N-archive/citra
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

renderer_vulkan: Rewrite descriptor management

Browse source 
* Switch to batched vkUpdateDescriptorSets from cached descriptor sets with templates
This commit is contained in:
GPUCode 2024-02-09 01:56:20 +02:00
parent 4a63fc2ca2
commit 0c9037f075
36 changed files with 786 additions and 740 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
externals/vulkan-headers vendored

@ -1 +1 @@
Subproject commit 217e93c664ec6704ec2d8c36fa116c1a4a1e2d40 Subproject commit 5a5c9a643484d888873e32c5d7d484fae8e71d3d

4
src/video_core/CMakeLists.txt
View file

@ -160,8 +160,8 @@ if (ENABLE_VULKAN)
renderer_vulkan/vk_blit_helper.h renderer_vulkan/vk_blit_helper.h
renderer_vulkan/vk_common.cpp renderer_vulkan/vk_common.cpp
renderer_vulkan/vk_common.h renderer_vulkan/vk_common.h
renderer_vulkan/vk_descriptor_pool.cpp renderer_vulkan/vk_descriptor_update_queue.cpp
renderer_vulkan/vk_descriptor_pool.h renderer_vulkan/vk_descriptor_update_queue.h
renderer_vulkan/vk_graphics_pipeline.cpp renderer_vulkan/vk_graphics_pipeline.cpp
renderer_vulkan/vk_graphics_pipeline.h renderer_vulkan/vk_graphics_pipeline.h
renderer_vulkan/vk_master_semaphore.cpp renderer_vulkan/vk_master_semaphore.cpp

2
src/video_core/custom_textures/custom_tex_manager.cpp
View file

@ -385,7 +385,7 @@ std::vector<FileUtil::FSTEntry> CustomTexManager::GetTextures(u64 title_id) {
} }
void CustomTexManager::CreateWorkers() { void CustomTexManager::CreateWorkers() {
const std::size_t num_workers = std::max(std::thread::hardware_concurrency(), 2U) - 1; const std::size_t num_workers = std::max(std::thread::hardware_concurrency(), 2U) >> 1;
workers = std::make_unique<Common::ThreadWorker>(num_workers, "Custom textures"); workers = std::make_unique<Common::ThreadWorker>(num_workers, "Custom textures");
} }

8
src/video_core/pica/regs_texturing.h
View file

@ -176,15 +176,15 @@ struct TexturingRegs {
INSERT_PADDING_WORDS(0x9); INSERT_PADDING_WORDS(0x9);
struct FullTextureConfig { struct FullTextureConfig {
const bool enabled; const u32 enabled;
const TextureConfig config; const TextureConfig config;
const TextureFormat format; const TextureFormat format;
}; };
const std::array<FullTextureConfig, 3> GetTextures() const { const std::array<FullTextureConfig, 3> GetTextures() const {
return {{ return {{
{static_cast<bool>(main_config.texture0_enable), texture0, texture0_format}, {main_config.texture0_enable, texture0, texture0_format},
{static_cast<bool>(main_config.texture1_enable), texture1, texture1_format}, {main_config.texture1_enable, texture1, texture1_format},
{static_cast<bool>(main_config.texture2_enable), texture2, texture2_format}, {main_config.texture2_enable, texture2, texture2_format},
}}; }};
} }

63
src/video_core/rasterizer_cache/rasterizer_cache.h
View file

@ -600,14 +600,43 @@ typename T::Surface& RasterizerCache<T>::GetTextureCube(const TextureCubeConfig&
auto [it, new_surface] = texture_cube_cache.try_emplace(config); auto [it, new_surface] = texture_cube_cache.try_emplace(config);
TextureCube& cube = it->second; TextureCube& cube = it->second;
const std::array addresses = {config.px, config.nx, config.py, config.ny, config.pz, config.nz};
if (new_surface) { if (new_surface) {
Pica::Texture::TextureInfo info = {
.width = config.width,
.height = config.width,
.format = config.format,
};
info.SetDefaultStride();
u32 res_scale = 1;
for (u32 i = 0; i < addresses.size(); i++) {
if (!addresses[i]) {
continue;
}
SurfaceId& face_id = cube.face_ids[i];
if (!face_id) {
info.physical_address = addresses[i];
face_id = GetTextureSurface(info, config.levels - 1);
Surface& surface = slot_surfaces[face_id];
ASSERT_MSG(
surface.levels >= config.levels,
"Texture cube face levels are not enough to validate the levels requested");
surface.flags |= SurfaceFlagBits::Tracked;
}
Surface& surface = slot_surfaces[face_id];
res_scale = std::max(surface.res_scale, res_scale);
}
SurfaceParams cube_params = { SurfaceParams cube_params = {
.addr = config.px, .addr = config.px,
.width = config.width, .width = config.width,
.height = config.width, .height = config.width,
.stride = config.width, .stride = config.width,
.levels = config.levels, .levels = config.levels,
.res_scale = filter != Settings::TextureFilter::None ? resolution_scale_factor : 1, .res_scale = res_scale,
.texture_type = TextureType::CubeMap, .texture_type = TextureType::CubeMap,
.pixel_format = PixelFormatFromTextureFormat(config.format), .pixel_format = PixelFormatFromTextureFormat(config.format),
.type = SurfaceType::Texture, .type = SurfaceType::Texture,
@ -616,38 +645,20 @@ typename T::Surface& RasterizerCache<T>::GetTextureCube(const TextureCubeConfig&
cube.surface_id = CreateSurface(cube_params); cube.surface_id = CreateSurface(cube_params);
} }
const u32 scaled_size = slot_surfaces[cube.surface_id].GetScaledWidth(); Surface& cube_surface = slot_surfaces[cube.surface_id];
const std::array addresses = {config.px, config.nx, config.py, config.ny, config.pz, config.nz};
Pica::Texture::TextureInfo info = {
.width = config.width,
.height = config.width,
.format = config.format,
};
info.SetDefaultStride();
for (u32 i = 0; i < addresses.size(); i++) { for (u32 i = 0; i < addresses.size(); i++) {
if (!addresses[i]) { if (!addresses[i]) {
continue; continue;
} }
Surface& surface = slot_surfaces[cube.face_ids[i]];
SurfaceId& face_id = cube.face_ids[i];
if (!face_id) {
info.physical_address = addresses[i];
face_id = GetTextureSurface(info, config.levels - 1);
ASSERT_MSG(slot_surfaces[face_id].levels >= config.levels,
"Texture cube face levels are not enough to validate the levels requested");
}
Surface& surface = slot_surfaces[face_id];
surface.flags |= SurfaceFlagBits::Tracked;
if (cube.ticks[i] == surface.modification_tick) { if (cube.ticks[i] == surface.modification_tick) {
continue; continue;
} }
cube.ticks[i] = surface.modification_tick; cube.ticks[i] = surface.modification_tick;
Surface& cube_surface = slot_surfaces[cube.surface_id]; boost::container::small_vector<TextureCopy, 8> upload_copies;
for (u32 level = 0; level < config.levels; level++) { for (u32 level = 0; level < config.levels; level++) {
const u32 width_lod = scaled_size >> level; const u32 width_lod = surface.GetScaledWidth() >> level;
const TextureCopy texture_copy = { upload_copies.push_back({
.src_level = level, .src_level = level,
.dst_level = level, .dst_level = level,
.src_layer = 0, .src_layer = 0,
@ -655,9 +666,9 @@ typename T::Surface& RasterizerCache<T>::GetTextureCube(const TextureCubeConfig&
.src_offset = {0, 0}, .src_offset = {0, 0},
.dst_offset = {0, 0}, .dst_offset = {0, 0},
.extent = {width_lod, width_lod}, .extent = {width_lod, width_lod},
}; });
runtime.CopyTextures(surface, cube_surface, texture_copy);
} }
runtime.CopyTextures(surface, cube_surface, upload_copies);
} }
return slot_surfaces[cube.surface_id]; return slot_surfaces[cube.surface_id];

13
src/video_core/renderer_opengl/gl_texture_runtime.cpp
View file

@ -260,16 +260,19 @@ void TextureRuntime::ClearTexture(Surface& surface, const VideoCore::TextureClea
} }
bool TextureRuntime::CopyTextures(Surface& source, Surface& dest, bool TextureRuntime::CopyTextures(Surface& source, Surface& dest,
const VideoCore::TextureCopy& copy) { std::span<const VideoCore::TextureCopy> copies) {
const GLenum src_textarget = source.texture_type == VideoCore::TextureType::CubeMap const GLenum src_textarget = source.texture_type == VideoCore::TextureType::CubeMap
? GL_TEXTURE_CUBE_MAP ? GL_TEXTURE_CUBE_MAP
: GL_TEXTURE_2D; : GL_TEXTURE_2D;
const GLenum dest_textarget = const GLenum dest_textarget =
dest.texture_type == VideoCore::TextureType::CubeMap ? GL_TEXTURE_CUBE_MAP : GL_TEXTURE_2D; dest.texture_type == VideoCore::TextureType::CubeMap ? GL_TEXTURE_CUBE_MAP : GL_TEXTURE_2D;
glCopyImageSubData(source.Handle(), src_textarget, copy.src_level, copy.src_offset.x,
copy.src_offset.y, copy.src_layer, dest.Handle(), dest_textarget, for (const auto& copy : copies) {
copy.dst_level, copy.dst_offset.x, copy.dst_offset.y, copy.dst_layer, glCopyImageSubData(source.Handle(), src_textarget, copy.src_level, copy.src_offset.x,
copy.extent.width, copy.extent.height, 1); copy.src_offset.y, copy.src_layer, dest.Handle(), dest_textarget,
copy.dst_level, copy.dst_offset.x, copy.dst_offset.y, copy.dst_layer,
copy.extent.width, copy.extent.height, 1);
}
return true; return true;
} }

7
src/video_core/renderer_opengl/gl_texture_runtime.h
View file

@ -65,7 +65,12 @@ public:
void ClearTexture(Surface& surface, const VideoCore::TextureClear& clear); void ClearTexture(Surface& surface, const VideoCore::TextureClear& clear);
/// Copies a rectangle of source to another rectange of dest /// Copies a rectangle of source to another rectange of dest
bool CopyTextures(Surface& source, Surface& dest, const VideoCore::TextureCopy& copy); bool CopyTextures(Surface& source, Surface& dest,
std::span<const VideoCore::TextureCopy> copies);
bool CopyTextures(Surface& source, Surface& dest, const VideoCore::TextureCopy& copy) {
return CopyTextures(source, dest, std::array{copy});
}
/// Blits a rectangle of source to another rectange of dest /// Blits a rectangle of source to another rectange of dest
bool BlitTextures(Surface& source, Surface& dest, const VideoCore::TextureBlit& blit); bool BlitTextures(Surface& source, Surface& dest, const VideoCore::TextureBlit& blit);

44
src/video_core/renderer_vulkan/renderer_vulkan.cpp
View file

@ -54,21 +54,21 @@ RendererVulkan::RendererVulkan(Core::System& system, Pica::PicaCore& pica_,
Frontend::EmuWindow& window, Frontend::EmuWindow* secondary_window) Frontend::EmuWindow& window, Frontend::EmuWindow* secondary_window)
: RendererBase{system, window, secondary_window}, memory{system.Memory()}, pica{pica_}, : RendererBase{system, window, secondary_window}, memory{system.Memory()}, pica{pica_},
instance{system.TelemetrySession(), window, Settings::values.physical_device.GetValue()}, instance{system.TelemetrySession(), window, Settings::values.physical_device.GetValue()},
scheduler{instance}, render_manager{instance, scheduler}, pool{instance}, scheduler{instance}, render_manager{instance, scheduler}, main_window{window, instance,
main_window{window, instance, scheduler}, scheduler},
vertex_buffer{instance, scheduler, vk::BufferUsageFlagBits::eVertexBuffer, vertex_buffer{instance, scheduler, vk::BufferUsageFlagBits::eVertexBuffer,
VERTEX_BUFFER_SIZE}, VERTEX_BUFFER_SIZE},
rasterizer{memory, update_queue{instance}, rasterizer{memory,
pica, pica,
system.CustomTexManager(), system.CustomTexManager(),
*this, *this,
render_window, render_window,
instance, instance,
scheduler, scheduler,
pool, render_manager,
render_manager, update_queue,
main_window.ImageCount()}, main_window.ImageCount()},
present_set_provider{instance, pool, PRESENT_BINDINGS} { present_heap{instance, scheduler.GetMasterSemaphore(), PRESENT_BINDINGS, 32} {
CompileShaders(); CompileShaders();
BuildLayouts(); BuildLayouts();
BuildPipelines(); BuildPipelines();
@ -127,16 +127,14 @@ void RendererVulkan::PrepareRendertarget() {
void RendererVulkan::PrepareDraw(Frame* frame, const Layout::FramebufferLayout& layout) { void RendererVulkan::PrepareDraw(Frame* frame, const Layout::FramebufferLayout& layout) {
const auto sampler = present_samplers[!Settings::values.filter_mode.GetValue()]; const auto sampler = present_samplers[!Settings::values.filter_mode.GetValue()];
std::transform(screen_infos.begin(), screen_infos.end(), present_textures.begin(), const auto present_set = present_heap.Commit();
[&](auto& info) { for (u32 index = 0; index < screen_infos.size(); index++) {
return DescriptorData{vk::DescriptorImageInfo{sampler, info.image_view, update_queue.AddImageSampler(present_set, 0, index, screen_infos[index].image_view,
vk::ImageLayout::eGeneral}}; sampler);
}); }
const auto descriptor_set = present_set_provider.Acquire(present_textures);
render_manager.EndRendering(); render_manager.EndRendering();
scheduler.Record([this, layout, frame, descriptor_set, renderpass = main_window.Renderpass(), scheduler.Record([this, layout, frame, present_set, renderpass = main_window.Renderpass(),
index = current_pipeline](vk::CommandBuffer cmdbuf) { index = current_pipeline](vk::CommandBuffer cmdbuf) {
const vk::Viewport viewport = { const vk::Viewport viewport = {
.x = 0.0f, .x = 0.0f,
@ -171,7 +169,7 @@ void RendererVulkan::PrepareDraw(Frame* frame, const Layout::FramebufferLayout&
cmdbuf.beginRenderPass(renderpass_begin_info, vk::SubpassContents::eInline); cmdbuf.beginRenderPass(renderpass_begin_info, vk::SubpassContents::eInline);
cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, present_pipelines[index]); cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, present_pipelines[index]);
cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eGraphics, layout, 0, descriptor_set, {}); cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eGraphics, layout, 0, present_set, {});
}); });
} }
@ -258,7 +256,7 @@ void RendererVulkan::BuildLayouts() {
.size = sizeof(PresentUniformData), .size = sizeof(PresentUniformData),
}; };
const auto descriptor_set_layout = present_set_provider.Layout(); const auto descriptor_set_layout = present_heap.Layout();
const vk::PipelineLayoutCreateInfo layout_info = { const vk::PipelineLayoutCreateInfo layout_info = {
.setLayoutCount = 1, .setLayoutCount = 1,
.pSetLayouts = &descriptor_set_layout, .pSetLayouts = &descriptor_set_layout,

