diff options
Diffstat (limited to 'src/video_core')
30 files changed, 191 insertions, 83 deletions
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 58a45ab67..b5ed3380f 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -115,7 +115,34 @@ void BufferCache<P>::WriteMemory(VAddr cpu_addr, u64 size) { template <class P> void BufferCache<P>::CachedWriteMemory(VAddr cpu_addr, u64 size) { - memory_tracker.CachedCpuWrite(cpu_addr, size); + const bool is_dirty = IsRegionRegistered(cpu_addr, size); + if (!is_dirty) { + return; + } + VAddr aligned_start = Common::AlignDown(cpu_addr, YUZU_PAGESIZE); + VAddr aligned_end = Common::AlignUp(cpu_addr + size, YUZU_PAGESIZE); + if (!IsRegionGpuModified(aligned_start, aligned_end - aligned_start)) { + WriteMemory(cpu_addr, size); + return; + } + + tmp_buffer.resize_destructive(size); + cpu_memory.ReadBlockUnsafe(cpu_addr, tmp_buffer.data(), size); + + InlineMemoryImplementation(cpu_addr, size, tmp_buffer); +} + +template <class P> +bool BufferCache<P>::OnCPUWrite(VAddr cpu_addr, u64 size) { + const bool is_dirty = IsRegionRegistered(cpu_addr, size); + if (!is_dirty) { + return false; + } + if (memory_tracker.IsRegionGpuModified(cpu_addr, size)) { + return true; + } + WriteMemory(cpu_addr, size); + return false; } template <class P> @@ -1553,6 +1580,14 @@ bool BufferCache<P>::InlineMemory(VAddr dest_address, size_t copy_size, return false; } + InlineMemoryImplementation(dest_address, copy_size, inlined_buffer); + + return true; +} + +template <class P> +void BufferCache<P>::InlineMemoryImplementation(VAddr dest_address, size_t copy_size, + std::span<const u8> inlined_buffer) { const IntervalType subtract_interval{dest_address, dest_address + copy_size}; ClearDownload(subtract_interval); common_ranges.subtract(subtract_interval); @@ -1574,8 +1609,6 @@ bool BufferCache<P>::InlineMemory(VAddr dest_address, size_t copy_size, } else { buffer.ImmediateUpload(buffer.Offset(dest_address), inlined_buffer.first(copy_size)); } - - return true; } template <class P> diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h index fe6068cfe..460fc7551 100644 --- a/src/video_core/buffer_cache/buffer_cache_base.h +++ b/src/video_core/buffer_cache/buffer_cache_base.h @@ -245,6 +245,8 @@ public: void CachedWriteMemory(VAddr cpu_addr, u64 size); + bool OnCPUWrite(VAddr cpu_addr, u64 size); + void DownloadMemory(VAddr cpu_addr, u64 size); std::optional<VideoCore::RasterizerDownloadArea> GetFlushArea(VAddr cpu_addr, u64 size); @@ -543,6 +545,9 @@ private: void ClearDownload(IntervalType subtract_interval); + void InlineMemoryImplementation(VAddr dest_address, size_t copy_size, + std::span<const u8> inlined_buffer); + VideoCore::RasterizerInterface& rasterizer; Core::Memory::Memory& cpu_memory; diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index a290d6ea7..f8598fd98 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -174,8 +174,7 @@ void MaxwellDMA::CopyBlockLinearToPitch() { src_operand.address = regs.offset_in; DMA::BufferOperand dst_operand; - u32 abs_pitch_out = std::abs(static_cast<s32>(regs.pitch_out)); - dst_operand.pitch = abs_pitch_out; + dst_operand.pitch = static_cast<u32>(std::abs(regs.pitch_out)); dst_operand.width = regs.line_length_in; dst_operand.height = regs.line_count; dst_operand.address = regs.offset_out; @@ -222,7 +221,7 @@ void MaxwellDMA::CopyBlockLinearToPitch() { const size_t src_size = CalculateSize(true, bytes_per_pixel, width, height, depth, block_height, block_depth); - const size_t dst_size = static_cast<size_t>(abs_pitch_out) * regs.line_count; + const size_t dst_size = dst_operand.pitch * regs.line_count; read_buffer.resize_destructive(src_size); write_buffer.resize_destructive(dst_size); @@ -231,7 +230,7 @@ void MaxwellDMA::CopyBlockLinearToPitch() { UnswizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, width, height, depth, x_offset, src_params.origin.y, x_elements, regs.line_count, block_height, block_depth, - abs_pitch_out); + dst_operand.pitch); memory_manager.WriteBlockCached(regs.offset_out, write_buffer.data(), dst_size); } diff --git a/src/video_core/fence_manager.h b/src/video_core/fence_manager.h index 35d699bbf..ab20ff30f 100644 --- a/src/video_core/fence_manager.h +++ b/src/video_core/fence_manager.h @@ -69,7 +69,6 @@ public: } void SignalFence(std::function<void()>&& func) { - rasterizer.InvalidateGPUCache(); bool delay_fence = Settings::IsGPULevelHigh(); if constexpr (!can_async_check) { TryReleasePendingFences<false>(); @@ -96,6 +95,7 @@ public: guard.unlock(); cv.notify_all(); } + rasterizer.InvalidateGPUCache(); } void SignalSyncPoint(u32 value) { diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index db385076d..c192e33b2 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -95,7 +95,9 @@ struct GPU::Impl { /// Synchronizes CPU writes with Host GPU memory. void InvalidateGPUCache() { - rasterizer->InvalidateGPUCache(); + std::function<void(VAddr, size_t)> callback_writes( + [this](VAddr address, size_t size) { rasterizer->OnCacheInvalidation(address, size); }); + system.GatherGPUDirtyMemory(callback_writes); } /// Signal the ending of command list. @@ -299,6 +301,10 @@ struct GPU::Impl { gpu_thread.InvalidateRegion(addr, size); } + bool OnCPUWrite(VAddr addr, u64 size) { + return rasterizer->OnCPUWrite(addr, size); + } + /// Notify rasterizer that any caches of the specified region should be flushed and invalidated void FlushAndInvalidateRegion(VAddr addr, u64 size) { gpu_thread.FlushAndInvalidateRegion(addr, size); @@ -561,6 +567,10 @@ void GPU::InvalidateRegion(VAddr addr, u64 size) { impl->InvalidateRegion(addr, size); } +bool GPU::OnCPUWrite(VAddr addr, u64 size) { + return impl->OnCPUWrite(addr, size); +} + void GPU::FlushAndInvalidateRegion(VAddr addr, u64 size) { impl->FlushAndInvalidateRegion(addr, size); } diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index e49c40cf2..ba2838b89 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -250,6 +250,10 @@ public: /// Notify rasterizer that any caches of the specified region should be invalidated void InvalidateRegion(VAddr addr, u64 size); + /// Notify rasterizer that CPU is trying to write this area. It returns true if the area is + /// sensible, false otherwise + bool OnCPUWrite(VAddr addr, u64 size); + /// Notify rasterizer that any caches of the specified region should be flushed and invalidated void FlushAndInvalidateRegion(VAddr addr, u64 size); diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp index 889144f38..2f0f9f593 100644 --- a/src/video_core/gpu_thread.cpp +++ b/src/video_core/gpu_thread.cpp @@ -47,7 +47,7 @@ static void RunThread(std::stop_token stop_token, Core::System& system, } else if (const auto* flush = std::get_if<FlushRegionCommand>(&next.data)) { rasterizer->FlushRegion(flush->addr, flush->size); } else if (const auto* invalidate = std::get_if<InvalidateRegionCommand>(&next.data)) { - rasterizer->OnCPUWrite(invalidate->addr, invalidate->size); + rasterizer->OnCacheInvalidation(invalidate->addr, invalidate->size); } else { ASSERT(false); } @@ -102,12 +102,12 @@ void ThreadManager::TickGPU() { } void ThreadManager::InvalidateRegion(VAddr addr, u64 size) { - rasterizer->OnCPUWrite(addr, size); + rasterizer->OnCacheInvalidation(addr, size); } void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) { // Skip