Plexi09/eden-miror
1
0
Fork 0
mirror of https://git.eden-emu.dev/eden-emu/eden synced 2026-06-28 09:45:45 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

[vulkan] Simplification of Acquire policy for descriptors usage

Browse source
This commit is contained in:
CamilleLaVey 2026-06-26 21:58:43 -04:00
parent fb53c4e135
commit 63aa33a544
8 changed files with 33 additions and 17 deletions
Download patch file Download diff file Expand all files Collapse all files

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

@ -22,6 +22,15 @@ namespace Vulkan {
using Shader::Backend::SPIRV::NUM_TEXTURE_AND_IMAGE_SCALING_WORDS;
[[nodiscard]] inline u32 NumDescriptorEntries(const Shader::Info& info) {
return Shader::NumDescriptors(info.constant_buffer_descriptors) +
Shader::NumDescriptors(info.storage_buffers_descriptors) +
Shader::NumDescriptors(info.texture_buffer_descriptors) +
Shader::NumDescriptors(info.image_buffer_descriptors) +
Shader::NumDescriptors(info.texture_descriptors) +
Shader::NumDescriptors(info.image_descriptors);
}
class DescriptorLayoutBuilder {
public:
DescriptorLayoutBuilder(const Device& device_) : device{&device_} {}

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

@ -326,7 +326,7 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer
const u32 staging_size = static_cast<u32>(num_vertices * sizeof(u16));
const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
compute_pass_descriptor_queue.Acquire(scheduler);
compute_pass_descriptor_queue.Acquire(scheduler, 2);
compute_pass_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices);
compute_pass_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size);
const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
@ -384,7 +384,7 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
const std::size_t staging_size = num_tri_vertices * sizeof(u32);
const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
compute_pass_descriptor_queue.Acquire(scheduler);
compute_pass_descriptor_queue.Acquire(scheduler, 2);
compute_pass_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size);
compute_pass_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size);
const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
@ -429,7 +429,7 @@ void ConditionalRenderingResolvePass::Resolve(VkBuffer dst_buffer, VkBuffer src_
}
const size_t compare_size = compare_to_zero ? 8 : 24;
compute_pass_descriptor_queue.Acquire(scheduler);
compute_pass_descriptor_queue.Acquire(scheduler, 2);
compute_pass_descriptor_queue.AddBuffer(src_buffer, src_offset, compare_size);
compute_pass_descriptor_queue.AddBuffer(dst_buffer, 0, sizeof(u32));
const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
@ -498,7 +498,7 @@ void QueriesPrefixScanPass::Run(VkBuffer accumulation_buffer, VkBuffer dst_buffe
static constexpr size_t DISPATCH_SIZE = 2048U;
size_t runs_to_do = std::min<size_t>(current_runs, DISPATCH_SIZE);
current_runs -= runs_to_do;
compute_pass_descriptor_queue.Acquire(scheduler);
compute_pass_descriptor_queue.Acquire(scheduler, 3);
compute_pass_descriptor_queue.AddBuffer(src_buffer, 0, number_of_sums * sizeof(u64));
compute_pass_descriptor_queue.AddBuffer(dst_buffer, 0, number_of_sums * sizeof(u64));
compute_pass_descriptor_queue.AddBuffer(accumulation_buffer, 0, sizeof(u64));
@ -600,7 +600,7 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U);
const u32 num_dispatches_z = image.info.resources.layers;
compute_pass_descriptor_queue.Acquire(scheduler);
compute_pass_descriptor_queue.Acquire(scheduler, 2);
compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset,
image.guest_size_bytes - swizzle.buffer_offset);
compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
@ -821,7 +821,7 @@ void BlockLinearUnswizzle3DPass::UnswizzleChunk(
pc.blocks_dim[1] = blocks_y;
pc.blocks_dim[2] = z_count; // Only process the count
compute_pass_descriptor_queue.Acquire(scheduler);
compute_pass_descriptor_queue.Acquire(scheduler, 3);
compute_pass_descriptor_queue.AddBuffer(*image.runtime->swizzle_table_buffer, 0,
image.runtime->swizzle_table_size);
compute_pass_descriptor_queue.AddBuffer(swizzled.buffer,
@ -989,7 +989,7 @@ void MSAACopyPass::CopyImage(Image& dst_image, Image& src_image,
ASSERT(copy.dst_subresource.base_layer == 0);
ASSERT(copy.dst_subresource.num_layers == 1);
compute_pass_descriptor_queue.Acquire(scheduler);
compute_pass_descriptor_queue.Acquire(scheduler, 2);
compute_pass_descriptor_queue.AddImage(
src_image.StorageImageView(copy.src_subresource.base_level));
compute_pass_descriptor_queue.AddImage(

