GetCustomVariable(u32 id) { - return MakeNode<CustomVarNode>(id); -} - -Node ShaderIR::GetImmediate19(Instruction instr) { - return Immediate(instr.alu.GetImm20_19()); -} - -Node ShaderIR::GetImmediate32(Instruction instr) { - return Immediate(instr.alu.GetImm20_32()); -} - -Node ShaderIR::GetConstBuffer(u64 index_, u64 offset_) { - const auto index = static_cast<u32>(index_); - const auto offset = static_cast<u32>(offset_); - - used_cbufs.try_emplace(index).first->second.MarkAsUsed(offset); - - return MakeNode<CbufNode>(index, Immediate(offset)); -} - -Node ShaderIR::GetConstBufferIndirect(u64 index_, u64 offset_, Node node) { - const auto index = static_cast<u32>(index_); - const auto offset = static_cast<u32>(offset_); - - used_cbufs.try_emplace(index).first->second.MarkAsUsedIndirect(); - - Node final_offset = [&] { - // Attempt to inline constant buffer without a variable offset. This is done to allow - // tracking LDC calls. - if (const auto gpr = std::get_if<GprNode>(&*node)) { - if (gpr->GetIndex() == Register::ZeroIndex) { - return Immediate(offset); - } - } - return Operation(OperationCode::UAdd, NO_PRECISE, std::move(node), Immediate(offset)); - }(); - return MakeNode<CbufNode>(index, std::move(final_offset)); -} - -Node ShaderIR::GetPredicate(u64 pred_, bool negated) { - const auto pred = static_cast<Pred>(pred_); - if (pred != Pred::UnusedIndex && pred != Pred::NeverExecute) { - used_predicates.insert(pred); - } - - return MakeNode<PredicateNode>(pred, negated); -} - -Node ShaderIR::GetPredicate(bool immediate) { - return GetPredicate(static_cast<u64>(immediate ? Pred::UnusedIndex : Pred::NeverExecute)); -} - -Node ShaderIR::GetInputAttribute(Attribute::Index index, u64 element, Node buffer) { - MarkAttributeUsage(index, element); - used_input_attributes.emplace(index); - return MakeNode<AbufNode>(index, static_cast<u32>(element), std::move(buffer)); -} - -Node ShaderIR::GetPhysicalInputAttribute(Tegra::Shader::Register physical_address, Node buffer) { - uses_physical_attributes = true; - return MakeNode<AbufNode>(GetRegister(physical_address), buffer); -} - -Node ShaderIR::GetOutputAttribute(Attribute::Index index, u64 element, Node buffer) { - MarkAttributeUsage(index, element); - used_output_attributes.insert(index); - return MakeNode<AbufNode>(index, static_cast<u32>(element), std::move(buffer)); -} - -Node ShaderIR::GetInternalFlag(InternalFlag flag, bool negated) const { - Node node = MakeNode<InternalFlagNode>(flag); - if (negated) { - return Operation(OperationCode::LogicalNegate, std::move(node)); - } - return node; -} - -Node ShaderIR::GetLocalMemory(Node address) { - return MakeNode<LmemNode>(std::move(address)); -} - -Node ShaderIR::GetSharedMemory(Node address) { - return MakeNode<SmemNode>(std::move(address)); -} - -Node ShaderIR::GetTemporary(u32 id) { - return GetRegister(Register::ZeroIndex + 1 + id); -} - -Node ShaderIR::GetOperandAbsNegFloat(Node value, bool absolute, bool negate) { - if (absolute) { - value = Operation(OperationCode::FAbsolute, NO_PRECISE, std::move(value)); - } - if (negate) { - value = Operation(OperationCode::FNegate, NO_PRECISE, std::move(value)); - } - return value; -} - -Node ShaderIR::GetSaturatedFloat(Node value, bool saturate) { - if (!saturate) { - return value; - } - - Node positive_zero = Immediate(std::copysignf(0, 1)); - Node positive_one = Immediate(1.0f); - return Operation(OperationCode::FClamp, NO_PRECISE, std::move(value), std::move(positive_zero), - std::move(positive_one)); -} - -Node ShaderIR::ConvertIntegerSize(Node value, Register::Size size, bool is_signed) { - switch (size) { - case Register::Size::Byte: - value = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, NO_PRECISE, - std::move(value), Immediate(24)); - value = SignedOperation(OperationCode::IArithmeticShiftRight, is_signed, NO_PRECISE, - std::move(value), Immediate(24)); - return value; - case Register::Size::Short: - value = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, NO_PRECISE, - std::move(value), Immediate(16)); - value = SignedOperation(OperationCode::IArithmeticShiftRight, is_signed, NO_PRECISE, - std::move(value), Immediate(16)); - return value; - case Register::Size::Word: - // Default - do nothing - return value; - default: - UNREACHABLE_MSG("Unimplemented conversion size: {}", size); - return value; - } -} - -Node ShaderIR::GetOperandAbsNegInteger(Node value, bool absolute, bool negate, bool is_signed) { - if (!is_signed) { - // Absolute or negate on an unsigned is pointless - return value; - } - if (absolute) { - value = Operation(OperationCode::IAbsolute, NO_PRECISE, std::move(value)); - } - if (negate) { - value = Operation(OperationCode::INegate, NO_PRECISE, std::move(value)); - } - return value; -} - -Node ShaderIR::UnpackHalfImmediate(Instruction instr, bool has_negation) { - Node value = Immediate(instr.half_imm.PackImmediates()); - if (!has_negation) { - return value; - } - - Node first_negate = GetPredicate(instr.half_imm.first_negate != 0); - Node second_negate = GetPredicate(instr.half_imm.second_negate != 0); - - return Operation(OperationCode::HNegate, NO_PRECISE, std::move(value), std::move(first_negate), - std::move(second_negate)); -} - -Node ShaderIR::UnpackHalfFloat(Node value, Tegra::Shader::HalfType type) { - return Operation(OperationCode::HUnpack, type, std::move(value)); -} - -Node ShaderIR::HalfMerge(Node dest, Node src, Tegra::Shader::HalfMerge merge) { - switch (merge) { - case Tegra::Shader::HalfMerge::H0_H1: - return src; - case Tegra::Shader::HalfMerge::F32: - return Operation(OperationCode::HMergeF32, std::move(src)); - case Tegra::Shader::HalfMerge::Mrg_H0: - return Operation(OperationCode::HMergeH0, std::move(dest), std::move(src)); - case Tegra::Shader::HalfMerge::Mrg_H1: - return Operation(OperationCode::HMergeH1, std::move(dest), std::move(src)); - } - UNREACHABLE(); - return src; -} - -Node ShaderIR::GetOperandAbsNegHalf(Node value, bool absolute, bool negate) { - if (absolute) { - value = Operation(OperationCode::HAbsolute, NO_PRECISE, std::move(value)); - } - if (negate) { - value = Operation(OperationCode::HNegate, NO_PRECISE, std::move(value), GetPredicate(true), - GetPredicate(true)); - } - return value; -} - -Node ShaderIR::GetSaturatedHalfFloat(Node value, bool saturate) { - if (!saturate) { - return value; - } - - Node positive_zero = Immediate(std::copysignf(0, 1)); - Node positive_one = Immediate(1.0f); - return Operation(OperationCode::HClamp, NO_PRECISE, std::move(value), std::move(positive_zero), - std::move(positive_one)); -} - -Node ShaderIR::GetPredicateComparisonFloat(PredCondition condition, Node op_a, Node op_b) { - if (condition == PredCondition::T) { - return GetPredicate(true); - } else if (condition == PredCondition::F) { - return GetPredicate(false); - } - - static constexpr std::array comparison_table{ - OperationCode(0), - OperationCode::LogicalFOrdLessThan, // LT - OperationCode::LogicalFOrdEqual, // EQ - OperationCode::LogicalFOrdLessEqual, // LE - OperationCode::LogicalFOrdGreaterThan, // GT - OperationCode::LogicalFOrdNotEqual, // NE - OperationCode::LogicalFOrdGreaterEqual, // GE - OperationCode::LogicalFOrdered, // NUM - OperationCode::LogicalFUnordered, // NAN - OperationCode::LogicalFUnordLessThan, // LTU - OperationCode::LogicalFUnordEqual, // EQU - OperationCode::LogicalFUnordLessEqual, // LEU - OperationCode::LogicalFUnordGreaterThan, // GTU - OperationCode::LogicalFUnordNotEqual, // NEU - OperationCode::LogicalFUnordGreaterEqual, // GEU - }; - const std::size_t index = static_cast<std::size_t>(condition); - ASSERT_MSG(index < std::size(comparison_table), "Invalid condition={}", index); - - return Operation(comparison_table[index], op_a, op_b); -} - -Node ShaderIR::GetPredicateComparisonInteger(PredCondition condition, bool is_signed, Node op_a, - Node op_b) { - static constexpr std::array comparison_table{ - std::pair{PredCondition::LT, OperationCode::LogicalILessThan}, - std::pair{PredCondition::EQ, OperationCode::LogicalIEqual}, - std::pair{PredCondition::LE, OperationCode::LogicalILessEqual}, - std::pair{PredCondition::GT, OperationCode::LogicalIGreaterThan}, - std::pair{PredCondition::NE, OperationCode::LogicalINotEqual}, - std::pair{PredCondition::GE, OperationCode::LogicalIGreaterEqual}, - }; - - const auto comparison = - std::find_if(comparison_table.cbegin(), comparison_table.cend(), - [condition](const auto entry) { return condition == entry.first; }); - UNIMPLEMENTED_IF_MSG(comparison == comparison_table.cend(), - "Unknown predicate comparison operation"); - - return SignedOperation(comparison->second, is_signed, NO_PRECISE, std::move(op_a), - std::move(op_b)); -} - -Node ShaderIR::GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition, Node op_a, - Node op_b) { - static constexpr std::array comparison_table{ - std::pair{PredCondition::LT, OperationCode::Logical2HLessThan}, - std::pair{PredCondition::EQ, OperationCode::Logical2HEqual}, - std::pair{PredCondition::LE, OperationCode::Logical2HLessEqual}, - std::pair{PredCondition::GT, OperationCode::Logical2HGreaterThan}, - std::pair{PredCondition::NE, OperationCode::Logical2HNotEqual}, - std::pair{PredCondition::GE, OperationCode::Logical2HGreaterEqual}, - std::pair{PredCondition::LTU, OperationCode::Logical2HLessThanWithNan}, - std::pair{PredCondition::LEU, OperationCode::Logical2HLessEqualWithNan}, - std::pair{PredCondition::GTU, OperationCode::Logical2HGreaterThanWithNan}, - std::pair{PredCondition::NEU, OperationCode::Logical2HNotEqualWithNan}, - std::pair{PredCondition::GEU, OperationCode::Logical2HGreaterEqualWithNan}, - }; - - const auto comparison = - std::find_if(comparison_table.cbegin(), comparison_table.cend(), - [condition](const auto entry) { return condition == entry.first; }); - UNIMPLEMENTED_IF_MSG(comparison == comparison_table.cend(), - "Unknown predicate comparison operation"); - - return Operation(comparison->second, NO_PRECISE, std::move(op_a), std::move(op_b)); -} - -OperationCode ShaderIR::GetPredicateCombiner(PredOperation operation) { - static constexpr std::array operation_table{ - OperationCode::LogicalAnd, - OperationCode::LogicalOr, - OperationCode::LogicalXor, - }; - - const auto index = static_cast<std::size_t>(operation); - if (index >= operation_table.size()) { - UNIMPLEMENTED_MSG("Unknown predicate operation."); - return {}; - } - - return operation_table[index]; -} - -Node ShaderIR::GetConditionCode(ConditionCode cc) const { - switch (cc) { - case ConditionCode::NEU: - return GetInternalFlag(InternalFlag::Zero, true); - case ConditionCode::FCSM_TR: - UNIMPLEMENTED_MSG("EXIT.FCSM_TR is not implemented"); - return MakeNode<PredicateNode>(Pred::NeverExecute, false); - default: - UNIMPLEMENTED_MSG("Unimplemented condition code: {}", cc); - return MakeNode<PredicateNode>(Pred::NeverExecute, false); - } -} - -void ShaderIR::SetRegister(NodeBlock& bb, Register dest, Node src) { - bb.push_back(Operation(OperationCode::Assign, GetRegister(dest), std::move(src))); -} - -void ShaderIR::SetPredicate(NodeBlock& bb, u64 dest, Node src) { - bb.push_back(Operation(OperationCode::LogicalAssign, GetPredicate(dest), std::move(src))); -} - -void ShaderIR::SetInternalFlag(NodeBlock& bb, InternalFlag flag, Node value) { - bb.push_back(Operation(OperationCode::LogicalAssign, GetInternalFlag(flag), std::move(value))); -} - -void ShaderIR::SetLocalMemory(NodeBlock& bb, Node address, Node value) { - bb.push_back( - Operation(OperationCode::Assign, GetLocalMemory(std::move(address)), std::move(value))); -} - -void ShaderIR::SetSharedMemory(NodeBlock& bb, Node address, Node value) { - bb.push_back( - Operation(OperationCode::Assign, GetSharedMemory(std::move(address)), std::move(value))); -} - -void ShaderIR::SetTemporary(NodeBlock& bb, u32 id, Node value) { - SetRegister(bb, Register::ZeroIndex + 1 + id, std::move(value)); -} - -void ShaderIR::SetInternalFlagsFromFloat(NodeBlock& bb, Node value, bool sets_cc) { - if (!sets_cc) { - return; - } - Node zerop = Operation(OperationCode::LogicalFOrdEqual, std::move(value), Immediate(0.0f)); - SetInternalFlag(bb, InternalFlag::Zero, std::move(zerop)); - LOG_WARNING(HW_GPU, "Condition codes implementation is incomplete"); -} - -void ShaderIR::SetInternalFlagsFromInteger(NodeBlock& bb, Node value, bool sets_cc) { - if (!sets_cc) { - return; - } - Node zerop = Operation(OperationCode::LogicalIEqual, std::move(value), Immediate(0)); - SetInternalFlag(bb, InternalFlag::Zero, std::move(zerop)); - LOG_WARNING(HW_GPU, "Condition codes implementation is incomplete"); -} - -Node ShaderIR::BitfieldExtract(Node value, u32 offset, u32 bits) { - return Operation(OperationCode::UBitfieldExtract, NO_PRECISE, std::move(value), - Immediate(offset), Immediate(bits)); -} - -Node ShaderIR::BitfieldInsert(Node base, Node insert, u32 offset, u32 bits) { - return Operation(OperationCode::UBitfieldInsert, NO_PRECISE, base, insert, Immediate(offset), - Immediate(bits)); -} - -void ShaderIR::MarkAttributeUsage(Attribute::Index index, u64 element) { - switch (index) { - case Attribute::Index::LayerViewportPointSize: - switch (element) { - case 0: - UNIMPLEMENTED(); - break; - case 1: - uses_layer = true; - break; - case 2: - uses_viewport_index = true; - break; - case 3: - uses_point_size = true; - break; - } - break; - case Attribute::Index::TessCoordInstanceIDVertexID: - switch (element) { - case 2: - uses_instance_id = true; - break; - case 3: - uses_vertex_id = true; - break; - } - break; - case Attribute::Index::ClipDistances0123: - case Attribute::Index::ClipDistances4567: { - const u64 clip_index = (index == Attribute::Index::ClipDistances4567 ? 4 : 0) + element; - used_clip_distances.at(clip_index) = true; - break; - } - case Attribute::Index::FrontColor: - case Attribute::Index::FrontSecondaryColor: - case Attribute::Index::BackColor: - case Attribute::Index::BackSecondaryColor: - uses_legacy_varyings = true; - break; - default: - if (index >= Attribute::Index::TexCoord_0 && index <= Attribute::Index::TexCoord_7) { - uses_legacy_varyings = true; - } - break; - } -} - -std::size_t ShaderIR::DeclareAmend(Node new_amend) { - const auto id = amend_code.size(); - amend_code.push_back(std::move(new_amend)); - return id; -} - -u32 ShaderIR::NewCustomVariable() { - return num_custom_variables++; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h deleted file mode 100644 index 1cd7c14d7..000000000 --- a/src/video_core/shader/shader_ir.h +++ /dev/null @@ -1,479 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <array> -#include <list> -#include <map> -#include <optional> -#include <set> -#include <tuple> -#include <vector> - -#include "common/common_types.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/engines/shader_header.h" -#include "video_core/shader/ast.h" -#include "video_core/shader/compiler_settings.h" -#include "video_core/shader/memory_util.h" -#include "video_core/shader/node.h" -#include "video_core/shader/registry.h" - -namespace VideoCommon::Shader { - -struct ShaderBlock; - -constexpr u32 MAX_PROGRAM_LENGTH = 0x1000; - -struct ConstBuffer { - constexpr explicit ConstBuffer(u32 max_offset_, bool is_indirect_) - : max_offset{max_offset_}, is_indirect{is_indirect_} {} - - constexpr ConstBuffer() = default; - - void MarkAsUsed(u64 offset) { - max_offset = std::max(max_offset, static_cast<u32>(offset)); - } - - void MarkAsUsedIndirect() { - is_indirect = true; - } - - bool IsIndirect() const { - return is_indirect; - } - - u32 GetSize() const { - return max_offset + static_cast<u32>(sizeof(float)); - } - - u32 GetMaxOffset() const { - return max_offset; - } - -private: - u32 max_offset = 0; - bool is_indirect = false; -}; - -struct GlobalMemoryUsage { - bool is_read{}; - bool is_written{}; -}; - -class ShaderIR final { -public: - explicit ShaderIR(const ProgramCode& program_code_, u32 main_offset_, - CompilerSettings settings_, Registry& registry_); - ~ShaderIR(); - - const std::map<u32, NodeBlock>& GetBasicBlocks() const { - return basic_blocks; - } - - const std::set<u32>& GetRegisters() const { - return used_registers; - } - - const std::set<Tegra::Shader::Pred>& GetPredicates() const { - return used_predicates; - } - - const std::set<Tegra::Shader::Attribute::Index>& GetInputAttributes() const { - return used_input_attributes; - } - - const std::set<Tegra::Shader::Attribute::Index>& GetOutputAttributes() const { - return used_output_attributes; - } - - const std::map<u32, ConstBuffer>& GetConstantBuffers() const { - return used_cbufs; - } - - const std::list<SamplerEntry>& GetSamplers() const { - return used_samplers; - } - - const std::list<ImageEntry>& GetImages() const { - return used_images; - } - - const std::array<bool, Tegra::Engines::Maxwell3D::Regs::NumClipDistances>& GetClipDistances() - const { - return used_clip_distances; - } - - const std::map<GlobalMemoryBase, GlobalMemoryUsage>& GetGlobalMemory() const { - return used_global_memory; - } - - std::size_t GetLength() const { - return static_cast<std::size_t>(coverage_end * sizeof(u64)); - } - - bool UsesLayer() const { - return uses_layer; - } - - bool UsesViewportIndex() const { - return uses_viewport_index; - } - - bool UsesPointSize() const { - return uses_point_size; - } - - bool UsesInstanceId() const { - return uses_instance_id; - } - - bool UsesVertexId() const { - return uses_vertex_id; - } - - bool UsesLegacyVaryings() const { - return uses_legacy_varyings; - } - - bool UsesYNegate() const { - return uses_y_negate; - } - - bool UsesWarps() const { - return uses_warps; - } - - bool HasPhysicalAttributes() const { - return uses_physical_attributes; - } - - const Tegra::Shader::Header& GetHeader() const { - return header; - } - - bool IsFlowStackDisabled() const { - return disable_flow_stack; - } - - bool IsDecompiled() const { - return decompiled; - } - - const ASTManager& GetASTManager() const { - return program_manager; - } - - ASTNode GetASTProgram() const { - return program_manager.GetProgram(); - } - - u32 GetASTNumVariables() const { - return program_manager.GetVariables(); - } - - u32 ConvertAddressToNvidiaSpace(u32 address) const { - return (address - main_offset) * static_cast<u32>(sizeof(Tegra::Shader::Instruction)); - } - - /// Returns a condition code evaluated from internal flags - Node GetConditionCode(Tegra::Shader::ConditionCode cc) const; - - const Node& GetAmendNode(std::size_t index) const { - return amend_code[index]; - } - - u32 GetNumCustomVariables() const { - return num_custom_variables; - } - -private: - friend class ASTDecoder; - - struct SamplerInfo { - std::optional<Tegra::Shader::TextureType> type; - std::optional<bool> is_array; - std::optional<bool> is_shadow; - std::optional<bool> is_buffer; - - constexpr bool IsComplete() const noexcept { - return type && is_array && is_shadow && is_buffer; - } - }; - - void Decode(); - void PostDecode(); - - NodeBlock DecodeRange(u32 begin, u32 end); - void DecodeRangeInner(NodeBlock& bb, u32 begin, u32 end); - void InsertControlFlow(NodeBlock& bb, const ShaderBlock& block); - - /** - * Decodes a single instruction from Tegra to IR. - * @param bb Basic block where the nodes will be written to. - * @param pc Program counter. Offset to decode. - * @return Next address to decode. - */ - u32 DecodeInstr(NodeBlock& bb, u32 pc); - - u32 DecodeArithmetic(NodeBlock& bb, u32 pc); - u32 DecodeArithmeticImmediate(NodeBlock& bb, u32 pc); - u32 DecodeBfe(NodeBlock& bb, u32 pc); - u32 DecodeBfi(NodeBlock& bb, u32 pc); - u32 DecodeShift(NodeBlock& bb, u32 pc); - u32 DecodeArithmeticInteger(NodeBlock& bb, u32 pc); - u32 DecodeArithmeticIntegerImmediate(NodeBlock& bb, u32 pc); - u32 DecodeArithmeticHalf(NodeBlock& bb, u32 pc); - u32 DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc); - u32 DecodeFfma(NodeBlock& bb, u32 pc); - u32 DecodeHfma2(NodeBlock& bb, u32 pc); - u32 DecodeConversion(NodeBlock& bb, u32 pc); - u32 DecodeWarp(NodeBlock& bb, u32 pc); - u32 DecodeMemory(NodeBlock& bb, u32 pc); - u32 DecodeTexture(NodeBlock& bb, u32 pc); - u32 DecodeImage(NodeBlock& bb, u32 pc); - u32 DecodeFloatSetPredicate(NodeBlock& bb, u32 pc); - u32 DecodeIntegerSetPredicate(NodeBlock& bb, u32 pc); - u32 DecodeHalfSetPredicate(NodeBlock& bb, u32 pc); - u32 DecodePredicateSetRegister(NodeBlock& bb, u32 pc); - u32 DecodePredicateSetPredicate(NodeBlock& bb, u32 pc); - u32 DecodeRegisterSetPredicate(NodeBlock& bb, u32 pc); - u32 DecodeFloatSet(NodeBlock& bb, u32 pc); - u32 DecodeIntegerSet(NodeBlock& bb, u32 pc); - u32 DecodeHalfSet(NodeBlock& bb, u32 pc); - u32 DecodeVideo(NodeBlock& bb, u32 pc); - u32 DecodeXmad(NodeBlock& bb, u32 pc); - u32 DecodeOther(NodeBlock& bb, u32 pc); - - /// Generates a node for a passed register. - Node GetRegister(Tegra::Shader::Register reg); - /// Generates a node for a custom variable - Node GetCustomVariable(u32 id); - /// Generates a node representing a 19-bit immediate value - Node GetImmediate19(Tegra::Shader::Instruction instr); - /// Generates a node representing a 32-bit immediate value - Node GetImmediate32(Tegra::Shader::Instruction instr); - /// Generates a node representing a constant buffer - Node GetConstBuffer(u64 index, u64 offset); - /// Generates a node representing a constant buffer with a variadic offset - Node GetConstBufferIndirect(u64 index, u64 offset, Node node); - /// Generates a node for a passed predicate. It can be optionally negated - Node GetPredicate(u64 pred, bool negated = false); - /// Generates a predicate node for an immediate true or false value - Node GetPredicate(bool immediate); - /// Generates a node representing an input attribute. Keeps track of used attributes. - Node GetInputAttribute(Tegra::Shader::Attribute::Index index, u64 element, Node buffer = {}); - /// Generates a node representing a physical input attribute. - Node GetPhysicalInputAttribute(Tegra::Shader::Register physical_address, Node buffer = {}); - /// Generates a node representing an output attribute. Keeps track of used attributes. - Node GetOutputAttribute(Tegra::Shader::Attribute::Index index, u64 element, Node buffer); - /// Generates a node representing an internal flag - Node GetInternalFlag(InternalFlag flag, bool negated = false) const; - /// Generates a node representing a local memory address - Node GetLocalMemory(Node address); - /// Generates a node representing a shared memory address - Node GetSharedMemory(Node address); - /// Generates a temporary, internally it uses a post-RZ register - Node GetTemporary(u32 id); - - /// Sets a register. src value must be a number-evaluated node. - void SetRegister(NodeBlock& bb, Tegra::Shader::Register dest, Node src); - /// Sets a predicate. src value must be a bool-evaluated node - void SetPredicate(NodeBlock& bb, u64 dest, Node src); - /// Sets an internal flag. src value must be a bool-evaluated node - void SetInternalFlag(NodeBlock& bb, InternalFlag flag, Node value); - /// Sets a local memory address with a value. - void SetLocalMemory(NodeBlock& bb, Node address, Node value); - /// Sets a shared memory address with a value. - void SetSharedMemory(NodeBlock& bb, Node address, Node value); - /// Sets a temporary. Internally it uses a post-RZ register - void SetTemporary(NodeBlock& bb, u32 id, Node value); - - /// Sets internal flags from a float - void SetInternalFlagsFromFloat(NodeBlock& bb, Node value, bool sets_cc = true); - /// Sets internal flags from an integer - void SetInternalFlagsFromInteger(NodeBlock& bb, Node value, bool sets_cc = true); - - /// Conditionally absolute/negated float. Absolute is applied first - Node GetOperandAbsNegFloat(Node value, bool absolute, bool negate); - /// Conditionally saturates a float - Node GetSaturatedFloat(Node value, bool saturate = true); - - /// Converts an integer to different sizes. - Node ConvertIntegerSize(Node value, Tegra::Shader::Register::Size size, bool is_signed); - /// Conditionally absolute/negated integer. Absolute is applied first - Node GetOperandAbsNegInteger(Node value, bool absolute, bool negate, bool is_signed); - - /// Unpacks a half immediate from an instruction - Node UnpackHalfImmediate(Tegra::Shader::Instruction instr, bool has_negation); - /// Unpacks a binary value into a half float pair with a type format - Node UnpackHalfFloat(Node value, Tegra::Shader::HalfType type); - /// Merges a half pair into another value - Node HalfMerge(Node dest, Node src, Tegra::Shader::HalfMerge merge); - /// Conditionally absolute/negated half float pair. Absolute is applied first - Node GetOperandAbsNegHalf(Node value, bool absolute, bool negate); - /// Conditionally saturates a half float pair - Node GetSaturatedHalfFloat(Node value, bool saturate = true); - - /// Get image component value by type and size - std::pair<Node, bool> GetComponentValue(Tegra::Texture::ComponentType component_type, - u32 component_size, Node original_value); - - /// Returns a predicate comparing two floats - Node GetPredicateComparisonFloat(Tegra::Shader::PredCondition condition, Node op_a, Node op_b); - /// Returns a predicate comparing two integers - Node GetPredicateComparisonInteger(Tegra::Shader::PredCondition condition, bool is_signed, - Node op_a, Node op_b); - /// Returns a predicate comparing two half floats. meta consumes how both pairs will be compared - Node GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition, Node op_a, Node op_b); - - /// Returns a predicate combiner operation - OperationCode GetPredicateCombiner(Tegra::Shader::PredOperation operation); - - /// Queries the missing sampler info from the execution context. - SamplerInfo GetSamplerInfo(SamplerInfo info, - std::optional<Tegra::Engines::SamplerDescriptor> sampler); - - /// Accesses a texture sampler. - std::optional<SamplerEntry> GetSampler(Tegra::Shader::Sampler sampler, SamplerInfo info); - - /// Accesses a texture sampler for a bindless texture. - std::optional<SamplerEntry> GetBindlessSampler(Tegra::Shader::Register reg, SamplerInfo info, - Node& index_var); - - /// Accesses an image. - ImageEntry& GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type); - - /// Access a bindless image sampler. - ImageEntry& GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type); - - /// Extracts a sequence of bits from a node - Node BitfieldExtract(Node value, u32 offset, u32 bits); - - /// Inserts a sequence of bits from a node - Node BitfieldInsert(Node base, Node insert, u32 offset, u32 bits); - - /// Marks the usage of a input or output attribute. - void MarkAttributeUsage(Tegra::Shader::Attribute::Index index, u64 element); - - /// Decodes VMNMX instruction and inserts its code into the passed basic block. - void DecodeVMNMX(NodeBlock& bb, Tegra::Shader::Instruction instr); - - void WriteTexInstructionFloat(NodeBlock& bb, Tegra::Shader::Instruction instr, - const Node4& components); - - void WriteTexsInstructionFloat(NodeBlock& bb, Tegra::Shader::Instruction instr, - const Node4& components, bool ignore_mask = false); - void WriteTexsInstructionHalfFloat(NodeBlock& bb, Tegra::Shader::Instruction instr, - const Node4& components, bool ignore_mask = false); - - Node4 GetTexCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type, - Tegra::Shader::TextureProcessMode process_mode, bool depth_compare, - bool is_array, bool is_aoffi, - std::optional<Tegra::Shader::Register> bindless_reg); - - Node4 GetTexsCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type, - Tegra::Shader::TextureProcessMode process_mode, bool depth_compare, - bool is_array); - - Node4 GetTld4Code(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type, - bool depth_compare, bool is_array, bool is_aoffi, bool is_ptp, - bool is_bindless); - - Node4 GetTldCode(Tegra::Shader::Instruction instr); - - Node4 GetTldsCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type, - bool is_array); - - std::tuple<std::size_t, std::size_t> ValidateAndGetCoordinateElement( - Tegra::Shader::TextureType texture_type, bool depth_compare, bool is_array, - bool lod_bias_enabled, std::size_t max_coords, std::size_t max_inputs); - - std::vector<Node> GetAoffiCoordinates(Node aoffi_reg, std::size_t coord_count, bool is_tld4); - - std::vector<Node> GetPtpCoordinates(std::array<Node, 2> ptp_regs); - - Node4 GetTextureCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type, - Tegra::Shader::TextureProcessMode process_mode, std::vector<Node> coords, - Node array, Node depth_compare, u32 bias_offset, std::vector<Node> aoffi, - std::optional<Tegra::Shader::Register> bindless_reg); - - Node GetVideoOperand(Node op, bool is_chunk, bool is_signed, Tegra::Shader::VideoType type, - u64 byte_height); - - void WriteLogicOperation(NodeBlock& bb, Tegra::Shader::Register dest, - Tegra::Shader::LogicOperation logic_op, Node op_a, Node op_b, - Tegra::Shader::PredicateResultMode predicate_mode, - Tegra::Shader::Pred predicate, bool sets_cc); - void WriteLop3Instruction(NodeBlock& bb, Tegra::Shader::Register dest, Node op_a, Node op_b, - Node op_c, Node imm_lut, bool sets_cc); - - std::tuple<Node, u32, u32> TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) const; - - std::pair<Node, TrackSampler> TrackBindlessSampler(Node tracked, const NodeBlock& code, - s64 cursor); - - std::pair<Node, TrackSampler> HandleBindlessIndirectRead(const CbufNode& cbuf, - const OperationNode& operation, - Node gpr, Node base_offset, - Node tracked, const NodeBlock& code, - s64 cursor); - - std::optional<u32> TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor) const; - - std::pair<Node, s64> TrackRegister(const GprNode* tracked, const NodeBlock& code, - s64 cursor) const; - - std::tuple<Node, Node, GlobalMemoryBase> TrackGlobalMemory(NodeBlock& bb, - Tegra::Shader::Instruction instr, - bool is_read, bool is_write); - - /// Register new amending code and obtain the reference id. - std::size_t DeclareAmend(Node new_amend); - - u32 NewCustomVariable(); - - const ProgramCode& program_code; - const u32 main_offset; - const CompilerSettings settings; - Registry& registry; - - bool decompiled{}; - bool disable_flow_stack{}; - - u32 coverage_begin{}; - u32 coverage_end{}; - - std::map<u32, NodeBlock> basic_blocks; - NodeBlock global_code; - ASTManager program_manager{true, true}; - std::vector<Node> amend_code; - u32 num_custom_variables{}; - - std::set<u32> used_registers; - std::set<Tegra::Shader::Pred> used_predicates; - std::set<Tegra::Shader::Attribute::Index> used_input_attributes; - std::set<Tegra::Shader::Attribute::Index> used_output_attributes; - std::map<u32, ConstBuffer> used_cbufs; - std::list<SamplerEntry> used_samplers; - std::list<ImageEntry> used_images; - std::array<bool, Tegra::Engines::Maxwell3D::Regs::NumClipDistances> used_clip_distances{}; - std::map<GlobalMemoryBase, GlobalMemoryUsage> used_global_memory; - bool uses_layer{}; - bool uses_viewport_index{}; - bool uses_point_size{}; - bool uses_physical_attributes{}; // Shader uses AL2P or physical attribute read/writes - bool uses_instance_id{}; - bool uses_vertex_id{}; - bool uses_legacy_varyings{}; - bool uses_y_negate{}; - bool uses_warps{}; - bool uses_indexed_samplers{}; - - Tegra::Shader::Header header; -}; - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/track.cpp b/src/video_core/shader/track.cpp deleted file mode 100644 index 6be3ea92b..000000000 --- a/src/video_core/shader/track.cpp +++ /dev/null @@ -1,236 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <utility> -#include <variant> - -#include "common/common_types.h" -#include "video_core/shader/node.h" -#include "video_core/shader/node_helper.h" -#include "video_core/shader/shader_ir.h" - -namespace VideoCommon::Shader { - -namespace { - -std::pair<Node, s64> FindOperation(const NodeBlock& code, s64 cursor, - OperationCode operation_code) { - for (; cursor >= 0; --cursor) { - Node node = code.at(cursor); - - if (const auto operation = std::get_if<OperationNode>(&*node)) { - if (operation->GetCode() == operation_code) { - return {std::move(node), cursor}; - } - } - - if (const auto conditional = std::get_if<ConditionalNode>(&*node)) { - const auto& conditional_code = conditional->GetCode(); - auto result = FindOperation( - conditional_code, static_cast<s64>(conditional_code.size() - 1), operation_code); - auto& found = result.first; - if (found) { - return {std::move(found), cursor}; - } - } - } - return {}; -} - -std::optional<std::pair<Node, Node>> DecoupleIndirectRead(const OperationNode& operation) { - if (operation.GetCode() != OperationCode::UAdd) { - return std::nullopt; - } - Node gpr; - Node offset; - ASSERT(operation.GetOperandsCount() == 2); - for (std::size_t i = 0; i < operation.GetOperandsCount(); i++) { - Node operand = operation[i]; - if (std::holds_alternative<ImmediateNode>(*operand)) { - offset = operation[i]; - } else if (std::holds_alternative<GprNode>(*operand)) { - gpr = operation[i]; - } - } - if (offset && gpr) { - return std::make_pair(gpr, offset); - } - return std::nullopt; -} - -bool AmendNodeCv(std::size_t amend_index, Node node) { - if (const auto operation = std::get_if<OperationNode>(&*node)) { - operation->SetAmendIndex(amend_index); - return true; - } - if (const auto conditional = std::get_if<ConditionalNode>(&*node)) { - conditional->SetAmendIndex(amend_index); - return true; - } - return false; -} - -} // Anonymous namespace - -std::pair<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, const NodeBlock& code, - s64 cursor) { - if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) { - const u32 cbuf_index = cbuf->GetIndex(); - - // Constant buffer found, test if it's an immediate - const auto& offset = cbuf->GetOffset(); - if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) { - auto track = MakeTrackSampler<BindlessSamplerNode>(cbuf_index, immediate->GetValue()); - return {tracked, track}; - } - if (const auto operation = std::get_if<OperationNode>(&*offset)) { - const u32 bound_buffer = registry.GetBoundBuffer(); - if (bound_buffer != cbuf_index) { - return {}; - } - if (const std::optional pair = DecoupleIndirectRead(*operation)) { - auto [gpr, base_offset] = *pair; - return HandleBindlessIndirectRead(*cbuf, *operation, gpr, base_offset, tracked, - code, cursor); - } - } - return {}; - } - if (const auto gpr = std::get_if<GprNode>(&*tracked)) { - if (gpr->GetIndex() == Tegra::Shader::Register::ZeroIndex) { - return {}; - } - // Reduce the cursor in one to avoid infinite loops when the instruction sets the same - // register that it uses as operand - const auto [source, new_cursor] = TrackRegister(gpr, code, cursor - 1); - if (!source) { - return {}; - } - return TrackBindlessSampler(source, code, new_cursor); - } - if (const auto operation = std::get_if<OperationNode>(&*tracked)) { - const OperationNode& op = *operation; - - const OperationCode opcode = operation->GetCode(); - if (opcode == OperationCode::IBitwiseOr || opcode == OperationCode::UBitwiseOr) { - ASSERT(op.GetOperandsCount() == 2); - auto [node_a, index_a, offset_a] = TrackCbuf(op[0], code, cursor); - auto [node_b, index_b, offset_b] = TrackCbuf(op[1], code, cursor); - if (node_a && node_b) { - auto track = MakeTrackSampler<SeparateSamplerNode>(std::pair{index_a, index_b}, - std::pair{offset_a, offset_b}); - return {tracked, std::move(track)}; - } - } - std::size_t i = op.GetOperandsCount(); - while (i--) { - if (auto found = TrackBindlessSampler(op[i - 1], code, cursor); std::get<0>(found)) { - // Constant buffer found in operand. - return found; - } - } - return {}; - } - if (const auto conditional = std::get_if<ConditionalNode>(&*tracked)) { - const auto& conditional_code = conditional->GetCode(); - return TrackBindlessSampler(tracked, conditional_code, - static_cast<s64>(conditional_code.size())); - } - return {}; -} - -std::pair<Node, TrackSampler> ShaderIR::HandleBindlessIndirectRead( - const CbufNode& cbuf, const OperationNode& operation, Node gpr, Node base_offset, Node tracked, - const NodeBlock& code, s64 cursor) { - const auto offset_imm = std::get<ImmediateNode>(*base_offset); - const auto& gpu_driver = registry.AccessGuestDriverProfile(); - const u32 bindless_cv = NewCustomVariable(); - const u32 texture_handler_size = gpu_driver.GetTextureHandlerSize(); - Node op = Operation(OperationCode::UDiv, gpr, Immediate(texture_handler_size)); - - Node cv_node = GetCustomVariable(bindless_cv); - Node amend_op = Operation(OperationCode::Assign, std::move(cv_node), std::move(op)); - const std::size_t amend_index = DeclareAmend(std::move(amend_op)); - AmendNodeCv(amend_index, code[cursor]); - - // TODO: Implement bindless index custom variable - auto track = - MakeTrackSampler<ArraySamplerNode>(cbuf.GetIndex(), offset_imm.GetValue(), bindless_cv); - return {tracked, track}; -} - -std::tuple<Node, u32, u32> ShaderIR::TrackCbuf(Node tracked, const NodeBlock& code, - s64 cursor) const { - if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) { - // Constant buffer found, test if it's an immediate - const auto& offset = cbuf->GetOffset(); - if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) { - return {tracked, cbuf->GetIndex(), immediate->GetValue()}; - } - return {}; - } - if (const auto gpr = std::get_if<GprNode>(&*tracked)) { - if (gpr->GetIndex() == Tegra::Shader::Register::ZeroIndex) { - return {}; - } - // Reduce the cursor in one to avoid infinite loops when the instruction sets the same - // register that it uses as operand - const auto [source, new_cursor] = TrackRegister(gpr, code, cursor - 1); - if (!source) { - return {}; - } - return TrackCbuf(source, code, new_cursor); - } - if (const auto operation = std::get_if<OperationNode>(&*tracked)) { - for (std::size_t i = operation->GetOperandsCount(); i > 0; --i) { - if (auto found = TrackCbuf((*operation)[i - 1], code, cursor); std::get<0>(found)) { - // Cbuf found in operand. - return found; - } - } - return {}; - } - if (const auto conditional = std::get_if<ConditionalNode>(&*tracked)) { - const auto& conditional_code = conditional->GetCode(); - return TrackCbuf(tracked, conditional_code, static_cast<s64>(conditional_code.size())); - } - return {}; -} - -std::optional<u32> ShaderIR::TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor) const { - // Reduce the cursor in one to avoid infinite loops when the instruction sets the same register - // that it uses as operand - const auto result = TrackRegister(&std::get<GprNode>(*tracked), code, cursor - 1); - const auto& found = result.first; - if (!found) { - return std::nullopt; - } - if (const auto immediate = std::get_if<ImmediateNode>(&*found)) { - return immediate->GetValue(); - } - return std::nullopt; -} - -std::pair<Node, s64> ShaderIR::TrackRegister(const GprNode* tracked, const NodeBlock& code, - s64 cursor) const { - for (; cursor >= 0; --cursor) { - const auto [found_node, new_cursor] = FindOperation(code, cursor, OperationCode::Assign); - if (!found_node) { - return {}; - } - const auto operation = std::get_if<OperationNode>(&*found_node); - ASSERT(operation); - - const auto& target = (*operation)[0]; - if (const auto gpr_target = std::get_if<GprNode>(&*target)) { - if (gpr_target->GetIndex() == tracked->GetIndex()) { - return {(*operation)[1], new_cursor}; - } - } - } - return {}; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/transform_feedback.cpp b/src/video_core/shader/transform_feedback.cpp deleted file mode 100644 index 22a933761..000000000 --- a/src/video_core/shader/transform_feedback.cpp +++ /dev/null @@ -1,115 +0,0 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <array> -#include <unordered_map> - -#include "common/assert.h" -#include "common/common_types.