flush on asynch mode, as FlushAndInvalidateRegion is not used for anything too important - rasterizer->OnCPUWrite(addr, size); + rasterizer->OnCacheInvalidation(addr, size); } u64 ThreadManager::PushCommand(CommandData&& command_data, bool block) { diff --git a/src/video_core/host1x/codecs/codec.cpp b/src/video_core/host1x/codecs/codec.cpp index cd6a3a9b8..da07a556f 100644 --- a/src/video_core/host1x/codecs/codec.cpp +++ b/src/video_core/host1x/codecs/codec.cpp @@ -290,7 +290,7 @@ void Codec::Decode() { return vp9_decoder->GetFrameBytes(); default: ASSERT(false); - return std::vector<u8>{}; + return std::span<const u8>{}; } }(); AVPacketPtr packet{av_packet_alloc(), AVPacketDeleter}; diff --git a/src/video_core/host1x/codecs/h264.cpp b/src/video_core/host1x/codecs/h264.cpp index ce827eb6c..862904e39 100644 --- a/src/video_core/host1x/codecs/h264.cpp +++ b/src/video_core/host1x/codecs/h264.cpp @@ -29,15 +29,15 @@ H264::H264(Host1x::Host1x& host1x_) : host1x{host1x_} {} H264::~H264() = default; -const std::vector<u8>& H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state, - bool is_first_frame) { +std::span<const u8> H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state, + bool is_first_frame) { H264DecoderContext context; host1x.MemoryManager().ReadBlock(state.picture_info_offset, &context, sizeof(H264DecoderContext)); const s64 frame_number = context.h264_parameter_set.frame_number.Value(); if (!is_first_frame && frame_number != 0) { - frame.resize(context.stream_len); + frame.resize_destructive(context.stream_len); host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size()); return frame; } @@ -135,14 +135,14 @@ const std::vector<u8>& H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegist for (s32 index = 0; index < 6; index++) { writer.WriteBit(true); std::span<const u8> matrix{context.weight_scale}; - writer.WriteScalingList(matrix, index * 16, 16); + writer.WriteScalingList(scan, matrix, index * 16, 16); } if (context.h264_parameter_set.transform_8x8_mode_flag) { for (s32 index = 0; index < 2; index++) { writer.WriteBit(true); std::span<const u8> matrix{context.weight_scale_8x8}; - writer.WriteScalingList(matrix, index * 64, 64); + writer.WriteScalingList(scan, matrix, index * 64, 64); } } @@ -188,8 +188,8 @@ void H264BitWriter::WriteBit(bool state) { WriteBits(state ? 1 : 0, 1); } -void H264BitWriter::WriteScalingList(std::span<const u8> list, s32 start, s32 count) { - static Common::ScratchBuffer<u8> scan{}; +void H264BitWriter::WriteScalingList(Common::ScratchBuffer<u8>& scan, std::span<const u8> list, + s32 start, s32 count) { scan.resize_destructive(count); if (count == 16) { std::memcpy(scan.data(), zig_zag_scan.data(), scan.size()); diff --git a/src/video_core/host1x/codecs/h264.h b/src/video_core/host1x/codecs/h264.h index 5cc86454e..d6b556322 100644 --- a/src/video_core/host1x/codecs/h264.h +++ b/src/video_core/host1x/codecs/h264.h @@ -5,9 +5,11 @@ #include <span> #include <vector> + #include "common/bit_field.h" #include "common/common_funcs.h" #include "common/common_types.h" +#include "common/scratch_buffer.h" #include "video_core/host1x/nvdec_common.h" namespace Tegra { @@ -37,7 +39,8 @@ public: /// Based on section 7.3.2.1.1.1 and Table 7-4 in the H.264 specification /// Writes the scaling matrices of the sream - void WriteScalingList(std::span<const u8> list, s32 start, s32 count); + void WriteScalingList(Common::ScratchBuffer<u8>& scan, std::span<const u8> list, s32 start, + s32 count); /// Return the bitstream as a vector. [[nodiscard]] std::vector<u8>& GetByteArray(); @@ -63,11 +66,12 @@ public: ~H264(); /// Compose the H264 frame for FFmpeg decoding - [[nodiscard]] const std::vector<u8>& ComposeFrame( - const Host1x::NvdecCommon::NvdecRegisters& state, bool is_first_frame = false); + [[nodiscard]] std::span<const u8> ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state, + bool is_first_frame = false); private: - std::vector<u8> frame; + Common::ScratchBuffer<u8> frame; + Common::ScratchBuffer<u8> scan; Host1x::Host1x& host1x; struct H264ParameterSet { diff --git a/src/video_core/host1x/codecs/vp8.cpp b/src/video_core/host1x/codecs/vp8.cpp index 28fb12cb8..ee6392ff9 100644 --- a/src/video_core/host1x/codecs/vp8.cpp +++ b/src/video_core/host1x/codecs/vp8.cpp @@ -12,7 +12,7 @@ VP8::VP8(Host1x::Host1x& host1x_) : host1x{host1x_} {} VP8::~VP8() = default; -const std::vector<u8>& VP8::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state) { +std::span<const u8> VP8::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state) { VP8PictureInfo info; host1x.MemoryManager().ReadBlock(state.picture_info_offset, &info, sizeof(VP8PictureInfo)); diff --git a/src/video_core/host1x/codecs/vp8.h b/src/video_core/host1x/codecs/vp8.h index 5bf07ecab..7926b73f3 100644 --- a/src/video_core/host1x/codecs/vp8.h +++ b/src/video_core/host1x/codecs/vp8.h @@ -4,10 +4,11 @@ #pragma once #include <array> -#include <vector> +#include <span> #include "common/common_funcs.h" #include "common/common_types.h" +#include "common/scratch_buffer.h" #include "video_core/host1x/nvdec_common.h" namespace Tegra { @@ -24,11 +25,11 @@ public: ~VP8(); /// Compose the VP8 frame for FFmpeg decoding - [[nodiscard]] const std::vector<u8>& ComposeFrame( + [[nodiscard]] std::span<const u8> ComposeFrame( const Host1x::NvdecCommon::NvdecRegisters& state); private: - std::vector<u8> frame; + Common::ScratchBuffer<u8> frame; Host1x::Host1x& host1x; struct VP8PictureInfo { diff --git a/src/video_core/host1x/codecs/vp9.cpp b/src/video_core/host1x/codecs/vp9.cpp index cf40c9012..306c3d0e8 100644 --- a/src/video_core/host1x/codecs/vp9.cpp +++ b/src/video_core/host1x/codecs/vp9.cpp @@ -3,6 +3,7 @@ #include <algorithm> // for std::copy #include <numeric> + #include "common/assert.h" #include "video_core/host1x/codecs/vp9.h" #include "video_core/host1x/host1x.h" diff --git a/src/video_core/host1x/codecs/vp9.h b/src/video_core/host1x/codecs/vp9.h index d4083e8d3..f1ed19508 100644 --- a/src/video_core/host1x/codecs/vp9.h +++ b/src/video_core/host1x/codecs/vp9.h @@ -4,9 +4,11 @@ #pragma once #include <array> +#include <span> #include <vector> #include "common/common_types.h" +#include "common/scratch_buffer.h" #include "common/stream.h" #include "video_core/host1x/codecs/vp9_types.h" #include "video_core/host1x/nvdec_common.h" @@ -128,8 +130,8 @@ public: return !current_frame_info.show_frame; } - /// Returns a const reference to the composed frame data. - [[nodiscard]] const std::vector<u8>& GetFrameBytes() const { + /// Returns a const span to the composed frame data. + [[nodiscard]] std::span<const u8> GetFrameBytes() const { return frame; } @@ -181,7 +183,7 @@ private: [[nodiscard]] VpxBitStreamWriter ComposeUncompressedHeader(); Host1x::Host1x& host1x; - std::vector<u8> frame; + Common::ScratchBuffer<u8> frame; std::array<s8, 4> loop_filter_ref_deltas{}; std::array<s8, 2> loop_filter_mode_deltas{}; diff --git a/src/video_core/host1x/codecs/vp9_types.h b/src/video_core/host1x/codecs/vp9_types.h index adad8ed7e..cc9b25690 100644 --- a/src/video_core/host1x/codecs/vp9_types.h +++ b/src/video_core/host1x/codecs/vp9_types.h @@ -5,6 +5,7 @@ #include <array> #include <vector> + #include "common/common_funcs.h" #include "common/common_types.h" diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index 7566a8c4e..cb8029a4f 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -109,7 +109,9 @@ public: } /// Notify rasterizer that any caches of the specified region are desync with guest - virtual void OnCPUWrite(VAddr addr, u64 size) = 0; + virtual void OnCacheInvalidation(VAddr addr, u64 size) = 0; + + virtual bool OnCPUWrite(VAddr addr, u64 size) = 0; /// Sync memory between guest and host. virtual void InvalidateGPUCache() = 0; diff --git a/src/video_core/renderer_null/null_rasterizer.cpp b/src/video_core/renderer_null/null_rasterizer.cpp index bf2ce4c49..92ecf6682 100644 --- a/src/video_core/renderer_null/null_rasterizer.cpp +++ b/src/video_core/renderer_null/null_rasterizer.cpp @@ -47,7 +47,10 @@ bool RasterizerNull::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheTyp return false; } void RasterizerNull::InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType) {} -void RasterizerNull::OnCPUWrite(VAddr addr, u64 size) {} +bool RasterizerNull::OnCPUWrite(VAddr addr, u64 size) { + return false; +} +void RasterizerNull::OnCacheInvalidation(VAddr addr, u64 size) {} VideoCore::RasterizerDownloadArea RasterizerNull::GetFlushArea(VAddr addr, u64 size) { VideoCore::RasterizerDownloadArea new_area{ .start_address = Common::AlignDown(addr, Core::Memory::YUZU_PAGESIZE), diff --git a/src/video_core/renderer_null/null_rasterizer.h b/src/video_core/renderer_null/null_rasterizer.h index a8d35d2c1..93b9a6971 100644 --- a/src/video_core/renderer_null/null_rasterizer.h +++ b/src/video_core/renderer_null/null_rasterizer.h @@ -53,7 +53,8 @@ public: VideoCommon::CacheType which = VideoCommon::CacheType::All) override; void InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override; - void OnCPUWrite(VAddr addr, u64 size) override; + void OnCacheInvalidation(VAddr addr, u64 size) override; + bool OnCPUWrite(VAddr addr, u64 size) override; VideoCore::RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) override; void InvalidateGPUCache() override; void UnmapMemory(VAddr addr, u64 size) override; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index edf527f2d..aadd6967c 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -485,12 +485,33 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size, VideoCommon::Cache } } -void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) { +bool RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) { + MICROPROFILE_SCOPE(OpenGL_CacheManagement); + if (addr == 0 || size == 0) { + return false; + } + + { + std::scoped_lock lock{buffer_cache.mutex}; + if (buffer_cache.OnCPUWrite(addr, size)) { + return true; + } + } + + { + std::scoped_lock lock{texture_cache.mutex}; + texture_cache.WriteMemory(addr, size); + } + + shader_cache.InvalidateRegion(addr, size); + return false; +} + +void RasterizerOpenGL::OnCacheInvalidation(VAddr addr, u64 size) { MICROPROFILE_SCOPE(OpenGL_CacheManagement); if (addr == 0 || size == 0) { return; } - shader_cache.OnCPUWrite(addr, size); { std::scoped_lock lock{texture_cache.mutex}; texture_cache.WriteMemory(addr, size); @@ -499,15 +520,11 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) { std::scoped_lock lock{buffer_cache.mutex}; buffer_cache.CachedWriteMemory(addr, size); } + shader_cache.InvalidateRegion(addr, size); } void RasterizerOpenGL::InvalidateGPUCache() { - MICROPROFILE_SCOPE(OpenGL_CacheManagement); - shader_cache.SyncGuestHost(); - { - std::scoped_lock lock{buffer_cache.mutex}; - buffer_cache.FlushCachedWrites(); - } + gpu.InvalidateGPUCache(); } void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) { @@ -519,7 +536,7 @@ void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) { std::scoped_lock lock{buffer_cache.mutex}; buffer_cache.WriteMemory(addr, size); } - shader_cache.OnCPUWrite(addr, size); + shader_cache.OnCacheInvalidation(addr, size); } void RasterizerOpenGL::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) { diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index a73ad15c1..8eda2ddba 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -98,7 +98,8 @@ public: VideoCore::RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) override; void InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override; - void OnCPUWrite(VAddr addr, u64 size) override; + void OnCacheInvalidation(VAddr addr, u64 size) override; + bool OnCPUWrite(VAddr addr, u64 size) override; void InvalidateGPUCache() override; void UnmapMemory(VAddr addr, u64 size) override; void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override; diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index b72f95235..51df18ec3 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -591,7 +591,7 @@ void BufferCacheRuntime::ReserveNullBuffer() { .flags = 0, .size = 4, .usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT | - VK_BUFFER_USAGE_TRANSFER_DST_BIT, + VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT, .sharingMode = VK_SHARING_MODE_EXCLUSIVE, .queueFamilyIndexCount = 0, .pQueueFamilyIndices = nullptr, @@ -599,7 +599,6 @@ void BufferCacheRuntime::ReserveNullBuffer() { if (device.IsExtTransformFeedbackSupported()) { create_info.usage |= VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_BUFFER_BIT_EXT; } - create_info.usage |= VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT; null_buffer = memory_allocator.CreateBuffer(create_info, MemoryUsage::DeviceLocal); if (device.HasDebuggingToolAttached()) { null_buffer.SetObjectNameEXT("Null buffer"); diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index f7c0d939a..456bb040e 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -566,11 +566,32 @@ void RasterizerVulkan::InnerInvalidation(std::span<const std::pair<VAddr, std::s } } -void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) { +bool RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) { + if (addr == 0 || size == 0) { + return false; + } + + { + std::scoped_lock lock{buffer_cache.mutex}; + if (buffer_cache.OnCPUWrite(addr, size)) { + return true; + } + } + + { + std::scoped_lock lock{texture_cache.mutex}; + texture_cache.WriteMemory(addr, size); + } + + pipeline_cache.InvalidateRegion(addr, size); + return false; +} + +void RasterizerVulkan::OnCacheInvalidation(VAddr addr, u64 size) { if (addr == 0 || size == 0) { return; } - pipeline_cache.OnCPUWrite(addr, size); + { std::scoped_lock lock{texture_cache.mutex}; texture_cache.WriteMemory(addr, size); @@ -579,14 +600,11 @@ void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) { std::scoped_lock lock{buffer_cache.mutex}; buffer_cache.CachedWriteMemory(addr, size); } + pipeline_cache.InvalidateRegion(addr, size); } void RasterizerVulkan::InvalidateGPUCache() { - pipeline_cache.SyncGuestHost(); - { - std::scoped_lock lock{buffer_cache.mutex}; - buffer_cache.FlushCachedWrites(); - } + gpu.InvalidateGPUCache(); } void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) { @@ -598,7 +616,7 @@ void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) { std::scoped_lock lock{buffer_cache.mutex}; buffer_cache.WriteMemory(addr, size); } - pipeline_cache.OnCPUWrite(addr, size); + pipeline_cache.OnCacheInvalidation(addr, size); } void