diff options
Diffstat (limited to 'src/video_core/renderer_opengl')
30 files changed, 833 insertions, 930 deletions
diff --git a/src/video_core/renderer_opengl/gl_arb_decompiler.cpp b/src/video_core/renderer_opengl/gl_arb_decompiler.cpp index eb5158407..b7e9ed2e9 100644 --- a/src/video_core/renderer_opengl/gl_arb_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_arb_decompiler.cpp @@ -185,10 +185,6 @@ std::string TextureType(const MetaTexture& meta) { return type; } -std::string GlobalMemoryName(const GlobalMemoryBase& base) { - return fmt::format("gmem{}_{}", base.cbuf_index, base.cbuf_offset); -} - class ARBDecompiler final { public: explicit ARBDecompiler(const Device& device, const ShaderIR& ir, const Registry& registry, @@ -199,6 +195,8 @@ public: } private: + void DefineGlobalMemory(); + void DeclareHeader(); void DeclareVertex(); void DeclareGeometry(); @@ -228,6 +226,7 @@ private: std::pair<std::string, std::size_t> BuildCoords(Operation); std::string BuildAoffi(Operation); + std::string GlobalMemoryPointer(const GmemNode& gmem); void Exit(); std::string Assign(Operation); @@ -378,10 +377,8 @@ private: std::string address; std::string_view opname; if (const auto gmem = std::get_if<GmemNode>(&*operation[0])) { - AddLine("SUB.U {}, {}, {};", temporary, Visit(gmem->GetRealAddress()), - Visit(gmem->GetBaseAddress())); - address = fmt::format("{}[{}]", GlobalMemoryName(gmem->GetDescriptor()), temporary); - opname = "ATOMB"; + address = GlobalMemoryPointer(*gmem); + opname = "ATOM"; } else if (const auto smem = std::get_if<SmemNode>(&*operation[0])) { address = fmt::format("shared_mem[{}]", Visit(smem->GetAddress())); opname = "ATOMS"; @@ -456,9 +453,13 @@ private: shader_source += '\n'; } - std::string AllocTemporary() { - max_temporaries = std::max(max_temporaries, num_temporaries + 1); - return fmt::format("T{}.x", num_temporaries++); + 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() { @@ -466,8 +467,13 @@ private: 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; @@ -478,6 +484,11 @@ private: 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"; @@ -784,6 +795,8 @@ private: 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];"); @@ -829,12 +842,20 @@ std::string_view HeaderStageName(ShaderType stage) { } } +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_storage_buffer;"); AddLine("OPTION NV_shader_thread_group;"); if (ir.UsesWarps() && device.HasWarpIntrinsics()) { AddLine("OPTION NV_shader_thread_shuffle;"); @@ -892,11 +913,19 @@ void ARBDecompiler::DeclareCompute() { 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) { - const u32 size_in_bytes = info.shared_memory_size_in_words * 4; - AddLine("SHARED_MEMORY {};", size_in_bytes); - AddLine("SHARED shared_mem[] = {{program.sharedmem}};"); + 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() { @@ -951,11 +980,10 @@ void ARBDecompiler::DeclareLocalMemory() { } void ARBDecompiler::DeclareGlobalMemory() { - u32 binding = 0; // device.GetBaseBindings(stage).shader_storage_buffer; - for (const auto& pair : ir.GetGlobalMemory()) { - const auto& base = pair.first; - AddLine("STORAGE {}[] = {{ program.storage[{}] }};", GlobalMemoryName(base), binding); - ++binding; + const std::size_t num_entries = ir.GetGlobalMemory().size(); + if (num_entries > 0) { + const std::size_t num_vectors = Common::AlignUp(num_entries, 2) / 2; + AddLine("PARAM c[{}] = {{ program.local[0..{}] }};", num_vectors, num_vectors - 1); } } @@ -977,6 +1005,9 @@ 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() { @@ -1260,13 +1291,6 @@ std::string ARBDecompiler::Visit(const Node& node) { return "{0, 0, 0, 0}.x"; } - const auto buffer_index = [this, &abuf]() -> std::string { - if (stage != ShaderType::Geometry) { - return ""; - } - return fmt::format("[{}]", Visit(abuf->GetBuffer())); - }; - const Attribute::Index index = abuf->GetIndex(); const u32 element = abuf->GetElement(); const char swizzle = Swizzle(element); @@ -1339,10 +1363,7 @@ std::string ARBDecompiler::Visit(const Node& node) { if (const auto gmem = std::get_if<GmemNode>(&*node)) { std::string temporary = AllocTemporary(); - AddLine("SUB.U {}, {}, {};", temporary, Visit(gmem->GetRealAddress()), - Visit(gmem->GetBaseAddress())); - AddLine("LDB.U32 {}, {}[{}];", temporary, GlobalMemoryName(gmem->GetDescriptor()), - temporary); + AddLine("LOAD.U32 {}, {};", temporary, GlobalMemoryPointer(*gmem)); return temporary; } @@ -1375,7 +1396,7 @@ std::string ARBDecompiler::Visit(const Node& node) { return {}; } - if (const auto cmt = std::get_if<CommentNode>(&*node)) { + if ([[maybe_unused]] const auto cmt = std::get_if<CommentNode>(&*node)) { // Uncommenting this will generate invalid code. GLASM lacks comments. // AddLine("// {}", cmt->GetText()); return {}; @@ -1419,6 +1440,22 @@ std::string ARBDecompiler::BuildAoffi(Operation operation) { return fmt::format(", offset({})", temporary); } +std::string ARBDecompiler::GlobalMemoryPointer(const GmemNode& gmem) { + const u32 binding = global_memory_names.at(gmem.GetDescriptor()); + const char result_swizzle = binding % 2 == 0 ? 'x' : 'y'; + + const std::string pointer = AllocLongVectorTemporary(); + std::string temporary = AllocTemporary(); + + const u32 local_index = binding / 2; + AddLine("PK64.U {}, c[{}];", pointer, local_index); + AddLine("SUB.U {}, {}, {};", temporary, Visit(gmem.GetRealAddress()), + Visit(gmem.GetBaseAddress())); + AddLine("CVT.U64.U32 {}.z, {};", pointer, temporary); + AddLine("ADD.U64 {}.x, {}.{}, {}.z;", pointer, pointer, result_swizzle, pointer); + return fmt::format("{}.x", pointer); +} + void ARBDecompiler::Exit() { if (stage != ShaderType::Fragment) { AddLine("RET;"); @@ -1515,11 +1552,7 @@ std::string ARBDecompiler::Assign(Operation operation) { ResetTemporaries(); return {}; } else if (const auto gmem = std::get_if<GmemNode>(&*dest)) { - const std::string temporary = AllocTemporary(); - AddLine("SUB.U {}, {}, {};", temporary, Visit(gmem->GetRealAddress()), - Visit(gmem->GetBaseAddress())); - AddLine("STB.U32 {}, {}[{}];", Visit(src), GlobalMemoryName(gmem->GetDescriptor()), - temporary); + AddLine("STORE.U32 {}, {};", Visit(src), GlobalMemoryPointer(*gmem)); ResetTemporaries(); return {}; } else { @@ -1671,7 +1704,7 @@ std::string ARBDecompiler::HCastFloat(Operation operation) { } std::string ARBDecompiler::HUnpack(Operation operation) { - const std::string operand = Visit(operation[0]); + std::string operand = Visit(operation[0]); switch (std::get<Tegra::Shader::HalfType>(operation.GetMeta())) { case Tegra::Shader::HalfType::H0_H1: return operand; @@ -2021,7 +2054,7 @@ std::string ARBDecompiler::InvocationId(Operation) { std::string ARBDecompiler::YNegate(Operation) { LOG_WARNING(Render_OpenGL, "(STUBBED)"); - const std::string temporary = AllocTemporary(); + std::string temporary = AllocTemporary(); AddLine("MOV.F {}, 1;", temporary); return temporary; } @@ -2044,10 +2077,6 @@ std::string ARBDecompiler::ShuffleIndexed(Operation operation) { } std::string ARBDecompiler::Barrier(Operation) { - if (!ir.IsDecompiled()) { - LOG_ERROR(Render_OpenGL, "BAR used but shader is not decompiled"); - return {}; - } AddLine("BAR;"); return {}; } diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp index e461e4c70..b1c4cd62f 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -26,7 +26,7 @@ Buffer::Buffer(const Device& device, VAddr cpu_addr, std::size_t size) : VideoCommon::BufferBlock{cpu_addr, size} { gl_buffer.Create(); glNamedBufferData(gl_buffer.handle, static_cast<GLsizeiptr>(size), nullptr, GL_DYNAMIC_DRAW); - if (device.HasVertexBufferUnifiedMemory()) { + if (device.UseAssemblyShaders() || device.HasVertexBufferUnifiedMemory()) { glMakeNamedBufferResidentNV(gl_buffer.handle, GL_READ_WRITE); glGetNamedBufferParameterui64vNV(gl_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, &gpu_address); } @@ -59,9 +59,10 @@ void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst static_cast<GLintptr>(dst_offset), static_cast<GLsizeiptr>(size)); } -OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system, +OGLBufferCache::OGLBufferCache(VideoCore::RasterizerInterface& rasterizer, + Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory, const Device& device_, std::size_t stream_size) - : GenericBufferCache{rasterizer, system, + : GenericBufferCache{rasterizer, gpu_memory, cpu_memory, std::make_unique<OGLStreamBuffer>(device_, stream_size, true)}, device{device_} { if (!device.HasFastBufferSubData()) { diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h index 88fdc0536..f75b32e31 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.h +++ b/src/video_core/renderer_opengl/gl_buffer_cache.h @@ -52,7 +52,8 @@ private: using GenericBufferCache = VideoCommon::BufferCache<Buffer, GLuint, OGLStreamBuffer>; class OGLBufferCache final : public GenericBufferCache { public: - explicit OGLBufferCache(RasterizerOpenGL& rasterizer, Core::System& system, + explicit OGLBufferCache(VideoCore::RasterizerInterface& rasterizer, + Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory, const Device& device, std::size_t stream_size); ~OGLBufferCache(); diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp index c1f20f0ab..a94e4f72e 100644 --- a/src/video_core/renderer_opengl/gl_device.cpp +++ b/src/video_core/renderer_opengl/gl_device.cpp @@ -193,7 +193,6 @@ bool IsASTCSupported() { Device::Device() : max_uniform_buffers{BuildMaxUniformBuffers()}, base_bindings{BuildBaseBindings()} { const std::string_view vendor = reinterpret_cast<const char*>(glGetString(GL_VENDOR)); - const std::string_view renderer = reinterpret_cast<const char*>(glGetString(GL_RENDERER)); const std::string_view version = reinterpret_cast<const char*>(glGetString(GL_VERSION)); const std::vector extensions = GetExtensions(); @@ -212,6 +211,7 @@ Device::Device() shader_storage_alignment = GetInteger<std::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); 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; @@ -233,6 +233,8 @@ Device::Device() GLAD_GL_NV_gpu_program5 && GLAD_GL_NV_compute_program5 && GLAD_GL_NV_transform_feedback && GLAD_GL_NV_transform_feedback2; + use_asynchronous_shaders = Settings::values.use_asynchronous_shaders.GetValue(); + LOG_INFO(Render_OpenGL, "Renderer_VariableAOFFI: {}", has_variable_aoffi); LOG_INFO(Render_OpenGL, "Renderer_ComponentIndexingBug: {}", has_component_indexing_bug); LOG_INFO(Render_OpenGL, "Renderer_PreciseBug: {}", has_precise_bug); @@ -248,6 +250,7 @@ Device::Device(std::nullptr_t) { 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; diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h index e1d811966..8a4b6b9fc 100644 --- a/src/video_core/renderer_opengl/gl_device.h +++ b/src/video_core/renderer_opengl/gl_device.h @@ -52,6 +52,10 @@ public: return max_varyings; } + u32 GetMaxComputeSharedMemorySize() const { + return max_compute_shared_memory_size; + } + bool HasWarpIntrinsics() const { return has_warp_intrinsics; } @@ -104,6 +108,10 @@ public: return use_assembly_shaders; } + bool UseAsynchronousShaders() const { + return use_asynchronous_shaders; + } + private: static bool TestVariableAoffi(); static bool TestPreciseBug(); @@ -114,6 +122,7 @@ private: std::size_t shader_storage_alignment{}; u32 max_vertex_attributes{}; u32 max_varyings{}; + u32 max_compute_shared_memory_size{}; bool has_warp_intrinsics{}; bool has_shader_ballot{}; bool has_vertex_viewport_layer{}; @@ -127,6 +136,7 @@ private: bool has_fast_buffer_sub_data{}; bool has_nv_viewport_array2{}; bool use_assembly_shaders{}; + bool use_asynchronous_shaders{}; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_fence_manager.cpp b/src/video_core/renderer_opengl/gl_fence_manager.cpp index ec5421afa..b532fdcc2 100644 --- a/src/video_core/renderer_opengl/gl_fence_manager.cpp +++ b/src/video_core/renderer_opengl/gl_fence_manager.cpp @@ -4,16 +4,17 @@ #include "common/assert.h" +#include <glad/glad.h> + #include "video_core/renderer_opengl/gl_buffer_cache.h" #include "video_core/renderer_opengl/gl_fence_manager.h" namespace OpenGL { -GLInnerFence::GLInnerFence(u32 payload, bool is_stubbed) - : VideoCommon::FenceBase(payload, is_stubbed), sync_object{} {} +GLInnerFence::GLInnerFence(u32 payload, bool is_stubbed) : FenceBase(payload, is_stubbed) {} GLInnerFence::GLInnerFence(GPUVAddr address, u32 payload, bool is_stubbed) - : VideoCommon::FenceBase(address, payload, is_stubbed), sync_object{} {} + : FenceBase(address, payload, is_stubbed) {} GLInnerFence::~GLInnerFence() = default; @@ -44,11 +45,10 @@ void GLInnerFence::Wait() { glClientWaitSync(sync_object.handle, 0, GL_TIMEOUT_IGNORED); } -FenceManagerOpenGL::FenceManagerOpenGL(Core::System& system, - VideoCore::RasterizerInterface& rasterizer, +FenceManagerOpenGL::FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer, Tegra::GPU& gpu, TextureCacheOpenGL& texture_cache, OGLBufferCache& buffer_cache, QueryCache& query_cache) - : GenericFenceManager(system, rasterizer, texture_cache, buffer_cache, query_cache) {} + : GenericFenceManager{rasterizer, gpu, texture_cache, buffer_cache, query_cache} {} Fence FenceManagerOpenGL::CreateFence(u32 value, bool is_stubbed) { return std::make_shared<GLInnerFence>(value, is_stubbed); diff --git a/src/video_core/renderer_opengl/gl_fence_manager.h b/src/video_core/renderer_opengl/gl_fence_manager.h index c917b3343..da1dcdace 100644 --- a/src/video_core/renderer_opengl/gl_fence_manager.h +++ b/src/video_core/renderer_opengl/gl_fence_manager.h @@ -5,7 +5,6 @@ #pragma once #include <memory> -#include <glad/glad.h> #include "common/common_types.h" #include "video_core/fence_manager.h" @@ -38,9 +37,9 @@ using GenericFenceManager = class FenceManagerOpenGL final : public GenericFenceManager { public: - FenceManagerOpenGL(Core::System& system, VideoCore::RasterizerInterface& rasterizer, - TextureCacheOpenGL& texture_cache, OGLBufferCache& buffer_cache, - QueryCache& query_cache); + explicit FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer, Tegra::GPU& gpu, + TextureCacheOpenGL& texture_cache, OGLBufferCache& buffer_cache, + QueryCache& query_cache); protected: Fence CreateFence(u32 value, bool is_stubbed) override; diff --git a/src/video_core/renderer_opengl/gl_query_cache.cpp b/src/video_core/renderer_opengl/gl_query_cache.cpp index d7ba57aca..1a3d9720e 100644 --- a/src/video_core/renderer_opengl/gl_query_cache.cpp +++ b/src/video_core/renderer_opengl/gl_query_cache.cpp @@ -30,12 +30,11 @@ constexpr GLenum GetTarget(VideoCore::QueryType type) { } // Anonymous namespace -QueryCache::QueryCache(Core::System& system, RasterizerOpenGL& gl_rasterizer) - : VideoCommon::QueryCacheBase< - QueryCache, CachedQuery, CounterStream, HostCounter, - std::vector<OGLQuery>>{system, - static_cast<VideoCore::RasterizerInterface&>(gl_rasterizer)}, - gl_rasterizer{gl_rasterizer} {} +QueryCache::QueryCache(RasterizerOpenGL& rasterizer, Tegra::Engines::Maxwell3D& maxwell3d, + Tegra::MemoryManager& gpu_memory) + : VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter>( + rasterizer, maxwell3d, gpu_memory), + gl_rasterizer{rasterizer} {} QueryCache::~QueryCache() = default; @@ -90,6 +89,8 @@ u64 HostCounter::BlockingQuery() const { CachedQuery::CachedQuery(QueryCache& cache, VideoCore::QueryType type, VAddr cpu_addr, u8* host_ptr) : VideoCommon::CachedQueryBase<HostCounter>{cpu_addr, host_ptr}, cache{&cache}, type{type} {} +CachedQuery::~CachedQuery() = default; + CachedQuery::CachedQuery(CachedQuery&& rhs) noexcept : VideoCommon::CachedQueryBase<HostCounter>(std::move(rhs)), cache{rhs.cache}, type{rhs.type} {} diff --git a/src/video_core/renderer_opengl/gl_query_cache.h b/src/video_core/renderer_opengl/gl_query_cache.h index d8e7052a1..82cac51ee 100644 --- a/src/video_core/renderer_opengl/gl_query_cache.h +++ b/src/video_core/renderer_opengl/gl_query_cache.h @@ -26,10 +26,11 @@ class RasterizerOpenGL; using CounterStream = VideoCommon::CounterStreamBase<QueryCache, HostCounter>; -class QueryCache final : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, - HostCounter, std::vector<OGLQuery>> { +class QueryCache final + : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter> { public: - explicit QueryCache(Core::System& system, RasterizerOpenGL& rasterizer); + explicit QueryCache(RasterizerOpenGL& rasterizer, Tegra::Engines::Maxwell3D& maxwell3d, + Tegra::MemoryManager& gpu_memory); ~QueryCache(); OGLQuery AllocateQuery(VideoCore::QueryType type); @@ -40,6 +41,7 @@ public: private: RasterizerOpenGL& gl_rasterizer; + std::array<std::vector<OGLQuery>, VideoCore::NumQueryTypes> query_pools; }; class HostCounter final : public VideoCommon::HostCounterBase<QueryCache, HostCounter> { @@ -62,10 +64,12 @@ class CachedQuery final : public VideoCommon::CachedQueryBase<HostCounter> { public: explicit CachedQuery(QueryCache& cache, VideoCore::QueryType type, VAddr cpu_addr, u8* host_ptr); - CachedQuery(CachedQuery&& rhs) noexcept; - CachedQuery(const CachedQuery&) = delete; + ~CachedQuery() override; + CachedQuery(CachedQuery&& rhs) noexcept; CachedQuery& operator=(CachedQuery&& rhs) noexcept; + + CachedQuery(const CachedQuery&) = delete; CachedQuery& operator=(const CachedQuery&) = delete; void Flush() override; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index e960a0ef1..bbb2eb17c 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -139,17 +139,33 @@ void oglEnable(GLenum cap, bool state) { (state ? glEnable : glDisable)(cap); } +void UpdateBindlessPointers(GLenum target, GLuint64EXT* pointers, std::size_t num_entries) { + if (num_entries == 0) { + return; + } + if (num_entries % 2 == 1) { + pointers[num_entries] = 0; + } + const GLsizei num_vectors = static_cast<GLsizei>((num_entries + 1) / 2); + glProgramLocalParametersI4uivNV(target, 0, num_vectors, + reinterpret_cast<const GLuint*>(pointers)); +} + } // Anonymous namespace -RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWindow& emu_window, - const Device& device, ScreenInfo& info, - ProgramManager& program_manager, StateTracker& state_tracker) - : RasterizerAccelerated{system.Memory()}, device{device}, texture_cache{system, *this, device, - state_tracker}, - shader_cache{*this, system, emu_window, device}, query_cache{system, *this}, - buffer_cache{*this, system, device, STREAM_BUFFER_SIZE}, - fence_manager{system, *this, texture_cache, buffer_cache, query_cache}, system{system}, - screen_info{info}, program_manager{program_manager}, state_tracker{state_tracker} { +RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window, Tegra::GPU& gpu_, + Core::Memory::Memory& cpu_memory, const Device& device_, + ScreenInfo& screen_info_, ProgramManager& program_manager_, + StateTracker& state_tracker_) + : RasterizerAccelerated{cpu_memory}, gpu(gpu_), maxwell3d(gpu.Maxwell3D()), + kepler_compute(gpu.KeplerCompute()), gpu_memory(gpu.MemoryManager()), device(device_), + screen_info(screen_info_), program_manager(program_manager_), state_tracker(state_tracker_), + texture_cache(*this, maxwell3d, gpu_memory, device, state_tracker), + shader_cache(*this, emu_window, gpu, maxwell3d, kepler_compute, gpu_memory, device), + query_cache(*this, maxwell3d, gpu_memory), + buffer_cache(*this, gpu_memory, cpu_memory, device, STREAM_BUFFER_SIZE), + fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache), + async_shaders(emu_window) { CheckExtensions(); unified_uniform_buffer.Create(); @@ -162,6 +178,10 @@ RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWind nullptr, 0); } } + + if (device.UseAsynchronousShaders()) { + async_shaders.AllocateWorkers(); + } } RasterizerOpenGL::~RasterizerOpenGL() { @@ -179,8 +199,7 @@ void RasterizerOpenGL::CheckExtensions() { } void RasterizerOpenGL::SetupVertexFormat() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::VertexFormats]) { return; } @@ -200,7 +219,7 @@ void RasterizerOpenGL::SetupVertexFormat() { } flags[Dirty::VertexFormat0 + index] = false; - const auto attrib = gpu.regs.vertex_attrib_format[index]; + const auto attrib = maxwell3d.regs.vertex_attrib_format[index]; const auto gl_index = static_cast<GLuint>(index); // Disable constant attributes. @@ -224,8 +243,7 @@ void RasterizerOpenGL::SetupVertexFormat() { } void RasterizerOpenGL::SetupVertexBuffer() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::VertexBuffers]) { return; } @@ -236,7 +254,7 @@ void RasterizerOpenGL::SetupVertexBuffer() { const bool use_unified_memory = device.HasVertexBufferUnifiedMemory(); // Upload all guest vertex arrays sequentially to our buffer - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; for (std::size_t index = 0; index < NUM_SUPPORTED_VERTEX_BINDINGS; ++index) { if (!flags[Dirty::VertexBuffer0 + index]) { continue; @@ -273,14 +291,13 @@ void RasterizerOpenGL::SetupVertexBuffer() { } void RasterizerOpenGL::SetupVertexInstances() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::VertexInstances]) { return; } flags[Dirty::VertexInstances] = false; - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; for (std::size_t index = 0; index < NUM_SUPPORTED_VERTEX_ATTRIBUTES; ++index) { if (!flags[Dirty::VertexInstance0 + index]) { continue; @@ -296,7 +313,7 @@ void RasterizerOpenGL::SetupVertexInstances() { GLintptr RasterizerOpenGL::SetupIndexBuffer() { MICROPROFILE_SCOPE(OpenGL_Index); - const auto& regs = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; const std::size_t size = CalculateIndexBufferSize(); const auto info = buffer_cache.UploadMemory(regs.index_array.IndexStart(), size); glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, info.handle); @@ -305,16 +322,14 @@ GLintptr RasterizerOpenGL::SetupIndexBuffer() { void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { MICROPROFILE_SCOPE(OpenGL_Shader); - auto& gpu = system.GPU().Maxwell3D(); - std::size_t num_ssbos = 0; u32 clip_distances = 0; for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { - const auto& shader_config = gpu.regs.shader_config[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 (!gpu.regs.IsShaderConfigEnabled(index)) { + if (!maxwell3d.regs.IsShaderConfigEnabled(index)) { switch (program) { case Maxwell::ShaderProgram::Geometry: program_manager.UseGeometryShader(0); @@ -329,31 +344,15 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { } // Currently this stages are not supported in the OpenGL backend. - // Todo(Blinkhawk): Port tesselation shaders from Vulkan to OpenGL - if (program == Maxwell::ShaderProgram::TesselationControl) { - continue; - } else if (program == Maxwell::ShaderProgram::TesselationEval) { + // 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); - - if (device.UseAssemblyShaders()) { - // Check for ARB limitation. We only have 16 SSBOs per context state. To workaround this - // all stages share the same bindings. - const std::size_t num_stage_ssbos = shader->GetEntries().global_memory_entries.size(); - ASSERT_MSG(num_stage_ssbos == 0 || num_ssbos == 0, "SSBOs on more than one stage"); - num_ssbos += num_stage_ssbos; - } - - // Stage indices are 0 - 5 - const std::size_t stage = index == 0 ? 0 : index - 1; - SetupDrawConstBuffers(stage, shader); - SetupDrawGlobalMemory(stage, shader); - SetupDrawTextures(stage, shader); - SetupDrawImages(stage, shader); + Shader* const shader = shader_cache.GetStageProgram(program, async_shaders); - const GLuint program_handle = shader->GetHandle(); + const GLuint program_handle = shader->IsBuilt() ? shader->GetHandle() : 0; switch (program) { case Maxwell::ShaderProgram::VertexA: case Maxwell::ShaderProgram::VertexB: @@ -370,6 +369,13 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { shader_config.enable.Value(), shader_config.offset); } + // Stage indices are 0 - 5 + const std::size_t stage = index == 0 ? 0 : index - 1; + SetupDrawConstBuffers(stage, shader); + SetupDrawGlobalMemory(stage, shader); + SetupDrawTextures(stage, shader); + SetupDrawImages(stage, shader); + // 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 @@ -384,11 +390,11 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { } SyncClipEnabled(clip_distances); - gpu.dirty.flags[Dirty::Shaders] = false; + maxwell3d.dirty.flags[Dirty::Shaders] = false; } std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const { - const auto& regs = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; std::size_t size = 0; for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) { @@ -406,34 +412,27 @@ std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const { } std::size_t RasterizerOpenGL::CalculateIndexBufferSize() const { - const auto& regs = system.GPU().Maxwell3D().regs; - - return static_cast<std::size_t>(regs.index_array.count) * - static_cast<std::size_t>(regs.index_array.FormatSizeInBytes()); + return static_cast<std::size_t>(maxwell3d.regs.index_array.count) * + static_cast<std::size_t>(maxwell3d.regs.index_array.FormatSizeInBytes()); } -void RasterizerOpenGL::LoadDiskResources(const std::atomic_bool& stop_loading, +void RasterizerOpenGL::LoadDiskResources(u64 title_id, const std::atomic_bool& stop_loading, const VideoCore::DiskResourceLoadCallback& callback) { - shader_cache.LoadDiskCache(stop_loading, callback); -} - -void RasterizerOpenGL::SetupDirtyFlags() { - state_tracker.Initialize(); + shader_cache.LoadDiskCache(title_id, stop_loading, callback); } void RasterizerOpenGL::ConfigureFramebuffers() { MICROPROFILE_SCOPE(OpenGL_Framebuffer); - auto& gpu = system.GPU().Maxwell3D(); - if (!gpu.dirty.flags[VideoCommon::Dirty::RenderTargets]) { + if (!maxwell3d.dirty.flags[VideoCommon::Dirty::RenderTargets]) { return; } - gpu.dirty.flags[VideoCommon::Dirty::RenderTargets] = false; + maxwell3d.dirty.flags[VideoCommon::Dirty::RenderTargets] = false; texture_cache.GuardRenderTargets(true); View depth_surface = texture_cache.GetDepthBufferSurface(true); - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; UNIMPLEMENTED_IF(regs.rt_separate_frag_data == 0); // Bind the framebuffer surfaces @@ -465,8 +464,7 @@ void RasterizerOpenGL::ConfigureFramebuffers() { } void RasterizerOpenGL::ConfigureClearFramebuffer(bool using_color, bool using_depth_stencil) { - auto& gpu = system.GPU().Maxwell3D(); - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; texture_cache.GuardRenderTargets(true); View color_surface; @@ -516,12 +514,11 @@ void RasterizerOpenGL::ConfigureClearFramebuffer(bool using_color, bool using_de } void RasterizerOpenGL::Clear() { - const auto& gpu = system.GPU().Maxwell3D(); - if (!gpu.ShouldExecute()) { + if (!maxwell3d.ShouldExecute()) { return; } - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; bool use_color{}; bool use_depth{}; bool use_stencil{}; @@ -586,7 +583,6 @@ void RasterizerOpenGL::Clear() { void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { MICROPROFILE_SCOPE(OpenGL_Drawing); - auto& gpu = system.GPU().Maxwell3D(); query_cache.UpdateCounters(); @@ -634,7 +630,7 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { if (invalidated) { // When the stream buffer has been invalidated, we have to consider vertex buffers as dirty - auto& dirty = gpu.dirty.flags; + auto& dirty = maxwell3d.dirty.flags; dirty[Dirty::VertexBuffers] = true; for (int index = Dirty::VertexBuffer0; index <= Dirty::VertexBuffer31; ++index) { dirty[index] = true; @@ -655,7 +651,7 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { // Setup emulation uniform buffer. if (!device.UseAssemblyShaders()) { MaxwellUniformData ubo; - ubo.SetFromRegs(gpu); + ubo.SetFromRegs(maxwell3d); const auto info = buffer_cache.UploadHostMemory(&ubo, sizeof(ubo), device.GetUniformBufferAlignment()); glBindBufferRange(GL_UNIFORM_BUFFER, EmulationUniformBlockBinding, info.handle, info.offset, @@ -664,7 +660,7 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { // Setup shaders and their used resources. texture_cache.GuardSamplers(true); - const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(gpu.regs.draw.topology); + const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d.regs.draw.topology); SetupShaders(primitive_mode); texture_cache.GuardSamplers(false); @@ -681,14 +677,14 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { BeginTransformFeedback(primitive_mode); - const GLuint base_instance = static_cast<GLuint>(gpu.regs.vb_base_instance); + const GLuint base_instance = static_cast<GLuint>(maxwell3d.regs.vb_base_instance); const GLsizei num_instances = - static_cast<GLsizei>(is_instanced ? gpu.mme_draw.instance_count : 1); + static_cast<GLsizei>(is_instanced ? maxwell3d.mme_draw.instance_count : 1); if (is_indexed) { - const GLint base_vertex = static_cast<GLint>(gpu.regs.vb_element_base); - const GLsizei num_vertices = static_cast<GLsizei>(gpu.regs.index_array.count); + const GLint base_vertex = static_cast<GLint>(maxwell3d.regs.vb_element_base); + const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d.regs.index_array.count); const GLvoid* offset = reinterpret_cast<const GLvoid*>(index_buffer_offset); - const GLenum format = MaxwellToGL::IndexFormat(gpu.regs.index_array.format); + const GLenum format = MaxwellToGL::IndexFormat(maxwell3d.regs.index_array.format); if (num_instances == 1 && base_instance == 0 && base_vertex == 0) { glDrawElements(primitive_mode, num_vertices, format, offset); } else if (num_instances == 1 && base_instance == 0) { @@ -707,8 +703,8 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { base_instance); } } else { - const GLint base_vertex = static_cast<GLint>(gpu.regs.vertex_buffer.first); - const GLsizei num_vertices = static_cast<GLsizei>(gpu.regs.vertex_buffer.count); + const GLint base_vertex = static_cast<GLint>(maxwell3d.regs.vertex_buffer.first); + const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d.regs.vertex_buffer.count); if (num_instances == 1 && base_instance == 0) { glDrawArrays(primitive_mode, base_vertex, num_vertices); } else if (base_instance == 0) { @@ -723,7 +719,7 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { ++num_queued_commands; - system.GPU().TickWork(); + gpu.TickWork(); } void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { @@ -731,6 +727,8 @@ void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { current_cbuf = 0; auto kernel = shader_cache.GetComputeKernel(code_addr); + program_manager.BindCompute(kernel->GetHandle()); + SetupComputeTextures(kernel); SetupComputeImages(kernel); @@ -744,7 +742,7 @@ void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { buffer_cache.Unmap(); - const auto& launch_desc = system.GPU().KeplerCompute().launch_description; + const auto& launch_desc = kepler_compute.launch_description; program_manager.BindCompute(kernel->GetHandle()); glDispatchCompute(launch_desc.grid_dim_x, launch_desc.grid_dim_y, launch_desc.grid_dim_z); ++num_queued_commands; @@ -807,17 +805,14 @@ void RasterizerOpenGL::SyncGuestHost() { } void RasterizerOpenGL::SignalSemaphore(GPUVAddr addr, u32 value) { - auto& gpu{system.GPU()}; if (!gpu.IsAsync()) { - auto& memory_manager{gpu.MemoryManager()}; - memory_manager.Write<u32>(addr, value); + gpu_memory.Write<u32>(addr, value); return; } fence_manager.SignalSemaphore(addr, value); } void RasterizerOpenGL::SignalSyncPoint(u32 value) { - auto& gpu{system.GPU()}; if (!gpu.IsAsync()) { gpu.IncrementSyncPoint(value); return; @@ -826,7 +821,6 @@ void RasterizerOpenGL::SignalSyncPoint(u32 value) { } void RasterizerOpenGL::ReleaseFences() { - auto& gpu{system.GPU()}; if (!gpu.IsAsync()) { return; } @@ -912,7 +906,7 @@ void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, Shader* sh GL_FRAGMENT_PROGRAM_PARAMETER_BUFFER_NV}; MICROPROFILE_SCOPE(OpenGL_UBO); - const auto& stages = system.GPU().Maxwell3D().state.shader_stages; + const auto& stages = maxwell3d.state.shader_stages; const auto& shader_stage = stages[stage_index]; const auto& entries = shader->GetEntries(); const bool use_unified = entries.use_unified_uniforms; @@ -937,7 +931,7 @@ void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, Shader* sh void RasterizerOpenGL::SetupComputeConstBuffers(Shader* kernel) { MICROPROFILE_SCOPE(OpenGL_UBO); - const auto& launch_desc = system.GPU().KeplerCompute().launch_description; + const auto& launch_desc = kepler_compute.launch_description; const auto& entries = kernel->GetEntries(); const bool use_unified = entries.use_unified_uniforms; @@ -1005,45 +999,66 @@ void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding, } void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, Shader* shader) { - auto& gpu{system.GPU()}; - auto& memory_manager{gpu.MemoryManager()}; - const auto cbufs{gpu.Maxwell3D().state.shader_stages[stage_index]}; + static constexpr std::array TARGET_LUT = { + GL_VERTEX_PROGRAM_NV, GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV, + GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV, + }; + + const auto& cbufs{maxwell3d.state.shader_stages[stage_index]}; + const auto& entries{shader->GetEntries().global_memory_entries}; - u32 binding = - device.UseAssemblyShaders() ? 0 : device.GetBaseBindings(stage_index).shader_storage_buffer; - for (const auto& entry : shader->GetEntries().global_memory_entries) { + std::array<GLuint64EXT, 32> pointers; + ASSERT(entries.size() < pointers.size()); + + const bool assembly_shaders = device.UseAssemblyShaders(); + u32 binding = assembly_shaders ? 0 : device.GetBaseBindings(stage_index).shader_storage_buffer; + for (const auto& entry : entries) { const GPUVAddr addr{cbufs.const_buffers[entry.cbuf_index].address + entry.cbuf_offset}; - const GPUVAddr gpu_addr{memory_manager.Read<u64>(addr)}; - const u32 size{memory_manager.Read<u32>(addr + 8)}; - SetupGlobalMemory(binding++, entry, gpu_addr, size); + const GPUVAddr gpu_addr{gpu_memory.Read<u64>(addr)}; + const u32 size{gpu_memory.Read<u32>(addr + 8)}; + SetupGlobalMemory(binding, entry, gpu_addr, size, &pointers[binding]); + ++binding; + } + if (assembly_shaders) { + UpdateBindlessPointers(TARGET_LUT[stage_index], pointers.data(), entries.size()); } } void RasterizerOpenGL::SetupComputeGlobalMemory(Shader* kernel) { - auto& gpu{system.GPU()}; - auto& memory_manager{gpu.MemoryManager()}; - const auto cbufs{gpu.KeplerCompute().launch_description.const_buffer_config}; + const auto& cbufs{kepler_compute.launch_description.const_buffer_config}; + const auto& entries{kernel->GetEntries().global_memory_entries}; + + std::array<GLuint64EXT, 32> pointers; + ASSERT(entries.size() < pointers.size()); u32 binding = 0; - for (const auto& entry : kernel->GetEntries().global_memory_entries) { - const auto addr{cbufs[entry.cbuf_index].Address() + entry.cbuf_offset}; - const auto gpu_addr{memory_manager.Read<u64>(addr)}; - const auto size{memory_manager.Read<u32>(addr + 8)}; - SetupGlobalMemory(binding++, entry, gpu_addr, size); + for (const auto& entry : entries) { + const GPUVAddr addr{cbufs[entry.cbuf_index].Address() + entry.cbuf_offset}; + const GPUVAddr gpu_addr{gpu_memory.Read<u64>(addr)}; + const u32 size{gpu_memory.Read<u32>(addr + 8)}; + SetupGlobalMemory(binding, entry, gpu_addr, size, &pointers[binding]); + ++binding; + } + if (device.UseAssemblyShaders()) { + UpdateBindlessPointers(GL_COMPUTE_PROGRAM_NV, pointers.data(), entries.size()); } } void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry, - GPUVAddr gpu_addr, std::size_t size) { - const auto alignment{device.GetShaderStorageBufferAlignment()}; + GPUVAddr gpu_addr, std::size_t size, + GLuint64EXT* pointer) { + const std::size_t alignment{device.GetShaderStorageBufferAlignment()}; const auto info = buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.is_written); - glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, info.handle, info.offset, - static_cast<GLsizeiptr>(size)); + if (device.UseAssemblyShaders()) { + *pointer = info.address + info.offset; + } else { + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, info.handle, info.offset, + static_cast<GLsizeiptr>(size)); + } } void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, Shader* shader) { MICROPROFILE_SCOPE(OpenGL_Texture); - const auto& maxwell3d = system.GPU().Maxwell3D(); u32 binding = device.GetBaseBindings(stage_index).sampler; for (const auto& entry : shader->GetEntries().samplers) { const auto shader_type = static_cast<ShaderType>(stage_index); @@ -1056,11 +1071,10 @@ void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, Shader* shader void RasterizerOpenGL::SetupComputeTextures(Shader* kernel) { MICROPROFILE_SCOPE(OpenGL_Texture); - const auto& compute = system.GPU().KeplerCompute(); u32 binding = 0; for (const auto& entry : kernel->GetEntries().samplers) { for (std::size_t i = 0; i < entry.size; ++i) { - const auto texture = GetTextureInfo(compute, entry, ShaderType::Compute, i); + const auto texture = GetTextureInfo(kepler_compute, entry, ShaderType::Compute, i); SetupTexture(binding++, texture, entry); } } @@ -1084,20 +1098,18 @@ void RasterizerOpenGL::SetupTexture(u32 binding, const Tegra::Texture::FullTextu } void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, Shader* shader) { - const auto& maxwell3d = system.GPU().Maxwell3D(); u32 binding = device.GetBaseBindings(stage_index).image; for (const auto& entry : shader->GetEntries().images) { - const auto shader_type = static_cast<Tegra::Engines::ShaderType>(stage_index); + const auto shader_type = static_cast<ShaderType>(stage_index); const auto tic = GetTextureInfo(maxwell3d, entry, shader_type).tic; SetupImage(binding++, tic, entry); } } void RasterizerOpenGL::SetupComputeImages(Shader* shader) { - const auto& compute = system.GPU().KeplerCompute(); u32 binding = 0; for (const auto& entry : shader->GetEntries().images) { - const auto tic = GetTextureInfo(compute, entry, Tegra::Engines::ShaderType::Compute).tic; + const auto tic = GetTextureInfo(kepler_compute, entry, ShaderType::Compute).tic; SetupImage(binding++, tic, entry); } } @@ -1117,9 +1129,8 @@ void RasterizerOpenGL::SetupImage(u32 binding, const Tegra::Texture::TICEntry& t } void RasterizerOpenGL::SyncViewport() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; - const auto& regs = gpu.regs; + auto& flags = maxwell3d.dirty.flags; + const auto& regs = maxwell3d.regs; const bool dirty_viewport = flags[Dirty::Viewports]; const bool dirty_clip_control = flags[Dirty::ClipControl]; @@ -1191,25 +1202,23 @@ void RasterizerOpenGL::SyncViewport() { } void RasterizerOpenGL::SyncDepthClamp() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::DepthClampEnabled]) { return; } flags[Dirty::DepthClampEnabled] = false; - oglEnable(GL_DEPTH_CLAMP, gpu.regs.view_volume_clip_control.depth_clamp_disabled == 0); + oglEnable(GL_DEPTH_CLAMP, maxwell3d.regs.view_volume_clip_control.depth_clamp_disabled == 0); } void RasterizerOpenGL::SyncClipEnabled(u32 clip_mask) { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::ClipDistances] && !flags[Dirty::Shaders]) { return; } flags[Dirty::ClipDistances] = false; - clip_mask &= gpu.regs.clip_distance_enabled; + clip_mask &= maxwell3d.regs.clip_distance_enabled; if (clip_mask == last_clip_distance_mask) { return; } @@ -1225,9 +1234,8 @@ void RasterizerOpenGL::SyncClipCoef() { } void RasterizerOpenGL::SyncCullMode() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; - const auto& regs = gpu.regs; + auto& flags = maxwell3d.dirty.flags; + const auto& regs = maxwell3d.regs; if (flags[Dirty::CullTest]) { flags[Dirty::CullTest] = false; @@ -1242,26 +1250,24 @@ void RasterizerOpenGL::SyncCullMode() { } void RasterizerOpenGL::SyncPrimitiveRestart() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::PrimitiveRestart]) { return; } flags[Dirty::PrimitiveRestart] = false; - if (gpu.regs.primitive_restart.enabled) { + if (maxwell3d.regs.primitive_restart.enabled) { glEnable(GL_PRIMITIVE_RESTART); - glPrimitiveRestartIndex(gpu.regs.primitive_restart.index); + glPrimitiveRestartIndex(maxwell3d.regs.primitive_restart.index); } else { glDisable(GL_PRIMITIVE_RESTART); } } void RasterizerOpenGL::SyncDepthTestState() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; + const auto& regs = maxwell3d.regs; - const auto& regs = gpu.regs; if (flags[Dirty::DepthMask]) { flags[Dirty::DepthMask] = false; glDepthMask(regs.depth_write_enabled ? GL_TRUE : GL_FALSE); @@ -1279,14 +1285,13 @@ void RasterizerOpenGL::SyncDepthTestState() { } void RasterizerOpenGL::SyncStencilTestState() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::StencilTest]) { return; } flags[Dirty::StencilTest] = false; - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; oglEnable(GL_STENCIL_TEST, regs.stencil_enable); glStencilFuncSeparate(GL_FRONT, MaxwellToGL::ComparisonOp(regs.stencil_front_func_func), @@ -1311,25 +1316,24 @@ void RasterizerOpenGL::SyncStencilTestState() { } void RasterizerOpenGL::SyncRasterizeEnable() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::RasterizeEnable]) { return; } flags[Dirty::RasterizeEnable] = false; - oglEnable(GL_RASTERIZER_DISCARD, gpu.regs.rasterize_enable == 0); + oglEnable(GL_RASTERIZER_DISCARD, maxwell3d.regs.rasterize_enable == 0); } void RasterizerOpenGL::SyncPolygonModes() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::PolygonModes]) { return; } flags[Dirty::PolygonModes] = false; - if (gpu.regs.fill_rectangle) { + const auto& regs = maxwell3d.regs; + if (regs.fill_rectangle) { if (!GLAD_GL_NV_fill_rectangle) { LOG_ERROR(Render_OpenGL, "GL_NV_fill_rectangle used and not supported"); glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); @@ -1342,27 +1346,26 @@ void RasterizerOpenGL::SyncPolygonModes() { return; } - if (gpu.regs.polygon_mode_front == gpu.regs.polygon_mode_back) { + if (regs.polygon_mode_front == regs.polygon_mode_back) { flags[Dirty::PolygonModeFront] = false; flags[Dirty::PolygonModeBack] = false; - glPolygonMode(GL_FRONT_AND_BACK, MaxwellToGL::PolygonMode(gpu.regs.polygon_mode_front)); + glPolygonMode(GL_FRONT_AND_BACK, MaxwellToGL::PolygonMode(regs.polygon_mode_front)); return; } if (flags[Dirty::PolygonModeFront]) { flags[Dirty::PolygonModeFront] = false; - glPolygonMode(GL_FRONT, MaxwellToGL::PolygonMode(gpu.regs.polygon_mode_front)); + glPolygonMode(GL_FRONT, MaxwellToGL::PolygonMode(regs.polygon_mode_front)); } if (flags[Dirty::PolygonModeBack]) { flags[Dirty::PolygonModeBack] = false; - glPolygonMode(GL_BACK, MaxwellToGL::PolygonMode(gpu.regs.polygon_mode_back)); + glPolygonMode(GL_BACK, MaxwellToGL::PolygonMode(regs.polygon_mode_back)); } } void RasterizerOpenGL::SyncColorMask() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::ColorMasks]) { return; } @@ -1371,7 +1374,7 @@ void RasterizerOpenGL::SyncColorMask() { const