diff options
Diffstat (limited to 'src/video_core/renderer_vulkan')
46 files changed, 921 insertions, 1057 deletions
diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp index 81a39a3b8..da5c550ea 100644 --- a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp +++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp @@ -58,6 +58,7 @@ void FixedPipelineState::Fill(const Maxwell& regs, bool has_extended_dynamic_sta logic_op_enable.Assign(regs.logic_op.enable != 0 ? 1 : 0); logic_op.Assign(PackLogicOp(regs.logic_op.operation)); rasterize_enable.Assign(regs.rasterize_enable != 0 ? 1 : 0); + topology.Assign(regs.draw.topology); std::memcpy(&point_size, ®s.point_size, sizeof(point_size)); // TODO: C++20 std::bit_cast @@ -131,7 +132,6 @@ void FixedPipelineState::BlendingAttachment::Fill(const Maxwell& regs, std::size } void FixedPipelineState::DynamicState::Fill(const Maxwell& regs) { - const u32 topology_index = static_cast<u32>(regs.draw.topology.Value()); u32 packed_front_face = PackFrontFace(regs.front_face); if (regs.screen_y_control.triangle_rast_flip != 0) { // Flip front face @@ -161,7 +161,6 @@ void FixedPipelineState::DynamicState::Fill(const Maxwell& regs) { depth_test_enable.Assign(regs.depth_test_enable); front_face.Assign(packed_front_face); depth_test_func.Assign(PackComparisonOp(regs.depth_test_func)); - topology.Assign(topology_index); cull_face.Assign(PackCullFace(regs.cull_face)); cull_enable.Assign(regs.cull_test_enabled != 0 ? 1 : 0); diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.h b/src/video_core/renderer_vulkan/fixed_pipeline_state.h index cdcbb65f5..2c18eeaae 100644 --- a/src/video_core/renderer_vulkan/fixed_pipeline_state.h +++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.h @@ -150,9 +150,8 @@ struct FixedPipelineState { }; union { u32 raw2; - BitField<0, 4, u32> topology; - BitField<4, 2, u32> cull_face; - BitField<6, 1, u32> cull_enable; + BitField<0, 2, u32> cull_face; + BitField<2, 1, u32> cull_enable; }; std::array<VertexBinding, Maxwell::NumVertexArrays> vertex_bindings; @@ -169,10 +168,6 @@ struct FixedPipelineState { Maxwell::FrontFace FrontFace() const noexcept { return UnpackFrontFace(front_face.Value()); } - - constexpr Maxwell::PrimitiveTopology Topology() const noexcept { - return static_cast<Maxwell::PrimitiveTopology>(topology.Value()); - } }; union { @@ -190,6 +185,7 @@ struct FixedPipelineState { BitField<18, 1, u32> logic_op_enable; BitField<19, 4, u32> logic_op; BitField<23, 1, u32> rasterize_enable; + BitField<24, 4, Maxwell::PrimitiveTopology> topology; }; u32 point_size; std::array<u32, Maxwell::NumVertexArrays> binding_divisors; diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp index f8c77f4fa..d22de1d81 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp @@ -78,9 +78,10 @@ VkSamplerAddressMode WrapMode(const VKDevice& device, Tegra::Texture::WrapMode w case Tegra::Texture::WrapMode::MirrorOnceBorder: UNIMPLEMENTED(); return VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE; + default: + UNIMPLEMENTED_MSG("Unimplemented wrap mode={}", static_cast<u32>(wrap_mode)); + return {}; } - UNIMPLEMENTED_MSG("Unimplemented wrap mode={}", static_cast<u32>(wrap_mode)); - return {}; } VkCompareOp DepthCompareFunction(Tegra::Texture::DepthCompareFunc depth_compare_func) { @@ -298,9 +299,10 @@ VkPrimitiveTopology PrimitiveTopology([[maybe_unused]] const VKDevice& device, return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST; case Maxwell::PrimitiveTopology::Patches: return VK_PRIMITIVE_TOPOLOGY_PATCH_LIST; + default: + UNIMPLEMENTED_MSG("Unimplemented topology={}", static_cast<u32>(topology)); + return {}; } - UNIMPLEMENTED_MSG("Unimplemented topology={}", static_cast<u32>(topology)); - return {}; } VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttribute::Size size) { @@ -325,6 +327,8 @@ VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttrib return VK_FORMAT_R16G16B16A16_UNORM; case Maxwell::VertexAttribute::Size::Size_10_10_10_2: return VK_FORMAT_A2B10G10R10_UNORM_PACK32; + default: + break; } break; case Maxwell::VertexAttribute::Type::SignedNorm: @@ -347,6 +351,8 @@ VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttrib return VK_FORMAT_R16G16B16A16_SNORM; case Maxwell::VertexAttribute::Size::Size_10_10_10_2: return VK_FORMAT_A2B10G10R10_SNORM_PACK32; + default: + break; } break; case Maxwell::VertexAttribute::Type::UnsignedScaled: @@ -369,6 +375,8 @@ VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttrib return VK_FORMAT_R16G16B16A16_USCALED; case Maxwell::VertexAttribute::Size::Size_10_10_10_2: return VK_FORMAT_A2B10G10R10_USCALED_PACK32; + default: + break; } break; case Maxwell::VertexAttribute::Type::SignedScaled: @@ -391,6 +399,8 @@ VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttrib return VK_FORMAT_R16G16B16A16_SSCALED; case Maxwell::VertexAttribute::Size::Size_10_10_10_2: return VK_FORMAT_A2B10G10R10_SSCALED_PACK32; + default: + break; } break; case Maxwell::VertexAttribute::Type::UnsignedInt: @@ -421,6 +431,8 @@ VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttrib return VK_FORMAT_R32G32B32A32_UINT; case Maxwell::VertexAttribute::Size::Size_10_10_10_2: return VK_FORMAT_A2B10G10R10_UINT_PACK32; + default: + break; } break; case Maxwell::VertexAttribute::Type::SignedInt: @@ -451,6 +463,8 @@ VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttrib return VK_FORMAT_R32G32B32A32_SINT; case Maxwell::VertexAttribute::Size::Size_10_10_10_2: return VK_FORMAT_A2B10G10R10_SINT_PACK32; + default: + break; } break; case Maxwell::VertexAttribute::Type::Float: @@ -471,6 +485,8 @@ VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttrib return VK_FORMAT_R32G32B32_SFLOAT; case Maxwell::VertexAttribute::Size::Size_32_32_32_32: return VK_FORMAT_R32G32B32A32_SFLOAT; + default: + break; } break; } diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp index 7ffc90cd0..f2610868e 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp +++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp @@ -25,9 +25,9 @@ #include "video_core/renderer_vulkan/renderer_vulkan.h" #include "video_core/renderer_vulkan/vk_blit_screen.h" #include "video_core/renderer_vulkan/vk_device.h" +#include "video_core/renderer_vulkan/vk_master_semaphore.h" #include "video_core/renderer_vulkan/vk_memory_manager.h" #include "video_core/renderer_vulkan/vk_rasterizer.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_state_tracker.h" #include "video_core/renderer_vulkan/vk_swapchain.h" @@ -56,7 +56,7 @@ VkBool32 DebugCallback(VkDebugUtilsMessageSeverityFlagBitsEXT severity, VkDebugUtilsMessageTypeFlagsEXT type, const VkDebugUtilsMessengerCallbackDataEXT* data, [[maybe_unused]] void* user_data) { - const char* message{data->pMessage}; + const char* const message{data->pMessage}; if (severity & VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT) { LOG_CRITICAL(Render_Vulkan, "{}", message); @@ -86,7 +86,7 @@ Common::DynamicLibrary OpenVulkanLibrary() { if (!library.Open(filename.c_str())) { // Android devices may not have libvulkan.so.1, only libvulkan.so. filename = Common::DynamicLibrary::GetVersionedFilename("vulkan"); - library.Open(filename.c_str()); + (void)library.Open(filename.c_str()); } #endif return library; @@ -240,8 +240,12 @@ std::string BuildCommaSeparatedExtensions(std::vector<std::string> available_ext } // Anonymous namespace -RendererVulkan::RendererVulkan(Core::Frontend::EmuWindow& window, Core::System& system) - : RendererBase(window), system{system} {} +RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_, + Core::Frontend::EmuWindow& emu_window, + Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_, + std::unique_ptr<Core::Frontend::GraphicsContext> context) + : RendererBase{emu_window, std::move(context)}, telemetry_session{telemetry_session_}, + cpu_memory{cpu_memory_}, gpu{gpu_} {} RendererVulkan::~RendererVulkan() { ShutDown(); @@ -268,11 +272,11 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { scheduler->WaitWorker(); swapchain->AcquireNextImage(); - const auto [fence, render_semaphore] = blit_screen->Draw(*framebuffer, use_accelerated); + const VkSemaphore render_semaphore = blit_screen->Draw(*framebuffer, use_accelerated); - scheduler->Flush(false, render_semaphore); + scheduler->Flush(render_semaphore); - if (swapchain->Present(render_semaphore, fence)) { + if (swapchain->Present(render_semaphore)) { blit_screen->Recreate(); } @@ -282,11 +286,6 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { render_window.PollEvents(); } -bool RendererVulkan::TryPresent(int /*timeout_ms*/) { - // TODO (bunnei): ImplementMe - return true; -} - bool RendererVulkan::Init() { library = OpenVulkanLibrary(); std::tie(instance, instance_version) = CreateInstance( @@ -299,23 +298,21 @@ bool RendererVulkan::Init() { memory_manager = std::make_unique<VKMemoryManager>(*device); - resource_manager = std::make_unique<VKResourceManager>(*device); + state_tracker = std::make_unique<StateTracker>(gpu); + + scheduler = std::make_unique<VKScheduler>(*device, *state_tracker); const auto& framebuffer = render_window.GetFramebufferLayout(); - swapchain = std::make_unique<VKSwapchain>(*surface, *device); + swapchain = std::make_unique<VKSwapchain>(*surface, *device, *scheduler); swapchain->Create(framebuffer.width, framebuffer.height, false); - state_tracker = std::make_unique<StateTracker>(system); - - scheduler = std::make_unique<VKScheduler>(*device, *resource_manager, *state_tracker); - - rasterizer = std::make_unique<RasterizerVulkan>(system, render_window, screen_info, *device, - *resource_manager, *memory_manager, - *state_tracker, *scheduler); + rasterizer = std::make_unique<RasterizerVulkan>(render_window, gpu, gpu.MemoryManager(), + cpu_memory, screen_info, *device, + *memory_manager, *state_tracker, *scheduler); - blit_screen = std::make_unique<VKBlitScreen>(system, render_window, *rasterizer, *device, - *resource_manager, *memory_manager, *swapchain, - *scheduler, screen_info); + blit_screen = + std::make_unique<VKBlitScreen>(cpu_memory, render_window, *rasterizer, *device, + *memory_manager, *swapchain, *scheduler, screen_info); return true; } @@ -333,7 +330,6 @@ void RendererVulkan::ShutDown() { scheduler.reset(); swapchain.reset(); memory_manager.reset(); - resource_manager.reset(); device.reset(); } @@ -442,8 +438,7 @@ void RendererVulkan::Report() const { LOG_INFO(Render_Vulkan, "Device: {}", model_name); LOG_INFO(Render_Vulkan, "Vulkan: {}", api_version); - auto& telemetry_session = system.TelemetrySession(); - constexpr auto field = Common::Telemetry::FieldType::UserSystem; + static constexpr auto field = Common::Telemetry::FieldType::UserSystem; telemetry_session.AddField(field, "GPU_Vendor", vendor_name); telemetry_session.AddField(field, "GPU_Model", model_name); telemetry_session.AddField(field, "GPU_Vulkan_Driver", driver_name); diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.h b/src/video_core/renderer_vulkan/renderer_vulkan.h index 9617a93e9..1044ca124 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.h +++ b/src/video_core/renderer_vulkan/renderer_vulkan.h @@ -14,7 +14,15 @@ #include "video_core/renderer_vulkan/wrapper.h" namespace Core { -class System; +class TelemetrySession; +} + +namespace Core::Memory { +class Memory; +} + +namespace Tegra { +class GPU; } namespace Vulkan { @@ -22,9 +30,7 @@ namespace Vulkan { class StateTracker; class VKBlitScreen; class VKDevice; -class VKFence; class VKMemoryManager; -class VKResourceManager; class VKSwapchain; class VKScheduler; class VKImage; @@ -38,13 +44,15 @@ struct VKScreenInfo { class RendererVulkan final : public VideoCore::RendererBase { public: - explicit RendererVulkan(Core::Frontend::EmuWindow& window, Core::System& system); + explicit RendererVulkan(Core::TelemetrySession& telemtry_session, + Core::Frontend::EmuWindow& emu_window, Core::Memory::Memory& cpu_memory, + Tegra::GPU& gpu, + std::unique_ptr<Core::Frontend::GraphicsContext> context); ~RendererVulkan() override; bool Init() override; void ShutDown() override; void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override; - bool TryPresent(int timeout_ms) override; static std::vector<std::string> EnumerateDevices(); @@ -57,7 +65,9 @@ private: void Report() const; - Core::System& system; + Core::TelemetrySession& telemetry_session; + Core::Memory::Memory& cpu_memory; + Tegra::GPU& gpu; Common::DynamicLibrary library; vk::InstanceDispatch dld; @@ -71,11 +81,10 @@ private: vk::DebugCallback debug_callback; std::unique_ptr<VKDevice> device; - std::unique_ptr<VKSwapchain> swapchain; std::unique_ptr<VKMemoryManager> memory_manager; - std::unique_ptr<VKResourceManager> resource_manager; std::unique_ptr<StateTracker> state_tracker; std::unique_ptr<VKScheduler> scheduler; + std::unique_ptr<VKSwapchain> swapchain; std::unique_ptr<VKBlitScreen> blit_screen; }; diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp index a551e3de8..b5b60309e 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp +++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp @@ -12,11 +12,9 @@ #include "common/assert.h" #include "common/common_types.h" #include "common/math_util.h" - #include "core/core.h" #include "core/frontend/emu_window.h" #include "core/memory.h" - #include "video_core/gpu.h" #include "video_core/morton.h" #include "video_core/rasterizer_interface.h" @@ -24,8 +22,8 @@ #include "video_core/renderer_vulkan/vk_blit_screen.h" #include "video_core/renderer_vulkan/vk_device.h" #include "video_core/renderer_vulkan/vk_image.h" +#include "video_core/renderer_vulkan/vk_master_semaphore.h" #include "video_core/renderer_vulkan/vk_memory_manager.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_shader_util.h" #include "video_core/renderer_vulkan/vk_swapchain.h" @@ -210,17 +208,15 @@ struct VKBlitScreen::BufferData { // Unaligned image data goes here }; -VKBlitScreen::VKBlitScreen(Core::System& system, Core::Frontend::EmuWindow& render_window, - VideoCore::RasterizerInterface& rasterizer, const VKDevice& device, - VKResourceManager& resource_manager, VKMemoryManager& memory_manager, - VKSwapchain& swapchain, VKScheduler& scheduler, - const VKScreenInfo& screen_info) - : system{system}, render_window{render_window}, rasterizer{rasterizer}, device{device}, - resource_manager{resource_manager}, memory_manager{memory_manager}, swapchain{swapchain}, - scheduler{scheduler}, image_count{swapchain.GetImageCount()}, screen_info{screen_info} { - watches.resize(image_count); - std::generate(watches.begin(), watches.end(), - []() { return std::make_unique<VKFenceWatch>(); }); +VKBlitScreen::VKBlitScreen(Core::Memory::Memory& cpu_memory_, + Core::Frontend::EmuWindow& render_window_, + VideoCore::RasterizerInterface& rasterizer_, const VKDevice& device_, + VKMemoryManager& memory_manager_, VKSwapchain& swapchain_, + VKScheduler& scheduler_, const VKScreenInfo& screen_info_) + : cpu_memory{cpu_memory_}, render_window{render_window_}, rasterizer{rasterizer_}, + device{device_}, memory_manager{memory_manager_}, swapchain{swapchain_}, + scheduler{scheduler_}, image_count{swapchain.GetImageCount()}, screen_info{screen_info_} { + resource_ticks.resize(image_count); CreateStaticResources(); CreateDynamicResources(); @@ -232,15 +228,16 @@ void VKBlitScreen::Recreate() { CreateDynamicResources(); } -std::tuple<VKFence&, VkSemaphore> VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, - bool use_accelerated) { +VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool use_accelerated) { RefreshResources(framebuffer); // Finish any pending renderpass scheduler.RequestOutsideRenderPassOperationContext(); const std::size_t image_index = swapchain.GetImageIndex(); - watches[image_index]->Watch(scheduler.GetFence()); + + scheduler.Wait(resource_ticks[image_index]); + resource_ticks[image_index] = scheduler.CurrentTick(); VKImage* blit_image = use_accelerated ? screen_info.image : raw_images[image_index].get(); @@ -259,7 +256,7 @@ std::tuple<VKFence&, VkSemaphore> VKBlitScreen::Draw(const Tegra::FramebufferCon const auto pixel_format = VideoCore::Surface::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format); const VAddr framebuffer_addr = framebuffer.address + framebuffer.offset; - const auto host_ptr = system.Memory().GetPointer(framebuffer_addr); + const auto host_ptr = cpu_memory.GetPointer(framebuffer_addr); rasterizer.FlushRegion(ToCacheAddr(host_ptr), GetSizeInBytes(framebuffer)); // TODO(Rodrigo): Read this from HLE @@ -343,7 +340,7 @@ std::tuple<VKFence&, VkSemaphore> VKBlitScreen::Draw(const Tegra::FramebufferCon cmdbuf.EndRenderPass(); }); - return {scheduler.GetFence(), *semaphores[image_index]}; + return *semaphores[image_index]; } void VKBlitScreen::CreateStaticResources() { @@ -711,7 +708,7 @@ void VKBlitScreen::CreateFramebuffers() { void VKBlitScreen::ReleaseRawImages() { for (std::size_t i = 0; i < raw_images.size(); ++i) { - watches[i]->Wait(); + scheduler.Wait(resource_ticks.at(i)); } raw_images.clear(); raw_buffer_commits.clear(); diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.h b/src/video_core/renderer_vulkan/vk_blit_screen.h index 243640fab..8f2839214 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.h +++ b/src/video_core/renderer_vulkan/vk_blit_screen.h @@ -5,16 +5,18 @@ #pragma once #include <memory> -#include <tuple> #include "video_core/renderer_vulkan/vk_memory_manager.