Merge pull request #9913 from ameerj/acc-dma-refactor

AccelerateDMA: Refactor Buffer/Image copy code and implement for OGL

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -63,7 +63,7 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra
       buffer_cache(*this, cpu_memory_, buffer_cache_runtime),
       shader_cache(*this, emu_window_, device, texture_cache, buffer_cache, program_manager,
                    state_tracker, gpu.ShaderNotify()),
-      query_cache(*this), accelerate_dma(buffer_cache),
+      query_cache(*this), accelerate_dma(buffer_cache, texture_cache),
       fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache),
       blit_image(program_manager_) {}
 
@@ -1262,7 +1262,8 @@ void RasterizerOpenGL::ReleaseChannel(s32 channel_id) {
     query_cache.EraseChannel(channel_id);
 }
 
-AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_) : buffer_cache{buffer_cache_} {}
+AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_, TextureCache& texture_cache_)
+    : buffer_cache{buffer_cache_}, texture_cache{texture_cache_} {}
 
 bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) {
     std::scoped_lock lock{buffer_cache.mutex};
@@ -1274,4 +1275,44 @@ bool AccelerateDMA::BufferClear(GPUVAddr src_address, u64 amount, u32 value) {
     return buffer_cache.DMAClear(src_address, amount, value);
 }
 
+template <bool IS_IMAGE_UPLOAD>
+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);
+    if (image_id == VideoCommon::NULL_IMAGE_ID) {
+        return false;
+    }
+    const u32 buffer_size = static_cast<u32>(buffer_operand.pitch * buffer_operand.height);
+    static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize;
+    const auto post_op = IS_IMAGE_UPLOAD ? VideoCommon::ObtainBufferOperation::DoNothing
+                                         : VideoCommon::ObtainBufferOperation::MarkAsWritten;
+    const auto [buffer, offset] =
+        buffer_cache.ObtainBuffer(buffer_operand.address, buffer_size, sync_info, post_op);
+
+    const auto [image, copy] = texture_cache.DmaBufferImageCopy(
+        copy_info, buffer_operand, image_operand, image_id, IS_IMAGE_UPLOAD);
+    const std::span copy_span{&copy, 1};
+
+    if constexpr (IS_IMAGE_UPLOAD) {
+        image->UploadMemory(buffer->Handle(), offset, copy_span);
+    } else {
+        image->DownloadMemory(buffer->Handle(), offset, copy_span);
+    }
+    return true;
+}
+
+bool AccelerateDMA::ImageToBuffer(const Tegra::DMA::ImageCopy& copy_info,
+                                  const Tegra::DMA::ImageOperand& image_operand,
+                                  const Tegra::DMA::BufferOperand& buffer_operand) {
+    return DmaBufferImageCopy<false>(copy_info, buffer_operand, image_operand);
+}
+
+bool AccelerateDMA::BufferToImage(const Tegra::DMA::ImageCopy& copy_info,
+                                  const Tegra::DMA::BufferOperand& buffer_operand,
+                                  const Tegra::DMA::ImageOperand& image_operand) {
+    return DmaBufferImageCopy<true>(copy_info, buffer_operand, image_operand);
+}
+
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -50,24 +50,26 @@ static_assert(sizeof(BindlessSSBO) * CHAR_BIT == 128);
 
 class AccelerateDMA : public Tegra::Engines::AccelerateDMAInterface {
 public:
-    explicit AccelerateDMA(BufferCache& buffer_cache);
+    explicit AccelerateDMA(BufferCache& buffer_cache, TextureCache& texture_cache);
 
     bool BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) override;
 
     bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) override;
 
     bool ImageToBuffer(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::ImageOperand& src,
-                       const Tegra::DMA::BufferOperand& dst) override {
-        return false;
-    }
+                       const Tegra::DMA::BufferOperand& dst) override;
 
