[vk] batched draws

Signed-off-by: lizzie <lizzie@eden-emu.dev>

src/video_core/engines/draw_manager.cpp

@@ -69,6 +69,7 @@ void DrawManager::ProcessMethodCall(u32 method, u32 argument) {
 
 void DrawManager::Clear(u32 layer_count) {
     if (maxwell3d->ShouldExecute()) {
+        maxwell3d->rasterizer->FlushBatchedDraws();
         maxwell3d->rasterizer->Clear(layer_count);
     }
 }
@@ -144,6 +145,8 @@ void DrawManager::SetInlineIndexBuffer(u32 index) {
 }
 
 void DrawManager::DrawBegin() {
+    maxwell3d->rasterizer->FlushBatchedDraws();
+
     const auto& regs{maxwell3d->regs};
     auto reset_instance_count = regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::First;
     auto increment_instance_count =
@@ -192,6 +195,7 @@ void DrawManager::DrawEnd(u32 instance_count, bool force_draw) {
         draw_state.inline_index_draw_indexes.clear();
         break;
     }
+    maxwell3d->rasterizer->FlushBatchedDraws();
 }
 
 void DrawManager::DrawIndexSmall(u32 argument) {

src/video_core/rasterizer_interface.h

@@ -39,6 +39,9 @@ class RasterizerInterface {
 public:
     virtual ~RasterizerInterface() = default;
 
+    /// Flushes draws batched by Draw()
+    virtual void FlushBatchedDraws() = 0;
+
     /// Dispatches a draw invocation
     virtual void Draw(bool is_indexed, u32 instance_count) = 0;
 

src/video_core/renderer_null/null_rasterizer.cpp

@@ -22,6 +22,7 @@ RasterizerNull::RasterizerNull(Tegra::GPU& gpu) : m_gpu{gpu} {}
 RasterizerNull::~RasterizerNull() = default;
 
 void RasterizerNull::Draw(bool is_indexed, u32 instance_count) {}
+void RasterizerNull::FlushBatchedDraws() {}
 void RasterizerNull::DrawTexture() {}
 void RasterizerNull::Clear(u32 layer_count) {}
 void RasterizerNull::DispatchCompute() {}

src/video_core/renderer_null/null_rasterizer.h

@@ -37,6 +37,7 @@ public:
     explicit RasterizerNull(Tegra::GPU& gpu);
     ~RasterizerNull() override;
 
+    void FlushBatchedDraws() override;
     void Draw(bool is_indexed, u32 instance_count) override;
     void DrawTexture() override;
     void Clear(u32 layer_count) override;

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -258,6 +258,10 @@ void RasterizerOpenGL::PrepareDraw(bool is_indexed, Func&& draw_func) {
     has_written_global_memory |= pipeline->WritesGlobalMemory();
 }
 
+void RasterizerOpenGL::FlushBatchedDraws() {
+
+}
+
 void RasterizerOpenGL::Draw(bool is_indexed, u32 instance_count) {
     PrepareDraw(is_indexed, [this, is_indexed, instance_count](GLenum primitive_mode) {
         const auto& draw_state = maxwell3d->draw_manager->GetDrawState();

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -81,6 +81,7 @@ public:
                               StateTracker& state_tracker_);
     ~RasterizerOpenGL() override;
 
+    void FlushBatchedDraws() override;
     void Draw(bool is_indexed, u32 instance_count) override;
     void DrawIndirect() override;
     void DrawTexture() override;

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -310,6 +310,8 @@ void GraphicsPipeline::AddTransition(GraphicsPipeline* transition) {
 
 template <typename Spec>
 bool GraphicsPipeline::ConfigureImpl(bool is_indexed) {
+    maxwell3d->Rasterizer().FlushBatchedDraws(); //DRAW THINGS FROM PREVIOUS PIPELINE!!!!!!
+
     std::array<VideoCommon::ImageViewInOut, MAX_IMAGE_ELEMENTS> views;
     std::array<VideoCommon::SamplerId, MAX_IMAGE_ELEMENTS> samplers;
     size_t sampler_index{};
@@ -507,8 +509,7 @@ bool GraphicsPipeline::ConfigureImpl(bool is_indexed) {
     return true;
 }
 
-void GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling,
-                                     const RenderAreaPushConstant& render_area) {
+void GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling, const RenderAreaPushConstant& render_area) {
     scheduler.RequestRenderpass(texture_cache.GetFramebuffer());
     if (!is_built.load(std::memory_order::relaxed)) {
         // Wait for the pipeline to be built