bool force = flags[Dirty::ColorMaskCommon]; flags[Dirty::ColorMaskCommon] = false; - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; if (regs.color_mask_common) { if (!force && !flags[Dirty::ColorMask0]) { return; @@ -1396,33 +1399,30 @@ void RasterizerOpenGL::SyncColorMask() { } void RasterizerOpenGL::SyncMultiSampleState() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::MultisampleControl]) { return; } flags[Dirty::MultisampleControl] = false; - const auto& regs = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; oglEnable(GL_SAMPLE_ALPHA_TO_COVERAGE, regs.multisample_control.alpha_to_coverage); oglEnable(GL_SAMPLE_ALPHA_TO_ONE, regs.multisample_control.alpha_to_one); } void RasterizerOpenGL::SyncFragmentColorClampState() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::FragmentClampColor]) { return; } flags[Dirty::FragmentClampColor] = false; - glClampColor(GL_CLAMP_FRAGMENT_COLOR, gpu.regs.frag_color_clamp ? GL_TRUE : GL_FALSE); + glClampColor(GL_CLAMP_FRAGMENT_COLOR, maxwell3d.regs.frag_color_clamp ? GL_TRUE : GL_FALSE); } void RasterizerOpenGL::SyncBlendState() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; - const auto& regs = gpu.regs; + auto& flags = maxwell3d.dirty.flags; + const auto& regs = maxwell3d.regs; if (flags[Dirty::BlendColor]) { flags[Dirty::BlendColor] = false; @@ -1479,14 +1479,13 @@ void RasterizerOpenGL::SyncBlendState() { } void RasterizerOpenGL::SyncLogicOpState() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::LogicOp]) { return; } flags[Dirty::LogicOp] = false; - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; if (regs.logic_op.enable) { glEnable(GL_COLOR_LOGIC_OP); glLogicOp(MaxwellToGL::LogicOp(regs.logic_op.operation)); @@ -1496,14 +1495,13 @@ void RasterizerOpenGL::SyncLogicOpState() { } void RasterizerOpenGL::SyncScissorTest() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::Scissors]) { return; } flags[Dirty::Scissors] = false; - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; for (std::size_t index = 0; index < Maxwell::NumViewports; ++index) { if (!flags[Dirty::Scissor0 + index]) { continue; @@ -1522,16 +1520,15 @@ void RasterizerOpenGL::SyncScissorTest() { } void RasterizerOpenGL::SyncPointState() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::PointSize]) { return; } flags[Dirty::PointSize] = false; - oglEnable(GL_POINT_SPRITE, gpu.regs.point_sprite_enable); + oglEnable(GL_POINT_SPRITE, maxwell3d.regs.point_sprite_enable); - if (gpu.regs.vp_point_size.enable) { + if (maxwell3d.regs.vp_point_size.enable) { // By definition of GL_POINT_SIZE, it only matters if GL_PROGRAM_POINT_SIZE is disabled. glEnable(GL_PROGRAM_POINT_SIZE); return; @@ -1539,32 +1536,30 @@ void RasterizerOpenGL::SyncPointState() { // Limit the point size to 1 since nouveau sometimes sets a point size of 0 (and that's invalid // in OpenGL). - glPointSize(std::max(1.0f, gpu.regs.point_size)); + glPointSize(std::max(1.0f, maxwell3d.regs.point_size)); glDisable(GL_PROGRAM_POINT_SIZE); } void RasterizerOpenGL::SyncLineState() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::LineWidth]) { return; } flags[Dirty::LineWidth] = false; - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; oglEnable(GL_LINE_SMOOTH, regs.line_smooth_enable); glLineWidth(regs.line_smooth_enable ? regs.line_width_smooth : regs.line_width_aliased); } void RasterizerOpenGL::SyncPolygonOffset() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::PolygonOffset]) { return; } flags[Dirty::PolygonOffset] = false; - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; oglEnable(GL_POLYGON_OFFSET_FILL, regs.polygon_offset_fill_enable); oglEnable(GL_POLYGON_OFFSET_LINE, regs.polygon_offset_line_enable); oglEnable(GL_POLYGON_OFFSET_POINT, regs.polygon_offset_point_enable); @@ -1578,14 +1573,13 @@ void RasterizerOpenGL::SyncPolygonOffset() { } void RasterizerOpenGL::SyncAlphaTest() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::AlphaTest]) { return; } flags[Dirty::AlphaTest] = false; - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; if (regs.alpha_test_enabled && regs.rt_control.count > 1) { LOG_WARNING(Render_OpenGL, "Alpha testing with more than one render target is not tested"); } @@ -1599,20 +1593,19 @@ void RasterizerOpenGL::SyncAlphaTest() { } void RasterizerOpenGL::SyncFramebufferSRGB() { - auto& gpu = system.GPU().Maxwell3D(); - auto& flags = gpu.dirty.flags; + auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::FramebufferSRGB]) { return; } flags[Dirty::FramebufferSRGB] = false; - oglEnable(GL_FRAMEBUFFER_SRGB, gpu.regs.framebuffer_srgb); + 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 = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; static constexpr std::size_t STRIDE = 3; std::array<GLint, 128 * STRIDE * Maxwell::NumTransformFeedbackBuffers> attribs; @@ -1664,7 +1657,7 @@ void RasterizerOpenGL::SyncTransformFeedback() { } void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) { - const auto& regs = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; if (regs.tfb_enabled == 0) { return; } @@ -1707,7 +1700,7 @@ void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) { } void RasterizerOpenGL::EndTransformFeedback() { - const auto& regs = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; if (regs.tfb_enabled == 0) { return; } diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index 4f082592f..f451404b2 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -33,10 +33,11 @@ #include "video_core/renderer_opengl/gl_state_tracker.h" #include "video_core/renderer_opengl/gl_texture_cache.h" #include "video_core/renderer_opengl/utils.h" +#include "video_core/shader/async_shaders.h" #include "video_core/textures/texture.h" -namespace Core { -class System; +namespace Core::Memory { +class Memory; } namespace Core::Frontend { @@ -54,9 +55,10 @@ struct DrawParameters; class RasterizerOpenGL : public VideoCore::RasterizerAccelerated { public: - explicit RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWindow& emu_window, - const Device& device, ScreenInfo& info, - ProgramManager& program_manager, StateTracker& state_tracker); + explicit RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window, Tegra::GPU& gpu, + Core::Memory::Memory& cpu_memory, const Device& device, + ScreenInfo& screen_info, ProgramManager& program_manager, + StateTracker& state_tracker); ~RasterizerOpenGL() override; void Draw(bool is_indexed, bool is_instanced) override; @@ -82,15 +84,22 @@ public: const Tegra::Engines::Fermi2D::Config& copy_config) override; bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) override; - void LoadDiskResources(const std::atomic_bool& stop_loading, + void LoadDiskResources(u64 title_id, const std::atomic_bool& stop_loading, const VideoCore::DiskResourceLoadCallback& callback) override; - void SetupDirtyFlags() override; /// Returns true when there are commands queued to the OpenGL server. bool AnyCommandQueued() const { return num_queued_commands > 0; } + VideoCommon::Shader::AsyncShaders& GetAsyncShaders() { + return async_shaders; + } + + const VideoCommon::Shader::AsyncShaders& GetAsyncShaders() const { + return async_shaders; + } + private: /// Configures the color and depth framebuffer states. void ConfigureFramebuffers(); @@ -115,9 +124,9 @@ private: /// Configures the current global memory entries to use for the kernel invocation. void SetupComputeGlobalMemory(Shader* kernel); - /// Configures a constant buffer. + /// Configures a global memory buffer. void SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry, GPUVAddr gpu_addr, - std::size_t size); + std::size_t size, GLuint64EXT* pointer); /// Configures the current textures to use for the draw command. void SetupDrawTextures(std::size_t stage_index, Shader* shader); @@ -228,7 +237,15 @@ private: void SetupShaders(GLenum primitive_mode); + Tegra::GPU& gpu; + Tegra::Engines::Maxwell3D& maxwell3d; + Tegra::Engines::KeplerCompute& kepler_compute; + Tegra::MemoryManager& gpu_memory; + const Device& device; + ScreenInfo& screen_info; + ProgramManager& program_manager; + StateTracker& state_tracker; TextureCacheOpenGL texture_cache; ShaderCacheOpenGL shader_cache; @@ -238,10 +255,7 @@ private: OGLBufferCache buffer_cache; FenceManagerOpenGL fence_manager; - Core::System& system; - ScreenInfo& screen_info; - ProgramManager& program_manager; - StateTracker& state_tracker; + VideoCommon::Shader::AsyncShaders async_shaders; static constexpr std::size_t STREAM_BUFFER_SIZE = 128 * 1024 * 1024; diff --git a/src/video_core/renderer_opengl/gl_resource_manager.cpp b/src/video_core/renderer_opengl/gl_resource_manager.cpp index a787e27d2..0ebcec427 100644 --- a/src/video_core/renderer_opengl/gl_resource_manager.cpp +++ b/src/video_core/renderer_opengl/gl_resource_manager.cpp @@ -2,6 +2,7 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <string_view> #include <utility> #include <glad/glad.h> #include "common/common_types.h" @@ -82,11 +83,13 @@ void OGLSampler::Release() { handle = 0; } -void OGLShader::Create(const char* source, GLenum type) { - if (handle != 0) +void OGLShader::Create(std::string_view source, GLenum type) { + if (handle != 0) { return; - if (source == nullptr) + } + if (source.empty()) { return; + } MICROPROFILE_SCOPE(OpenGL_ResourceCreation); handle = GLShader::LoadShader(source, type); diff --git a/src/video_core/renderer_opengl/gl_resource_manager.h b/src/video_core/renderer_opengl/gl_resource_manager.h index f8b322227..f48398669 100644 --- a/src/video_core/renderer_opengl/gl_resource_manager.h +++ b/src/video_core/renderer_opengl/gl_resource_manager.h @@ -4,6 +4,7 @@ #pragma once +#include <string_view> #include <utility> #include <glad/glad.h> #include "common/common_types.h" @@ -127,7 +128,7 @@ public: return *this; } - void Create(const char* source, GLenum type); + void Create(std::string_view source, GLenum type); void Release(); @@ -177,6 +178,12 @@ public: Release(); } + OGLAssemblyProgram& operator=(OGLAssemblyProgram&& o) noexcept { + Release(); + handle = std::exchange(o.handle, 0); + return *this; + } + /// 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 c6a3bf3a1..bd56bed0c 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -22,6 +22,7 @@ #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" @@ -31,6 +32,7 @@ #include "video_core/shader/registry.h" #include "video_core/shader/shader_ir.h" #include "video_core/shader_cache.h" +#include "video_core/shader_notify.h" namespace OpenGL { @@ -125,7 +127,7 @@ 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}; - const auto registry = std::make_shared<Registry>(entry.type, 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); @@ -140,9 +142,24 @@ std::shared_ptr<Registry> MakeRegistry(const ShaderDiskCacheEntry& entry) { return registry; } +std::unordered_set<GLenum> GetSupportedFormats() { + GLint num_formats; + glGetIntegerv(GL_NUM_PROGRAM_BINARY_FORMATS, &num_formats); + + std::vector<GLint> formats(num_formats); + glGetIntegerv(GL_PROGRAM_BINARY_FORMATS, formats.data()); + + std::unordered_set<GLenum> supported_formats; + for (const GLint format : formats) { + supported_formats.insert(static_cast<GLenum>(format)); + } + 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 = false) { + const ShaderIR& ir, const Registry& registry, bool hint_retrievable) { const std::string shader_id = MakeShaderID(unique_identifier, shader_type); LOG_INFO(Render_OpenGL, "{}", shader_id); @@ -181,30 +198,17 @@ ProgramSharedPtr BuildShader(const Device& device, ShaderType shader_type, u64 u return program; } -std::unordered_set<GLenum> GetSupportedFormats() { - GLint num_formats; - glGetIntegerv(GL_NUM_PROGRAM_BINARY_FORMATS, &num_formats); - - std::vector<GLint> formats(num_formats); - glGetIntegerv(GL_PROGRAM_BINARY_FORMATS, formats.data()); - - std::unordered_set<GLenum> supported_formats; - for (const GLint format : formats) { - supported_formats.insert(static_cast<GLenum>(format)); - } - return supported_formats; -} - -} // Anonymous namespace - Shader::Shader(std::shared_ptr<VideoCommon::Shader::Registry> registry_, ShaderEntries entries_, - ProgramSharedPtr program_) - : registry{std::move(registry_)}, entries{std::move(entries_)}, program{std::move(program_)} { + 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; } - ASSERT(handle != 0); + if (is_built) { + ASSERT(handle != 0); + } } Shader::~Shader() = default; @@ -214,43 +218,78 @@ GLuint Shader::GetHandle() const { return handle; } -std::unique_ptr<Shader> Shader::CreateStageFromMemory(const ShaderParameters& params, - Maxwell::ShaderProgram program_type, - ProgramCode code, ProgramCode code_b) { +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); - const std::size_t size_in_bytes = code.size() * sizeof(u64); - auto registry = std::make_shared<Registry>(shader_type, params.system.GPU().Maxwell3D()); - 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); + 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); - 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)); + 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); - return std::unique_ptr<Shader>(new Shader( - std::move(registry), MakeEntries(params.device, ir, shader_type), std::move(program))); + auto program = std::make_shared<ProgramHandle>(); + return std::unique_ptr<Shader>( + new Shader(std::move(registry), std::move(entries), std::move(program), false)); + } } std::unique_ptr<Shader> Shader::CreateKernelFromMemory(const ShaderParameters& params, ProgramCode code) { - const std::size_t size_in_bytes = code.size() * sizeof(u64); + auto& gpu = params.gpu; + gpu.ShaderNotify().MarkSharderBuilding(); - auto& engine = params.system.GPU().KeplerCompute(); - auto registry = std::make_shared<Registry>(ShaderType::Compute, engine); + 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); @@ -266,6 +305,8 @@ std::unique_ptr<Shader> Shader::CreateKernelFromMemory(const ShaderParameters& p 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))); @@ -277,15 +318,20 @@ std::unique_ptr<Shader> Shader::CreateFromCache(const