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

renderer_vulkan: separate guest and host compute descriptor queues

Browse source
This commit is contained in:
Liam 2023-05-10 13:35:25 -04:00
parent b7f60e9123
commit 67fd1df762
16 changed files with 81 additions and 75 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
src/video_core/renderer_vulkan/pipeline_helper.h
View file

@ -176,7 +176,7 @@ public:
}; };
inline void PushImageDescriptors(TextureCache& texture_cache, inline void PushImageDescriptors(TextureCache& texture_cache,
UpdateDescriptorQueue& update_descriptor_queue, GuestDescriptorQueue& guest_descriptor_queue,
const Shader::Info& info, RescalingPushConstant& rescaling, const Shader::Info& info, RescalingPushConstant& rescaling,
const VkSampler*& samplers, const VkSampler*& samplers,
const VideoCommon::ImageViewInOut*& views) { const VideoCommon::ImageViewInOut*& views) {
@ -190,7 +190,7 @@ inline void PushImageDescriptors(TextureCache& texture_cache,
const VkSampler sampler{*(samplers++)}; const VkSampler sampler{*(samplers++)};
ImageView& image_view{texture_cache.GetImageView(image_view_id)}; ImageView& image_view{texture_cache.GetImageView(image_view_id)};
const VkImageView vk_image_view{image_view.Handle(desc.type)}; const VkImageView vk_image_view{image_view.Handle(desc.type)};
update_descriptor_queue.AddSampledImage(vk_image_view, sampler); guest_descriptor_queue.AddSampledImage(vk_image_view, sampler);
rescaling.PushTexture(texture_cache.IsRescaling(image_view)); rescaling.PushTexture(texture_cache.IsRescaling(image_view));
} }
} }
@ -201,7 +201,7 @@ inline void PushImageDescriptors(TextureCache& texture_cache,
texture_cache.MarkModification(image_view.image_id); texture_cache.MarkModification(image_view.image_id);
} }
const VkImageView vk_image_view{image_view.StorageView(desc.type, desc.format)}; const VkImageView vk_image_view{image_view.StorageView(desc.type, desc.format)};
update_descriptor_queue.AddImage(vk_image_view); guest_descriptor_queue.AddImage(vk_image_view);
rescaling.PushImage(texture_cache.IsRescaling(image_view)); rescaling.PushImage(texture_cache.IsRescaling(image_view));
} }
} }

10
src/video_core/renderer_vulkan/vk_buffer_cache.cpp
View file

@ -298,12 +298,14 @@ private:
BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_allocator_, BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_allocator_,
Scheduler& scheduler_, StagingBufferPool& staging_pool_, Scheduler& scheduler_, StagingBufferPool& staging_pool_,
UpdateDescriptorQueue& update_descriptor_queue_, GuestDescriptorQueue& guest_descriptor_queue_,
ComputePassDescriptorQueue& compute_pass_descriptor_queue,
DescriptorPool& descriptor_pool) DescriptorPool& descriptor_pool)
: device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_}, : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_},
staging_pool{staging_pool_}, update_descriptor_queue{update_descriptor_queue_}, staging_pool{staging_pool_}, guest_descriptor_queue{guest_descriptor_queue_},
uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue), uint8_pass(device, scheduler, descriptor_pool, staging_pool, compute_pass_descriptor_queue),
quad_index_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue) { quad_index_pass(device, scheduler, descriptor_pool, staging_pool,
compute_pass_descriptor_queue) {
quad_array_index_buffer = std::make_shared<QuadArrayIndexBuffer>(device_, memory_allocator_, quad_array_index_buffer = std::make_shared<QuadArrayIndexBuffer>(device_, memory_allocator_,
scheduler_, staging_pool_); scheduler_, staging_pool_);
quad_strip_index_buffer = std::make_shared<QuadStripIndexBuffer>(device_, memory_allocator_, quad_strip_index_buffer = std::make_shared<QuadStripIndexBuffer>(device_, memory_allocator_,