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" #include "video_core/renderer_vulkan/wrapper.h" namespace Core { class System; } +namespace Core::Memory { +class Memory; +} + namespace Core::Frontend { class EmuWindow; } @@ -30,26 +32,26 @@ class RasterizerInterface; namespace Vulkan { struct ScreenInfo; + class RasterizerVulkan; class VKDevice; -class VKFence; class VKImage; class VKScheduler; class VKSwapchain; class VKBlitScreen final { public: - explicit VKBlitScreen(Core::System& system, Core::Frontend::EmuWindow& render_window, + explicit VKBlitScreen(Core::Memory::Memory& cpu_memory, + Core::Frontend::EmuWindow& render_window, VideoCore::RasterizerInterface& rasterizer, const VKDevice& device, - VKResourceManager& resource_manager, VKMemoryManager& memory_manager, - VKSwapchain& swapchain, VKScheduler& scheduler, - const VKScreenInfo& screen_info); + VKMemoryManager& memory_manager, VKSwapchain& swapchain, + VKScheduler& scheduler, const VKScreenInfo& screen_info); ~VKBlitScreen(); void Recreate(); - std::tuple<VKFence&, VkSemaphore> Draw(const Tegra::FramebufferConfig& framebuffer, - bool use_accelerated); + [[nodiscard]] VkSemaphore Draw(const Tegra::FramebufferConfig& framebuffer, + bool use_accelerated); private: struct BufferData; @@ -81,11 +83,10 @@ private: u64 GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer, std::size_t image_index) const; - Core::System& system; + Core::Memory::Memory& cpu_memory; Core::Frontend::EmuWindow& render_window; VideoCore::RasterizerInterface& rasterizer; const VKDevice& device; - VKResourceManager& resource_manager; VKMemoryManager& memory_manager; VKSwapchain& swapchain; VKScheduler& scheduler; @@ -106,7 +107,7 @@ private: vk::Buffer buffer; VKMemoryCommit buffer_commit; - std::vector<std::unique_ptr<VKFenceWatch>> watches; + std::vector<u64> resource_ticks; std::vector<vk::Semaphore> semaphores; std::vector<std::unique_ptr<VKImage>> raw_images; diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 1d2f8b557..d9d3da9ea 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -145,14 +145,15 @@ void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst }); } -VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system, - const VKDevice& device, VKMemoryManager& memory_manager, - VKScheduler& scheduler, VKStagingBufferPool& staging_pool) - : VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer>{rasterizer, system, - CreateStreamBuffer(device, - scheduler)}, - device{device}, memory_manager{memory_manager}, scheduler{scheduler}, staging_pool{ - staging_pool} {} +VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer, + Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory, + const VKDevice& device_, VKMemoryManager& memory_manager_, + VKScheduler& scheduler_, VKStagingBufferPool& staging_pool_) + : VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer>{rasterizer, gpu_memory, cpu_memory, + CreateStreamBuffer(device_, + scheduler_)}, + device{device_}, memory_manager{memory_manager_}, scheduler{scheduler_}, staging_pool{ + staging_pool_} {} VKBufferCache::~VKBufferCache() = default; diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index 991ee451c..7fb5ceedf 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -13,10 +13,6 @@ #include "video_core/renderer_vulkan/vk_stream_buffer.h" #include "video_core/renderer_vulkan/wrapper.h" -namespace Core { -class System; -} - namespace Vulkan { class VKDevice; @@ -53,7 +49,8 @@ private: class VKBufferCache final : public VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer> { public: - explicit VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system, + explicit VKBufferCache(VideoCore::RasterizerInterface& rasterizer, + Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory, const VKDevice& device, VKMemoryManager& memory_manager, VKScheduler& scheduler, VKStagingBufferPool& staging_pool); ~VKBufferCache(); diff --git a/src/video_core/renderer_vulkan/vk_command_pool.cpp b/src/video_core/renderer_vulkan/vk_command_pool.cpp new file mode 100644 index 000000000..6339f4fe0 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_command_pool.cpp @@ -0,0 +1,46 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <cstddef> + +#include "video_core/renderer_vulkan/vk_command_pool.h" +#include "video_core/renderer_vulkan/vk_device.h" +#include "video_core/renderer_vulkan/wrapper.h" + +namespace Vulkan { + +constexpr size_t COMMAND_BUFFER_POOL_SIZE = 0x1000; + +struct CommandPool::Pool { + vk::CommandPool handle; + vk::CommandBuffers cmdbufs; +}; + +CommandPool::CommandPool(MasterSemaphore& master_semaphore, const VKDevice& device) + : ResourcePool(master_semaphore, COMMAND_BUFFER_POOL_SIZE), device{device} {} + +CommandPool::~CommandPool() = default; + +void CommandPool::Allocate(size_t begin, size_t end) { + // Command buffers are going to be commited, recorded, executed every single usage cycle. + // They are also going to be reseted when commited. + Pool& pool = pools.emplace_back(); + pool.handle = device.GetLogical().CreateCommandPool({ + .sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO, + .pNext = nullptr, + .flags = + VK_COMMAND_POOL_CREATE_TRANSIENT_BIT | VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, + .queueFamilyIndex = device.GetGraphicsFamily(), + }); + pool.cmdbufs = pool.handle.Allocate(COMMAND_BUFFER_POOL_SIZE); +} + +VkCommandBuffer CommandPool::Commit() { + const size_t index = CommitResource(); + const auto pool_index = index / COMMAND_BUFFER_POOL_SIZE; + const auto sub_index = index % COMMAND_BUFFER_POOL_SIZE; + return pools[pool_index].cmdbufs[sub_index]; +} + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_command_pool.h b/src/video_core/renderer_vulkan/vk_command_pool.h new file mode 100644 index 000000000..b9cb3fb5d --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_command_pool.h @@ -0,0 +1,34 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <cstddef> +#include <vector> + +#include "video_core/renderer_vulkan/vk_resource_pool.h" +#include "video_core/renderer_vulkan/wrapper.h" + +namespace Vulkan { + +class MasterSemaphore; +class VKDevice; + +class CommandPool final : public ResourcePool { +public: + explicit CommandPool(MasterSemaphore& master_semaphore, const VKDevice& device); + ~CommandPool() override; + + void Allocate(size_t begin, size_t end) override; + + VkCommandBuffer Commit(); + +private: + struct Pool; + + const VKDevice& device; + std::vector<Pool> pools; +}; + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 182461ed9..9637c6059 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -112,7 +112,8 @@ constexpr u8 quad_array[] = { 0xf9, 0x00, 0x02, 0x00, 0x21, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x23, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x4b, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x4e, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x4c, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x4b, 0x00, 0x00, 0x00, - 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00}; + 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00, +}; VkDescriptorSetLayoutBinding BuildQuadArrayPassDescriptorSetLayoutBinding() { return { @@ -218,7 +219,8 @@ constexpr u8 uint8_pass[] = { 0x2a, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, - 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00}; + 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00, +}; // Quad indexed SPIR-V module. Generated from the "shaders/" directory. constexpr u8 QUAD_INDEXED_SPV[] = { @@ -341,7 +343,8 @@ constexpr u8 QUAD_INDEXED_SPV[] = { 0xf9, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x37, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x73, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x76, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x73, 0x00, 0x00, 0x00, - 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00}; + 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00, +}; std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBindings() { return {{ @@ -448,12 +451,12 @@ VKComputePass::VKComputePass(const VKDevice& device, VKDescriptorPool& descripto VKComputePass::~VKComputePass() = default; -VkDescriptorSet VKComputePass::CommitDescriptorSet(VKUpdateDescriptorQueue& update_descriptor_queue, - VKFence& fence) { +VkDescriptorSet VKComputePass::CommitDescriptorSet( + VKUpdateDescriptorQueue& update_descriptor_queue) { if (!descriptor_template) { return nullptr; } - const auto set = descriptor_allocator->Commit(fence); + const VkDescriptorSet set = descriptor_allocator->Commit(); update_descriptor_queue.Send(*descriptor_template, set); return set; } @@ -477,7 +480,7 @@ std::pair<VkBuffer, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32 update_descriptor_queue.Acquire(); update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size); - const auto set = CommitDescriptorSet(update_descriptor_queue, scheduler.GetFence()); + const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue); scheduler.RequestOutsideRenderPassOperationContext(); @@ -520,13 +523,13 @@ Uint8Pass::~Uint8Pass() = default; std::pair<VkBuffer, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buffer, u64 src_offset) { - const auto staging_size = static_cast<u32>(num_vertices * sizeof(u16)); + const u32 staging_size = static_cast<u32>(num_vertices * sizeof(u16)); auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false); update_descriptor_queue.Acquire(); update_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices); update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size); - const auto set = CommitDescriptorSet(update_descriptor_queue, scheduler.GetFence()); + const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue); scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, set, @@ -589,7 +592,7 @@ std::pair<VkBuffer, u64> QuadIndexedPass::Assemble( update_descriptor_queue.Acquire(); update_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size); update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size); - const auto set = CommitDescriptorSet(update_descriptor_queue, scheduler.GetFence()); + const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue); scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, set, diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h index 230b526bc..acc94f27e 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.h +++ b/src/video_core/renderer_vulkan/vk_compute_pass.h @@ -15,7 +15,6 @@ namespace Vulkan { class VKDevice; -class VKFence; class VKScheduler; class VKStagingBufferPool; class VKUpdateDescriptorQueue; @@ -30,8 +29,7 @@ public: ~VKComputePass(); protected: - VkDescriptorSet CommitDescriptorSet(VKUpdateDescriptorQueue& update_descriptor_queue, - VKFence& fence); + VkDescriptorSet CommitDescriptorSet(VKUpdateDescriptorQueue& update_descriptor_queue); vk::DescriptorUpdateTemplateKHR descriptor_template; vk::PipelineLayout layout; diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp index ed9d2991c..9be72dc9b 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp @@ -32,7 +32,7 @@ VkDescriptorSet VKComputePipeline::CommitDescriptorSet() { if (!descriptor_template) { return {}; } - const auto set = descriptor_allocator.Commit(scheduler.GetFence()); + const VkDescriptorSet set = descriptor_allocator.Commit(); update_descriptor_queue.Send(*descriptor_template, set); return set; } diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp index ac4a0884e..f38e089d5 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp @@ -7,7 +7,8 @@ #include "common/common_types.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/renderer_vulkan/vk_device.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" +#include "video_core/renderer_vulkan/vk_resource_pool.h" +#include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/wrapper.h" namespace Vulkan { @@ -15,14 +16,15 @@ namespace Vulkan { // Prefer small grow rates to avoid saturating the descriptor pool with barely used pipelines. constexpr std::size_t SETS_GROW_RATE = 0x20; -DescriptorAllocator::DescriptorAllocator(VKDescriptorPool& descriptor_pool, - VkDescriptorSetLayout layout) - : VKFencedPool{SETS_GROW_RATE}, descriptor_pool{descriptor_pool}, layout{layout} {} +DescriptorAllocator::DescriptorAllocator(VKDescriptorPool& descriptor_pool_, + VkDescriptorSetLayout layout_) + : ResourcePool(descriptor_pool_.master_semaphore, SETS_GROW_RATE), + descriptor_pool{descriptor_pool_}, layout{layout_} {} DescriptorAllocator::~DescriptorAllocator() = default; -VkDescriptorSet DescriptorAllocator::Commit(VKFence& fence) { - const std::size_t index = CommitResource(fence); +VkDescriptorSet DescriptorAllocator::Commit() { + const std::size_t index = CommitResource(); return descriptors_allocations[index / SETS_GROW_RATE][index % SETS_GROW_RATE]; } @@ -30,8 +32,9 @@ void DescriptorAllocator::Allocate(std::size_t begin, std::size_t end) { descriptors_allocations.push_back(descriptor_pool.AllocateDescriptors(layout, end - begin)); } -VKDescriptorPool::VKDescriptorPool(const VKDevice& device) - : device{device}, active_pool{AllocateNewPool()} {} +VKDescriptorPool::VKDescriptorPool(const VKDevice& device_, VKScheduler& scheduler) + : device{device_}, master_semaphore{scheduler.GetMasterSemaphore()}, active_pool{ + AllocateNewPool()} {} VKDescriptorPool::~VKDescriptorPool() = default; diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.h b/src/video_core/renderer_vulkan/vk_descriptor_pool.h index 9efa66bef..544f32a20 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.h +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.h @@ -6,21 +6,24 @@ #include <vector> -#include "video_core/renderer_vulkan/vk_resource_manager.h" +#include "video_core/renderer_vulkan/vk_resource_pool.h" #include "video_core/renderer_vulkan/wrapper.h" namespace Vulkan { +class VKDevice; class VKDescriptorPool; +class VKScheduler; -class DescriptorAllocator final : public VKFencedPool { +class DescriptorAllocator final : public ResourcePool { public: explicit DescriptorAllocator(VKDescriptorPool& descriptor_pool, VkDescriptorSetLayout layout); ~DescriptorAllocator() override; + DescriptorAllocator& operator=(const DescriptorAllocator&) = delete; DescriptorAllocator(const DescriptorAllocator&) = delete; - VkDescriptorSet Commit(VKFence& fence); + VkDescriptorSet Commit(); protected: void Allocate(std::size_t begin, std::size_t end) override; @@ -36,15 +39,19 @@ class VKDescriptorPool final { friend DescriptorAllocator; public: - explicit VKDescriptorPool(const VKDevice& device); + explicit VKDescriptorPool(const VKDevice& device, VKScheduler& scheduler); ~VKDescriptorPool(); + VKDescriptorPool(const VKDescriptorPool&) = delete; + VKDescriptorPool& operator=(const VKDescriptorPool&) = delete; + private: vk::DescriptorPool* AllocateNewPool(); vk::DescriptorSets AllocateDescriptors(VkDescriptorSetLayout layout, std::size_t count); const VKDevice& device; + MasterSemaphore& master_semaphore; std::vector<vk::DescriptorPool> pools; vk::DescriptorPool* active_pool; diff --git a/src/video_core/renderer_vulkan/vk_device.cpp b/src/video_core/renderer_vulkan/vk_device.cpp index 90916ee0e..e1217ca83 100644 --- a/src/video_core/renderer_vulkan/vk_device.cpp +++ b/src/video_core/renderer_vulkan/vk_device.cpp @@ -45,6 +45,7 @@ constexpr std::array REQUIRED_EXTENSIONS{ VK_KHR_8BIT_STORAGE_EXTENSION_NAME, VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME, VK_KHR_DESCRIPTOR_UPDATE_TEMPLATE_EXTENSION_NAME, + VK_KHR_TIMELINE_SEMAPHORE_EXTENSION_NAME, VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME, VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME, VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME, @@ -81,6 +82,21 @@ VkFormatFeatureFlags GetFormatFeatures(VkFormatProperties properties, FormatType } } +[[nodiscard]] bool IsRDNA(std::string_view device_name, VkDriverIdKHR driver_id) { + static constexpr std::array RDNA_DEVICES{ + "5700", + "5600", + "5500", + "5300", + }; + if (driver_id != VK_DRIVER_ID_AMD_PROPRIETARY_KHR) { + return false; + } + return std::any_of(RDNA_DEVICES.begin(), RDNA_DEVICES.end(), [device_name](const char* name) { + return device_name.find(name) != std::string_view::npos; + }); +} + std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties( vk::PhysicalDevice physical, const vk::InstanceDispatch& dld) { static constexpr std::array formats{ @@ -253,6 +269,13 @@ bool VKDevice::Create() { .inheritedQueries = false, }; + VkPhysicalDeviceTimelineSemaphoreFeaturesKHR timeline_semaphore{ + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES_KHR, + .pNext = nullptr, + .timelineSemaphore = true, + }; + SetNext(next, timeline_semaphore); + VkPhysicalDevice16BitStorageFeaturesKHR bit16_storage{ .