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

@ -7,7 +7,6 @@
#include "common/common_types.h" #include "common/common_types.h"
#include "common/math_util.h" #include "common/math_util.h"
#include "video_core/renderer_base.h" #include "video_core/renderer_base.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_instance.h" #include "video_core/renderer_vulkan/vk_instance.h"
#include "video_core/renderer_vulkan/vk_present_window.h" #include "video_core/renderer_vulkan/vk_present_window.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h" #include "video_core/renderer_vulkan/vk_rasterizer.h"
@ -119,14 +118,14 @@ private:
Instance instance; Instance instance;
Scheduler scheduler; Scheduler scheduler;
RenderManager render_manager; RenderManager render_manager;
DescriptorPool pool;
PresentWindow main_window; PresentWindow main_window;
StreamBuffer vertex_buffer; StreamBuffer vertex_buffer;
DescriptorUpdateQueue update_queue;
RasterizerVulkan rasterizer; RasterizerVulkan rasterizer;
std::unique_ptr<PresentWindow> second_window; std::unique_ptr<PresentWindow> second_window;
DescriptorHeap present_heap;
vk::UniquePipelineLayout present_pipeline_layout; vk::UniquePipelineLayout present_pipeline_layout;
DescriptorSetProvider present_set_provider;
std::array<vk::Pipeline, PRESENT_PIPELINES> present_pipelines; std::array<vk::Pipeline, PRESENT_PIPELINES> present_pipelines;
std::array<vk::ShaderModule, PRESENT_PIPELINES> present_shaders; std::array<vk::ShaderModule, PRESENT_PIPELINES> present_shaders;
std::array<vk::Sampler, 2> present_samplers; std::array<vk::Sampler, 2> present_samplers;
@ -134,7 +133,6 @@ private:
u32 current_pipeline = 0; u32 current_pipeline = 0;
std::array<ScreenInfo, 3> screen_infos{}; std::array<ScreenInfo, 3> screen_infos{};
std::array<DescriptorData, 3> present_textures{};
PresentUniformData draw_info{}; PresentUniformData draw_info{};
vk::ClearColorValue clear_color{}; vk::ClearColorValue clear_color{};
}; };

79
src/video_core/renderer_vulkan/vk_blit_helper.cpp
View file

@ -4,6 +4,7 @@
#include "common/vector_math.h" #include "common/vector_math.h"
#include "video_core/renderer_vulkan/vk_blit_helper.h" #include "video_core/renderer_vulkan/vk_blit_helper.h"
#include "video_core/renderer_vulkan/vk_descriptor_update_queue.h"
#include "video_core/renderer_vulkan/vk_instance.h" #include "video_core/renderer_vulkan/vk_instance.h"
#include "video_core/renderer_vulkan/vk_render_manager.h" #include "video_core/renderer_vulkan/vk_render_manager.h"
#include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_scheduler.h"
@ -177,12 +178,13 @@ constexpr vk::PipelineShaderStageCreateInfo MakeStages(vk::ShaderModule compute_
} // Anonymous namespace } // Anonymous namespace
BlitHelper::BlitHelper(const Instance& instance_, Scheduler& scheduler_, DescriptorPool& pool, BlitHelper::BlitHelper(const Instance& instance_, Scheduler& scheduler_,
RenderManager& render_manager_) RenderManager& render_manager_, DescriptorUpdateQueue& update_queue_)
: instance{instance_}, scheduler{scheduler_}, render_manager{render_manager_}, : instance{instance_}, scheduler{scheduler_}, render_manager{render_manager_},
device{instance.GetDevice()}, compute_provider{instance, pool, COMPUTE_BINDINGS}, update_queue{update_queue_}, device{instance.GetDevice()},
compute_buffer_provider{instance, pool, COMPUTE_BUFFER_BINDINGS}, compute_provider{instance, scheduler.GetMasterSemaphore(), COMPUTE_BINDINGS},
two_textures_provider{instance, pool, TWO_TEXTURES_BINDINGS}, compute_buffer_provider{instance, scheduler.GetMasterSemaphore(), COMPUTE_BUFFER_BINDINGS},
two_textures_provider{instance, scheduler.GetMasterSemaphore(), TWO_TEXTURES_BINDINGS, 16},
compute_pipeline_layout{ compute_pipeline_layout{
device.createPipelineLayout(PipelineLayoutCreateInfo(&compute_provider.Layout(), true))}, device.createPipelineLayout(PipelineLayoutCreateInfo(&compute_provider.Layout(), true))},
compute_buffer_pipeline_layout{device.createPipelineLayout( compute_buffer_pipeline_layout{device.createPipelineLayout(
@ -282,24 +284,13 @@ bool BlitHelper::BlitDepthStencil(Surface& source, Surface& dest,
.extent = {dest.GetScaledWidth(), dest.GetScaledHeight()}, .extent = {dest.GetScaledWidth(), dest.GetScaledHeight()},
}; };
std::array<DescriptorData, 2> textures{}; const auto descriptor_set = two_textures_provider.Commit();
textures[0].image_info = vk::DescriptorImageInfo{ update_queue.AddImageSampler(descriptor_set, 0, 0, source.DepthView(), nearest_sampler);
.sampler = nearest_sampler, update_queue.AddImageSampler(descriptor_set, 1, 0, source.StencilView(), nearest_sampler);
.imageView = source.DepthView(),
.imageLayout = vk::ImageLayout::eGeneral,
};
textures[1].image_info = vk::DescriptorImageInfo{
.sampler = nearest_sampler,
.imageView = source.StencilView(),
.imageLayout = vk::ImageLayout::eGeneral,
};
const auto descriptor_set = two_textures_provider.Acquire(textures);
const RenderPass depth_pass = { const RenderPass depth_pass = {
.framebuffer = dest.Framebuffer(), .framebuffer = dest.Framebuffer(),
.render_pass = .render_pass = render_manager.GetRenderpass(PixelFormat::Invalid, dest.pixel_format, false),
render_manager.GetRenderpass(PixelFormat::Invalid, dest.pixel_format, false),
.render_area = dst_render_area, .render_area = dst_render_area,
}; };
render_manager.BeginRendering(depth_pass); render_manager.BeginRendering(depth_pass);
@ -318,21 +309,12 @@ bool BlitHelper::BlitDepthStencil(Surface& source, Surface& dest,
bool BlitHelper::ConvertDS24S8ToRGBA8(Surface& source, Surface& dest, bool BlitHelper::ConvertDS24S8ToRGBA8(Surface& source, Surface& dest,
const VideoCore::TextureCopy& copy) { const VideoCore::TextureCopy& copy) {
std::array<DescriptorData, 3> textures{}; const auto descriptor_set = compute_provider.Commit();
textures[0].image_info = vk::DescriptorImageInfo{ update_queue.AddImageSampler(descriptor_set, 0, 0, source.DepthView(), VK_NULL_HANDLE,
.imageView = source.DepthView(), vk::ImageLayout::eDepthStencilReadOnlyOptimal);
.imageLayout = vk::ImageLayout::eDepthStencilReadOnlyOptimal, update_queue.AddImageSampler(descriptor_set, 1, 0, source.StencilView(), VK_NULL_HANDLE,
}; vk::ImageLayout::eDepthStencilReadOnlyOptimal);
textures[1].image_info = vk::DescriptorImageInfo{ update_queue.AddStorageImage(descriptor_set, 2, dest.ImageView());
.imageView = source.StencilView(),
.imageLayout = vk::ImageLayout::eDepthStencilReadOnlyOptimal,
};
textures[2].image_info = vk::DescriptorImageInfo{
.imageView = dest.ImageView(),
.imageLayout = vk::ImageLayout::eGeneral,
};
const auto descriptor_set = compute_provider.Acquire(textures);
render_manager.EndRendering(); render_manager.EndRendering();
scheduler.Record([this, descriptor_set, copy, src_image = source.Image(), scheduler.Record([this, descriptor_set, copy, src_image = source.Image(),
@ -438,24 +420,13 @@ bool BlitHelper::ConvertDS24S8ToRGBA8(Surface& source, Surface& dest,
bool BlitHelper::DepthToBuffer(Surface& source, vk::Buffer buffer, bool BlitHelper::DepthToBuffer(Surface& source, vk::Buffer buffer,
const VideoCore::BufferTextureCopy& copy) { const VideoCore::BufferTextureCopy& copy) {
std::array<DescriptorData, 3> textures{}; const auto descriptor_set = compute_buffer_provider.Commit();
textures[0].image_info = vk::DescriptorImageInfo{ update_queue.AddImageSampler(descriptor_set, 0, 0, source.DepthView(), nearest_sampler,
.sampler = nearest_sampler, vk::ImageLayout::eDepthStencilReadOnlyOptimal);
.imageView = source.DepthView(), update_queue.AddImageSampler(descriptor_set, 1, 0, source.StencilView(), nearest_sampler,
.imageLayout = vk::ImageLayout::eDepthStencilReadOnlyOptimal, vk::ImageLayout::eDepthStencilReadOnlyOptimal);
}; update_queue.AddBuffer(descriptor_set, 2, buffer, copy.buffer_offset, copy.buffer_size,
textures[1].image_info = vk::DescriptorImageInfo{ vk::DescriptorType::eStorageBuffer);
.sampler = nearest_sampler,
.imageView = source.StencilView(),
.imageLayout = vk::ImageLayout::eDepthStencilReadOnlyOptimal,
};
textures[2].buffer_info = vk::DescriptorBufferInfo{
.buffer = buffer,
.offset = copy.buffer_offset,
.range = copy.buffer_size,
};
const auto descriptor_set = compute_buffer_provider.Acquire(textures);
render_manager.EndRendering(); render_manager.EndRendering();
scheduler.Record([this, descriptor_set, copy, src_image = source.Image(), scheduler.Record([this, descriptor_set, copy, src_image = source.Image(),
@ -544,7 +515,7 @@ vk::Pipeline BlitHelper::MakeDepthStencilBlitPipeline() {
const std::array stages = MakeStages(full_screen_vert, blit_depth_stencil_frag); const std::array stages = MakeStages(full_screen_vert, blit_depth_stencil_frag);
const auto renderpass = render_manager.GetRenderpass(VideoCore::PixelFormat::Invalid, const auto renderpass = render_manager.GetRenderpass(VideoCore::PixelFormat::Invalid,
VideoCore::PixelFormat::D24S8, false); VideoCore::PixelFormat::D24S8, false);
vk::GraphicsPipelineCreateInfo depth_stencil_info = { vk::GraphicsPipelineCreateInfo depth_stencil_info = {
.stageCount = static_cast<u32>(stages.size()), .stageCount = static_cast<u32>(stages.size()),
.pStages = stages.data(), .pStages = stages.data(),