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -48,19 +48,8 @@ namespace Vulkan {
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 using MaxwellDrawState = Tegra::Engines::DrawManager::State;
 using VideoCommon::ImageViewId;
-using VideoCommon::ImageViewType;
-
 
 namespace {
-struct DrawParams {
-    u32 base_instance;
-    u32 num_instances;
-    u32 base_vertex;
-    u32 num_vertices;
-    u32 first_index;
-    bool is_indexed;
-};
-
 VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t index, float scale) {
     const auto& src = regs.viewport_transform[index];
     const auto conv = [scale](float value) {
@@ -151,8 +140,8 @@ VkRect2D GetScissorState(const Maxwell& regs, size_t index, u32 up_scale = 1, u3
     return scissor;
 }
 
-DrawParams MakeDrawParams(const MaxwellDrawState& draw_state, u32 num_instances, bool is_indexed) {
-    DrawParams params{
+RasterizerVulkan::DrawParams MakeDrawParams(const MaxwellDrawState& draw_state, u32 num_instances, bool is_indexed) {
+    RasterizerVulkan::DrawParams params{
         .base_instance = draw_state.base_instance,
         .num_instances = num_instances,
         .base_vertex = is_indexed ? draw_state.base_index : draw_state.vertex_buffer.first,
@@ -215,6 +204,8 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) {
     SCOPE_EXIT {
         gpu.TickWork();
     };
+    if (state_tracker.flags->count() > 0)
+        FlushBatchedDraws(); // Dirty state changes the way Vulkan would behave
     FlushWork();
     gpu_memory->FlushCaching();
 
@@ -225,75 +216,94 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) {
     std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
     // update engine as channel may be different.
     pipeline->SetEngine(maxwell3d, gpu_memory);
-    if (!pipeline->Configure(is_indexed))
-        return;
-
-    UpdateDynamicStates();
-
-    HandleTransformFeedback();
-    query_cache.CounterEnable(VideoCommon::QueryType::ZPassPixelCount64,
-                              maxwell3d->regs.zpass_pixel_count_enable);
-
-    draw_func();
+    if (pipeline->Configure(is_indexed)) {
+        UpdateDynamicStates();
+        HandleTransformFeedback();
+        query_cache.CounterEnable(VideoCommon::QueryType::ZPassPixelCount64, maxwell3d->regs.zpass_pixel_count_enable);
+        draw_func();
+    }
 }
 
