58b0ae84b5
This reworks how host<->device synchronization works on the Vulkan backend. Instead of "protecting" resources with a fence and signalling these as free when the fence is known to be signalled by the host GPU, use timeline semaphores. Vulkan timeline semaphores allow use to work on a subset of D3D12 fences. As far as we are concerned, timeline semaphores are a value set by the host or the device that can be waited by either of them. Taking advantange of this, we can have a monolithically increasing atomic value for each submission to the graphics queue. Instead of protecting resources with a fence, we simply store the current logical tick (the atomic value stored in CPU memory). When we want to know if a resource is free, it can be compared to the current GPU tick. This greatly simplifies resource management code and the free status of resources should have less false negatives. To workaround bugs in validation layers, when these are attached there's a thread waiting for timeline semaphores.
147 lines
4.7 KiB
C++
147 lines
4.7 KiB
C++
// Copyright 2020 yuzu Emulator Project
|
|
// Licensed under GPLv2 or any later version
|
|
// Refer to the license.txt file included.
|
|
|
|
#pragma once
|
|
|
|
#include <condition_variable>
|
|
#include <memory>
|
|
#include <shared_mutex>
|
|
#include <thread>
|
|
|
|
// This header includes both Vulkan and OpenGL headers, this has to be fixed
|
|
// Unfortunately, including OpenGL will include Windows.h that defines macros that can cause issues.
|
|
// Forcefully include glad early and undefine macros
|
|
#include <glad/glad.h>
|
|
#ifdef CreateEvent
|
|
#undef CreateEvent
|
|
#endif
|
|
#ifdef CreateSemaphore
|
|
#undef CreateSemaphore
|
|
#endif
|
|
|
|
#include "common/common_types.h"
|
|
#include "video_core/renderer_opengl/gl_device.h"
|
|
#include "video_core/renderer_opengl/gl_resource_manager.h"
|
|
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
|
|
#include "video_core/renderer_vulkan/vk_device.h"
|
|
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
|
|
#include "video_core/renderer_vulkan/vk_scheduler.h"
|
|
|
|
namespace Core::Frontend {
|
|
class EmuWindow;
|
|
class GraphicsContext;
|
|
} // namespace Core::Frontend
|
|
|
|
namespace Tegra {
|
|
class GPU;
|
|
}
|
|
|
|
namespace Vulkan {
|
|
class VKPipelineCache;
|
|
}
|
|
|
|
namespace VideoCommon::Shader {
|
|
|
|
class AsyncShaders {
|
|
public:
|
|
enum class Backend {
|
|
OpenGL,
|
|
GLASM,
|
|
Vulkan,
|
|
};
|
|
|
|
struct ResultPrograms {
|
|
OpenGL::OGLProgram opengl;
|
|
OpenGL::OGLAssemblyProgram glasm;
|
|
};
|
|
|
|
struct Result {
|
|
u64 uid;
|
|
VAddr cpu_address;
|
|
Backend backend;
|
|
ResultPrograms program;
|
|
std::vector<u64> code;
|
|
std::vector<u64> code_b;
|
|
Tegra::Engines::ShaderType shader_type;
|
|
};
|
|
|
|
explicit AsyncShaders(Core::Frontend::EmuWindow& emu_window);
|
|
~AsyncShaders();
|
|
|
|
/// Start up shader worker threads
|
|
void AllocateWorkers();
|
|
|
|
/// Clear the shader queue and kill all worker threads
|
|
void FreeWorkers();
|
|
|
|
// Force end all threads
|
|
void KillWorkers();
|
|
|
|
/// Check to see if any shaders have actually been compiled
|
|
[[nodiscard]] bool HasCompletedWork() const;
|
|
|
|
/// Deduce if a shader can be build on another thread of MUST be built in sync. We cannot build
|
|
/// every shader async as some shaders are only built and executed once. We try to "guess" which
|
|
/// shader would be used only once
|
|
[[nodiscard]] bool IsShaderAsync(const Tegra::GPU& gpu) const;
|
|
|
|
/// Pulls completed compiled shaders
|
|
[[nodiscard]] std::vector<Result> GetCompletedWork();
|
|
|
|
void QueueOpenGLShader(const OpenGL::Device& device, Tegra::Engines::ShaderType shader_type,
|
|
u64 uid, std::vector<u64> code, std::vector<u64> code_b, u32 main_offset,
|
|
CompilerSettings compiler_settings, const Registry& registry,
|
|
VAddr cpu_addr);
|
|
|
|
void QueueVulkanShader(Vulkan::VKPipelineCache* pp_cache, const Vulkan::VKDevice& device,
|
|
Vulkan::VKScheduler& scheduler,
|
|
Vulkan::VKDescriptorPool& descriptor_pool,
|
|
Vulkan::VKUpdateDescriptorQueue& update_descriptor_queue,
|
|
Vulkan::VKRenderPassCache& renderpass_cache,
|
|
std::vector<VkDescriptorSetLayoutBinding> bindings,
|
|
Vulkan::SPIRVProgram program, Vulkan::GraphicsPipelineCacheKey key);
|
|
|
|
private:
|
|
void ShaderCompilerThread(Core::Frontend::GraphicsContext* context);
|
|
|
|
/// Check our worker queue to see if we have any work queued already
|
|
[[nodiscard]] bool HasWorkQueued() const;
|
|
|
|
struct WorkerParams {
|
|
Backend backend;
|
|
// For OGL
|
|
const OpenGL::Device* device;
|
|
Tegra::Engines::ShaderType shader_type;
|
|
u64 uid;
|
|
std::vector<u64> code;
|
|
std::vector<u64> code_b;
|
|
u32 main_offset;
|
|
CompilerSettings compiler_settings;
|
|
std::optional<Registry> registry;
|
|
VAddr cpu_address;
|
|
|
|
// For Vulkan
|
|
Vulkan::VKPipelineCache* pp_cache;
|
|
const Vulkan::VKDevice* vk_device;
|
|
Vulkan::VKScheduler* scheduler;
|
|
Vulkan::VKDescriptorPool* descriptor_pool;
|
|
Vulkan::VKUpdateDescriptorQueue* update_descriptor_queue;
|
|
Vulkan::VKRenderPassCache* renderpass_cache;
|
|
std::vector<VkDescriptorSetLayoutBinding> bindings;
|
|
Vulkan::SPIRVProgram program;
|
|
Vulkan::GraphicsPipelineCacheKey key;
|
|
};
|
|
|
|
std::condition_variable cv;
|
|
mutable std::mutex queue_mutex;
|
|
mutable std::shared_mutex completed_mutex;
|
|
std::atomic<bool> is_thread_exiting{};
|
|
std::vector<std::unique_ptr<Core::Frontend::GraphicsContext>> context_list;
|
|
std::vector<std::thread> worker_threads;
|
|
std::queue<WorkerParams> pending_queue;
|
|
std::vector<Result> finished_work;
|
|
Core::Frontend::EmuWindow& emu_window;
|
|
};
|
|
|
|
} // namespace VideoCommon::Shader
|