     bool BufferToImage(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& src,
-                       const Tegra::DMA::ImageOperand& dst) override {
-        return false;
-    }
+                       const Tegra::DMA::ImageOperand& dst) override;
 
 private:
+    template <bool IS_IMAGE_UPLOAD>
+    bool DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
+                            const Tegra::DMA::BufferOperand& src,
+                            const Tegra::DMA::ImageOperand& dst);
+
     BufferCache& buffer_cache;
+    TextureCache& texture_cache;
 };
 
 class RasterizerOpenGL : public VideoCore::RasterizerAccelerated,

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -763,14 +763,14 @@ Image::Image(const VideoCommon::NullImageParams& params) : VideoCommon::ImageBas
 
 Image::~Image() = default;
 
-void Image::UploadMemory(const ImageBufferMap& map,
+void Image::UploadMemory(GLuint buffer_handle, size_t buffer_offset,
                          std::span<const VideoCommon::BufferImageCopy> copies) {
     const bool is_rescaled = True(flags & ImageFlagBits::Rescaled);
     if (is_rescaled) {
         ScaleDown(true);
     }
-    glBindBuffer(GL_PIXEL_UNPACK_BUFFER, map.buffer);
-    glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, map.offset, unswizzled_size_bytes);
+    glBindBuffer(GL_PIXEL_UNPACK_BUFFER, buffer_handle);
+    glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, buffer_offset, unswizzled_size_bytes);
 
     glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
 
@@ -789,21 +789,26 @@ void Image::UploadMemory(const ImageBufferMap& map,
             current_image_height = copy.buffer_image_height;
             glPixelStorei(GL_UNPACK_IMAGE_HEIGHT, current_image_height);
         }
-        CopyBufferToImage(copy, map.offset);
+        CopyBufferToImage(copy, buffer_offset);
     }
     if (is_rescaled) {
         ScaleUp();
     }
 }
 
-void Image::DownloadMemory(ImageBufferMap& map,
+void Image::UploadMemory(const ImageBufferMap& map,
+                         std::span<const VideoCommon::BufferImageCopy> copies) {
+    UploadMemory(map.buffer, map.offset, copies);
+}
+
+void Image::DownloadMemory(GLuint buffer_handle, size_t buffer_offset,
                            std::span<const VideoCommon::BufferImageCopy> copies) {
     const bool is_rescaled = True(flags & ImageFlagBits::Rescaled);
     if (is_rescaled) {
         ScaleDown();
     }
     glMemoryBarrier(GL_PIXEL_BUFFER_BARRIER_BIT); // TODO: Move this to its own API
-    glBindBuffer(GL_PIXEL_PACK_BUFFER, map.buffer);
+    glBindBuffer(GL_PIXEL_PACK_BUFFER, buffer_handle);
     glPixelStorei(GL_PACK_ALIGNMENT, 1);
 
     u32 current_row_length = std::numeric_limits<u32>::max();
@@ -821,13 +826,18 @@ void Image::DownloadMemory(ImageBufferMap& map,
             current_image_height = copy.buffer_image_height;
             glPixelStorei(GL_PACK_IMAGE_HEIGHT, current_image_height);
         }
-        CopyImageToBuffer(copy, map.offset);
+        CopyImageToBuffer(copy, buffer_offset);
     }
     if (is_rescaled) {
         ScaleUp(true);
     }
 }
 
+void Image::DownloadMemory(ImageBufferMap& map,
+                           std::span<const VideoCommon::BufferImageCopy> copies) {
+    DownloadMemory(map.buffer, map.offset, copies);
+}
+
 GLuint Image::StorageHandle() noexcept {
     switch (info.format) {
     case PixelFormat::A8B8G8R8_SRGB:

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -206,9 +206,15 @@ public:
     Image(Image&&) = default;
     Image& operator=(Image&&) = default;
 
