video_core: Fix destruction sequence

Rejoice! Vulkan no longer leaks memory

CMakeModules/VcpkgCmakeUtils.cmake

@@ -19,7 +19,7 @@ endif()
 # Install a package using the command line interface
 function(vcpkg_add_package PKG_NAME)
     # Run the executable to install dependencies
-    set(VCPKG_TARGET_TRIPLET_FLAG "--triplet=${VCPKG_TARGET_TRIPLET}")
+    set(VCPKG_TARGET_TRIPLET_FLAG "--triplet=x64-windows-static")
     message(STATUS "VCPKG: Installing ${PKG_NAME}")
     execute_process(COMMAND ${VCPKG_EXECUTABLE} ${VCPKG_TARGET_TRIPLET_FLAG} ${VCPKG_RECURSE_REBUILD_FLAG} --disable-metrics install "${PKG_NAME}" WORKING_DIRECTORY ${CMAKE_SOURCE_DIR} RESULT_VARIABLE VCPKG_INSTALL_OK)
     if (NOT VCPKG_INSTALL_OK EQUAL "0")

src/video_core/common/backend.h

@@ -32,6 +32,9 @@ public:
     // Triggers a swapchain buffer swap
     virtual void EndPresent() = 0;
 
+    // Submits any pending work and blocks the host until it completes
+    virtual void Flush() = 0;
+
     // Returns the framebuffer created from the swapchain images
     virtual FramebufferHandle GetWindowFramebuffer() = 0;
 

src/video_core/common/renderer.cpp

@@ -56,7 +56,8 @@ layout (std140, push_constant) uniform PresentUniformData {
 };
 
 void main() {
-    color = texture(sampler2D(top_screen, screen_sampler), frag_tex_coord);
+    //color = texture(sampler2D(top_screen, screen_sampler), frag_tex_coord);
+    color = vec4(1.f, 0.f, 0.f, 1.f);
 }
 )";
 
@@ -158,7 +159,7 @@ DisplayRenderer::DisplayRenderer(Frontend::EmuWindow& window) : render_window(wi
 
     // Create vertex buffer for the screen rectangle
     const BufferInfo vertex_info = {
-        .capacity = sizeof(ScreenRectVertex) * 10,
+        .capacity = sizeof(ScreenRectVertex) * 16,
         .usage = BufferUsage::Vertex
     };
 
@@ -207,6 +208,11 @@ DisplayRenderer::DisplayRenderer(Frontend::EmuWindow& window) : render_window(wi
     present_pipelines[0]->BindSampler(1, 0, screen_sampler);
 }
 
+DisplayRenderer::~DisplayRenderer() {
+    // Must flush the backend before destroying any pipelines!
+    backend->Flush();
+}
+
 void DisplayRenderer::PrepareRendertarget() {
     for (int i = 0; i < 3; i++) {
         int fb_id = i == 2 ? 1 : 0;
@@ -387,8 +393,8 @@ void DisplayRenderer::DrawSingleScreen(u32 screen, bool rotate, float x, float y
     }
 
     const u32 size = sizeof(ScreenRectVertex) * vertices.size();
-    const u64 mapped_offset = vertex_buffer->GetCurrentOffset();
     auto vertex_data = vertex_buffer->Map(size);
+    const u64 mapped_offset = vertex_buffer->GetCurrentOffset();
 
     // Copy vertex data
     std::memcpy(vertex_data.data(), vertices.data(), size);

src/video_core/common/renderer.h

@@ -67,7 +67,7 @@ constexpr u32 PRESENT_PIPELINES = 3;
 class DisplayRenderer {
 public:
     DisplayRenderer(Frontend::EmuWindow& window);
-    ~DisplayRenderer() = default;
+    ~DisplayRenderer();
 
     void SwapBuffers();
     void TryPresent(int timeout_ms) {}
@@ -121,8 +121,8 @@ private:
     void LoadColorToActiveTexture(u8 color_r, u8 color_g, u8 color_b, const ScreenInfo& screen);
 
 private:
-    std::unique_ptr<VideoCore::Rasterizer> rasterizer;
     std::unique_ptr<BackendBase> backend;
+    std::unique_ptr<VideoCore::Rasterizer> rasterizer;
     Frontend::EmuWindow& render_window;
     Common::Vec4f clear_color;
     f32 m_current_fps = 0.0f;

src/video_core/renderer_vulkan/vk_backend.cpp

@@ -81,7 +81,11 @@ Backend::Backend(Frontend::EmuWindow& window) : BackendBase(window),
 }
 
