tacosukoneko/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

vk_pipeline_cache,shader_notify: Add shader notifications

Browse source
This commit is contained in:
ReinUsesLisp 2021-06-06 00:11:36 -03:00 committed by ameerj
parent 48aad8dc05
commit cffd4716c5
10 changed files with 126 additions and 95 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

12
src/video_core/renderer_vulkan/vk_compute_pipeline.cpp
View file

@ -14,6 +14,7 @@
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/shader_notify.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@ -24,14 +25,18 @@ using Tegra::Texture::TexturePair;
ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue_,
Common::ThreadWorker* thread_worker, const Shader::Info& info_,
Common::ThreadWorker* thread_worker,
VideoCore::ShaderNotify* shader_notify, const Shader::Info& info_,
vk::ShaderModule spv_module_)
: device{device_}, update_descriptor_queue{update_descriptor_queue_}, info{info_},
spv_module(std::move(spv_module_)) {
if (shader_notify) {
shader_notify->MarkShaderBuilding();
}
std::copy_n(info.constant_buffer_used_sizes.begin(), uniform_buffer_sizes.size(),
uniform_buffer_sizes.begin());
auto func{[this, &descriptor_pool] {
auto func{[this, &descriptor_pool, shader_notify] {
DescriptorLayoutBuilder builder{device.GetLogical()};
builder.Add(info, VK_SHADER_STAGE_COMPUTE_BIT);
@ -66,6 +71,9 @@ ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descript
std::lock_guard lock{build_mutex};
is_built = true;
build_condvar.notify_one();
if (shader_notify) {
shader_notify->MarkShaderComplete();
}
}};
if (thread_worker) {
thread_worker->QueueWork(std::move(func));

7
src/video_core/renderer_vulkan/vk_compute_pipeline.h
View file

@ -18,6 +18,10 @@
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace VideoCore {
class ShaderNotify;
}
namespace Vulkan {
class Device;
@ -27,7 +31,8 @@ class ComputePipeline {
public:
explicit ComputePipeline(const Device& device, DescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue,
Common::ThreadWorker* thread_worker, const Shader::Info& info,
Common::ThreadWorker* thread_worker,
VideoCore::ShaderNotify* shader_notify, const Shader::Info& info,
vk::ShaderModule spv_module);
ComputePipeline& operator=(ComputePipeline&&) noexcept = delete;

30
src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
View file

@ -17,6 +17,7 @@
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/shader_notify.h"
#include "video_core/vulkan_common/vulkan_device.h"
#if defined(_MSC_VER) && defined(NDEBUG)
@ -203,30 +204,30 @@ ConfigureFuncPtr ConfigureFunc(const std::array<vk::ShaderModule, NUM_STAGES>& m
}
} // Anonymous namespace
GraphicsPipeline::GraphicsPipeline(Tegra::Engines::Maxwell3D& maxwell3d_,
Tegra::MemoryManager& gpu_memory_, VKScheduler& scheduler_,
BufferCache& buffer_cache_, TextureCache& texture_cache_,
const Device& device_, DescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue_,
Common::ThreadWorker* worker_thread,
RenderPassCache& render_pass_cache,
const GraphicsPipelineCacheKey& key_,
std::array<vk::ShaderModule, NUM_STAGES> stages,
const std::array<const Shader::Info*, NUM_STAGES>& infos)
GraphicsPipeline::GraphicsPipeline(
Tegra::Engines::Maxwell3D& maxwell3d_, Tegra::MemoryManager& gpu_memory_,
VKScheduler& scheduler_, BufferCache& buffer_cache_, TextureCache& texture_cache_,
VideoCore::ShaderNotify* shader_notify, const Device& device_, DescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue_, Common::ThreadWorker* worker_thread,
RenderPassCache& render_pass_cache, const GraphicsPipelineCacheKey& key_,
std::array<vk::ShaderModule, NUM_STAGES> stages,
const std::array<const Shader::Info*, NUM_STAGES>& infos)
: key{key_}, maxwell3d{maxwell3d_}, gpu_memory{gpu_memory_}, device{device_},
texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, scheduler{scheduler_},
update_descriptor_queue{update_descriptor_queue_}, spv_modules{std::move(stages)} {
std::ranges::transform(infos, stage_infos.begin(),
[](const Shader::Info* info) { return info ? *info : Shader::Info{}; });
if (shader_notify) {
shader_notify->MarkShaderBuilding();
}
for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
const Shader::Info* const info{infos[stage]};
if (!info) {
continue;
}
stage_infos[stage] = *info;
enabled_uniform_buffer_masks[stage] = info->constant_buffer_mask;
std::ranges::copy(info->constant_buffer_used_sizes, uniform_buffer_sizes[stage].begin());
}
auto func{[this, &render_pass_cache, &descriptor_pool] {
auto func{[this, shader_notify, &render_pass_cache, &descriptor_pool] {
DescriptorLayoutBuilder builder{MakeBuilder(device, stage_infos)};
descriptor_set_layout = builder.CreateDescriptorSetLayout();
descriptor_allocator = descriptor_pool.Allocator(*descriptor_set_layout, stage_infos);
@ -242,6 +243,9 @@ GraphicsPipeline::GraphicsPipeline(Tegra::Engines::Maxwell3D& maxwell3d_,
std::lock_guard lock{build_mutex};
is_built = true;
build_condvar.notify_one();
if (shader_notify) {
shader_notify->MarkShaderComplete();
}
}};
if (worker_thread) {
worker_thread->QueueWork(std::move(func));