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/shader/registry.h" -#include "video_core/shader/transform_feedback.h" - -namespace VideoCommon::Shader { - -namespace { - -using Maxwell = Tegra::Engines::Maxwell3D::Regs; - -// TODO(Rodrigo): Change this to constexpr std::unordered_set in C++20 - -/// Attribute offsets that describe a vector -constexpr std::array VECTORS = { - 28, // gl_Position - 32, // Generic 0 - 36, // Generic 1 - 40, // Generic 2 - 44, // Generic 3 - 48, // Generic 4 - 52, // Generic 5 - 56, // Generic 6 - 60, // Generic 7 - 64, // Generic 8 - 68, // Generic 9 - 72, // Generic 10 - 76, // Generic 11 - 80, // Generic 12 - 84, // Generic 13 - 88, // Generic 14 - 92, // Generic 15 - 96, // Generic 16 - 100, // Generic 17 - 104, // Generic 18 - 108, // Generic 19 - 112, // Generic 20 - 116, // Generic 21 - 120, // Generic 22 - 124, // Generic 23 - 128, // Generic 24 - 132, // Generic 25 - 136, // Generic 26 - 140, // Generic 27 - 144, // Generic 28 - 148, // Generic 29 - 152, // Generic 30 - 156, // Generic 31 - 160, // gl_FrontColor - 164, // gl_FrontSecondaryColor - 160, // gl_BackColor - 164, // gl_BackSecondaryColor - 192, // gl_TexCoord[0] - 196, // gl_TexCoord[1] - 200, // gl_TexCoord[2] - 204, // gl_TexCoord[3] - 208, // gl_TexCoord[4] - 212, // gl_TexCoord[5] - 216, // gl_TexCoord[6] - 220, // gl_TexCoord[7] -}; -} // namespace - -std::unordered_map<u8, VaryingTFB> BuildTransformFeedback(const GraphicsInfo& info) { - - std::unordered_map<u8, VaryingTFB> tfb; - - for (std::size_t buffer = 0; buffer < Maxwell::NumTransformFeedbackBuffers; ++buffer) { - const auto& locations = info.tfb_varying_locs[buffer]; - const auto& layout = info.tfb_layouts[buffer]; - const std::size_t varying_count = layout.varying_count; - - std::size_t highest = 0; - - for (std::size_t offset = 0; offset < varying_count; ++offset) { - const std::size_t base_offset = offset; - const u8 location = locations[offset]; - - VaryingTFB varying; - varying.buffer = layout.stream; - varying.stride = layout.stride; - varying.offset = offset * sizeof(u32); - varying.components = 1; - - if (std::find(VECTORS.begin(), VECTORS.end(), location / 4 * 4) != VECTORS.end()) { - UNIMPLEMENTED_IF_MSG(location % 4 != 0, "Unaligned TFB"); - - const u8 base_index = location / 4; - while (offset + 1 < varying_count && base_index == locations[offset + 1] / 4) { - ++offset; - ++varying.components; - } - } - - [[maybe_unused]] const bool inserted = tfb.emplace(location, varying).second; - UNIMPLEMENTED_IF_MSG(!inserted, "Varying already stored"); - - highest = std::max(highest, (base_offset + varying.components) * sizeof(u32)); - } - - UNIMPLEMENTED_IF(highest != layout.stride); - } - return tfb; -} - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/transform_feedback.h b/src/video_core/shader/transform_feedback.h deleted file mode 100644 index 77d05f64c..000000000 --- a/src/video_core/shader/transform_feedback.h +++ /dev/null @@ -1,23 +0,0 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <unordered_map> - -#include "common/common_types.h" -#include "video_core/shader/registry.h" - -namespace VideoCommon::Shader { - -struct VaryingTFB { - std::size_t buffer; - std::size_t stride; - std::size_t offset; - std::size_t components; -}; - -std::unordered_map<u8, VaryingTFB> BuildTransformFeedback(const GraphicsInfo& info); - -} // namespace VideoCommon::Shader diff --git a/src/video_core/shader_cache.cpp b/src/video_core/shader_cache.cpp new file mode 100644 index 000000000..78bf90c48 --- /dev/null +++ b/src/video_core/shader_cache.cpp @@ -0,0 +1,250 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <algorithm> +#include <array> +#include <vector> + +#include "common/assert.h" +#include "shader_recompiler/frontend/maxwell/control_flow.h" +#include "shader_recompiler/object_pool.h" +#include "video_core/dirty_flags.h" +#include "video_core/engines/kepler_compute.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/memory_manager.h" +#include "video_core/shader_cache.h" +#include "video_core/shader_environment.h" + +namespace VideoCommon { + +void ShaderCache::InvalidateRegion(VAddr addr, size_t size) { + std::scoped_lock lock{invalidation_mutex}; + InvalidatePagesInRegion(addr, size); + RemovePendingShaders(); +} + +void ShaderCache::OnCPUWrite(VAddr addr, size_t size) { + std::lock_guard lock{invalidation_mutex}; + InvalidatePagesInRegion(addr, size); +} + +void ShaderCache::SyncGuestHost() { + std::scoped_lock lock{invalidation_mutex}; + RemovePendingShaders(); +} + +ShaderCache::ShaderCache(VideoCore::RasterizerInterface& rasterizer_, + Tegra::MemoryManager& gpu_memory_, Tegra::Engines::Maxwell3D& maxwell3d_, + Tegra::Engines::KeplerCompute& kepler_compute_) + : gpu_memory{gpu_memory_}, maxwell3d{maxwell3d_}, kepler_compute{kepler_compute_}, + rasterizer{rasterizer_} {} + +bool ShaderCache::RefreshStages(std::array<u64, 6>& unique_hashes) { + auto& dirty{maxwell3d.dirty.flags}; + if (!dirty[VideoCommon::Dirty::Shaders]) { + return last_shaders_valid; + } + dirty[VideoCommon::Dirty::Shaders] = false; + + const GPUVAddr base_addr{maxwell3d.regs.code_address.CodeAddress()}; + for (size_t index = 0; index < Tegra::Engines::Maxwell3D::Regs::MaxShaderProgram; ++index) { + if (!maxwell3d.regs.IsShaderConfigEnabled(index)) { + unique_hashes[index] = 0; + continue; + } + const auto& shader_config{maxwell3d.regs.shader_config[index]}; + const auto program{static_cast<Tegra::Engines::Maxwell3D::Regs::ShaderProgram>(index)}; + const GPUVAddr shader_addr{base_addr + shader_config.offset}; + const std::optional<VAddr> cpu_shader_addr{gpu_memory.GpuToCpuAddress(shader_addr)}; + if (!cpu_shader_addr) { + LOG_ERROR(HW_GPU, "Invalid GPU address for shader 0x{:016x}", shader_addr); + last_shaders_valid = false; + return false; + } + const ShaderInfo* shader_info{TryGet(*cpu_shader_addr)}; + if (!shader_info) { + const u32 start_address{shader_config.offset}; + GraphicsEnvironment env{maxwell3d, gpu_memory, program, base_addr, start_address}; + shader_info = MakeShaderInfo(env, *cpu_shader_addr); + } + shader_infos[index] = shader_info; + unique_hashes[index] = shader_info->unique_hash; + } + last_shaders_valid = true; + return true; +} + +const ShaderInfo* ShaderCache::ComputeShader() { + const GPUVAddr program_base{kepler_compute.regs.code_loc.Address()}; + const auto& qmd{kepler_compute.launch_description}; + const GPUVAddr shader_addr{program_base + qmd.program_start}; + const std::optional<VAddr> cpu_shader_addr{gpu_memory.GpuToCpuAddress(shader_addr)}; + if (!cpu_shader_addr) { + LOG_ERROR(HW_GPU, "Invalid GPU address for shader 0x{:016x}", shader_addr); + return nullptr; + } + if (const ShaderInfo* const shader = TryGet(*cpu_shader_addr)) { + return shader; + } + ComputeEnvironment env{kepler_compute, gpu_memory, program_base, qmd.program_start}; + return MakeShaderInfo(env, *cpu_shader_addr); +} + +void ShaderCache::GetGraphicsEnvironments(GraphicsEnvironments& result, + const std::array<u64, NUM_PROGRAMS>& unique_hashes) { + size_t env_index{}; + const GPUVAddr base_addr{maxwell3d.regs.code_address.CodeAddress()}; + for (size_t index = 0; index < NUM_PROGRAMS; ++index) { + if (unique_hashes[index] == 0) { + continue; + } + const auto program{static_cast<Tegra::Engines::Maxwell3D::Regs::ShaderProgram>(index)}; + auto& env{result.envs[index]}; + const u32 start_address{maxwell3d.regs.shader_config[index].offset}; + env = GraphicsEnvironment{maxwell3d, gpu_memory, program, base_addr, start_address}; + env.SetCachedSize(shader_infos[index]->size_bytes); + result.env_ptrs[env_index++] = &env; + } +} + +ShaderInfo* ShaderCache::TryGet(VAddr addr) const { + std::scoped_lock lock{lookup_mutex}; + + const auto it = lookup_cache.find(addr); + if (it == lookup_cache.end()) { + return nullptr; + } + return it->second->data; +} + +void ShaderCache::Register(std::unique_ptr<ShaderInfo> data, VAddr addr, size_t size) { + std::scoped_lock lock{invalidation_mutex, lookup_mutex}; + + const VAddr addr_end = addr + size; + Entry* const entry = NewEntry(addr, addr_end, data.get()); + + const u64 page_end = (addr_end + PAGE_SIZE - 1) >> PAGE_BITS; + for (u64 page = addr >> PAGE_BITS; page < page_end; ++page) { + invalidation_cache[page].push_back(entry); + } + + storage.push_back(std::move(data)); + + rasterizer.UpdatePagesCachedCount(addr, size, 1); +} + +void ShaderCache::InvalidatePagesInRegion(VAddr addr, size_t size) { + const VAddr addr_end = addr + size; + const u64 page_end = (addr_end + PAGE_SIZE - 1) >> PAGE_BITS; + for (u64 page = addr >> PAGE_BITS; page < page_end; ++page) { + auto it = invalidation_cache.find(page); + if (it == invalidation_cache.end()) { + continue; + } + InvalidatePageEntries(it->second, addr, addr_end); + } +} + +void ShaderCache::RemovePendingShaders() { + if (marked_for_removal.empty()) { + return; + } + // Remove duplicates + std::ranges::sort(marked_for_removal); + marked_for_removal.erase(std::unique(marked_for_removal.begin(), marked_for_removal.end()), + marked_for_removal.end()); + + std::vector<ShaderInfo*> removed_shaders; + removed_shaders.reserve(marked_for_removal.size()); + + std::scoped_lock lock{lookup_mutex}; + + for (Entry* const entry : marked_for_removal) { + removed_shaders.push_back(entry->data); + + const auto it = lookup_cache.find(entry->addr_start); + ASSERT(it != lookup_cache.end()); + lookup_cache.erase(it); + } + marked_for_removal.clear(); + + if (!removed_shaders.empty()) { + RemoveShadersFromStorage(std::move(removed_shaders)); + } +} + +void ShaderCache::InvalidatePageEntries(std::vector<Entry*>& entries, VAddr addr, VAddr addr_end) { + size_t index = 0; + while (index < entries.size()) { + Entry* const entry = entries[index]; + if (!entry->Overlaps(addr, addr_end)) { + ++index; + continue; + } + + UnmarkMemory(entry); + RemoveEntryFromInvalidationCache(entry); + marked_for_removal.push_back(entry); + } +} + +void ShaderCache::RemoveEntryFromInvalidationCache(const Entry* entry) { + const u64 page_end = (entry->addr_end + PAGE_SIZE - 1) >> PAGE_BITS; + for (u64 page = entry->addr_start >> PAGE_BITS; page < page_end; ++page) { + const auto entries_it = invalidation_cache.find(page); + ASSERT(entries_it != invalidation_cache.end()); + std::vector<Entry*>& entries = entries_it->second; + + const auto entry_it = std::ranges::find(entries, entry); + ASSERT(entry_it != entries.end()); + entries.erase(entry_it); + } +} + +void ShaderCache::UnmarkMemory(Entry* entry) { + if (!entry->is_memory_marked) { + return; + } + entry->is_memory_marked = false; + + const VAddr addr = entry->addr_start; + const size_t size = entry->addr_end - addr; + rasterizer.UpdatePagesCachedCount(addr, size, -1); +} + +void ShaderCache::RemoveShadersFromStorage(std::vector<ShaderInfo*> removed_shaders) { + // Remove them from the cache + std::erase_if(storage, [&removed_shaders](const std::unique_ptr<ShaderInfo>& shader) { + return std::ranges::find(removed_shaders, shader.get()) != removed_shaders.end(); + }); +} + +ShaderCache::Entry* ShaderCache::NewEntry(VAddr addr, VAddr addr_end, ShaderInfo* data) { + auto entry = std::make_unique<Entry>(Entry{addr, addr_end, data}); + Entry* const entry_pointer = entry.get(); + + lookup_cache.emplace(addr, std::move(entry)); + return entry_pointer; +} + +const ShaderInfo* ShaderCache::MakeShaderInfo(GenericEnvironment& env, VAddr cpu_addr) { + auto info = std::make_unique<ShaderInfo>(); + if (const std::optional<u64> cached_hash{env.Analyze()}) { + info->unique_hash = *cached_hash; + info->size_bytes = env.CachedSize(); + } else { + // Slow path, not really hit on commercial games + // Build a control flow graph to get the real shader size + Shader::ObjectPool<Shader::Maxwell::Flow::Block> flow_block; + Shader::Maxwell::Flow::CFG cfg{env, flow_block, env.StartAddress()}; + info->unique_hash = env.CalculateHash(); + info->size_bytes = env.ReadSize(); + } + const size_t size_bytes{info->size_bytes}; + const ShaderInfo* const result{info.get()}; + Register(std::move(info), cpu_addr, size_bytes); + return result; +} + +} // namespace VideoCommon diff --git a/src/video_core/shader_cache.h b/src/video_core/shader_cache.h index 015a789d6..136fe294c 100644 --- a/src/video_core/shader_cache.h +++ b/src/video_core/shader_cache.h @@ -5,226 +5,147 @@ #pragma once #include <algorithm> +#include <array> #include <memory> #include <mutex> +#include <span> #include <unordered_map> #include <utility> #include <vector> -#include "common/assert.h" #include "common/common_types.h" #include "video_core/rasterizer_interface.h" +#include "video_core/shader_environment.h" + +namespace Tegra { +class MemoryManager; +} namespace VideoCommon { -template <class T> +class GenericEnvironment; + +struct ShaderInfo { + u64 unique_hash{}; + size_t size_bytes{}; +}; + class ShaderCache { static constexpr u64 PAGE_BITS = 14; static constexpr u64 PAGE_SIZE = u64(1) << PAGE_BITS; + static constexpr size_t NUM_PROGRAMS = 6; + struct Entry { VAddr addr_start; VAddr addr_end; - T* data; + ShaderInfo* data; bool is_memory_marked = true; - constexpr bool Overlaps(VAddr start, VAddr end) const noexcept { + bool Overlaps(VAddr start, VAddr end) const noexcept { return start < addr_end && addr_start < end; } }; public: - virtual ~ShaderCache() = default; - /// @brief Removes shaders inside a given region /// @note Checks for ranges /// @param addr Start address of the invalidation /// @param size Number of bytes of the invalidation - void InvalidateRegion(VAddr addr, std::size_t size) { - std::scoped_lock lock{invalidation_mutex}; - InvalidatePagesInRegion(addr, size); - RemovePendingShaders(); - } + void InvalidateRegion(VAddr addr, size_t size); /// @brief Unmarks a memory region as cached and marks it for removal /// @param addr Start address of the CPU write operation /// @param size Number of bytes of the CPU write operation - void OnCPUWrite(VAddr addr, std::size_t size) { - std::lock_guard lock{invalidation_mutex}; - InvalidatePagesInRegion(addr, size); - } + void OnCPUWrite(VAddr addr, size_t size); /// @brief Flushes delayed removal operations - void SyncGuestHost() { - std::scoped_lock lock{invalidation_mutex}; - RemovePendingShaders(); - } + void SyncGuestHost(); - /// @brief Tries to obtain a cached shader starting in a given address - /// @note Doesn't check for ranges, the given address has to be the start of the shader - /// @param addr Start address of the shader, this doesn't cache for region - /// @return Pointer to a valid shader, nullptr when nothing is found - T* TryGet(VAddr addr) const { - std::scoped_lock lock{lookup_mutex}; +protected: + struct GraphicsEnvironments { + std::array<GraphicsEnvironment, NUM_PROGRAMS> envs; + std::array<Shader::Environment*, NUM_PROGRAMS> env_ptrs; - const auto it = lookup_cache.find(addr); - if (it == lookup_cache.end()) { - return nullptr; + std::span<Shader::Environment* const> Span() const noexcept { + return std::span(env_ptrs.begin(), std::ranges::find(env_ptrs, nullptr)); } - return it->second->data; - } - -protected: - explicit ShaderCache(VideoCore::RasterizerInterface& rasterizer_) : rasterizer{rasterizer_} {} + }; - /// @brief Register in the cache a given entry - /// @param data Shader to store in the cache - /// @param addr Start address of the shader that will be registered - /// @param size Size in bytes of the shader - void Register(std::unique_ptr<T> data, VAddr addr, std::size_t size) { - std::scoped_lock lock{invalidation_mutex, lookup_mutex}; + explicit ShaderCache(VideoCore::RasterizerInterface& rasterizer_, + Tegra::MemoryManager& gpu_memory_, Tegra::Engines::Maxwell3D& maxwell3d_, + Tegra::Engines::KeplerCompute& kepler_compute_); - const VAddr addr_end = addr + size; - Entry* const entry = NewEntry(addr, addr_end, data.get()); + /// @brief Update the hashes and information of shader stages + /// @param unique_hashes Shader hashes to store into when a stage is enabled + /// @return True no success, false on error + bool RefreshStages(std::array<u64, NUM_PROGRAMS>& unique_hashes); - const u64 page_end = (addr_end + PAGE_SIZE - 1) >> PAGE_BITS; - for (u64 page = addr >> PAGE_BITS; page < page_end; ++page) { - invalidation_cache[page].push_back(entry); - } + /// @brief Returns information about the current compute shader + /// @return Pointer to a valid shader, nullptr on error + const ShaderInfo* ComputeShader(); - storage.push_back(std::move(data)); + /// @brief Collect the current graphics environments + void GetGraphicsEnvironments(GraphicsEnvironments& result, + const std::array<u64, NUM_PROGRAMS>& unique_hashes); - rasterizer.UpdatePagesCachedCount(addr, size, 1); - } + Tegra::MemoryManager& gpu_memory; + Tegra::Engines::Maxwell3D& maxwell3d; + Tegra::Engines::KeplerCompute& kepler_compute; - /// @brief Called when a shader is going to be removed - /// @param shader