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

@ -4,7 +4,7 @@
#pragma once #pragma once
#include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/renderer_vulkan/vk_resource_pool.h"
namespace VideoCore { namespace VideoCore {
struct TextureBlit; struct TextureBlit;
@ -18,13 +18,14 @@ class Instance;
class RenderManager; class RenderManager;
class Scheduler; class Scheduler;
class Surface; class Surface;
class DescriptorUpdateQueue;
class BlitHelper { class BlitHelper {
friend class TextureRuntime; friend class TextureRuntime;
public: public:
BlitHelper(const Instance& instance, Scheduler& scheduler, DescriptorPool& pool, explicit BlitHelper(const Instance& instance, Scheduler& scheduler,
RenderManager& render_manager); RenderManager& render_manager, DescriptorUpdateQueue& update_queue);
~BlitHelper(); ~BlitHelper();
bool BlitDepthStencil(Surface& source, Surface& dest, const VideoCore::TextureBlit& blit); bool BlitDepthStencil(Surface& source, Surface& dest, const VideoCore::TextureBlit& blit);
@ -42,13 +43,14 @@ private:
const Instance& instance; const Instance& instance;
Scheduler& scheduler; Scheduler& scheduler;
RenderManager& render_manager; RenderManager& render_manager;
DescriptorUpdateQueue& update_queue;
vk::Device device; vk::Device device;
vk::RenderPass r32_renderpass; vk::RenderPass r32_renderpass;
DescriptorSetProvider compute_provider; DescriptorHeap compute_provider;
DescriptorSetProvider compute_buffer_provider; DescriptorHeap compute_buffer_provider;
DescriptorSetProvider two_textures_provider; DescriptorHeap two_textures_provider;
vk::PipelineLayout compute_pipeline_layout; vk::PipelineLayout compute_pipeline_layout;
vk::PipelineLayout compute_buffer_pipeline_layout; vk::PipelineLayout compute_buffer_pipeline_layout;
vk::PipelineLayout two_textures_pipeline_layout; vk::PipelineLayout two_textures_pipeline_layout;

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

@ -9,7 +9,6 @@
#define VK_NO_PROTOTYPES #define VK_NO_PROTOTYPES
#define VULKAN_HPP_DISPATCH_LOADER_DYNAMIC 1 #define VULKAN_HPP_DISPATCH_LOADER_DYNAMIC 1
#define VULKAN_HPP_NO_CONSTRUCTORS #define VULKAN_HPP_NO_CONSTRUCTORS
#define VULKAN_HPP_NO_UNION_CONSTRUCTORS
#define VULKAN_HPP_NO_STRUCT_SETTERS #define VULKAN_HPP_NO_STRUCT_SETTERS
#include <vulkan/vulkan.hpp> #include <vulkan/vulkan.hpp>

141
src/video_core/renderer_vulkan/vk_descriptor_pool.cpp
View file

@ -1,141 +0,0 @@
// Copyright 2023 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/microprofile.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_instance.h"
namespace Vulkan {
MICROPROFILE_DEFINE(Vulkan_DescriptorSetAcquire, "Vulkan", "Descriptor Set Acquire",
MP_RGB(64, 128, 256));
constexpr u32 MAX_BATCH_SIZE = 8;
DescriptorPool::DescriptorPool(const Instance& instance_) : instance{instance_} {
auto& pool = pools.emplace_back();
pool = CreatePool();
}
DescriptorPool::~DescriptorPool() = default;
std::vector<vk::DescriptorSet> DescriptorPool::Allocate(vk::DescriptorSetLayout layout,
u32 num_sets) {
std::array<vk::DescriptorSetLayout, MAX_BATCH_SIZE> layouts;
layouts.fill(layout);
u32 current_pool = 0;
vk::DescriptorSetAllocateInfo alloc_info = {
.descriptorPool = *pools[current_pool],
.descriptorSetCount = num_sets,
.pSetLayouts = layouts.data(),
};
while (true) {
try {
return instance.GetDevice().allocateDescriptorSets(alloc_info);
} catch (const vk::OutOfPoolMemoryError&) {
current_pool++;
if (current_pool == pools.size()) {
LOG_INFO(Render_Vulkan, "Run out of pools, creating new one!");
auto& pool = pools.emplace_back();
pool = CreatePool();
}
alloc_info.descriptorPool = *pools[current_pool];
}
}
}
vk::DescriptorSet DescriptorPool::Allocate(vk::DescriptorSetLayout layout) {
const auto sets = Allocate(layout, 1);
return sets[0];
}
vk::UniqueDescriptorPool DescriptorPool::CreatePool() {
// Choose a sane pool size good for most games
static constexpr std::array<vk::DescriptorPoolSize, 6> pool_sizes = {{
{vk::DescriptorType::eUniformBufferDynamic, 64},
{vk::DescriptorType::eUniformTexelBuffer, 64},
{vk::DescriptorType::eCombinedImageSampler, 4096},
{vk::DescriptorType::eSampledImage, 256},
{vk::DescriptorType::eStorageImage, 256},
{vk::DescriptorType::eStorageBuffer, 32},
}};
const vk::DescriptorPoolCreateInfo descriptor_pool_info = {
.maxSets = 4098,
.poolSizeCount = static_cast<u32>(pool_sizes.size()),
.pPoolSizes = pool_sizes.data(),
};
return instance.GetDevice().createDescriptorPoolUnique(descriptor_pool_info);
}
DescriptorSetProvider::DescriptorSetProvider(
const Instance& instance, DescriptorPool& pool_,
std::span<const vk::DescriptorSetLayoutBinding> bindings)
: pool{pool_}, device{instance.GetDevice()} {
std::array<vk::DescriptorUpdateTemplateEntry, MAX_DESCRIPTORS> update_entries;
for (u32 i = 0; i < bindings.size(); i++) {
update_entries[i] = vk::DescriptorUpdateTemplateEntry{
.dstBinding = bindings[i].binding,
.dstArrayElement = 0,
.descriptorCount = bindings[i].descriptorCount,
.descriptorType = bindings[i].descriptorType,
.offset = i * sizeof(DescriptorData),
.stride = sizeof(DescriptorData),
};
}
const vk::DescriptorSetLayoutCreateInfo layout_info = {
.bindingCount = static_cast<u32>(bindings.size()),
.pBindings = bindings.data(),
};
layout = device.createDescriptorSetLayoutUnique(layout_info);
const vk::DescriptorUpdateTemplateCreateInfo template_info = {
.descriptorUpdateEntryCount = static_cast<u32>(bindings.size()),
.pDescriptorUpdateEntries = update_entries.data(),
.templateType = vk::DescriptorUpdateTemplateType::eDescriptorSet,
.descriptorSetLayout = *layout,
};
update_template = device.createDescriptorUpdateTemplateUnique(template_info);
}
DescriptorSetProvider::~DescriptorSetProvider() = default;
vk::DescriptorSet DescriptorSetProvider::Acquire(std::span<const DescriptorData> data) {
MICROPROFILE_SCOPE(Vulkan_DescriptorSetAcquire);
DescriptorSetData key{};
std::memcpy(key.data(), data.data(), data.size_bytes());
const auto [it, new_set] = descriptor_set_map.try_emplace(key);
if (!new_set) {
return it->second;
}
if (free_sets.empty()) {
free_sets = pool.Allocate(*layout, MAX_BATCH_SIZE);
}
it.value() = free_sets.back();
free_sets.pop_back();
device.updateDescriptorSetWithTemplate(it->second, *update_template, data[0]);
return it->second;
}
void DescriptorSetProvider::FreeWithImage(vk::ImageView image_view) {
for (auto it = descriptor_set_map.begin(); it != descriptor_set_map.end();) {
const auto& [data, set] = *it;
const bool has_image = std::any_of(data.begin(), data.end(), [image_view](auto& info) {
return info.image_info.imageView == image_view;
});
if (has_image) {
free_sets.push_back(set);
it = descriptor_set_map.erase(it);
} else {
it++;
}
}
}
} // namespace Vulkan

92
src/video_core/renderer_vulkan/vk_descriptor_pool.h
View file

@ -1,92 +0,0 @@
// Copyright 2023 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <span>
#include <vector>
#include <tsl/robin_map.h>
#include "common/hash.h"
#include "video_core/renderer_vulkan/vk_common.h"
namespace Vulkan {
class Instance;
constexpr u32 MAX_DESCRIPTORS = 7;
union DescriptorData {
vk::DescriptorImageInfo image_info;
vk::DescriptorBufferInfo buffer_info;
vk::BufferView buffer_view;
bool operator==(const DescriptorData& other) const noexcept {
return std::memcmp(this, &other, sizeof(DescriptorData)) == 0;
}
};
using DescriptorSetData = std::array<DescriptorData, MAX_DESCRIPTORS>;
struct DataHasher {
u64 operator()(const DescriptorSetData& data) const noexcept {
return Common::ComputeHash64(data.data(), sizeof(data));
}
};
/**
* An interface for allocating descriptor sets that manages a collection of descriptor pools.
*/
class DescriptorPool {
public:
explicit DescriptorPool(const Instance& instance);
~DescriptorPool();
std::vector<vk::DescriptorSet> Allocate(vk::DescriptorSetLayout layout, u32 num_sets);
vk::DescriptorSet Allocate(vk::DescriptorSetLayout layout);
private:
vk::UniqueDescriptorPool CreatePool();
private:
const Instance& instance;
std::vector<vk::UniqueDescriptorPool> pools;
};
/**
* Allocates and caches descriptor sets of a specific layout.
*/
class DescriptorSetProvider {
public:
explicit DescriptorSetProvider(const Instance& instance, DescriptorPool& pool,
std::span<const vk::DescriptorSetLayoutBinding> bindings);
~DescriptorSetProvider();
vk::DescriptorSet Acquire(std::span<const DescriptorData> data);
void FreeWithImage(vk::ImageView image_view);
[[nodiscard]] vk::DescriptorSetLayout Layout() const noexcept {
return *layout;
}
[[nodiscard]] vk::DescriptorSetLayout& Layout() noexcept {
return layout.get();
}
[[nodiscard]] vk::DescriptorUpdateTemplate UpdateTemplate() const noexcept {
return *update_template;
}
private:
DescriptorPool& pool;
vk::Device device;
vk::UniqueDescriptorSetLayout layout;
vk::UniqueDescriptorUpdateTemplate update_template;
std::vector<vk::DescriptorSet> free_sets;
tsl::robin_map<DescriptorSetData, vk::DescriptorSet, DataHasher> descriptor_set_map;
};
} // namespace Vulkan

109
src/video_core/renderer_vulkan/vk_descriptor_update_queue.cpp Normal file
View file

@ -0,0 +1,109 @@
// Copyright 2024 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "video_core/renderer_vulkan/vk_descriptor_update_queue.h"
#include "video_core/renderer_vulkan/vk_instance.h"
namespace Vulkan {
DescriptorUpdateQueue::DescriptorUpdateQueue(const Instance& instance, u32 descriptor_write_max_)
: device{instance.GetDevice()}, descriptor_write_max{descriptor_write_max_} {
descriptor_infos = std::make_unique<DescriptorInfoUnion[]>(descriptor_write_max);
descriptor_writes = std::make_unique<vk::WriteDescriptorSet[]>(descriptor_write_max);
}
void DescriptorUpdateQueue::Flush() {
if (descriptor_write_end == 0) {
return;
}
device.updateDescriptorSets({std::span(descriptor_writes.get(), descriptor_write_end)}, {});
descriptor_write_end = 0;
}
void DescriptorUpdateQueue::AddStorageImage(vk::DescriptorSet target, u8 binding,
vk::ImageView image_view,
vk::ImageLayout image_layout) {
if (descriptor_write_end >= descriptor_write_max) [[unlikely]] {
Flush();
}
auto& image_info = descriptor_infos[descriptor_write_end].image_info;
image_info.sampler = VK_NULL_HANDLE;
image_info.imageView = image_view;
image_info.imageLayout = image_layout;
descriptor_writes[descriptor_write_end++] = vk::WriteDescriptorSet{
.dstSet = target,
.dstBinding = binding,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = vk::DescriptorType::eStorageImage,
.pImageInfo = &image_info,
};
}
void DescriptorUpdateQueue::AddImageSampler(vk::DescriptorSet target, u8 binding, u8 array_index,
vk::ImageView image_view, vk::Sampler sampler,
vk::ImageLayout image_layout) {
if (descriptor_write_end >= descriptor_write_max) [[unlikely]] {
Flush();
}
auto& image_info = descriptor_infos[descriptor_write_end].image_info;
image_info.sampler = sampler;
image_info.imageView = image_view;
image_info.imageLayout = image_layout;
descriptor_writes[descriptor_write_end++] = vk::WriteDescriptorSet{
.dstSet = target,
.dstBinding = binding,
.dstArrayElement = array_index,
.descriptorCount = 1,
.descriptorType =
sampler ? vk::DescriptorType::eCombinedImageSampler : vk::DescriptorType::eSampledImage,
.pImageInfo = &image_info,
};
}
void DescriptorUpdateQueue::AddBuffer(vk::DescriptorSet target, u8 binding, vk::Buffer buffer,
vk::DeviceSize offset, vk::DeviceSize size,
vk::DescriptorType type) {
if (descriptor_write_end >= descriptor_write_max) [[unlikely]] {
Flush();
}
auto& buffer_info = descriptor_infos[descriptor_write_end].buffer_info;
buffer_info.buffer = buffer;
buffer_info.offset = offset;
buffer_info.range = size;
descriptor_writes[descriptor_write_end++] = vk::WriteDescriptorSet{
.dstSet = target,
.dstBinding = binding,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = type,
.pBufferInfo = &buffer_info,
};
}
void DescriptorUpdateQueue::AddTexelBuffer(vk::DescriptorSet target, u8 binding,
vk::BufferView buffer_view) {
if (descriptor_write_end >= descriptor_write_max) [[unlikely]] {
Flush();
}
auto& buffer_info = descriptor_infos[descriptor_write_end].buffer_view;
buffer_info = buffer_view;
descriptor_writes[descriptor_write_end++] = vk::WriteDescriptorSet{
.dstSet = target,
.dstBinding = binding,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = vk::DescriptorType::eUniformTexelBuffer,
.pTexelBufferView = &buffer_info,
};
}
} // namespace Vulkan