RasterizerVulkan::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) { diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index b39710b3c..73257d964 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -96,7 +96,8 @@ public: void InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override; void InnerInvalidation(std::span<const std::pair<VAddr, std::size_t>> sequences) override; - void OnCPUWrite(VAddr addr, u64 size) override; + void OnCacheInvalidation(VAddr addr, u64 size) override; + bool OnCPUWrite(VAddr addr, u64 size) override; void InvalidateGPUCache() override; void UnmapMemory(VAddr addr, u64 size) override; void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override; diff --git a/src/video_core/shader_cache.cpp b/src/video_core/shader_cache.cpp index 4db948b6d..01701201d 100644 --- a/src/video_core/shader_cache.cpp +++ b/src/video_core/shader_cache.cpp @@ -24,7 +24,7 @@ void ShaderCache::InvalidateRegion(VAddr addr, size_t size) { RemovePendingShaders(); } -void ShaderCache::OnCPUWrite(VAddr addr, size_t size) { +void ShaderCache::OnCacheInvalidation(VAddr addr, size_t size) { std::scoped_lock lock{invalidation_mutex}; InvalidatePagesInRegion(addr, size); } diff --git a/src/video_core/shader_cache.h b/src/video_core/shader_cache.h index f3cc4c70b..de8e08002 100644 --- a/src/video_core/shader_cache.h +++ b/src/video_core/shader_cache.h @@ -62,7 +62,7 @@ public: /// @brief Unmarks a memory region as cached and marks it for removal /// @param addr Start address of the CPU write operation /// @param size Number of bytes of the CPU write operation - void OnCPUWrite(VAddr addr, size_t size); + void OnCacheInvalidation(VAddr addr, size_t size); /// @brief Flushes delayed removal operations void SyncGuestHost(); diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index d3f03a995..79f158db4 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -598,14 +598,6 @@ void TextureCache<P>::UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t siz [&](ImageId id, Image&) { deleted_images.push_back(id); }); for (const ImageId id : deleted_images) { Image& image = slot_images[id]; - if (True(image.flags & ImageFlagBits::CpuModified)) { - return; - } - image.flags |= ImageFlagBits::CpuModified; - if (True(image.flags & ImageFlagBits::Tracked)) { - UntrackImage(image, id); - } - /* if (True(image.flags & ImageFlagBits::Remapped)) { continue; } @@ -613,7 +605,6 @@ void TextureCache<P>::UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t siz if (True(image.flags & ImageFlagBits::Tracked)) { UntrackImage(image, id); } - */ } } @@ -874,11 +865,19 @@ void TextureCache<P>::PopAsyncFlushes() { template <class P> ImageId TextureCache<P>::DmaImageId(const Tegra::DMA::ImageOperand& operand, bool is_upload) { const ImageInfo dst_info(operand); - const ImageId image_id = FindDMAImage(dst_info, operand.address); - if (!image_id) { + const ImageId dst_id = FindDMAImage(dst_info, operand.address); + if (!dst_id) { + return NULL_IMAGE_ID; + } + auto& image = slot_images[dst_id]; + if (False(image.flags & ImageFlagBits::GpuModified)) { + // No need to waste time on an image that's synced with guest + return NULL_IMAGE_ID; + } + if (image.info.type == ImageType::e3D) { + // Don't accelerate 3D images. return NULL_IMAGE_ID; } - auto& image = slot_images[image_id]; if (!is_upload && !image.info.dma_downloaded) { // Force a full sync. image.info.dma_downloaded = true; @@ -888,7 +887,7 @@ ImageId TextureCache<P>::DmaImageId(const Tegra::DMA::ImageOperand& operand, boo if (!base) { return NULL_IMAGE_ID; } - return image_id; + return dst_id; } template <class P> diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 70436cf1c..421e71e5a 