Shader that will be removed - /// @pre invalidation_cache is locked - /// @pre lookup_mutex is locked - virtual void OnShaderRemoval([[maybe_unused]] T* shader) {} + std::array<const ShaderInfo*, NUM_PROGRAMS> shader_infos{}; + bool last_shaders_valid = false; private: + /// @brief Tries to obtain a cached shader starting in a given address + /// @note Doesn't check for ranges, the given address has to be the start of the shader + /// @param addr Start address of the shader, this doesn't cache for region + /// @return Pointer to a valid shader, nullptr when nothing is found + ShaderInfo* TryGet(VAddr addr) const; + + /// @brief Register in the cache a given entry + /// @param data Shader to store in the cache + /// @param addr Start address of the shader that will be registered + /// @param size Size in bytes of the shader + void Register(std::unique_ptr<ShaderInfo> data, VAddr addr, size_t size); + /// @brief Invalidate pages in a given region /// @pre invalidation_mutex is locked - void InvalidatePagesInRegion(VAddr addr, std::size_t size) { - const VAddr addr_end = addr + size; - const u64 page_end = (addr_end + PAGE_SIZE - 1) >> PAGE_BITS; - for (u64 page = addr >> PAGE_BITS; page < page_end; ++page) { - auto it = invalidation_cache.find(page); - if (it == invalidation_cache.end()) { - continue; - } - InvalidatePageEntries(it->second, addr, addr_end); - } - } + void InvalidatePagesInRegion(VAddr addr, size_t size); /// @brief Remove shaders marked for deletion /// @pre invalidation_mutex is locked - void RemovePendingShaders() { - if (marked_for_removal.empty()) { - return; - } - // Remove duplicates - std::sort(marked_for_removal.begin(), marked_for_removal.end()); - marked_for_removal.erase(std::unique(marked_for_removal.begin(), marked_for_removal.end()), - marked_for_removal.end()); - - std::vector<T*> removed_shaders; - removed_shaders.reserve(marked_for_removal.size()); - - std::scoped_lock lock{lookup_mutex}; - - for (Entry* const entry : marked_for_removal) { - removed_shaders.push_back(entry->data); - - const auto it = lookup_cache.find(entry->addr_start); - ASSERT(it != lookup_cache.end()); - lookup_cache.erase(it); - } - marked_for_removal.clear(); - - if (!removed_shaders.empty()) { - RemoveShadersFromStorage(std::move(removed_shaders)); - } - } + void RemovePendingShaders(); /// @brief Invalidates entries in a given range for the passed page /// @param entries Vector of entries in the page, it will be modified on overlaps /// @param addr Start address of the invalidation /// @param addr_end Non-inclusive end address of the invalidation /// @pre invalidation_mutex is locked - void InvalidatePageEntries(std::vector<Entry*>& entries, VAddr addr, VAddr addr_end) { - std::size_t index = 0; - while (index < entries.size()) { - Entry* const entry = entries[index]; - if (!entry->Overlaps(addr, addr_end)) { - ++index; - continue; - } - - UnmarkMemory(entry); - RemoveEntryFromInvalidationCache(entry); - marked_for_removal.push_back(entry); - } - } + void InvalidatePageEntries(std::vector<Entry*>& entries, VAddr addr, VAddr addr_end); /// @brief Removes all references to an entry in the invalidation cache /// @param entry Entry to remove from the invalidation cache /// @pre invalidation_mutex is locked - void RemoveEntryFromInvalidationCache(const Entry* entry) { - const u64 page_end = (entry->addr_end + PAGE_SIZE - 1) >> PAGE_BITS; - for (u64 page = entry->addr_start >> PAGE_BITS; page < page_end; ++page) { - const auto entries_it = invalidation_cache.find(page); - ASSERT(entries_it != invalidation_cache.end()); - std::vector<Entry*>& entries = entries_it->second; - - const auto entry_it = std::find(entries.begin(), entries.end(), entry); - ASSERT(entry_it != entries.end()); - entries.erase(entry_it); - } - } + void RemoveEntryFromInvalidationCache(const Entry* entry); /// @brief Unmarks an entry from the rasterizer cache /// @param entry Entry to unmark from memory - void UnmarkMemory(Entry* entry) { - if (!entry->is_memory_marked) { - return; - } - entry->is_memory_marked = false; - - const VAddr addr = entry->addr_start; - const std::size_t size = entry->addr_end - addr; - rasterizer.UpdatePagesCachedCount(addr, size, -1); - } + void UnmarkMemory(Entry* entry); /// @brief Removes a vector of shaders from a list /// @param removed_shaders Shaders to be removed from the storage /// @pre invalidation_mutex is locked /// @pre lookup_mutex is locked - void RemoveShadersFromStorage(std::vector<T*> removed_shaders) { - // Notify removals - for (T* const shader : removed_shaders) { - OnShaderRemoval(shader); - } - - // Remove them from the cache - const auto is_removed = [&removed_shaders](const std::unique_ptr<T>& shader) { - return std::find(removed_shaders.begin(), removed_shaders.end(), shader.get()) != - removed_shaders.end(); - }; - std::erase_if(storage, is_removed); - } + void RemoveShadersFromStorage(std::vector<ShaderInfo*> removed_shaders); /// @brief Creates a new entry in the lookup cache and returns its pointer /// @pre lookup_mutex is locked - Entry* NewEntry(VAddr addr, VAddr addr_end, T* data) { - auto entry = std::make_unique<Entry>(Entry{addr, addr_end, data}); - Entry* const entry_pointer = entry.get(); + Entry* NewEntry(VAddr addr, VAddr addr_end, ShaderInfo* data); - lookup_cache.emplace(addr, std::move(entry)); - return entry_pointer; - } + /// @brief Create a new shader entry and register it + const ShaderInfo* MakeShaderInfo(GenericEnvironment& env, VAddr cpu_addr); VideoCore::RasterizerInterface& rasterizer; @@ -233,7 +154,7 @@ private: std::unordered_map<u64, std::unique_ptr<Entry>> lookup_cache; std::unordered_map<u64, std::vector<Entry*>> invalidation_cache; - std::vector<std::unique_ptr<T>> storage; + std::vector<std::unique_ptr<ShaderInfo>> storage; std::vector<Entry*> marked_for_removal; }; diff --git a/src/video_core/shader_environment.cpp b/src/video_core/shader_environment.cpp new file mode 100644 index 000000000..8a4581c19 --- /dev/null +++ b/src/video_core/shader_environment.cpp @@ -0,0 +1,460 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <filesystem> +#include <fstream> +#include <memory> +#include <optional> +#include <utility> + +#include "common/assert.h" +#include "common/cityhash.h" +#include "common/common_types.h" +#include "common/div_ceil.h" +#include "common/fs/fs.h" +#include "common/logging/log.h" +#include "shader_recompiler/environment.h" +#include "video_core/memory_manager.h" +#include "video_core/shader_environment.h" +#include "video_core/textures/texture.h" + +namespace VideoCommon { + +constexpr std::array<char, 8> MAGIC_NUMBER{'y', 'u', 'z', 'u', 'c', 'a', 'c', 'h'}; + +constexpr size_t INST_SIZE = sizeof(u64); + +using Maxwell = Tegra::Engines::Maxwell3D::Regs; + +static u64 MakeCbufKey(u32 index, u32 offset) { + return (static_cast<u64>(index) << 32) | offset; +} + +static Shader::TextureType ConvertType(const Tegra::Texture::TICEntry& entry) { + switch (entry.texture_type) { + case Tegra::Texture::TextureType::Texture1D: + return Shader::TextureType::Color1D; + case Tegra::Texture::TextureType::Texture2D: + case Tegra::Texture::TextureType::Texture2DNoMipmap: + return Shader::TextureType::Color2D; + case Tegra::Texture::TextureType::Texture3D: + return Shader::TextureType::Color3D; + case Tegra::Texture::TextureType::TextureCubemap: + return Shader::TextureType::ColorCube; + case Tegra::Texture::TextureType::Texture1DArray: + return Shader::TextureType::ColorArray1D; + case Tegra::Texture::TextureType::Texture2DArray: + return Shader::TextureType::ColorArray2D; + case Tegra::Texture::TextureType::Texture1DBuffer: + return Shader::TextureType::Buffer; + case Tegra::Texture::TextureType::TextureCubeArray: + return Shader::TextureType::ColorArrayCube; + default: + throw Shader::NotImplementedException("Unknown texture type"); + } +} + +GenericEnvironment::GenericEnvironment(Tegra::MemoryManager& gpu_memory_, GPUVAddr program_base_, + u32 start_address_) + : gpu_memory{&gpu_memory_}, program_base{program_base_} { + start_address = start_address_; +} + +GenericEnvironment::~GenericEnvironment() = default; + +u32 GenericEnvironment::TextureBoundBuffer() const { + return texture_bound; +} + +u32 GenericEnvironment::LocalMemorySize() const { + return local_memory_size; +} + +u32 GenericEnvironment::SharedMemorySize() const { + return shared_memory_size; +} + +std::array<u32, 3> GenericEnvironment::WorkgroupSize() const { + return workgroup_size; +} + +u64 GenericEnvironment::ReadInstruction(u32 address) { + read_lowest = std::min(read_lowest, address); + read_highest = std::max(read_highest, address); + + if (address >= cached_lowest && address < cached_highest) { + return code[(address - cached_lowest) / INST_SIZE]; + } + has_unbound_instructions = true; + return gpu_memory->Read<u64>(program_base + address); +} + +std::optional<u64> GenericEnvironment::Analyze() { + const std::optional<u64> size{TryFindSize()}; + if (!size) { + return std::nullopt; + } + cached_lowest = start_address; + cached_highest = start_address + static_cast<u32>(*size); + return Common::CityHash64(reinterpret_cast<const char*>(code.data()), *size); +} + +void GenericEnvironment::SetCachedSize(size_t size_bytes) { + cached_lowest = start_address; + cached_highest = start_address + static_cast<u32>(size_bytes); + code.resize(CachedSize()); + gpu_memory->ReadBlock(program_base + cached_lowest, code.data(), code.size() * sizeof(u64)); +} + +size_t GenericEnvironment::CachedSize() const noexcept { + return cached_highest - cached_lowest + INST_SIZE; +} + +size_t GenericEnvironment::ReadSize() const noexcept { + return read_highest - read_lowest + INST_SIZE; +} + +bool GenericEnvironment::CanBeSerialized() const noexcept { + return !has_unbound_instructions; +} + +u64 GenericEnvironment::CalculateHash() const { + const size_t size{ReadSize()}; + const auto data{std::make_unique<char[]>(size)}; + gpu_memory->ReadBlock(program_base + read_lowest, data.get(), size); + return Common::CityHash64(data.get(), size); +} + +void GenericEnvironment::Serialize(std::ofstream& file) const { + const u64 code_size{static_cast<u64>(CachedSize())}; + const u64 num_texture_types{static_cast<u64>(texture_types.size())}; + const u64 num_cbuf_values{static_cast<u64>(cbuf_values.size())}; + + file.write(reinterpret_cast<const char*>(&code_size), sizeof(code_size)) + .write(reinterpret_cast<const char*>(&num_texture_types), sizeof(num_texture_types)) + .write(reinterpret_cast<const char*>(&num_cbuf_values), sizeof(num_cbuf_values)) + .write(reinterpret_cast<const char*>(&local_memory_size), sizeof(local_memory_size)) + .write(reinterpret_cast<const char*>(&texture_bound), sizeof(texture_bound)) + .write(reinterpret_cast<const char*>(&start_address), sizeof(start_address)) + .write(reinterpret_cast<const char*>(&cached_lowest), sizeof(cached_lowest)) + .write(reinterpret_cast<const char*>(&cached_highest), sizeof(cached_highest)) + .write(reinterpret_cast<const char*>(&stage), sizeof(stage)) + .write(reinterpret_cast<const char*>(code.data()), code_size); + for (const auto [key, type] : texture_types) { + file.write(reinterpret_cast<const char*>(&key), sizeof(key)) + .write(reinterpret_cast<const char*>(&type), sizeof(type)); + } + for (const auto [key, type] : cbuf_values) { + file.write(reinterpret_cast<const char*>(&key), sizeof(key)) + .write(reinterpret_cast<const char*>(&type), sizeof(type)); + } + if (stage == Shader::Stage::Compute) { + file.write(reinterpret_cast<const char*>(&workgroup_size), sizeof(workgroup_size)) + .write(reinterpret_cast<const char*>(&shared_memory_size), sizeof(shared_memory_size)); + } else { + file.write(reinterpret_cast<const char*>(&sph), sizeof(sph)); + if (stage == Shader::Stage::Geometry) { + file.write(reinterpret_cast<const char*>(&gp_passthrough_mask), + sizeof(gp_passthrough_mask)); + } + } +} + +std::optional<u64> GenericEnvironment::TryFindSize() { + static constexpr size_t BLOCK_SIZE = 0x1000; + static constexpr size_t MAXIMUM_SIZE = 0x100000; + + static constexpr u64 SELF_BRANCH_A = 0xE2400FFFFF87000FULL; + static constexpr u64 SELF_BRANCH_B = 0xE2400FFFFF07000FULL; + + GPUVAddr guest_addr{program_base + start_address}; + size_t offset{0}; + size_t size{BLOCK_SIZE}; + while (size <= MAXIMUM_SIZE) { + code.resize(size / INST_SIZE); + u64* const data = code.data() + offset / INST_SIZE; + gpu_memory->ReadBlock(guest_addr, data, BLOCK_SIZE); + for (size_t index = 0; index < BLOCK_SIZE; index += INST_SIZE) { + const u64 inst = data[index / INST_SIZE]; + if (inst == SELF_BRANCH_A || inst == SELF_BRANCH_B) { + return offset + index; + } + } + guest_addr += BLOCK_SIZE; + size += BLOCK_SIZE; + offset += BLOCK_SIZE; + } + return std::nullopt; +} + +Shader::TextureType GenericEnvironment::ReadTextureTypeImpl(GPUVAddr tic_addr, u32 tic_limit, + bool via_header_index, u32 raw) { + const auto handle{Tegra::Texture::TexturePair(raw, via_header_index)}; + const GPUVAddr descriptor_addr{tic_addr + handle.first * sizeof(Tegra::Texture::TICEntry)}; + Tegra::Texture::TICEntry entry; + gpu_memory->ReadBlock(descriptor_addr, &entry, sizeof(entry)); + const Shader::TextureType result{ConvertType(entry)}; + texture_types.emplace(raw, result); + return result; +} + +GraphicsEnvironment::GraphicsEnvironment(Tegra::Engines::Maxwell3D& maxwell3d_, + Tegra::MemoryManager& gpu_memory_, + Maxwell::ShaderProgram program, GPUVAddr program_base_, + u32 start_address_) + : GenericEnvironment{gpu_memory_, program_base_, start_address_}, maxwell3d{&maxwell3d_} { + gpu_memory->ReadBlock(program_base + start_address, &sph, sizeof(sph)); + gp_passthrough_mask = maxwell3d->regs.gp_passthrough_mask; + switch (program) { + case Maxwell::ShaderProgram::VertexA: + stage = Shader::Stage::VertexA; + stage_index = 0; + break; + case Maxwell::ShaderProgram::VertexB: + stage = Shader::Stage::VertexB; + stage_index = 0; + break; + case Maxwell::ShaderProgram::TesselationControl: + stage = Shader::Stage::TessellationControl; + stage_index = 1; + break; + case Maxwell::ShaderProgram::TesselationEval: + stage = Shader::Stage::TessellationEval; + stage_index = 2; + break; + case Maxwell::ShaderProgram::Geometry: + stage = Shader::Stage::Geometry; + stage_index = 3; + break; + case Maxwell::ShaderProgram::Fragment: + stage = Shader::Stage::Fragment; + stage_index = 4; + break; + default: + UNREACHABLE_MSG("Invalid program={}", program); + break; + } + const u64 local_size{sph.LocalMemorySize()}; + ASSERT(local_size <= std::numeric_limits<u32>::max()); + local_memory_size = static_cast<u32>(local_size) + sph.common3.shader_local_memory_crs_size; + texture_bound = maxwell3d->regs.tex_cb_index; +} + +u32 GraphicsEnvironment::ReadCbufValue(u32 cbuf_index, u32 cbuf_offset) { + const auto& cbuf{maxwell3d->state.shader_stages[stage_index].const_buffers[cbuf_index]}; + ASSERT(cbuf.enabled); + u32 value{}; + if (cbuf_offset < cbuf.size) { + value = gpu_memory->Read<u32>(cbuf.address + cbuf_offset); + } + cbuf_values.emplace(MakeCbufKey(cbuf_index, cbuf_offset), value); + return value; +} + +Shader::TextureType GraphicsEnvironment::ReadTextureType(u32 handle) { + const auto& regs{maxwell3d->regs}; + const bool via_header_index{regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex}; + return ReadTextureTypeImpl(regs.tic.Address(), regs.tic.limit, via_header_index, handle); +} + +ComputeEnvironment::ComputeEnvironment(Tegra::Engines::KeplerCompute& kepler_compute_, + Tegra::MemoryManager& gpu_memory_, GPUVAddr program_base_, + u32 start_address_) + : GenericEnvironment{gpu_memory_, program_base_, start_address_}, kepler_compute{ + &kepler_compute_} { + const auto& qmd{kepler_compute->launch_description}; + stage = Shader::Stage::Compute; + local_memory_size = qmd.local_pos_alloc + qmd.local_crs_alloc; + texture_bound = kepler_compute->regs.tex_cb_index; + shared_memory_size = qmd.shared_alloc; + workgroup_size = {qmd.block_dim_x, qmd.block_dim_y, qmd.block_dim_z}; +} + +u32 ComputeEnvironment::ReadCbufValue(u32 cbuf_index, u32 cbuf_offset) { + const auto& qmd{kepler_compute->launch_description}; + ASSERT(((qmd.const_buffer_enable_mask.Value() >> cbuf_index) & 1) != 0); + const auto& cbuf{qmd.const_buffer_config[cbuf_index]}; + u32 value{}; + if (cbuf_offset < cbuf.size) { + value = gpu_memory->Read<u32>(cbuf.Address() + cbuf_offset); + } + cbuf_values.emplace(MakeCbufKey(cbuf_index, cbuf_offset), value); + return value; +} + +Shader::TextureType ComputeEnvironment::ReadTextureType(u32 handle) { + const auto& regs{kepler_compute->regs}; + const auto& qmd{kepler_compute->launch_description}; + return ReadTextureTypeImpl(regs.tic.Address(), regs.tic.limit, qmd.linked_tsc != 0, handle); +} + +void FileEnvironment::Deserialize(std::ifstream& file) { + u64 code_size{}; + u64 num_texture_types{}; + u64 num_cbuf_values{}; + file.read(reinterpret_cast<char*>(&code_size), sizeof(code_size)) + .read(reinterpret_cast<char*>(&num_texture_types), sizeof(num_texture_types)) + .read(reinterpret_cast<char*>(&num_cbuf_values), sizeof(num_cbuf_values)) + .read(reinterpret_cast<char*>(&local_memory_size), sizeof(local_memory_size)) + .read(reinterpret_cast<char*>(&texture_bound), sizeof(texture_bound)) + .read(reinterpret_cast<char*>(&start_address), sizeof(start_address)) + .read(reinterpret_cast<char*>(&read_lowest), sizeof(read_lowest)) + .read(reinterpret_cast<char*>(&read_highest), sizeof(read_highest)) + .read(reinterpret_cast<char*>(&stage), sizeof(stage)); + code = std::make_unique<u64[]>(Common::DivCeil(code_size, sizeof(u64))); + file.read(reinterpret_cast<char*>(code.get()), code_size); + for (size_t i = 0; i < num_texture_types; ++i) { + u32 key; + Shader::TextureType type; + file.read(reinterpret_cast<char*>(&key), sizeof(key)) + .read(reinterpret_cast<char*>(&type), sizeof(type)); + texture_types.emplace(key, type); + } + for (size_t i = 0; i < num_cbuf_values; ++i) { + u64 key; + u32 value; + file.read(reinterpret_cast<char*>(&key), sizeof(key)) + .read(reinterpret_cast<char*>(&value), sizeof(value)); + cbuf_values.emplace(key, value); + } + if (stage == Shader::Stage::Compute) { + file.read(reinterpret_cast<char*>(&workgroup_size), sizeof(workgroup_size)) + .read(reinterpret_cast<char*>(&shared_memory_size), sizeof(shared_memory_size)); + } else { + file.read(reinterpret_cast<char*>(&sph), sizeof(sph)); + if (stage == Shader::Stage::Geometry) { + file.read(reinterpret_cast<char*>(&gp_passthrough_mask), sizeof(gp_passthrough_mask)); + } + } +} + +u64 FileEnvironment::ReadInstruction(u32 address) { + if (address < read_lowest || address > read_highest) { + throw Shader::LogicError("Out of bounds address {}", address); + } + return code[(address - read_lowest) / sizeof(u64)]; +} + +u32 FileEnvironment::ReadCbufValue(u32 cbuf_index, u32 