53
src/video_core/renderer_vulkan/vk_descriptor_update_queue.h Normal file
View file

@ -0,0 +1,53 @@
// Copyright 2024 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <memory>
#include <variant>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/vk_common.h"
namespace Vulkan {
class Instance;
struct DescriptorInfoUnion {
DescriptorInfoUnion() {}
union {
vk::DescriptorImageInfo image_info;
vk::DescriptorBufferInfo buffer_info;
vk::BufferView buffer_view;
};
};
class DescriptorUpdateQueue {
public:
explicit DescriptorUpdateQueue(const Instance& instance, u32 descriptor_write_max = 2048);
~DescriptorUpdateQueue() = default;
void Flush();
void AddStorageImage(vk::DescriptorSet target, u8 binding, vk::ImageView image_view,
vk::ImageLayout image_layout = vk::ImageLayout::eGeneral);
void AddImageSampler(vk::DescriptorSet target, u8 binding, u8 array_index,
vk::ImageView image_view, vk::Sampler sampler,
vk::ImageLayout imageLayout = vk::ImageLayout::eGeneral);
void AddBuffer(vk::DescriptorSet target, u8 binding, vk::Buffer buffer, vk::DeviceSize offset,
vk::DeviceSize size = VK_WHOLE_SIZE,
vk::DescriptorType type = vk::DescriptorType::eUniformBufferDynamic);
void AddTexelBuffer(vk::DescriptorSet target, u8 binding, vk::BufferView buffer_view);
private:
const vk::Device device;
const u32 descriptor_write_max;
std::unique_ptr<DescriptorInfoUnion[]> descriptor_infos;
std::unique_ptr<vk::WriteDescriptorSet[]> descriptor_writes;
u32 descriptor_write_end = 0;
};
} // namespace Vulkan

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

@ -126,7 +126,7 @@ struct AttachmentInfo {
}; };
/** /**
* Information about a graphics/compute pipeline * Information about a graphics pipeline
*/ */
struct PipelineInfo { struct PipelineInfo {
BlendingState blending; BlendingState blending;

155
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
View file

@ -11,6 +11,7 @@
#include "common/scope_exit.h" #include "common/scope_exit.h"
#include "common/settings.h" #include "common/settings.h"
#include "video_core/renderer_vulkan/pica_to_vk.h" #include "video_core/renderer_vulkan/pica_to_vk.h"
#include "video_core/renderer_vulkan/vk_descriptor_update_queue.h"
#include "video_core/renderer_vulkan/vk_instance.h" #include "video_core/renderer_vulkan/vk_instance.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h" #include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_render_manager.h" #include "video_core/renderer_vulkan/vk_render_manager.h"
@ -62,34 +63,34 @@ constexpr std::array<vk::DescriptorSetLayoutBinding, 6> BUFFER_BINDINGS = {{
{5, vk::DescriptorType::eUniformTexelBuffer, 1, vk::ShaderStageFlagBits::eFragment}, {5, vk::DescriptorType::eUniformTexelBuffer, 1, vk::ShaderStageFlagBits::eFragment},
}}; }};
template <u32 NumTex0>
constexpr std::array<vk::DescriptorSetLayoutBinding, 3> TEXTURE_BINDINGS = {{ constexpr std::array<vk::DescriptorSetLayoutBinding, 3> TEXTURE_BINDINGS = {{
{0, vk::DescriptorType::eCombinedImageSampler, 1, vk::ShaderStageFlagBits::eFragment}, {0, vk::DescriptorType::eCombinedImageSampler, NumTex0,
{1, vk::DescriptorType::eCombinedImageSampler, 1, vk::ShaderStageFlagBits::eFragment}, vk::ShaderStageFlagBits::eFragment}, // tex0
{2, vk::DescriptorType::eCombinedImageSampler, 1, vk::ShaderStageFlagBits::eFragment}, {1, vk::DescriptorType::eCombinedImageSampler, 1, vk::ShaderStageFlagBits::eFragment}, // tex1
{2, vk::DescriptorType::eCombinedImageSampler, 1, vk::ShaderStageFlagBits::eFragment}, // tex2
}}; }};
// TODO: Use descriptor array for shadow cube constexpr std::array<vk::DescriptorSetLayoutBinding, 2> UTILITY_BINDINGS = {{
constexpr std::array<vk::DescriptorSetLayoutBinding, 7> SHADOW_BINDINGS = {{ {0, vk::DescriptorType::eStorageImage, 1, vk::ShaderStageFlagBits::eFragment}, // shadow_buffer
{0, vk::DescriptorType::eStorageImage, 1, vk::ShaderStageFlagBits::eFragment}, {1, vk::DescriptorType::eCombinedImageSampler, 1,
{1, vk::DescriptorType::eStorageImage, 1, vk::ShaderStageFlagBits::eFragment}, vk::ShaderStageFlagBits::eFragment}, // tex_normal
{2, vk::DescriptorType::eStorageImage, 1, vk::ShaderStageFlagBits::eFragment},
{3, vk::DescriptorType::eStorageImage, 1, vk::ShaderStageFlagBits::eFragment},
{4, vk::DescriptorType::eStorageImage, 1, vk::ShaderStageFlagBits::eFragment},
{5, vk::DescriptorType::eStorageImage, 1, vk::ShaderStageFlagBits::eFragment},
{6, vk::DescriptorType::eStorageImage, 1, vk::ShaderStageFlagBits::eFragment},
}}; }};
PipelineCache::PipelineCache(const Instance& instance_, Scheduler& scheduler_, PipelineCache::PipelineCache(const Instance& instance_, Scheduler& scheduler_,
RenderManager& render_manager_, DescriptorPool& pool_) RenderManager& render_manager_, DescriptorUpdateQueue& update_queue_)
: instance{instance_}, scheduler{scheduler_}, render_manager{render_manager_}, pool{pool_}, : instance{instance_}, scheduler{scheduler_}, render_manager{render_manager_},
num_worker_threads{std::max(std::thread::hardware_concurrency(), 2U)}, update_queue{update_queue_},
num_worker_threads{std::max(std::thread::hardware_concurrency(), 2U) >> 1},
workers{num_worker_threads, "Pipeline workers"}, workers{num_worker_threads, "Pipeline workers"},
descriptor_set_providers{DescriptorSetProvider{instance, pool, BUFFER_BINDINGS}, descriptor_heaps{
DescriptorSetProvider{instance, pool, TEXTURE_BINDINGS}, DescriptorHeap{instance, scheduler.GetMasterSemaphore(), BUFFER_BINDINGS, 32},
DescriptorSetProvider{instance, pool, SHADOW_BINDINGS}}, DescriptorHeap{instance, scheduler.GetMasterSemaphore(), TEXTURE_BINDINGS<1>},
DescriptorHeap{instance, scheduler.GetMasterSemaphore(), UTILITY_BINDINGS, 32}},
trivial_vertex_shader{ trivial_vertex_shader{
instance, vk::ShaderStageFlagBits::eVertex, instance, vk::ShaderStageFlagBits::eVertex,
GLSL::GenerateTrivialVertexShader(instance.IsShaderClipDistanceSupported(), true)} { GLSL::GenerateTrivialVertexShader(instance.IsShaderClipDistanceSupported(), true)} {
scheduler.RegisterOnDispatch([this] { update_queue.Flush(); });
profile = Pica::Shader::Profile{ profile = Pica::Shader::Profile{
.has_separable_shaders = true, .has_separable_shaders = true,
.has_clip_planes = instance.IsShaderClipDistanceSupported(), .has_clip_planes = instance.IsShaderClipDistanceSupported(),
@ -106,13 +107,13 @@ PipelineCache::PipelineCache(const Instance& instance_, Scheduler& scheduler_,
} }
void PipelineCache::BuildLayout() { void PipelineCache::BuildLayout() {
std::array<vk::DescriptorSetLayout, NUM_RASTERIZER_SETS> descriptor_set_layouts; std::array<vk::DescriptorSetLayout, NumRasterizerSets> descriptor_set_layouts;
std::transform(descriptor_set_providers.begin(), descriptor_set_providers.end(), descriptor_set_layouts[0] = descriptor_heaps[0].Layout();
descriptor_set_layouts.begin(), descriptor_set_layouts[1] = descriptor_heaps[1].Layout();
[](const auto& provider) { return provider.Layout(); }); descriptor_set_layouts[2] = descriptor_heaps[2].Layout();
const vk::PipelineLayoutCreateInfo layout_info = { const vk::PipelineLayoutCreateInfo layout_info = {
.setLayoutCount = NUM_RASTERIZER_SETS, .setLayoutCount = NumRasterizerSets,
.pSetLayouts = descriptor_set_layouts.data(), .pSetLayouts = descriptor_set_layouts.data(),
.pushConstantRangeCount = 0, .pushConstantRangeCount = 0,
.pPushConstantRanges = nullptr, .pPushConstantRanges = nullptr,
@ -214,55 +215,11 @@ bool PipelineCache::BindPipeline(const PipelineInfo& info, bool wait_built) {
return false; return false;
} }
u32 new_descriptors_start = 0;
std::span<vk::DescriptorSet> new_descriptors_span{};
std::span<u32> new_offsets_span{};
// Ensure all the descriptor sets are set at least once at the beginning.
if (scheduler.IsStateDirty(StateFlags::DescriptorSets)) {
set_dirty.set();
}
if (set_dirty.any()) {
for (u32 i = 0; i < NUM_RASTERIZER_SETS; i++) {
if (!set_dirty.test(i)) {
continue;
}
bound_descriptor_sets[i] = descriptor_set_providers[i].Acquire(update_data[i]);
}
new_descriptors_span = bound_descriptor_sets;
// Only send new offsets if the buffer descriptor-set changed.
if (set_dirty.test(0)) {
new_offsets_span = offsets;
}
// Try to compact the number of updated descriptor-set slots to the ones that have actually
// changed
if (!set_dirty.all()) {
const u64 dirty_mask = set_dirty.to_ulong();
new_descriptors_start = static_cast<u32>(std::countr_zero(dirty_mask));
const u32 new_descriptors_end = 64u - static_cast<u32>(std::countl_zero(dirty_mask));
const u32 new_descriptors_size = new_descriptors_end - new_descriptors_start;
new_descriptors_span =
new_descriptors_span.subspan(new_descriptors_start, new_descriptors_size);
}
set_dirty.reset();
}
boost::container::static_vector<vk::DescriptorSet, NUM_RASTERIZER_SETS> new_descriptors(
new_descriptors_span.begin(), new_descriptors_span.end());
boost::container::static_vector<u32, NUM_DYNAMIC_OFFSETS> new_offsets(new_offsets_span.begin(),
new_offsets_span.end());
const bool is_dirty = scheduler.IsStateDirty(StateFlags::Pipeline); const bool is_dirty = scheduler.IsStateDirty(StateFlags::Pipeline);
const bool pipeline_dirty = (current_pipeline != pipeline) || is_dirty; const bool pipeline_dirty = (current_pipeline != pipeline) || is_dirty;
scheduler.Record([this, is_dirty, pipeline_dirty, pipeline, scheduler.Record([this, is_dirty, pipeline_dirty, pipeline,
current_dynamic = current_info.dynamic, dynamic = info.dynamic, current_dynamic = current_info.dynamic, dynamic = info.dynamic,
new_descriptors_start, descriptor_sets = std::move(new_descriptors), descriptor_sets = bound_descriptor_sets, offsets = offsets,
offsets = std::move(new_offsets),
current_rasterization = current_info.rasterization, current_rasterization = current_info.rasterization,
current_depth_stencil = current_info.depth_stencil, current_depth_stencil = current_info.depth_stencil,
rasterization = info.rasterization, rasterization = info.rasterization,
@ -364,10 +321,8 @@ bool PipelineCache::BindPipeline(const PipelineInfo& info, bool wait_built) {
cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, pipeline->Handle()); cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, pipeline->Handle());
} }
if (descriptor_sets.size()) { cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eGraphics, *pipeline_layout, 0,
cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eGraphics, *pipeline_layout, descriptor_sets, offsets);
new_descriptors_start, descriptor_sets, offsets);
}
}); });
current_info = info; current_info = info;
@ -385,7 +340,6 @@ bool PipelineCache::UseProgrammableVertexShader(const Pica::RegsInternal& regs,
// We also don't need the geometry shader if we have the barycentric extension. // We also don't need the geometry shader if we have the barycentric extension.
const bool use_geometry_shader = instance.UseGeometryShaders() && !regs.lighting.disable && const bool use_geometry_shader = instance.UseGeometryShaders() && !regs.lighting.disable &&
!instance.IsFragmentShaderBarycentricSupported(); !instance.IsFragmentShaderBarycentricSupported();
PicaVSConfig config{regs, setup, instance.IsShaderClipDistanceSupported(), use_geometry_shader}; PicaVSConfig config{regs, setup, instance.IsShaderClipDistanceSupported(), use_geometry_shader};
for (u32 i = 0; i < layout.attribute_count; i++) { for (u32 i = 0; i < layout.attribute_count; i++) {
@ -402,7 +356,7 @@ bool PipelineCache::UseProgrammableVertexShader(const Pica::RegsInternal& regs,
} }
} }
auto [it, new_config] = programmable_vertex_map.try_emplace(config); const auto [it, new_config] = programmable_vertex_map.try_emplace(config);
if (new_config) { if (new_config) {
auto program = GLSL::GenerateVertexShader(setup, config, true); auto program = GLSL::GenerateVertexShader(setup, config, true);
if (program.empty()) { if (program.empty()) {
@ -497,59 +451,6 @@ void PipelineCache::UseFragmentShader(const Pica::RegsInternal& regs,
shader_hashes[ProgramType::FS] = fs_config.Hash(); shader_hashes[ProgramType::FS] = fs_config.Hash();
} }
void PipelineCache::BindTexture(u32 binding, vk::ImageView image_view, vk::Sampler sampler) {
auto& info = update_data[1][binding].image_info;
if (info.imageView == image_view && info.sampler == sampler) {
return;
}
set_dirty[1] = true;
info = vk::DescriptorImageInfo{
.sampler = sampler,
.imageView = image_view,
.imageLayout = vk::ImageLayout::eGeneral,
};
}
void PipelineCache::BindStorageImage(u32 binding, vk::ImageView image_view) {
auto& info = update_data[2][binding].image_info;
if (info.imageView == image_view) {
return;
}
set_dirty[2] = true;
info = vk::DescriptorImageInfo{
.imageView = image_view,
.imageLayout = vk::ImageLayout::eGeneral,
};
}
void PipelineCache::BindBuffer(u32 binding, vk::Buffer buffer, u32 offset, u32 size) {
auto& info = update_data[0][binding].buffer_info;
if (info.buffer == buffer && info.offset == offset && info.range == size) {
return;
}
set_dirty[0] = true;
info = vk::DescriptorBufferInfo{
.buffer = buffer,
.offset = offset,
.range = size,
};
}
void PipelineCache::BindTexelBuffer(u32 binding, vk::BufferView buffer_view) {
auto& view = update_data[0][binding].buffer_view;
if (view != buffer_view) {
set_dirty[0] = true;
view = buffer_view;
}
}
void PipelineCache::SetBufferOffset(u32 binding, std::size_t offset) {
if (offsets[binding] != static_cast<u32>(offset)) {
offsets[binding] = static_cast<u32>(offset);
set_dirty[0] = true;
}
}
bool PipelineCache::IsCacheValid(std::span<const u8> data) const { bool PipelineCache::IsCacheValid(std::span<const u8> data) const {
if (data.size() < sizeof(vk::PipelineCacheHeaderVersionOne)) { if (data.size() < sizeof(vk::PipelineCacheHeaderVersionOne)) {
LOG_ERROR(Render_Vulkan, "Pipeline cache failed validation: Invalid header"); LOG_ERROR(Render_Vulkan, "Pipeline cache failed validation: Invalid header");