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

@ -63,7 +63,8 @@ class BufferCacheRuntime {
public: public:
explicit BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_manager_, explicit BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_manager_,
Scheduler& scheduler_, StagingBufferPool& staging_pool_, Scheduler& scheduler_, StagingBufferPool& staging_pool_,
UpdateDescriptorQueue& update_descriptor_queue_, GuestDescriptorQueue& guest_descriptor_queue,
ComputePassDescriptorQueue& compute_pass_descriptor_queue,
DescriptorPool& descriptor_pool); DescriptorPool& descriptor_pool);
void Finish(); void Finish();
@ -116,12 +117,12 @@ public:
void BindTextureBuffer(Buffer& buffer, u32 offset, u32 size, void BindTextureBuffer(Buffer& buffer, u32 offset, u32 size,
VideoCore::Surface::PixelFormat format) { VideoCore::Surface::PixelFormat format) {
update_descriptor_queue.AddTexelBuffer(buffer.View(offset, size, format)); guest_descriptor_queue.AddTexelBuffer(buffer.View(offset, size, format));
} }
private: private:
void BindBuffer(VkBuffer buffer, u32 offset, u32 size) { void BindBuffer(VkBuffer buffer, u32 offset, u32 size) {
update_descriptor_queue.AddBuffer(buffer, offset, size); guest_descriptor_queue.AddBuffer(buffer, offset, size);
} }
void ReserveNullBuffer(); void ReserveNullBuffer();
@ -130,7 +131,7 @@ private:
MemoryAllocator& memory_allocator; MemoryAllocator& memory_allocator;
Scheduler& scheduler; Scheduler& scheduler;
StagingBufferPool& staging_pool; StagingBufferPool& staging_pool;
UpdateDescriptorQueue& update_descriptor_queue; GuestDescriptorQueue& guest_descriptor_queue;
std::shared_ptr<QuadArrayIndexBuffer> quad_array_index_buffer; std::shared_ptr<QuadArrayIndexBuffer> quad_array_index_buffer;
std::shared_ptr<QuadStripIndexBuffer> quad_strip_index_buffer; std::shared_ptr<QuadStripIndexBuffer> quad_strip_index_buffer;

37
src/video_core/renderer_vulkan/vk_compute_pass.cpp
View file

@ -200,12 +200,12 @@ ComputePass::~ComputePass() = default;
Uint8Pass::Uint8Pass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool, Uint8Pass::Uint8Pass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool,
StagingBufferPool& staging_buffer_pool_, StagingBufferPool& staging_buffer_pool_,
UpdateDescriptorQueue& update_descriptor_queue_) ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
: ComputePass(device_, descriptor_pool, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS, : ComputePass(device_, descriptor_pool, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS,
INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, {}, INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, {},
VULKAN_UINT8_COMP_SPV), VULKAN_UINT8_COMP_SPV),
scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
update_descriptor_queue{update_descriptor_queue_} {} compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
Uint8Pass::~Uint8Pass() = default; Uint8Pass::~Uint8Pass() = default;
@ -214,10 +214,10 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer
const u32 staging_size = static_cast<u32>(num_vertices * sizeof(u16)); const u32 staging_size = static_cast<u32>(num_vertices * sizeof(u16));
const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal); const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
update_descriptor_queue.Acquire(); compute_pass_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices); compute_pass_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices);
update_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size); compute_pass_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size);
const void* const descriptor_data{update_descriptor_queue.UpdateData()}; const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
scheduler.RequestOutsideRenderPassOperationContext(); scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([this, descriptor_data, num_vertices](vk::CommandBuffer cmdbuf) { scheduler.Record([this, descriptor_data, num_vertices](vk::CommandBuffer cmdbuf) {
@ -242,12 +242,12 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer
QuadIndexedPass::QuadIndexedPass(const Device& device_, Scheduler& scheduler_, QuadIndexedPass::QuadIndexedPass(const Device& device_, Scheduler& scheduler_,
DescriptorPool& descriptor_pool_, DescriptorPool& descriptor_pool_,
StagingBufferPool& staging_buffer_pool_, StagingBufferPool& staging_buffer_pool_,
UpdateDescriptorQueue& update_descriptor_queue_) ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
: ComputePass(device_, descriptor_pool_, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS, : ComputePass(device_, descriptor_pool_, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS,
INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO,
COMPUTE_PUSH_CONSTANT_RANGE<sizeof(u32) * 3>, VULKAN_QUAD_INDEXED_COMP_SPV), COMPUTE_PUSH_CONSTANT_RANGE<sizeof(u32) * 3>, VULKAN_QUAD_INDEXED_COMP_SPV),
scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
update_descriptor_queue{update_descriptor_queue_} {} compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
QuadIndexedPass::~QuadIndexedPass() = default; QuadIndexedPass::~QuadIndexedPass() = default;
@ -272,10 +272,10 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
const std::size_t staging_size = num_tri_vertices * sizeof(u32); const std::size_t staging_size = num_tri_vertices * sizeof(u32);
const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal); const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
update_descriptor_queue.Acquire(); compute_pass_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size); compute_pass_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size);
update_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size); compute_pass_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size);
const void* const descriptor_data{update_descriptor_queue.UpdateData()}; const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
scheduler.RequestOutsideRenderPassOperationContext(); scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([this, descriptor_data, num_tri_vertices, base_vertex, index_shift, scheduler.Record([this, descriptor_data, num_tri_vertices, base_vertex, index_shift,
@ -304,13 +304,14 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
ASTCDecoderPass::ASTCDecoderPass(const Device& device_, Scheduler& scheduler_, ASTCDecoderPass::ASTCDecoderPass(const Device& device_, Scheduler& scheduler_,
DescriptorPool& descriptor_pool_, DescriptorPool& descriptor_pool_,
StagingBufferPool& staging_buffer_pool_, StagingBufferPool& staging_buffer_pool_,
UpdateDescriptorQueue& update_descriptor_queue_, ComputePassDescriptorQueue& compute_pass_descriptor_queue_,
MemoryAllocator& memory_allocator_) MemoryAllocator& memory_allocator_)
: ComputePass(device_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, : ComputePass(device_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS,
ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO,
COMPUTE_PUSH_CONSTANT_RANGE<sizeof(AstcPushConstants)>, ASTC_DECODER_COMP_SPV), COMPUTE_PUSH_CONSTANT_RANGE<sizeof(AstcPushConstants)>, ASTC_DECODER_COMP_SPV),
scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
update_descriptor_queue{update_descriptor_queue_}, memory_allocator{memory_allocator_} {} compute_pass_descriptor_queue{compute_pass_descriptor_queue_}, memory_allocator{
memory_allocator_} {}
ASTCDecoderPass::~ASTCDecoderPass() = default; ASTCDecoderPass::~ASTCDecoderPass() = default;
@ -358,11 +359,11 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U); const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U);
const u32 num_dispatches_z = image.info.resources.layers; const u32 num_dispatches_z = image.info.resources.layers;
update_descriptor_queue.Acquire(); compute_pass_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(map.buffer, input_offset, compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset,
image.guest_size_bytes - swizzle.buffer_offset); image.guest_size_bytes - swizzle.buffer_offset);
update_descriptor_queue.AddImage(image.StorageImageView(swizzle.level)); compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
const void* const descriptor_data{update_descriptor_queue.UpdateData()}; const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
// To unswizzle the ASTC data // To unswizzle the ASTC data
const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info); const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);