+    void UploadMemory(GLuint buffer_handle, size_t buffer_offset,
+                      std::span<const VideoCommon::BufferImageCopy> copies);
+
     void UploadMemory(const ImageBufferMap& map,
                       std::span<const VideoCommon::BufferImageCopy> copies);
 
+    void DownloadMemory(GLuint buffer_handle, size_t buffer_offset,
+                        std::span<const VideoCommon::BufferImageCopy> copies);
+
     void DownloadMemory(ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies);
 
     GLuint StorageHandle() noexcept;

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -770,232 +770,44 @@ bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64
     return buffer_cache.DMACopy(src_address, dest_address, amount);
 }
 
-bool AccelerateDMA::ImageToBuffer(const Tegra::DMA::ImageCopy& copy_info,
-                                  const Tegra::DMA::ImageOperand& src,
-                                  const Tegra::DMA::BufferOperand& dst) {
+template <bool IS_IMAGE_UPLOAD>
+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};
-    auto query_image = texture_cache.ObtainImage(src, false);
-    if (!query_image) {
+    const auto image_id = texture_cache.DmaImageId(image_operand);
+    if (image_id == VideoCommon::NULL_IMAGE_ID) {
         return false;
     }
-    auto* image = query_image->first;
-    auto [level, base] = query_image->second;
-    const u32 buffer_size = static_cast<u32>(dst.pitch * dst.height);
-    const auto [buffer, offset] = buffer_cache.ObtainBuffer(
-        dst.address, buffer_size, VideoCommon::ObtainBufferSynchronize::FullSynchronize,
-        VideoCommon::ObtainBufferOperation::MarkAsWritten);
+    const u32 buffer_size = static_cast<u32>(buffer_operand.pitch * buffer_operand.height);
+    static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize;
+    const auto post_op = IS_IMAGE_UPLOAD ? VideoCommon::ObtainBufferOperation::DoNothing
+                                         : VideoCommon::ObtainBufferOperation::MarkAsWritten;
+    const auto [buffer, offset] =
+        buffer_cache.ObtainBuffer(buffer_operand.address, buffer_size, sync_info, post_op);
 
-    const bool is_rescaled = image->IsRescaled();
-    if (is_rescaled) {
-        image->ScaleDown();
-    }
-    VkImageSubresourceLayers subresources{
-        .aspectMask = image->AspectMask(),
-        .mipLevel = level,
-        .baseArrayLayer = base,
-        .layerCount = 1,
-    };
-    const u32 bpp = VideoCore::Surface::BytesPerBlock(image->info.format);
-    const auto convert = [old_bpp = src.bytes_per_pixel, bpp](u32 value) {
-        return (old_bpp * value) / bpp;
-    };
-    const u32 base_x = convert(src.params.origin.x.Value());
-    const u32 base_y = src.params.origin.y.Value();
-    const u32 length_x = convert(copy_info.length_x);
-    const u32 length_y = copy_info.length_y;
-    VkOffset3D image_offset{
-        .x = static_cast<s32>(base_x),
-        .y = static_cast<s32>(base_y),
-        .z = 0,
-    };
-    VkExtent3D image_extent{
-        .width = length_x,
-        .height = length_y,
-        .depth = 1,
-    };
-    auto buff_info(dst);
-    buff_info.pitch = convert(dst.pitch);
-    scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([src_image = image->Handle(), dst_buffer = buffer->Handle(),
-                      buffer_offset = offset, subresources, image_offset, image_extent,
-                      buff_info](vk::CommandBuffer cmdbuf) {
-        const std::array buffer_copy_info{
-            VkBufferImageCopy{
-                .bufferOffset = buffer_offset,
-                .bufferRowLength = buff_info.pitch,
-                .bufferImageHeight = buff_info.height,
-                .imageSubresource = subresources,
-                .imageOffset = image_offset,
-                .imageExtent = image_extent,
-            },
-        };
-        const VkImageSubresourceRange range{
-            .aspectMask = subresources.aspectMask,
-            .baseMipLevel = subresources.mipLevel,
-            .levelCount = 1,
-            .baseArrayLayer = subresources.baseArrayLayer,
-            .layerCount = 1,
-        };
-        static constexpr VkMemoryBarrier WRITE_BARRIER{
-            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
-            .pNext = nullptr,
-            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
-        };
-        const std::array pre_barriers{
-            VkImageMemoryBarrier{
-                .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-                .pNext = nullptr,
-                .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-                                 VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT |
-                                 VK_ACCESS_TRANSFER_WRITE_BIT,
-                .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
-                .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
-                .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
-                .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .image = src_image,
-                .subresourceRange = range,
-            },
-        };
-        const std::array post_barriers{
-            VkImageMemoryBarrier{
-                .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-                .pNext = nullptr,
-                .srcAccessMask = 0,
-                .dstAccessMask = 0,
-                .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
-                .newLayout = VK_IMAGE_LAYOUT_GENERAL,
-                .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .image = src_image,
-                .subresourceRange = range,
-            },
-        };
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
-                               0, {}, {}, pre_barriers);
-        cmdbuf.CopyImageToBuffer(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_buffer,
-                                 buffer_copy_info);
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
-                               0, WRITE_BARRIER, nullptr, post_barriers);
-    });
-    if (is_rescaled) {
-        image->ScaleUp(true);
+    const auto [image, copy] = texture_cache.DmaBufferImageCopy(
+        copy_info, buffer_operand, image_operand, image_id, IS_IMAGE_UPLOAD);
+    const std::span copy_span{&copy, 1};
+
+    if constexpr (IS_IMAGE_UPLOAD) {
+        image->UploadMemory(buffer->Handle(), offset, copy_span);
+    } else {
+        image->DownloadMemory(buffer->Handle(), offset, copy_span);
     }
     return true;
 }
 