-void RasterizerVulkan::Draw(bool is_indexed, u32 instance_count) {
-    PrepareDraw(is_indexed, [this, is_indexed, instance_count] {
-        const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
-        const u32 num_instances{instance_count};
-        const DrawParams draw_params{MakeDrawParams(draw_state, num_instances, is_indexed)};
-
-        // Use VK_EXT_multi_draw if available (single draw becomes multi-draw with count=1)
-        if (device.IsExtMultiDrawSupported()) {
-            scheduler.Record([draw_params](vk::CommandBuffer cmdbuf) {
-                if (draw_params.is_indexed) {
-                    // Use multi-draw indexed with single draw
-                    const VkMultiDrawIndexedInfoEXT multi_draw_info{
-                        .firstIndex = draw_params.first_index,
-                        .indexCount = draw_params.num_vertices,
-                    };
-                    const int32_t vertex_offset = static_cast<int32_t>(draw_params.base_vertex);
-                    cmdbuf.DrawMultiIndexedEXT(1, &multi_draw_info, draw_params.num_instances,
-                                               draw_params.base_instance,
-                                               sizeof(VkMultiDrawIndexedInfoEXT), &vertex_offset);
-                } else {
-                    // Use multi-draw with single draw
-                    const VkMultiDrawInfoEXT multi_draw_info{
-                        .firstVertex = draw_params.base_vertex,
-                        .vertexCount = draw_params.num_vertices,
-                    };
-                    cmdbuf.DrawMultiEXT(1, &multi_draw_info, draw_params.num_instances,
-                                        draw_params.base_instance,
-                                        sizeof(VkMultiDrawInfoEXT));
+void RasterizerVulkan::FlushBatchedDraws() {
+    auto const log_draw = [](RasterizerVulkan::DrawParams draw_params) {
+        // Log draw call
+        if (Settings::values.gpu_logging_enabled.GetValue() && Settings::values.gpu_log_vulkan_calls.GetValue()) {
+            const std::string params = draw_params.is_indexed ?
+                fmt::format("vertices={}, instances={}, firstIndex={}, baseVertex={}, baseInstance={}", draw_params.num_vertices, draw_params.num_instances, draw_params.first_index, draw_params.base_vertex, draw_params.base_instance) :
+                fmt::format("vertices={}, instances={}, firstVertex={}, firstInstance={}", draw_params.num_vertices, draw_params.num_instances, draw_params.base_vertex, draw_params.base_instance);
+            GPU::Logging::GPULogger::GetInstance().LogVulkanCall(draw_params.is_indexed ? "vkCmdDrawIndexed" : "vkCmdDraw", params, VK_SUCCESS);
+        }
+    };
+    (void)log_draw;
+    auto const& list = batched_draw_params;
+    if (list.empty())
+        return;
+    // Use VK_EXT_multi_draw if available (single draw becomes multi-draw with count=1)
+    for (size_t i = 0; i < list.size(); ) {
+        auto const first_i = i;
+        bool last_is_indexed = list[i].is_indexed;
+        u32 last_num_instances = list[i].num_instances;
+        u32 last_base_instance = list[i].base_instance;
+        // now for the sad draw part...
+        u32 count = 0;
+        if (last_is_indexed) {
+            std::vector<VkMultiDrawIndexedInfoEXT> multi_draw_info;
+            std::vector<int32_t> vertex_offset;
+            for (; i < list.size()
+            && list[i].is_indexed == last_is_indexed
+            && list[i].num_instances == last_num_instances
+            && list[i].base_instance == last_base_instance; ++i) {
+                multi_draw_info.push_back({
+                    .firstIndex = list[i].first_index,
+                    .indexCount = list[i].num_vertices
+                });
+                vertex_offset.push_back(list[i].base_vertex);
+                ++count;
+            }
+            if (device.IsExtMultiDrawSupported() && multi_draw_info.size() > 1) {
+                scheduler.Record([last_num_instances, last_base_instance, count, multi_draw_info, vertex_offset](vk::CommandBuffer cmdbuf) {
+                    cmdbuf.DrawMultiIndexedEXT(count, multi_draw_info.data(), last_num_instances, last_base_instance, sizeof(VkMultiDrawIndexedInfoEXT), vertex_offset.data());
+                });
+            } else {
+                for (size_t j = first_i; j < i; ++j) {
+                    scheduler.Record([draw_params = list[j]](vk::CommandBuffer cmdbuf) {
+                        cmdbuf.DrawIndexed(draw_params.num_vertices, draw_params.num_instances, draw_params.first_index, draw_params.base_vertex, draw_params.base_instance);
+                    });
                 }
-            });
+            }
         } else {
-            // Fallback to standard draw calls
-            scheduler.Record([draw_params](vk::CommandBuffer cmdbuf) {
-                if (draw_params.is_indexed) {
-                    cmdbuf.DrawIndexed(draw_params.num_vertices, draw_params.num_instances,
-                                       draw_params.first_index, draw_params.base_vertex,
-                                       draw_params.base_instance);
-                } else {
-                    cmdbuf.Draw(draw_params.num_vertices, draw_params.num_instances,
-                                draw_params.base_vertex, draw_params.base_instance);
+            std::vector<VkMultiDrawInfoEXT> multi_draw_info;
+            for (; i < list.size()
+            && list[i].is_indexed == last_is_indexed
+            && list[i].num_instances == last_num_instances
+            && list[i].base_instance == last_base_instance; ++i) {
+                multi_draw_info.push_back({
+                    .firstVertex = list[i].base_vertex,
+                    .vertexCount = list[i].num_vertices
+                });
+                ++count;
+            }
+            if (device.IsExtMultiDrawSupported() && multi_draw_info.size() > 1) {
+                scheduler.Record([last_num_instances, last_base_instance, count, multi_draw_info](vk::CommandBuffer cmdbuf) {
+                    cmdbuf.DrawMultiEXT(count, multi_draw_info.data(), last_num_instances, last_base_instance, sizeof(VkMultiDrawInfoEXT));
+                });
+            } else {
+                for (size_t j = first_i; j < i; ++j) {
+                    scheduler.Record([draw_params = list[j]](vk::CommandBuffer cmdbuf) {
+                        cmdbuf.Draw(draw_params.num_vertices, draw_params.num_instances, draw_params.base_vertex, draw_params.base_instance);
+                    });
                 }
-            });
+            }
         }
+    }
+    batched_draw_params.clear();
+}
 