cbuf_offset) { + const auto it{cbuf_values.find(MakeCbufKey(cbuf_index, cbuf_offset))}; + if (it == cbuf_values.end()) { + throw Shader::LogicError("Uncached read texture type"); + } + return it->second; +} + +Shader::TextureType FileEnvironment::ReadTextureType(u32 handle) { + const auto it{texture_types.find(handle)}; + if (it == texture_types.end()) { + throw Shader::LogicError("Uncached read texture type"); + } + return it->second; +} + +u32 FileEnvironment::LocalMemorySize() const { + return local_memory_size; +} + +u32 FileEnvironment::SharedMemorySize() const { + return shared_memory_size; +} + +u32 FileEnvironment::TextureBoundBuffer() const { + return texture_bound; +} + +std::array<u32, 3> FileEnvironment::WorkgroupSize() const { + return workgroup_size; +} + +void SerializePipeline(std::span<const char> key, std::span<const GenericEnvironment* const> envs, + const std::filesystem::path& filename, u32 cache_version) try { + std::ofstream file(filename, std::ios::binary | std::ios::ate | std::ios::app); + file.exceptions(std::ifstream::failbit); + if (!file.is_open()) { + LOG_ERROR(Common_Filesystem, "Failed to open pipeline cache file {}", + Common::FS::PathToUTF8String(filename)); + return; + } + if (file.tellp() == 0) { + // Write header + file.write(MAGIC_NUMBER.data(), MAGIC_NUMBER.size()) + .write(reinterpret_cast<const char*>(&cache_version), sizeof(cache_version)); + } + if (!std::ranges::all_of(envs, &GenericEnvironment::CanBeSerialized)) { + return; + } + const u32 num_envs{static_cast<u32>(envs.size())}; + file.write(reinterpret_cast<const char*>(&num_envs), sizeof(num_envs)); + for (const GenericEnvironment* const env : envs) { + env->Serialize(file); + } + file.write(key.data(), key.size_bytes()); + +} catch (const std::ios_base::failure& e) { + LOG_ERROR(Common_Filesystem, "{}", e.what()); + if (!Common::FS::RemoveFile(filename)) { + LOG_ERROR(Common_Filesystem, "Failed to delete pipeline cache file {}", + Common::FS::PathToUTF8String(filename)); + } +} + +void LoadPipelines( + std::stop_token stop_loading, const std::filesystem::path& filename, u32 expected_cache_version, + Common::UniqueFunction<void, std::ifstream&, FileEnvironment> load_compute, + Common::UniqueFunction<void, std::ifstream&, std::vector<FileEnvironment>> load_graphics) try { + std::ifstream file(filename, std::ios::binary | std::ios::ate); + if (!file.is_open()) { + return; + } + file.exceptions(std::ifstream::failbit); + const auto end{file.tellg()}; + file.seekg(0, std::ios::beg); + + std::array<char, 8> magic_number; + u32 cache_version; + file.read(magic_number.data(), magic_number.size()) + .read(reinterpret_cast<char*>(&cache_version), sizeof(cache_version)); + if (magic_number != MAGIC_NUMBER || cache_version != expected_cache_version) { + file.close(); + if (Common::FS::RemoveFile(filename)) { + if (magic_number != MAGIC_NUMBER) { + LOG_ERROR(Common_Filesystem, "Invalid pipeline cache file"); + } + if (cache_version != expected_cache_version) { + LOG_INFO(Common_Filesystem, "Deleting old pipeline cache"); + } + } else { + LOG_ERROR(Common_Filesystem, + "Invalid pipeline cache file and failed to delete it in \"{}\"", + Common::FS::PathToUTF8String(filename)); + } + return; + } + while (file.tellg() != end) { + if (stop_loading.stop_requested()) { + return; + } + u32 num_envs{}; + file.read(reinterpret_cast<char*>(&num_envs), sizeof(num_envs)); + std::vector<FileEnvironment> envs(num_envs); + for (FileEnvironment& env : envs) { + env.Deserialize(file); + } + if (envs.front().ShaderStage() == Shader::Stage::Compute) { + load_compute(file, std::move(envs.front())); + } else { + load_graphics(file, std::move(envs)); + } + } + +} catch (const std::ios_base::failure& e) { + LOG_ERROR(Common_Filesystem, "{}", e.what()); + if (!Common::FS::RemoveFile(filename)) { + LOG_ERROR(Common_Filesystem, "Failed to delete pipeline cache file {}", + Common::FS::PathToUTF8String(filename)); + } +} + +} // namespace VideoCommon diff --git a/src/video_core/shader_environment.h b/src/video_core/shader_environment.h new file mode 100644 index 000000000..2079979db --- /dev/null +++ b/src/video_core/shader_environment.h @@ -0,0 +1,183 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <array> +#include <atomic> +#include <filesystem> +#include <iosfwd> +#include <limits> +#include <memory> +#include <optional> +#include <span> +#include <type_traits> +#include <unordered_map> +#include <vector> + +#include "common/common_types.h" +#include "common/unique_function.h" +#include "shader_recompiler/environment.h" +#include "video_core/engines/kepler_compute.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/textures/texture.h" + +namespace Tegra { +class Memorymanager; +} + +namespace VideoCommon { + +class GenericEnvironment : public Shader::Environment { +public: + explicit GenericEnvironment() = default; + explicit GenericEnvironment(Tegra::MemoryManager& gpu_memory_, GPUVAddr program_base_, + u32 start_address_); + + ~GenericEnvironment() override; + + [[nodiscard]] u32 TextureBoundBuffer() const final; + + [[nodiscard]] u32 LocalMemorySize() const final; + + [[nodiscard]] u32 SharedMemorySize() const final; + + [[nodiscard]] std::array<u32, 3> WorkgroupSize() const final; + + [[nodiscard]] u64 ReadInstruction(u32 address) final; + + [[nodiscard]] std::optional<u64> Analyze(); + + void SetCachedSize(size_t size_bytes); + + [[nodiscard]] size_t CachedSize() const noexcept; + + [[nodiscard]] size_t ReadSize() const noexcept; + + [[nodiscard]] bool CanBeSerialized() const noexcept; + + [[nodiscard]] u64 CalculateHash() const; + + void Serialize(std::ofstream& file) const; + +protected: + std::optional<u64> TryFindSize(); + + Shader::TextureType ReadTextureTypeImpl(GPUVAddr tic_addr, u32 tic_limit, bool via_header_index, + u32 raw); + + Tegra::MemoryManager* gpu_memory{}; + GPUVAddr program_base{}; + + std::vector<u64> code; + std::unordered_map<u32, Shader::TextureType> texture_types; + std::unordered_map<u64, u32> cbuf_values; + + u32 local_memory_size{}; + u32 texture_bound{}; + u32 shared_memory_size{}; + std::array<u32, 3> workgroup_size{}; + + u32 read_lowest = std::numeric_limits<u32>::max(); + u32 read_highest = 0; + + u32 cached_lowest = std::numeric_limits<u32>::max(); + u32 cached_highest = 0; + + bool has_unbound_instructions = false; +}; + +class GraphicsEnvironment final : public GenericEnvironment { +public: + explicit GraphicsEnvironment() = default; + explicit GraphicsEnvironment(Tegra::Engines::Maxwell3D& maxwell3d_, + Tegra::MemoryManager& gpu_memory_, + Tegra::Engines::Maxwell3D::Regs::ShaderProgram program, + GPUVAddr program_base_, u32 start_address_); + + ~GraphicsEnvironment() override = default; + + u32 ReadCbufValue(u32 cbuf_index, u32 cbuf_offset) override; + + Shader::TextureType ReadTextureType(u32 handle) override; + +private: + Tegra::Engines::Maxwell3D* maxwell3d{}; + size_t stage_index{}; +}; + +class ComputeEnvironment final : public GenericEnvironment { +public: + explicit ComputeEnvironment() = default; + explicit ComputeEnvironment(Tegra::Engines::KeplerCompute& kepler_compute_, + Tegra::MemoryManager& gpu_memory_, GPUVAddr program_base_, + u32 start_address_); + + ~ComputeEnvironment() override = default; + + u32 ReadCbufValue(u32 cbuf_index, u32 cbuf_offset) override; + + Shader::TextureType ReadTextureType(u32 handle) override; + +private: + Tegra::Engines::KeplerCompute* kepler_compute{}; +}; + +class FileEnvironment final : public Shader::Environment { +public: + FileEnvironment() = default; + ~FileEnvironment() override = default; + + FileEnvironment& operator=(FileEnvironment&&) noexcept = default; + FileEnvironment(FileEnvironment&&) noexcept = default; + + FileEnvironment& operator=(const FileEnvironment&) = delete; + FileEnvironment(const FileEnvironment&) = delete; + + void Deserialize(std::ifstream& file); + + [[nodiscard]] u64 ReadInstruction(u32 address) override; + + [[nodiscard]] u32 ReadCbufValue(u32 cbuf_index, u32 cbuf_offset) override; + + [[nodiscard]] Shader::TextureType ReadTextureType(u32 handle) override; + + [[nodiscard]] u32 LocalMemorySize() const override; + + [[nodiscard]] u32 SharedMemorySize() const override; + + [[nodiscard]] u32 TextureBoundBuffer() const override; + + [[nodiscard]] std::array<u32, 3> WorkgroupSize() const override; + +private: + std::unique_ptr<u64[]> code; + std::unordered_map<u32, Shader::TextureType> texture_types; + std::unordered_map<u64, u32> cbuf_values; + std::array<u32, 3> workgroup_size{}; + u32 local_memory_size{}; + u32 shared_memory_size{}; + u32 texture_bound{}; + u32 read_lowest{}; + u32 read_highest{}; +}; + +void SerializePipeline(std::span<const char> key, std::span<const GenericEnvironment* const> envs, + const std::filesystem::path& filename, u32 cache_version); + +template <typename Key, typename Envs> +void SerializePipeline(const Key& key, const Envs& envs, const std::filesystem::path& filename, + u32 cache_version) { + static_assert(std::is_trivially_copyable_v<Key>); + static_assert(std::has_unique_object_representations_v<Key>); + SerializePipeline(std::span(reinterpret_cast<const char*>(&key), sizeof(key)), + std::span(envs.data(), envs.size()), filename, cache_version); +} + +void LoadPipelines( + std::stop_token stop_loading, const std::filesystem::path& filename, u32 expected_cache_version, + Common::UniqueFunction<void, std::ifstream&, FileEnvironment> load_compute, + Common::UniqueFunction<void, std::ifstream&, std::vector<FileEnvironment>> load_graphics); + +} // namespace VideoCommon diff --git a/src/video_core/shader_notify.cpp b/src/video_core/shader_notify.cpp index 693e47158..dc6995b46 100644 --- a/src/video_core/shader_notify.cpp +++ b/src/video_core/shader_notify.cpp @@ -2,42 +2,35 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. -#include <mutex> +#include <atomic> +#include <chrono> +#include <optional> + #include "video_core/shader_notify.h" using namespace std::chrono_literals; namespace VideoCore { -namespace { -constexpr auto UPDATE_TICK = 32ms; -} - -ShaderNotify::ShaderNotify() = default; -ShaderNotify::~ShaderNotify() = default; -std::size_t ShaderNotify::GetShadersBuilding() { - const auto now = std::chrono::high_resolution_clock::now(); - const auto diff = now - last_update; - if (diff > UPDATE_TICK) { - std::shared_lock lock(mutex); - last_updated_count = accurate_count; +const auto TIME_TO_STOP_REPORTING = 2s; + +int ShaderNotify::ShadersBuilding() noexcept { + const int now_complete = num_complete.load(std::memory_order::relaxed); + const int now_building = num_building.load(std::memory_order::relaxed); + if (now_complete == now_building) { + const auto now = std::chrono::high_resolution_clock::now(); + if (completed && num_complete == num_when_completed) { + if (now - complete_time > TIME_TO_STOP_REPORTING) { + report_base = now_complete; + completed = false; + } + } else { + completed = true; + num_when_completed = num_complete; + complete_time = now; + } } - return last_updated_count; -} - -std::size_t ShaderNotify::GetShadersBuildingAccurate() { - std::shared_lock lock{mutex}; - return accurate_count; -} - -void ShaderNotify::MarkShaderComplete() { - std::unique_lock lock{mutex}; - accurate_count--; -} - -void ShaderNotify::MarkSharderBuilding() { - std::unique_lock lock{mutex}; - accurate_count++; + return now_building - report_base; } } // namespace VideoCore diff --git a/src/video_core/shader_notify.h b/src/video_core/shader_notify.h index a9c92d179..ad363bfb5 100644 --- a/src/video_core/shader_notify.h +++ b/src/video_core/shader_notify.h @@ -4,26 +4,30 @@ #pragma once +#include <atomic> #include <chrono> -#include <shared_mutex> -#include "common/common_types.h" +#include <optional> namespace VideoCore { class ShaderNotify { public: - ShaderNotify(); - ~ShaderNotify(); + [[nodiscard]] int ShadersBuilding() noexcept; - std::size_t GetShadersBuilding(); - std::size_t GetShadersBuildingAccurate(); + void MarkShaderComplete() noexcept { + ++num_complete; + } - void MarkShaderComplete(); - void MarkSharderBuilding(); + void MarkShaderBuilding() noexcept { + ++num_building; + } private: - std::size_t last_updated_count{}; - std::size_t accurate_count{}; - std::shared_mutex mutex; - std::chrono::high_resolution_clock::time_point last_update{}; + std::atomic_int num_building{}; + std::atomic_int num_complete{}; + int report_base{}; + + bool completed{}; + int num_when_completed{}; + std::chrono::high_resolution_clock::time_point complete_time; }; } // namespace VideoCore diff --git a/src/video_core/texture_cache/formatter.cpp b/src/video_core/texture_cache/formatter.cpp index d10ba4ccd..249cc4d0f 100644 --- a/src/video_core/texture_cache/formatter.cpp +++ b/src/video_core/texture_cache/formatter.cpp @@ -43,7 +43,7 @@ std::string Name(const ImageBase& image) { return "Invalid"; } -std::string Name(const ImageViewBase& image_view, std::optional<ImageViewType> type) { +std::string Name(const ImageViewBase& image_view) { const u32 width = image_view.size.width; const u32 height = image_view.size.height; const u32 depth = image_view.size.depth; @@ -51,7 +51,7 @@ std::string Name(const ImageViewBase& image_view, std::optional<ImageViewType> t const u32 num_layers = image_view.range.extent.layers; const std::string level = num_levels > 1 ? fmt::format(":{}", num_levels) : ""; - switch (type.value_or(image_view.type)) { + switch (image_view.type) { case ImageViewType::e1D: return fmt::format("ImageView 1D {}{}", width, level); case ImageViewType::e2D: diff --git a/src/video_core/texture_cache/formatter.h b/src/video_core/texture_cache/formatter.h index a48413983..c6cf0583f 100644 --- a/src/video_core/texture_cache/formatter.h +++ b/src/video_core/texture_cache/formatter.h @@ -255,8 +255,7 @@ struct RenderTargets; [[nodiscard]] std::string Name(const ImageBase& image); -[[nodiscard]] std::string Name(const ImageViewBase& image_view, - std::optional<ImageViewType> type = std::nullopt); +[[nodiscard]] std::string Name(const ImageViewBase& image_view); [[nodiscard]] std::string Name(const RenderTargets& render_targets); diff --git a/src/video_core/texture_cache/image_view_base.cpp b/src/video_core/texture_cache/image_view_base.cpp index e8d632f9e..450becbeb 100644 --- a/src/video_core/texture_cache/image_view_base.cpp +++ b/src/video_core/texture_cache/image_view_base.cpp @@ -36,6 +36,15 @@ ImageViewBase::ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_i } } +ImageViewBase::ImageViewBase(const ImageInfo& info, const ImageViewInfo& view_info) + : format{info.format}, type{ImageViewType::Buffer}, size{ + .width = info.size.width, + .height = 1, + .depth = 1, + } { + ASSERT_MSG(view_info.type == ImageViewType::Buffer, "Expected texture buffer"); +} + ImageViewBase::ImageViewBase(const NullImageParams&) {} } // namespace VideoCommon diff --git a/src/video_core/texture_cache/image_view_base.h b/src/video_core/texture_cache/image_view_base.h index 73954167e..903f715c5 100644 --- a/src/video_core/texture_cache/image_view_base.h +++ b/src/video_core/texture_cache/image_view_base.h @@ -27,6 +27,7 @@ DECLARE_ENUM_FLAG_OPERATORS(ImageViewFlagBits) struct ImageViewBase { explicit ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_info, ImageId image_id); + explicit ImageViewBase(const ImageInfo& info, const ImageViewInfo& view_info); explicit ImageViewBase(const NullImageParams&); [[nodiscard]] bool IsBuffer() const noexcept { diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 01de2d498..f34c9d9ca 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -117,6 +117,9 @@ public: /// Return a reference to the given image view id [[nodiscard]] ImageView& GetImageView(ImageViewId id) noexcept; + /// Mark an image as modified from the GPU + void MarkModification(ImageId id) noexcept; + /// Fill image_view_ids with the graphics images in indices void FillGraphicsImageViews(std::span<const u32> indices, std::span<ImageViewId> image_view_ids); @@ -527,6 +530,11 @@ typename P::ImageView& TextureCache<P>::GetImageView(ImageViewId id) noexcept { } template <class P> +void TextureCache<P>::MarkModification(ImageId id) noexcept { + MarkModification(slot_images[id]); +} + +template <class P> void TextureCache<P>::FillGraphicsImageViews(std::span<const u32> indices, std::span<ImageViewId> image_view_ids) { FillImageViews(graphics_image_table, graphics_image_view_ids, indices, image_view_ids); @@ -540,13 +548,13 @@ void TextureCache<P>::FillComputeImageViews(std::span<const u32> indices, template <class P> typename P::Sampler* TextureCache<P>::GetGraphicsSampler(u32 index) { - [[unlikely]] if (index > graphics_sampler_table.Limit()) { - LOG_ERROR(HW_GPU, "Invalid sampler index={}", index); + if (index > graphics_sampler_table.Limit()) { + LOG_DEBUG(HW_GPU, "Invalid sampler index={}", index); return &slot_samplers[NULL_SAMPLER_ID]; } const auto [descriptor, is_new] = graphics_sampler_table.Read(index); SamplerId& id = graphics_sampler_ids[index]; - [[unlikely]] if (is_new) { + if (is_new) { id = FindSampler(descriptor); } return &slot_samplers[id]; @@ -554,13 +562,13 @@ typename P::Sampler* TextureCache<P>::GetGraphicsSampler(u32 index) { template <class P> typename P::Sampler* TextureCache<P>::GetComputeSampler(u32 index) { - [[unlikely]] if (index > compute_sampler_table.Limit()) { - LOG_ERROR(HW_GPU, "Invalid sampler index={}", index); + if (index > compute_sampler_table.Limit()) { + LOG_DEBUG(HW_GPU, "Invalid sampler index={}", index); return &slot_samplers[NULL_SAMPLER_ID]; } const auto [descriptor, is_new] = compute_sampler_table.Read(index); SamplerId& id = compute_sampler_ids[index]; - [[unlikely]] if (is_new) { + if (is_new) { id = FindSampler(descriptor); } return &slot_samplers[id]; @@ -599,6 +607,12 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) { using namespace VideoCommon::Dirty; auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::RenderTargets]) { + for (size_t index = 0; index < NUM_RT; ++index) { + ImageViewId& color_buffer_id = render_targets.color_buffer_ids[index]; + PrepareImageView(color_buffer_id, true, is_clear && IsFullClear(color_buffer_id)); + } + const ImageViewId depth_buffer_id = render_targets.depth_buffer_id; + PrepareImageView(depth_buffer_id, true, is_clear && IsFullClear(depth_buffer_id)); return; } flags[Dirty::RenderTargets] = false; @@ -655,7 +669,7 @@ ImageViewId TextureCache<P>::VisitImageView(DescriptorTable<TICEntry>& table, std::span<ImageViewId> cached_image_view_ids, u32 index) { if (index > table.Limit()) { - LOG_ERROR(HW_GPU, "Invalid image view index={}", index); + LOG_DEBUG(HW_GPU, "Invalid image view index={}", index); return NULL_IMAGE_VIEW_ID; } const auto [descriptor, is_new] = table.Read(index); @@ -962,9 +976,6 @@ void TextureCache<P>::UploadImageContents(Image& image, StagingBuffer& staging) auto