55
src/video_core/renderer_vulkan/vk_pipeline_cache.h
View file

@ -7,8 +7,8 @@
#include <bitset> #include <bitset>
#include <tsl/robin_map.h> #include <tsl/robin_map.h>
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_graphics_pipeline.h" #include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
#include "video_core/renderer_vulkan/vk_resource_pool.h"
#include "video_core/shader/generator/pica_fs_config.h" #include "video_core/shader/generator/pica_fs_config.h"
#include "video_core/shader/generator/profile.h" #include "video_core/shader/generator/profile.h"
#include "video_core/shader/generator/shader_gen.h" #include "video_core/shader/generator/shader_gen.h"
@ -23,22 +23,38 @@ namespace Vulkan {
class Instance; class Instance;
class Scheduler; class Scheduler;
class RenderManager; class RenderManager;
class DescriptorPool; class DescriptorUpdateQueue;
constexpr u32 NUM_RASTERIZER_SETS = 3; enum class DescriptorHeapType : u32 {
constexpr u32 NUM_DYNAMIC_OFFSETS = 3; Buffer,
Texture,
Utility,
};
/** /**
* Stores a collection of rasterizer pipelines used during rendering. * Stores a collection of rasterizer pipelines used during rendering.
*/ */
class PipelineCache { class PipelineCache {
static constexpr u32 NumRasterizerSets = 3;
static constexpr u32 NumDescriptorHeaps = 3;
static constexpr u32 NumDynamicOffsets = 3;
public: public:
explicit PipelineCache(const Instance& instance, Scheduler& scheduler, explicit PipelineCache(const Instance& instance, Scheduler& scheduler,
RenderManager& render_manager, DescriptorPool& pool); RenderManager& render_manager, DescriptorUpdateQueue& update_queue);
~PipelineCache(); ~PipelineCache();
[[nodiscard]] DescriptorSetProvider& TextureProvider() noexcept { /// Acquires and binds a free descriptor set from the appropriate heap.
return descriptor_set_providers[1]; vk::DescriptorSet Acquire(DescriptorHeapType type) {
const u32 index = static_cast<u32>(type);
const auto descriptor_set = descriptor_heaps[index].Commit();
bound_descriptor_sets[index] = descriptor_set;
return descriptor_set;
}
/// Sets the dynamic offset for the uniform buffer at binding
void UpdateRange(u8 binding, u32 offset) {
offsets[binding] = offset;
} }
/// Loads the pipeline cache stored to disk /// Loads the pipeline cache stored to disk
@ -66,21 +82,6 @@ public:
/// Binds a fragment shader generated from PICA state /// Binds a fragment shader generated from PICA state
void UseFragmentShader(const Pica::RegsInternal& regs, const Pica::Shader::UserConfig& user); void UseFragmentShader(const Pica::RegsInternal& regs, const Pica::Shader::UserConfig& user);
/// Binds a texture to the specified binding
void BindTexture(u32 binding, vk::ImageView image_view, vk::Sampler sampler);
/// Binds a storage image to the specified binding
void BindStorageImage(u32 binding, vk::ImageView image_view);
/// Binds a buffer to the specified binding
void BindBuffer(u32 binding, vk::Buffer buffer, u32 offset, u32 size);
/// Binds a buffer to the specified binding
void BindTexelBuffer(u32 binding, vk::BufferView buffer_view);
/// Sets the dynamic offset for the uniform buffer at binding
void SetBufferOffset(u32 binding, std::size_t offset);
private: private:
/// Builds the rasterizer pipeline layout /// Builds the rasterizer pipeline layout
void BuildLayout(); void BuildLayout();
@ -98,7 +99,7 @@ private:
const Instance& instance; const Instance& instance;
Scheduler& scheduler; Scheduler& scheduler;
RenderManager& render_manager; RenderManager& render_manager;
DescriptorPool& pool; DescriptorUpdateQueue& update_queue;
Pica::Shader::Profile profile{}; Pica::Shader::Profile profile{};
vk::UniquePipelineCache pipeline_cache; vk::UniquePipelineCache pipeline_cache;
@ -110,11 +111,9 @@ private:
tsl::robin_map<u64, std::unique_ptr<GraphicsPipeline>, Common::IdentityHash<u64>> tsl::robin_map<u64, std::unique_ptr<GraphicsPipeline>, Common::IdentityHash<u64>>
graphics_pipelines; graphics_pipelines;
std::array<DescriptorSetProvider, NUM_RASTERIZER_SETS> descriptor_set_providers; std::array<DescriptorHeap, NumDescriptorHeaps> descriptor_heaps;
std::array<DescriptorSetData, NUM_RASTERIZER_SETS> update_data{}; std::array<vk::DescriptorSet, NumRasterizerSets> bound_descriptor_sets{};
std::array<vk::DescriptorSet, NUM_RASTERIZER_SETS> bound_descriptor_sets{}; std::array<u32, NumDynamicOffsets> offsets{};
std::array<u32, NUM_DYNAMIC_OFFSETS> offsets{};
std::bitset<NUM_RASTERIZER_SETS> set_dirty{};
std::array<u64, MAX_SHADER_STAGES> shader_hashes; std::array<u64, MAX_SHADER_STAGES> shader_hashes;
std::array<Shader*, MAX_SHADER_STAGES> current_shaders; std::array<Shader*, MAX_SHADER_STAGES> current_shaders;

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

@ -31,8 +31,9 @@ static VKAPI_ATTR VkBool32 VKAPI_CALL DebugUtilsCallback(
VkDebugUtilsMessageSeverityFlagBitsEXT severity, VkDebugUtilsMessageTypeFlagsEXT type, VkDebugUtilsMessageSeverityFlagBitsEXT severity, VkDebugUtilsMessageTypeFlagsEXT type,
const VkDebugUtilsMessengerCallbackDataEXT* callback_data, void* user_data) { const VkDebugUtilsMessengerCallbackDataEXT* callback_data, void* user_data) {
switch (callback_data->messageIdNumber) { switch (static_cast<u32>(callback_data->messageIdNumber)) {
case 0x609a13b: // Vertex attribute at location not consumed by shader case 0x609a13b: // Vertex attribute at location not consumed by shader
case 0xc81ad50e:
return VK_FALSE; return VK_FALSE;
default: default:
break; break;

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

@ -138,11 +138,11 @@ PresentWindow::PresentWindow(Frontend::EmuWindow& emu_window_, const Instance& i
if (instance.HasDebuggingToolAttached()) { if (instance.HasDebuggingToolAttached()) {
for (u32 i = 0; i < num_images; ++i) { for (u32 i = 0; i < num_images; ++i) {
Vulkan::SetObjectName(device, swap_chain[i].cmdbuf, "Swapchain Command Buffer {}", i); SetObjectName(device, swap_chain[i].cmdbuf, "Swapchain Command Buffer {}", i);
Vulkan::SetObjectName(device, swap_chain[i].render_ready, SetObjectName(device, swap_chain[i].render_ready,
"Swapchain Semaphore: render_ready {}", i); "Swapchain Semaphore: render_ready {}", i);
Vulkan::SetObjectName(device, swap_chain[i].present_done, SetObjectName(device, swap_chain[i].present_done, "Swapchain Fence: present_done {}",
"Swapchain Fence: present_done {}", i); i);
} }
} }