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES_KHR, .pNext = nullptr, @@ -383,6 +406,15 @@ bool VKDevice::Create() { CollectTelemetryParameters(); + if (ext_extended_dynamic_state && IsRDNA(properties.deviceName, driver_id)) { + // AMD's proprietary driver supports VK_EXT_extended_dynamic_state but on RDNA devices it + // seems to cause stability issues + LOG_WARNING( + Render_Vulkan, + "Blacklisting AMD proprietary on RDNA devices from VK_EXT_extended_dynamic_state"); + ext_extended_dynamic_state = false; + } + graphics_queue = logical.GetQueue(graphics_family); present_queue = logical.GetQueue(present_family); diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.cpp b/src/video_core/renderer_vulkan/vk_fence_manager.cpp index d7f65d435..5babbdd0b 100644 --- a/src/video_core/renderer_vulkan/vk_fence_manager.cpp +++ b/src/video_core/renderer_vulkan/vk_fence_manager.cpp @@ -29,8 +29,8 @@ void InnerFence::Queue() { } ASSERT(!event); - event = device.GetLogical().CreateNewEvent(); - ticks = scheduler.Ticks(); + event = device.GetLogical().CreateEvent(); + ticks = scheduler.CurrentTick(); scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([event = *event](vk::CommandBuffer cmdbuf) { @@ -52,7 +52,7 @@ void InnerFence::Wait() { } ASSERT(event); - if (ticks >= scheduler.Ticks()) { + if (ticks >= scheduler.CurrentTick()) { scheduler.Flush(); } while (!IsEventSignalled()) { @@ -71,12 +71,12 @@ bool InnerFence::IsEventSignalled() const { } } -VKFenceManager::VKFenceManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer, - const VKDevice& device, VKScheduler& scheduler, - VKTextureCache& texture_cache, VKBufferCache& buffer_cache, - VKQueryCache& query_cache) - : GenericFenceManager(system, rasterizer, texture_cache, buffer_cache, query_cache), - device{device}, scheduler{scheduler} {} +VKFenceManager::VKFenceManager(VideoCore::RasterizerInterface& rasterizer, Tegra::GPU& gpu, + Tegra::MemoryManager& memory_manager, VKTextureCache& texture_cache, + VKBufferCache& buffer_cache, VKQueryCache& query_cache, + const VKDevice& device_, VKScheduler& scheduler_) + : GenericFenceManager(rasterizer, gpu, texture_cache, buffer_cache, query_cache), + device{device_}, scheduler{scheduler_} {} Fence VKFenceManager::CreateFence(u32 value, bool is_stubbed) { return std::make_shared<InnerFence>(device, scheduler, value, is_stubbed); diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.h b/src/video_core/renderer_vulkan/vk_fence_manager.h index 043fe7947..1547d6d30 100644 --- a/src/video_core/renderer_vulkan/vk_fence_manager.h +++ b/src/video_core/renderer_vulkan/vk_fence_manager.h @@ -55,10 +55,10 @@ using GenericFenceManager = class VKFenceManager final : public GenericFenceManager { public: - explicit VKFenceManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer, - const VKDevice& device, VKScheduler& scheduler, - VKTextureCache& texture_cache, VKBufferCache& buffer_cache, - VKQueryCache& query_cache); + explicit VKFenceManager(VideoCore::RasterizerInterface& rasterizer, Tegra::GPU& gpu, + Tegra::MemoryManager& memory_manager, VKTextureCache& texture_cache, + VKBufferCache& buffer_cache, VKQueryCache& query_cache, + const VKDevice& device, VKScheduler& scheduler); protected: Fence CreateFence(u32 value, bool is_stubbed) override; diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index 2e46c6278..696eaeb5f 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -93,7 +93,7 @@ VkDescriptorSet VKGraphicsPipeline::CommitDescriptorSet() { if (!descriptor_template) { return {}; } - const auto set = descriptor_allocator.Commit(scheduler.GetFence()); + const VkDescriptorSet set = descriptor_allocator.Commit(); update_descriptor_queue.Send(*descriptor_template, set); return set; } @@ -261,12 +261,12 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa vertex_input_ci.pNext = &input_divisor_ci; } - const auto input_assembly_topology = MaxwellToVK::PrimitiveTopology(device, dynamic.Topology()); + const auto input_assembly_topology = MaxwellToVK::PrimitiveTopology(device, state.topology); const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO, .pNext = nullptr, .flags = 0, - .topology = MaxwellToVK::PrimitiveTopology(device, dynamic.Topology()), + .topology = MaxwellToVK::PrimitiveTopology(device, state.topology), .primitiveRestartEnable = state.primitive_restart_enable != 0 && SupportsPrimitiveRestart(input_assembly_topology), }; @@ -400,7 +400,6 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa static constexpr std::array extended{ VK_DYNAMIC_STATE_CULL_MODE_EXT, VK_DYNAMIC_STATE_FRONT_FACE_EXT, - VK_DYNAMIC_STATE_PRIMITIVE_TOPOLOGY_EXT, VK_DYNAMIC_STATE_VERTEX_INPUT_BINDING_STRIDE_EXT, VK_DYNAMIC_STATE_DEPTH_TEST_ENABLE_EXT, VK_DYNAMIC_STATE_DEPTH_WRITE_ENABLE_EXT, diff --git a/src/video_core/renderer_vulkan/vk_master_semaphore.cpp b/src/video_core/renderer_vulkan/vk_master_semaphore.cpp new file mode 100644 index 000000000..ae26e558d --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_master_semaphore.cpp @@ -0,0 +1,56 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <atomic> +#include <chrono> + +#include "core/settings.h" +#include "video_core/renderer_vulkan/vk_device.h" +#include "video_core/renderer_vulkan/vk_master_semaphore.h" +#include "video_core/renderer_vulkan/wrapper.h" + +namespace Vulkan { + +using namespace std::chrono_literals; + +MasterSemaphore::MasterSemaphore(const VKDevice& device) { + static constexpr VkSemaphoreTypeCreateInfoKHR semaphore_type_ci{ + .sType = VK_STRUCTURE_TYPE_SEMAPHORE_TYPE_CREATE_INFO_KHR, + .pNext = nullptr, + .semaphoreType = VK_SEMAPHORE_TYPE_TIMELINE_KHR, + .initialValue = 0, + }; + static constexpr VkSemaphoreCreateInfo semaphore_ci{ + .sType = VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO, + .pNext = &semaphore_type_ci, + .flags = 0, + }; + semaphore = device.GetLogical().CreateSemaphore(semaphore_ci); + + if (!Settings::values.renderer_debug) { + return; + } + // Validation layers have a bug where they fail to track resource usage when using timeline + // semaphores and synchronizing with GetSemaphoreCounterValueKHR. To workaround this issue, have + // a separate thread waiting for each timeline semaphore value. + debug_thread = std::thread([this] { + u64 counter = 0; + while (!shutdown) { + if (semaphore.Wait(counter, 10'000'000)) { + ++counter; + } + } + }); +} + +MasterSemaphore::~MasterSemaphore() { + shutdown = true; + + // This thread might not be started + if (debug_thread.joinable()) { + debug_thread.join(); + } +} + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_master_semaphore.h b/src/video_core/renderer_vulkan/vk_master_semaphore.h new file mode 100644 index 000000000..0e93706d7 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_master_semaphore.h @@ -0,0 +1,70 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <atomic> +#include <thread> + +#include "common/common_types.h" +#include "video_core/renderer_vulkan/wrapper.h" + +namespace Vulkan { + +class VKDevice; + +class MasterSemaphore { +public: + explicit MasterSemaphore(const VKDevice& device); + ~MasterSemaphore(); + + /// Returns the current logical tick. + [[nodiscard]] u64 CurrentTick() const noexcept { + return current_tick; + } + + /// Returns the timeline semaphore handle. + [[nodiscard]] VkSemaphore Handle() const noexcept { + return *semaphore; + } + + /// Returns true when a tick has been hit by the GPU. + [[nodiscard]] bool IsFree(u64 tick) { + return gpu_tick >= tick; + } + + /// Advance to the logical tick. + void NextTick() noexcept { + ++current_tick; + } + + /// Refresh the known GPU tick + void Refresh() { + gpu_tick = semaphore.GetCounter(); + } + + /// Waits for a tick to be hit on the GPU + void Wait(u64 tick) { + // No need to wait if the GPU is ahead of the tick + if (IsFree(tick)) { + return; + } + // Update the GPU tick and try again + Refresh(); + if (IsFree(tick)) { + return; + } + // If none of the above is hit, fallback to a regular wait + semaphore.Wait(tick); + } + +private: + vk::Semaphore semaphore; ///< Timeline semaphore. + std::atomic<u64> gpu_tick{0}; ///< Current known GPU tick. + std::atomic<u64> current_tick{1}; ///< Current logical tick. + std::atomic<bool> shutdown{false}; ///< True when the object is being destroyed. + std::thread debug_thread; ///< Debug thread to workaround validation layer bugs. +}; + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index cfdcdd6ab..dedc9c466 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -135,64 +135,56 @@ bool ComputePipelineCacheKey::operator==(const ComputePipelineCacheKey& rhs) con return std::memcmp(&rhs, this, sizeof *this) == 0; } -Shader::Shader(Core::System& system, Tegra::Engines::ShaderType stage, GPUVAddr gpu_addr, - VideoCommon::Shader::ProgramCode program_code, u32 main_offset) - : gpu_addr{gpu_addr}, program_code{std::move(program_code)}, - registry{stage, GetEngine(system, stage)}, shader_ir{this->program_code, main_offset, - compiler_settings, registry}, - entries{GenerateShaderEntries(shader_ir)} {} +Shader::Shader(Tegra::Engines::ConstBufferEngineInterface& engine, Tegra::Engines::ShaderType stage, + GPUVAddr gpu_addr_, VAddr cpu_addr, VideoCommon::Shader::ProgramCode program_code_, + u32 main_offset) + : gpu_addr(gpu_addr_), program_code(std::move(program_code_)), registry(stage, engine), + shader_ir(program_code, main_offset, compiler_settings, registry), + entries(GenerateShaderEntries(shader_ir)) {} Shader::~Shader() = default; -Tegra::Engines::ConstBufferEngineInterface& Shader::GetEngine(Core::System& system, - Tegra::Engines::ShaderType stage) { - if (stage == ShaderType::Compute) { - return system.GPU().KeplerCompute(); - } else { - return system.GPU().Maxwell3D(); - } -} - -VKPipelineCache::VKPipelineCache(Core::System& system, RasterizerVulkan& rasterizer, - const VKDevice& device, VKScheduler& scheduler, - VKDescriptorPool& descriptor_pool, - VKUpdateDescriptorQueue& update_descriptor_queue, - VKRenderPassCache& renderpass_cache) - : VideoCommon::ShaderCache<Shader>{rasterizer}, system{system}, device{device}, - scheduler{scheduler}, descriptor_pool{descriptor_pool}, - update_descriptor_queue{update_descriptor_queue}, renderpass_cache{renderpass_cache} {} +VKPipelineCache::VKPipelineCache(RasterizerVulkan& rasterizer, Tegra::GPU& gpu_, + Tegra::Engines::Maxwell3D& maxwell3d_, + Tegra::Engines::KeplerCompute& kepler_compute_, + Tegra::MemoryManager& gpu_memory_, const VKDevice& device_, + VKScheduler& scheduler_, VKDescriptorPool& descriptor_pool_, + VKUpdateDescriptorQueue& update_descriptor_queue_, + VKRenderPassCache& renderpass_cache_) + : VideoCommon::ShaderCache<Shader>{rasterizer}, gpu{gpu_}, maxwell3d{maxwell3d_}, + kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_}, device{device_}, + scheduler{scheduler_}, descriptor_pool{descriptor_pool_}, + update_descriptor_queue{update_descriptor_queue_}, renderpass_cache{renderpass_cache_} {} VKPipelineCache::~VKPipelineCache() = default; std::array<Shader*, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() { - const auto& gpu = system.GPU().Maxwell3D(); - std::array<Shader*, Maxwell::MaxShaderProgram> shaders{}; + for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { const auto program{static_cast<Maxwell::ShaderProgram>(index)}; // Skip stages that are not enabled - if (!gpu.regs.IsShaderConfigEnabled(index)) { + if (!maxwell3d.regs.IsShaderConfigEnabled(index)) { continue; } - auto& memory_manager{system.GPU().MemoryManager()}; - const GPUVAddr program_addr{GetShaderAddress(system, program)}; - const std::optional cpu_addr = memory_manager.GpuToCpuAddress(program_addr); + const GPUVAddr gpu_addr{GetShaderAddress(maxwell3d, program)}; + const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); ASSERT(cpu_addr); Shader* result = cpu_addr ? TryGet(*cpu_addr) : null_shader.get(); if (!result) { - const auto host_ptr{memory_manager.GetPointer(program_addr)}; + const u8* const host_ptr{gpu_memory.GetPointer(gpu_addr)}; // No shader found - create a new one - constexpr u32 stage_offset = STAGE_MAIN_OFFSET; + static constexpr u32 stage_offset = STAGE_MAIN_OFFSET; const auto stage = static_cast<ShaderType>(index == 0 ? 0 : index - 1); - ProgramCode code = GetShaderCode(memory_manager, program_addr, host_ptr, false); + ProgramCode code = GetShaderCode(gpu_memory, gpu_addr, host_ptr, false); const std::size_t size_in_bytes = code.size() * sizeof(u64); - auto shader = std::make_unique<Shader>(system, stage, program_addr, std::move(code), - stage_offset); + auto shader = std::make_unique<Shader>(maxwell3d, stage, gpu_addr, *cpu_addr, + std::move(code), stage_offset); result = shader.get(); if (cpu_addr) { @@ -215,11 +207,11 @@ VKGraphicsPipeline* VKPipelineCache::GetGraphicsPipeline( } last_graphics_key = key; - if (device.UseAsynchronousShaders() && async_shaders.IsShaderAsync(system.GPU())) { + if (device.UseAsynchronousShaders() && async_shaders.IsShaderAsync(gpu)) { std::unique_lock lock{pipeline_cache}; const auto [pair, is_cache_miss] = graphics_cache.try_emplace(key); if (is_cache_miss) { - system.GPU().ShaderNotify().MarkSharderBuilding(); + gpu.ShaderNotify().MarkSharderBuilding(); LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash()); const auto [program, bindings] = DecompileShaders(key.fixed_state); async_shaders.QueueVulkanShader(this, device, scheduler, descriptor_pool, @@ -233,13 +225,13 @@ VKGraphicsPipeline* VKPipelineCache::GetGraphicsPipeline( const auto [pair, is_cache_miss] = graphics_cache.try_emplace(key); auto& entry = pair->second; if (is_cache_miss) { - system.GPU().ShaderNotify().MarkSharderBuilding(); + gpu.ShaderNotify().MarkSharderBuilding(); LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash()); const auto [program, bindings] = DecompileShaders(key.fixed_state); entry = std::make_unique<VKGraphicsPipeline>(device, scheduler, descriptor_pool, update_descriptor_queue, renderpass_cache, key, bindings, program); - system.GPU().ShaderNotify().MarkShaderComplete(); + gpu.ShaderNotify().MarkShaderComplete(); } last_graphics_pipeline = entry.get(); return last_graphics_pipeline; @@ -255,22 +247,21 @@ VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCach } LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash()); - auto& memory_manager = system.GPU().MemoryManager(); - const auto program_addr = key.shader; + const GPUVAddr gpu_addr = key.shader; - const auto cpu_addr = memory_manager.GpuToCpuAddress(program_addr); + const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); ASSERT(cpu_addr); Shader* shader = cpu_addr ? TryGet(*cpu_addr) : null_kernel.get(); if (!shader) { // No shader found - create a new one - const auto host_ptr = memory_manager.GetPointer(program_addr); + const auto host_ptr = gpu_memory.GetPointer(gpu_addr); - ProgramCode code = GetShaderCode(memory_manager, program_addr, host_ptr, true); + ProgramCode code = GetShaderCode(gpu_memory, gpu_addr, host_ptr, true); const std::size_t size_in_bytes = code.size() * sizeof(u64); - auto shader_info = std::make_unique<Shader>(system, ShaderType::Compute, program_addr, - std::move(code), KERNEL_MAIN_OFFSET); + auto shader_info = std::make_unique<Shader>(kepler_compute, ShaderType::Compute, gpu_addr, + *cpu_addr, std::move(code), KERNEL_MAIN_OFFSET); shader = shader_info.get(); if (cpu_addr) { @@ -298,7 +289,7 @@ VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCach } void VKPipelineCache::EmplacePipeline(std::unique_ptr<VKGraphicsPipeline> pipeline) { - system.GPU().ShaderNotify().MarkShaderComplete(); + gpu.ShaderNotify().MarkShaderComplete(); std::unique_lock lock{pipeline_cache}; graphics_cache.at(pipeline->GetCacheKey()) = std::move(pipeline); } @@ -339,12 +330,8 @@ void VKPipelineCache::OnShaderRemoval(Shader* shader) { std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>> VKPipelineCache::DecompileShaders(const FixedPipelineState& fixed_state) { - auto& memory_manager = system.GPU().MemoryManager(); - const auto& gpu = system.GPU().Maxwell3D(); - Specialization specialization; - if (fixed_state.dynamic_state.Topology() == Maxwell::PrimitiveTopology::Points || - device.IsExtExtendedDynamicStateSupported()) { + if (fixed_state.topology == Maxwell::PrimitiveTopology::Points) { float point_size; std::memcpy(&point_size, &fixed_state.point_size, sizeof(float)); specialization.point_size = point_size; @@ -364,12 +351,12 @@ VKPipelineCache::DecompileShaders(const FixedPipelineState& fixed_state) { const auto program_enum = static_cast<Maxwell::ShaderProgram>(index); // Skip stages that are not enabled - if (!gpu.regs.IsShaderConfigEnabled(index)) { + if (!maxwell3d.regs.IsShaderConfigEnabled(index)) { continue; } - const GPUVAddr gpu_addr = GetShaderAddress(system, program_enum); - const std::optional<VAddr> cpu_addr = memory_manager.GpuToCpuAddress(gpu_addr); + const GPUVAddr gpu_addr = GetShaderAddress(maxwell3d, program_enum); + const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); Shader* const shader = cpu_addr ? TryGet(*cpu_addr) : null_shader.get(); const std::size_t stage = index == 0 ? 