copies = UnswizzleImage(gpu_memory, gpu_addr, image.info, unswizzled_data); ConvertImage(unswizzled_data, image.info, mapped_span, copies); image.UploadMemory(staging, copies); - } else if (image.info.type == ImageType::Buffer) { - const std::array copies{UploadBufferCopy(gpu_memory, gpu_addr, image, mapped_span)}; - image.UploadMemory(staging, copies); } else { const auto copies = UnswizzleImage(gpu_memory, gpu_addr, image.info, mapped_span); image.UploadMemory(staging, copies); @@ -987,7 +998,12 @@ ImageViewId TextureCache<P>::FindImageView(const TICEntry& config) { template <class P> ImageViewId TextureCache<P>::CreateImageView(const TICEntry& config) { const ImageInfo info(config); - const GPUVAddr image_gpu_addr = config.Address() - config.BaseLayer() * info.layer_stride; + if (info.type == ImageType::Buffer) { + const ImageViewInfo view_info(config, 0); + return slot_image_views.insert(runtime, info, view_info, config.Address()); + } + const u32 layer_offset = config.BaseLayer() * info.layer_stride; + const GPUVAddr image_gpu_addr = config.Address() - layer_offset; const ImageId image_id = FindOrInsertImage(info, image_gpu_addr); if (!image_id) { return NULL_IMAGE_VIEW_ID; @@ -1795,6 +1811,9 @@ void TextureCache<P>::PrepareImageView(ImageViewId image_view_id, bool is_modifi return; } const ImageViewBase& image_view = slot_image_views[image_view_id]; + if (image_view.IsBuffer()) { + return; + } PrepareImage(image_view.image_id, is_modification, invalidate); } diff --git a/src/video_core/texture_cache/types.h b/src/video_core/texture_cache/types.h index 9fbdc1ac6..47a11cb2f 100644 --- a/src/video_core/texture_cache/types.h +++ b/src/video_core/texture_cache/types.h @@ -133,8 +133,8 @@ struct BufferImageCopy { }; struct BufferCopy { - size_t src_offset; - size_t dst_offset; + u64 src_offset; + u64 dst_offset; size_t size; }; diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp index c872517b8..59cf2f561 100644 --- a/src/video_core/texture_cache/util.cpp +++ b/src/video_core/texture_cache/util.cpp @@ -169,23 +169,6 @@ template <u32 GOB_EXTENT> return Common::DivCeil(AdjustMipSize(size, level), block_size); } -[[nodiscard]] constexpr std::pair<int, int> Samples(int num_samples) { - switch (num_samples) { - case 1: - return {1, 1}; - case 2: - return {2, 1}; - case 4: - return {2, 2}; - case 8: - return {4, 2}; - case 16: - return {4, 4}; - } - UNREACHABLE_MSG("Invalid number of samples={}", num_samples); - return {1, 1}; -} - [[nodiscard]] constexpr Extent2D DefaultBlockSize(PixelFormat format) { return {DefaultBlockWidth(format), DefaultBlockHeight(format)}; } @@ -283,14 +266,13 @@ template <u32 GOB_EXTENT> } [[nodiscard]] constexpr LevelInfo MakeLevelInfo(PixelFormat format, Extent3D size, Extent3D block, - u32 num_samples, u32 tile_width_spacing) { - const auto [samples_x, samples_y] = Samples(num_samples); + u32 tile_width_spacing) { const u32 bytes_per_block = BytesPerBlock(format); return { .size = { - .width = size.width * samples_x, - .height = size.height * samples_y, + .width = size.width, + .height = size.height, .depth = size.depth, }, .block = block, @@ -301,14 +283,12 @@ template <u32 GOB_EXTENT> } [[nodiscard]] constexpr LevelInfo MakeLevelInfo(const ImageInfo& info) { - return MakeLevelInfo(info.format, info.size, info.block, info.num_samples, - info.tile_width_spacing); + return MakeLevelInfo(info.format, info.size, info.block, info.tile_width_spacing); } [[nodiscard]] constexpr u32 CalculateLevelOffset(PixelFormat format, Extent3D size, Extent3D block, - u32 num_samples, u32 tile_width_spacing, - u32 level) { - const LevelInfo info = MakeLevelInfo(format, size, block, num_samples, tile_width_spacing); + u32 tile_width_spacing, u32 level) { + const LevelInfo info = MakeLevelInfo(format, size, block, tile_width_spacing); u32 offset = 0; for (u32 current_level = 0; current_level < level; ++current_level) { offset += CalculateLevelSize(info, current_level); @@ -645,8 +625,8 @@ u32 CalculateLayerStride(const ImageInfo& info) noexcept { u32 CalculateLayerSize(const ImageInfo& info) noexcept { ASSERT(info.type != ImageType::Linear); - return CalculateLevelOffset(info.format, info.size, info.block, info.num_samples, - info.tile_width_spacing, info.resources.levels); + return CalculateLevelOffset(info.format, info.size, info.block, info.tile_width_spacing, + info.resources.levels); } LevelArray CalculateMipLevelOffsets(const ImageInfo& info) noexcept { @@ -1195,37 +1175,37 @@ static_assert(CalculateLevelSize(LevelInfo{{1920, 1080, 1}, {0, 2, 0}, {1, 1}, 2 0x7f8000); static_assert(CalculateLevelSize(LevelInfo{{32, 32, 1}, {0, 0, 4}, {1, 1}, 4, 0}, 0) == 0x4000); -static_assert(CalculateLevelOffset(PixelFormat::R8_SINT, {1920, 1080, 1}, {0, 2, 0}, 1, 0, 7) == +static_assert(CalculateLevelOffset(PixelFormat::R8_SINT, {1920, 1080, 1}, {0, 2, 0}, 0, 7) == 0x2afc00); -static_assert(CalculateLevelOffset(PixelFormat::ASTC_2D_12X12_UNORM, {8192, 4096, 1}, {0, 2, 0}, 1, - 0, 12) == 0x50d200); - -static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, - 0) == 0); -static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, - 1) == 0x400000); -static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, - 2) == 0x500000); -static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, - 3) == 0x540000); -static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, - 4) == 0x550000); -static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, - 5) == 0x554000); -static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, - 6) == 0x555000); -static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, - 7) == 0x555400); -static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, - 8) == 0x555600); -static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, - 9) == 0x555800); +static_assert(CalculateLevelOffset(PixelFormat::ASTC_2D_12X12_UNORM, {8192, 4096, 1}, {0, 2, 0}, 0, + 12) == 0x50d200); + +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 0) == + 0); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 1) == + 0x400000); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 2) == + 0x500000); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 3) == + 0x540000); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 4) == + 0x550000); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 5) == + 0x554000); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 6) == + 0x555000); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 7) == + 0x555400); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 8) == + 0x555600); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 9) == + 0x555800); constexpr u32 ValidateLayerSize(PixelFormat format, u32 width, u32 height, u32 block_height, u32 tile_width_spacing, u32 level) { const Extent3D size{width, height, 1}; const Extent3D block{0, block_height, 0}; - const u32 offset = CalculateLevelOffset(format, size, block, 1, tile_width_spacing, level); + const u32 offset = CalculateLevelOffset(format, size, block, tile_width_spacing, level); return AlignLayerSize(offset, size, block, DefaultBlockHeight(format), tile_width_spacing); } diff --git a/src/video_core/textures/texture.h b/src/video_core/textures/texture.h index c1d14335e..1a9399455 100644 --- a/src/video_core/textures/texture.h +++ b/src/video_core/textures/texture.h @@ -154,6 +154,15 @@ union TextureHandle { }; static_assert(sizeof(TextureHandle) == 4, "TextureHandle has wrong size"); +[[nodiscard]] inline std::pair<u32, u32> TexturePair(u32 raw, bool via_header_index) { + if (via_header_index) { + return {raw, raw}; + } else { + const Tegra::Texture::TextureHandle handle{raw}; + return {handle.tic_id, via_header_index ? handle.tic_id : handle.tsc_id}; + } +} + struct TICEntry { union { struct { diff --git a/src/video_core/transform_feedback.cpp b/src/video_core/transform_feedback.cpp new file mode 100644 index 000000000..ba26ac3f1 --- /dev/null +++ b/src/video_core/transform_feedback.cpp @@ -0,0 +1,99 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <algorithm> +#include <array> +#include <vector> + +#include "common/alignment.h" +#include "common/assert.h" +#include "shader_recompiler/shader_info.h" +#include "video_core/transform_feedback.h" + +namespace VideoCommon { + +std::vector<Shader::TransformFeedbackVarying> MakeTransformFeedbackVaryings( + const TransformFeedbackState& state) { + static constexpr std::array VECTORS{ + 28, // gl_Position + 32, // Generic 0 + 36, // Generic 1 + 40, // Generic 2 + 44, // Generic 3 + 48, // Generic 4 + 52, // Generic 5 + 56, // Generic 6 + 60, // Generic 7 + 64, // Generic 8 + 68, // Generic 9 + 72, // Generic 10 + 76, // Generic 11 + 80, // Generic 12 + 84, // Generic 13 + 88, // Generic 14 + 92, // Generic 15 + 96, // Generic 16 + 100, // Generic 17 + 104, // Generic 18 + 108, // Generic 19 + 112, // Generic 20 + 116, // Generic 21 + 120, // Generic 22 + 124, // Generic 23 + 128, // Generic 24 + 132, // Generic 25 + 136, // Generic 26 + 140, // Generic 27 + 144, // Generic 28 + 148, // Generic 29 + 152, // Generic 30 + 156, // Generic 31 + 160, // gl_FrontColor + 164, // gl_FrontSecondaryColor + 160, // gl_BackColor + 164, // gl_BackSecondaryColor + 192, // gl_TexCoord[0] + 196, // gl_TexCoord[1] + 200, // gl_TexCoord[2] + 204, // gl_TexCoord[3] + 208, // gl_TexCoord[4] + 212, // gl_TexCoord[5] + 216, // gl_TexCoord[6] + 220, // gl_TexCoord[7] + }; + std::vector<Shader::TransformFeedbackVarying> xfb(256); + for (size_t buffer = 0; buffer < state.layouts.size(); ++buffer) { + const auto& locations = state.varyings[buffer]; + const auto& layout = state.layouts[buffer]; + const u32 varying_count = layout.varying_count; + u32 highest = 0; + for (u32 offset = 0; offset < varying_count; ++offset) { + const u32 base_offset = offset; + const u8 location = locations[offset]; + + UNIMPLEMENTED_IF_MSG(layout.stream != 0, "Stream is not zero: {}", layout.stream); + Shader::TransformFeedbackVarying varying{ + .buffer = static_cast<u32>(buffer), + .stride = layout.stride, + .offset = offset * 4, + .components = 1, + }; + if (std::ranges::find(VECTORS, Common::AlignDown(location, 4)) != VECTORS.end()) { + UNIMPLEMENTED_IF_MSG(location % 4 != 0, "Unaligned TFB"); + + const u8 base_index = location / 4; + while (offset + 1 < varying_count && base_index == locations[offset + 1] / 4) { + ++offset; + ++varying.components; + } + } + xfb[location] = varying; + highest = std::max(highest, (base_offset + varying.components) * 4); + } + UNIMPLEMENTED_IF(highest != layout.stride); + } + return xfb; +} + +} // namespace VideoCommon diff --git a/src/video_core/transform_feedback.h b/src/video_core/transform_feedback.h new file mode 100644 index 000000000..8f6946d65 --- /dev/null +++ b/src/video_core/transform_feedback.h @@ -0,0 +1,30 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <array> +#include <vector> + +#include "common/common_types.h" +#include "shader_recompiler/runtime_info.h" +#include "video_core/engines/maxwell_3d.h" + +namespace VideoCommon { + +struct TransformFeedbackState { + struct Layout { + u32 stream; + u32 varying_count; + u32 stride; + }; + std::array<Layout, Tegra::Engines::Maxwell3D::Regs::NumTransformFeedbackBuffers> layouts; + std::array<std::array<u8, 128>, Tegra::Engines::Maxwell3D::Regs::NumTransformFeedbackBuffers> + varyings; +}; + +std::vector<Shader::TransformFeedbackVarying> MakeTransformFeedbackVaryings( + const TransformFeedbackState& state); + +} // namespace VideoCommon diff --git a/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp b/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp index 758c038ba..fdd1a5081 100644 --- a/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp +++ b/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp @@ -73,12 +73,11 @@ NsightAftermathTracker::~NsightAftermathTracker() { } } -void NsightAftermathTracker::SaveShader(const std::vector<u32>& spirv) const { +void NsightAftermathTracker::SaveShader(std::span<const u32> spirv) const { if (!initialized) { return; } - - std::vector<u32> spirv_copy = spirv; + std::vector<u32> spirv_copy(spirv.begin(), spirv.end()); GFSDK_Aftermath_SpirvCode shader; shader.pData = spirv_copy.data(); shader.size = static_cast<u32>(spirv_copy.size() * 4); @@ -100,7 +99,7 @@ void NsightAftermathTracker::SaveShader(const std::vector<u32>& spirv) const { LOG_ERROR(Render_Vulkan, "Failed to dump SPIR-V module with hash={:016x}", hash.hash); return; } - if (file.Write(spirv) != spirv.size()) { + if (file.WriteSpan(spirv) != spirv.size()) { LOG_ERROR(Render_Vulkan, "Failed to write SPIR-V module with hash={:016x}", hash.hash); return; } diff --git a/src/video_core/vulkan_common/nsight_aftermath_tracker.h b/src/video_core/vulkan_common/nsight_aftermath_tracker.h index 4fe2b14d9..eae1891dd 100644 --- a/src/video_core/vulkan_common/nsight_aftermath_tracker.h +++ b/src/video_core/vulkan_common/nsight_aftermath_tracker.h @@ -6,6 +6,7 @@ #include <filesystem> #include <mutex> +#include <span> #include <string> #include <vector> @@ -33,7 +34,7 @@ public: NsightAftermathTracker(NsightAftermathTracker&&) = delete; NsightAftermathTracker& operator=(NsightAftermathTracker&&) = delete; - void SaveShader(const std::vector<u32>& spirv) const; + void SaveShader(std::span<const u32> spirv) const; private: #ifdef HAS_NSIGHT_AFTERMATH @@ -61,21 +62,21 @@ private: bool initialized = false; Common::DynamicLibrary dl; - PFN_GFSDK_Aftermath_DisableGpuCrashDumps GFSDK_Aftermath_DisableGpuCrashDumps; - PFN_GFSDK_Aftermath_EnableGpuCrashDumps GFSDK_Aftermath_EnableGpuCrashDumps; - PFN_GFSDK_Aftermath_GetShaderDebugInfoIdentifier GFSDK_Aftermath_GetShaderDebugInfoIdentifier; - PFN_GFSDK_Aftermath_GetShaderHashSpirv GFSDK_Aftermath_GetShaderHashSpirv; - PFN_GFSDK_Aftermath_GpuCrashDump_CreateDecoder GFSDK_Aftermath_GpuCrashDump_CreateDecoder; - PFN_GFSDK_Aftermath_GpuCrashDump_DestroyDecoder GFSDK_Aftermath_GpuCrashDump_DestroyDecoder; - PFN_GFSDK_Aftermath_GpuCrashDump_GenerateJSON GFSDK_Aftermath_GpuCrashDump_GenerateJSON; - PFN_GFSDK_Aftermath_GpuCrashDump_GetJSON GFSDK_Aftermath_GpuCrashDump_GetJSON; + PFN_GFSDK_Aftermath_DisableGpuCrashDumps GFSDK_Aftermath_DisableGpuCrashDumps{}; + PFN_GFSDK_Aftermath_EnableGpuCrashDumps GFSDK_Aftermath_EnableGpuCrashDumps{}; + PFN_GFSDK_Aftermath_GetShaderDebugInfoIdentifier GFSDK_Aftermath_GetShaderDebugInfoIdentifier{}; + PFN_GFSDK_Aftermath_GetShaderHashSpirv GFSDK_Aftermath_GetShaderHashSpirv{}; + PFN_GFSDK_Aftermath_GpuCrashDump_CreateDecoder GFSDK_Aftermath_GpuCrashDump_CreateDecoder{}; + PFN_GFSDK_Aftermath_GpuCrashDump_DestroyDecoder GFSDK_Aftermath_GpuCrashDump_DestroyDecoder{}; + PFN_GFSDK_Aftermath_GpuCrashDump_GenerateJSON GFSDK_Aftermath_GpuCrashDump_GenerateJSON{}; + PFN_GFSDK_Aftermath_GpuCrashDump_GetJSON GFSDK_Aftermath_GpuCrashDump_GetJSON{}; #endif }; #ifndef HAS_NSIGHT_AFTERMATH inline NsightAftermathTracker::NsightAftermathTracker() = default; inline NsightAftermathTracker::~NsightAftermathTracker() = default; -inline void NsightAftermathTracker::SaveShader(const std::vector<u32>&) const {} +inline void NsightAftermathTracker::SaveShader(std::span<const u32>) const {} #endif } // namespace Vulkan diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index f214510da..44afdc1cd 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -2,6 +2,7 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <algorithm> #include <bitset> #include <chrono> #include <optional> @@ -33,6 +34,12 @@ constexpr std::array DEPTH16_UNORM_STENCIL8_UINT{ }; } // namespace Alternatives +enum class NvidiaArchitecture { + AmpereOrNewer, + Turing, + VoltaOrOlder, +}; + constexpr std::array REQUIRED_EXTENSIONS{ VK_KHR_MAINTENANCE1_EXTENSION_NAME, VK_KHR_STORAGE_BUFFER_STORAGE_CLASS_EXTENSION_NAME, @@ -43,11 +50,14 @@ constexpr std::array REQUIRED_EXTENSIONS{ VK_KHR_DESCRIPTOR_UPDATE_TEMPLATE_EXTENSION_NAME, VK_KHR_TIMELINE_SEMAPHORE_EXTENSION_NAME, VK_KHR_SAMPLER_MIRROR_CLAMP_TO_EDGE_EXTENSION_NAME, + VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME, + VK_KHR_VARIABLE_POINTERS_EXTENSION_NAME, VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME, VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME, VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME, VK_EXT_ROBUSTNESS_2_EXTENSION_NAME, VK_EXT_HOST_QUERY_RESET_EXTENSION_NAME, + VK_EXT_SHADER_DEMOTE_TO_HELPER_INVOCATION_EXTENSION_NAME, #ifdef _WIN32 VK_KHR_EXTERNAL_MEMORY_WIN32_EXTENSION_NAME, #endif @@ -112,6 +122,7 @@ std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties(vk::Physica VK_FORMAT_R16G16_SFLOAT, VK_FORMAT_R16G16_SINT, VK_FORMAT_R16_UNORM, + VK_FORMAT_R16_SNORM, VK_FORMAT_R16_UINT, VK_FORMAT_R8G8B8A8_SRGB, VK_FORMAT_R8G8_UNORM, @@ -191,15 +202,47 @@ std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties(vk::Physica return format_properties; } +std::vector<std::string> GetSupportedExtensions(vk::PhysicalDevice physical) { + const std::vector extensions = physical.EnumerateDeviceExtensionProperties(); + std::vector<std::string> supported_extensions; + supported_extensions.reserve(extensions.size()); + for (const auto& extension : extensions) { + supported_extensions.emplace_back(extension.extensionName); + } + return supported_extensions; +} + +NvidiaArchitecture GetNvidiaArchitecture(vk::PhysicalDevice physical, + std::span<const std::string> exts) { + if (std::ranges::find(exts, VK_KHR_FRAGMENT_SHADING_RATE_EXTENSION_NAME) != exts.end()) { + VkPhysicalDeviceFragmentShadingRatePropertiesKHR shading_rate_props{}; + shading_rate_props.sType = + VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FRAGMENT_SHADING_RATE_PROPERTIES_KHR; + VkPhysicalDeviceProperties2KHR physical_properties{}; + physical_properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2_KHR; + physical_properties.pNext = &shading_rate_props; + physical.GetProperties2KHR(physical_properties); + if (shading_rate_props.primitiveFragmentShadingRateWithMultipleViewports) { + // Only Ampere and newer support this feature + return NvidiaArchitecture::AmpereOrNewer; + } + } + if (std::ranges::find(exts, VK_NV_SHADING_RATE_IMAGE_EXTENSION_NAME) != exts.end()) { + return NvidiaArchitecture::Turing; + } + return NvidiaArchitecture::VoltaOrOlder; +} } // Anonymous namespace Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR surface, const vk::InstanceDispatch& dld_) : instance{instance_}, dld{dld_}, physical{physical_}, properties{physical.GetProperties()}, - format_properties{GetFormatProperties(physical)} { + supported_extensions{GetSupportedExtensions(physical)}, + format_properties(GetFormatProperties(physical)) { CheckSuitability(surface != nullptr); SetupFamilies(surface); SetupFeatures(); + SetupProperties(); const auto queue_cis = GetDeviceQueueCreateInfos(); const std::vector extensions = LoadExtensions(surface != nullptr); @@ -214,16 +257,16 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR .independentBlend = true, .geometryShader = true, .tessellationShader = true, - .sampleRateShading = false, - .dualSrcBlend = false, + .sampleRateShading = true, + .dualSrcBlend = true, .logicOp = false, .multiDrawIndirect = false, .drawIndirectFirstInstance = false, .depthClamp = true, .depthBiasClamp = true, - .fillModeNonSolid = false, - .depthBounds = false, - .wideLines = false, + .fillModeNonSolid = true, + .depthBounds = is_depth_bounds_supported, + .wideLines = true, .largePoints = true, .alphaToOne = false, .multiViewport = true, @@ -245,11 +288,11 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR .shaderSampledImageArrayDynamicIndexing = false, .shaderStorageBufferArrayDynamicIndexing = false, .shaderStorageImageArrayDynamicIndexing = false, - .shaderClipDistance = false, - .shaderCullDistance = false, - .shaderFloat64 = false, - .shaderInt64 = false, - .shaderInt16 = false, + .shaderClipDistance = true, + .shaderCullDistance = true, + .shaderFloat64 = is_shader_float64_supported, + .shaderInt64 = is_shader_int64_supported, + .shaderInt16 = is_shader_int16_supported, .shaderResourceResidency = false, .shaderResourceMinLod = false, .sparseBinding = false, @@ -278,7 +321,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR VkPhysicalDevice16BitStorageFeaturesKHR bit16_storage{ .