-        // Log draw call
-        if (Settings::values.gpu_logging_enabled.GetValue() &&
-            Settings::values.gpu_log_vulkan_calls.GetValue()) {
-            const std::string params = is_indexed ?
-                fmt::format("vertices={}, instances={}, firstIndex={}, baseVertex={}, baseInstance={}",
-                    draw_params.num_vertices, draw_params.num_instances,
-                    draw_params.first_index, draw_params.base_vertex, draw_params.base_instance) :
-                fmt::format("vertices={}, instances={}, firstVertex={}, firstInstance={}",
-                    draw_params.num_vertices, draw_params.num_instances,
-                    draw_params.base_vertex, draw_params.base_instance);
-            GPU::Logging::GPULogger::GetInstance().LogVulkanCall(
-                is_indexed ? "vkCmdDrawIndexed" : "vkCmdDraw", params, VK_SUCCESS);
-        }
+void RasterizerVulkan::Draw(bool is_indexed, u32 instance_count) {
+    PrepareDraw(is_indexed, [this, is_indexed, instance_count] {
+        auto const& draw_state = maxwell3d->draw_manager->GetDrawState();
+        auto const num_instances = instance_count;
+        batched_draw_params.push_back(MakeDrawParams(draw_state, num_instances, is_indexed));
     });
 }
 
@@ -356,16 +366,16 @@ void RasterizerVulkan::DrawIndirect() {
 }
 
 void RasterizerVulkan::DrawTexture() {
-
     SCOPE_EXIT {
         gpu.TickWork();
     };
+    if (state_tracker.flags->count() > 0)
+        FlushBatchedDraws(); // Dirty state changes the way Vulkan would behave
     FlushWork();
 
     std::scoped_lock l{texture_cache.mutex};
     texture_cache.SynchronizeGraphicsDescriptors();
     texture_cache.UpdateRenderTargets(false);
-
     UpdateDynamicStates();
 
     query_cache.CounterEnable(VideoCommon::QueryType::ZPassPixelCount64,
@@ -403,6 +413,8 @@ void RasterizerVulkan::DrawTexture() {
 }
 
 void RasterizerVulkan::Clear(u32 layer_count) {
+    if (state_tracker.flags->count() > 0)
+        FlushBatchedDraws(); // Dirty state changes the way Vulkan would behave
     FlushWork();
     gpu_memory->FlushCaching();
 
@@ -559,6 +571,8 @@ void RasterizerVulkan::Clear(u32 layer_count) {
 }
 
 void RasterizerVulkan::DispatchCompute() {
+    if (state_tracker.flags->count() > 0)
+        FlushBatchedDraws(); // Dirty state changes the way Vulkan would behave
     FlushWork();
     gpu_memory->FlushCaching();
 
@@ -934,7 +948,6 @@ void RasterizerVulkan::FlushWork() {
     static constexpr u32 DRAWS_TO_DISPATCH = 4096;
     static constexpr u32 CHECK_MASK = 7;
 #endif // ANDROID
-
     static_assert(DRAWS_TO_DISPATCH % (CHECK_MASK + 1) == 0);
     if ((++draw_counter & CHECK_MASK) != CHECK_MASK) {
         return;

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -52,57 +52,48 @@ class StateTracker;
 
 class AccelerateDMA : public Tegra::Engines::AccelerateDMAInterface {
 public:
-    explicit AccelerateDMA(BufferCache& buffer_cache, TextureCache& texture_cache,
-                           Scheduler& scheduler);
-
+    explicit AccelerateDMA(BufferCache& buffer_cache, TextureCache& texture_cache, Scheduler& scheduler);
     bool BufferCopy(GPUVAddr start_address, GPUVAddr end_address, u64 amount) override;
-
     bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) override;
-
-    bool ImageToBuffer(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::ImageOperand& src,
-                       const Tegra::DMA::BufferOperand& dst) override;
-
-    bool BufferToImage(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& src,
-                       const Tegra::DMA::ImageOperand& dst) override;
-
+    bool ImageToBuffer(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::ImageOperand& src, const Tegra::DMA::BufferOperand& dst) override;
+    bool BufferToImage(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& src, const Tegra::DMA::ImageOperand& dst) override;
 private:
     template <bool IS_IMAGE_UPLOAD>
-    bool DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info,
-                            const Tegra::DMA::BufferOperand& src,
-                            const Tegra::DMA::ImageOperand& dst);
-
+    bool DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& src, const Tegra::DMA::ImageOperand& dst);
     BufferCache& buffer_cache;
     TextureCache& texture_cache;
     Scheduler& scheduler;
 };
 