ShaderParameters& params, precompiled_shader.registry, precompiled_shader.entries, precompiled_shader.program)); } -ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer, Core::System& system, - Core::Frontend::EmuWindow& emu_window, const Device& device) - : VideoCommon::ShaderCache<Shader>{rasterizer}, system{system}, - emu_window{emu_window}, device{device}, disk_cache{system} {} +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_) + : VideoCommon::ShaderCache<Shader>{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(const std::atomic_bool& stop_loading, +void ShaderCacheOpenGL::LoadDiskCache(u64 title_id, const std::atomic_bool& stop_loading, const VideoCore::DiskResourceLoadCallback& callback) { + disk_cache.BindTitleID(title_id); const std::optional transferable = disk_cache.LoadTransferable(); if (!transferable) { return; @@ -361,7 +407,7 @@ void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading, } }; - const auto num_workers{static_cast<std::size_t>(std::thread::hardware_concurrency() + 1ULL)}; + 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); @@ -436,42 +482,77 @@ ProgramSharedPtr ShaderCacheOpenGL::GeneratePrecompiledProgram( return program; } -Shader* ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) { - if (!system.GPU().Maxwell3D().dirty.flags[Dirty::Shaders]) { - return last_shaders[static_cast<std::size_t>(program)]; +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; + } } - auto& memory_manager{system.GPU().MemoryManager()}; - const GPUVAddr address{GetShaderAddress(system, program)}; + const GPUVAddr address{GetShaderAddress(maxwell3d, program)}; + + 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; + } + 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)); + } + + 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)); + } + } // Look up shader in the cache based on address - const auto cpu_addr{memory_manager.GpuToCpuAddress(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 auto host_ptr{memory_manager.GetPointer(address)}; + const u8* const host_ptr{gpu_memory.GetPointer(address)}; // No shader found - create a new one - ProgramCode code{GetShaderCode(memory_manager, address, host_ptr, false)}; + ProgramCode code{GetShaderCode(gpu_memory, address, host_ptr, false)}; ProgramCode code_b; if (program == Maxwell::ShaderProgram::VertexA) { - const GPUVAddr address_b{GetShaderAddress(system, Maxwell::ShaderProgram::VertexB)}; - const u8* host_ptr_b = memory_manager.GetPointer(address_b); - code_b = GetShaderCode(memory_manager, address_b, host_ptr_b, false); + 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{system, disk_cache, device, - *cpu_addr, host_ptr, unique_identifier}; + 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)); + 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); } @@ -487,21 +568,20 @@ Shader* ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) { } Shader* ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) { - auto& memory_manager{system.GPU().MemoryManager()}; - const auto cpu_addr{memory_manager.GpuToCpuAddress(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; } - const auto host_ptr{memory_manager.GetPointer(code_addr)}; // No kernel found, create a new one - ProgramCode code{GetShaderCode(memory_manager, code_addr, host_ptr, true)}; + 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{system, disk_cache, device, - *cpu_addr, host_ptr, unique_identifier}; + const ShaderParameters params{gpu, kepler_compute, disk_cache, device, + *cpu_addr, host_ptr, unique_identifier}; std::unique_ptr<Shader> kernel; const auto found = runtime_cache.find(unique_identifier); diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h index 994aaeaf2..1708af06a 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_cache.h @@ -25,14 +25,18 @@ #include "video_core/shader/shader_ir.h" #include "video_core/shader_cache.h" -namespace Core { -class System; +namespace Tegra { +class MemoryManager; } namespace Core::Frontend { class EmuWindow; } +namespace VideoCommon::Shader { +class AsyncShaders; +} + namespace OpenGL { class Device; @@ -53,14 +57,20 @@ struct PrecompiledShader { }; struct ShaderParameters { - Core::System& system; + Tegra::GPU& gpu; + Tegra::Engines::ConstBufferEngineInterface& engine; ShaderDiskCacheOpenGL& disk_cache; const Device& device; VAddr cpu_addr; - u8* host_ptr; + 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 { public: ~Shader(); @@ -68,15 +78,28 @@ public: /// 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; } - static std::unique_ptr<Shader> CreateStageFromMemory(const ShaderParameters& params, - Maxwell::ShaderProgram program_type, - ProgramCode program_code, - ProgramCode program_code_b); + const VideoCommon::Shader::Registry& GetRegistry() const { + return *registry; + } + + /// Mark a OpenGL shader as built + void AsyncOpenGLBuilt(OGLProgram new_program); + + /// Mark a GLASM shader as built + void AsyncGLASMBuilt(OGLAssemblyProgram new_program); + + 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); + static std::unique_ptr<Shader> CreateKernelFromMemory(const ShaderParameters& params, ProgramCode code); @@ -85,26 +108,30 @@ public: private: explicit Shader(std::shared_ptr<VideoCommon::Shader::Registry> registry, ShaderEntries entries, - ProgramSharedPtr program); + ProgramSharedPtr program, bool is_built = true); std::shared_ptr<VideoCommon::Shader::Registry> registry; ShaderEntries entries; ProgramSharedPtr program; GLuint handle = 0; + bool is_built{}; }; class ShaderCacheOpenGL final : public VideoCommon::ShaderCache<Shader> { public: - explicit ShaderCacheOpenGL(RasterizerOpenGL& rasterizer, Core::System& system, - Core::Frontend::EmuWindow& emu_window, const Device& device); + 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; /// Loads disk cache for the current game - void LoadDiskCache(const std::atomic_bool& stop_loading, + void LoadDiskCache(u64 title_id, const std::atomic_bool& stop_loading, const VideoCore::DiskResourceLoadCallback& callback); /// Gets the current specified shader stage program - Shader* GetStageProgram(Maxwell::ShaderProgram program); + Shader* GetStageProgram(Maxwell::ShaderProgram program, + VideoCommon::Shader::AsyncShaders& async_shaders); /// Gets a compute kernel in the passed address Shader* GetComputeKernel(GPUVAddr code_addr); @@ -114,9 +141,13 @@ private: const ShaderDiskCacheEntry& entry, const ShaderDiskCachePrecompiled& precompiled_entry, const std::unordered_set<GLenum>& supported_formats); - Core::System& system; Core::Frontend::EmuWindow& emu_window; + Tegra::GPU& gpu; + Tegra::MemoryManager& gpu_memory; + Tegra::Engines::Maxwell3D& maxwell3d; + Tegra::Engines::KeplerCompute& kepler_compute; const Device& device; + ShaderDiskCacheOpenGL disk_cache; std::unordered_map<u64, PrecompiledShader> runtime_cache; diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 2c49aeaac..bbb8fb095 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -602,8 +602,15 @@ private: return; } const auto& info = registry.GetComputeInfo(); - if (const u32 size = info.shared_memory_size_in_words; size > 0) { - code.AddLine("shared uint smem[{}];", size); + 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;", @@ -806,7 +813,7 @@ private: 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 std::nullopt; } return it->second.components; } @@ -1288,21 +1295,21 @@ private: switch (element) { case 0: UNIMPLEMENTED(); - return {}; + return std::nullopt; case 1: if (stage == ShaderType::Vertex && !device.HasVertexViewportLayer()) { - return {}; + return std::nullopt; } return {{"gl_Layer", Type::Int}}; case 2: if (stage == ShaderType::Vertex && !device.HasVertexViewportLayer()) { - return {}; + return std::nullopt; } return {{"gl_ViewportIndex", Type::Int}}; case 3: return {{"gl_PointSize", Type::Float}}; } - return {}; + return std::nullopt; case Attribute::Index::FrontColor: return {{"gl_FrontColor"s + GetSwizzle(element), Type::Float}}; case Attribute::Index::FrontSecondaryColor: @@ -1325,7 +1332,7 @@ private: Type::Float}}; } UNIMPLEMENTED_MSG("Unhandled output attribute: {}", static_cast<u32>(attribute)); - return {}; + return std::nullopt; } } @@ -1436,8 +1443,10 @@ private: return expr + ", vec2(0.0), vec2(0.0))"; case TextureType::TextureCube: return expr + ", vec3(0.0), vec3(0.0))"; + default: + UNREACHABLE(); + break; } - UNREACHABLE(); } for (const auto& variant : extras) { @@ -1912,7 +1921,7 @@ private: Expression Comparison(Operation operation) { static_assert(!unordered || type == Type::Float); - const Expression expr = GenerateBinaryInfix(operation, op, Type::Bool, type, type); + 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 @@ -1952,10 +1961,6 @@ private: return {fmt::format("({} != 0)", carry), Type::Bool}; } - Expression LogicalFIsNan(Operation operation) { - return GenerateUnary(operation, "isnan", Type::Bool, Type::Float); - } - Expression LogicalAssign(Operation operation) { const Node& dest = operation[0]; const Node& src = operation[1]; @@ -2771,15 +2776,6 @@ private: return std::min<u32>(device.GetMaxVaryings(), Maxwell::NumVaryings); } - bool IsRenderTargetEnabled(u32 render_target) const { - for (u32 component = 0; component < 4; ++component) { - if (header.ps.IsColorComponentOutputEnabled(render_target, component)) { - return true; - } - } - return false; - } - const Device& device; const ShaderIR& ir; const Registry& registry; diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp index 2dcc2b0eb..166ee34e1 100644 --- a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp @@ -73,7 +73,7 @@ ShaderDiskCacheEntry::ShaderDiskCacheEntry() = default; ShaderDiskCacheEntry::~ShaderDiskCacheEntry() = default; -bool ShaderDiskCacheEntry::Load(FileUtil::IOFile& file) { +bool ShaderDiskCacheEntry::Load(Common::FS::IOFile& file) { if (file.ReadBytes(&type, sizeof(u32)) != sizeof(u32)) { return false; } @@ -144,7 +144,7 @@ bool ShaderDiskCacheEntry::Load(FileUtil::IOFile& file) { return true; } -bool ShaderDiskCacheEntry::Save(FileUtil::IOFile& file) const { +bool ShaderDiskCacheEntry::Save(Common::FS::IOFile& file) const { if (file.WriteObject(static_cast<u32>(type)) != 1 || file.WriteObject(static_cast<u32>(code.size())) != 1 || file.WriteObject(static_cast<u32>(code_b.size())) != 1) { @@ -206,28 +206,32 @@ bool ShaderDiskCacheEntry::Save(FileUtil::IOFile& file) const { flat_bindless_samplers.size(); } -ShaderDiskCacheOpenGL::ShaderDiskCacheOpenGL(Core::System& system) : system{system} {} +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 = system.CurrentProcess()->GetTitleID() != 0; + const bool has_title_id = title_id != 0; if (!Settings::values.use_disk_shader_cache.GetValue() || !has_title_id) { - return {}; + return std::nullopt; } - FileUtil::IOFile file(GetTransferablePath(), "rb"); + Common::FS::IOFile file(GetTransferablePath(), "rb"); if (!file.IsOpen()) { LOG_INFO(Render_OpenGL, "No transferable shader cache found"); is_usable = true; - return {}; + return std::nullopt; } u32 version{}; if (file.ReadBytes(&version, sizeof(version)) != sizeof(version)) { LOG_ERROR(Render_OpenGL, "Failed to get transferable cache version, skipping it"); - return {}; + return std::nullopt; } if (version < NativeVersion) { @@ -235,12 +239,12 @@ std::optional<std::vector<ShaderDiskCacheEntry>> ShaderDiskCacheOpenGL::LoadTran file.Close(); InvalidateTransferable(); is_usable = true; - return {}; + return std::nullopt; } if (version > NativeVersion) { LOG_WARNING(Render_OpenGL, "Transferable shader cache was generated with a newer version " "of the emulator, skipping"); - return {}; + return std::nullopt; } // Version is valid, load the shaders @@ -249,7 +253,7 @@ std::optional<std::vector<ShaderDiskCacheEntry>> ShaderDiskCacheOpenGL::LoadTran ShaderDiskCacheEntry& entry = entries.emplace_back(); if (!entry.Load(file)) { LOG_ERROR(Render_OpenGL, "Failed to load transferable raw entry, skipping"); - return {}; + return std::nullopt; } } @@ -262,7 +266,7 @@ std::vector<ShaderDiskCachePrecompiled> ShaderDiskCacheOpenGL::LoadPrecompiled() return {}; } - FileUtil::IOFile file(GetPrecompiledPath(), "rb"); + Common::FS::IOFile file(GetPrecompiledPath(), "rb"); if (!file.IsOpen()) { LOG_INFO(Render_OpenGL, "No precompiled shader cache found"); return {}; @@ -279,7 +283,7 @@ std::vector<ShaderDiskCachePrecompiled> ShaderDiskCacheOpenGL::LoadPrecompiled() } std::optional<std::vector<ShaderDiskCachePrecompiled>> ShaderDiskCacheOpenGL::LoadPrecompiledFile( - FileUtil::IOFile& file) { + Common::FS::IOFile& file) { // Read compressed file from disk and decompress to virtual precompiled cache file std::vector<u8> compressed(file.GetSize()); file.ReadBytes(compressed.data(), compressed.size()); @@ -290,12 +294,12 @@ std::optional<std::vector<ShaderDiskCachePrecompiled>> ShaderDiskCacheOpenGL::Lo ShaderCacheVersionHash file_hash{}; if (!LoadArrayFromPrecompiled(file_hash.data(), file_hash.size())) { precompiled_cache_virtual_file_offset = 0; - return {}; + 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 {}; + return std::nullopt; } std::vector<ShaderDiskCachePrecompiled> entries; @@ -305,19 +309,20 @@ std::optional<std::vector<ShaderDiskCachePrecompiled>> ShaderDiskCacheOpenGL::Lo if (!LoadObjectFromPrecompiled(entry.unique_identifier) || !LoadObjectFromPrecompiled(entry.binary_format) || !LoadObjectFromPrecompiled(binary_size)) { - return {}; + return std::nullopt; } entry.binary.resize(binary_size); if (!LoadArrayFromPrecompiled(entry.binary.data(), entry.binary.size())) { - return {}; + return std::nullopt; } } - return entries; + + return std::move(entries); } void ShaderDiskCacheOpenGL::InvalidateTransferable() { - if (!FileUtil::Delete(GetTransferablePath())) { + if (!Common::FS::Delete(GetTransferablePath())) { LOG_ERROR(Render_OpenGL, "Failed to invalidate transferable file={}", GetTransferablePath()); } @@ -328,7 +333,7 @@ void ShaderDiskCacheOpenGL::InvalidatePrecompiled() { // Clear virtaul precompiled cache file precompiled_cache_virtual_file.Resize(0); - if (!FileUtil::Delete(GetPrecompiledPath())) { + if (!Common::FS::Delete(GetPrecompiledPath())) { LOG_ERROR(Render_OpenGL, "Failed to invalidate precompiled file={}", GetPrecompiledPath()); } } @@ -344,7 +349,7 @@ void ShaderDiskCacheOpenGL::SaveEntry(const ShaderDiskCacheEntry& entry) { return; } - FileUtil::IOFile file = AppendTransferableFile(); + Common::FS::IOFile file = AppendTransferableFile(); if (!file.IsOpen()) { return; } @@ -386,15 +391,15 @@ void ShaderDiskCacheOpenGL::SavePrecompiled(u64 unique_identifier, GLuint progra } } -FileUtil::IOFile ShaderDiskCacheOpenGL::AppendTransferableFile() const { +Common::FS::IOFile ShaderDiskCacheOpenGL::AppendTransferableFile() const { if (!EnsureDirectories()) { return {}; } const auto transferable_path{GetTransferablePath()}; - const bool existed = FileUtil::Exists(transferable_path); + const bool existed = Common::FS::Exists(transferable_path); - FileUtil::IOFile file(transferable_path, "ab"); + Common::FS::IOFile file(transferable_path, "ab"); if (!file.IsOpen()) { LOG_ERROR(Render_OpenGL, "Failed to open transferable cache in path={}", transferable_path); return {}; @@ -426,7 +431,7 @@ void ShaderDiskCacheOpenGL::SaveVirtualPrecompiledFile() { Common::Compression::CompressDataZSTDDefault(uncompressed.data(), uncompressed.size()); const auto precompiled_path{GetPrecompiledPath()}; - FileUtil::IOFile file(precompiled_path, "wb"); + Common::FS::IOFile file(precompiled_path, "wb"); if (!file.IsOpen()) { LOG_ERROR(Render_OpenGL, "Failed to open precompiled cache in path={}", precompiled_path); @@ -440,24 +445,24 @@ void ShaderDiskCacheOpenGL::SaveVirtualPrecompiledFile() { bool ShaderDiskCacheOpenGL::EnsureDirectories() const { const auto CreateDir = [](const std::string& dir) { - if (!FileUtil::CreateDir(dir)) { + if (!Common::FS::CreateDir(dir)) { LOG_ERROR(Render_OpenGL, "Failed to create directory={}", dir); return false; } return true; }; - return CreateDir(FileUtil::GetUserPath(FileUtil::UserPath::ShaderDir)) && + return CreateDir(Common::FS::GetUserPath(Common::FS::UserPath::ShaderDir)) && CreateDir(GetBaseDir()) && CreateDir(GetTransferableDir()) && CreateDir(GetPrecompiledDir()); } std::string ShaderDiskCacheOpenGL::GetTransferablePath() const { - return FileUtil::SanitizePath(GetTransferableDir() + DIR_SEP_CHR + GetTitleID() + ".bin"); + return Common::FS::SanitizePath(GetTransferableDir() + DIR_SEP_CHR + GetTitleID() + ".bin"); } std::string ShaderDiskCacheOpenGL::GetPrecompiledPath() const { - return FileUtil::SanitizePath(GetPrecompiledDir() + DIR_SEP_CHR + GetTitleID() + ".bin"); + return Common::FS::SanitizePath(GetPrecompiledDir() + DIR_SEP_CHR + GetTitleID() + ".bin"); } std::string ShaderDiskCacheOpenGL::GetTransferableDir() const { @@ -469,11 +474,11 @@ std::string ShaderDiskCacheOpenGL::GetPrecompiledDir() const { } std::string ShaderDiskCacheOpenGL::GetBaseDir() const { - return FileUtil::GetUserPath(FileUtil::UserPath::ShaderDir) + DIR_SEP "opengl"; + return Common::FS::GetUserPath(Common::FS::UserPath::ShaderDir) + DIR_SEP "opengl"; } std::string ShaderDiskCacheOpenGL::GetTitleID() const { - return fmt::format("{:016X}", system.CurrentProcess()->GetTitleID()); + 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 index a79cef0e9..aef841c1d 100644 --- a/src/video_core/renderer_opengl/gl_shader_disk_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.h @@ -21,11 +21,7 @@ #include "video_core/engines/shader_type.h" #include "video_core/shader/registry.h" -namespace Core { -class System; -} - -namespace FileUtil { +namespace Common::FS { class IOFile; } @@ -38,9 +34,9 @@ struct ShaderDiskCacheEntry { ShaderDiskCacheEntry(); ~ShaderDiskCacheEntry(); - bool Load(FileUtil::IOFile& file); + bool Load(Common::FS::IOFile& file); - bool Save(FileUtil::IOFile& file) const; + bool Save(Common::FS::IOFile& file) const; bool HasProgramA() const { return !code.empty() && !code_b.empty(); @@ -70,9 +66,12 @@ struct ShaderDiskCachePrecompiled { class ShaderDiskCacheOpenGL { public: - explicit ShaderDiskCacheOpenGL(Core::System& system); + 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(); @@ -97,10 +96,10 @@ public: private: /// Loads the transferable cache. Returns empty on failure. std::optional<std::vector<ShaderDiskCachePrecompiled>> LoadPrecompiledFile( - FileUtil::IOFile& file); + Common::FS::IOFile& file); /// Opens current game's transferable file and write it's header if it doesn't exist - FileUtil::IOFile AppendTransferableFile() const; + Common::FS::IOFile AppendTransferableFile() const; /// Save precompiled header to precompiled_cache_in_memory void SavePrecompiledHeaderToVirtualPrecompiledCache(); @@ -157,8 +156,6 @@ private: return LoadArrayFromPrecompiled(&object, 1); } - Core::System& system; - // Stores whole precompiled cache which will be read from or saved to the precompiled chache // file FileSys::VectorVfsFile precompiled_cache_virtual_file; @@ -168,8 +165,11 @@ private: // 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{}; + 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 8e754fa90..691c6c79b 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.cpp +++ b/src/video_core/renderer_opengl/gl_shader_manager.cpp @@ -11,8 +11,30 @@ namespace OpenGL { -ProgramManager::ProgramManager(const Device& device) { - use_assembly_programs = device.UseAssemblyShaders(); +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 { @@ -33,9 +55,7 @@ void ProgramManager::BindCompute(GLuint program) { } void ProgramManager::BindGraphicsPipeline() { - if (use_assembly_programs) { - UpdateAssemblyPrograms(); - } else { + if (!use_assembly_programs) { UpdateSourcePrograms(); } } @@ -63,32 +83,25 @@ void ProgramManager::RestoreGuestPipeline() { } } -void ProgramManager::UpdateAssemblyPrograms() { - const auto update_state = [](GLenum stage, bool& enabled, GLuint current, GLuint old) { - if (current == old) { - return; - } - if (current == 0) { - if (enabled) { - enabled = false; - glDisable(stage); - } - return; - } - if (!enabled) { - enabled = true; - glEnable(stage); - } - glBindProgramARB(stage, current); - }; +void ProgramManager::UseVertexShader(GLuint program) { + if (use_assembly_programs) { + BindProgram(GL_VERTEX_PROGRAM_NV, program, current_state.vertex, vertex_enabled); + } + current_state.vertex = program; +} - update_state(GL_VERTEX_PROGRAM_NV, vertex_enabled, current_state.vertex, old_state.vertex); - update_state(GL_GEOMETRY_PROGRAM_NV, geometry_enabled, current_state.geometry, - old_state.geometry); - update_state(GL_FRAGMENT_PROGRAM_NV, fragment_enabled, current_state.fragment, - old_state.fragment); +void ProgramManager::UseGeometryShader(GLuint program) { + if (use_assembly_programs) { + BindProgram(GL_GEOMETRY_PROGRAM_NV, program, current_state.vertex, geometry_enabled); + } + current_state.geometry = program; +} - old_state = current_state; +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() { diff --git a/src/video_core/renderer_opengl/gl_shader_manager.h b/src/video_core/renderer_opengl/gl_shader_manager.h index 0f03b4f12..950e0dfcb 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.h +++ b/src/video_core/renderer_opengl/gl_shader_manager.h @@ -45,17 +45,9 @@ public: /// Rewinds BindHostPipeline state changes. void RestoreGuestPipeline(); - void UseVertexShader(GLuint program) { - current_state.vertex = program; - } - - void UseGeometryShader(GLuint program) { - current_state.geometry = program; - } - - void UseFragmentShader(GLuint program) { - current_state.fragment = program; - } + void UseVertexShader(GLuint program); + void UseGeometryShader(GLuint program); + void UseFragmentShader(GLuint program); private: struct PipelineState { @@ -64,9 +56,6 @@ private: GLuint fragment = 0; }; - /// Update NV_gpu_program5 programs. - void UpdateAssemblyPrograms(); - /// Update GLSL programs. void UpdateSourcePrograms(); diff --git a/src/video_core/renderer_opengl/gl_shader_util.cpp b/src/video_core/renderer_opengl/gl_shader_util.cpp index 9e74eda0d..4bf0d6090 100644 --- a/src/video_core/renderer_opengl/gl_shader_util.cpp +++ b/src/video_core/renderer_opengl/gl_shader_util.cpp @@ -2,6 +2,7 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <string_view> #include <vector> #include <glad/glad.h> #include "common/assert.h" @@ -11,7 +12,8 @@ namespace OpenGL::GLShader { namespace { -const char* GetStageDebugName(GLenum type) { + +std::string_view StageDebugName(GLenum type) { switch (type) { case GL_VERTEX_SHADER: return "vertex"; @@ -25,12 +27,17 @@ const char* GetStageDebugName(GLenum type) { UNIMPLEMENTED(); return "unknown"; } + } // Anonymous namespace -GLuint LoadShader(const char* source, GLenum type) { - const char* debug_type = GetStageDebugName(type); +GLuint LoadShader(std::string_view source, GLenum type) { + const std::string_view debug_type = StageDebugName(type); const GLuint shader_id = glCreateShader(type); - glShaderSource(shader_id, 1, &source, nullptr); + + const GLchar* source_string = source.data(); + const GLint source_length = static_cast<GLint>(source.size()); + + glShaderSource(shader_id, 1, &source_string, &source_length); LOG_DEBUG(Render_OpenGL, "Compiling {} shader...", debug_type); glCompileShader(shader_id); diff --git a/src/video_core/renderer_opengl/gl_shader_util.h b/src/video_core/renderer_opengl/gl_shader_util.h index 03b7548c2..1b770532e 100644 --- a/src/video_core/renderer_opengl/gl_shader_util.h +++ b/src/video_core/renderer_opengl/gl_shader_util.h @@ -38,7 +38,7 @@ void LogShaderSource(T... shaders) { * @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(const char* source, GLenum type); +GLuint LoadShader(std::string_view source, GLenum type); /** * Utility function to create and compile an OpenGL GLSL shader program (vertex + fragment shader) diff --git a/src/video_core/renderer_opengl/gl_state_tracker.cpp b/src/video_core/renderer_opengl/gl_state_tracker.cpp index d24fad3de..6bcf831f2 100644 --- a/src/video_core/renderer_opengl/gl_state_tracker.cpp +++ b/src/video_core/renderer_opengl/gl_state_tracker.cpp @@ -214,10 +214,8 @@ void SetupDirtyMisc(Tables& tables) { } // Anonymous namespace -StateTracker::StateTracker(Core::System& system) : system{system} {} - -void StateTracker::Initialize() { - auto& dirty = system.GPU().Maxwell3D().dirty; +StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags} { + auto& dirty = gpu.Maxwell3D().dirty; auto& tables = dirty.tables; SetupDirtyRenderTargets(tables); SetupDirtyColorMasks(tables); diff --git a/src/video_core/renderer_opengl/gl_state_tracker.h b/src/video_core/renderer_opengl/gl_state_tracker.h index 0f823288e..9d127548f 100644 --- a/src/video_core/renderer_opengl/gl_state_tracker.h +++ b/src/video_core/renderer_opengl/gl_state_tracker.h @@ -13,8 +13,8 @@ #include "video_core/dirty_flags.h" #include "video_core/engines/maxwell_3d.h" -namespace Core { -class System; +namespace Tegra { +class GPU; } namespace OpenGL { @@ -90,9 +90,7 @@ static_assert(Last <= std::numeric_limits<u8>::max()); class StateTracker { public: - explicit StateTracker(Core::System& system); - - void Initialize(); + explicit StateTracker(Tegra::GPU& gpu); void BindIndexBuffer(GLuint new_index_buffer) { if (index_buffer == new_index_buffer) { @@ -103,7 +101,6 @@ public: } void NotifyScreenDrawVertexArray() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::VertexFormats] = true; flags[OpenGL::Dirty::VertexFormat0 + 0] = true; flags[OpenGL::Dirty::VertexFormat0 + 1] = true; @@ -117,98 +114,81 @@ public: } void NotifyPolygonModes() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::PolygonModes] = true; flags[OpenGL::Dirty::PolygonModeFront] = true; flags[OpenGL::Dirty::PolygonModeBack] = true; } void NotifyViewport0() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::Viewports] = true; flags[OpenGL::Dirty::Viewport0] = true; } void NotifyScissor0() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::Scissors] = true; flags[OpenGL::Dirty::Scissor0] = true; } void NotifyColorMask0() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::ColorMasks] = true; flags[OpenGL::Dirty::ColorMask0] = true; } void NotifyBlend0() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::BlendStates] = true; flags[OpenGL::Dirty::BlendState0] = true; } void NotifyFramebuffer() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[VideoCommon::Dirty::RenderTargets] = true; } void NotifyFrontFace() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::FrontFace] = true; } void NotifyCullTest() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::CullTest] = true; } void NotifyDepthMask() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::DepthMask] = true; } void NotifyDepthTest() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::DepthTest] = true; } void NotifyStencilTest() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::StencilTest] = true; } void NotifyPolygonOffset() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::PolygonOffset] = true; } void NotifyRasterizeEnable() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::RasterizeEnable] = true; } void NotifyFramebufferSRGB() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::FramebufferSRGB] = true; } void NotifyLogicOp() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::LogicOp] = true; } void NotifyClipControl() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::ClipControl] = true; } void NotifyAlphaTest() { - auto& flags = system.GPU().Maxwell3D().dirty.flags; flags[OpenGL::Dirty::AlphaTest] = true; } private: - Core::System& system; + Tegra::Engines::Maxwell3D::DirtyState::Flags& flags; GLuint index_buffer = 0; }; diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.cpp b/src/video_core/renderer_opengl/gl_stream_buffer.cpp index 3655ff629..887995cf4 100644 --- a/src/video_core/renderer_opengl/gl_stream_buffer.cpp +++ b/src/video_core/renderer_opengl/gl_stream_buffer.cpp @@ -35,7 +35,7 @@ OGLStreamBuffer::OGLStreamBuffer(const Device& device, GLsizeiptr size, bool ver mapped_ptr = static_cast<u8*>( glMapNamedBufferRange(gl_buffer.handle, 0, buffer_size, flags | GL_MAP_FLUSH_EXPLICIT_BIT)); - if (device.HasVertexBufferUnifiedMemory()) { + if (device.UseAssemblyShaders() || device.HasVertexBufferUnifiedMemory()) { glMakeNamedBufferResidentNV(gl_buffer.handle, GL_READ_ONLY); glGetNamedBufferParameterui64vNV(gl_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, &gpu_address); } diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index 