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES_KHR, .pNext = nullptr, - .storageBuffer16BitAccess = false, + .storageBuffer16BitAccess = true, .uniformAndStorageBuffer16BitAccess = true, .storagePushConstant16 = false, .storageInputOutput16 = false, @@ -310,6 +353,21 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR }; SetNext(next, host_query_reset); + VkPhysicalDeviceVariablePointerFeaturesKHR variable_pointers{ + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES_KHR, + .pNext = nullptr, + .variablePointersStorageBuffer = VK_TRUE, + .variablePointers = VK_TRUE, + }; + SetNext(next, variable_pointers); + + VkPhysicalDeviceShaderDemoteToHelperInvocationFeaturesEXT demote{ + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DEMOTE_TO_HELPER_INVOCATION_FEATURES_EXT, + .pNext = nullptr, + .shaderDemoteToHelperInvocation = true, + }; + SetNext(next, demote); + VkPhysicalDeviceFloat16Int8FeaturesKHR float16_int8; if (is_float16_supported) { float16_int8 = { @@ -327,6 +385,14 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR LOG_INFO(Render_Vulkan, "Device doesn't support viewport swizzles"); } + if (!nv_viewport_array2) { + LOG_INFO(Render_Vulkan, "Device doesn't support viewport masks"); + } + + if (!nv_geometry_shader_passthrough) { + LOG_INFO(Render_Vulkan, "Device doesn't support passthrough geometry shaders"); + } + VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR std430_layout; if (khr_uniform_buffer_standard_layout) { std430_layout = { @@ -389,12 +455,83 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR LOG_INFO(Render_Vulkan, "Device doesn't support extended dynamic state"); } + VkPhysicalDeviceLineRasterizationFeaturesEXT line_raster; + if (ext_line_rasterization) { + line_raster = { + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_LINE_RASTERIZATION_FEATURES_EXT, + .pNext = nullptr, + .rectangularLines = VK_TRUE, + .bresenhamLines = VK_FALSE, + .smoothLines = VK_TRUE, + .stippledRectangularLines = VK_FALSE, + .stippledBresenhamLines = VK_FALSE, + .stippledSmoothLines = VK_FALSE, + }; + SetNext(next, line_raster); + } else { + LOG_INFO(Render_Vulkan, "Device doesn't support smooth lines"); + } + + if (!ext_conservative_rasterization) { + LOG_INFO(Render_Vulkan, "Device doesn't support conservative rasterization"); + } + + VkPhysicalDeviceProvokingVertexFeaturesEXT provoking_vertex; + if (ext_provoking_vertex) { + provoking_vertex = { + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROVOKING_VERTEX_FEATURES_EXT, + .pNext = nullptr, + .provokingVertexLast = VK_TRUE, + .transformFeedbackPreservesProvokingVertex = VK_TRUE, + }; + SetNext(next, provoking_vertex); + } else { + LOG_INFO(Render_Vulkan, "Device doesn't support provoking vertex last"); + } + + VkPhysicalDeviceVertexInputDynamicStateFeaturesEXT vertex_input_dynamic; + if (ext_vertex_input_dynamic_state) { + vertex_input_dynamic = { + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_INPUT_DYNAMIC_STATE_FEATURES_EXT, + .pNext = nullptr, + .vertexInputDynamicState = VK_TRUE, + }; + SetNext(next, vertex_input_dynamic); + } else { + LOG_INFO(Render_Vulkan, "Device doesn't support vertex input dynamic state"); + } + + VkPhysicalDeviceShaderAtomicInt64FeaturesKHR atomic_int64; + if (ext_shader_atomic_int64) { + atomic_int64 = { + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_INT64_FEATURES_KHR, + .pNext = nullptr, + .shaderBufferInt64Atomics = VK_TRUE, + .shaderSharedInt64Atomics = VK_TRUE, + }; + SetNext(next, atomic_int64); + } + + VkPhysicalDeviceWorkgroupMemoryExplicitLayoutFeaturesKHR workgroup_layout; + if (khr_workgroup_memory_explicit_layout) { + workgroup_layout = { + .sType = + VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_FEATURES_KHR, + .pNext = nullptr, + .workgroupMemoryExplicitLayout = VK_TRUE, + .workgroupMemoryExplicitLayoutScalarBlockLayout = VK_TRUE, + .workgroupMemoryExplicitLayout8BitAccess = VK_TRUE, + .workgroupMemoryExplicitLayout16BitAccess = VK_TRUE, + }; + SetNext(next, workgroup_layout); + } + if (!ext_depth_range_unrestricted) { LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted"); } VkDeviceDiagnosticsConfigCreateInfoNV diagnostics_nv; - if (nv_device_diagnostics_config) { + if (Settings::values.enable_nsight_aftermath && nv_device_diagnostics_config) { nsight_aftermath_tracker = std::make_unique<NsightAftermathTracker>(); diagnostics_nv = { @@ -412,11 +549,33 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR CollectTelemetryParameters(); CollectToolingInfo(); + if (driver_id == VK_DRIVER_ID_NVIDIA_PROPRIETARY_KHR) { + const auto arch = GetNvidiaArchitecture(physical, supported_extensions); + switch (arch) { + case NvidiaArchitecture::AmpereOrNewer: + LOG_WARNING(Render_Vulkan, "Blacklisting Ampere devices from float16 math"); + is_float16_supported = false; + break; + case NvidiaArchitecture::Turing: + break; + case NvidiaArchitecture::VoltaOrOlder: + LOG_WARNING(Render_Vulkan, "Blacklisting Volta and older from VK_KHR_push_descriptor"); + khr_push_descriptor = false; + break; + } + } if (ext_extended_dynamic_state && driver_id == VK_DRIVER_ID_MESA_RADV) { - LOG_WARNING( - Render_Vulkan, - "Blacklisting RADV for VK_EXT_extended_dynamic state, likely due to a bug in yuzu"); - ext_extended_dynamic_state = false; + // Mask driver version variant + const u32 version = (properties.driverVersion << 3) >> 3; + if (version < VK_MAKE_API_VERSION(0, 21, 2, 0)) { + LOG_WARNING(Render_Vulkan, + "RADV versions older than 21.2 have broken VK_EXT_extended_dynamic_state"); + ext_extended_dynamic_state = false; + } + } + if (ext_vertex_input_dynamic_state && driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS) { + LOG_WARNING(Render_Vulkan, "Blacklisting Intel for VK_EXT_vertex_input_dynamic_state"); + ext_vertex_input_dynamic_state = false; } if (is_float16_supported && driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS) { // Intel's compiler crashes when using fp16 on Astral Chain, disable it for the time being. @@ -426,8 +585,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR graphics_queue = logical.GetQueue(graphics_family); present_queue = logical.GetQueue(present_family); - - use_asynchronous_shaders = Settings::values.use_asynchronous_shaders.GetValue(); } Device::~Device() = default; @@ -471,7 +628,7 @@ void Device::ReportLoss() const { std::this_thread::sleep_for(std::chrono::seconds{15}); } -void Device::SaveShader(const std::vector<u32>& spirv) const { +void Device::SaveShader(std::span<const u32> spirv) const { if (nsight_aftermath_tracker) { nsight_aftermath_tracker->SaveShader(spirv); } @@ -597,10 +754,20 @@ void Device::CheckSuitability(bool requires_swapchain) const { throw vk::Exception(VK_ERROR_FEATURE_NOT_PRESENT); } } + VkPhysicalDeviceShaderDemoteToHelperInvocationFeaturesEXT demote{}; + demote.sType = + VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DEMOTE_TO_HELPER_INVOCATION_FEATURES_EXT; + demote.pNext = nullptr; + + VkPhysicalDeviceVariablePointerFeaturesKHR variable_pointers{}; + variable_pointers.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES_KHR; + variable_pointers.pNext = &demote; + VkPhysicalDeviceRobustness2FeaturesEXT robustness2{}; robustness2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_FEATURES_EXT; + robustness2.pNext = &variable_pointers; - VkPhysicalDeviceFeatures2 features2{}; + VkPhysicalDeviceFeatures2KHR features2{}; features2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2; features2.pNext = &robustness2; @@ -610,7 +777,6 @@ void Device::CheckSuitability(bool requires_swapchain) const { const std::array feature_report{ std::make_pair(features.robustBufferAccess, "robustBufferAccess"), std::make_pair(features.vertexPipelineStoresAndAtomics, "vertexPipelineStoresAndAtomics"), - std::make_pair(features.robustBufferAccess, "robustBufferAccess"), std::make_pair(features.imageCubeArray, "imageCubeArray"), std::make_pair(features.independentBlend, "independentBlend"), std::make_pair(features.depthClamp, "depthClamp"), @@ -618,13 +784,23 @@ void Device::CheckSuitability(bool requires_swapchain) const { std::make_pair(features.largePoints, "largePoints"), std::make_pair(features.multiViewport, "multiViewport"), std::make_pair(features.depthBiasClamp, "depthBiasClamp"), + std::make_pair(features.fillModeNonSolid, "fillModeNonSolid"), + std::make_pair(features.wideLines, "wideLines"), std::make_pair(features.geometryShader, "geometryShader"), std::make_pair(features.tessellationShader, "tessellationShader"), + std::make_pair(features.sampleRateShading, "sampleRateShading"), + std::make_pair(features.dualSrcBlend, "dualSrcBlend"), std::make_pair(features.occlusionQueryPrecise, "occlusionQueryPrecise"), std::make_pair(features.fragmentStoresAndAtomics, "fragmentStoresAndAtomics"), std::make_pair(features.shaderImageGatherExtended, "shaderImageGatherExtended"), std::make_pair(features.shaderStorageImageWriteWithoutFormat, "shaderStorageImageWriteWithoutFormat"), + std::make_pair(features.shaderClipDistance, "shaderClipDistance"), + std::make_pair(features.shaderCullDistance, "shaderCullDistance"), + std::make_pair(demote.shaderDemoteToHelperInvocation, "shaderDemoteToHelperInvocation"), + std::make_pair(variable_pointers.variablePointers, "variablePointers"), + std::make_pair(variable_pointers.variablePointersStorageBuffer, + "variablePointersStorageBuffer"), std::make_pair(robustness2.robustBufferAccess2, "robustBufferAccess2"), std::make_pair(robustness2.robustImageAccess2, "robustImageAccess2"), std::make_pair(robustness2.nullDescriptor, "nullDescriptor"), @@ -647,14 +823,19 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) { } bool has_khr_shader_float16_int8{}; + bool has_khr_workgroup_memory_explicit_layout{}; bool has_ext_subgroup_size_control{}; bool has_ext_transform_feedback{}; bool has_ext_custom_border_color{}; bool has_ext_extended_dynamic_state{}; - for (const VkExtensionProperties& extension : physical.EnumerateDeviceExtensionProperties()) { + bool has_ext_shader_atomic_int64{}; + bool has_ext_provoking_vertex{}; + bool has_ext_vertex_input_dynamic_state{}; + bool has_ext_line_rasterization{}; + for (const std::string& extension : supported_extensions) { const auto test = [&](std::optional<std::reference_wrapper<bool>> status, const char* name, bool push) { - if (extension.extensionName != std::string_view(name)) { + if (extension != name) { return; } if (push) { @@ -665,8 +846,13 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) { } }; test(nv_viewport_swizzle, VK_NV_VIEWPORT_SWIZZLE_EXTENSION_NAME, true); + test(nv_viewport_array2, VK_NV_VIEWPORT_ARRAY2_EXTENSION_NAME, true); + test(nv_geometry_shader_passthrough, VK_NV_GEOMETRY_SHADER_PASSTHROUGH_EXTENSION_NAME, + true); test(khr_uniform_buffer_standard_layout, VK_KHR_UNIFORM_BUFFER_STANDARD_LAYOUT_EXTENSION_NAME, true); + test(khr_spirv_1_4, VK_KHR_SPIRV_1_4_EXTENSION_NAME, true); + test(khr_push_descriptor, VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME, true); test(has_khr_shader_float16_int8, VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME, false); test(ext_depth_range_unrestricted, VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME, true); test(ext_index_type_uint8, VK_EXT_INDEX_TYPE_UINT8_EXTENSION_NAME, true); @@ -675,16 +861,25 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) { true); test(ext_tooling_info, VK_EXT_TOOLING_INFO_EXTENSION_NAME, true); test(ext_shader_stencil_export, VK_EXT_SHADER_STENCIL_EXPORT_EXTENSION_NAME, true); + test(ext_conservative_rasterization, VK_EXT_CONSERVATIVE_RASTERIZATION_EXTENSION_NAME, + true); test(has_ext_transform_feedback, VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME, false); test(has_ext_custom_border_color, VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME, false); test(has_ext_extended_dynamic_state, VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME, false); test(has_ext_subgroup_size_control, VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME, false); - if (Settings::values.renderer_debug) { + test(has_ext_provoking_vertex, VK_EXT_PROVOKING_VERTEX_EXTENSION_NAME, false); + test(has_ext_vertex_input_dynamic_state, VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME, + false); + test(has_ext_shader_atomic_int64, VK_KHR_SHADER_ATOMIC_INT64_EXTENSION_NAME, false); + test(has_khr_workgroup_memory_explicit_layout, + VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME, false); + test(has_ext_line_rasterization, VK_EXT_LINE_RASTERIZATION_EXTENSION_NAME, false); + if (Settings::values.enable_nsight_aftermath) { test(nv_device_diagnostics_config, VK_NV_DEVICE_DIAGNOSTICS_CONFIG_EXTENSION_NAME, true); } } - VkPhysicalDeviceFeatures2KHR features; + VkPhysicalDeviceFeatures2KHR features{}; features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2_KHR; VkPhysicalDeviceProperties2KHR physical_properties; @@ -722,10 +917,49 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) { subgroup_properties.maxSubgroupSize >= GuestWarpSize) { extensions.push_back(VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME); guest_warp_stages = subgroup_properties.requiredSubgroupSizeStages; + ext_subgroup_size_control = true; } } else { is_warp_potentially_bigger = true; } + if (has_ext_provoking_vertex) { + VkPhysicalDeviceProvokingVertexFeaturesEXT provoking_vertex; + provoking_vertex.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROVOKING_VERTEX_FEATURES_EXT; + provoking_vertex.pNext = nullptr; + features.pNext = &provoking_vertex; + physical.GetFeatures2KHR(features); + + if (provoking_vertex.provokingVertexLast && + provoking_vertex.transformFeedbackPreservesProvokingVertex) { + extensions.push_back(VK_EXT_PROVOKING_VERTEX_EXTENSION_NAME); + ext_provoking_vertex = true; + } + } + if (has_ext_vertex_input_dynamic_state) { + VkPhysicalDeviceVertexInputDynamicStateFeaturesEXT vertex_input; + vertex_input.sType = + VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_INPUT_DYNAMIC_STATE_FEATURES_EXT; + vertex_input.pNext = nullptr; + features.pNext = &vertex_input; + physical.GetFeatures2KHR(features); + + if (vertex_input.vertexInputDynamicState) { + extensions.push_back(VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME); + ext_vertex_input_dynamic_state = true; + } + } + if (has_ext_shader_atomic_int64) { + VkPhysicalDeviceShaderAtomicInt64Features atomic_int64; + atomic_int64.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTENDED_DYNAMIC_STATE_FEATURES_EXT; + atomic_int64.pNext = nullptr; + features.pNext = &atomic_int64; + physical.GetFeatures2KHR(features); + + if (atomic_int64.shaderBufferInt64Atomics && atomic_int64.shaderSharedInt64Atomics) { + extensions.push_back(VK_KHR_SHADER_ATOMIC_INT64_EXTENSION_NAME); + ext_shader_atomic_int64 = true; + } + } if (has_ext_transform_feedback) { VkPhysicalDeviceTransformFeedbackFeaturesEXT tfb_features; tfb_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT; @@ -760,17 +994,55 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) { } } if (has_ext_extended_dynamic_state) { - VkPhysicalDeviceExtendedDynamicStateFeaturesEXT dynamic_state; - dynamic_state.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTENDED_DYNAMIC_STATE_FEATURES_EXT; - dynamic_state.pNext = nullptr; - features.pNext = &dynamic_state; + VkPhysicalDeviceExtendedDynamicStateFeaturesEXT extended_dynamic_state; + extended_dynamic_state.sType = + VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTENDED_DYNAMIC_STATE_FEATURES_EXT; + extended_dynamic_state.pNext = nullptr; + features.pNext = &extended_dynamic_state; physical.GetFeatures2KHR(features); - if (dynamic_state.extendedDynamicState) { + if (extended_dynamic_state.extendedDynamicState) { extensions.push_back(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME); ext_extended_dynamic_state = true; } } + if (has_ext_line_rasterization) { + VkPhysicalDeviceLineRasterizationFeaturesEXT line_raster; + line_raster.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_LINE_RASTERIZATION_FEATURES_EXT; + line_raster.pNext = nullptr; + features.pNext = &line_raster; + physical.GetFeatures2KHR(features); + if (line_raster.rectangularLines && line_raster.smoothLines) { + extensions.push_back(VK_EXT_LINE_RASTERIZATION_EXTENSION_NAME); + ext_line_rasterization = true; + } + } + if (has_khr_workgroup_memory_explicit_layout) { + VkPhysicalDeviceWorkgroupMemoryExplicitLayoutFeaturesKHR layout; + layout.sType = + VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_FEATURES_KHR; + layout.pNext = nullptr; + features.pNext = &layout; + physical.GetFeatures2KHR(features); + + if (layout.workgroupMemoryExplicitLayout && + layout.workgroupMemoryExplicitLayout8BitAccess && + layout.workgroupMemoryExplicitLayout16BitAccess && + layout.workgroupMemoryExplicitLayoutScalarBlockLayout) { + extensions.push_back(VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME); + khr_workgroup_memory_explicit_layout = true; + } + } + if (khr_push_descriptor) { + VkPhysicalDevicePushDescriptorPropertiesKHR push_descriptor; + push_descriptor.