3
src/video_core/renderer_vulkan/vk_compute_pipeline.cpp
View file

@ -45,6 +45,7 @@ ComputePipeline::ComputePipeline(const Device& device_, Scheduler& scheduler, vk
}
std::copy_n(info.constant_buffer_used_sizes.begin(), uniform_buffer_sizes.size(),
uniform_buffer_sizes.begin());
num_descriptor_entries = NumDescriptorEntries(info);
auto func{[this, &scheduler, &descriptor_pool, shader_notify, pipeline_statistics] {
DescriptorLayoutBuilder builder{device};
@ -113,7 +114,7 @@ ComputePipeline::ComputePipeline(const Device& device_, Scheduler& scheduler, vk
void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
Tegra::MemoryManager& gpu_memory, Scheduler& scheduler,
BufferCache& buffer_cache, TextureCache& texture_cache) {
guest_descriptor_queue.Acquire(scheduler);
guest_descriptor_queue.Acquire(scheduler, num_descriptor_entries);
buffer_cache.SetComputeUniformBufferState(info.constant_buffer_mask, &uniform_buffer_sizes);
buffer_cache.UnbindComputeStorageBuffers();

1
src/video_core/renderer_vulkan/vk_compute_pipeline.h
View file

@ -53,6 +53,7 @@ private:
vk::PipelineCache& pipeline_cache;
GuestDescriptorQueue& guest_descriptor_queue;
Shader::Info info;
u32 num_descriptor_entries{};
VideoCommon::ComputeUniformBufferSizes uniform_buffer_sizes{};

3
src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
View file

@ -268,6 +268,7 @@ GraphicsPipeline::GraphicsPipeline(
num_textures += Shader::NumDescriptors(info->texture_descriptors);
num_image_elements += Shader::NumDescriptors(info->texture_descriptors);
num_image_elements += Shader::NumDescriptors(info->image_descriptors);
num_descriptor_entries += NumDescriptorEntries(*info);
}
fragment_has_color0_output = stage_infos[NUM_STAGES - 1].stores_frag_color[0];
auto func{[this, shader_notify, &render_pass_cache, &descriptor_pool, pipeline_statistics] {
@ -473,7 +474,7 @@ bool GraphicsPipeline::ConfigureImpl(bool is_indexed) {
buffer_cache.UpdateGraphicsBuffers(is_indexed);
buffer_cache.BindHostGeometryBuffers(is_indexed);
guest_descriptor_queue.Acquire(scheduler);
guest_descriptor_queue.Acquire(scheduler, num_descriptor_entries);
RescalingPushConstant rescaling;
RenderAreaPushConstant render_area;

1
src/video_core/renderer_vulkan/vk_graphics_pipeline.h
View file

@ -159,6 +159,7 @@ private:
std::array<Shader::Info, NUM_STAGES> stage_infos;
std::array<u32, 5> enabled_uniform_buffer_masks{};
VideoCommon::UniformBufferSizes uniform_buffer_sizes{};
u32 num_descriptor_entries{};
size_t num_image_elements{};
u32 num_textures{};
bool fragment_has_color0_output{};

17
src/video_core/renderer_vulkan/vk_update_descriptor.cpp
View file

@ -7,6 +7,7 @@
#include <variant>
#include <boost/container/static_vector.hpp>
#include "common/assert.h"
#include "common/logging.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
@ -32,13 +33,15 @@ void UpdateDescriptorQueue::TickFrame() {
payload_cursor = payload_start;
}
void UpdateDescriptorQueue::Acquire(Scheduler& scheduler) {
// Minimum number of entries required.
// This is the maximum number of entries a single draw call might use.
static constexpr size_t MIN_ENTRIES = 0x400;
if (std::distance(payload_start, payload_cursor) + MIN_ENTRIES >= FRAME_PAYLOAD_SIZE) {
LOG_WARNING(Render_Vulkan, "Payload overflow, waiting for worker thread");
void UpdateDescriptorQueue::Acquire(Scheduler& scheduler, size_t required_entries) {
static constexpr size_t DEFAULT_REQUIRED_ENTRIES = 0x400;
const size_t reserve = required_entries > 0 ? required_entries : DEFAULT_REQUIRED_ENTRIES;
ASSERT_MSG(reserve < FRAME_PAYLOAD_SIZE, "Descriptor reservation {} >= frame capacity {}",
reserve, FRAME_PAYLOAD_SIZE);
const size_t used = static_cast<size_t>(std::distance(payload_start, payload_cursor));
if (used + reserve >= FRAME_PAYLOAD_SIZE) {
LOG_WARNING(Render_Vulkan, "Payload overflow (used={}, reserve={}, capacity={})",
used, reserve, FRAME_PAYLOAD_SIZE);
scheduler.WaitWorker();
payload_cursor = payload_start;
}

2
src/video_core/renderer_vulkan/vk_update_descriptor.h
View file

@ -38,7 +38,7 @@ public:
~UpdateDescriptorQueue();
void TickFrame();
void Acquire(Scheduler& scheduler);
void Acquire(Scheduler& scheduler, size_t required_entries = 0);
const DescriptorUpdateEntry* UpdateData() const noexcept {
return upload_start;