Mark accelerted DMA destination buffers and images as GPU-modified

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -1335,7 +1335,8 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
     }
     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 = VideoCommon::ObtainBufferOperation::DoNothing;
+    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);
 
@@ -1344,8 +1345,12 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
     const std::span copy_span{&copy, 1};
 
     if constexpr (IS_IMAGE_UPLOAD) {
+        texture_cache.PrepareImage(image_id, true, false);
         image->UploadMemory(buffer->Handle(), offset, copy_span);
     } else {
+        if (offset % BytesPerBlock(image->info.format)) {
+            return false;
+        }
         texture_cache.DownloadImageIntoBuffer(image, buffer->Handle(), offset, copy_span,
                                               buffer_operand.address, buffer_size);
     }

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -830,7 +830,8 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
     }
     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 = VideoCommon::ObtainBufferOperation::DoNothing;
+    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);
 
@@ -839,8 +840,12 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
     const std::span copy_span{&copy, 1};
 
     if constexpr (IS_IMAGE_UPLOAD) {
+        texture_cache.PrepareImage(image_id, true, false);
         image->UploadMemory(buffer->Handle(), offset, copy_span);
     } else {
+        if (offset % BytesPerBlock(image->info.format)) {
+            return false;
+        }
         texture_cache.DownloadImageIntoBuffer(image, buffer->Handle(), offset, copy_span,
                                               buffer_operand.address, buffer_size);
     }

src/video_core/texture_cache/texture_cache_base.h

@@ -243,6 +243,9 @@ public:
     /// Create channel state.
     void CreateChannel(Tegra::Control::ChannelState& channel) final override;
 
+    /// Prepare an image to be used
+    void PrepareImage(ImageId image_id, bool is_modification, bool invalidate);
+
     std::recursive_mutex mutex;
 
 private:
@@ -387,9 +390,6 @@ private:
     /// Synchronize image aliases, copying data if needed
     void SynchronizeAliases(ImageId image_id);
 
-    /// Prepare an image to be used
-    void PrepareImage(ImageId image_id, bool is_modification, bool invalidate);
-
     /// Prepare an image view to be used
     void PrepareImageView(ImageViewId image_view_id, bool is_modification, bool invalidate);