14
src/video_core/renderer_vulkan/vk_compute_pass.h
View file

@ -9,6 +9,7 @@
#include "common/common_types.h" #include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h" #include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h" #include "video_core/vulkan_common/vulkan_wrapper.h"
@ -21,7 +22,6 @@ namespace Vulkan {
class Device; class Device;
class StagingBufferPool; class StagingBufferPool;
class Scheduler; class Scheduler;
class UpdateDescriptorQueue;
class Image; class Image;
struct StagingBufferRef; struct StagingBufferRef;
@ -50,7 +50,7 @@ class Uint8Pass final : public ComputePass {
public: public:
explicit Uint8Pass(const Device& device_, Scheduler& scheduler_, explicit Uint8Pass(const Device& device_, Scheduler& scheduler_,
DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_,
UpdateDescriptorQueue& update_descriptor_queue_); ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
~Uint8Pass(); ~Uint8Pass();
/// Assemble uint8 indices into an uint16 index buffer /// Assemble uint8 indices into an uint16 index buffer
@ -61,7 +61,7 @@ public:
private: private:
Scheduler& scheduler; Scheduler& scheduler;
StagingBufferPool& staging_buffer_pool; StagingBufferPool& staging_buffer_pool;
UpdateDescriptorQueue& update_descriptor_queue; ComputePassDescriptorQueue& compute_pass_descriptor_queue;
}; };
class QuadIndexedPass final : public ComputePass { class QuadIndexedPass final : public ComputePass {
@ -69,7 +69,7 @@ public:
explicit QuadIndexedPass(const Device& device_, Scheduler& scheduler_, explicit QuadIndexedPass(const Device& device_, Scheduler& scheduler_,
DescriptorPool& descriptor_pool_, DescriptorPool& descriptor_pool_,
StagingBufferPool& staging_buffer_pool_, StagingBufferPool& staging_buffer_pool_,
UpdateDescriptorQueue& update_descriptor_queue_); ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
~QuadIndexedPass(); ~QuadIndexedPass();
std::pair<VkBuffer, VkDeviceSize> Assemble( std::pair<VkBuffer, VkDeviceSize> Assemble(
@ -79,7 +79,7 @@ public:
private: private:
Scheduler& scheduler; Scheduler& scheduler;
StagingBufferPool& staging_buffer_pool; StagingBufferPool& staging_buffer_pool;
UpdateDescriptorQueue& update_descriptor_queue; ComputePassDescriptorQueue& compute_pass_descriptor_queue;
}; };
class ASTCDecoderPass final : public ComputePass { class ASTCDecoderPass final : public ComputePass {
@ -87,7 +87,7 @@ public:
explicit ASTCDecoderPass(const Device& device_, Scheduler& scheduler_, explicit ASTCDecoderPass(const Device& device_, Scheduler& scheduler_,
DescriptorPool& descriptor_pool_, DescriptorPool& descriptor_pool_,
StagingBufferPool& staging_buffer_pool_, StagingBufferPool& staging_buffer_pool_,
UpdateDescriptorQueue& update_descriptor_queue_, ComputePassDescriptorQueue& compute_pass_descriptor_queue_,
MemoryAllocator& memory_allocator_); MemoryAllocator& memory_allocator_);
~ASTCDecoderPass(); ~ASTCDecoderPass();
@ -97,7 +97,7 @@ public:
private: private:
Scheduler& scheduler; Scheduler& scheduler;
StagingBufferPool& staging_buffer_pool; StagingBufferPool& staging_buffer_pool;
UpdateDescriptorQueue& update_descriptor_queue; ComputePassDescriptorQueue& compute_pass_descriptor_queue;
MemoryAllocator& memory_allocator; MemoryAllocator& memory_allocator;
}; };