147
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View file

@ -58,13 +58,15 @@ RasterizerVulkan::RasterizerVulkan(Memory::MemorySystem& memory, Pica::PicaCore&
VideoCore::CustomTexManager& custom_tex_manager, VideoCore::CustomTexManager& custom_tex_manager,
VideoCore::RendererBase& renderer, VideoCore::RendererBase& renderer,
Frontend::EmuWindow& emu_window, const Instance& instance, Frontend::EmuWindow& emu_window, const Instance& instance,
Scheduler& scheduler, DescriptorPool& pool, Scheduler& scheduler, RenderManager& render_manager,
RenderManager& render_manager, u32 image_count) DescriptorUpdateQueue& update_queue_, u32 image_count)
: RasterizerAccelerated{memory, pica}, instance{instance}, scheduler{scheduler}, : RasterizerAccelerated{memory, pica}, instance{instance}, scheduler{scheduler},
render_manager{render_manager}, pipeline_cache{instance, scheduler, render_manager, render_manager{render_manager}, update_queue{update_queue_},
pool}, pipeline_cache{instance, scheduler, render_manager, update_queue}, runtime{instance,
runtime{instance, scheduler, render_manager, pool, pipeline_cache.TextureProvider(), scheduler,
image_count}, render_manager,
update_queue,
image_count},
res_cache{memory, custom_tex_manager, runtime, regs, renderer}, res_cache{memory, custom_tex_manager, runtime, regs, renderer},
stream_buffer{instance, scheduler, BUFFER_USAGE, STREAM_BUFFER_SIZE}, stream_buffer{instance, scheduler, BUFFER_USAGE, STREAM_BUFFER_SIZE},
uniform_buffer{instance, scheduler, vk::BufferUsageFlagBits::eUniformBuffer, uniform_buffer{instance, scheduler, vk::BufferUsageFlagBits::eUniformBuffer,
@ -80,9 +82,9 @@ RasterizerVulkan::RasterizerVulkan(Memory::MemorySystem& memory, Pica::PicaCore&
// Query uniform buffer alignment. // Query uniform buffer alignment.
uniform_buffer_alignment = instance.UniformMinAlignment(); uniform_buffer_alignment = instance.UniformMinAlignment();
uniform_size_aligned_vs_pica = uniform_size_aligned_vs_pica =
Common::AlignUp(sizeof(VSPicaUniformData), uniform_buffer_alignment); Common::AlignUp<u32>(sizeof(VSPicaUniformData), uniform_buffer_alignment);
uniform_size_aligned_vs = Common::AlignUp(sizeof(VSUniformData), uniform_buffer_alignment); uniform_size_aligned_vs = Common::AlignUp<u32>(sizeof(VSUniformData), uniform_buffer_alignment);
uniform_size_aligned_fs = Common::AlignUp(sizeof(FSUniformData), uniform_buffer_alignment); uniform_size_aligned_fs = Common::AlignUp<u32>(sizeof(FSUniformData), uniform_buffer_alignment);
// Define vertex layout for software shaders // Define vertex layout for software shaders
MakeSoftwareVertexLayout(); MakeSoftwareVertexLayout();
@ -108,28 +110,32 @@ RasterizerVulkan::RasterizerVulkan(Memory::MemorySystem& memory, Pica::PicaCore&
.range = VK_WHOLE_SIZE, .range = VK_WHOLE_SIZE,
}); });
scheduler.RegisterOnSubmit([&render_manager] { scheduler.RegisterOnSubmit([&render_manager] { render_manager.EndRendering(); });
render_manager.EndRendering();
});
// Since we don't have access to VK_EXT_descriptor_indexing we need to intiallize // Prepare the static buffer descriptor set.
// all descriptor sets even the ones we don't use. const auto buffer_set = pipeline_cache.Acquire(DescriptorHeapType::Buffer);
pipeline_cache.BindBuffer(0, uniform_buffer.Handle(), 0, sizeof(VSPicaUniformData)); update_queue.AddBuffer(buffer_set, 0, uniform_buffer.Handle(), 0, sizeof(VSPicaUniformData));
pipeline_cache.BindBuffer(1, uniform_buffer.Handle(), 0, sizeof(VSUniformData)); update_queue.AddBuffer(buffer_set, 1, uniform_buffer.Handle(), 0, sizeof(VSUniformData));
pipeline_cache.BindBuffer(2, uniform_buffer.Handle(), 0, sizeof(FSUniformData)); update_queue.AddBuffer(buffer_set, 2, uniform_buffer.Handle(), 0, sizeof(FSUniformData));
pipeline_cache.BindTexelBuffer(3, *texture_lf_view); update_queue.AddTexelBuffer(buffer_set, 3, *texture_lf_view);
pipeline_cache.BindTexelBuffer(4, *texture_rg_view); update_queue.AddTexelBuffer(buffer_set, 4, *texture_rg_view);
pipeline_cache.BindTexelBuffer(5, *texture_rgba_view); update_queue.AddTexelBuffer(buffer_set, 5, *texture_rgba_view);
const auto texture_set = pipeline_cache.Acquire(DescriptorHeapType::Texture);
Surface& null_surface = res_cache.GetSurface(VideoCore::NULL_SURFACE_ID); Surface& null_surface = res_cache.GetSurface(VideoCore::NULL_SURFACE_ID);
Sampler& null_sampler = res_cache.GetSampler(VideoCore::NULL_SAMPLER_ID); Sampler& null_sampler = res_cache.GetSampler(VideoCore::NULL_SAMPLER_ID);
// Prepare texture and utility descriptor sets.
for (u32 i = 0; i < 3; i++) { for (u32 i = 0; i < 3; i++) {
pipeline_cache.BindTexture(i, null_surface.ImageView(), null_sampler.Handle()); update_queue.AddImageSampler(texture_set, i, 0, null_surface.ImageView(),
null_sampler.Handle());
} }
for (u32 i = 0; i < 7; i++) { const auto utility_set = pipeline_cache.Acquire(DescriptorHeapType::Utility);
pipeline_cache.BindStorageImage(i, null_surface.StorageView()); update_queue.AddStorageImage(utility_set, 0, null_surface.StorageView());
} update_queue.AddImageSampler(utility_set, 1, 0, null_surface.ImageView(),
null_sampler.Handle());
update_queue.Flush();
SyncEntireState(); SyncEntireState();
} }
@ -482,13 +488,6 @@ bool RasterizerVulkan::Draw(bool accelerate, bool is_indexed) {
pipeline_info.attachments.color = framebuffer->Format(SurfaceType::Color); pipeline_info.attachments.color = framebuffer->Format(SurfaceType::Color);
pipeline_info.attachments.depth = framebuffer->Format(SurfaceType::Depth); pipeline_info.attachments.depth = framebuffer->Format(SurfaceType::Depth);
if (shadow_rendering) {
pipeline_cache.BindStorageImage(6, framebuffer->ImageView(SurfaceType::Color));
} else {
Surface& null_surface = res_cache.GetSurface(VideoCore::NULL_SURFACE_ID);
pipeline_cache.BindStorageImage(6, null_surface.StorageView());
}
// Update scissor uniforms // Update scissor uniforms
const auto [scissor_x1, scissor_y2, scissor_x2, scissor_y1] = fb_helper.Scissor(); const auto [scissor_x1, scissor_y2, scissor_x2, scissor_y1] = fb_helper.Scissor();
if (fs_uniform_block_data.data.scissor_x1 != scissor_x1 || if (fs_uniform_block_data.data.scissor_x1 != scissor_x1 ||
@ -505,6 +504,7 @@ bool RasterizerVulkan::Draw(bool accelerate, bool is_indexed) {
// Sync and bind the texture surfaces // Sync and bind the texture surfaces
SyncTextureUnits(framebuffer); SyncTextureUnits(framebuffer);
SyncUtilityTextures(framebuffer);
// Sync and bind the shader // Sync and bind the shader
if (shader_dirty) { if (shader_dirty) {
@ -538,8 +538,8 @@ bool RasterizerVulkan::Draw(bool accelerate, bool is_indexed) {
} else { } else {
pipeline_cache.BindPipeline(pipeline_info, true); pipeline_cache.BindPipeline(pipeline_info, true);
const u64 vertex_size = vertex_batch.size() * sizeof(HardwareVertex);
const u32 vertex_count = static_cast<u32>(vertex_batch.size()); const u32 vertex_count = static_cast<u32>(vertex_batch.size());
const u32 vertex_size = vertex_count * sizeof(HardwareVertex);
const auto [buffer, offset, _] = stream_buffer.Map(vertex_size, sizeof(HardwareVertex)); const auto [buffer, offset, _] = stream_buffer.Map(vertex_size, sizeof(HardwareVertex));
std::memcpy(buffer, vertex_batch.data(), vertex_size); std::memcpy(buffer, vertex_batch.data(), vertex_size);
@ -559,6 +559,11 @@ void RasterizerVulkan::SyncTextureUnits(const Framebuffer* framebuffer) {
using TextureType = Pica::TexturingRegs::TextureConfig::TextureType; using TextureType = Pica::TexturingRegs::TextureConfig::TextureType;
const auto pica_textures = regs.texturing.GetTextures(); const auto pica_textures = regs.texturing.GetTextures();
const bool use_cube_heap =
pica_textures[0].enabled && pica_textures[0].config.type == TextureType::ShadowCube;
const auto texture_set = pipeline_cache.Acquire(use_cube_heap ? DescriptorHeapType::Texture
: DescriptorHeapType::Texture);
for (u32 texture_index = 0; texture_index < pica_textures.size(); ++texture_index) { for (u32 texture_index = 0; texture_index < pica_textures.size(); ++texture_index) {
const auto& texture = pica_textures[texture_index]; const auto& texture = pica_textures[texture_index];
@ -566,8 +571,8 @@ void RasterizerVulkan::SyncTextureUnits(const Framebuffer* framebuffer) {
if (!texture.enabled) { if (!texture.enabled) {
const Surface& null_surface = res_cache.GetSurface(VideoCore::NULL_SURFACE_ID); const Surface& null_surface = res_cache.GetSurface(VideoCore::NULL_SURFACE_ID);
const Sampler& null_sampler = res_cache.GetSampler(VideoCore::NULL_SAMPLER_ID); const Sampler& null_sampler = res_cache.GetSampler(VideoCore::NULL_SAMPLER_ID);
pipeline_cache.BindTexture(texture_index, null_surface.ImageView(), update_queue.AddImageSampler(texture_set, texture_index, 0, null_surface.ImageView(),
null_sampler.Handle()); null_sampler.Handle());
continue; continue;
} }
@ -576,20 +581,21 @@ void RasterizerVulkan::SyncTextureUnits(const Framebuffer* framebuffer) {
switch (texture.config.type.Value()) { switch (texture.config.type.Value()) {
case TextureType::Shadow2D: { case TextureType::Shadow2D: {
Surface& surface = res_cache.GetTextureSurface(texture); Surface& surface = res_cache.GetTextureSurface(texture);
Sampler& sampler = res_cache.GetSampler(texture.config);
surface.flags |= VideoCore::SurfaceFlagBits::ShadowMap; surface.flags |= VideoCore::SurfaceFlagBits::ShadowMap;
pipeline_cache.BindStorageImage(0, surface.StorageView()); update_queue.AddImageSampler(texture_set, texture_index, 0, surface.StorageView(),
sampler.Handle());
continue; continue;
} }
case TextureType::ShadowCube: { case TextureType::ShadowCube: {
BindShadowCube(texture); BindShadowCube(texture, texture_set);
continue; continue;
} }
case TextureType::TextureCube: { case TextureType::TextureCube: {
BindTextureCube(texture); BindTextureCube(texture, texture_set);
continue; continue;
} }
default: default:
UnbindSpecial();
break; break;
} }
} }
@ -597,13 +603,26 @@ void RasterizerVulkan::SyncTextureUnits(const Framebuffer* framebuffer) {
// Bind the texture provided by the rasterizer cache // Bind the texture provided by the rasterizer cache
Surface& surface = res_cache.GetTextureSurface(texture); Surface& surface = res_cache.GetTextureSurface(texture);
Sampler& sampler = res_cache.GetSampler(texture.config); Sampler& sampler = res_cache.GetSampler(texture.config);
if (!IsFeedbackLoop(texture_index, framebuffer, surface, sampler)) { const vk::ImageView color_view = framebuffer->ImageView(SurfaceType::Color);
pipeline_cache.BindTexture(texture_index, surface.ImageView(), sampler.Handle()); const bool is_feedback_loop = color_view == surface.ImageView();
} const vk::ImageView texture_view =
is_feedback_loop ? surface.CopyImageView() : surface.ImageView();
update_queue.AddImageSampler(texture_set, texture_index, 0, texture_view, sampler.Handle());
} }
} }
void RasterizerVulkan::BindShadowCube(const Pica::TexturingRegs::FullTextureConfig& texture) { void RasterizerVulkan::SyncUtilityTextures(const Framebuffer* framebuffer) {
const bool shadow_rendering = regs.framebuffer.IsShadowRendering();
if (!shadow_rendering) {
return;
}
const auto utility_set = pipeline_cache.Acquire(DescriptorHeapType::Utility);
update_queue.AddStorageImage(utility_set, 0, framebuffer->ImageView(SurfaceType::Color));
}
void RasterizerVulkan::BindShadowCube(const Pica::TexturingRegs::FullTextureConfig& texture,
vk::DescriptorSet texture_set) {
using CubeFace = Pica::TexturingRegs::CubeFace; using CubeFace = Pica::TexturingRegs::CubeFace;
auto info = Pica::Texture::TextureInfo::FromPicaRegister(texture.config, texture.format); auto info = Pica::Texture::TextureInfo::FromPicaRegister(texture.config, texture.format);
constexpr std::array faces = { constexpr std::array faces = {
@ -611,6 +630,8 @@ void RasterizerVulkan::BindShadowCube(const Pica::TexturingRegs::FullTextureConf
CubeFace::NegativeY, CubeFace::PositiveZ, CubeFace::NegativeZ, CubeFace::NegativeY, CubeFace::PositiveZ, CubeFace::NegativeZ,
}; };
Sampler& sampler = res_cache.GetSampler(texture.config);
for (CubeFace face : faces) { for (CubeFace face : faces) {
const u32 binding = static_cast<u32>(face); const u32 binding = static_cast<u32>(face);
info.physical_address = regs.texturing.GetCubePhysicalAddress(face); info.physical_address = regs.texturing.GetCubePhysicalAddress(face);
@ -618,11 +639,13 @@ void RasterizerVulkan::BindShadowCube(const Pica::TexturingRegs::FullTextureConf
const VideoCore::SurfaceId surface_id = res_cache.GetTextureSurface(info); const VideoCore::SurfaceId surface_id = res_cache.GetTextureSurface(info);
Surface& surface = res_cache.GetSurface(surface_id); Surface& surface = res_cache.GetSurface(surface_id);
surface.flags |= VideoCore::SurfaceFlagBits::ShadowMap; surface.flags |= VideoCore::SurfaceFlagBits::ShadowMap;
pipeline_cache.BindStorageImage(binding, surface.StorageView()); update_queue.AddImageSampler(texture_set, 0, binding, surface.StorageView(),
sampler.Handle());
} }
} }
void RasterizerVulkan::BindTextureCube(const Pica::TexturingRegs::FullTextureConfig& texture) { void RasterizerVulkan::BindTextureCube(const Pica::TexturingRegs::FullTextureConfig& texture,
vk::DescriptorSet texture_set) {
using CubeFace = Pica::TexturingRegs::CubeFace; using CubeFace = Pica::TexturingRegs::CubeFace;
const VideoCore::TextureCubeConfig config = { const VideoCore::TextureCubeConfig config = {
.px = regs.texturing.GetCubePhysicalAddress(CubeFace::PositiveX), .px = regs.texturing.GetCubePhysicalAddress(CubeFace::PositiveX),
@ -638,27 +661,7 @@ void RasterizerVulkan::BindTextureCube(const Pica::TexturingRegs::FullTextureCon
Surface& surface = res_cache.GetTextureCube(config); Surface& surface = res_cache.GetTextureCube(config);
Sampler& sampler = res_cache.GetSampler(texture.config); Sampler& sampler = res_cache.GetSampler(texture.config);
pipeline_cache.BindTexture(0, surface.ImageView(), sampler.Handle()); update_queue.AddImageSampler(texture_set, 0, 0, surface.ImageView(), sampler.Handle());
}
bool RasterizerVulkan::IsFeedbackLoop(u32 texture_index, const Framebuffer* framebuffer,
Surface& surface, Sampler& sampler) {
const vk::ImageView color_view = framebuffer->ImageView(SurfaceType::Color);
const bool is_feedback_loop = color_view == surface.ImageView();
if (!is_feedback_loop) {
return false;
}
// Make a temporary copy of the framebuffer to sample from
pipeline_cache.BindTexture(texture_index, surface.CopyImageView(), sampler.Handle());
return true;
}
void RasterizerVulkan::UnbindSpecial() {
Surface& null_surface = res_cache.GetSurface(VideoCore::NULL_SURFACE_ID);
for (u32 i = 0; i < 6; i++) {
pipeline_cache.BindStorageImage(i, null_surface.StorageView());
}
} }
void RasterizerVulkan::NotifyFixedFunctionPicaRegisterChanged(u32 id) { void RasterizerVulkan::NotifyFixedFunctionPicaRegisterChanged(u32 id) {
@ -1096,7 +1099,7 @@ void RasterizerVulkan::UploadUniforms(bool accelerate_draw) {
return; return;
} }
const u64 uniform_size = const u32 uniform_size =
uniform_size_aligned_vs_pica + uniform_size_aligned_vs + uniform_size_aligned_fs; uniform_size_aligned_vs_pica + uniform_size_aligned_vs + uniform_size_aligned_fs;
auto [uniforms, offset, invalidate] = auto [uniforms, offset, invalidate] =
uniform_buffer.Map(uniform_size, uniform_buffer_alignment); uniform_buffer.Map(uniform_size, uniform_buffer_alignment);
@ -1107,18 +1110,18 @@ void RasterizerVulkan::UploadUniforms(bool accelerate_draw) {
std::memcpy(uniforms + used_bytes, &vs_uniform_block_data.data, std::memcpy(uniforms + used_bytes, &vs_uniform_block_data.data,
sizeof(vs_uniform_block_data.data)); sizeof(vs_uniform_block_data.data));
pipeline_cache.SetBufferOffset(1, offset + used_bytes); pipeline_cache.UpdateRange(1, offset + used_bytes);
vs_uniform_block_data.dirty = false; vs_uniform_block_data.dirty = false;
used_bytes += static_cast<u32>(uniform_size_aligned_vs); used_bytes += uniform_size_aligned_vs;
} }
if (sync_fs || invalidate) { if (sync_fs || invalidate) {
std::memcpy(uniforms + used_bytes, &fs_uniform_block_data.data, std::memcpy(uniforms + used_bytes, &fs_uniform_block_data.data,
sizeof(fs_uniform_block_data.data)); sizeof(fs_uniform_block_data.data));
pipeline_cache.SetBufferOffset(2, offset + used_bytes); pipeline_cache.UpdateRange(2, offset + used_bytes);
fs_uniform_block_data.dirty = false; fs_uniform_block_data.dirty = false;
used_bytes += static_cast<u32>(uniform_size_aligned_fs); used_bytes += uniform_size_aligned_fs;
} }
if (sync_vs_pica) { if (sync_vs_pica) {
@ -1126,8 +1129,8 @@ void RasterizerVulkan::UploadUniforms(bool accelerate_draw) {
vs_uniforms.uniforms.SetFromRegs(regs.vs, pica.vs_setup); vs_uniforms.uniforms.SetFromRegs(regs.vs, pica.vs_setup);
std::memcpy(uniforms + used_bytes, &vs_uniforms, sizeof(vs_uniforms)); std::memcpy(uniforms + used_bytes, &vs_uniforms, sizeof(vs_uniforms));
pipeline_cache.SetBufferOffset(0, offset + used_bytes); pipeline_cache.UpdateRange(0, offset + used_bytes);
used_bytes += static_cast<u32>(uniform_size_aligned_vs_pica); used_bytes += uniform_size_aligned_vs_pica;
} }
uniform_buffer.Commit(used_bytes); uniform_buffer.Commit(used_bytes);

