Merge pull request #6629 from FernandoS27/accel-dma-2

DMAEngine: Accelerate BufferClear [accelerateDMA Part 2]

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -136,6 +136,30 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer,
     });
 }
 
+void BufferCacheRuntime::ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t size, u32 value) {
+    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,
+    };
+    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,
+    };
+
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([dest_buffer, offset, size, value](vk::CommandBuffer cmdbuf) {
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
+                               0, READ_BARRIER);
+        cmdbuf.FillBuffer(dest_buffer, offset, size, value);
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+                               0, WRITE_BARRIER);
+    });
+}
+
 void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format,
                                          u32 base_vertex, u32 num_indices, VkBuffer buffer,
                                          u32 offset, [[maybe_unused]] u32 size) {

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -60,6 +60,8 @@ public:
     void CopyBuffer(VkBuffer src_buffer, VkBuffer dst_buffer,
                     std::span<const VideoCommon::BufferCopy> copies);
 
+    void ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t size, u32 value);
+
     void BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format, u32 num_indices,
                          u32 base_vertex, VkBuffer buffer, u32 offset, u32 size);
 

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -706,6 +706,11 @@ void RasterizerVulkan::FlushWork() {
 
 AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_) : buffer_cache{buffer_cache_} {}
 
+bool AccelerateDMA::BufferClear(GPUVAddr src_address, u64 amount, u32 value) {
+    std::scoped_lock lock{buffer_cache.mutex};
+    return buffer_cache.DMAClear(src_address, amount, value);
+}
+
 bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) {
     std::scoped_lock lock{buffer_cache.mutex};
     return buffer_cache.DMACopy(src_address, dest_address, amount);

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -56,6 +56,8 @@ public:
 
     bool BufferCopy(GPUVAddr start_address, GPUVAddr end_address, u64 amount) override;
 
+    bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) override;
+
 private:
     BufferCache& buffer_cache;
 };