 Backend::~Backend() {
+
+    // Wait for all GPU operations to finish before continuing
     vk::Device device = instance.GetDevice();
+    device.waitIdle();
+
     device.destroyPipelineCache(pipeline_cache);
 
     for (u32 pool = 0; pool < SCHEDULER_COMMAND_COUNT; pool++) {
@@ -100,10 +104,14 @@ bool Backend::BeginPresent() {
 }
 
 void Backend::EndPresent() {
-    scheduler.Submit(false, swapchain.GetAvailableSemaphore(), swapchain.GetPresentSemaphore());
+    scheduler.Submit(false, true, swapchain.GetAvailableSemaphore(), swapchain.GetPresentSemaphore());
     swapchain.Present();
 }
 
+void Backend::Flush() {
+    scheduler.Submit(true);
+}
+
 FramebufferHandle Backend::GetWindowFramebuffer() {
     auto framebuffer = swapchain.GetCurrentFramebuffer();
     auto extent = swapchain.GetExtent();

src/video_core/renderer_vulkan/vk_backend.h

@@ -21,11 +21,10 @@ public:
 
     bool BeginPresent() override;
     void EndPresent() override;
+    void Flush() override;
 
     FramebufferHandle GetWindowFramebuffer() override;
-
     u64 QueryDriver(Query query) override { return 0; }
-
     u64 PipelineInfoHash(const PipelineInfo& info) override;
 
     BufferHandle CreateBuffer(BufferInfo info) override;
@@ -40,10 +39,8 @@ public:
 
     void Draw(PipelineHandle pipeline, FramebufferHandle draw_framebuffer,
               u32 base_vertex, u32 num_vertices) override;
-
     void DrawIndexed(PipelineHandle pipeline, FramebufferHandle draw_framebuffer,
                      u32 base_index, u32 num_indices, u32 base_vertex) override;
-
     void DispatchCompute(PipelineHandle pipeline, Common::Vec3<u32> groupsize,
                          Common::Vec3<u32> groups) override {}
 

src/video_core/renderer_vulkan/vk_buffer.cpp

@@ -148,7 +148,7 @@ void Buffer::Commit(u32 size) {
     if (info.usage == BufferUsage::Staging && size > 0) {
         vmaFlushAllocation(allocator, allocation, buffer_offset, size);
     } else {
-        vk::CommandBuffer command_buffer = scheduler.GetUploadCommandBuffer();
+        vk::CommandBuffer command_buffer = scheduler.GetRenderCommandBuffer();
         Buffer& staging = scheduler.GetCommandUploadBuffer();
 
         const vk::BufferCopy copy_region = {
@@ -158,6 +158,7 @@ void Buffer::Commit(u32 size) {
         };
 
         // Copy staging buffer to device local buffer
+        staging.Commit(size);
         command_buffer.copyBuffer(staging.GetHandle(), buffer, copy_region);
 
         vk::AccessFlags access_mask;
@@ -184,6 +185,8 @@ void Buffer::Commit(u32 size) {
         const vk::BufferMemoryBarrier buffer_barrier = {
             .srcAccessMask = vk::AccessFlagBits::eTransferWrite,
             .dstAccessMask = access_mask,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
             .buffer = buffer,
             .offset = buffer_offset,
             .size = size

src/video_core/renderer_vulkan/vk_instance.cpp

@@ -60,7 +60,9 @@ Instance::Instance(Frontend::EmuWindow& window) {
 
 Instance::~Instance() {
     device.waitIdle();
+    vmaDestroyAllocator(allocator);
     device.destroy();
+    instance.destroySurfaceKHR(surface);
     instance.destroy();
 }
 