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

@ -26,13 +26,13 @@ using Tegra::Texture::TexturePair;
ComputePipeline::ComputePipeline(const Device& device_, vk::PipelineCache& pipeline_cache_, ComputePipeline::ComputePipeline(const Device& device_, vk::PipelineCache& pipeline_cache_,
DescriptorPool& descriptor_pool, DescriptorPool& descriptor_pool,
UpdateDescriptorQueue& update_descriptor_queue_, GuestDescriptorQueue& guest_descriptor_queue_,
Common::ThreadWorker* thread_worker, Common::ThreadWorker* thread_worker,
PipelineStatistics* pipeline_statistics, PipelineStatistics* pipeline_statistics,
VideoCore::ShaderNotify* shader_notify, const Shader::Info& info_, VideoCore::ShaderNotify* shader_notify, const Shader::Info& info_,
vk::ShaderModule spv_module_) vk::ShaderModule spv_module_)
: device{device_}, pipeline_cache(pipeline_cache_), : device{device_},
update_descriptor_queue{update_descriptor_queue_}, info{info_}, pipeline_cache(pipeline_cache_), guest_descriptor_queue{guest_descriptor_queue_}, info{info_},
spv_module(std::move(spv_module_)) { spv_module(std::move(spv_module_)) {
if (shader_notify) { if (shader_notify) {
shader_notify->MarkShaderBuilding(); shader_notify->MarkShaderBuilding();
@ -99,7 +99,7 @@ ComputePipeline::ComputePipeline(const Device& device_, vk::PipelineCache& pipel
void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
Tegra::MemoryManager& gpu_memory, Scheduler& scheduler, Tegra::MemoryManager& gpu_memory, Scheduler& scheduler,
BufferCache& buffer_cache, TextureCache& texture_cache) { BufferCache& buffer_cache, TextureCache& texture_cache) {
update_descriptor_queue.Acquire(); guest_descriptor_queue.Acquire();
buffer_cache.SetComputeUniformBufferState(info.constant_buffer_mask, &uniform_buffer_sizes); buffer_cache.SetComputeUniformBufferState(info.constant_buffer_mask, &uniform_buffer_sizes);
buffer_cache.UnbindComputeStorageBuffers(); buffer_cache.UnbindComputeStorageBuffers();
@ -194,7 +194,7 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
RescalingPushConstant rescaling; RescalingPushConstant rescaling;
const VkSampler* samplers_it{samplers.data()}; const VkSampler* samplers_it{samplers.data()};
const VideoCommon::ImageViewInOut* views_it{views.data()}; const VideoCommon::ImageViewInOut* views_it{views.data()};
PushImageDescriptors(texture_cache, update_descriptor_queue, info, rescaling, samplers_it, PushImageDescriptors(texture_cache, guest_descriptor_queue, info, rescaling, samplers_it,
views_it); views_it);
if (!is_built.load(std::memory_order::relaxed)) { if (!is_built.load(std::memory_order::relaxed)) {
@ -204,7 +204,7 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
build_condvar.wait(lock, [this] { return is_built.load(std::memory_order::relaxed); }); build_condvar.wait(lock, [this] { return is_built.load(std::memory_order::relaxed); });
}); });
} }
const void* const descriptor_data{update_descriptor_queue.UpdateData()}; const void* const descriptor_data{guest_descriptor_queue.UpdateData()};
const bool is_rescaling = !info.texture_descriptors.empty() || !info.image_descriptors.empty(); const bool is_rescaling = !info.texture_descriptors.empty() || !info.image_descriptors.empty();
scheduler.Record([this, descriptor_data, is_rescaling, scheduler.Record([this, descriptor_data, is_rescaling,
rescaling_data = rescaling.Data()](vk::CommandBuffer cmdbuf) { rescaling_data = rescaling.Data()](vk::CommandBuffer cmdbuf) {

