GPU: Refactor synchronization on Async GPU

src/video_core/dma_pusher.cpp

@@ -32,6 +32,7 @@ void DmaPusher::DispatchCalls() {
         }
     }
     gpu.FlushCommands();
+    gpu.SyncGuestHost();
 }
 
 bool DmaPusher::Step() {

src/video_core/engines/maxwell_3d.cpp

@@ -403,9 +403,13 @@ void Maxwell3D::ProcessQueryGet() {
                "Units other than CROP are unimplemented");
 
     switch (regs.query.query_get.operation) {
-    case Regs::QueryOperation::Release:
-        StampQueryResult(regs.query.query_sequence, regs.query.query_get.short_query == 0);
+    case Regs::QueryOperation::Release: {
+        rasterizer.FlushCommands();
+        rasterizer.SyncGuestHost();
+        const u64 result = regs.query.query_sequence;
+        StampQueryResult(result, regs.query.query_get.short_query == 0);
         break;
+    }
     case Regs::QueryOperation::Acquire:
         // TODO(Blinkhawk): Under this operation, the GPU waits for the CPU to write a value that
         // matches the current payload.

src/video_core/gpu.cpp

@@ -142,6 +142,9 @@ void GPU::FlushCommands() {
     renderer->Rasterizer().FlushCommands();
 }
 
+void GPU::SyncGuestHost() {
+    renderer->Rasterizer().SyncGuestHost();
+}
 // Note that, traditionally, methods are treated as 4-byte addressable locations, and hence
 // their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4.
 // So the values you see in docs might be multiplied by 4.

src/video_core/gpu.h

@@ -156,6 +156,7 @@ public:
     void CallMethod(const MethodCall& method_call);
 
     void FlushCommands();
+    void SyncGuestHost();
 
     /// Returns a reference to the Maxwell3D GPU engine.
     Engines::Maxwell3D& Maxwell3D();

src/video_core/gpu_thread.cpp

@@ -40,7 +40,7 @@ static void RunThread(VideoCore::RendererBase& renderer, Core::Frontend::Graphic
         } else if (const auto data = std::get_if<FlushRegionCommand>(&next.data)) {
             renderer.Rasterizer().FlushRegion(data->addr, data->size);
         } else if (const auto data = std::get_if<InvalidateRegionCommand>(&next.data)) {
-            renderer.Rasterizer().InvalidateRegion(data->addr, data->size);
+            renderer.Rasterizer().OnCPUWrite(data->addr, data->size);
         } else if (std::holds_alternative<EndProcessingCommand>(next.data)) {
             return;
         } else {
@@ -82,12 +82,12 @@ void ThreadManager::FlushRegion(VAddr addr, u64 size) {
 }
 
 void ThreadManager::InvalidateRegion(VAddr addr, u64 size) {
-    system.Renderer().Rasterizer().InvalidateRegion(addr, size);
+    system.Renderer().Rasterizer().OnCPUWrite(addr, size);
 }
 
 void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) {
     // Skip flush on asynch mode, as FlushAndInvalidateRegion is not used for anything too important
-    InvalidateRegion(addr, size);
+    system.Renderer().Rasterizer().OnCPUWrite(addr, size);
 }
 
 void ThreadManager::WaitIdle() const {

src/video_core/rasterizer_interface.h

@@ -58,6 +58,12 @@ public:
     /// Notify rasterizer that any caches of the specified region should be invalidated
     virtual void InvalidateRegion(VAddr addr, u64 size) = 0;
 
+    /// Notify rasterizer that any caches of the specified region are desync with guest
+    virtual void OnCPUWrite(VAddr addr, u64 size) = 0;
+
+    /// Sync memory between guest and host.
+    virtual void SyncGuestHost() = 0;
+
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
     /// and invalidated
     virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0;

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -660,6 +660,22 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
     query_cache.InvalidateRegion(addr, size);
 }
 
+void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
+    MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+    if (!addr || !size) {
+        return;
+    }
+    texture_cache.OnCPUWrite(addr, size);
+    shader_cache.InvalidateRegion(addr, size);
+    buffer_cache.InvalidateRegion(addr, size);
+}
+
+void RasterizerOpenGL::SyncGuestHost() {
+    MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+    texture_cache.SyncGuestHost();
+    // buffer_cache.SyncGuestHost();
+}
+
 void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) {
     if (Settings::IsGPULevelExtreme()) {
         FlushRegion(addr, size);

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -67,6 +67,8 @@ public:
     void FlushAll() override;
     void FlushRegion(VAddr addr, u64 size) override;
     void InvalidateRegion(VAddr addr, u64 size) override;
+    void OnCPUWrite(VAddr addr, u64 size) override;
+    void SyncGuestHost() override;
     void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
     void FlushCommands() override;
     void TickFrame() override;

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -524,6 +524,20 @@ void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size) {
     query_cache.InvalidateRegion(addr, size);
 }
 
+void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
+    if (!addr || !size) {
+        return;
+    }
+    texture_cache.OnCPUWrite(addr, size);
+    pipeline_cache.InvalidateRegion(addr, size);
+    buffer_cache.InvalidateRegion(addr, size);
+}
+
+void RasterizerVulkan::SyncGuestHost() {
+    texture_cache.SyncGuestHost();
+    // buffer_cache.SyncGuestHost();
+}
+
 void RasterizerVulkan::FlushAndInvalidateRegion(VAddr addr, u64 size) {
     FlushRegion(addr, size);
     InvalidateRegion(addr, size);

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -119,6 +119,8 @@ public:
     void FlushAll() override;
     void FlushRegion(VAddr addr, u64 size) override;
     void InvalidateRegion(VAddr addr, u64 size) override;
+    void OnCPUWrite(VAddr addr, u64 size) override;
+    void SyncGuestHost() override;
     void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
     void FlushCommands() override;
     void TickFrame() override;

src/video_core/texture_cache/texture_cache.h

@@ -63,7 +63,7 @@ public:
         }
     }
 
-    void OnCPUWrite(CacheAddr addr, std::size_t size) {
+    void OnCPUWrite(VAddr addr, std::size_t size) {
         std::lock_guard lock{mutex};
 
         for (const auto& surface : GetSurfacesInRegion(addr, size)) {
@@ -549,7 +549,7 @@ private:
         }
         const auto& final_params = new_surface->GetSurfaceParams();
         if (cr_params.type != final_params.type) {
-            if (Settings::values.use_accurate_gpu_emulation) {
+            if (Settings::IsGPULevelExtreme()) {
                 BufferCopy(current_surface, new_surface);
             }
         } else {