-class RasterizerVulkan final : public VideoCore::RasterizerInterface,
-                               protected VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> {
+class RasterizerVulkan final : public VideoCore::RasterizerInterface, protected VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> {
 public:
-    explicit RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
-                              Tegra::MaxwellDeviceMemoryManager& device_memory_,
-                              const Device& device_, MemoryAllocator& memory_allocator_,
-                              StateTracker& state_tracker_, Scheduler& scheduler_);
+    struct DrawParams {
+        u32 base_instance;
+        u32 num_instances;
+        u32 base_vertex;
+        u32 num_vertices;
+        u32 first_index;
+        bool is_indexed;
+    };
+
+    explicit RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, Tegra::MaxwellDeviceMemoryManager& device_memory_, const Device& device_, MemoryAllocator& memory_allocator_, StateTracker& state_tracker_, Scheduler& scheduler_);
     ~RasterizerVulkan() override;
 
+    void FlushBatchedDraws() override;
     void Draw(bool is_indexed, u32 instance_count) override;
     void DrawIndirect() override;
     void DrawTexture() override;
     void Clear(u32 layer_count) override;
     void DispatchCompute() override;
     void ResetCounter(VideoCommon::QueryType type) override;
-    void Query(GPUVAddr gpu_addr, VideoCommon::QueryType type,
-               VideoCommon::QueryPropertiesFlags flags, u32 payload, u32 subreport) override;
+    void Query(GPUVAddr gpu_addr, VideoCommon::QueryType type, VideoCommon::QueryPropertiesFlags flags, u32 payload, u32 subreport) override;
     void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
     void DisableGraphicsUniformBuffer(size_t stage, u32 index) override;
     void FlushAll() override;
-    void FlushRegion(DAddr addr, u64 size,
-                     VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
-    bool MustFlushRegion(DAddr addr, u64 size,
-                         VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
+    void FlushRegion(DAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
+    bool MustFlushRegion(DAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
     VideoCore::RasterizerDownloadArea GetFlushArea(DAddr addr, u64 size) override;
-    void InvalidateRegion(DAddr addr, u64 size,
-                          VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
+    void InvalidateRegion(DAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
     void InnerInvalidation(std::span<const std::pair<DAddr, std::size_t>> sequences) override;
     void OnCacheInvalidation(DAddr addr, u64 size) override;
     bool OnCPUWrite(DAddr addr, u64 size) override;
@@ -114,8 +105,7 @@ public:
     void SignalSyncPoint(u32 value) override;
     void SignalReference() override;
     void ReleaseFences(bool force = true) override;
-    void FlushAndInvalidateRegion(
-        DAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
+    void FlushAndInvalidateRegion(DAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
     void WaitForIdle() override;
     void FragmentBarrier() override;
     void TiledCacheBarrier() override;
@@ -136,9 +126,7 @@ public:
     void BindChannel(Tegra::Control::ChannelState& channel) override;
 
     void ReleaseChannel(s32 channel_id) override;
-    std::optional<FramebufferTextureInfo> AccelerateDisplay(const Tegra::FramebufferConfig& config,
-                                                            VAddr framebuffer_addr,
-                                                            u32 pixel_stride);
+    std::optional<FramebufferTextureInfo> AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride);
 
 private:
     static constexpr const u64 NEEDS_D24[] = {
@@ -223,6 +211,7 @@ private:
     boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices;
     std::array<VideoCommon::ImageViewId, MAX_IMAGE_VIEWS> image_view_ids;
     boost::container::static_vector<VkSampler, MAX_TEXTURES> sampler_handles;
+    std::vector<DrawParams> batched_draw_params;
 
     u32 draw_counter = 0;
 };

src/video_core/renderer_vulkan/vk_state_tracker.h

@@ -301,6 +301,7 @@ private:
         return is_dirty;
     }
 
+public:
     struct StencilProperties {
         u32 ref = 0;
         u32 write_mask = 0;