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR; + push_descriptor.pNext = nullptr; + + physical_properties.pNext = &push_descriptor; + physical.GetProperties2KHR(physical_properties); + + max_push_descriptors = push_descriptor.maxPushDescriptors; + } return extensions; } @@ -806,11 +1078,25 @@ void Device::SetupFamilies(VkSurfaceKHR surface) { } void Device::SetupFeatures() { - const auto supported_features{physical.GetFeatures()}; - is_formatless_image_load_supported = supported_features.shaderStorageImageReadWithoutFormat; - is_shader_storage_image_multisample = supported_features.shaderStorageImageMultisample; + const VkPhysicalDeviceFeatures features{physical.GetFeatures()}; + is_depth_bounds_supported = features.depthBounds; + is_formatless_image_load_supported = features.shaderStorageImageReadWithoutFormat; + is_shader_float64_supported = features.shaderFloat64; + is_shader_int64_supported = features.shaderInt64; + is_shader_int16_supported = features.shaderInt16; + is_shader_storage_image_multisample = features.shaderStorageImageMultisample; is_blit_depth_stencil_supported = TestDepthStencilBlits(); - is_optimal_astc_supported = IsOptimalAstcSupported(supported_features); + is_optimal_astc_supported = IsOptimalAstcSupported(features); +} + +void Device::SetupProperties() { + float_controls.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT_CONTROLS_PROPERTIES_KHR; + + VkPhysicalDeviceProperties2KHR properties2{}; + properties2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2_KHR; + properties2.pNext = &float_controls; + + physical.GetProperties2KHR(properties2); } void Device::CollectTelemetryParameters() { @@ -832,12 +1118,6 @@ void Device::CollectTelemetryParameters() { driver_id = driver.driverID; vendor_name = driver.driverName; - - const std::vector extensions = physical.EnumerateDeviceExtensionProperties(); - reported_extensions.reserve(std::size(extensions)); - for (const auto& extension : extensions) { - reported_extensions.emplace_back(extension.extensionName); - } } void Device::CollectPhysicalMemoryInfo() { diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 96c0f8c60..df394e384 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -4,6 +4,7 @@ #pragma once +#include <span> #include <string> #include <string_view> #include <unordered_map> @@ -43,7 +44,7 @@ public: void ReportLoss() const; /// Reports a shader to Nsight Aftermath. - void SaveShader(const std::vector<u32>& spirv) const; + void SaveShader(std::span<const u32> spirv) const; /// Returns the name of the VkDriverId reported from Vulkan. std::string GetDriverName() const; @@ -128,6 +129,11 @@ public: return properties.limits.maxComputeSharedMemorySize; } + /// Returns float control properties of the device. + const VkPhysicalDeviceFloatControlsPropertiesKHR& FloatControlProperties() const { + return float_controls; + } + /// Returns true if ASTC is natively supported. bool IsOptimalAstcSupported() const { return is_optimal_astc_supported; @@ -148,11 +154,31 @@ public: return guest_warp_stages & stage; } + /// Returns the maximum number of push descriptors. + u32 MaxPushDescriptors() const { + return max_push_descriptors; + } + /// Returns true if formatless image load is supported. bool IsFormatlessImageLoadSupported() const { return is_formatless_image_load_supported; } + /// Returns true if shader int64 is supported. + bool IsShaderInt64Supported() const { + return is_shader_int64_supported; + } + + /// Returns true if shader int16 is supported. + bool IsShaderInt16Supported() const { + return is_shader_int16_supported; + } + + // Returns true if depth bounds is supported. + bool IsDepthBoundsSupported() const { + return is_depth_bounds_supported; + } + /// Returns true when blitting from and to depth stencil images is supported. bool IsBlitDepthStencilSupported() const { return is_blit_depth_stencil_supported; @@ -163,11 +189,36 @@ public: return nv_viewport_swizzle; } - /// Returns true if the device supports VK_EXT_scalar_block_layout. + /// Returns true if the device supports VK_NV_viewport_array2. + bool IsNvViewportArray2Supported() const { + return nv_viewport_array2; + } + + /// Returns true if the device supports VK_NV_geometry_shader_passthrough. + bool IsNvGeometryShaderPassthroughSupported() const { + return nv_geometry_shader_passthrough; + } + + /// Returns true if the device supports VK_KHR_uniform_buffer_standard_layout. bool IsKhrUniformBufferStandardLayoutSupported() const { return khr_uniform_buffer_standard_layout; } + /// Returns true if the device supports VK_KHR_spirv_1_4. + bool IsKhrSpirv1_4Supported() const { + return khr_spirv_1_4; + } + + /// Returns true if the device supports VK_KHR_push_descriptor. + bool IsKhrPushDescriptorSupported() const { + return khr_push_descriptor; + } + + /// Returns true if the device supports VK_KHR_workgroup_memory_explicit_layout. + bool IsKhrWorkgroupMemoryExplicitLayoutSupported() const { + return khr_workgroup_memory_explicit_layout; + } + /// Returns true if the device supports VK_EXT_index_type_uint8. bool IsExtIndexTypeUint8Supported() const { return ext_index_type_uint8; @@ -188,6 +239,11 @@ public: return ext_shader_viewport_index_layer; } + /// Returns true if the device supports VK_EXT_subgroup_size_control. + bool IsExtSubgroupSizeControlSupported() const { + return ext_subgroup_size_control; + } + /// Returns true if the device supports VK_EXT_transform_feedback. bool IsExtTransformFeedbackSupported() const { return ext_transform_feedback; @@ -203,11 +259,36 @@ public: return ext_extended_dynamic_state; } + /// Returns true if the device supports VK_EXT_line_rasterization. + bool IsExtLineRasterizationSupported() const { + return ext_line_rasterization; + } + + /// Returns true if the device supports VK_EXT_vertex_input_dynamic_state. + bool IsExtVertexInputDynamicStateSupported() const { + return ext_vertex_input_dynamic_state; + } + /// Returns true if the device supports VK_EXT_shader_stencil_export. bool IsExtShaderStencilExportSupported() const { return ext_shader_stencil_export; } + /// Returns true if the device supports VK_EXT_conservative_rasterization. + bool IsExtConservativeRasterizationSupported() const { + return ext_conservative_rasterization; + } + + /// Returns true if the device supports VK_EXT_provoking_vertex. + bool IsExtProvokingVertexSupported() const { + return ext_provoking_vertex; + } + + /// Returns true if the device supports VK_KHR_shader_atomic_int64. + bool IsExtShaderAtomicInt64Supported() const { + return ext_shader_atomic_int64; + } + /// Returns true when a known debugging tool is attached. bool HasDebuggingToolAttached() const { return has_renderdoc || has_nsight_graphics; @@ -220,12 +301,7 @@ public: /// Returns the list of available extensions. const std::vector<std::string>& GetAvailableExtensions() const { - return reported_extensions; - } - - /// Returns true if the setting for async shader compilation is enabled. - bool UseAsynchronousShaders() const { - return use_asynchronous_shaders; + return supported_extensions; } u64 GetDeviceLocalMemory() const { @@ -245,6 +321,9 @@ private: /// Sets up device features. void SetupFeatures(); + /// Sets up device properties. + void SetupProperties(); + /// Collects telemetry information from the device. void CollectTelemetryParameters(); @@ -267,46 +346,60 @@ private: bool IsFormatSupported(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage, FormatType format_type) const; - VkInstance instance; ///< Vulkan instance. - vk::DeviceDispatch dld; ///< Device function pointers. - vk::PhysicalDevice physical; ///< Physical device. - VkPhysicalDeviceProperties properties; ///< Device properties. - vk::Device logical; ///< Logical device. - vk::Queue graphics_queue; ///< Main graphics queue. - vk::Queue present_queue; ///< Main present queue. - u32 instance_version{}; ///< Vulkan onstance version. + VkInstance instance; ///< Vulkan instance. + vk::DeviceDispatch dld; ///< Device function pointers. + vk::PhysicalDevice physical; ///< Physical device. + VkPhysicalDeviceProperties properties; ///< Device properties. + VkPhysicalDeviceFloatControlsPropertiesKHR float_controls{}; ///< Float control properties. + vk::Device logical; ///< Logical device. + vk::Queue graphics_queue; ///< Main graphics queue. + vk::Queue present_queue; ///< Main present queue. + u32 instance_version{}; ///< Vulkan onstance version. u32 graphics_family{}; ///< Main graphics queue family index. u32 present_family{}; ///< Main present queue family index. VkDriverIdKHR driver_id{}; ///< Driver ID. VkShaderStageFlags guest_warp_stages{}; ///< Stages where the guest warp size can be forced. u64 device_access_memory{}; ///< Total size of device local memory in bytes. + u32 max_push_descriptors{}; ///< Maximum number of push descriptors bool is_optimal_astc_supported{}; ///< Support for native ASTC. bool is_float16_supported{}; ///< Support for float16 arithmetics. bool is_warp_potentially_bigger{}; ///< Host warp size can be bigger than guest. bool is_formatless_image_load_supported{}; ///< Support for shader image read without format. + bool is_depth_bounds_supported{}; ///< Support for depth bounds. + bool is_shader_float64_supported{}; ///< Support for float64. + bool is_shader_int64_supported{}; ///< Support for int64. + bool is_shader_int16_supported{}; ///< Support for int16. bool is_shader_storage_image_multisample{}; ///< Support for image operations on MSAA images. bool is_blit_depth_stencil_supported{}; ///< Support for blitting from and to depth stencil. bool nv_viewport_swizzle{}; ///< Support for VK_NV_viewport_swizzle. - bool khr_uniform_buffer_standard_layout{}; ///< Support for std430 on UBOs. - bool ext_index_type_uint8{}; ///< Support for VK_EXT_index_type_uint8. - bool ext_sampler_filter_minmax{}; ///< Support for VK_EXT_sampler_filter_minmax. - bool ext_depth_range_unrestricted{}; ///< Support for VK_EXT_depth_range_unrestricted. - bool ext_shader_viewport_index_layer{}; ///< Support for VK_EXT_shader_viewport_index_layer. - bool ext_tooling_info{}; ///< Support for VK_EXT_tooling_info. - bool ext_transform_feedback{}; ///< Support for VK_EXT_transform_feedback. - bool ext_custom_border_color{}; ///< Support for VK_EXT_custom_border_color. - bool ext_extended_dynamic_state{}; ///< Support for VK_EXT_extended_dynamic_state. - bool ext_shader_stencil_export{}; ///< Support for VK_EXT_shader_stencil_export. - bool nv_device_diagnostics_config{}; ///< Support for VK_NV_device_diagnostics_config. - bool has_renderdoc{}; ///< Has RenderDoc attached - bool has_nsight_graphics{}; ///< Has Nsight Graphics attached - - // Asynchronous Graphics Pipeline setting - bool use_asynchronous_shaders{}; ///< Setting to use asynchronous shaders/graphics pipeline + bool nv_viewport_array2{}; ///< Support for VK_NV_viewport_array2. + bool nv_geometry_shader_passthrough{}; ///< Support for VK_NV_geometry_shader_passthrough. + bool khr_uniform_buffer_standard_layout{}; ///< Support for scalar uniform buffer layouts. + bool khr_spirv_1_4{}; ///< Support for VK_KHR_spirv_1_4. + bool khr_workgroup_memory_explicit_layout{}; ///< Support for explicit workgroup layouts. + bool khr_push_descriptor{}; ///< Support for VK_KHR_push_descritor. + bool ext_index_type_uint8{}; ///< Support for VK_EXT_index_type_uint8. + bool ext_sampler_filter_minmax{}; ///< Support for VK_EXT_sampler_filter_minmax. + bool ext_depth_range_unrestricted{}; ///< Support for VK_EXT_depth_range_unrestricted. + bool ext_shader_viewport_index_layer{}; ///< Support for VK_EXT_shader_viewport_index_layer. + bool ext_tooling_info{}; ///< Support for VK_EXT_tooling_info. + bool ext_subgroup_size_control{}; ///< Support for VK_EXT_subgroup_size_control. + bool ext_transform_feedback{}; ///< Support for VK_EXT_transform_feedback. + bool ext_custom_border_color{}; ///< Support for VK_EXT_custom_border_color. + bool ext_extended_dynamic_state{}; ///< Support for VK_EXT_extended_dynamic_state. + bool ext_line_rasterization{}; ///< Support for VK_EXT_line_rasterization. + bool ext_vertex_input_dynamic_state{}; ///< Support for VK_EXT_vertex_input_dynamic_state. + bool ext_shader_stencil_export{}; ///< Support for VK_EXT_shader_stencil_export. + bool ext_shader_atomic_int64{}; ///< Support for VK_KHR_shader_atomic_int64. + bool ext_conservative_rasterization{}; ///< Support for VK_EXT_conservative_rasterization. + bool ext_provoking_vertex{}; ///< Support for VK_EXT_provoking_vertex. + bool nv_device_diagnostics_config{}; ///< Support for VK_NV_device_diagnostics_config. + bool has_renderdoc{}; ///< Has RenderDoc attached + bool has_nsight_graphics{}; ///< Has Nsight Graphics attached // Telemetry parameters - std::string vendor_name; ///< Device's driver name. - std::vector<std::string> reported_extensions; ///< Reported Vulkan extensions. + std::string vendor_name; ///< Device's driver name. + std::vector<std::string> supported_extensions; ///< Reported Vulkan extensions. /// Format properties dictionary. std::unordered_map<VkFormat, VkFormatProperties> format_properties; diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index 2aa0ffbe6..bbf0fccae 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -103,6 +103,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdFillBuffer); X(vkCmdPipelineBarrier); X(vkCmdPushConstants); + X(vkCmdPushDescriptorSetWithTemplateKHR); X(vkCmdSetBlendConstants); X(vkCmdSetDepthBias); X(vkCmdSetDepthBounds); @@ -120,9 +121,11 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdSetDepthTestEnableEXT); X(vkCmdSetDepthWriteEnableEXT); X(vkCmdSetFrontFaceEXT); + X(vkCmdSetLineWidth); X(vkCmdSetPrimitiveTopologyEXT); X(vkCmdSetStencilOpEXT); X(vkCmdSetStencilTestEnableEXT); + X(vkCmdSetVertexInputEXT); X(vkCmdResolveImage); X(vkCreateBuffer); X(vkCreateBufferView); @@ -311,8 +314,6 @@ const char* ToString(VkResult result) noexcept { return "VK_ERROR_FULL_SCREEN_EXCLUSIVE_MODE_LOST_EXT"; case VkResult::VK_ERROR_UNKNOWN: return "VK_ERROR_UNKNOWN"; - case VkResult::VK_ERROR_INCOMPATIBLE_VERSION_KHR: - return "VK_ERROR_INCOMPATIBLE_VERSION_KHR"; case VkResult::VK_THREAD_IDLE_KHR: return "VK_THREAD_IDLE_KHR"; case VkResult::VK_THREAD_DONE_KHR: diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index 3e36d356a..d76bb4324 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -193,15 +193,16 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkBeginCommandBuffer vkBeginCommandBuffer{}; PFN_vkBindBufferMemory vkBindBufferMemory{}; PFN_vkBindImageMemory vkBindImageMemory{}; + PFN_vkCmdBeginDebugUtilsLabelEXT vkCmdBeginDebugUtilsLabelEXT{}; PFN_vkCmdBeginQuery vkCmdBeginQuery{}; PFN_vkCmdBeginRenderPass vkCmdBeginRenderPass{}; PFN_vkCmdBeginTransformFeedbackEXT vkCmdBeginTransformFeedbackEXT{}; - PFN_vkCmdBeginDebugUtilsLabelEXT vkCmdBeginDebugUtilsLabelEXT{}; PFN_vkCmdBindDescriptorSets vkCmdBindDescriptorSets{}; PFN_vkCmdBindIndexBuffer vkCmdBindIndexBuffer{}; PFN_vkCmdBindPipeline vkCmdBindPipeline{}; PFN_vkCmdBindTransformFeedbackBuffersEXT vkCmdBindTransformFeedbackBuffersEXT{}; PFN_vkCmdBindVertexBuffers vkCmdBindVertexBuffers{}; + PFN_vkCmdBindVertexBuffers2EXT vkCmdBindVertexBuffers2EXT{}; PFN_vkCmdBlitImage vkCmdBlitImage{}; PFN_vkCmdClearAttachments vkCmdClearAttachments{}; PFN_vkCmdCopyBuffer vkCmdCopyBuffer{}; @@ -211,34 +212,36 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdDispatch vkCmdDispatch{}; PFN_vkCmdDraw vkCmdDraw{}; PFN_vkCmdDrawIndexed vkCmdDrawIndexed{}; + PFN_vkCmdEndDebugUtilsLabelEXT vkCmdEndDebugUtilsLabelEXT{}; PFN_vkCmdEndQuery vkCmdEndQuery{}; PFN_vkCmdEndRenderPass vkCmdEndRenderPass{}; PFN_vkCmdEndTransformFeedbackEXT vkCmdEndTransformFeedbackEXT{}; - PFN_vkCmdEndDebugUtilsLabelEXT vkCmdEndDebugUtilsLabelEXT{}; PFN_vkCmdFillBuffer vkCmdFillBuffer{}; PFN_vkCmdPipelineBarrier vkCmdPipelineBarrier{}; PFN_vkCmdPushConstants vkCmdPushConstants{}; + PFN_vkCmdPushDescriptorSetWithTemplateKHR vkCmdPushDescriptorSetWithTemplateKHR{}; + PFN_vkCmdResolveImage vkCmdResolveImage{}; PFN_vkCmdSetBlendConstants vkCmdSetBlendConstants{}; + PFN_vkCmdSetCullModeEXT vkCmdSetCullModeEXT{}; PFN_vkCmdSetDepthBias vkCmdSetDepthBias{}; PFN_vkCmdSetDepthBounds vkCmdSetDepthBounds{}; - PFN_vkCmdSetEvent vkCmdSetEvent{}; - PFN_vkCmdSetScissor vkCmdSetScissor{}; - PFN_vkCmdSetStencilCompareMask vkCmdSetStencilCompareMask{}; - PFN_vkCmdSetStencilReference vkCmdSetStencilReference{}; - PFN_vkCmdSetStencilWriteMask vkCmdSetStencilWriteMask{}; - PFN_vkCmdSetViewport vkCmdSetViewport{}; - PFN_vkCmdWaitEvents vkCmdWaitEvents{}; - PFN_vkCmdBindVertexBuffers2EXT vkCmdBindVertexBuffers2EXT{}; - PFN_vkCmdSetCullModeEXT vkCmdSetCullModeEXT{}; PFN_vkCmdSetDepthBoundsTestEnableEXT vkCmdSetDepthBoundsTestEnableEXT{}; PFN_vkCmdSetDepthCompareOpEXT vkCmdSetDepthCompareOpEXT{}; PFN_vkCmdSetDepthTestEnableEXT vkCmdSetDepthTestEnableEXT{}; PFN_vkCmdSetDepthWriteEnableEXT vkCmdSetDepthWriteEnableEXT{}; + PFN_vkCmdSetEvent vkCmdSetEvent{}; PFN_vkCmdSetFrontFaceEXT vkCmdSetFrontFaceEXT{}; + PFN_vkCmdSetLineWidth vkCmdSetLineWidth{}; PFN_vkCmdSetPrimitiveTopologyEXT vkCmdSetPrimitiveTopologyEXT{}; + PFN_vkCmdSetScissor vkCmdSetScissor{}; + PFN_vkCmdSetStencilCompareMask vkCmdSetStencilCompareMask{}; PFN_vkCmdSetStencilOpEXT vkCmdSetStencilOpEXT{}; + PFN_vkCmdSetStencilReference vkCmdSetStencilReference{}; PFN_vkCmdSetStencilTestEnableEXT vkCmdSetStencilTestEnableEXT{}; - PFN_vkCmdResolveImage vkCmdResolveImage{}; + PFN_vkCmdSetStencilWriteMask vkCmdSetStencilWriteMask{}; + PFN_vkCmdSetVertexInputEXT vkCmdSetVertexInputEXT{}; + PFN_vkCmdSetViewport vkCmdSetViewport{}; + PFN_vkCmdWaitEvents vkCmdWaitEvents{}; PFN_vkCreateBuffer vkCreateBuffer{}; PFN_vkCreateBufferView vkCreateBufferView{}; PFN_vkCreateCommandPool vkCreateCommandPool{}; @@ -989,6 +992,12 @@ public: dynamic_offsets.size(), dynamic_offsets.data()); } + void PushDescriptorSetWithTemplateKHR(VkDescriptorUpdateTemplateKHR update_template, + VkPipelineLayout layout, u32 set, + const void* data) const noexcept { + dld->vkCmdPushDescriptorSetWithTemplateKHR(handle, update_template, layout, set, data); + } + void BindPipeline(VkPipelineBindPoint bind_point, VkPipeline pipeline) const noexcept { dld->vkCmdBindPipeline(handle, bind_point, pipeline); } @@ -1190,6 +1199,10 @@ public: dld->vkCmdSetFrontFaceEXT(handle, front_face); } + void SetLineWidth(float line_width) const noexcept { + dld->vkCmdSetLineWidth(handle, line_width); + } + void SetPrimitiveTopologyEXT(VkPrimitiveTopology primitive_topology) const noexcept { dld->vkCmdSetPrimitiveTopologyEXT(handle, primitive_topology); } @@ -1203,6 +1216,13 @@ public: dld->vkCmdSetStencilTestEnableEXT(handle, enable ? VK_TRUE : VK_FALSE); } + void SetVertexInputEXT( + vk::Span<VkVertexInputBindingDescription2EXT> bindings, + vk::Span<VkVertexInputAttributeDescription2EXT> attributes) const noexcept { + dld->vkCmdSetVertexInputEXT(handle, bindings.size(), bindings.data(), attributes.size(), + attributes.data()); + } + void BindTransformFeedbackBuffersEXT(u32 first, u32 count, const VkBuffer* buffers, const VkDeviceSize* offsets, const VkDeviceSize* sizes) const noexcept { |