4
src/video_core/renderer_vulkan/vk_compute_pipeline.h
View file

@ -30,7 +30,7 @@ class ComputePipeline {
public: public:
explicit ComputePipeline(const Device& device, vk::PipelineCache& pipeline_cache, explicit ComputePipeline(const Device& device, vk::PipelineCache& pipeline_cache,
DescriptorPool& descriptor_pool, DescriptorPool& descriptor_pool,
UpdateDescriptorQueue& update_descriptor_queue, GuestDescriptorQueue& guest_descriptor_queue,
Common::ThreadWorker* thread_worker, Common::ThreadWorker* thread_worker,
PipelineStatistics* pipeline_statistics, PipelineStatistics* pipeline_statistics,
VideoCore::ShaderNotify* shader_notify, const Shader::Info& info, VideoCore::ShaderNotify* shader_notify, const Shader::Info& info,
@ -48,7 +48,7 @@ public:
private: private:
const Device& device; const Device& device;
vk::PipelineCache& pipeline_cache; vk::PipelineCache& pipeline_cache;
UpdateDescriptorQueue& update_descriptor_queue; GuestDescriptorQueue& guest_descriptor_queue;
Shader::Info info; Shader::Info info;
VideoCommon::ComputeUniformBufferSizes uniform_buffer_sizes{}; VideoCommon::ComputeUniformBufferSizes uniform_buffer_sizes{};

10
src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
View file

@ -236,13 +236,13 @@ GraphicsPipeline::GraphicsPipeline(
Scheduler& scheduler_, BufferCache& buffer_cache_, TextureCache& texture_cache_, Scheduler& scheduler_, BufferCache& buffer_cache_, TextureCache& texture_cache_,
vk::PipelineCache& pipeline_cache_, VideoCore::ShaderNotify* shader_notify, vk::PipelineCache& pipeline_cache_, VideoCore::ShaderNotify* shader_notify,
const Device& device_, DescriptorPool& descriptor_pool, const Device& device_, DescriptorPool& descriptor_pool,
UpdateDescriptorQueue& update_descriptor_queue_, Common::ThreadWorker* worker_thread, GuestDescriptorQueue& guest_descriptor_queue_, Common::ThreadWorker* worker_thread,
PipelineStatistics* pipeline_statistics, RenderPassCache& render_pass_cache, PipelineStatistics* pipeline_statistics, RenderPassCache& render_pass_cache,
const GraphicsPipelineCacheKey& key_, std::array<vk::ShaderModule, NUM_STAGES> stages, const GraphicsPipelineCacheKey& key_, std::array<vk::ShaderModule, NUM_STAGES> stages,
const std::array<const Shader::Info*, NUM_STAGES>& infos) const std::array<const Shader::Info*, NUM_STAGES>& infos)
: key{key_}, device{device_}, texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, : key{key_}, device{device_}, texture_cache{texture_cache_}, buffer_cache{buffer_cache_},
pipeline_cache(pipeline_cache_), scheduler{scheduler_}, pipeline_cache(pipeline_cache_), scheduler{scheduler_},
update_descriptor_queue{update_descriptor_queue_}, spv_modules{std::move(stages)} { guest_descriptor_queue{guest_descriptor_queue_}, spv_modules{std::move(stages)} {
if (shader_notify) { if (shader_notify) {
shader_notify->MarkShaderBuilding(); shader_notify->MarkShaderBuilding();
} }
@ -449,7 +449,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) {
buffer_cache.UpdateGraphicsBuffers(is_indexed); buffer_cache.UpdateGraphicsBuffers(is_indexed);
buffer_cache.BindHostGeometryBuffers(is_indexed); buffer_cache.BindHostGeometryBuffers(is_indexed);
update_descriptor_queue.Acquire(); guest_descriptor_queue.Acquire();
RescalingPushConstant rescaling; RescalingPushConstant rescaling;
RenderAreaPushConstant render_area; RenderAreaPushConstant render_area;
@ -457,7 +457,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) {
const VideoCommon::ImageViewInOut* views_it{views.data()}; const VideoCommon::ImageViewInOut* views_it{views.data()};
const auto prepare_stage{[&](size_t stage) LAMBDA_FORCEINLINE { const auto prepare_stage{[&](size_t stage) LAMBDA_FORCEINLINE {
buffer_cache.BindHostStageBuffers(stage); buffer_cache.BindHostStageBuffers(stage);
PushImageDescriptors(texture_cache, update_descriptor_queue, stage_infos[stage], rescaling, PushImageDescriptors(texture_cache, guest_descriptor_queue, stage_infos[stage], rescaling,
samplers_it, views_it); samplers_it, views_it);
const auto& info{stage_infos[0]}; const auto& info{stage_infos[0]};
if (info.uses_render_area) { if (info.uses_render_area) {
@ -499,7 +499,7 @@ void GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling,
const bool is_rescaling{texture_cache.IsRescaling()}; const bool is_rescaling{texture_cache.IsRescaling()};
const bool update_rescaling{scheduler.UpdateRescaling(is_rescaling)}; const bool update_rescaling{scheduler.UpdateRescaling(is_rescaling)};
const bool bind_pipeline{scheduler.UpdateGraphicsPipeline(this)}; const bool bind_pipeline{scheduler.UpdateGraphicsPipeline(this)};
const void* const descriptor_data{update_descriptor_queue.UpdateData()}; const void* const descriptor_data{guest_descriptor_queue.UpdateData()};
scheduler.Record([this, descriptor_data, bind_pipeline, rescaling_data = rescaling.Data(), scheduler.Record([this, descriptor_data, bind_pipeline, rescaling_data = rescaling.Data(),
is_rescaling, update_rescaling, is_rescaling, update_rescaling,
uses_render_area = render_area.uses_render_area, uses_render_area = render_area.uses_render_area,