0 : index - 1; // Stage indices are 0 - 5 diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.h b/src/video_core/renderer_vulkan/vk_pipeline_cache.h index c04829e77..e558e6658 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.h +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.h @@ -38,7 +38,6 @@ class RasterizerVulkan; class VKComputePipeline; class VKDescriptorPool; class VKDevice; -class VKFence; class VKScheduler; class VKUpdateDescriptorQueue; @@ -85,7 +84,8 @@ namespace Vulkan { class Shader { public: - explicit Shader(Core::System& system, Tegra::Engines::ShaderType stage, GPUVAddr gpu_addr, + explicit Shader(Tegra::Engines::ConstBufferEngineInterface& engine, + Tegra::Engines::ShaderType stage, GPUVAddr gpu_addr, VAddr cpu_addr, VideoCommon::Shader::ProgramCode program_code, u32 main_offset); ~Shader(); @@ -97,22 +97,19 @@ public: return shader_ir; } - const VideoCommon::Shader::Registry& GetRegistry() const { - return registry; - } - const VideoCommon::Shader::ShaderIR& GetIR() const { return shader_ir; } + const VideoCommon::Shader::Registry& GetRegistry() const { + return registry; + } + const ShaderEntries& GetEntries() const { return entries; } private: - static Tegra::Engines::ConstBufferEngineInterface& GetEngine(Core::System& system, - Tegra::Engines::ShaderType stage); - GPUVAddr gpu_addr{}; VideoCommon::Shader::ProgramCode program_code; VideoCommon::Shader::Registry registry; @@ -122,9 +119,11 @@ private: class VKPipelineCache final : public VideoCommon::ShaderCache<Shader> { public: - explicit VKPipelineCache(Core::System& system, RasterizerVulkan& rasterizer, - const VKDevice& device, VKScheduler& scheduler, - VKDescriptorPool& descriptor_pool, + explicit VKPipelineCache(RasterizerVulkan& rasterizer, Tegra::GPU& gpu, + Tegra::Engines::Maxwell3D& maxwell3d, + Tegra::Engines::KeplerCompute& kepler_compute, + Tegra::MemoryManager& gpu_memory, const VKDevice& device, + VKScheduler& scheduler, VKDescriptorPool& descriptor_pool, VKUpdateDescriptorQueue& update_descriptor_queue, VKRenderPassCache& renderpass_cache); ~VKPipelineCache() override; @@ -145,7 +144,11 @@ private: std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>> DecompileShaders( const FixedPipelineState& fixed_state); - Core::System& system; + Tegra::GPU& gpu; + Tegra::Engines::Maxwell3D& maxwell3d; + Tegra::Engines::KeplerCompute& kepler_compute; + Tegra::MemoryManager& gpu_memory; + const VKDevice& device; VKScheduler& scheduler; VKDescriptorPool& descriptor_pool; diff --git a/src/video_core/renderer_vulkan/vk_query_cache.cpp b/src/video_core/renderer_vulkan/vk_query_cache.cpp index 6cd63d090..ee2d871e3 100644 --- a/src/video_core/renderer_vulkan/vk_query_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_query_cache.cpp @@ -9,35 +9,33 @@ #include "video_core/renderer_vulkan/vk_device.h" #include "video_core/renderer_vulkan/vk_query_cache.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" +#include "video_core/renderer_vulkan/vk_resource_pool.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/wrapper.h" namespace Vulkan { +using VideoCore::QueryType; + namespace { constexpr std::array QUERY_TARGETS = {VK_QUERY_TYPE_OCCLUSION}; -constexpr VkQueryType GetTarget(VideoCore::QueryType type) { +constexpr VkQueryType GetTarget(QueryType type) { return QUERY_TARGETS[static_cast<std::size_t>(type)]; } } // Anonymous namespace -QueryPool::QueryPool() : VKFencedPool{GROW_STEP} {} +QueryPool::QueryPool(const VKDevice& device_, VKScheduler& scheduler, QueryType type_) + : ResourcePool{scheduler.GetMasterSemaphore(), GROW_STEP}, device{device_}, type{type_} {} QueryPool::~QueryPool() = default; -void QueryPool::Initialize(const VKDevice& device_, VideoCore::QueryType type_) { - device = &device_; - type = type_; -} - -std::pair<VkQueryPool, u32> QueryPool::Commit(VKFence& fence) { +std::pair<VkQueryPool, u32> QueryPool::Commit() { std::size_t index; do { - index = CommitResource(fence); + index = CommitResource(); } while (usage[index]); usage[index] = true; @@ -47,7 +45,7 @@ std::pair<VkQueryPool, u32> QueryPool::Commit(VKFence& fence) { void QueryPool::Allocate(std::size_t begin, std::size_t end) { usage.resize(end); - pools.push_back(device->GetLogical().CreateQueryPool({ + pools.push_back(device.GetLogical().CreateQueryPool({ .sType = VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO, .pNext = nullptr, .flags = 0, @@ -68,30 +66,39 @@ void QueryPool::Reserve(std::pair<VkQueryPool, u32> query) { usage[pool_index * GROW_STEP + static_cast<std::ptrdiff_t>(query.second)] = false; } -VKQueryCache::VKQueryCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer, +VKQueryCache::VKQueryCache(VideoCore::RasterizerInterface& rasterizer, + Tegra::Engines::Maxwell3D& maxwell3d, Tegra::MemoryManager& gpu_memory, const VKDevice& device, VKScheduler& scheduler) - : VideoCommon::QueryCacheBase<VKQueryCache, CachedQuery, CounterStream, HostCounter, - QueryPool>{system, rasterizer}, - device{device}, scheduler{scheduler} { - for (std::size_t i = 0; i < static_cast<std::size_t>(VideoCore::NumQueryTypes); ++i) { - query_pools[i].Initialize(device, static_cast<VideoCore::QueryType>(i)); + : VideoCommon::QueryCacheBase<VKQueryCache, CachedQuery, CounterStream, + HostCounter>{rasterizer, maxwell3d, gpu_memory}, + device{device}, scheduler{scheduler}, query_pools{ + QueryPool{device, scheduler, + QueryType::SamplesPassed}, + } {} + +VKQueryCache::~VKQueryCache() { + // TODO(Rodrigo): This is a hack to destroy all HostCounter instances before the base class + // destructor is called. The query cache should be redesigned to have a proper ownership model + // instead of using shared pointers. + for (size_t query_type = 0; query_type < VideoCore::NumQueryTypes; ++query_type) { + auto& stream = Stream(static_cast<QueryType>(query_type)); + stream.Update(false); + stream.Reset(); } } -VKQueryCache::~VKQueryCache() = default; - -std::pair<VkQueryPool, u32> VKQueryCache::AllocateQuery(VideoCore::QueryType type) { - return query_pools[static_cast<std::size_t>(type)].Commit(scheduler.GetFence()); +std::pair<VkQueryPool, u32> VKQueryCache::AllocateQuery(QueryType type) { + return query_pools[static_cast<std::size_t>(type)].Commit(); } -void VKQueryCache::Reserve(VideoCore::QueryType type, std::pair<VkQueryPool, u32> query) { +void VKQueryCache::Reserve(QueryType type, std::pair<VkQueryPool, u32> query) { query_pools[static_cast<std::size_t>(type)].Reserve(query); } HostCounter::HostCounter(VKQueryCache& cache, std::shared_ptr<HostCounter> dependency, - VideoCore::QueryType type) + QueryType type) : VideoCommon::HostCounterBase<VKQueryCache, HostCounter>{std::move(dependency)}, cache{cache}, - type{type}, query{cache.AllocateQuery(type)}, ticks{cache.Scheduler().Ticks()} { + type{type}, query{cache.AllocateQuery(type)}, tick{cache.Scheduler().CurrentTick()} { const vk::Device* logical = &cache.Device().GetLogical(); cache.Scheduler().Record([logical, query = query](vk::CommandBuffer cmdbuf) { logical->ResetQueryPoolEXT(query.first, query.second, 1); @@ -109,7 +116,7 @@ void HostCounter::EndQuery() { } u64 HostCounter::BlockingQuery() const { - if (ticks >= cache.Scheduler().Ticks()) { + if (tick >= cache.Scheduler().CurrentTick()) { cache.Scheduler().Flush(); } u64 data; diff --git a/src/video_core/renderer_vulkan/vk_query_cache.h b/src/video_core/renderer_vulkan/vk_query_cache.h index 40119e6d3..2e57fb75d 100644 --- a/src/video_core/renderer_vulkan/vk_query_cache.h +++ b/src/video_core/renderer_vulkan/vk_query_cache.h @@ -11,7 +11,7 @@ #include "common/common_types.h" #include "video_core/query_cache.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" +#include "video_core/renderer_vulkan/vk_resource_pool.h" #include "video_core/renderer_vulkan/wrapper.h" namespace VideoCore { @@ -28,14 +28,12 @@ class VKScheduler; using CounterStream = VideoCommon::CounterStreamBase<VKQueryCache, HostCounter>; -class QueryPool final : public VKFencedPool { +class QueryPool final : public ResourcePool { public: - explicit QueryPool(); + explicit QueryPool(const VKDevice& device, VKScheduler& scheduler, VideoCore::QueryType type); ~QueryPool() override; - void Initialize(const VKDevice& device, VideoCore::QueryType type); - - std::pair<VkQueryPool, u32> Commit(VKFence& fence); + std::pair<VkQueryPool, u32> Commit(); void Reserve(std::pair<VkQueryPool, u32> query); @@ -45,18 +43,18 @@ protected: private: static constexpr std::size_t GROW_STEP = 512; - const VKDevice* device = nullptr; - VideoCore::QueryType type = {}; + const VKDevice& device; + const VideoCore::QueryType type; std::vector<vk::QueryPool> pools; std::vector<bool> usage; }; class VKQueryCache final - : public VideoCommon::QueryCacheBase<VKQueryCache, CachedQuery, CounterStream, HostCounter, - QueryPool> { + : public VideoCommon::QueryCacheBase<VKQueryCache, CachedQuery, CounterStream, HostCounter> { public: - explicit VKQueryCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer, + explicit VKQueryCache(VideoCore::RasterizerInterface& rasterizer, + Tegra::Engines::Maxwell3D& maxwell3d, Tegra::MemoryManager& gpu_memory, const VKDevice& device, VKScheduler& scheduler); ~VKQueryCache(); @@ -75,6 +73,7 @@ public: private: const VKDevice& device; VKScheduler& scheduler; + std::array<QueryPool, VideoCore::NumQueryTypes> query_pools; }; class HostCounter final : public VideoCommon::HostCounterBase<VKQueryCache, HostCounter> { @@ -91,7 +90,7 @@ private: VKQueryCache& cache; const VideoCore::QueryType type; const std::pair<VkQueryPool, u32> query; - const u64 ticks; + const u64 tick; }; class CachedQuery : public VideoCommon::CachedQueryBase<HostCounter> { diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 936f76195..e0fb8693f 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -31,7 +31,6 @@ #include "video_core/renderer_vulkan/vk_pipeline_cache.h" #include "video_core/renderer_vulkan/vk_rasterizer.h" #include "video_core/renderer_vulkan/vk_renderpass_cache.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" #include "video_core/renderer_vulkan/vk_sampler_cache.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" @@ -381,28 +380,28 @@ void RasterizerVulkan::DrawParameters::Draw(vk::CommandBuffer cmdbuf) const { } } -RasterizerVulkan::RasterizerVulkan(Core::System& system, Core::Frontend::EmuWindow& renderer, - VKScreenInfo& screen_info, const VKDevice& device, - VKResourceManager& resource_manager, - VKMemoryManager& memory_manager, StateTracker& state_tracker, - VKScheduler& scheduler) - : RasterizerAccelerated{system.Memory()}, system{system}, render_window{renderer}, - screen_info{screen_info}, device{device}, resource_manager{resource_manager}, - memory_manager{memory_manager}, state_tracker{state_tracker}, scheduler{scheduler}, - staging_pool(device, memory_manager, scheduler), descriptor_pool(device), - update_descriptor_queue(device, scheduler), renderpass_cache(device), +RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window, Tegra::GPU& gpu_, + Tegra::MemoryManager& gpu_memory_, + Core::Memory::Memory& cpu_memory, VKScreenInfo& screen_info_, + const VKDevice& device_, VKMemoryManager& memory_manager_, + StateTracker& state_tracker_, VKScheduler& scheduler_) + : RasterizerAccelerated(cpu_memory), gpu(gpu_), gpu_memory(gpu_memory_), + maxwell3d(gpu.Maxwell3D()), kepler_compute(gpu.KeplerCompute()), screen_info(screen_info_), + device(device_), memory_manager(memory_manager_), state_tracker(state_tracker_), + scheduler(scheduler_), staging_pool(device, memory_manager, scheduler), + descriptor_pool(device, scheduler_), update_descriptor_queue(device, scheduler), + renderpass_cache(device), quad_array_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue), quad_indexed_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue), uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue), - texture_cache(system, *this, device, resource_manager, memory_manager, scheduler, - staging_pool), - pipeline_cache(system, *this, device, scheduler, descriptor_pool, update_descriptor_queue, - renderpass_cache), - buffer_cache(*this, system, device, memory_manager, scheduler, staging_pool), - sampler_cache(device), - fence_manager(system, *this, device, scheduler, texture_cache, buffer_cache, query_cache), - query_cache(system, *this, device, scheduler), - wfi_event{device.GetLogical().CreateNewEvent()}, async_shaders{renderer} { + texture_cache(*this, maxwell3d, gpu_memory, device, memory_manager, scheduler, staging_pool), + pipeline_cache(*this, gpu, maxwell3d, kepler_compute, gpu_memory, device, scheduler, + descriptor_pool, update_descriptor_queue, renderpass_cache), + buffer_cache(*this, gpu_memory, cpu_memory, device, memory_manager, scheduler, staging_pool), + sampler_cache(device), query_cache(*this, maxwell3d, gpu_memory, device, scheduler), + fence_manager(*this, gpu, gpu_memory, texture_cache, buffer_cache, query_cache, device, + scheduler), + wfi_event(device.GetLogical().CreateEvent()), async_shaders(emu_window) { scheduler.SetQueryCache(query_cache); if (device.UseAsynchronousShaders()) { async_shaders.AllocateWorkers(); @@ -414,15 +413,13 @@ RasterizerVulkan::~RasterizerVulkan() = default; void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) { MICROPROFILE_SCOPE(Vulkan_Drawing); + SCOPE_EXIT({ gpu.TickWork(); }); FlushWork(); query_cache.UpdateCounters(); - SCOPE_EXIT({ system.GPU().TickWork(); }); - - const auto& gpu = system.GPU().Maxwell3D(); GraphicsPipelineCacheKey key; - key.fixed_state.Fill(gpu.regs, device.IsExtExtendedDynamicStateSupported()); + key.fixed_state.Fill(maxwell3d.regs, device.IsExtExtendedDynamicStateSupported()); buffer_cache.Map(CalculateGraphicsStreamBufferSize(is_indexed)); @@ -480,8 +477,7 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) { void RasterizerVulkan::Clear() { MICROPROFILE_SCOPE(Vulkan_Clearing); - const auto& gpu = system.GPU().Maxwell3D(); - if (!system.GPU().Maxwell3D().ShouldExecute()) { + if (!maxwell3d.ShouldExecute()) { return; } @@ -490,7 +486,7 @@ void RasterizerVulkan::Clear() { query_cache.UpdateCounters(); - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; const bool use_color = regs.clear_buffers.R || regs.clear_buffers.G || regs.clear_buffers.B || regs.clear_buffers.A; const bool use_depth = regs.clear_buffers.Z; @@ -559,7 +555,7 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) { query_cache.UpdateCounters(); - const auto& launch_desc = system.GPU().KeplerCompute().launch_description; + const auto& launch_desc = kepler_compute.launch_description; auto& pipeline = pipeline_cache.GetComputePipeline({ .shader = code_addr, .shared_memory_size = launch_desc.shared_alloc, @@ -655,16 +651,14 @@ void RasterizerVulkan::SyncGuestHost() { } void RasterizerVulkan::SignalSemaphore(GPUVAddr addr, u32 value) { - auto& gpu{system.GPU()}; if (!gpu.IsAsync()) { - gpu.MemoryManager().Write<u32>(addr, value); + gpu_memory.Write<u32>(addr, value); return; } fence_manager.SignalSemaphore(addr, value); } void RasterizerVulkan::SignalSyncPoint(u32 value) { - auto& gpu{system.GPU()}; if (!gpu.IsAsync()) { gpu.IncrementSyncPoint(value); return; @@ -673,7 +667,6 @@ void RasterizerVulkan::SignalSyncPoint(u32 value) { } void RasterizerVulkan::ReleaseFences() { - auto& gpu{system.GPU()}; if (!gpu.IsAsync()) { return; } @@ -751,10 +744,6 @@ bool RasterizerVulkan::AccelerateDisplay(const Tegra::FramebufferConfig& config, return true; } -void RasterizerVulkan::SetupDirtyFlags() { - state_tracker.Initialize(); -} - void RasterizerVulkan::FlushWork() { static constexpr u32 DRAWS_TO_DISPATCH = 4096; @@ -778,10 +767,9 @@ void RasterizerVulkan::FlushWork() { RasterizerVulkan::Texceptions RasterizerVulkan::UpdateAttachments(bool is_clear) { MICROPROFILE_SCOPE(Vulkan_RenderTargets); - auto& maxwell3d = system.GPU().Maxwell3D(); - auto& dirty = maxwell3d.dirty.flags; - auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d.regs; + auto& dirty = maxwell3d.dirty.flags; const bool update_rendertargets = dirty[VideoCommon::Dirty::RenderTargets]; dirty[VideoCommon::Dirty::RenderTargets] = false; @@ -844,7 +832,7 @@ std::tuple<VkFramebuffer, VkExtent2D> RasterizerVulkan::ConfigureFramebuffers( return