32
src/video_core/renderer_vulkan/vk_rasterizer.h
View file

@ -5,6 +5,7 @@
#pragma once #pragma once
#include "video_core/rasterizer_accelerated.h" #include "video_core/rasterizer_accelerated.h"
#include "video_core/renderer_vulkan/vk_descriptor_update_queue.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h" #include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_render_manager.h" #include "video_core/renderer_vulkan/vk_render_manager.h"
#include "video_core/renderer_vulkan/vk_stream_buffer.h" #include "video_core/renderer_vulkan/vk_stream_buffer.h"
@ -32,15 +33,15 @@ struct ScreenInfo;
class Instance; class Instance;
class Scheduler; class Scheduler;
class RenderManager; class RenderManager;
class DescriptorPool;
class RasterizerVulkan : public VideoCore::RasterizerAccelerated { class RasterizerVulkan : public VideoCore::RasterizerAccelerated {
public: public:
explicit RasterizerVulkan(Memory::MemorySystem& memory, Pica::PicaCore& pica, explicit RasterizerVulkan(Memory::MemorySystem& memory, Pica::PicaCore& pica,
VideoCore::CustomTexManager& custom_tex_manager, VideoCore::CustomTexManager& custom_tex_manager,
VideoCore::RendererBase& renderer, Frontend::EmuWindow& emu_window, VideoCore::RendererBase& renderer, Frontend::EmuWindow& emu_window,
const Instance& instance, Scheduler& scheduler, DescriptorPool& pool, const Instance& instance, Scheduler& scheduler,
RenderManager& render_manager, u32 image_count); RenderManager& render_manager, DescriptorUpdateQueue& update_queue,
u32 image_count);
~RasterizerVulkan() override; ~RasterizerVulkan() override;
void TickFrame(); void TickFrame();
@ -102,18 +103,16 @@ private:
/// Syncs all enabled PICA texture units /// Syncs all enabled PICA texture units
void SyncTextureUnits(const Framebuffer* framebuffer); void SyncTextureUnits(const Framebuffer* framebuffer);
/// Syncs all utility textures in the fragment shader.
void SyncUtilityTextures(const Framebuffer* framebuffer);
/// Binds the PICA shadow cube required for shadow mapping /// Binds the PICA shadow cube required for shadow mapping
void BindShadowCube(const Pica::TexturingRegs::FullTextureConfig& texture); void BindShadowCube(const Pica::TexturingRegs::FullTextureConfig& texture,
vk::DescriptorSet texture_set);
/// Binds a texture cube to texture unit 0 /// Binds a texture cube to texture unit 0
void BindTextureCube(const Pica::TexturingRegs::FullTextureConfig& texture); void BindTextureCube(const Pica::TexturingRegs::FullTextureConfig& texture,
vk::DescriptorSet texture_set);
/// Makes a temporary copy of the framebuffer if a feedback loop is detected
bool IsFeedbackLoop(u32 texture_index, const Framebuffer* framebuffer, Surface& surface,
Sampler& sampler);
/// Unbinds all special texture unit 0 texture configurations
void UnbindSpecial();
/// Upload the uniform blocks to the uniform buffer object /// Upload the uniform blocks to the uniform buffer object
void UploadUniforms(bool accelerate_draw); void UploadUniforms(bool accelerate_draw);
@ -146,6 +145,7 @@ private:
const Instance& instance; const Instance& instance;
Scheduler& scheduler; Scheduler& scheduler;
RenderManager& render_manager; RenderManager& render_manager;
DescriptorUpdateQueue& update_queue;
PipelineCache pipeline_cache; PipelineCache pipeline_cache;
TextureRuntime runtime; TextureRuntime runtime;
RasterizerCache res_cache; RasterizerCache res_cache;
@ -164,10 +164,10 @@ private:
vk::UniqueBufferView texture_lf_view; vk::UniqueBufferView texture_lf_view;
vk::UniqueBufferView texture_rg_view; vk::UniqueBufferView texture_rg_view;
vk::UniqueBufferView texture_rgba_view; vk::UniqueBufferView texture_rgba_view;
u64 uniform_buffer_alignment; vk::DeviceSize uniform_buffer_alignment;
u64 uniform_size_aligned_vs_pica; u32 uniform_size_aligned_vs_pica;
u64 uniform_size_aligned_vs; u32 uniform_size_aligned_vs;
u64 uniform_size_aligned_fs; u32 uniform_size_aligned_fs;
bool async_shaders{false}; bool async_shaders{false};
}; };

11
src/video_core/renderer_vulkan/vk_render_manager.cpp
View file

@ -1,4 +1,4 @@
// Copyright 2023 Citra Emulator Project // Copyright 2024 Citra Emulator Project
// Licensed under GPLv2 or any later version // Licensed under GPLv2 or any later version
// Refer to the license.txt file included. // Refer to the license.txt file included.
@ -23,7 +23,7 @@ RenderManager::RenderManager(const Instance& instance, Scheduler& scheduler)
RenderManager::~RenderManager() = default; RenderManager::~RenderManager() = default;
void RenderManager::BeginRendering(const Framebuffer* framebuffer, void RenderManager::BeginRendering(const Framebuffer* framebuffer,
Common::Rectangle<u32> draw_rect) { Common::Rectangle<u32> draw_rect) {
const vk::Rect2D render_area = { const vk::Rect2D render_area = {
.offset{ .offset{
.x = static_cast<s32>(draw_rect.left), .x = static_cast<s32>(draw_rect.left),
@ -107,6 +107,9 @@ void RenderManager::EndRendering() {
}; };
} }
cmdbuf.endRenderPass(); cmdbuf.endRenderPass();
if (num_barriers == 0) {
return;
}
cmdbuf.pipelineBarrier(pipeline_flags, cmdbuf.pipelineBarrier(pipeline_flags,
vk::PipelineStageFlagBits::eFragmentShader | vk::PipelineStageFlagBits::eFragmentShader |
vk::PipelineStageFlagBits::eTransfer, vk::PipelineStageFlagBits::eTransfer,
@ -128,7 +131,7 @@ void RenderManager::EndRendering() {
} }
vk::RenderPass RenderManager::GetRenderpass(VideoCore::PixelFormat color, vk::RenderPass RenderManager::GetRenderpass(VideoCore::PixelFormat color,
VideoCore::PixelFormat depth, bool is_clear) { VideoCore::PixelFormat depth, bool is_clear) {
std::scoped_lock lock{cache_mutex}; std::scoped_lock lock{cache_mutex};
const u32 color_index = const u32 color_index =
@ -153,7 +156,7 @@ vk::RenderPass RenderManager::GetRenderpass(VideoCore::PixelFormat color,
} }
vk::UniqueRenderPass RenderManager::CreateRenderPass(vk::Format color, vk::Format depth, vk::UniqueRenderPass RenderManager::CreateRenderPass(vk::Format color, vk::Format depth,
vk::AttachmentLoadOp load_op) const { vk::AttachmentLoadOp load_op) const {
u32 attachment_count = 0; u32 attachment_count = 0;
std::array<vk::AttachmentDescription, 2> attachments; std::array<vk::AttachmentDescription, 2> attachments;

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

@ -1,4 +1,4 @@
// Copyright 2023 Citra Emulator Project // Copyright 2024 Citra Emulator Project
// Licensed under GPLv2 or any later version // Licensed under GPLv2 or any later version
// Refer to the license.txt file included. // Refer to the license.txt file included.
@ -27,7 +27,10 @@ struct RenderPass {
u32 do_clear; u32 do_clear;
bool operator==(const RenderPass& other) const noexcept { bool operator==(const RenderPass& other) const noexcept {
return std::memcmp(this, &other, sizeof(RenderPass)) == 0; return std::tie(framebuffer, render_pass, render_area, do_clear) ==
std::tie(other.framebuffer, other.render_pass, other.render_area,
other.do_clear) &&
std::memcmp(&clear, &other.clear, sizeof(vk::ClearValue)) == 0;
} }
}; };