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -528,6 +528,14 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR } sets_per_pool = 64; + if (extensions.extended_dynamic_state3 && is_amd_driver && + properties.properties.driverVersion >= VK_MAKE_API_VERSION(0, 2, 0, 270)) { + LOG_WARNING(Render_Vulkan, + "AMD drivers after 23.5.2 have broken extendedDynamicState3ColorBlendEquation"); + features.extended_dynamic_state3.extendedDynamicState3ColorBlendEnable = false; + features.extended_dynamic_state3.extendedDynamicState3ColorBlendEquation = false; + dynamic_state3_blending = false; + } if (is_amd_driver) { // AMD drivers need a higher amount of Sets per Pool in certain circumstances like in XC2. sets_per_pool = 96; diff --git a/src/video_core/vulkan_common/vulkan_instance.cpp b/src/video_core/vulkan_common/vulkan_instance.cpp index 7624a9b32..6a294c1da 100644 --- a/src/video_core/vulkan_common/vulkan_instance.cpp +++ b/src/video_core/vulkan_common/vulkan_instance.cpp @@ -19,11 +19,9 @@ #include <windows.h> // ensure include order #include <vulkan/vulkan_win32.h> -#elif defined(__APPLE__) -#include <vulkan/vulkan_macos.h> #elif defined(__ANDROID__) #include <vulkan/vulkan_android.h> -#else +#elif !defined(__APPLE__) #include <X11/Xlib.h> #include <vulkan/vulkan_wayland.h> #include <vulkan/vulkan_xlib.h> @@ -68,7 +66,7 @@ namespace { break; #elif defined(__APPLE__) case Core::Frontend::WindowSystemType::Cocoa: - extensions.push_back(VK_MVK_MACOS_SURFACE_EXTENSION_NAME); + extensions.push_back(VK_EXT_METAL_SURFACE_EXTENSION_NAME); break; #elif defined(__ANDROID__) case Core::Frontend::WindowSystemType::Android: diff --git a/src/video_core/vulkan_common/vulkan_surface.cpp b/src/video_core/vulkan_common/vulkan_surface.cpp index c34599365..cfea4cd7b 100644 --- a/src/video_core/vulkan_common/vulkan_surface.cpp +++ b/src/video_core/vulkan_common/vulkan_surface.cpp @@ -11,11 +11,9 @@ #include <windows.h> // ensure include order #include <vulkan/vulkan_win32.h> -#elif defined(__APPLE__) -#include <vulkan/vulkan_macos.h> #elif defined(__ANDROID__) #include <vulkan/vulkan_android.h> -#else +#elif !defined(__APPLE__) #include <X11/Xlib.h> #include <vulkan/vulkan_wayland.h> #include <vulkan/vulkan_xlib.h> @@ -44,12 +42,13 @@ vk::SurfaceKHR CreateSurface( } #elif defined(__APPLE__) if (window_info.type == Core::Frontend::WindowSystemType::Cocoa) { - const VkMacOSSurfaceCreateInfoMVK mvk_ci{VK_STRUCTURE_TYPE_MACOS_SURFACE_CREATE_INFO_MVK, - nullptr, 0, window_info.render_surface}; - const auto vkCreateMacOSSurfaceMVK = reinterpret_cast<PFN_vkCreateMacOSSurfaceMVK>( - dld.vkGetInstanceProcAddr(*instance, "vkCreateMacOSSurfaceMVK")); - if (!vkCreateMacOSSurfaceMVK || - vkCreateMacOSSurfaceMVK(*instance, &mvk_ci, nullptr, &unsafe_surface) != VK_SUCCESS) { + const VkMetalSurfaceCreateInfoEXT macos_ci = { + .pLayer = static_cast<const CAMetalLayer*>(window_info.render_surface), + }; + const auto vkCreateMetalSurfaceEXT = reinterpret_cast<PFN_vkCreateMetalSurfaceEXT>( + dld.vkGetInstanceProcAddr(*instance, "vkCreateMetalSurfaceEXT")); + if (!vkCreateMetalSurfaceEXT || + vkCreateMetalSurfaceEXT(*instance, &macos_ci, nullptr, &unsafe_surface) != VK_SUCCESS) { LOG_ERROR(Render_Vulkan, "Failed to initialize Metal surface"); throw vk::Exception(VK_ERROR_INITIALIZATION_FAILED); } diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index b5e70fcd4..32bd75ad8 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -15,6 +15,8 @@ #define VK_NO_PROTOTYPES #ifdef _WIN32 #define VK_USE_PLATFORM_WIN32_KHR +#elif defined(__APPLE__) +#define VK_USE_PLATFORM_METAL_EXT #endif #include <vulkan/vulkan.h> |