true; }; - const auto& regs = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; const std::size_t num_attachments = static_cast<std::size_t>(regs.rt_control.count); for (std::size_t index = 0; index < num_attachments; ++index) { if (try_push(color_attachments[index])) { @@ -880,13 +868,12 @@ RasterizerVulkan::DrawParameters RasterizerVulkan::SetupGeometry(FixedPipelineSt bool is_instanced) { MICROPROFILE_SCOPE(Vulkan_Geometry); - const auto& gpu = system.GPU().Maxwell3D(); - const auto& regs = gpu.regs; + const auto& regs = maxwell3d.regs; SetupVertexArrays(buffer_bindings); const u32 base_instance = regs.vb_base_instance; - const u32 num_instances = is_instanced ? gpu.mme_draw.instance_count : 1; + const u32 num_instances = is_instanced ? maxwell3d.mme_draw.instance_count : 1; const u32 base_vertex = is_indexed ? regs.vb_element_base : regs.vertex_buffer.first; const u32 num_vertices = is_indexed ? regs.index_array.count : regs.vertex_buffer.count; @@ -947,7 +934,7 @@ void RasterizerVulkan::SetupImageTransitions( } void RasterizerVulkan::UpdateDynamicStates() { - auto& regs = system.GPU().Maxwell3D().regs; + auto& regs = maxwell3d.regs; UpdateViewportsState(regs); UpdateScissorsState(regs); UpdateDepthBias(regs); @@ -961,14 +948,13 @@ void RasterizerVulkan::UpdateDynamicStates() { UpdateDepthWriteEnable(regs); UpdateDepthCompareOp(regs); UpdateFrontFace(regs); - UpdatePrimitiveTopology(regs); UpdateStencilOp(regs); UpdateStencilTestEnable(regs); } } void RasterizerVulkan::BeginTransformFeedback() { - const auto& regs = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; if (regs.tfb_enabled == 0) { return; } @@ -1000,7 +986,7 @@ void RasterizerVulkan::BeginTransformFeedback() { } void RasterizerVulkan::EndTransformFeedback() { - const auto& regs = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; if (regs.tfb_enabled == 0) { return; } @@ -1013,7 +999,7 @@ void RasterizerVulkan::EndTransformFeedback() { } void RasterizerVulkan::SetupVertexArrays(BufferBindings& buffer_bindings) { - const auto& regs = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; for (std::size_t index = 0; index < Maxwell::NumVertexArrays; ++index) { const auto& vertex_array = regs.vertex_array[index]; @@ -1039,7 +1025,7 @@ void RasterizerVulkan::SetupIndexBuffer(BufferBindings& buffer_bindings, DrawPar if (params.num_vertices == 0) { return; } - const auto& regs = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; switch (regs.draw.topology) { case Maxwell::PrimitiveTopology::Quads: { if (!params.is_indexed) { @@ -1087,8 +1073,7 @@ void RasterizerVulkan::SetupIndexBuffer(BufferBindings& buffer_bindings, DrawPar void RasterizerVulkan::SetupGraphicsConstBuffers(const ShaderEntries& entries, std::size_t stage) { MICROPROFILE_SCOPE(Vulkan_ConstBuffers); - const auto& gpu = system.GPU().Maxwell3D(); - const auto& shader_stage = gpu.state.shader_stages[stage]; + const auto& shader_stage = maxwell3d.state.shader_stages[stage]; for (const auto& entry : entries.const_buffers) { SetupConstBuffer(entry, shader_stage.const_buffers[entry.GetIndex()]); } @@ -1096,8 +1081,7 @@ void RasterizerVulkan::SetupGraphicsConstBuffers(const ShaderEntries& entries, s void RasterizerVulkan::SetupGraphicsGlobalBuffers(const ShaderEntries& entries, std::size_t stage) { MICROPROFILE_SCOPE(Vulkan_GlobalBuffers); - auto& gpu{system.GPU()}; - const auto cbufs{gpu.Maxwell3D().state.shader_stages[stage]}; + const auto& cbufs{maxwell3d.state.shader_stages[stage]}; for (const auto& entry : entries.global_buffers) { const auto addr = cbufs.const_buffers[entry.GetCbufIndex()].address + entry.GetCbufOffset(); @@ -1107,19 +1091,17 @@ void RasterizerVulkan::SetupGraphicsGlobalBuffers(const ShaderEntries& entries, void RasterizerVulkan::SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage) { MICROPROFILE_SCOPE(Vulkan_Textures); - const auto& gpu = system.GPU().Maxwell3D(); for (const auto& entry : entries.uniform_texels) { - const auto image = GetTextureInfo(gpu, entry, stage).tic; + const auto image = GetTextureInfo(maxwell3d, entry, stage).tic; SetupUniformTexels(image, entry); } } void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, std::size_t stage) { MICROPROFILE_SCOPE(Vulkan_Textures); - const auto& gpu = system.GPU().Maxwell3D(); for (const auto& entry : entries.samplers) { for (std::size_t i = 0; i < entry.size; ++i) { - const auto texture = GetTextureInfo(gpu, entry, stage, i); + const auto texture = GetTextureInfo(maxwell3d, entry, stage, i); SetupTexture(texture, entry); } } @@ -1127,25 +1109,23 @@ void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, std:: void RasterizerVulkan::SetupGraphicsStorageTexels(const ShaderEntries& entries, std::size_t stage) { MICROPROFILE_SCOPE(Vulkan_Textures); - const auto& gpu = system.GPU().Maxwell3D(); for (const auto& entry : entries.storage_texels) { - const auto image = GetTextureInfo(gpu, entry, stage).tic; + const auto image = GetTextureInfo(maxwell3d, entry, stage).tic; SetupStorageTexel(image, entry); } } void RasterizerVulkan::SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage) { MICROPROFILE_SCOPE(Vulkan_Images); - const auto& gpu = system.GPU().Maxwell3D(); for (const auto& entry : entries.images) { - const auto tic = GetTextureInfo(gpu, entry, stage).tic; + const auto tic = GetTextureInfo(maxwell3d, entry, stage).tic; SetupImage(tic, entry); } } void RasterizerVulkan::SetupComputeConstBuffers(const ShaderEntries& entries) { MICROPROFILE_SCOPE(Vulkan_ConstBuffers); - const auto& launch_desc = system.GPU().KeplerCompute().launch_description; + const auto& launch_desc = kepler_compute.launch_description; for (const auto& entry : entries.const_buffers) { const auto& config = launch_desc.const_buffer_config[entry.GetIndex()]; const std::bitset<8> mask = launch_desc.const_buffer_enable_mask.Value(); @@ -1159,7 +1139,7 @@ void RasterizerVulkan::SetupComputeConstBuffers(const ShaderEntries& entries) { void RasterizerVulkan::SetupComputeGlobalBuffers(const ShaderEntries& entries) { MICROPROFILE_SCOPE(Vulkan_GlobalBuffers); - const auto cbufs{system.GPU().KeplerCompute().launch_description.const_buffer_config}; + const auto& cbufs{kepler_compute.launch_description.const_buffer_config}; for (const auto& entry : entries.global_buffers) { const auto addr{cbufs[entry.GetCbufIndex()].Address() + entry.GetCbufOffset()}; SetupGlobalBuffer(entry, addr); @@ -1168,19 +1148,17 @@ void RasterizerVulkan::SetupComputeGlobalBuffers(const ShaderEntries& entries) { void RasterizerVulkan::SetupComputeUniformTexels(const ShaderEntries& entries) { MICROPROFILE_SCOPE(Vulkan_Textures); - const auto& gpu = system.GPU().KeplerCompute(); for (const auto& entry : entries.uniform_texels) { - const auto image = GetTextureInfo(gpu, entry, ComputeShaderIndex).tic; + const auto image = GetTextureInfo(kepler_compute, entry, ComputeShaderIndex).tic; SetupUniformTexels(image, entry); } } void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) { MICROPROFILE_SCOPE(Vulkan_Textures); - const auto& gpu = system.GPU().KeplerCompute(); for (const auto& entry : entries.samplers) { for (std::size_t i = 0; i < entry.size; ++i) { - const auto texture = GetTextureInfo(gpu, entry, ComputeShaderIndex, i); + const auto texture = GetTextureInfo(kepler_compute, entry, ComputeShaderIndex, i); SetupTexture(texture, entry); } } @@ -1188,18 +1166,16 @@ void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) { void RasterizerVulkan::SetupComputeStorageTexels(const ShaderEntries& entries) { MICROPROFILE_SCOPE(Vulkan_Textures); - const auto& gpu = system.GPU().KeplerCompute(); for (const auto& entry : entries.storage_texels) { - const auto image = GetTextureInfo(gpu, entry, ComputeShaderIndex).tic; + const auto image = GetTextureInfo(kepler_compute, entry, ComputeShaderIndex).tic; SetupStorageTexel(image, entry); } } void RasterizerVulkan::SetupComputeImages(const ShaderEntries& entries) { MICROPROFILE_SCOPE(Vulkan_Images); - const auto& gpu = system.GPU().KeplerCompute(); for (const auto& entry : entries.images) { - const auto tic = GetTextureInfo(gpu, entry, ComputeShaderIndex).tic; + const auto tic = GetTextureInfo(kepler_compute, entry, ComputeShaderIndex).tic; SetupImage(tic, entry); } } @@ -1223,9 +1199,8 @@ void RasterizerVulkan::SetupConstBuffer(const ConstBufferEntry& entry, } void RasterizerVulkan::SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAddr address) { - auto& memory_manager{system.GPU().MemoryManager()}; - const auto actual_addr = memory_manager.Read<u64>(address); - const auto size = memory_manager.Read<u32>(address + 8); + const u64 actual_addr = gpu_memory.Read<u64>(address); + const u32 size = gpu_memory.Read<u32>(address + 8); if (size == 0) { // Sometimes global memory pointers don't have a proper size. Upload a dummy entry @@ -1442,16 +1417,6 @@ void RasterizerVulkan::UpdateFrontFace(Tegra::Engines::Maxwell3D::Regs& regs) { [front_face](vk::CommandBuffer cmdbuf) { cmdbuf.SetFrontFaceEXT(front_face); }); } -void RasterizerVulkan::UpdatePrimitiveTopology(Tegra::Engines::Maxwell3D::Regs& regs) { - if (!state_tracker.TouchPrimitiveTopology()) { - return; - } - const Maxwell::PrimitiveTopology primitive_topology = regs.draw.topology.Value(); - scheduler.Record([this, primitive_topology](vk::CommandBuffer cmdbuf) { - cmdbuf.SetPrimitiveTopologyEXT(MaxwellToVK::PrimitiveTopology(device, primitive_topology)); - }); -} - void RasterizerVulkan::UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs) { if (!state_tracker.TouchStencilOp()) { return; @@ -1508,7 +1473,7 @@ std::size_t RasterizerVulkan::CalculateComputeStreamBufferSize() const { } std::size_t RasterizerVulkan::CalculateVertexArraysSize() const { - const auto& regs = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; std::size_t size = 0; for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) { @@ -1523,9 +1488,8 @@ std::size_t RasterizerVulkan::CalculateVertexArraysSize() const { } std::size_t RasterizerVulkan::CalculateIndexBufferSize() const { - const auto& regs = system.GPU().Maxwell3D().regs; - return static_cast<std::size_t>(regs.index_array.count) * - static_cast<std::size_t>(regs.index_array.FormatSizeInBytes()); + return static_cast<std::size_t>(maxwell3d.regs.index_array.count) * + static_cast<std::size_t>(maxwell3d.regs.index_array.FormatSizeInBytes()); } std::size_t RasterizerVulkan::CalculateConstBufferSize( @@ -1540,7 +1504,7 @@ std::size_t RasterizerVulkan::CalculateConstBufferSize( } RenderPassParams RasterizerVulkan::GetRenderPassParams(Texceptions texceptions) const { - const auto& regs = system.GPU().Maxwell3D().regs; + const auto& regs = maxwell3d.regs; const std::size_t num_attachments = static_cast<std::size_t>(regs.rt_control.count); RenderPassParams params; diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index f640ba649..237e51fa4 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -25,7 +25,6 @@ #include "video_core/renderer_vulkan/vk_pipeline_cache.h" #include "video_core/renderer_vulkan/vk_query_cache.h" #include "video_core/renderer_vulkan/vk_renderpass_cache.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" #include "video_core/renderer_vulkan/vk_sampler_cache.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" @@ -106,10 +105,11 @@ struct ImageView { class RasterizerVulkan final : public VideoCore::RasterizerAccelerated { public: - explicit RasterizerVulkan(Core::System& system, Core::Frontend::EmuWindow& render_window, + explicit RasterizerVulkan(Core::Frontend::EmuWindow& emu_window, Tegra::GPU& gpu, + Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory, VKScreenInfo& screen_info, const VKDevice& device, - VKResourceManager& resource_manager, VKMemoryManager& memory_manager, - StateTracker& state_tracker, VKScheduler& scheduler); + VKMemoryManager& memory_manager, StateTracker& state_tracker, + VKScheduler& scheduler); ~RasterizerVulkan() override; void Draw(bool is_indexed, bool is_instanced) override; @@ -135,7 +135,6 @@ public: const Tegra::Engines::Fermi2D::Config& copy_config) override; bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) override; - void SetupDirtyFlags() override; VideoCommon::Shader::AsyncShaders& GetAsyncShaders() { return async_shaders; @@ -260,7 +259,6 @@ private: void UpdateDepthWriteEnable(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateDepthCompareOp(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateFrontFace(Tegra::Engines::Maxwell3D::Regs& regs); - void UpdatePrimitiveTopology(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs& regs); @@ -279,11 +277,13 @@ private: VkBuffer DefaultBuffer(); - Core::System& system; - Core::Frontend::EmuWindow& render_window; + Tegra::GPU& gpu; + Tegra::MemoryManager& gpu_memory; + Tegra::Engines::Maxwell3D& maxwell3d; + Tegra::Engines::KeplerCompute& kepler_compute; + VKScreenInfo& screen_info; const VKDevice& device; - VKResourceManager& resource_manager; VKMemoryManager& memory_manager; StateTracker& state_tracker; VKScheduler& scheduler; @@ -300,8 +300,8 @@ private: VKPipelineCache pipeline_cache; VKBufferCache buffer_cache; VKSamplerCache sampler_cache; - VKFenceManager fence_manager; VKQueryCache query_cache; + VKFenceManager fence_manager; vk::Buffer default_buffer; VKMemoryCommit default_buffer_commit; diff --git a/src/video_core/renderer_vulkan/vk_resource_manager.cpp b/src/video_core/renderer_vulkan/vk_resource_manager.cpp deleted file mode 100644 index f19330a36..000000000 --- a/src/video_core/renderer_vulkan/vk_resource_manager.cpp +++ /dev/null @@ -1,311 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <optional> -#include "common/assert.h" -#include "common/logging/log.h" -#include "video_core/renderer_vulkan/vk_device.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" -#include "video_core/renderer_vulkan/wrapper.h" - -namespace Vulkan { - -namespace { - -// TODO(Rodrigo): Fine tune these numbers. -constexpr std::size_t COMMAND_BUFFER_POOL_SIZE = 0x1000; -constexpr std::size_t FENCES_GROW_STEP = 0x40; - -constexpr VkFenceCreateInfo BuildFenceCreateInfo() { - return { - .sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - }; -} - -} // Anonymous namespace - -class CommandBufferPool final : public VKFencedPool { -public: - explicit CommandBufferPool(const VKDevice& device) - : VKFencedPool(COMMAND_BUFFER_POOL_SIZE), device{device} {} - - void Allocate(std::size_t begin, std::size_t end) override { - // Command buffers are going to be commited, recorded, executed every single usage cycle. - // They are also going to be reseted when commited. - Pool& pool = pools.emplace_back(); - pool.handle = device.GetLogical().CreateCommandPool({ - .sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO, - .pNext = nullptr, - .flags = VK_COMMAND_POOL_CREATE_TRANSIENT_BIT | - VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, - .queueFamilyIndex = device.GetGraphicsFamily(), - }); - pool.cmdbufs = pool.handle.Allocate(COMMAND_BUFFER_POOL_SIZE); - } - - VkCommandBuffer Commit(VKFence& fence) { - const std::size_t index = CommitResource(fence); - const auto pool_index = index / COMMAND_BUFFER_POOL_SIZE; - const auto sub_index = index % COMMAND_BUFFER_POOL_SIZE; - return pools[pool_index].cmdbufs[sub_index]; - } - -private: - struct Pool { - vk::CommandPool handle; - vk::CommandBuffers cmdbufs; - }; - - const VKDevice& device; - std::vector<Pool> pools; -}; - -VKResource::VKResource() = default; - -VKResource::~VKResource() = default; - -VKFence::VKFence(const VKDevice& device) - : device{device}, handle{device.GetLogical().CreateFence(BuildFenceCreateInfo())} {} - -VKFence::~VKFence() = default; - -void VKFence::Wait() { - switch (const VkResult result = handle.Wait()) { - case VK_SUCCESS: - return; - case VK_ERROR_DEVICE_LOST: - device.ReportLoss(); - [[fallthrough]]; - default: - throw vk::Exception(result); - } -} - -void VKFence::Release() { - ASSERT(is_owned); - is_owned = false; -} - -void VKFence::Commit() { - is_owned = true; - is_used = true; -} - -bool VKFence::Tick(bool gpu_wait, bool owner_wait) { - if (!is_used) { - // If a fence is not used it's always free. - return true; - } - if (is_owned && !owner_wait) { - // The fence is still being owned (Release has not been called) and ownership wait has - // not been asked. - return false; - } - - if (gpu_wait) { - // Wait for the fence if it has been requested. - (void)handle.Wait(); - } else { - if (handle.GetStatus() != VK_SUCCESS) { - // Vulkan fence is not ready, not much it can do here - return false; - } - } - - // Broadcast resources their free state. - for (auto* resource : protected_resources) { - resource->OnFenceRemoval(this); - } - protected_resources.clear(); - - // Prepare fence for reusage. - handle.Reset(); - is_used = false; - return true; -} - -void VKFence::Protect(VKResource* resource) { - protected_resources.push_back(resource); -} - -void