+bool AccelerateDMA::ImageToBuffer(const Tegra::DMA::ImageCopy& copy_info,
+                                  const Tegra::DMA::ImageOperand& image_operand,
+                                  const Tegra::DMA::BufferOperand& buffer_operand) {
+    return DmaBufferImageCopy<false>(copy_info, buffer_operand, image_operand);
+}
+
 bool AccelerateDMA::BufferToImage(const Tegra::DMA::ImageCopy& copy_info,
-                                  const Tegra::DMA::BufferOperand& src,
-                                  const Tegra::DMA::ImageOperand& dst) {
-    std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
-    auto query_image = texture_cache.ObtainImage(dst, true);
-    if (!query_image) {
-        return false;
-    }
-    auto* image = query_image->first;
-    auto [level, base] = query_image->second;
-    const u32 buffer_size = static_cast<u32>(src.pitch * src.height);
-    const auto [buffer, offset] = buffer_cache.ObtainBuffer(
-        src.address, buffer_size, VideoCommon::ObtainBufferSynchronize::FullSynchronize,
-        VideoCommon::ObtainBufferOperation::DoNothing);
-    const bool is_rescaled = image->IsRescaled();
-    if (is_rescaled) {
-        image->ScaleDown(true);
-    }
-    VkImageSubresourceLayers subresources{
-        .aspectMask = image->AspectMask(),
-        .mipLevel = level,
-        .baseArrayLayer = base,
-        .layerCount = 1,
-    };
-    const u32 bpp = VideoCore::Surface::BytesPerBlock(image->info.format);
-    const auto convert = [old_bpp = dst.bytes_per_pixel, bpp](u32 value) {
-        return (old_bpp * value) / bpp;
-    };
-    const u32 base_x = convert(dst.params.origin.x.Value());
-    const u32 base_y = dst.params.origin.y.Value();
-    const u32 length_x = convert(copy_info.length_x);
-    const u32 length_y = copy_info.length_y;
-    VkOffset3D image_offset{
-        .x = static_cast<s32>(base_x),
-        .y = static_cast<s32>(base_y),
-        .z = 0,
-    };
-    VkExtent3D image_extent{
-        .width = length_x,
-        .height = length_y,
-        .depth = 1,
-    };
-    auto buff_info(src);
-    buff_info.pitch = convert(src.pitch);
-    scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([dst_image = image->Handle(), src_buffer = buffer->Handle(),
-                      buffer_offset = offset, subresources, image_offset, image_extent,
-                      buff_info](vk::CommandBuffer cmdbuf) {
-        const std::array buffer_copy_info{
-            VkBufferImageCopy{
-                .bufferOffset = buffer_offset,
-                .bufferRowLength = buff_info.pitch,
-                .bufferImageHeight = buff_info.height,
-                .imageSubresource = subresources,
-                .imageOffset = image_offset,
-                .imageExtent = image_extent,
-            },
-        };
-        const VkImageSubresourceRange range{
-            .aspectMask = subresources.aspectMask,
-            .baseMipLevel = subresources.mipLevel,
-            .levelCount = 1,
-            .baseArrayLayer = subresources.baseArrayLayer,
-            .layerCount = 1,
-        };
-        static constexpr VkMemoryBarrier READ_BARRIER{
-            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
-            .pNext = nullptr,
-            .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
-        };
-        const std::array pre_barriers{
-            VkImageMemoryBarrier{
-                .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-                .pNext = nullptr,
-                .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-                                 VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT |
-                                 VK_ACCESS_TRANSFER_WRITE_BIT,
-                .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
-                .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
-                .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
-                .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .image = dst_image,
-                .subresourceRange = range,
-            },
-        };
-        const std::array post_barriers{
-            VkImageMemoryBarrier{
-                .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-                .pNext = nullptr,
-                .srcAccessMask = 0,
-                .dstAccessMask = 0,
-                .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
-                .newLayout = VK_IMAGE_LAYOUT_GENERAL,
-                .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .image = dst_image,
-                .subresourceRange = range,
-            },
-        };
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
-                               0, READ_BARRIER, {}, pre_barriers);
-        cmdbuf.CopyBufferToImage(src_buffer, dst_image, VK_IMAGE_LAYOUT_GENERAL, buffer_copy_info);
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
-                               0, nullptr, nullptr, post_barriers);
-    });
-    if (is_rescaled) {
-        image->ScaleUp();
-    }
-    return true;
+                                  const Tegra::DMA::BufferOperand& buffer_operand,
+                                  const Tegra::DMA::ImageOperand& image_operand) {
+    return DmaBufferImageCopy<true>(copy_info, buffer_operand, image_operand);
 }
 