22
src/video_core/renderer_vulkan/vk_graphics_pipeline.h
View file

@ -20,6 +20,10 @@
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace VideoCore {
class ShaderNotify;
}
namespace Vulkan {
struct GraphicsPipelineCacheKey {
@ -64,16 +68,14 @@ class GraphicsPipeline {
static constexpr size_t NUM_STAGES = Tegra::Engines::Maxwell3D::Regs::MaxShaderStage;
public:
explicit GraphicsPipeline(Tegra::Engines::Maxwell3D& maxwell3d,
Tegra::MemoryManager& gpu_memory, VKScheduler& scheduler,
BufferCache& buffer_cache, TextureCache& texture_cache,
const Device& device, DescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue,
Common::ThreadWorker* worker_thread,
RenderPassCache& render_pass_cache,
const GraphicsPipelineCacheKey& key,
std::array<vk::ShaderModule, NUM_STAGES> stages,
const std::array<const Shader::Info*, NUM_STAGES>& infos);
explicit GraphicsPipeline(
Tegra::Engines::Maxwell3D& maxwell3d, Tegra::MemoryManager& gpu_memory,
VKScheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache,
VideoCore::ShaderNotify* shader_notify, const Device& device,
DescriptorPool& descriptor_pool, VKUpdateDescriptorQueue& update_descriptor_queue,
Common::ThreadWorker* worker_thread, RenderPassCache& render_pass_cache,
const GraphicsPipelineCacheKey& key, std::array<vk::ShaderModule, NUM_STAGES> stages,
const std::array<const Shader::Info*, NUM_STAGES>& infos);
GraphicsPipeline& operator=(GraphicsPipeline&&) noexcept = delete;
GraphicsPipeline(GraphicsPipeline&&) noexcept = delete;