VKFence::Unprotect(VKResource* resource) { - const auto it = std::find(protected_resources.begin(), protected_resources.end(), resource); - ASSERT(it != protected_resources.end()); - - resource->OnFenceRemoval(this); - protected_resources.erase(it); -} - -void VKFence::RedirectProtection(VKResource* old_resource, VKResource* new_resource) noexcept { - std::replace(std::begin(protected_resources), std::end(protected_resources), old_resource, - new_resource); -} - -VKFenceWatch::VKFenceWatch() = default; - -VKFenceWatch::VKFenceWatch(VKFence& initial_fence) { - Watch(initial_fence); -} - -VKFenceWatch::VKFenceWatch(VKFenceWatch&& rhs) noexcept { - fence = std::exchange(rhs.fence, nullptr); - if (fence) { - fence->RedirectProtection(&rhs, this); - } -} - -VKFenceWatch& VKFenceWatch::operator=(VKFenceWatch&& rhs) noexcept { - fence = std::exchange(rhs.fence, nullptr); - if (fence) { - fence->RedirectProtection(&rhs, this); - } - return *this; -} - -VKFenceWatch::~VKFenceWatch() { - if (fence) { - fence->Unprotect(this); - } -} - -void VKFenceWatch::Wait() { - if (fence == nullptr) { - return; - } - fence->Wait(); - fence->Unprotect(this); -} - -void VKFenceWatch::Watch(VKFence& new_fence) { - Wait(); - fence = &new_fence; - fence->Protect(this); -} - -bool VKFenceWatch::TryWatch(VKFence& new_fence) { - if (fence) { - return false; - } - fence = &new_fence; - fence->Protect(this); - return true; -} - -void VKFenceWatch::OnFenceRemoval(VKFence* signaling_fence) { - ASSERT_MSG(signaling_fence == fence, "Removing the wrong fence"); - fence = nullptr; -} - -VKFencedPool::VKFencedPool(std::size_t grow_step) : grow_step{grow_step} {} - -VKFencedPool::~VKFencedPool() = default; - -std::size_t VKFencedPool::CommitResource(VKFence& fence) { - const auto Search = [&](std::size_t begin, std::size_t end) -> std::optional<std::size_t> { - for (std::size_t iterator = begin; iterator < end; ++iterator) { - if (watches[iterator]->TryWatch(fence)) { - // The resource is now being watched, a free resource was successfully found. - return iterator; - } - } - return {}; - }; - // Try to find a free resource from the hinted position to the end. - auto found = Search(free_iterator, watches.size()); - if (!found) { - // Search from beginning to the hinted position. - found = Search(0, free_iterator); - if (!found) { - // Both searches failed, the pool is full; handle it. - const std::size_t free_resource = ManageOverflow(); - - // Watch will wait for the resource to be free. - watches[free_resource]->Watch(fence); - found = free_resource; - } - } - // Free iterator is hinted to the resource after the one that's been commited. - free_iterator = (*found + 1) % watches.size(); - return *found; -} - -std::size_t VKFencedPool::ManageOverflow() { - const std::size_t old_capacity = watches.size(); - Grow(); - - // The last entry is guaranted to be free, since it's the first element of the freshly - // allocated resources. - return old_capacity; -} - -void VKFencedPool::Grow() { - const std::size_t old_capacity = watches.size(); - watches.resize(old_capacity + grow_step); - std::generate(watches.begin() + old_capacity, watches.end(), - []() { return std::make_unique<VKFenceWatch>(); }); - Allocate(old_capacity, old_capacity + grow_step); -} - -VKResourceManager::VKResourceManager(const VKDevice& device) : device{device} { - GrowFences(FENCES_GROW_STEP); - command_buffer_pool = std::make_unique<CommandBufferPool>(device); -} - -VKResourceManager::~VKResourceManager() = default; - -VKFence& VKResourceManager::CommitFence() { - const auto StepFences = [&](bool gpu_wait, bool owner_wait) -> VKFence* { - const auto Tick = [=](auto& fence) { return fence->Tick(gpu_wait, owner_wait); }; - const auto hinted = fences.begin() + fences_iterator; - - auto it = std::find_if(hinted, fences.end(), Tick); - if (it == fences.end()) { - it = std::find_if(fences.begin(), hinted, Tick); - if (it == hinted) { - return nullptr; - } - } - fences_iterator = std::distance(fences.begin(), it) + 1; - if (fences_iterator >= fences.size()) - fences_iterator = 0; - - auto& fence = *it; - fence->Commit(); - return fence.get(); - }; - - VKFence* found_fence = StepFences(false, false); - if (!found_fence) { - // Try again, this time waiting. - found_fence = StepFences(true, false); - - if (!found_fence) { - // Allocate new fences and try again. - LOG_INFO(Render_Vulkan, "Allocating new fences {} -> {}", fences.size(), - fences.size() + FENCES_GROW_STEP); - - GrowFences(FENCES_GROW_STEP); - found_fence = StepFences(true, false); - ASSERT(found_fence != nullptr); - } - } - return *found_fence; -} - -VkCommandBuffer VKResourceManager::CommitCommandBuffer(VKFence& fence) { - return command_buffer_pool->Commit(fence); -} - -void VKResourceManager::GrowFences(std::size_t new_fences_count) { - const std::size_t previous_size = fences.size(); - fences.resize(previous_size + new_fences_count); - - std::generate(fences.begin() + previous_size, fences.end(), - [this] { return std::make_unique<VKFence>(device); }); -} - -} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_resource_manager.h b/src/video_core/renderer_vulkan/vk_resource_manager.h deleted file mode 100644 index f683d2276..000000000 --- a/src/video_core/renderer_vulkan/vk_resource_manager.h +++ /dev/null @@ -1,196 +0,0 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <cstddef> -#include <memory> -#include <vector> -#include "video_core/renderer_vulkan/wrapper.h" - -namespace Vulkan { - -class VKDevice; -class VKFence; -class VKResourceManager; - -class CommandBufferPool; - -/// Interface for a Vulkan resource -class VKResource { -public: - explicit VKResource(); - virtual ~VKResource(); - - /** - * Signals the object that an owning fence has been signaled. - * @param signaling_fence Fence that signals its usage end. - */ - virtual void OnFenceRemoval(VKFence* signaling_fence) = 0; -}; - -/** - * Fences take ownership of objects, protecting them from GPU-side or driver-side concurrent access. - * They must be commited from the resource manager. Their usage flow is: commit the fence from the - * resource manager, protect resources with it and use them, send the fence to an execution queue - * and Wait for it if needed and then call Release. Used resources will automatically be signaled - * when they are free to be reused. - * @brief Protects resources for concurrent usage and signals its release. - */ -class VKFence { - friend class VKResourceManager; - -public: - explicit VKFence(const VKDevice& device); - ~VKFence(); - - /** - * Waits for the fence to be signaled. - * @warning You must have ownership of the fence and it has to be previously sent to a queue to - * call this function. - */ - void Wait(); - - /** - * Releases ownership of the fence. Pass after it has been sent to an execution queue. - * Unmanaged usage of the fence after the call will result in undefined behavior because it may - * be being used for something else. - */ - void Release(); - - /// Protects a resource with this fence. - void Protect(VKResource* resource); - - /// Removes protection for a resource. - void Unprotect(VKResource* resource); - - /// Redirects one protected resource to a new address. - void RedirectProtection(VKResource* old_resource, VKResource* new_resource) noexcept; - - /// Retreives the fence. - operator VkFence() const { - return *handle; - } - -private: - /// Take ownership of the fence. - void Commit(); - - /** - * Updates the fence status. - * @warning Waiting for the owner might soft lock the execution. - * @param gpu_wait Wait for the fence to be signaled by the driver. - * @param owner_wait Wait for the owner to signal its freedom. - * @returns True if the fence is free. Waiting for gpu and owner will always return true. - */ - bool Tick(bool gpu_wait, bool owner_wait); - - const VKDevice& device; ///< Device handler - vk::Fence handle; ///< Vulkan fence - std::vector<VKResource*> protected_resources; ///< List of resources protected by this fence - bool is_owned = false; ///< The fence has been commited but not released yet. - bool is_used = false; ///< The fence has been commited but it has not been checked to be free. -}; - -/** - * A fence watch is used to keep track of the usage of a fence and protect a resource or set of - * resources without having to inherit VKResource from their handlers. - */ -class VKFenceWatch final : public VKResource { -public: - explicit VKFenceWatch(); - VKFenceWatch(VKFence& initial_fence); - VKFenceWatch(VKFenceWatch&&) noexcept; - VKFenceWatch(const VKFenceWatch&) = delete; - ~VKFenceWatch() override; - - VKFenceWatch& operator=(VKFenceWatch&&) noexcept; - - /// Waits for the fence to be released. - void Wait(); - - /** - * Waits for a previous fence and watches a new one. - * @param new_fence New fence to wait to. - */ - void Watch(VKFence& new_fence); - - /** - * Checks if it's currently being watched and starts watching it if it's available. - * @returns True if a watch has started, false if it's being watched. - */ - bool TryWatch(VKFence& new_fence); - - void OnFenceRemoval(VKFence* signaling_fence) override; - - /** - * Do not use it paired with Watch. Use TryWatch instead. - * Returns true when the watch is free. - */ - bool IsUsed() const { - return fence != nullptr; - } - -private: - VKFence* fence{}; ///< Fence watching this resource. nullptr when the watch is free. -}; - -/** - * Handles a pool of resources protected by fences. Manages resource overflow allocating more - * resources. - */ -class VKFencedPool { -public: - explicit VKFencedPool(std::size_t grow_step); - virtual ~VKFencedPool(); - -protected: - /** - * Commits a free resource and protects it with a fence. It may allocate new resources. - * @param fence Fence that protects the commited resource. - * @returns Index of the resource commited. - */ - std::size_t CommitResource(VKFence& fence); - - /// Called when a chunk of resources have to be allocated. - virtual void Allocate(std::size_t begin, std::size_t end) = 0; - -private: - /// Manages pool overflow allocating new resources. - std::size_t ManageOverflow(); - - /// Allocates a new page of resources. - void Grow(); - - std::size_t grow_step = 0; ///< Number of new resources created after an overflow - std::size_t free_iterator = 0; ///< Hint to where the next free resources is likely to be found - std::vector<std::unique_ptr<VKFenceWatch>> watches; ///< Set of watched resources -}; - -/** - * The resource manager handles all resources that can be protected with a fence avoiding - * driver-side or GPU-side concurrent usage. Usage is documented in VKFence. - */ -class VKResourceManager final { -public: - explicit VKResourceManager(const VKDevice& device); - ~VKResourceManager(); - - /// Commits a fence. It has to be sent to a queue and released. - VKFence& CommitFence(); - - /// Commits an unused command buffer and protects it with a fence. - VkCommandBuffer CommitCommandBuffer(VKFence& fence); - -private: - /// Allocates new fences. - void GrowFences(std::size_t new_fences_count); - - const VKDevice& device; ///< Device handler. - std::size_t fences_iterator = 0; ///< Index where a free fence is likely to be found. - std::vector<std::unique_ptr<VKFence>> fences; ///< Pool of fences. - std::unique_ptr<CommandBufferPool> command_buffer_pool; ///< Pool of command buffers. -}; - -} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_resource_pool.cpp b/src/video_core/renderer_vulkan/vk_resource_pool.cpp new file mode 100644 index 000000000..ee274ac59 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_resource_pool.cpp @@ -0,0 +1,63 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <optional> + +#include "video_core/renderer_vulkan/vk_master_semaphore.h" +#include "video_core/renderer_vulkan/vk_resource_pool.h" + +namespace Vulkan { + +ResourcePool::ResourcePool(MasterSemaphore& master_semaphore_, size_t grow_step_) + : master_semaphore{master_semaphore_}, grow_step{grow_step_} {} + +ResourcePool::~ResourcePool() = default; + +size_t ResourcePool::CommitResource() { + // Refresh semaphore to query updated results + master_semaphore.Refresh(); + + const auto search = [this](size_t begin, size_t end) -> std::optional<size_t> { + for (size_t iterator = begin; iterator < end; ++iterator) { + if (master_semaphore.IsFree(ticks[iterator])) { + ticks[iterator] = master_semaphore.CurrentTick(); + return iterator; + } + } + return {}; + }; + // Try to find a free resource from the hinted position to the end. + auto found = search(free_iterator, ticks.size()); + if (!found) { + // Search from beginning to the hinted position. + found = search(0, free_iterator); + if (!found) { + // Both searches failed, the pool is full; handle it. + const size_t free_resource = ManageOverflow(); + + ticks[free_resource] = master_semaphore.CurrentTick(); + found = free_resource; + } + } + // Free iterator is hinted to the resource after the one that's been commited. + free_iterator = (*found + 1) % ticks.size(); + return *found; +} + +size_t ResourcePool::ManageOverflow() { + const size_t old_capacity = ticks.size(); + Grow(); + + // The last entry is guaranted to be free, since it's the first element of the freshly + // allocated resources. + return old_capacity; +} + +void ResourcePool::Grow() { + const size_t old_capacity = ticks.size(); + ticks.resize(old_capacity + grow_step); + Allocate(old_capacity, old_capacity + grow_step); +} + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_resource_pool.h b/src/video_core/renderer_vulkan/vk_resource_pool.h new file mode 100644 index 000000000..a018c7ec2 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_resource_pool.h @@ -0,0 +1,43 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <vector> + +#include "common/common_types.h" + +namespace Vulkan { + +class MasterSemaphore; + +/** + * Handles a pool of resources protected by fences. Manages resource overflow allocating more + * resources. + */ +class ResourcePool { +public: + explicit ResourcePool(MasterSemaphore& master_semaphore, size_t grow_step); + virtual ~ResourcePool(); + +protected: + size_t CommitResource(); + + /// Called when a chunk of resources have to be allocated. + virtual void Allocate(size_t begin, size_t end) = 0; + +private: + /// Manages pool overflow allocating new resources. + size_t ManageOverflow(); + + /// Allocates a new page of resources. + void Grow(); + + MasterSemaphore& master_semaphore; + size_t grow_step = 0; ///< Number of new resources created after an overflow + size_t free_iterator = 0; ///< Hint to where the next free resources is likely to be found + std::vector<u64> ticks; ///< Ticks for each resource +}; + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index dbbd0961a..1a483dc71 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -10,9 +10,10 @@ #include "common/microprofile.h" #include "common/thread.h" +#include "video_core/renderer_vulkan/vk_command_pool.h" #include "video_core/renderer_vulkan/vk_device.h" +#include "video_core/renderer_vulkan/vk_master_semaphore.h" #include "video_core/renderer_vulkan/vk_query_cache.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_state_tracker.h" #include "video_core/renderer_vulkan/wrapper.h" @@ -35,10 +36,10 @@ void VKScheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf) { last = nullptr; } -VKScheduler::VKScheduler(const VKDevice& device, VKResourceManager& resource_manager, - StateTracker& state_tracker) - : device{device}, resource_manager{resource_manager}, state_tracker{state_tracker}, - next_fence{&resource_manager.CommitFence()} { +VKScheduler::VKScheduler(const VKDevice& device_, StateTracker& state_tracker_) + : device{device_}, state_tracker{state_tracker_}, + master_semaphore{std::make_unique<MasterSemaphore>(device)}, + command_pool{std::make_unique<CommandPool>(*master_semaphore, device)} { AcquireNewChunk(); AllocateNewContext(); worker_thread = std::thread(&VKScheduler::WorkerThread, this); @@ -50,20 +51,27 @@ VKScheduler::~VKScheduler() { worker_thread.join(); } -void VKScheduler::Flush(bool release_fence, VkSemaphore semaphore) { +u64 VKScheduler::CurrentTick() const noexcept { + return master_semaphore->CurrentTick(); +} + +bool VKScheduler::IsFree(u64 tick) const noexcept { + return master_semaphore->IsFree(tick); +} + +void VKScheduler::Wait(u64 tick) { + master_semaphore->Wait(tick); +} + +void VKScheduler::Flush(VkSemaphore semaphore) { SubmitExecution(semaphore); - if (release_fence) { - current_fence->Release(); - } AllocateNewContext(); } -void VKScheduler::Finish(bool release_fence, VkSemaphore semaphore) { +void VKScheduler::Finish(VkSemaphore semaphore) { + const u64 presubmit_tick = CurrentTick(); SubmitExecution(semaphore); - current_fence->Wait(); - if (release_fence) { - current_fence->Release(); - } + Wait(presubmit_tick); AllocateNewContext(); } @@ -160,18 +168,38 @@ void VKScheduler::SubmitExecution(VkSemaphore semaphore) { current_cmdbuf.End(); + const VkSemaphore timeline_semaphore = master_semaphore->Handle(); + const u32 num_signal_semaphores = semaphore ? 