 void RasterizerVulkan::UpdateDynamicStates() {

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -59,6 +59,11 @@ public:
                        const Tegra::DMA::ImageOperand& dst) override;
 
 private:
+    template <bool IS_IMAGE_UPLOAD>
+    bool DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
+                            const Tegra::DMA::BufferOperand& src,
+                            const Tegra::DMA::ImageOperand& dst);
+
     BufferCache& buffer_cache;
     TextureCache& texture_cache;
     Scheduler& scheduler;

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -1315,15 +1315,16 @@ Image::Image(const VideoCommon::NullImageParams& params) : VideoCommon::ImageBas
 
 Image::~Image() = default;
 
-void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) {
+void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
+                         std::span<const VideoCommon::BufferImageCopy> copies) {
     // TODO: Move this to another API
     const bool is_rescaled = True(flags & ImageFlagBits::Rescaled);
     if (is_rescaled) {
         ScaleDown(true);
     }
     scheduler->RequestOutsideRenderPassOperationContext();
-    std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask);
-    const VkBuffer src_buffer = map.buffer;
+    std::vector vk_copies = TransformBufferImageCopies(copies, offset, aspect_mask);
+    const VkBuffer src_buffer = buffer;
     const VkImage vk_image = *original_image;
     const VkImageAspectFlags vk_aspect_mask = aspect_mask;
     const bool is_initialized = std::exchange(initialized, true);
@@ -1336,14 +1337,19 @@ void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImag
     }
 }
 
