From c6cac2ffaad4ac27f35cea25022d9c59c7ecfbf4 Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Sun, 30 Apr 2023 17:14:06 +0200 Subject: GPU: Add Reactive flushing --- src/video_core/renderer_opengl/gl_rasterizer.cpp | 23 +++++++++++++++++++++++ src/video_core/renderer_opengl/gl_rasterizer.h | 1 + 2 files changed, 24 insertions(+) (limited to 'src/video_core/renderer_opengl') diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 0089b4b27..3f07fe8bb 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -433,6 +433,29 @@ bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheT return false; } +VideoCore::RasterizerDownloadArea RasterizerOpenGL::GetFlushArea(VAddr addr, u64 size) { + { + std::scoped_lock lock{texture_cache.mutex}; + auto area = texture_cache.GetFlushArea(addr, size); + if (area) { + return *area; + } + } + { + std::scoped_lock lock{buffer_cache.mutex}; + auto area = buffer_cache.GetFlushArea(addr, size); + if (area) { + return *area; + } + } + VideoCore::RasterizerDownloadArea new_area{ + .start_address = Common::AlignDown(addr, Core::Memory::YUZU_PAGESIZE), + .end_address = Common::AlignUp(addr + size, Core::Memory::YUZU_PAGESIZE), + .preemtive = true, + }; + return new_area; +} + void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which) { MICROPROFILE_SCOPE(OpenGL_CacheManagement); if (addr == 0 || size == 0) { diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index ad6978bd0..410d8ffc5 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -95,6 +95,7 @@ public: VideoCommon::CacheType which = VideoCommon::CacheType::All) override; bool MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override; + 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; -- cgit v1.2.3 From 8014dd82594dbb40e13749203e67b21e8447733c Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Sun, 7 May 2023 23:34:52 +0200 Subject: Texture cache: Only force flush the dma downloads --- src/video_core/engines/maxwell_dma.cpp | 2 +- src/video_core/renderer_opengl/gl_rasterizer.cpp | 2 +- src/video_core/renderer_vulkan/vk_rasterizer.cpp | 2 +- src/video_core/texture_cache/texture_cache.h | 4 ++-- src/video_core/texture_cache/texture_cache_base.h | 2 +- 5 files changed, 6 insertions(+), 6 deletions(-) (limited to 'src/video_core/renderer_opengl') diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index f7400aac8..ebe5536de 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -288,7 +288,7 @@ void MaxwellDMA::CopyPitchToBlockLinear() { write_buffer.resize_destructive(dst_size); memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size); - memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size); + memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size); // If the input is linear and the output is tiled, swizzle the input and copy it over. SwizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, width, height, depth, x_offset, diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 3f07fe8bb..f5baa0f3c 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -1304,7 +1304,7 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand, const Tegra::DMA::ImageOperand& image_operand) { std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - const auto image_id = texture_cache.DmaImageId(image_operand); + const auto image_id = texture_cache.DmaImageId(image_operand, IS_IMAGE_UPLOAD); if (image_id == VideoCommon::NULL_IMAGE_ID) { return false; } diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 0bcc20544..628e1376f 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -793,7 +793,7 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand, const Tegra::DMA::ImageOperand& image_operand) { std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - const auto image_id = texture_cache.DmaImageId(image_operand); + const auto image_id = texture_cache.DmaImageId(image_operand, IS_IMAGE_UPLOAD); if (image_id == VideoCommon::NULL_IMAGE_ID) { return false; } diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index d49f3a7a0..e1198dcf8 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -811,7 +811,7 @@ void TextureCache

::PopAsyncFlushes() { } template -ImageId TextureCache

::DmaImageId(const Tegra::DMA::ImageOperand& operand) { +ImageId TextureCache

::DmaImageId(const Tegra::DMA::ImageOperand& operand, bool is_upload) { const ImageInfo dst_info(operand); const ImageId dst_id = FindDMAImage(dst_info, operand.address); if (!dst_id) { @@ -822,7 +822,7 @@ ImageId TextureCache

::DmaImageId(const Tegra::DMA::ImageOperand& operand) { // No need to waste time on an image that's synced with guest return NULL_IMAGE_ID; } - if (!image.info.dma_downloaded) { + if (!is_upload && !image.info.dma_downloaded) { // Force a full sync. image.info.dma_downloaded = true; return NULL_IMAGE_ID; diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 01f5ac588..0720494e5 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -207,7 +207,7 @@ public: /// Pop asynchronous downloads void PopAsyncFlushes(); - [[nodiscard]] ImageId DmaImageId(const Tegra::DMA::ImageOperand& operand); + [[nodiscard]] ImageId DmaImageId(const Tegra::DMA::ImageOperand& operand, bool is_upload); [[nodiscard]] std::pair DmaBufferImageCopy( const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand, -- cgit v1.2.3