5
src/video_core/renderer_vulkan/vk_graphics_pipeline.h
View file

@ -64,7 +64,6 @@ class RenderPassCache;
class RescalingPushConstant; class RescalingPushConstant;
class RenderAreaPushConstant; class RenderAreaPushConstant;
class Scheduler; class Scheduler;
class UpdateDescriptorQueue;
class GraphicsPipeline { class GraphicsPipeline {
static constexpr size_t NUM_STAGES = Tegra::Engines::Maxwell3D::Regs::MaxShaderStage; static constexpr size_t NUM_STAGES = Tegra::Engines::Maxwell3D::Regs::MaxShaderStage;
@ -74,7 +73,7 @@ public:
Scheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache, Scheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache,
vk::PipelineCache& pipeline_cache, VideoCore::ShaderNotify* shader_notify, vk::PipelineCache& pipeline_cache, VideoCore::ShaderNotify* shader_notify,
const Device& device, DescriptorPool& descriptor_pool, const Device& device, DescriptorPool& descriptor_pool,
UpdateDescriptorQueue& update_descriptor_queue, Common::ThreadWorker* worker_thread, GuestDescriptorQueue& guest_descriptor_queue, Common::ThreadWorker* worker_thread,
PipelineStatistics* pipeline_statistics, RenderPassCache& render_pass_cache, PipelineStatistics* pipeline_statistics, RenderPassCache& render_pass_cache,
const GraphicsPipelineCacheKey& key, std::array<vk::ShaderModule, NUM_STAGES> stages, const GraphicsPipelineCacheKey& key, std::array<vk::ShaderModule, NUM_STAGES> stages,
const std::array<const Shader::Info*, NUM_STAGES>& infos); const std::array<const Shader::Info*, NUM_STAGES>& infos);
@ -133,7 +132,7 @@ private:
BufferCache& buffer_cache; BufferCache& buffer_cache;
vk::PipelineCache& pipeline_cache; vk::PipelineCache& pipeline_cache;
Scheduler& scheduler; Scheduler& scheduler;
UpdateDescriptorQueue& update_descriptor_queue; GuestDescriptorQueue& guest_descriptor_queue;
void (*configure_func)(GraphicsPipeline*, bool){}; void (*configure_func)(GraphicsPipeline*, bool){};

8
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
View file

@ -277,11 +277,11 @@ bool GraphicsPipelineCacheKey::operator==(const GraphicsPipelineCacheKey& rhs) c
PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device_, PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device_,
Scheduler& scheduler_, DescriptorPool& descriptor_pool_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
UpdateDescriptorQueue& update_descriptor_queue_, GuestDescriptorQueue& guest_descriptor_queue_,
RenderPassCache& render_pass_cache_, BufferCache& buffer_cache_, RenderPassCache& render_pass_cache_, BufferCache& buffer_cache_,
TextureCache& texture_cache_, VideoCore::ShaderNotify& shader_notify_) TextureCache& texture_cache_, VideoCore::ShaderNotify& shader_notify_)
: VideoCommon::ShaderCache{rasterizer_}, device{device_}, scheduler{scheduler_}, : VideoCommon::ShaderCache{rasterizer_}, device{device_}, scheduler{scheduler_},
descriptor_pool{descriptor_pool_}, update_descriptor_queue{update_descriptor_queue_}, descriptor_pool{descriptor_pool_}, guest_descriptor_queue{guest_descriptor_queue_},
render_pass_cache{render_pass_cache_}, buffer_cache{buffer_cache_}, render_pass_cache{render_pass_cache_}, buffer_cache{buffer_cache_},
texture_cache{texture_cache_}, shader_notify{shader_notify_}, texture_cache{texture_cache_}, shader_notify{shader_notify_},
use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()}, use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()},
@ -643,7 +643,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr}; Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr};
return std::make_unique<GraphicsPipeline>( return std::make_unique<GraphicsPipeline>(
scheduler, buffer_cache, texture_cache, vulkan_pipeline_cache, &shader_notify, device, scheduler, buffer_cache, texture_cache, vulkan_pipeline_cache, &shader_notify, device,
descriptor_pool, update_descriptor_queue, thread_worker, statistics, render_pass_cache, key, descriptor_pool, guest_descriptor_queue, thread_worker, statistics, render_pass_cache, key,
std::move(modules), infos); std::move(modules), infos);
} catch (const Shader::Exception& exception) { } catch (const Shader::Exception& exception) {
@ -722,7 +722,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
} }
Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr}; Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr};
return std::make_unique<ComputePipeline>(device, vulkan_pipeline_cache, descriptor_pool, return std::make_unique<ComputePipeline>(device, vulkan_pipeline_cache, descriptor_pool,
update_descriptor_queue, thread_worker, statistics, guest_descriptor_queue, thread_worker, statistics,
&shader_notify, program.info, std::move(spv_module)); &shader_notify, program.info, std::move(spv_module));
} catch (const Shader::Exception& exception) { } catch (const Shader::Exception& exception) {