src/video_core/renderer_vulkan/vk_pipeline.cpp

@@ -183,7 +183,7 @@ PipelineOwner::~PipelineOwner() {
     u32 i = 0;
     while (set_layouts[i] && update_templates[i]) {
         device.destroyDescriptorSetLayout(set_layouts[i]);
-        device.destroyDescriptorUpdateTemplate(update_templates[i]);
+        device.destroyDescriptorUpdateTemplate(update_templates[i++]);
     }
 }
 

src/video_core/renderer_vulkan/vk_renderpass_cache.h

@@ -36,7 +36,7 @@ private:
     // Special renderpass used for rendering to the swapchain
     vk::RenderPass present_renderpass;
     // [color_format][depth_format][is_clear_pass]
-    vk::RenderPass cached_renderpasses[MAX_COLOR_FORMATS][MAX_DEPTH_FORMATS][2];
+    vk::RenderPass cached_renderpasses[MAX_COLOR_FORMATS+1][MAX_DEPTH_FORMATS+1][2];
 };
 
 } // namespace VideoCore::Vulkan

src/video_core/renderer_vulkan/vk_task_scheduler.cpp

@@ -64,6 +64,9 @@ CommandScheduler::~CommandScheduler() {
     vk::Device device = instance.GetDevice();
     VmaAllocator allocator = instance.GetAllocator();
 
+    // Submit any remaining work
+    Submit(true, false);
+
     for (auto& command : commands) {
         device.destroyFence(command.fence);
 
@@ -110,10 +113,11 @@ void CommandScheduler::SetSwitchCallback(std::function<void(u32)> callback) {
     switch_callback = callback;
 }
 
-void CommandScheduler::Submit(bool wait_completion, vk::Semaphore wait_semaphore, vk::Semaphore signal_semaphore) {
+void CommandScheduler::Submit(bool wait_completion, bool begin_next,
-    const CommandSlot& command = commands[current_command];
+                              vk::Semaphore wait_semaphore, vk::Semaphore signal_semaphore) {
 
     // End command buffers
+    const CommandSlot& command = commands[current_command];
     command.render_command_buffer.end();
     if (command.use_upload_buffer) {
         command.upload_command_buffer.end();
@@ -126,9 +130,14 @@ void CommandScheduler::Submit(bool wait_completion, vk::Semaphore wait_semaphore
 
     const u32 signal_semaphore_count = signal_semaphore ? 1u : 0u;
     const u32 wait_semaphore_count = wait_semaphore ? 1u : 0u;
-    const u32 command_buffer_count = command.use_upload_buffer ? 2u : 1u;
+    u32 command_buffer_count = 0;
-    const std::array command_buffers = { command.render_command_buffer,
+    std::array<vk::CommandBuffer, 2> command_buffers;
-                                         command.upload_command_buffer };
+
+    if (command.use_upload_buffer) {
+        command_buffers[command_buffer_count++] = command.upload_command_buffer;
+    }
+
+    command_buffers[command_buffer_count++] = command.render_command_buffer;
 
     // Prepeare submit info
     const vk::SubmitInfo submit_info = {
@@ -151,7 +160,9 @@ void CommandScheduler::Submit(bool wait_completion, vk::Semaphore wait_semaphore
     }
 
     // Switch to next cmdbuffer.
+    if (begin_next) {
         SwitchSlot();
+    }
 }
 
 void CommandScheduler::Schedule(std::function<void(vk::Device, VmaAllocator)>&& func) {

src/video_core/renderer_vulkan/vk_task_scheduler.h

@@ -37,8 +37,8 @@ public:
     void Schedule(std::function<void(vk::Device, VmaAllocator)>&& func);
 
     // Submits the current command to the graphics queue
-    void Submit(bool wait_completion = false, vk::Semaphore wait = VK_NULL_HANDLE,
+    void Submit(bool wait_completion = false, bool begin_next = true,
-                vk::Semaphore signal = VK_NULL_HANDLE);
+                vk::Semaphore wait = VK_NULL_HANDLE, vk::Semaphore signal = VK_NULL_HANDLE);
 
     // Returns the command buffer used for early upload operations.
     // This is useful for vertex/uniform buffer uploads that happen once per frame