2U : 1U; + + const u64 signal_value = master_semaphore->CurrentTick(); + const u64 wait_value = signal_value - 1; + const VkPipelineStageFlags wait_stage_mask = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT; + + master_semaphore->NextTick(); + + const std::array signal_values{signal_value, u64(0)}; + const std::array signal_semaphores{timeline_semaphore, semaphore}; + + const VkTimelineSemaphoreSubmitInfoKHR timeline_si{ + .sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR, + .pNext = nullptr, + .waitSemaphoreValueCount = 1, + .pWaitSemaphoreValues = &wait_value, + .signalSemaphoreValueCount = num_signal_semaphores, + .pSignalSemaphoreValues = signal_values.data(), + }; const VkSubmitInfo submit_info{ .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO, - .pNext = nullptr, - .waitSemaphoreCount = 0, - .pWaitSemaphores = nullptr, - .pWaitDstStageMask = nullptr, + .pNext = &timeline_si, + .waitSemaphoreCount = 1, + .pWaitSemaphores = &timeline_semaphore, + .pWaitDstStageMask = &wait_stage_mask, .commandBufferCount = 1, .pCommandBuffers = current_cmdbuf.address(), - .signalSemaphoreCount = semaphore ? 1U : 0U, - .pSignalSemaphores = &semaphore, + .signalSemaphoreCount = num_signal_semaphores, + .pSignalSemaphores = signal_semaphores.data(), }; - switch (const VkResult result = device.GetGraphicsQueue().Submit(submit_info, *current_fence)) { + switch (const VkResult result = device.GetGraphicsQueue().Submit(submit_info)) { case VK_SUCCESS: break; case VK_ERROR_DEVICE_LOST: @@ -183,14 +211,9 @@ void VKScheduler::SubmitExecution(VkSemaphore semaphore) { } void VKScheduler::AllocateNewContext() { - ++ticks; - std::unique_lock lock{mutex}; - current_fence = next_fence; - next_fence = &resource_manager.CommitFence(); - current_cmdbuf = vk::CommandBuffer(resource_manager.CommitCommandBuffer(*current_fence), - device.GetDispatchLoader()); + current_cmdbuf = vk::CommandBuffer(command_pool->Commit(), device.GetDispatchLoader()); current_cmdbuf.Begin({ .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, .pNext = nullptr, diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index 970a65566..7be8a19f0 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -16,42 +16,33 @@ namespace Vulkan { +class CommandPool; +class MasterSemaphore; class StateTracker; class VKDevice; -class VKFence; class VKQueryCache; -class VKResourceManager; - -class VKFenceView { -public: - VKFenceView() = default; - VKFenceView(VKFence* const& fence) : fence{fence} {} - - VKFence* operator->() const noexcept { - return fence; - } - - operator VKFence&() const noexcept { - return *fence; - } - -private: - VKFence* const& fence; -}; /// The scheduler abstracts command buffer and fence management with an interface that's able to do /// OpenGL-like operations on Vulkan command buffers. class VKScheduler { public: - explicit VKScheduler(const VKDevice& device, VKResourceManager& resource_manager, - StateTracker& state_tracker); + explicit VKScheduler(const VKDevice& device, StateTracker& state_tracker); ~VKScheduler(); + /// Returns the current command buffer tick. + [[nodiscard]] u64 CurrentTick() const noexcept; + + /// Returns true when a tick has been triggered by the GPU. + [[nodiscard]] bool IsFree(u64 tick) const noexcept; + + /// Waits for the given tick to trigger on the GPU. + void Wait(u64 tick); + /// Sends the current execution context to the GPU. - void Flush(bool release_fence = true, VkSemaphore semaphore = nullptr); + void Flush(VkSemaphore semaphore = nullptr); /// Sends the current execution context to the GPU and waits for it to complete. - void Finish(bool release_fence = true, VkSemaphore semaphore = nullptr); + void Finish(VkSemaphore semaphore = nullptr); /// Waits for the worker thread to finish executing everything. After this function returns it's /// safe to touch worker resources. @@ -86,14 +77,9 @@ public: (void)chunk->Record(command); } - /// Gets a reference to the current fence. - VKFenceView GetFence() const { - return current_fence; - } - - /// Returns the current command buffer tick. - u64 Ticks() const { - return ticks; + /// Returns the master timeline semaphore. + [[nodiscard]] MasterSemaphore& GetMasterSemaphore() const noexcept { + return *master_semaphore; } private: @@ -171,6 +157,13 @@ private: std::array<u8, 0x8000> data{}; }; + struct State { + VkRenderPass renderpass = nullptr; + VkFramebuffer framebuffer = nullptr; + VkExtent2D render_area = {0, 0}; + VkPipeline graphics_pipeline = nullptr; + }; + void WorkerThread(); void SubmitExecution(VkSemaphore semaphore); @@ -186,30 +179,23 @@ private: void AcquireNewChunk(); const VKDevice& device; - VKResourceManager& resource_manager; StateTracker& state_tracker; + std::unique_ptr<MasterSemaphore> master_semaphore; + std::unique_ptr<CommandPool> command_pool; + VKQueryCache* query_cache = nullptr; vk::CommandBuffer current_cmdbuf; - VKFence* current_fence = nullptr; - VKFence* next_fence = nullptr; - - struct State { - VkRenderPass renderpass = nullptr; - VkFramebuffer framebuffer = nullptr; - VkExtent2D render_area = {0, 0}; - VkPipeline graphics_pipeline = nullptr; - } state; std::unique_ptr<CommandChunk> chunk; std::thread worker_thread; + State state; Common::SPSCQueue<std::unique_ptr<CommandChunk>> chunk_queue; Common::SPSCQueue<std::unique_ptr<CommandChunk>> chunk_reserve; std::mutex mutex; std::condition_variable cv; - std::atomic<u64> ticks = 0; bool quit = false; }; diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp index 5eca0ab91..2fd3b7f39 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp @@ -10,36 +10,18 @@ #include "common/bit_util.h" #include "common/common_types.h" #include "video_core/renderer_vulkan/vk_device.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" #include "video_core/renderer_vulkan/wrapper.h" namespace Vulkan { -VKStagingBufferPool::StagingBuffer::StagingBuffer(std::unique_ptr<VKBuffer> buffer, VKFence& fence, - u64 last_epoch) - : buffer{std::move(buffer)}, watch{fence}, last_epoch{last_epoch} {} +VKStagingBufferPool::StagingBuffer::StagingBuffer(std::unique_ptr<VKBuffer> buffer_) + : buffer{std::move(buffer_)} {} -VKStagingBufferPool::StagingBuffer::StagingBuffer(StagingBuffer&& rhs) noexcept { - buffer = std::move(rhs.buffer); - watch = std::move(rhs.watch); - last_epoch = rhs.last_epoch; -} - -VKStagingBufferPool::StagingBuffer::~StagingBuffer() = default; - -VKStagingBufferPool::StagingBuffer& VKStagingBufferPool::StagingBuffer::operator=( - StagingBuffer&& rhs) noexcept { - buffer = std::move(rhs.buffer); - watch = std::move(rhs.watch); - last_epoch = rhs.last_epoch; - return *this; -} - -VKStagingBufferPool::VKStagingBufferPool(const VKDevice& device, VKMemoryManager& memory_manager, - VKScheduler& scheduler) - : device{device}, memory_manager{memory_manager}, scheduler{scheduler} {} +VKStagingBufferPool::VKStagingBufferPool(const VKDevice& device_, VKMemoryManager& memory_manager_, + VKScheduler& scheduler_) + : device{device_}, memory_manager{memory_manager_}, scheduler{scheduler_} {} VKStagingBufferPool::~VKStagingBufferPool() = default; @@ -51,7 +33,6 @@ VKBuffer& VKStagingBufferPool::GetUnusedBuffer(std::size_t size, bool host_visib } void VKStagingBufferPool::TickFrame() { - ++epoch; current_delete_level = (current_delete_level + 1) % NumLevels; ReleaseCache(true); @@ -59,11 +40,12 @@ void VKStagingBufferPool::TickFrame() { } VKBuffer* VKStagingBufferPool::TryGetReservedBuffer(std::size_t size, bool host_visible) { - for (auto& entry : GetCache(host_visible)[Common::Log2Ceil64(size)].entries) { - if (entry.watch.TryWatch(scheduler.GetFence())) { - entry.last_epoch = epoch; - return &*entry.buffer; + for (StagingBuffer& entry : GetCache(host_visible)[Common::Log2Ceil64(size)].entries) { + if (!scheduler.IsFree(entry.tick)) { + continue; } + entry.tick = scheduler.CurrentTick(); + return &*entry.buffer; } return nullptr; } @@ -86,8 +68,10 @@ VKBuffer& VKStagingBufferPool::CreateStagingBuffer(std::size_t size, bool host_v }); buffer->commit = memory_manager.Commit(buffer->handle, host_visible); - auto& entries = GetCache(host_visible)[log2].entries; - return *entries.emplace_back(std::move(buffer), scheduler.GetFence(), epoch).buffer; + std::vector<StagingBuffer>& entries = GetCache(host_visible)[log2].entries; + StagingBuffer& entry = entries.emplace_back(std::move(buffer)); + entry.tick = scheduler.CurrentTick(); + return *entry.buffer; } VKStagingBufferPool::StagingBuffersCache& VKStagingBufferPool::GetCache(bool host_visible) { @@ -109,9 +93,8 @@ u64 VKStagingBufferPool::ReleaseLevel(StagingBuffersCache& cache, std::size_t lo auto& entries = staging.entries; const std::size_t old_size = entries.size(); - const auto is_deleteable = [this](const auto& entry) { - static constexpr u64 epochs_to_destroy = 180; - return entry.last_epoch + epochs_to_destroy < epoch && !entry.watch.IsUsed(); + const auto is_deleteable = [this](const StagingBuffer& entry) { + return scheduler.IsFree(entry.tick); }; const std::size_t begin_offset = staging.delete_index; const std::size_t end_offset = std::min(begin_offset + deletions_per_tick, old_size); diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h index 3c4901437..2dd5049ac 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h @@ -10,13 +10,11 @@ #include "common/common_types.h" #include "video_core/renderer_vulkan/vk_memory_manager.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" #include "video_core/renderer_vulkan/wrapper.h" namespace Vulkan { class VKDevice; -class VKFenceWatch; class VKScheduler; struct VKBuffer final { @@ -36,16 +34,10 @@ public: private: struct StagingBuffer final { - explicit StagingBuffer(std::unique_ptr<VKBuffer> buffer, VKFence& fence, u64 last_epoch); - StagingBuffer(StagingBuffer&& rhs) noexcept; - StagingBuffer(const StagingBuffer&) = delete; - ~StagingBuffer(); - - StagingBuffer& operator=(StagingBuffer&& rhs) noexcept; + explicit StagingBuffer(std::unique_ptr<VKBuffer> buffer); std::unique_ptr<VKBuffer> buffer; - VKFenceWatch watch; - u64 last_epoch = 0; + u64 tick = 0; }; struct StagingBuffers final { @@ -73,8 +65,6 @@ private: StagingBuffersCache host_staging_buffers; StagingBuffersCache device_staging_buffers; - u64 epoch = 0; - std::size_t current_delete_level = 0; }; diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.cpp b/src/video_core/renderer_vulkan/vk_state_tracker.cpp index 9151d9fb1..5d2c4a796 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.cpp +++ b/src/video_core/renderer_vulkan/vk_state_tracker.cpp @@ -42,7 +42,6 @@ Flags MakeInvalidationFlags() { flags[DepthWriteEnable] = true; flags[DepthCompareOp] = true; flags[FrontFace] = true; - flags[PrimitiveTopology] = true; flags[StencilOp] = true; flags[StencilTestEnable] = true; return flags; @@ -112,10 +111,6 @@ void SetupDirtyFrontFace(Tables& tables) { table[OFF(screen_y_control)] = FrontFace; } -void SetupDirtyPrimitiveTopology(Tables& tables) { - tables[0][OFF(draw.topology)] = PrimitiveTopology; -} - void SetupDirtyStencilOp(Tables& tables) { auto& table = tables[0]; table[OFF(stencil_front_op_fail)] = StencilOp; @@ -137,12 +132,9 @@ void SetupDirtyStencilTestEnable(Tables& tables) { } // Anonymous namespace -StateTracker::StateTracker(Core::System& system) - : system{system}, invalidation_flags{MakeInvalidationFlags()} {} - -void StateTracker::Initialize() { - auto& dirty = system.GPU().Maxwell3D().dirty; - auto& tables = dirty.tables; +StateTracker::StateTracker(Tegra::GPU& gpu) + : flags{gpu.Maxwell3D().dirty.flags}, invalidation_flags{MakeInvalidationFlags()} { + auto& tables = gpu.Maxwell3D().dirty.tables; SetupDirtyRenderTargets(tables); SetupDirtyViewports(tables); SetupDirtyScissors(tables); @@ -156,13 +148,8 @@ void StateTracker::Initialize() { SetupDirtyDepthWriteEnable(tables); SetupDirtyDepthCompareOp(tables); SetupDirtyFrontFace(tables); - SetupDirtyPrimitiveTopology(tables); SetupDirtyStencilOp(tables); SetupDirtyStencilTestEnable(tables); } -void StateTracker::InvalidateCommandBufferState() { - system.GPU().Maxwell3D().dirty.flags |= invalidation_flags; -} - } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.h b/src/video_core/renderer_vulkan/vk_state_tracker.h index 54ca0d6c6..1de789e57 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.h +++ b/src/video_core/renderer_vulkan/vk_state_tracker.h @@ -32,7 +32,6 @@ enum : u8 { DepthWriteEnable, DepthCompareOp, FrontFace, - PrimitiveTopology, StencilOp, StencilTestEnable, @@ -43,12 +42,15 @@ static_assert(Last <= std::numeric_limits<u8>::max()); } // namespace Dirty class StateTracker { -public: - explicit StateTracker(Core::System& system); + using Maxwell = Tegra::Engines::Maxwell3D::Regs; - void Initialize(); +public: + explicit StateTracker(Tegra::GPU& gpu); - void InvalidateCommandBufferState(); + void InvalidateCommandBufferState() { + flags |= invalidation_flags; + current_topology = INVALID_TOPOLOGY; + } bool TouchViewports() { return Exchange(Dirty::Viewports, false); @@ -102,10 +104,6 @@ public: return Exchange(Dirty::FrontFace, false); } - bool TouchPrimitiveTopology() { - return Exchange(Dirty::PrimitiveTopology, false); - } - bool TouchStencilOp() { return Exchange(Dirty::StencilOp, false); } @@ -114,16 +112,24 @@ public: return Exchange(Dirty::StencilTestEnable, false); } + bool ChangePrimitiveTopology(Maxwell::PrimitiveTopology new_topology) { + const bool has_changed = current_topology != new_topology; + current_topology = new_topology; + return has_changed; + } + private: + static constexpr auto INVALID_TOPOLOGY = static_cast<Maxwell::PrimitiveTopology>(~0u); + bool Exchange(std::size_t id, bool new_value) const noexcept { - auto& flags = system.GPU().Maxwell3D().dirty.flags; const bool is_dirty = flags[id]; flags[id] = new_value; return is_dirty; } - Core::System& system; + Tegra::Engines::Maxwell3D::DirtyState::Flags& flags; Tegra::Engines::Maxwell3D::DirtyState::Flags invalidation_flags; + Maxwell::PrimitiveTopology current_topology = INVALID_TOPOLOGY; }; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_stream_buffer.cpp b/src/video_core/renderer_vulkan/vk_stream_buffer.cpp index a5526a3f5..1b59612b9 100644 --- a/src/video_core/renderer_vulkan/vk_stream_buffer.cpp +++ b/src/video_core/renderer_vulkan/vk_stream_buffer.cpp @@ -11,7 +11,6 @@ #include "common/alignment.h" #include "common/assert.h" #include "video_core/renderer_vulkan/vk_device.