5
src/video_core/renderer_vulkan/vk_pipeline_cache.h
View file

@ -82,7 +82,6 @@ class PipelineStatistics;
class RasterizerVulkan; class RasterizerVulkan;
class RenderPassCache; class RenderPassCache;
class Scheduler; class Scheduler;
class UpdateDescriptorQueue;
using VideoCommon::ShaderInfo; using VideoCommon::ShaderInfo;
@ -102,7 +101,7 @@ class PipelineCache : public VideoCommon::ShaderCache {
public: public:
explicit PipelineCache(RasterizerVulkan& rasterizer, const Device& device, Scheduler& scheduler, explicit PipelineCache(RasterizerVulkan& rasterizer, const Device& device, Scheduler& scheduler,
DescriptorPool& descriptor_pool, DescriptorPool& descriptor_pool,
UpdateDescriptorQueue& update_descriptor_queue, GuestDescriptorQueue& guest_descriptor_queue,
RenderPassCache& render_pass_cache, BufferCache& buffer_cache, RenderPassCache& render_pass_cache, BufferCache& buffer_cache,
TextureCache& texture_cache, VideoCore::ShaderNotify& shader_notify_); TextureCache& texture_cache, VideoCore::ShaderNotify& shader_notify_);
~PipelineCache(); ~PipelineCache();
@ -144,7 +143,7 @@ private:
const Device& device; const Device& device;
Scheduler& scheduler; Scheduler& scheduler;
DescriptorPool& descriptor_pool; DescriptorPool& descriptor_pool;
UpdateDescriptorQueue& update_descriptor_queue; GuestDescriptorQueue& guest_descriptor_queue;
RenderPassCache& render_pass_cache; RenderPassCache& render_pass_cache;
BufferCache& buffer_cache; BufferCache& buffer_cache;
TextureCache& texture_cache; TextureCache& texture_cache;

18
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View file

@ -160,17 +160,16 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
: RasterizerAccelerated{cpu_memory_}, gpu{gpu_}, screen_info{screen_info_}, device{device_}, : RasterizerAccelerated{cpu_memory_}, gpu{gpu_}, screen_info{screen_info_}, device{device_},
memory_allocator{memory_allocator_}, state_tracker{state_tracker_}, scheduler{scheduler_}, memory_allocator{memory_allocator_}, state_tracker{state_tracker_}, scheduler{scheduler_},
staging_pool(device, memory_allocator, scheduler), descriptor_pool(device, scheduler), staging_pool(device, memory_allocator, scheduler), descriptor_pool(device, scheduler),
update_descriptor_queue(device, scheduler), guest_descriptor_queue(device, scheduler), compute_pass_descriptor_queue(device, scheduler),
blit_image(device, scheduler, state_tracker, descriptor_pool), blit_image(device, scheduler, state_tracker, descriptor_pool), render_pass_cache(device),
render_pass_cache(device), texture_cache_runtime{device, scheduler, texture_cache_runtime{
memory_allocator, staging_pool, device, scheduler, memory_allocator, staging_pool,
blit_image, render_pass_cache, blit_image, render_pass_cache, descriptor_pool, compute_pass_descriptor_queue},
descriptor_pool, update_descriptor_queue},
texture_cache(texture_cache_runtime, *this), texture_cache(texture_cache_runtime, *this),
buffer_cache_runtime(device, memory_allocator, scheduler, staging_pool, buffer_cache_runtime(device, memory_allocator, scheduler, staging_pool,
update_descriptor_queue, descriptor_pool), guest_descriptor_queue, compute_pass_descriptor_queue, descriptor_pool),
buffer_cache(*this, cpu_memory_, buffer_cache_runtime), buffer_cache(*this, cpu_memory_, buffer_cache_runtime),
pipeline_cache(*this, device, scheduler, descriptor_pool, update_descriptor_queue, pipeline_cache(*this, device, scheduler, descriptor_pool, guest_descriptor_queue,
render_pass_cache, buffer_cache, texture_cache, gpu.ShaderNotify()), render_pass_cache, buffer_cache, texture_cache, gpu.ShaderNotify()),
query_cache{*this, cpu_memory_, device, scheduler}, query_cache{*this, cpu_memory_, device, scheduler},
accelerate_dma(buffer_cache, texture_cache, scheduler), accelerate_dma(buffer_cache, texture_cache, scheduler),
@ -669,7 +668,8 @@ void RasterizerVulkan::FlushCommands() {
void RasterizerVulkan::TickFrame() { void RasterizerVulkan::TickFrame() {
draw_counter = 0; draw_counter = 0;
update_descriptor_queue.TickFrame(); guest_descriptor_queue.TickFrame();
compute_pass_descriptor_queue.TickFrame();
fence_manager.TickFrame(); fence_manager.TickFrame();
staging_pool.TickFrame(); staging_pool.TickFrame();
{ {

