From 1209d428f14ffc6f1c8b6049b0431c30b26138d6 Mon Sep 17 00:00:00 2001 From: ameerj <52414509+ameerj@users.noreply.github.com> Date: Thu, 22 Dec 2022 17:01:39 -0500 Subject: [PATCH] texture_cache: Use pre-allocated buffer for texture downloads --- src/video_core/texture_cache/texture_cache.h | 9 ++++++--- src/video_core/texture_cache/util.cpp | 11 ++++++----- src/video_core/texture_cache/util.h | 3 ++- 3 files changed, 14 insertions(+), 9 deletions(-) diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index fccf4316d..6d7d8226f 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -96,7 +96,8 @@ void TextureCache

::RunGarbageCollector() { const auto copies = FullDownloadCopies(image.info); image.DownloadMemory(map, copies); runtime.Finish(); - SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span); + SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span, + swizzle_data_buffer); } if (True(image.flags & ImageFlagBits::Tracked)) { UntrackImage(image, image_id); @@ -467,7 +468,8 @@ void TextureCache

::DownloadMemory(VAddr cpu_addr, size_t size) { const auto copies = FullDownloadCopies(image.info); image.DownloadMemory(map, copies); runtime.Finish(); - SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span); + SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span, + swizzle_data_buffer); } } @@ -678,7 +680,8 @@ void TextureCache

::PopAsyncFlushes() { for (const ImageId image_id : download_ids) { const ImageBase& image = slot_images[image_id]; const auto copies = FullDownloadCopies(image.info); - SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, download_span); + SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, download_span, + swizzle_data_buffer); download_map.offset += image.unswizzled_size_bytes; download_span = download_span.subspan(image.unswizzled_size_bytes); } diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp index 4488fa9da..7999a7f06 100644 --- a/src/video_core/texture_cache/util.cpp +++ b/src/video_core/texture_cache/util.cpp @@ -505,7 +505,7 @@ void SwizzlePitchLinearImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr void SwizzleBlockLinearImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr, const ImageInfo& info, const BufferImageCopy& copy, - std::span input) { + std::span input, std::vector& tmp_buffer) { const Extent3D size = info.size; const LevelInfo level_info = MakeLevelInfo(info); const Extent2D tile_size = DefaultBlockSize(info.format); @@ -534,8 +534,8 @@ void SwizzleBlockLinearImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr tile_size.height, info.tile_width_spacing); const size_t subresource_size = sizes[level]; - const auto dst_data = std::make_unique(subresource_size); - const std::span dst(dst_data.get(), subresource_size); + tmp_buffer.resize(subresource_size); + const std::span dst(tmp_buffer); for (s32 layer = 0; layer < info.resources.layers; ++layer) { const std::span src = input.subspan(host_offset); @@ -977,13 +977,14 @@ std::vector FullUploadSwizzles(const ImageInfo& info) { } void SwizzleImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr, const ImageInfo& info, - std::span copies, std::span memory) { + std::span copies, std::span memory, + std::vector& tmp_buffer) { const bool is_pitch_linear = info.type == ImageType::Linear; for (const BufferImageCopy& copy : copies) { if (is_pitch_linear) { SwizzlePitchLinearImage(gpu_memory, gpu_addr, info, copy, memory); } else { - SwizzleBlockLinearImage(gpu_memory, gpu_addr, info, copy, memory); + SwizzleBlockLinearImage(gpu_memory, gpu_addr, info, copy, memory, tmp_buffer); } } } diff --git a/src/video_core/texture_cache/util.h b/src/video_core/texture_cache/util.h index ddf0b3b06..2c991b4d2 100644 --- a/src/video_core/texture_cache/util.h +++ b/src/video_core/texture_cache/util.h @@ -77,7 +77,8 @@ void ConvertImage(std::span input, const ImageInfo& info, std::span FullUploadSwizzles(const ImageInfo& info); void SwizzleImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr, const ImageInfo& info, - std::span copies, std::span memory); + std::span copies, std::span memory, + std::vector& tmp_buffer); [[nodiscard]] bool IsBlockLinearSizeCompatible(const ImageInfo& new_info, const ImageInfo& overlap_info, u32 new_level,