texture_cache: Use pre-allocated buffer for texture downloads

src/video_core/texture_cache/texture_cache.h

@@ -96,7 +96,8 @@ void TextureCache<P>::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<P>::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<P>::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);
     }

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<const u8> input) {
+                             std::span<const u8> input, std::vector<u8>& 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<u8[]>(subresource_size);
-    const std::span<u8> dst(dst_data.get(), subresource_size);
+    tmp_buffer.resize(subresource_size);
+    const std::span<u8> dst(tmp_buffer);
 
     for (s32 layer = 0; layer < info.resources.layers; ++layer) {
         const std::span<const u8> src = input.subspan(host_offset);
@@ -977,13 +977,14 @@ std::vector<SwizzleParameters> FullUploadSwizzles(const ImageInfo& info) {
 }
 
 void SwizzleImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr, const ImageInfo& info,
-                  std::span<const BufferImageCopy> copies, std::span<const u8> memory) {
+                  std::span<const BufferImageCopy> copies, std::span<const u8> memory,
+                  std::vector<u8>& 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);
         }
     }
 }

src/video_core/texture_cache/util.h

@@ -77,7 +77,8 @@ void ConvertImage(std::span<const u8> input, const ImageInfo& info, std::span<u8
 [[nodiscard]] std::vector<SwizzleParameters> FullUploadSwizzles(const ImageInfo& info);
 
 void SwizzleImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr, const ImageInfo& info,
-                  std::span<const BufferImageCopy> copies, std::span<const u8> memory);
+                  std::span<const BufferImageCopy> copies, std::span<const u8> memory,
+                  std::vector<u8>& tmp_buffer);
 
 [[nodiscard]] bool IsBlockLinearSizeCompatible(const ImageInfo& new_info,
                                                const ImageInfo& overlap_info, u32 new_level,