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_stream_buffer.h" #include "video_core/renderer_vulkan/wrapper.h" @@ -57,9 +56,9 @@ u32 GetMemoryType(const VkPhysicalDeviceMemoryProperties& properties, } // Anonymous namespace -VKStreamBuffer::VKStreamBuffer(const VKDevice& device, VKScheduler& scheduler, +VKStreamBuffer::VKStreamBuffer(const VKDevice& device_, VKScheduler& scheduler_, VkBufferUsageFlags usage) - : device{device}, scheduler{scheduler} { + : device{device_}, scheduler{scheduler_} { CreateBuffers(usage); ReserveWatches(current_watches, WATCHES_INITIAL_RESERVE); ReserveWatches(previous_watches, WATCHES_INITIAL_RESERVE); @@ -111,7 +110,7 @@ void VKStreamBuffer::Unmap(u64 size) { } auto& watch = current_watches[current_watch_cursor++]; watch.upper_bound = offset; - watch.fence.Watch(scheduler.GetFence()); + watch.tick = scheduler.CurrentTick(); } void VKStreamBuffer::CreateBuffers(VkBufferUsageFlags usage) { @@ -121,7 +120,8 @@ void VKStreamBuffer::CreateBuffers(VkBufferUsageFlags usage) { // Substract from the preferred heap size some bytes to avoid getting out of memory. const VkDeviceSize heap_size = memory_properties.memoryHeaps[preferred_heap].size; - const VkDeviceSize allocable_size = heap_size - 9 * 1024 * 1024; + // As per DXVK's example, using `heap_size / 2` + const VkDeviceSize allocable_size = heap_size / 2; buffer = device.GetLogical().CreateBuffer({ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, .pNext = nullptr, @@ -157,7 +157,7 @@ void VKStreamBuffer::WaitPendingOperations(u64 requested_upper_bound) { while (requested_upper_bound < wait_bound && wait_cursor < *invalidation_mark) { auto& watch = previous_watches[wait_cursor]; wait_bound = watch.upper_bound; - watch.fence.Wait(); + scheduler.Wait(watch.tick); ++wait_cursor; } } diff --git a/src/video_core/renderer_vulkan/vk_stream_buffer.h b/src/video_core/renderer_vulkan/vk_stream_buffer.h index 689f0d276..5e15ad78f 100644 --- a/src/video_core/renderer_vulkan/vk_stream_buffer.h +++ b/src/video_core/renderer_vulkan/vk_stream_buffer.h @@ -14,7 +14,6 @@ namespace Vulkan { class VKDevice; -class VKFence; class VKFenceWatch; class VKScheduler; @@ -44,8 +43,8 @@ public: } private: - struct Watch final { - VKFenceWatch fence; + struct Watch { + u64 tick{}; u64 upper_bound{}; }; diff --git a/src/video_core/renderer_vulkan/vk_swapchain.cpp b/src/video_core/renderer_vulkan/vk_swapchain.cpp index 6bfd2abae..9636a7c65 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.cpp +++ b/src/video_core/renderer_vulkan/vk_swapchain.cpp @@ -12,7 +12,7 @@ #include "core/core.h" #include "core/frontend/framebuffer_layout.h" #include "video_core/renderer_vulkan/vk_device.h" -#include "video_core/renderer_vulkan/vk_resource_manager.h" +#include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_swapchain.h" #include "video_core/renderer_vulkan/wrapper.h" @@ -56,8 +56,8 @@ VkExtent2D ChooseSwapExtent(const VkSurfaceCapabilitiesKHR& capabilities, u32 wi } // Anonymous namespace -VKSwapchain::VKSwapchain(VkSurfaceKHR surface, const VKDevice& device) - : surface{surface}, device{device} {} +VKSwapchain::VKSwapchain(VkSurfaceKHR surface_, const VKDevice& device_, VKScheduler& scheduler_) + : surface{surface_}, device{device_}, scheduler{scheduler_} {} VKSwapchain::~VKSwapchain() = default; @@ -75,21 +75,18 @@ void VKSwapchain::Create(u32 width, u32 height, bool srgb) { CreateSemaphores(); CreateImageViews(); - fences.resize(image_count, nullptr); + resource_ticks.clear(); + resource_ticks.resize(image_count); } void VKSwapchain::AcquireNextImage() { device.GetLogical().AcquireNextImageKHR(*swapchain, std::numeric_limits<u64>::max(), *present_semaphores[frame_index], {}, &image_index); - if (auto& fence = fences[image_index]; fence) { - fence->Wait(); - fence->Release(); - fence = nullptr; - } + scheduler.Wait(resource_ticks[image_index]); } -bool VKSwapchain::Present(VkSemaphore render_semaphore, VKFence& fence) { +bool VKSwapchain::Present(VkSemaphore render_semaphore) { const VkSemaphore present_semaphore{*present_semaphores[frame_index]}; const std::array<VkSemaphore, 2> semaphores{present_semaphore, render_semaphore}; const auto present_queue{device.GetPresentQueue()}; @@ -123,8 +120,7 @@ bool VKSwapchain::Present(VkSemaphore render_semaphore, VKFence& fence) { break; } - ASSERT(fences[image_index] == nullptr); - fences[image_index] = &fence; + resource_ticks[image_index] = scheduler.CurrentTick(); frame_index = (frame_index + 1) % static_cast<u32>(image_count); return recreated; } diff --git a/src/video_core/renderer_vulkan/vk_swapchain.h b/src/video_core/renderer_vulkan/vk_swapchain.h index a35d61345..6b39befdf 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.h +++ b/src/video_core/renderer_vulkan/vk_swapchain.h @@ -16,11 +16,11 @@ struct FramebufferLayout; namespace Vulkan { class VKDevice; -class VKFence; +class VKScheduler; class VKSwapchain { public: - explicit VKSwapchain(VkSurfaceKHR surface, const VKDevice& device); + explicit VKSwapchain(VkSurfaceKHR surface, const VKDevice& device, VKScheduler& scheduler); ~VKSwapchain(); /// Creates (or recreates) the swapchain with a given size. @@ -31,7 +31,7 @@ public: /// Presents the rendered image to the swapchain. Returns true when the swapchains had to be /// recreated. Takes responsability for the ownership of fence. - bool Present(VkSemaphore render_semaphore, VKFence& fence); + bool Present(VkSemaphore render_semaphore); /// Returns true when the framebuffer layout has changed. bool HasFramebufferChanged(const Layout::FramebufferLayout& framebuffer) const; @@ -74,6 +74,7 @@ private: const VkSurfaceKHR surface; const VKDevice& device; + VKScheduler& scheduler; vk::SwapchainKHR swapchain; @@ -81,7 +82,7 @@ private: std::vector<VkImage> images; std::vector<vk::ImageView> image_views; std::vector<vk::Framebuffer> framebuffers; - std::vector<VKFence*> fences; + std::vector<u64> resource_ticks; std::vector<vk::Semaphore> present_semaphores; u32 image_index{}; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 2c6f54101..f2c8f2ae1 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -188,12 +188,10 @@ u32 EncodeSwizzle(Tegra::Texture::SwizzleSource x_source, Tegra::Texture::Swizzl } // Anonymous namespace -CachedSurface::CachedSurface(Core::System& system, const VKDevice& device, - VKResourceManager& resource_manager, VKMemoryManager& memory_manager, +CachedSurface::CachedSurface(const VKDevice& device, VKMemoryManager& memory_manager, VKScheduler& scheduler, VKStagingBufferPool& staging_pool, GPUVAddr gpu_addr, const SurfaceParams& params) - : SurfaceBase<View>{gpu_addr, params, device.IsOptimalAstcSupported()}, system{system}, - device{device}, resource_manager{resource_manager}, + : SurfaceBase<View>{gpu_addr, params, device.IsOptimalAstcSupported()}, device{device}, memory_manager{memory_manager}, scheduler{scheduler}, staging_pool{staging_pool} { if (params.IsBuffer()) { buffer = CreateBuffer(device, params, host_memory_size); @@ -490,19 +488,20 @@ VkImageView CachedSurfaceView::GetAttachment() { return *render_target; } -VKTextureCache::VKTextureCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer, - const VKDevice& device, VKResourceManager& resource_manager, - VKMemoryManager& memory_manager, VKScheduler& scheduler, - VKStagingBufferPool& staging_pool) - : TextureCache(system, rasterizer, device.IsOptimalAstcSupported()), device{device}, - resource_manager{resource_manager}, memory_manager{memory_manager}, scheduler{scheduler}, - staging_pool{staging_pool} {} +VKTextureCache::VKTextureCache(VideoCore::RasterizerInterface& rasterizer, + Tegra::Engines::Maxwell3D& maxwell3d, + Tegra::MemoryManager& gpu_memory, const VKDevice& device_, + VKMemoryManager& memory_manager_, VKScheduler& scheduler_, + VKStagingBufferPool& staging_pool_) + : TextureCache(rasterizer, maxwell3d, gpu_memory, device_.IsOptimalAstcSupported()), + device{device_}, memory_manager{memory_manager_}, scheduler{scheduler_}, staging_pool{ + staging_pool_} {} VKTextureCache::~VKTextureCache() = default; Surface VKTextureCache::CreateSurface(GPUVAddr gpu_addr, const SurfaceParams& params) { - return std::make_shared<CachedSurface>(system, device, resource_manager, memory_manager, - scheduler, staging_pool, gpu_addr, params); + return std::make_shared<CachedSurface>(device, memory_manager, scheduler, staging_pool, + gpu_addr, params); } void VKTextureCache::ImageCopy(Surface& src_surface, Surface& dst_surface, diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 807e26c8a..39202feba 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -15,10 +15,6 @@ #include "video_core/texture_cache/surface_base.h" #include "video_core/texture_cache/texture_cache.h" -namespace Core { -class System; -} - namespace VideoCore { class RasterizerInterface; } @@ -27,7 +23,6 @@ namespace Vulkan { class RasterizerVulkan; class VKDevice; -class VKResourceManager; class VKScheduler; class VKStagingBufferPool; @@ -45,8 +40,7 @@ class CachedSurface final : public VideoCommon::SurfaceBase<View> { friend CachedSurfaceView; public: - explicit CachedSurface(Core::System& system, const VKDevice& device, - VKResourceManager& resource_manager, VKMemoryManager& memory_manager, + explicit CachedSurface(const VKDevice& device, VKMemoryManager& memory_manager, VKScheduler& scheduler, VKStagingBufferPool& staging_pool, GPUVAddr gpu_addr, const SurfaceParams& params); ~CachedSurface(); @@ -101,9 +95,7 @@ private: VkImageSubresourceRange GetImageSubresourceRange() const; - Core::System& system; const VKDevice& device; - VKResourceManager& resource_manager; VKMemoryManager& memory_manager; VKScheduler& scheduler; VKStagingBufferPool& staging_pool; @@ -201,10 +193,10 @@ private: class VKTextureCache final : public TextureCacheBase { public: - explicit VKTextureCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer, - const VKDevice& device, VKResourceManager& resource_manager, - VKMemoryManager& memory_manager, VKScheduler& scheduler, - VKStagingBufferPool& staging_pool); + explicit VKTextureCache(VideoCore::RasterizerInterface& rasterizer, + Tegra::Engines::Maxwell3D& maxwell3d, Tegra::MemoryManager& gpu_memory, + const VKDevice& device, VKMemoryManager& memory_manager, + VKScheduler& scheduler, VKStagingBufferPool& staging_pool); ~VKTextureCache(); private: @@ -219,7 +211,6 @@ private: void BufferCopy(Surface& src_surface, Surface& dst_surface) override; const VKDevice& device; - VKResourceManager& resource_manager; VKMemoryManager& memory_manager; VKScheduler& scheduler; VKStagingBufferPool& staging_pool; diff --git a/src/video_core/renderer_vulkan/wrapper.cpp b/src/video_core/renderer_vulkan/wrapper.cpp index 56055af1b..c034558a3 100644 --- a/src/video_core/renderer_vulkan/wrapper.cpp +++ b/src/video_core/renderer_vulkan/wrapper.cpp @@ -6,6 +6,7 @@ #include <exception> #include <memory> #include <optional> +#include <string_view> #include <utility> #include <vector> @@ -18,21 +19,42 @@ namespace Vulkan::vk { namespace { +template <typename Func> +void SortPhysicalDevices(std::vector<VkPhysicalDevice>& devices, const InstanceDispatch& dld, + Func&& func) { + // Calling GetProperties calls Vulkan more than needed. But they are supposed to be cheap + // functions. + std::stable_sort(devices.begin(), devices.end(), + [&dld, &func](VkPhysicalDevice lhs, VkPhysicalDevice rhs) { + return func(vk::PhysicalDevice(lhs, dld).GetProperties(), + vk::PhysicalDevice(rhs, dld).GetProperties()); + }); +} + +void SortPhysicalDevicesPerVendor(std::vector<VkPhysicalDevice>& devices, + const InstanceDispatch& dld, + std::initializer_list<u32> vendor_ids) { + for (auto it = vendor_ids.end(); it != vendor_ids.begin();) { + --it; + SortPhysicalDevices(devices, dld, [id = *it](const auto& lhs, const auto& rhs) { + return lhs.vendorID == id && rhs.vendorID != id; + }); + } +} + void SortPhysicalDevices(std::vector<VkPhysicalDevice>& devices, const InstanceDispatch& dld) { - std::stable_sort(devices.begin(), devices.end(), [&](auto lhs, auto rhs) { - // This will call Vulkan more than needed, but these calls are cheap. - const auto lhs_properties = vk::PhysicalDevice(lhs, dld).GetProperties(); - const auto rhs_properties = vk::PhysicalDevice(rhs, dld).GetProperties(); - - // Prefer discrete GPUs, Nvidia over AMD, AMD over Intel, Intel over the rest. - const bool preferred = - (lhs_properties.deviceType == VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU && - rhs_properties.deviceType != VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU) || - (lhs_properties.vendorID == 0x10DE && rhs_properties.vendorID != 0x10DE) || - (lhs_properties.vendorID == 0x1002 && rhs_properties.vendorID != 0x1002) || - (lhs_properties.vendorID == 0x8086 && rhs_properties.vendorID != 0x8086); - return !preferred; + // Sort by name, this will set a base and make GPUs with higher numbers appear first + // (e.g. GTX 1650 will intentionally be listed before a GTX 1080). + SortPhysicalDevices(devices, dld, [](const auto& lhs, const auto& rhs) { + return std::string_view{lhs.deviceName} > std::string_view{rhs.deviceName}; }); + // Prefer discrete over non-discrete + SortPhysicalDevices(devices, dld, [](const auto& lhs, const auto& rhs) { + return lhs.deviceType == VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU && + rhs.deviceType != VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU; + }); + // Prefer Nvidia over AMD, AMD over Intel, Intel over the rest. + SortPhysicalDevicesPerVendor(devices, dld, {0x10DE, 0x1002, 0x8086}); } template <typename T> @@ -149,6 +171,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkGetFenceStatus); X(vkGetImageMemoryRequirements); X(vkGetQueryPoolResults); + X(vkGetSemaphoreCounterValueKHR); X(vkMapMemory); X(vkQueueSubmit); X(vkResetFences); @@ -157,6 +180,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkUpdateDescriptorSetWithTemplateKHR); X(vkUpdateDescriptorSets); X(vkWaitForFences); + X(vkWaitSemaphoresKHR); #undef X } @@ -263,6 +287,22 @@ const char* ToString(VkResult result) noexcept { return "VK_ERROR_INVALID_DEVICE_ADDRESS_EXT"; case VkResult::VK_ERROR_FULL_SCREEN_EXCLUSIVE_MODE_LOST_EXT: return "VK_ERROR_FULL_SCREEN_EXCLUSIVE_MODE_LOST_EXT"; + case VkResult::VK_ERROR_UNKNOWN: + return "VK_ERROR_UNKNOWN"; + case VkResult::VK_ERROR_INCOMPATIBLE_VERSION_KHR: + return "VK_ERROR_INCOMPATIBLE_VERSION_KHR"; + case VkResult::VK_THREAD_IDLE_KHR: + return "VK_THREAD_IDLE_KHR"; + case VkResult::VK_THREAD_DONE_KHR: + return "VK_THREAD_DONE_KHR"; + case VkResult::VK_OPERATION_DEFERRED_KHR: + return "VK_OPERATION_DEFERRED_KHR"; + case VkResult::VK_OPERATION_NOT_DEFERRED_KHR: + return "VK_OPERATION_NOT_DEFERRED_KHR"; + case VkResult::VK_PIPELINE_COMPILE_REQUIRED_EXT: + return "VK_PIPELINE_COMPILE_REQUIRED_EXT"; + case VkResult::VK_RESULT_MAX_ENUM: + return "VK_RESULT_MAX_ENUM"; } return "Unknown"; } @@ -558,7 +598,10 @@ Semaphore Device::CreateSemaphore() const { .pNext = nullptr, .flags = 0, }; + return CreateSemaphore(ci); +} +Semaphore Device::CreateSemaphore(const VkSemaphoreCreateInfo& ci) const { VkSemaphore object; Check(dld->vkCreateSemaphore(handle, &ci, nullptr, &object)); return Semaphore(object, handle, *dld); @@ -644,7 +687,7 @@ ShaderModule Device::CreateShaderModule(const VkShaderModuleCreateInfo& ci) cons return ShaderModule(object, handle, *dld); } -Event Device::CreateNewEvent() const { +Event Device::CreateEvent() const { static constexpr VkEventCreateInfo ci{ .sType = VK_STRUCTURE_TYPE_EVENT_CREATE_INFO, .pNext = nullptr, diff --git a/src/video_core/renderer_vulkan/wrapper.h b/src/video_core/renderer_vulkan/wrapper.h index 748a94d2f..f64919623 100644 --- a/src/video_core/renderer_vulkan/wrapper.h +++ b/src/video_core/renderer_vulkan/wrapper.h @@ -267,6 +267,7 @@ struct DeviceDispatch : public InstanceDispatch { PFN_vkGetFenceStatus vkGetFenceStatus; PFN_vkGetImageMemoryRequirements vkGetImageMemoryRequirements; PFN_vkGetQueryPoolResults vkGetQueryPoolResults; + PFN_vkGetSemaphoreCounterValueKHR vkGetSemaphoreCounterValueKHR; PFN_vkMapMemory vkMapMemory; PFN_vkQueueSubmit vkQueueSubmit; PFN_vkResetFences vkResetFences; @@ -275,6 +276,7 @@ struct DeviceDispatch : public InstanceDispatch { PFN_vkUpdateDescriptorSetWithTemplateKHR vkUpdateDescriptorSetWithTemplateKHR; PFN_vkUpdateDescriptorSets vkUpdateDescriptorSets; PFN_vkWaitForFences vkWaitForFences; + PFN_vkWaitSemaphoresKHR vkWaitSemaphoresKHR; }; /// Loads instance agnostic function pointers. @@ -550,7 +552,6 @@ using PipelineLayout = Handle<VkPipelineLayout, VkDevice, DeviceDispatch>; using QueryPool = Handle<VkQueryPool, VkDevice, DeviceDispatch>; using RenderPass = Handle<VkRenderPass, VkDevice, DeviceDispatch>; using Sampler = Handle<VkSampler, VkDevice, DeviceDispatch>; -using Semaphore = Handle<VkSemaphore, VkDevice, DeviceDispatch>; using ShaderModule = Handle<VkShaderModule, VkDevice, DeviceDispatch>; using SurfaceKHR = Handle<VkSurfaceKHR, VkInstance, InstanceDispatch>; @@ -582,7 +583,8 @@ public: /// Construct a queue handle. constexpr Queue(VkQueue queue, const DeviceDispatch& dld) noexcept : queue{queue}, dld{&dld} {} - VkResult Submit(Span<VkSubmitInfo> submit_infos, VkFence fence) const noexcept { + VkResult Submit(Span<VkSubmitInfo> submit_infos, + VkFence fence = VK_NULL_HANDLE) const noexcept { return dld->vkQueueSubmit(queue, submit_infos.size(), submit_infos.data(), fence); } @@ -674,6 +676,44 @@ public: } }; +class Semaphore : public Handle<VkSemaphore, VkDevice, DeviceDispatch> { + using Handle<VkSemaphore, VkDevice, DeviceDispatch>::Handle; + +public: + [[nodiscard]] u64 GetCounter() const { + u64 value; + Check(dld->vkGetSemaphoreCounterValueKHR(owner, handle, &value)); + return value; + } + + /** + * Waits for a timeline semaphore on the host. + * + * @param value Value to wait + * @param timeout Time in nanoseconds to timeout + * @return True on successful wait, false on timeout + */ + bool Wait(u64 value, u64 timeout = std::numeric_limits<u64>::max()) const { + const VkSemaphoreWaitInfoKHR wait_info{ + .sType = VK_STRUCTURE_TYPE_SEMAPHORE_WAIT_INFO_KHR, + .pNext = nullptr, + .flags = 0, + .semaphoreCount = 1, + .pSemaphores = &handle, + .pValues = &value, + }; + const VkResult result = dld->vkWaitSemaphoresKHR(owner, &wait_info, timeout); + switch (result) { + case VK_SUCCESS: + return true; + case VK_TIMEOUT: + return false; + default: + throw Exception(result); + } + } +}; + class Device : public Handle<VkDevice, NoOwner, DeviceDispatch> { using Handle<VkDevice, NoOwner, DeviceDispatch>::Handle; @@ -694,6 +734,8 @@ public: Semaphore CreateSemaphore() const; + Semaphore CreateSemaphore(const VkSemaphoreCreateInfo& ci) const; + Fence CreateFence(const VkFenceCreateInfo& ci) const; DescriptorPool CreateDescriptorPool(const VkDescriptorPoolCreateInfo& ci) const; @@ -721,7 +763,7 @@ public: ShaderModule CreateShaderModule(const VkShaderModuleCreateInfo& ci) const; - Event CreateNewEvent() const; + Event CreateEvent() const; SwapchainKHR CreateSwapchainKHR(const VkSwapchainCreateInfoKHR& ci) const; |