3
src/video_core/renderer_vulkan/vk_rasterizer.h
View file

@ -184,7 +184,8 @@ private:
StagingBufferPool staging_pool; StagingBufferPool staging_pool;
DescriptorPool descriptor_pool; DescriptorPool descriptor_pool;
UpdateDescriptorQueue update_descriptor_queue; GuestDescriptorQueue guest_descriptor_queue;
ComputePassDescriptorQueue compute_pass_descriptor_queue;
BlitImageHelper blit_image; BlitImageHelper blit_image;
RenderPassCache render_pass_cache; RenderPassCache render_pass_cache;

4
src/video_core/renderer_vulkan/vk_texture_cache.cpp
View file

@ -798,13 +798,13 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched
BlitImageHelper& blit_image_helper_, BlitImageHelper& blit_image_helper_,
RenderPassCache& render_pass_cache_, RenderPassCache& render_pass_cache_,
DescriptorPool& descriptor_pool, DescriptorPool& descriptor_pool,
UpdateDescriptorQueue& update_descriptor_queue) ComputePassDescriptorQueue& compute_pass_descriptor_queue)
: device{device_}, scheduler{scheduler_}, memory_allocator{memory_allocator_}, : device{device_}, scheduler{scheduler_}, memory_allocator{memory_allocator_},
staging_buffer_pool{staging_buffer_pool_}, blit_image_helper{blit_image_helper_}, staging_buffer_pool{staging_buffer_pool_}, blit_image_helper{blit_image_helper_},
render_pass_cache{render_pass_cache_}, resolution{Settings::values.resolution_info} { render_pass_cache{render_pass_cache_}, resolution{Settings::values.resolution_info} {
if (Settings::values.accelerate_astc) { if (Settings::values.accelerate_astc) {
astc_decoder_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, astc_decoder_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool,
update_descriptor_queue, memory_allocator); compute_pass_descriptor_queue, memory_allocator);
} }
} }

3
src/video_core/renderer_vulkan/vk_texture_cache.h
View file

@ -34,7 +34,6 @@ class ImageView;
class Framebuffer; class Framebuffer;
class RenderPassCache; class RenderPassCache;
class StagingBufferPool; class StagingBufferPool;
class UpdateDescriptorQueue;
class Scheduler; class Scheduler;
class TextureCacheRuntime { class TextureCacheRuntime {
@ -45,7 +44,7 @@ public:
BlitImageHelper& blit_image_helper_, BlitImageHelper& blit_image_helper_,
RenderPassCache& render_pass_cache_, RenderPassCache& render_pass_cache_,
DescriptorPool& descriptor_pool, DescriptorPool& descriptor_pool,
UpdateDescriptorQueue& update_descriptor_queue); ComputePassDescriptorQueue& compute_pass_descriptor_queue);
void Finish(); void Finish();

6
src/video_core/renderer_vulkan/vk_update_descriptor.h
View file

@ -32,7 +32,7 @@ class UpdateDescriptorQueue final {
// This should be plenty for the vast majority of cases. Most desktop platforms only // This should be plenty for the vast majority of cases. Most desktop platforms only
// provide up to 3 swapchain images. // provide up to 3 swapchain images.
static constexpr size_t FRAMES_IN_FLIGHT = 5; static constexpr size_t FRAMES_IN_FLIGHT = 5;
static constexpr size_t FRAME_PAYLOAD_SIZE = 0x10000; static constexpr size_t FRAME_PAYLOAD_SIZE = 0x20000;
static constexpr size_t PAYLOAD_SIZE = FRAME_PAYLOAD_SIZE * FRAMES_IN_FLIGHT; static constexpr size_t PAYLOAD_SIZE = FRAME_PAYLOAD_SIZE * FRAMES_IN_FLIGHT;
public: public:
@ -86,4 +86,8 @@ private:
std::array<DescriptorUpdateEntry, PAYLOAD_SIZE> payload; std::array<DescriptorUpdateEntry, PAYLOAD_SIZE> payload;
}; };
// TODO: should these be separate classes instead?
using GuestDescriptorQueue = UpdateDescriptorQueue;
using ComputePassDescriptorQueue = UpdateDescriptorQueue;
} // namespace Vulkan } // namespace Vulkan