50
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
View file

@ -235,11 +235,11 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxw
VKScheduler& scheduler_, DescriptorPool& descriptor_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
RenderPassCache& render_pass_cache_, BufferCache& buffer_cache_,
TextureCache& texture_cache_)
TextureCache& texture_cache_, VideoCore::ShaderNotify& shader_notify_)
: VideoCommon::ShaderCache{rasterizer_, gpu_memory_, maxwell3d_, kepler_compute_},
device{device_}, scheduler{scheduler_}, descriptor_pool{descriptor_pool_},
update_descriptor_queue{update_descriptor_queue_}, render_pass_cache{render_pass_cache_},
buffer_cache{buffer_cache_}, texture_cache{texture_cache_},
buffer_cache{buffer_cache_}, texture_cache{texture_cache_}, shader_notify{shader_notify_},
use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()},
workers(std::max(std::thread::hardware_concurrency(), 2U) - 1, "yuzu:PipelineBuilder"),
serialization_thread(1, "yuzu:PipelineSerialization") {
@ -307,19 +307,7 @@ GraphicsPipeline* PipelineCache::CurrentGraphicsPipeline() {
return BuiltPipeline(current_pipeline);
}
}
const auto [pair, is_new]{graphics_cache.try_emplace(graphics_key)};
auto& pipeline{pair->second};
if (is_new) {
pipeline = CreateGraphicsPipeline();
}
if (!pipeline) {
return nullptr;
}
if (current_pipeline) {
current_pipeline->AddTransition(pipeline.get());
}
current_pipeline = pipeline.get();
return BuiltPipeline(current_pipeline);
return CurrentGraphicsPipelineSlowPath();
}
ComputePipeline* PipelineCache::CurrentComputePipeline() {
@ -416,6 +404,22 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
workers.WaitForRequests();
}
GraphicsPipeline* PipelineCache::CurrentGraphicsPipelineSlowPath() {
const auto [pair, is_new]{graphics_cache.try_emplace(graphics_key)};
auto& pipeline{pair->second};
if (is_new) {
pipeline = CreateGraphicsPipeline();
}
if (!pipeline) {
return nullptr;
}
if (current_pipeline) {
current_pipeline->AddTransition(pipeline.get());
}
current_pipeline = pipeline.get();
return BuiltPipeline(current_pipeline);
}
GraphicsPipeline* PipelineCache::BuiltPipeline(GraphicsPipeline* pipeline) const noexcept {
if (pipeline->IsBuilt()) {
return pipeline;
@ -484,14 +488,16 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
device.SaveShader(code);
modules[stage_index] = BuildShader(device, code);
if (device.HasDebuggingToolAttached()) {
const std::string name{fmt::format("{:016x}", key.unique_hashes[index])};
const std::string name{fmt::format("Shader {:016x}", key.unique_hashes[index])};
modules[stage_index].SetObjectNameEXT(name.c_str());
}
}
Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr};
return std::make_unique<GraphicsPipeline>(
maxwell3d, gpu_memory, scheduler, buffer_cache, texture_cache, device, descriptor_pool,
update_descriptor_queue, thread_worker, render_pass_cache, key, std::move(modules), infos);
VideoCore::ShaderNotify* const notify{build_in_parallel ? &shader_notify : nullptr};
return std::make_unique<GraphicsPipeline>(maxwell3d, gpu_memory, scheduler, buffer_cache,
texture_cache, notify, device, descriptor_pool,
update_descriptor_queue, thread_worker,
render_pass_cache, key, std::move(modules), infos);
} catch (const Shader::Exception& exception) {
LOG_ERROR(Render_Vulkan, "{}", exception.what());
@ -550,12 +556,14 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
device.SaveShader(code);
vk::ShaderModule spv_module{BuildShader(device, code)};
if (device.HasDebuggingToolAttached()) {
const auto name{fmt::format("{:016x}", key.unique_hash)};
const auto name{fmt::format("Shader {:016x}", key.unique_hash)};
spv_module.SetObjectNameEXT(name.c_str());
}
Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr};
VideoCore::ShaderNotify* const notify{build_in_parallel ? &shader_notify : nullptr};
return std::make_unique<ComputePipeline>(device, descriptor_pool, update_descriptor_queue,
thread_worker, program.info, std::move(spv_module));
thread_worker, notify, program.info,
std::move(spv_module));
} catch (const Shader::Exception& exception) {
LOG_ERROR(Render_Vulkan, "{}", exception.what());

9
src/video_core/renderer_vulkan/vk_pipeline_cache.h
View file

@ -38,6 +38,10 @@ namespace Shader::IR {
struct Program;
}
namespace VideoCore {
class ShaderNotify;
}
namespace Vulkan {
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
@ -104,7 +108,7 @@ public:
VKScheduler& scheduler, DescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue,
RenderPassCache& render_pass_cache, BufferCache& buffer_cache,
TextureCache& texture_cache);
TextureCache& texture_cache, VideoCore::ShaderNotify& shader_notify_);
~PipelineCache();
[[nodiscard]] GraphicsPipeline* CurrentGraphicsPipeline();
@ -115,6 +119,8 @@ public:
const VideoCore::DiskResourceLoadCallback& callback);
private:
[[nodiscard]] GraphicsPipeline* CurrentGraphicsPipelineSlowPath();
[[nodiscard]] GraphicsPipeline* BuiltPipeline(GraphicsPipeline* pipeline) const noexcept;
std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline();
@ -138,6 +144,7 @@ private:
RenderPassCache& render_pass_cache;
BufferCache& buffer_cache;
TextureCache& texture_cache;
VideoCore::ShaderNotify& shader_notify;
GraphicsPipelineCacheKey graphics_key{};
GraphicsPipeline* current_pipeline{};