-void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) {
+void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) {
+    UploadMemory(map.buffer, map.offset, copies);
+}
+
+void Image::DownloadMemory(VkBuffer buffer, VkDeviceSize offset,
+                           std::span<const VideoCommon::BufferImageCopy> copies) {
     const bool is_rescaled = True(flags & ImageFlagBits::Rescaled);
     if (is_rescaled) {
         ScaleDown();
     }
-    std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask);
+    std::vector vk_copies = TransformBufferImageCopies(copies, offset, aspect_mask);
     scheduler->RequestOutsideRenderPassOperationContext();
-    scheduler->Record([buffer = map.buffer, image = *original_image, aspect_mask = aspect_mask,
+    scheduler->Record([buffer, image = *original_image, aspect_mask = aspect_mask,
                        vk_copies](vk::CommandBuffer cmdbuf) {
         const VkImageMemoryBarrier read_barrier{
             .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
@@ -1398,6 +1404,10 @@ void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferIm
     }
 }
 
+void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) {
+    DownloadMemory(map.buffer, map.offset, copies);
+}
+
 bool Image::IsRescaled() const noexcept {
     return True(flags & ImageFlagBits::Rescaled);
 }

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -132,9 +132,15 @@ public:
     Image(Image&&) = default;
     Image& operator=(Image&&) = default;
 
+    void UploadMemory(VkBuffer buffer, VkDeviceSize offset,
+                      std::span<const VideoCommon::BufferImageCopy> copies);
+
     void UploadMemory(const StagingBufferRef& map,
                       std::span<const VideoCommon::BufferImageCopy> copies);
 
+    void DownloadMemory(VkBuffer buffer, VkDeviceSize offset,
+                        std::span<const VideoCommon::BufferImageCopy> copies);
+
     void DownloadMemory(const StagingBufferRef& map,
                         std::span<const VideoCommon::BufferImageCopy> copies);
 

src/video_core/texture_cache/texture_cache.h

@@ -744,6 +744,25 @@ void TextureCache<P>::PopAsyncFlushes() {
     }
 }
 
+template <class P>
+ImageId TextureCache<P>::DmaImageId(const Tegra::DMA::ImageOperand& operand) {
+    const ImageInfo dst_info(operand);
+    const ImageId dst_id = FindDMAImage(dst_info, operand.address);
+    if (!dst_id) {
+        return NULL_IMAGE_ID;
+    }
+    const 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;
+    }
+    const auto base = image.TryFindBase(operand.address);
+    if (!base) {
+        return NULL_IMAGE_ID;
+    }
+    return dst_id;
+}
+
 template <class P>
 bool TextureCache<P>::IsRescaling() const noexcept {
     return is_rescaling;
@@ -771,6 +790,49 @@ bool TextureCache<P>::IsRegionGpuModified(VAddr addr, size_t size) {
     return is_modified;
 }
 