61505879b..a863ef218 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -41,91 +41,103 @@ struct FormatTuple { }; constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex_format_tuples = {{ - {GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV}, // ABGR8U - {GL_RGBA8_SNORM, GL_RGBA, GL_BYTE}, // ABGR8S - {GL_RGBA8UI, GL_RGBA_INTEGER, GL_UNSIGNED_BYTE}, // ABGR8UI - {GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV}, // B5G6R5U - {GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV}, // A2B10G10R10U - {GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_1_5_5_5_REV}, // A1B5G5R5U - {GL_R8, GL_RED, GL_UNSIGNED_BYTE}, // R8U - {GL_R8UI, GL_RED_INTEGER, GL_UNSIGNED_BYTE}, // R8UI - {GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT}, // RGBA16F - {GL_RGBA16, GL_RGBA, GL_UNSIGNED_SHORT}, // RGBA16U - {GL_RGBA16_SNORM, GL_RGBA, GL_SHORT}, // RGBA16S - {GL_RGBA16UI, GL_RGBA_INTEGER, GL_UNSIGNED_SHORT}, // RGBA16UI - {GL_R11F_G11F_B10F, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV}, // R11FG11FB10F - {GL_RGBA32UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT}, // RGBA32UI - {GL_COMPRESSED_RGBA_S3TC_DXT1_EXT}, // DXT1 - {GL_COMPRESSED_RGBA_S3TC_DXT3_EXT}, // DXT23 - {GL_COMPRESSED_RGBA_S3TC_DXT5_EXT}, // DXT45 - {GL_COMPRESSED_RED_RGTC1}, // DXN1 - {GL_COMPRESSED_RG_RGTC2}, // DXN2UNORM - {GL_COMPRESSED_SIGNED_RG_RGTC2}, // DXN2SNORM - {GL_COMPRESSED_RGBA_BPTC_UNORM}, // BC7U - {GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT}, // BC6H_UF16 - {GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT}, // BC6H_SF16 - {GL_COMPRESSED_RGBA_ASTC_4x4_KHR}, // ASTC_2D_4X4 - {GL_RGBA8, GL_BGRA, GL_UNSIGNED_BYTE}, // BGRA8 - {GL_RGBA32F, GL_RGBA, GL_FLOAT}, // RGBA32F - {GL_RG32F, GL_RG, GL_FLOAT}, // RG32F - {GL_R32F, GL_RED, GL_FLOAT}, // R32F - {GL_R16F, GL_RED, GL_HALF_FLOAT}, // R16F - {GL_R16, GL_RED, GL_UNSIGNED_SHORT}, // R16U - {GL_R16_SNORM, GL_RED, GL_SHORT}, // R16S - {GL_R16UI, GL_RED_INTEGER, GL_UNSIGNED_SHORT}, // R16UI - {GL_R16I, GL_RED_INTEGER, GL_SHORT}, // R16I - {GL_RG16, GL_RG, GL_UNSIGNED_SHORT}, // RG16 - {GL_RG16F, GL_RG, GL_HALF_FLOAT}, // RG16F - {GL_RG16UI, GL_RG_INTEGER, GL_UNSIGNED_SHORT}, // RG16UI - {GL_RG16I, GL_RG_INTEGER, GL_SHORT}, // RG16I - {GL_RG16_SNORM, GL_RG, GL_SHORT}, // RG16S - {GL_RGB32F, GL_RGB, GL_FLOAT}, // RGB32F - {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV}, // RGBA8_SRGB - {GL_RG8, GL_RG, GL_UNSIGNED_BYTE}, // RG8U - {GL_RG8_SNORM, GL_RG, GL_BYTE}, // RG8S - {GL_RG8UI, GL_RG_INTEGER, GL_UNSIGNED_INT}, // RG8UI - {GL_RG32UI, GL_RG_INTEGER, GL_UNSIGNED_INT}, // RG32UI - {GL_RGB16F, GL_RGBA, GL_HALF_FLOAT}, // RGBX16F - {GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT}, // R32UI - {GL_R32I, GL_RED_INTEGER, GL_INT}, // R32I - {GL_COMPRESSED_RGBA_ASTC_8x8_KHR}, // ASTC_2D_8X8 - {GL_COMPRESSED_RGBA_ASTC_8x5_KHR}, // ASTC_2D_8X5 - {GL_COMPRESSED_RGBA_ASTC_5x4_KHR}, // ASTC_2D_5X4 - {GL_SRGB8_ALPHA8, GL_BGRA, GL_UNSIGNED_BYTE}, // BGRA8 + {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_BGRA, 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_BGRA, GL_UNSIGNED_BYTE}, // B8G8R8A8_UNORM // Compressed sRGB formats - {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT}, // DXT1_SRGB - {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT}, // DXT23_SRGB - {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT}, // DXT45_SRGB - {GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM}, // BC7U_SRGB - {GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4_REV}, // R4G4B4A4U + {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 + {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 + {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 + {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 + {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 + {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 + {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 + {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}, // E5B9G9R9F + {GL_RGB9_E5, GL_RGB, GL_UNSIGNED_INT_5_9_9_9_REV}, // E5B9G9R9_FLOAT // Depth formats - {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, // Z32F - {GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // Z16 + {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, // D32_FLOAT + {GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16_UNORM // DepthStencil formats - {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // Z24S8 - {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // S8Z24 - {GL_DEPTH32F_STENCIL8, GL_DEPTH_STENCIL, GL_FLOAT_32_UNSIGNED_INT_24_8_REV}, // Z32FS8 + {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 }}; const FormatTuple& GetFormatTuple(PixelFormat pixel_format) { @@ -178,10 +190,10 @@ GLint GetSwizzleSource(SwizzleSource source) { GLenum GetComponent(PixelFormat format, bool is_first) { switch (format) { - case PixelFormat::Z24S8: - case PixelFormat::Z32FS8: + case PixelFormat::D24_UNORM_S8_UINT: + case PixelFormat::D32_FLOAT_S8_UINT: return is_first ? GL_DEPTH_COMPONENT : GL_STENCIL_INDEX; - case PixelFormat::S8Z24: + case PixelFormat::S8_UINT_D24_UNORM: return is_first ? GL_STENCIL_INDEX : GL_DEPTH_COMPONENT; default: UNREACHABLE(); @@ -391,7 +403,7 @@ void CachedSurface::DecorateSurfaceName() { LabelGLObject(GL_TEXTURE, texture.handle, GetGpuAddr(), params.TargetName()); } -void CachedSurfaceView::DecorateViewName(GPUVAddr gpu_addr, std::string prefix) { +void CachedSurfaceView::DecorateViewName(GPUVAddr gpu_addr, const std::string& prefix) { LabelGLObject(GL_TEXTURE, main_view.handle, gpu_addr, prefix); } @@ -482,9 +494,9 @@ GLuint CachedSurfaceView::GetTexture(SwizzleSource x_source, SwizzleSource y_sou std::array swizzle{x_source, y_source, z_source, w_source}; switch (const PixelFormat format = GetSurfaceParams().pixel_format) { - case PixelFormat::Z24S8: - case PixelFormat::Z32FS8: - case PixelFormat::S8Z24: + case PixelFormat::D24_UNORM_S8_UINT: + case PixelFormat::D32_FLOAT_S8_UINT: + case PixelFormat::S8_UINT_D24_UNORM: UNIMPLEMENTED_IF(x_source != SwizzleSource::R && x_source != SwizzleSource::G); glTextureParameteri(view.handle, GL_DEPTH_STENCIL_TEXTURE_MODE, GetComponent(format, x_source == SwizzleSource::R)); @@ -520,10 +532,12 @@ OGLTextureView CachedSurfaceView::CreateTextureView() const { return texture_view; } -TextureCacheOpenGL::TextureCacheOpenGL(Core::System& system, - VideoCore::RasterizerInterface& rasterizer, - const Device& device, StateTracker& state_tracker) - : TextureCacheBase{system, rasterizer, device.HasASTC()}, state_tracker{state_tracker} { +TextureCacheOpenGL::TextureCacheOpenGL(VideoCore::RasterizerInterface& rasterizer, + Tegra::Engines::Maxwell3D& maxwell3d, + Tegra::MemoryManager& gpu_memory, const Device& device, + StateTracker& state_tracker_) + : TextureCacheBase{rasterizer, maxwell3d, gpu_memory, device.HasASTC()}, state_tracker{ + state_tracker_} { src_framebuffer.Create(); dst_framebuffer.Create(); } diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index bfc4ddf5d..7787134fc 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -90,7 +90,7 @@ public: Tegra::Texture::SwizzleSource z_source, Tegra::Texture::SwizzleSource w_source); - void DecorateViewName(GPUVAddr gpu_addr, std::string prefix); + void DecorateViewName(GPUVAddr gpu_addr, const std::string& prefix); void MarkAsModified(u64 tick) { surface.MarkAsModified(true, tick); @@ -129,8 +129,10 @@ private: class TextureCacheOpenGL final : public TextureCacheBase { public: - explicit TextureCacheOpenGL(Core::System& system, VideoCore::RasterizerInterface& rasterizer, - const Device& device, StateTracker& state_tracker); + explicit TextureCacheOpenGL(VideoCore::RasterizerInterface& rasterizer, + Tegra::Engines::Maxwell3D& maxwell3d, + Tegra::MemoryManager& gpu_memory, const Device& device, + StateTracker& state_tracker); ~TextureCacheOpenGL(); protected: diff --git a/src/video_core/renderer_opengl/maxwell_to_gl.h b/src/video_core/renderer_opengl/maxwell_to_gl.h index fe9bd4b5a..a8be2aa37 100644 --- a/src/video_core/renderer_opengl/maxwell_to_gl.h +++ b/src/video_core/renderer_opengl/maxwell_to_gl.h @@ -47,6 +47,8 @@ inline GLenum VertexFormat(Maxwell::VertexAttribute attrib) { return GL_UNSIGNED_INT; case Maxwell::VertexAttribute::Size::Size_10_10_10_2: return GL_UNSIGNED_INT_2_10_10_10_REV; + default: + break; } break; case Maxwell::VertexAttribute::Type::SignedNorm: @@ -70,6 +72,8 @@ inline GLenum VertexFormat(Maxwell::VertexAttribute attrib) { return GL_INT; case Maxwell::VertexAttribute::Size::Size_10_10_10_2: return GL_INT_2_10_10_10_REV; + default: + break; } break; case Maxwell::VertexAttribute::Type::Float: @@ -84,6 +88,8 @@ inline GLenum VertexFormat(Maxwell::VertexAttribute attrib) { case Maxwell::VertexAttribute::Size::Size_32_32_32: case Maxwell::VertexAttribute::Size::Size_32_32_32_32: return GL_FLOAT; + default: + break; } break; } diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index e66cdc083..2ccca1993 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -21,6 +21,8 @@ #include "core/perf_stats.h" #include "core/settings.h" #include "core/telemetry_session.h" +#include "video_core/host_shaders/opengl_present_frag.h" +#include "video_core/host_shaders/opengl_present_vert.h" #include "video_core/morton.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_shader_manager.h" @@ -30,60 +32,6 @@ namespace OpenGL { namespace { -constexpr std::size_t SWAP_CHAIN_SIZE = 3; - -struct Frame { - u32 width{}; /// Width of the frame (to detect resize) - u32 height{}; /// Height of the frame - bool color_reloaded{}; /// Texture attachment was recreated (ie: resized) - OpenGL::OGLRenderbuffer color{}; /// Buffer shared between the render/present FBO - OpenGL::OGLFramebuffer render{}; /// FBO created on the render thread - OpenGL::OGLFramebuffer present{}; /// FBO created on the present thread - GLsync render_fence{}; /// Fence created on the render thread - GLsync present_fence{}; /// Fence created on the presentation thread - bool is_srgb{}; /// Framebuffer is sRGB or RGB -}; - -constexpr char VERTEX_SHADER[] = R"( -#version 430 core - -out gl_PerVertex { - vec4 gl_Position; -}; - -layout (location = 0) in vec2 vert_position; -layout (location = 1) in vec2 vert_tex_coord; -layout (location = 0) out vec2 frag_tex_coord; - -// This is a truncated 3x3 matrix for 2D transformations: -// The upper-left 2x2 submatrix performs scaling/rotation/mirroring. -// The third column performs translation. -// The third row could be used for projection, which we don't need in 2D. It hence is assumed to -// implicitly be [0, 0, 1] -layout (location = 0) uniform mat3x2 modelview_matrix; - -void main() { - // Multiply input position by the rotscale part of the matrix and then manually translate by - // the last column. This is equivalent to using a full 3x3 matrix and expanding the vector - // to `vec3(vert_position.xy, 1.0)` - gl_Position = vec4(mat2(modelview_matrix) * vert_position + modelview_matrix[2], 0.0, 1.0); - frag_tex_coord = vert_tex_coord; -} -)"; - -constexpr char FRAGMENT_SHADER[] = R"( -#version 430 core - -layout (location = 0) in vec2 frag_tex_coord; -layout (location = 0) out vec4 color; - -layout (binding = 0) uniform sampler2D color_texture; - -void main() { - color = vec4(texture(color_texture, frag_tex_coord).rgb, 1.0f); -} -)"; - constexpr GLint PositionLocation = 0; constexpr GLint TexCoordLocation = 1; constexpr GLint ModelViewMatrixLocation = 0; @@ -96,24 +44,6 @@ struct ScreenRectVertex { std::array<GLfloat, 2> tex_coord; }; -/// Returns true if any debug tool is attached -bool HasDebugTool() { - const bool nsight = std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED"); - if (nsight) { - return true; - } - - GLint num_extensions; - glGetIntegerv(GL_NUM_EXTENSIONS, &num_extensions); - for (GLuint index = 0; index < static_cast<GLuint>(num_extensions); ++index) { - const auto name = reinterpret_cast<const char*>(glGetStringi(GL_EXTENSIONS, index)); - if (!std::strcmp(name, "GL_EXT_debug_tool")) { - return true; - } - } - return false; -} - /** * Defines a 1:1 pixel ortographic projection matrix with (0,0) on the top-left * corner and (width, height) on the lower-bottom. @@ -197,132 +127,15 @@ void APIENTRY DebugHandler(GLenum source, GLenum type, GLuint id, GLenum severit } // Anonymous namespace -/** - * For smooth Vsync rendering, we want to always present the latest frame that the core generates, - * but also make sure that rendering happens at the pace that the frontend dictates. This is a - * helper class that the renderer uses to sync frames between the render thread and the presentation - * thread - */ -class FrameMailbox { -public: - std::mutex swap_chain_lock; - std::condition_variable present_cv; - std::array<Frame, SWAP_CHAIN_SIZE> swap_chain{}; - std::queue<Frame*> free_queue; - std::deque<Frame*> present_queue; - Frame* previous_frame{}; - - FrameMailbox() { - for (auto& frame : swap_chain) { - free_queue.push(&frame); - } - } - - ~FrameMailbox() { - // lock the mutex and clear out the present and free_queues and notify any people who are - // blocked to prevent deadlock on shutdown - std::scoped_lock lock{swap_chain_lock}; - std::queue<Frame*>().swap(free_queue); - present_queue.clear(); - present_cv.notify_all(); - } - - void ReloadPresentFrame(Frame* frame, u32 height, u32 width) { - frame->present.Release(); - frame->present.Create(); - GLint previous_draw_fbo{}; - glGetIntegerv(GL_DRAW_FRAMEBUFFER_BINDING, &previous_draw_fbo); - glBindFramebuffer(GL_FRAMEBUFFER, frame->present.handle); - glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, - frame->color.handle); - if (glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) { - LOG_CRITICAL(Render_OpenGL, "Failed to recreate present FBO!"); - } - glBindFramebuffer(GL_DRAW_FRAMEBUFFER, previous_draw_fbo); - frame->color_reloaded = false; - } - - void ReloadRenderFrame(Frame* frame, u32 width, u32 height) { - // Recreate the color texture attachment - frame->color.Release(); - frame->color.Create(); - const GLenum internal_format = frame->is_srgb ? GL_SRGB8 : GL_RGB8; - glNamedRenderbufferStorage(frame->color.handle, internal_format, width, height); - - // Recreate the FBO for the render target - frame->render.Release(); - frame->render.Create(); - glBindFramebuffer(GL_FRAMEBUFFER, frame->render.handle); - glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, - frame->color.handle); - if (glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) { - LOG_CRITICAL(Render_OpenGL, "Failed to recreate render FBO!"); - } - - frame->width = width; - frame->height = height; - frame->color_reloaded = true; - } - - Frame* GetRenderFrame() { - std::unique_lock lock{swap_chain_lock}; - - // If theres no free frames, we will reuse the oldest render frame - if (free_queue.empty()) { - auto frame = present_queue.back(); - present_queue.pop_back(); - return frame; - } - - Frame* frame = free_queue.front(); - free_queue.pop(); - return frame; - } - - void ReleaseRenderFrame(Frame* frame) { - std::unique_lock lock{swap_chain_lock}; - present_queue.push_front(frame); - present_cv.notify_one(); - } - - Frame* TryGetPresentFrame(int timeout_ms) { - std::unique_lock lock{swap_chain_lock}; - // wait for new entries in the present_queue - present_cv.wait_for(lock, std::chrono::milliseconds(timeout_ms), - [&] { return !present_queue.empty(); }); - if (present_queue.empty()) { - // timed out waiting for a frame to draw so return the previous frame - return previous_frame; - } - - // free the previous frame and add it back to the free queue - if (previous_frame) { - free_queue.push(previous_frame); - } - - // the newest entries are pushed to the front of the queue - Frame* frame = present_queue.front(); - present_queue.pop_front(); - // remove all old entries from the present queue and move them back to the free_queue - for (auto f : present_queue) { - free_queue.push(f); - } - present_queue.clear(); - previous_frame = frame; - return frame; - } -}; - -RendererOpenGL::RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::System& system, - Core::Frontend::GraphicsContext& context) - : RendererBase{emu_window}, emu_window{emu_window}, system{system}, context{context}, - program_manager{device}, has_debug_tool{HasDebugTool()} {} +RendererOpenGL::RendererOpenGL(Core::TelemetrySession& telemetry_session_, + Core::Frontend::EmuWindow& emu_window_, + Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_, + std::unique_ptr<Core::Frontend::GraphicsContext> context) + : RendererBase{emu_window_, std::move(context)}, telemetry_session{telemetry_session_}, + emu_window{emu_window_}, cpu_memory{cpu_memory_}, gpu{gpu_}, program_manager{device} {} RendererOpenGL::~RendererOpenGL() = default; -MICROPROFILE_DEFINE(OpenGL_RenderFrame, "OpenGL", "Render Frame", MP_RGB(128, 128, 64)); -MICROPROFILE_DEFINE(OpenGL_WaitPresent, "OpenGL", "Wait For Present", MP_RGB(128, 128, 128)); - void RendererOpenGL::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { if (!framebuffer) { return; @@ -331,79 +144,34 @@ void RendererOpenGL::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { PrepareRendertarget(framebuffer); RenderScreenshot(); - Frame* frame; - { - MICROPROFILE_SCOPE(OpenGL_WaitPresent); - - frame = frame_mailbox->GetRenderFrame(); - - // Clean up sync objects before drawing + glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0); + DrawScreen(emu_window.GetFramebufferLayout()); - // INTEL driver workaround. We can't delete the previous render sync object until we are - // sure that the presentation is done - if (frame->present_fence) { - glClientWaitSync(frame->present_fence, 0, GL_TIMEOUT_IGNORED); - } - - // delete the draw fence if the frame wasn't presented - if (frame->render_fence) { - glDeleteSync(frame->render_fence); - frame->render_fence = 0; - } - - // wait for the presentation to be done - if (frame->present_fence) { - glWaitSync(frame->present_fence, 0, GL_TIMEOUT_IGNORED); - glDeleteSync(frame->present_fence); - frame->present_fence = 0; - } - } + ++m_current_frame; - { - MICROPROFILE_SCOPE(OpenGL_RenderFrame); - const auto& layout = render_window.GetFramebufferLayout(); - - // Recreate the frame if the size of the window has changed - if (layout.width != frame->width || layout.height != frame->height || - screen_info.display_srgb != frame->is_srgb) { - LOG_DEBUG(Render_OpenGL, "Reloading render frame"); - frame->is_srgb = screen_info.display_srgb; - frame_mailbox->ReloadRenderFrame(frame, layout.width, layout.height); - } - glBindFramebuffer(GL_DRAW_FRAMEBUFFER, frame->render.handle); - DrawScreen(layout); - // Create a fence for the frontend to wait on and swap this frame to OffTex - frame->render_fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0); - glFlush(); - frame_mailbox->ReleaseRenderFrame(frame); - m_current_frame++; - rasterizer->TickFrame(); - } + rasterizer->TickFrame(); render_window.PollEvents(); - if (has_debug_tool) { - glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0); - Present(0); - context.SwapBuffers(); - } + context->SwapBuffers(); } void RendererOpenGL::PrepareRendertarget(const Tegra::FramebufferConfig* framebuffer) { - if (framebuffer) { - // If framebuffer is provided, reload it from memory to a texture - if (screen_info.texture.width != static_cast<GLsizei>(framebuffer->width) || - screen_info.texture.height != static_cast<GLsizei>(framebuffer->height) || - screen_info.texture.pixel_format != framebuffer->pixel_format || - gl_framebuffer_data.empty()) { - // Reallocate texture if the framebuffer size has changed. - // This is expected to not happen very often and hence should not be a - // performance problem. - ConfigureFramebufferTexture(screen_info.texture, *framebuffer); - } - - // Load the framebuffer from memory, draw it to the screen, and swap buffers - LoadFBToScreenInfo(*framebuffer); + if (!framebuffer) { + return; + } + // If framebuffer is provided, reload it from memory to a texture + if (screen_info.texture.width != static_cast<GLsizei>(framebuffer->width) || + screen_info.texture.height != static_cast<GLsizei>(framebuffer->height) || + screen_info.texture.pixel_format != framebuffer->pixel_format || + gl_framebuffer_data.empty()) { + // Reallocate texture if the framebuffer size has changed. + // This is expected to not happen very often and hence should not be a + // performance problem. + ConfigureFramebufferTexture(screen_info.texture, *framebuffer); } + + // Load the framebuffer from memory, draw it to the screen, and swap buffers + LoadFBToScreenInfo(*framebuffer); } void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuffer) { @@ -423,7 +191,7 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf VideoCore::Surface::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format)}; const u32 bytes_per_pixel{VideoCore::Surface::GetBytesPerPixel(pixel_format)}; const u64 size_in_bytes{framebuffer.stride * framebuffer.height * bytes_per_pixel}; - u8* const host_ptr{system.Memory().GetPointer(framebuffer_addr)}; + u8* const host_ptr{cpu_memory.GetPointer(framebuffer_addr)}; rasterizer->FlushRegion(ToCacheAddr(host_ptr), size_in_bytes); // TODO(Rodrigo): Read this from HLE @@ -453,17 +221,15 @@ void RendererOpenGL::LoadColorToActiveGLTexture(u8 color_r, u8 color_g, u8 color } void RendererOpenGL::InitOpenGLObjects() { - frame_mailbox = std::make_unique<FrameMailbox>(); - 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(VERTEX_SHADER, GL_VERTEX_SHADER); + vertex_shader.Create(HostShaders::OPENGL_PRESENT_VERT, GL_VERTEX_SHADER); OGLShader fragment_shader; - fragment_shader.Create(FRAGMENT_SHADER, GL_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); @@ -508,18 +274,18 @@ void RendererOpenGL::AddTelemetryFields() { LOG_INFO(Render_OpenGL, "GL_VENDOR: {}", gpu_vendor); LOG_INFO(Render_OpenGL, "GL_RENDERER: {}", gpu_model); - auto& telemetry_session = system.TelemetrySession(); - telemetry_session.AddField(Telemetry::FieldType::UserSystem, "GPU_Vendor", gpu_vendor); - telemetry_session.AddField(Telemetry::FieldType::UserSystem, "GPU_Model", gpu_model); - telemetry_session.AddField(Telemetry::FieldType::UserSystem, "GPU_OpenGL_Version", gl_version); + constexpr auto user_system = Common::Telemetry::FieldType::UserSystem; + telemetry_session.AddField(user_system, "GPU_Vendor", gpu_vendor); + telemetry_session.AddField(user_system, "GPU_Model", gpu_model); + telemetry_session.AddField(user_system, "GPU_OpenGL_Version", gl_version); } void RendererOpenGL::CreateRasterizer() { if (rasterizer) { return; } - rasterizer = std::make_unique<RasterizerOpenGL>(system, emu_window, device, screen_info, - program_manager, state_tracker); + rasterizer = std::make_unique<RasterizerOpenGL>(emu_window, gpu, cpu_memory, device, + screen_info, program_manager, state_tracker); } void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture, @@ -535,12 +301,12 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture, GLint internal_format; switch (framebuffer.pixel_format) { - case Tegra::FramebufferConfig::PixelFormat::ABGR8: + case Tegra::FramebufferConfig::PixelFormat::A8B8G8R8_UNORM: internal_format = GL_RGBA8; texture.gl_format = GL_RGBA; texture.gl_type = GL_UNSIGNED_INT_8_8_8_8_REV; break; - case Tegra::FramebufferConfig::PixelFormat::RGB565: + case Tegra::FramebufferConfig::PixelFormat::RGB565_UNORM: internal_format = GL_RGB565; texture.gl_format = GL_RGB; texture.gl_type = GL_UNSIGNED_SHORT_5_6_5; @@ -682,51 +448,6 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { program_manager.RestoreGuestPipeline(); } -bool RendererOpenGL::TryPresent(int timeout_ms) { - if (has_debug_tool) { - LOG_DEBUG(Render_OpenGL, - "Skipping presentation because we are presenting on the main context"); - return false; - } - return Present(timeout_ms); -} - -bool RendererOpenGL::Present(int timeout_ms) { - const auto& layout = render_window.GetFramebufferLayout(); - auto frame = frame_mailbox->TryGetPresentFrame(timeout_ms); - if (!frame) { - LOG_DEBUG(Render_OpenGL, "TryGetPresentFrame returned no frame to present"); - return false; - } - - // Clearing before a full overwrite of a fbo can signal to drivers that they can avoid a - // readback since we won't be doing any blending - glClear(GL_COLOR_BUFFER_BIT); - - // Recreate the presentation FBO if the color attachment was changed - if (frame->color_reloaded) { - LOG_DEBUG(Render_OpenGL, "Reloading present frame"); - frame_mailbox->ReloadPresentFrame(frame, layout.width, layout.height); - } - glWaitSync(frame->render_fence, 0, GL_TIMEOUT_IGNORED); - // INTEL workaround. - // Normally we could just delete the draw fence here, but due to driver bugs, we can just delete - // it on the emulation thread without too much penalty - // glDeleteSync(frame.render_sync); - // frame.render_sync = 0; - - glBindFramebuffer(GL_READ_FRAMEBUFFER, frame->present.handle); - glBlitFramebuffer(0, 0, frame->width, frame->height, 0, 0, layout.width, layout.height, - GL_COLOR_BUFFER_BIT, GL_LINEAR); - - // Insert fence for the main thread to block on - frame->present_fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0); - glFlush(); - - glBindFramebuffer(GL_READ_FRAMEBUFFER, 0); - return true; -} - void RendererOpenGL::RenderScreenshot() { if (!renderer_settings.screenshot_requested) { return; @@ -741,7 +462,7 @@ void RendererOpenGL::RenderScreenshot() { screenshot_framebuffer.Create(); glBindFramebuffer(GL_FRAMEBUFFER, screenshot_framebuffer.handle); - Layout::FramebufferLayout layout{renderer_settings.screenshot_framebuffer_layout}; + const Layout::FramebufferLayout layout{renderer_settings.screenshot_framebuffer_layout}; GLuint renderbuffer; glGenRenderbuffers(1, &renderbuffer); diff --git a/src/video_core/renderer_opengl/renderer_opengl.h b/src/video_core/renderer_opengl/renderer_opengl.h index 8b18d32e6..9ef181f95 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.h +++ b/src/video_core/renderer_opengl/renderer_opengl.h @@ -16,16 +16,25 @@ namespace Core { class System; -} +class TelemetrySession; +} // namespace Core namespace Core::Frontend { class EmuWindow; } +namespace Core::Memory { +class Memory; +} + namespace Layout { struct FramebufferLayout; } +namespace Tegra { +class GPU; +} + namespace OpenGL { /// Structure used for storing information about the textures for the Switch screen @@ -46,24 +55,17 @@ struct ScreenInfo { TextureInfo texture; }; -struct PresentationTexture { - u32 width = 0; - u32 height = 0; - OGLTexture texture; -}; - -class FrameMailbox; - class RendererOpenGL final : public VideoCore::RendererBase { public: - explicit RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::System& system, - Core::Frontend::GraphicsContext& context); + explicit RendererOpenGL(Core::TelemetrySession& telemetry_session, + Core::Frontend::EmuWindow& emu_window, Core::Memory::Memory& cpu_memory, + Tegra::GPU& gpu, + std::unique_ptr<Core::Frontend::GraphicsContext> context); ~RendererOpenGL() override; bool Init() override; void ShutDown() override; void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override; - bool TryPresent(int timeout_ms) override; private: /// Initializes the OpenGL state and creates persistent objects. @@ -91,14 +93,13 @@ private: void PrepareRendertarget(const Tegra::FramebufferConfig* framebuffer); - bool Present(int timeout_ms); - + Core::TelemetrySession& telemetry_session; Core::Frontend::EmuWindow& emu_window; - Core::System& system; - Core::Frontend::GraphicsContext& context; - const Device device; + Core::Memory::Memory& cpu_memory; + Tegra::GPU& gpu; - StateTracker state_tracker{system}; + const Device device; + StateTracker state_tracker{gpu}; // OpenGL object IDs OGLBuffer vertex_buffer; @@ -120,13 +121,8 @@ private: std::vector<u8> gl_framebuffer_data; /// Used for transforming the framebuffer orientation - Tegra::FramebufferConfig::TransformFlags framebuffer_transform_flags; + Tegra::FramebufferConfig::TransformFlags framebuffer_transform_flags{}; Common::Rectangle<int> framebuffer_crop_rect; - - /// Frame presentation mailbox - std::unique_ptr<FrameMailbox> frame_mailbox; - - bool has_debug_tool = false; }; } // namespace OpenGL |