2
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View file

@ -140,7 +140,7 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
buffer_cache(*this, maxwell3d, kepler_compute, gpu_memory, cpu_memory_, buffer_cache_runtime),
pipeline_cache(*this, maxwell3d, kepler_compute, gpu_memory, device, scheduler,
descriptor_pool, update_descriptor_queue, render_pass_cache, buffer_cache,
texture_cache),
texture_cache, gpu.ShaderNotify()),
query_cache{*this, maxwell3d, gpu_memory, device, scheduler}, accelerate_dma{ buffer_cache },
fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler),
wfi_event(device.GetLogical().CreateEvent()) {

49
src/video_core/shader_notify.cpp
View file

@ -2,42 +2,35 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <mutex>
#include <atomic>
#include <chrono>
#include <optional>
#include "video_core/shader_notify.h"
using namespace std::chrono_literals;
namespace VideoCore {
namespace {
constexpr auto UPDATE_TICK = 32ms;
}
ShaderNotify::ShaderNotify() = default;
ShaderNotify::~ShaderNotify() = default;
const auto TIME_TO_STOP_REPORTING = 2s;
std::size_t ShaderNotify::GetShadersBuilding() {
const auto now = std::chrono::high_resolution_clock::now();
const auto diff = now - last_update;
if (diff > UPDATE_TICK) {
std::shared_lock lock(mutex);
last_updated_count = accurate_count;
int ShaderNotify::ShadersBuilding() noexcept {
const int now_complete = num_complete.load(std::memory_order::relaxed);
const int now_building = num_building.load(std::memory_order::relaxed);
if (now_complete == now_building) {
const auto now = std::chrono::high_resolution_clock::now();
if (completed && num_complete == num_when_completed) {
if (now - complete_time > TIME_TO_STOP_REPORTING) {
report_base = now_complete;
completed = false;
}
} else {
completed = true;
num_when_completed = num_complete;
complete_time = now;
}
}
return last_updated_count;
}
std::size_t ShaderNotify::GetShadersBuildingAccurate() {
std::shared_lock lock{mutex};
return accurate_count;
}
void ShaderNotify::MarkShaderComplete() {
std::unique_lock lock{mutex};
accurate_count--;
}
void ShaderNotify::MarkSharderBuilding() {
std::unique_lock lock{mutex};
accurate_count++;
return now_building - report_base;
}
} // namespace VideoCore

28
src/video_core/shader_notify.h
View file

@ -4,26 +4,30 @@
#pragma once
#include <atomic>
#include <chrono>
#include <shared_mutex>
#include "common/common_types.h"
#include <optional>
namespace VideoCore {
class ShaderNotify {
public:
ShaderNotify();
~ShaderNotify();
[[nodiscard]] int ShadersBuilding() noexcept;
std::size_t GetShadersBuilding();
std::size_t GetShadersBuildingAccurate();
void MarkShaderComplete() noexcept {
++num_complete;
}
void MarkShaderComplete();
void MarkSharderBuilding();
void MarkShaderBuilding() noexcept {
++num_building;
}
private:
std::size_t last_updated_count{};
std::size_t accurate_count{};
std::shared_mutex mutex;
std::chrono::high_resolution_clock::time_point last_update{};
std::atomic_int num_building{};
std::atomic_int num_complete{};
int report_base{};
bool completed{};
int num_when_completed{};
std::chrono::high_resolution_clock::time_point complete_time;
};
} // namespace VideoCore

12
src/yuzu/main.cpp
View file

@ -2900,13 +2900,13 @@ void GMainWindow::UpdateStatusBar() {
return;
}
auto results = Core::System::GetInstance().GetAndResetPerfStats();
auto& shader_notify = Core::System::GetInstance().GPU().ShaderNotify();
const auto shaders_building = shader_notify.GetShadersBuilding();
auto& system = Core::System::GetInstance();
auto results = system.GetAndResetPerfStats();
auto& shader_notify = system.GPU().ShaderNotify();
const int shaders_building = shader_notify.ShadersBuilding();
if (shaders_building != 0) {
shader_building_label->setText(
tr("Building: %n shader(s)", "", static_cast<int>(shaders_building)));
if (shaders_building > 0) {
shader_building_label->setText(tr("Building: %n shader(s)", "", shaders_building));
shader_building_label->setVisible(true);
} else {
shader_building_label->setVisible(false);