improve texture cache barrier

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -507,58 +507,83 @@ TransformBufferCopies(std::span<const VideoCommon::BufferCopy> copies, size_t bu
         return value;
     }
 }
+struct RangedBarrierRange {
+    u32 min_mip = std::numeric_limits<u32>::max();
+    u32 max_mip = std::numeric_limits<u32>::min();
+    u32 min_layer = std::numeric_limits<u32>::max();
+    u32 max_layer = std::numeric_limits<u32>::min();
 
+    void AddLayers(const VkImageSubresourceLayers& layers) {
+        min_mip = std::min(min_mip, layers.mipLevel);
+        max_mip = std::max(max_mip, layers.mipLevel + 1);
+        min_layer = std::min(min_layer, layers.baseArrayLayer);
+        max_layer = std::max(max_layer, layers.baseArrayLayer + layers.layerCount);
+    }
+
+    VkImageSubresourceRange SubresourceRange(VkImageAspectFlags aspect_mask) const noexcept {
+        return VkImageSubresourceRange{
+            .aspectMask = aspect_mask,
+            .baseMipLevel = min_mip,
+            .levelCount = max_mip - min_mip,
+            .baseArrayLayer = min_layer,
+            .layerCount = max_layer - min_layer,
+        };
+    }
+};
 void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage image,
                        VkImageAspectFlags aspect_mask, bool is_initialized,
                        std::span<const VkBufferImageCopy> copies) {
     static constexpr VkAccessFlags WRITE_ACCESS_FLAGS =
-        VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-        VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+                                           VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+                                           VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
     static constexpr VkAccessFlags READ_ACCESS_FLAGS = VK_ACCESS_SHADER_READ_BIT |
                                                        VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
                                                        VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT;
+
+    //  Compute exact mip/layer range being written to
+    RangedBarrierRange range;
+    for (const auto& region : copies) {
+        range.AddLayers(region.imageSubresource);
+    }
+    const VkImageSubresourceRange subresource_range = range.SubresourceRange(aspect_mask);
+
     const VkImageMemoryBarrier read_barrier{
-        .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-        .pNext = nullptr,
-        .srcAccessMask = WRITE_ACCESS_FLAGS,
-        .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-        .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
-        .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
-        .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-        .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-        .image = image,
-        .subresourceRange{
-            .aspectMask = aspect_mask,
-            .baseMipLevel = 0,
-            .levelCount = VK_REMAINING_MIP_LEVELS,
-            .baseArrayLayer = 0,
-            .layerCount = VK_REMAINING_ARRAY_LAYERS,
-        },
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = WRITE_ACCESS_FLAGS,
+            .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
+            .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = image,
+            .subresourceRange = subresource_range,
     };
+
     const VkImageMemoryBarrier write_barrier{
-        .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
-        .pNext = nullptr,
-        .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-        .dstAccessMask = WRITE_ACCESS_FLAGS | READ_ACCESS_FLAGS,
-        .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
-        .newLayout = VK_IMAGE_LAYOUT_GENERAL,
-        .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-        .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-        .image = image,
-        .subresourceRange{
-            .aspectMask = aspect_mask,
-            .baseMipLevel = 0,
-            .levelCount = VK_REMAINING_MIP_LEVELS,
-            .baseArrayLayer = 0,
-            .layerCount = VK_REMAINING_ARRAY_LAYERS,
-        },
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .dstAccessMask = WRITE_ACCESS_FLAGS | READ_ACCESS_FLAGS,
+            .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = image,
+            .subresourceRange = subresource_range,
     };
-    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
-                           read_barrier);
+
+    cmdbuf.PipelineBarrier(
+            VK_PIPELINE_STAGE_HOST_BIT,
+            VK_PIPELINE_STAGE_TRANSFER_BIT,
+            0, read_barrier);
+
     cmdbuf.CopyBufferToImage(src_buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copies);
-    // TODO: Move this to another API
-    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0,
-                           write_barrier);
+
+    cmdbuf.PipelineBarrier(
+            VK_PIPELINE_STAGE_TRANSFER_BIT,
+            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
+            0, write_barrier);
 }
 
 [[nodiscard]] VkImageBlit MakeImageBlit(const Region2D& dst_region, const Region2D& src_region,
@@ -651,29 +676,7 @@ void TryTransformSwizzleIfNeeded(PixelFormat format, std::array<SwizzleSource, 4
     }
 }
 
-struct RangedBarrierRange {
-    u32 min_mip = std::numeric_limits<u32>::max();
-    u32 max_mip = std::numeric_limits<u32>::min();
-    u32 min_layer = std::numeric_limits<u32>::max();
-    u32 max_layer = std::numeric_limits<u32>::min();
 
-    void AddLayers(const VkImageSubresourceLayers& layers) {
-        min_mip = std::min(min_mip, layers.mipLevel);
-        max_mip = std::max(max_mip, layers.mipLevel + 1);
-        min_layer = std::min(min_layer, layers.baseArrayLayer);
-        max_layer = std::max(max_layer, layers.baseArrayLayer + layers.layerCount);
-    }
-
-    VkImageSubresourceRange SubresourceRange(VkImageAspectFlags aspect_mask) const noexcept {
-        return VkImageSubresourceRange{
-            .aspectMask = aspect_mask,
-            .baseMipLevel = min_mip,
-            .levelCount = max_mip - min_mip,
-            .baseArrayLayer = min_layer,
-            .layerCount = max_layer - min_layer,
-        };
-    }
-};
 
 [[nodiscard]] VkFormat Format(Shader::ImageFormat format) {
     switch (format) {
@@ -1457,12 +1460,18 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src,
                 .subresourceRange = dst_range.SubresourceRange(aspect_mask),
             },
         };
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
-                               0, {}, {}, pre_barriers);
+        cmdbuf.PipelineBarrier(
+                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
+                VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT,
+                VK_PIPELINE_STAGE_TRANSFER_BIT,
+                0, nullptr, nullptr, pre_barriers);
         cmdbuf.CopyImage(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image,
                          VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, vk_copies);
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
-                               0, {}, {}, post_barriers);
+        cmdbuf.PipelineBarrier(
+                VK_PIPELINE_STAGE_TRANSFER_BIT,
+                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
+                VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT,
+                0, nullptr, nullptr, post_barriers);
     });
 }
 
@@ -2352,7 +2361,7 @@ void TextureCacheRuntime::TransitionImageLayout(Image& image) {
         };
         scheduler.RequestOutsideRenderPassOperationContext();
         scheduler.Record([barrier](vk::CommandBuffer cmdbuf) {
-            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
                                    VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, barrier);
         });
     }