+template <class P>
+std::pair<typename TextureCache<P>::Image*, BufferImageCopy> TextureCache<P>::DmaBufferImageCopy(
+    const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand,
+    const Tegra::DMA::ImageOperand& image_operand, ImageId image_id, bool modifies_image) {
+    const auto [level, base] = PrepareDmaImage(image_id, image_operand.address, modifies_image);
+    auto* image = &slot_images[image_id];
+    const u32 buffer_size = static_cast<u32>(buffer_operand.pitch * buffer_operand.height);
+    const u32 bpp = VideoCore::Surface::BytesPerBlock(image->info.format);
+    const auto convert = [old_bpp = image_operand.bytes_per_pixel, bpp](u32 value) {
+        return (old_bpp * value) / bpp;
+    };
+    const u32 base_x = convert(image_operand.params.origin.x.Value());
+    const u32 base_y = image_operand.params.origin.y.Value();
+    const u32 length_x = convert(copy_info.length_x);
+    const u32 length_y = copy_info.length_y;
+
+    const BufferImageCopy copy{
+        .buffer_offset = 0,
+        .buffer_size = buffer_size,
+        .buffer_row_length = convert(buffer_operand.pitch),
+        .buffer_image_height = buffer_operand.height,
+        .image_subresource =
+            {
+                .base_level = static_cast<s32>(level),
+                .base_layer = static_cast<s32>(base),
+                .num_layers = 1,
+            },
+        .image_offset =
+            {
+                .x = static_cast<s32>(base_x),
+                .y = static_cast<s32>(base_y),
+                .z = 0,
+            },
+        .image_extent =
+            {
+                .width = length_x,
+                .height = length_y,
+                .depth = 1,
+            },
+    };
+    return {image, copy};
+}
+
 template <class P>
 void TextureCache<P>::RefreshContents(Image& image, ImageId image_id) {
     if (False(image.flags & ImageFlagBits::CpuModified)) {
@@ -1405,26 +1467,14 @@ ImageId TextureCache<P>::FindDMAImage(const ImageInfo& info, GPUVAddr gpu_addr)
 }
 
 template <class P>
-std::optional<std::pair<typename TextureCache<P>::Image*, std::pair<u32, u32>>>
-TextureCache<P>::ObtainImage(const Tegra::DMA::ImageOperand& operand, bool mark_as_modified) {
-    ImageInfo dst_info(operand);
-    ImageId dst_id = FindDMAImage(dst_info, operand.address);
-    if (!dst_id) {
-        return std::nullopt;
-    }
-    auto& image = slot_images[dst_id];
-    auto base = image.TryFindBase(operand.address);
-    if (!base) {
-        return std::nullopt;
-    }
-    if (False(image.flags & ImageFlagBits::GpuModified)) {
-        // No need to waste time on an image that's synced with guest
-        return std::nullopt;
-    }
+std::pair<u32, u32> TextureCache<P>::PrepareDmaImage(ImageId dst_id, GPUVAddr base_addr,
+                                                     bool mark_as_modified) {
+    const auto& image = slot_images[dst_id];
+    const auto base = image.TryFindBase(base_addr);
     PrepareImage(dst_id, mark_as_modified, false);
-    auto& new_image = slot_images[dst_id];
+    const auto& new_image = slot_images[dst_id];
     lru_cache.Touch(new_image.lru_index, frame_tick);
-    return std::make_pair(&new_image, std::make_pair(base->level, base->layer));
+    return std::make_pair(base->level, base->layer);
 }
 
 template <class P>

src/video_core/texture_cache/texture_cache_base.h

@@ -209,8 +209,11 @@ public:
     /// Pop asynchronous downloads
     void PopAsyncFlushes();
 
-    [[nodiscard]] std::optional<std::pair<Image*, std::pair<u32, u32>>> ObtainImage(
-        const Tegra::DMA::ImageOperand& operand, bool mark_as_modified);
+    [[nodiscard]] ImageId DmaImageId(const Tegra::DMA::ImageOperand& operand);
+
+    [[nodiscard]] std::pair<Image*, BufferImageCopy> DmaBufferImageCopy(
+        const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand,
+        const Tegra::DMA::ImageOperand& image_operand, ImageId image_id, bool modifies_image);
 
     /// Return true when a CPU region is modified from the GPU
     [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
@@ -386,6 +389,9 @@ private:
     /// Returns true if the current clear parameters clear the whole image of a given image view
     [[nodiscard]] bool IsFullClear(ImageViewId id);
 
+    [[nodiscard]] std::pair<u32, u32> PrepareDmaImage(ImageId dst_id, GPUVAddr base_addr,
+                                                      bool mark_as_modified);
+
     bool ImageCanRescale(ImageBase& image);
     void InvalidateScale(Image& image);
     bool ScaleUp(Image& image);