157
src/video_core/renderer_vulkan/vk_resource_pool.cpp
View file

@ -4,6 +4,7 @@
#include <cstddef> #include <cstddef>
#include <optional> #include <optional>
#include <unordered_map>
#include "video_core/renderer_vulkan/vk_instance.h" #include "video_core/renderer_vulkan/vk_instance.h"
#include "video_core/renderer_vulkan/vk_master_semaphore.h" #include "video_core/renderer_vulkan/vk_master_semaphore.h"
#include "video_core/renderer_vulkan/vk_resource_pool.h" #include "video_core/renderer_vulkan/vk_resource_pool.h"
@ -14,9 +15,7 @@ ResourcePool::ResourcePool(MasterSemaphore* master_semaphore_, std::size_t grow_
: master_semaphore{master_semaphore_}, grow_step{grow_step_} {} : master_semaphore{master_semaphore_}, grow_step{grow_step_} {}
std::size_t ResourcePool::CommitResource() { std::size_t ResourcePool::CommitResource() {
// Refresh semaphore to query updated results u64 gpu_tick = master_semaphore->KnownGpuTick();
master_semaphore->Refresh();
const u64 gpu_tick = master_semaphore->KnownGpuTick();
const auto search = [this, gpu_tick](std::size_t begin, const auto search = [this, gpu_tick](std::size_t begin,
std::size_t end) -> std::optional<std::size_t> { std::size_t end) -> std::optional<std::size_t> {
for (std::size_t iterator = begin; iterator < end; ++iterator) { for (std::size_t iterator = begin; iterator < end; ++iterator) {
@ -29,7 +28,13 @@ std::size_t ResourcePool::CommitResource() {
}; };
// Try to find a free resource from the hinted position to the end. // Try to find a free resource from the hinted position to the end.
std::optional<std::size_t> found = search(hint_iterator, ticks.size()); auto found = search(hint_iterator, ticks.size());
if (!found) {
// Refresh semaphore to query updated results
master_semaphore->Refresh();
gpu_tick = master_semaphore->KnownGpuTick();
found = search(hint_iterator, ticks.size());
}
if (!found) { if (!found) {
// Search from beginning to the hinted position. // Search from beginning to the hinted position.
found = search(0, hint_iterator); found = search(0, hint_iterator);
@ -48,75 +53,137 @@ std::size_t ResourcePool::CommitResource() {
} }
std::size_t ResourcePool::ManageOverflow() { std::size_t ResourcePool::ManageOverflow() {
const std::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 std::size_t old_capacity = ticks.size(); const std::size_t old_capacity = ticks.size();
ticks.resize(old_capacity + grow_step); ticks.resize(old_capacity + grow_step);
Allocate(old_capacity, old_capacity + grow_step); Allocate(old_capacity, old_capacity + grow_step);
return old_capacity;
} }
constexpr std::size_t COMMAND_BUFFER_POOL_SIZE = 4; constexpr std::size_t COMMAND_BUFFER_POOL_SIZE = 4;
struct CommandPool::Pool {
vk::CommandPool handle;
std::array<vk::CommandBuffer, COMMAND_BUFFER_POOL_SIZE> cmdbufs;
};
CommandPool::CommandPool(const Instance& instance, MasterSemaphore* master_semaphore) CommandPool::CommandPool(const Instance& instance, MasterSemaphore* master_semaphore)
: ResourcePool{master_semaphore, COMMAND_BUFFER_POOL_SIZE}, instance{instance} {} : ResourcePool{master_semaphore, COMMAND_BUFFER_POOL_SIZE}, instance{instance} {
CommandPool::~CommandPool() {
vk::Device device = instance.GetDevice();
for (Pool& pool : pools) {
device.destroyCommandPool(pool.handle);
}
}
void CommandPool::Allocate(std::size_t begin, std::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();
const vk::CommandPoolCreateInfo pool_create_info = { const vk::CommandPoolCreateInfo pool_create_info = {
.flags = vk::CommandPoolCreateFlagBits::eTransient | .flags = vk::CommandPoolCreateFlagBits::eTransient |
vk::CommandPoolCreateFlagBits::eResetCommandBuffer, vk::CommandPoolCreateFlagBits::eResetCommandBuffer,
.queueFamilyIndex = instance.GetGraphicsQueueFamilyIndex(), .queueFamilyIndex = instance.GetGraphicsQueueFamilyIndex(),
}; };
const vk::Device device = instance.GetDevice();
cmd_pool = device.createCommandPoolUnique(pool_create_info);
if (instance.HasDebuggingToolAttached()) {
SetObjectName(device, *cmd_pool, "CommandPool");
}
}
vk::Device device = instance.GetDevice(); CommandPool::~CommandPool() = default;
pool.handle = device.createCommandPool(pool_create_info);
void CommandPool::Allocate(std::size_t begin, std::size_t end) {
cmd_buffers.resize(end);
const vk::CommandBufferAllocateInfo buffer_alloc_info = { const vk::CommandBufferAllocateInfo buffer_alloc_info = {
.commandPool = pool.handle, .commandPool = *cmd_pool,
.level = vk::CommandBufferLevel::ePrimary, .level = vk::CommandBufferLevel::ePrimary,
.commandBufferCount = COMMAND_BUFFER_POOL_SIZE, .commandBufferCount = COMMAND_BUFFER_POOL_SIZE,
}; };
auto buffers = device.allocateCommandBuffers(buffer_alloc_info); const vk::Device device = instance.GetDevice();
std::copy(buffers.begin(), buffers.end(), pool.cmdbufs.begin()); const auto result =
device.allocateCommandBuffers(&buffer_alloc_info, cmd_buffers.data() + begin);
ASSERT(result == vk::Result::eSuccess);
if (instance.HasDebuggingToolAttached()) { if (instance.HasDebuggingToolAttached()) {
Vulkan::SetObjectName(device, pool.handle, "CommandPool: Pool({})", for (std::size_t i = begin; i < end; ++i) {
COMMAND_BUFFER_POOL_SIZE); SetObjectName(device, cmd_buffers[i], "CommandPool: Command Buffer {}", i);
for (u32 i = 0; i < pool.cmdbufs.size(); ++i) {
Vulkan::SetObjectName(device, pool.cmdbufs[i], "CommandPool: Command Buffer {}", i);
} }
} }
} }
vk::CommandBuffer CommandPool::Commit() { vk::CommandBuffer CommandPool::Commit() {
const std::size_t index = CommitResource(); const std::size_t index = CommitResource();
const auto pool_index = index / COMMAND_BUFFER_POOL_SIZE; return cmd_buffers[index];
const auto sub_index = index % COMMAND_BUFFER_POOL_SIZE; }
return pools[pool_index].cmdbufs[sub_index];
constexpr u32 DESCRIPTOR_SET_BATCH = 32;
DescriptorHeap::DescriptorHeap(const Instance& instance, MasterSemaphore* master_semaphore,
std::span<const vk::DescriptorSetLayoutBinding> bindings,
u32 descriptor_heap_count_)
: ResourcePool{master_semaphore, DESCRIPTOR_SET_BATCH}, device{instance.GetDevice()},
descriptor_heap_count{descriptor_heap_count_} {
// Create descriptor set layout.
const vk::DescriptorSetLayoutCreateInfo layout_ci = {
.bindingCount = static_cast<u32>(bindings.size()),
.pBindings = bindings.data(),
};
descriptor_set_layout = device.createDescriptorSetLayoutUnique(layout_ci);
if (instance.HasDebuggingToolAttached()) {
SetObjectName(device, *descriptor_set_layout, "DescriptorSetLayout");
}
// Build descriptor set pool counts.
std::unordered_map<vk::DescriptorType, u16> descriptor_type_counts;
for (const auto& binding : bindings) {
descriptor_type_counts[binding.descriptorType] += binding.descriptorCount;
}
for (const auto& [type, count] : descriptor_type_counts) {
auto& pool_size = pool_sizes.emplace_back();
pool_size.descriptorCount = count * descriptor_heap_count;
pool_size.type = type;
}
// Create descriptor pool
AppendDescriptorPool();
}
DescriptorHeap::~DescriptorHeap() = default;
void DescriptorHeap::Allocate(std::size_t begin, std::size_t end) {
ASSERT(end - begin == DESCRIPTOR_SET_BATCH);
descriptor_sets.resize(end);
std::array<vk::DescriptorSetLayout, DESCRIPTOR_SET_BATCH> layouts;
layouts.fill(*descriptor_set_layout);
u32 current_pool = 0;
vk::DescriptorSetAllocateInfo alloc_info = {
.descriptorPool = *pools[current_pool],
.descriptorSetCount = DESCRIPTOR_SET_BATCH,
.pSetLayouts = layouts.data(),
};
// Attempt to allocate the descriptor set batch. If the pool has run out of space, use a new
// one.
while (true) {
const auto result =
device.allocateDescriptorSets(&alloc_info, descriptor_sets.data() + begin);
if (result == vk::Result::eSuccess) {
break;
}
if (result == vk::Result::eErrorOutOfPoolMemory) {
current_pool++;
if (current_pool == pools.size()) {
LOG_INFO(Render_Vulkan, "Run out of pools, creating new one!");
AppendDescriptorPool();
}
alloc_info.descriptorPool = *pools[current_pool];
}
}
}
vk::DescriptorSet DescriptorHeap::Commit() {
const std::size_t index = CommitResource();
return descriptor_sets[index];
}
void DescriptorHeap::AppendDescriptorPool() {
const vk::DescriptorPoolCreateInfo pool_info = {
.flags = vk::DescriptorPoolCreateFlagBits::eFreeDescriptorSet,
.maxSets = descriptor_heap_count,
.poolSizeCount = static_cast<u32>(pool_sizes.size()),
.pPoolSizes = pool_sizes.data(),
};
auto& pool = pools.emplace_back();
pool = device.createDescriptorPoolUnique(pool_info);
} }
} // namespace Vulkan } // namespace Vulkan

34
src/video_core/renderer_vulkan/vk_resource_pool.h
View file

@ -39,9 +39,6 @@ private:
/// Manages pool overflow allocating new resources. /// Manages pool overflow allocating new resources.
std::size_t ManageOverflow(); std::size_t ManageOverflow();
/// Allocates a new page of resources.
void Grow();
protected: protected:
MasterSemaphore* master_semaphore{nullptr}; MasterSemaphore* master_semaphore{nullptr};
std::size_t grow_step = 0; ///< Number of new resources created after an overflow std::size_t grow_step = 0; ///< Number of new resources created after an overflow
@ -59,9 +56,36 @@ public:
vk::CommandBuffer Commit(); vk::CommandBuffer Commit();
private: private:
struct Pool;
const Instance& instance; const Instance& instance;
std::vector<Pool> pools; vk::UniqueCommandPool cmd_pool;
std::vector<vk::CommandBuffer> cmd_buffers;
};
class DescriptorHeap final : public ResourcePool {
public:
explicit DescriptorHeap(const Instance& instance, MasterSemaphore* master_semaphore,
std::span<const vk::DescriptorSetLayoutBinding> bindings,
u32 descriptor_heap_count = 1024);
~DescriptorHeap() override;
const vk::DescriptorSetLayout& Layout() const {
return *descriptor_set_layout;
}
void Allocate(std::size_t begin, std::size_t end) override;
vk::DescriptorSet Commit();
private:
void AppendDescriptorPool();
private:
vk::Device device;
vk::UniqueDescriptorSetLayout descriptor_set_layout;
u32 descriptor_heap_count;
std::vector<vk::DescriptorPoolSize> pool_sizes;
std::vector<vk::UniqueDescriptorPool> pools;
std::vector<vk::DescriptorSet> descriptor_sets;
}; };
} // namespace Vulkan } // namespace Vulkan

6
src/video_core/renderer_vulkan/vk_scheduler.cpp
View file

@ -5,10 +5,8 @@
#include <mutex> #include <mutex>
#include <utility> #include <utility>
#include "common/microprofile.h" #include "common/microprofile.h"
#include "common/settings.h"
#include "common/thread.h" #include "common/thread.h"
#include "video_core/renderer_vulkan/vk_instance.h" #include "video_core/renderer_vulkan/vk_instance.h"
#include "video_core/renderer_vulkan/vk_render_manager.h"
#include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_scheduler.h"
MICROPROFILE_DEFINE(Vulkan_WaitForWorker, "Vulkan", "Wait for worker", MP_RGB(255, 192, 192)); MICROPROFILE_DEFINE(Vulkan_WaitForWorker, "Vulkan", "Wait for worker", MP_RGB(255, 192, 192));
@ -98,7 +96,7 @@ void Scheduler::DispatchWork() {
return; return;
} }
//on_dispatch(); on_dispatch();
{ {
std::scoped_lock ql{queue_mutex}; std::scoped_lock ql{queue_mutex};
@ -183,6 +181,8 @@ void Scheduler::SubmitExecution(vk::Semaphore signal_semaphore, vk::Semaphore wa
master_semaphore->SubmitWork(cmdbuf, wait_semaphore, signal_semaphore, signal_value); master_semaphore->SubmitWork(cmdbuf, wait_semaphore, signal_semaphore, signal_value);
}); });
master_semaphore->Refresh();
if (!use_worker_thread) { if (!use_worker_thread) {
AllocateWorkerCommandBuffers(); AllocateWorkerCommandBuffers();
} else { } else {

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

@ -4,9 +4,9 @@
#pragma once #pragma once
#include <functional>
#include <memory> #include <memory>
#include <utility> #include <utility>
#include <functional>
#include "common/alignment.h" #include "common/alignment.h"
#include "common/common_funcs.h" #include "common/common_funcs.h"
#include "common/polyfill_thread.h" #include "common/polyfill_thread.h"

1
src/video_core/renderer_vulkan/vk_shader_util.cpp
View file

@ -182,6 +182,7 @@ vk::ShaderModule Compile(std::string_view code, vk::ShaderStageFlagBits stage, v
includer)) [[unlikely]] { includer)) [[unlikely]] {
LOG_INFO(Render_Vulkan, "Shader Info Log:\n{}\n{}", shader->getInfoLog(), LOG_INFO(Render_Vulkan, "Shader Info Log:\n{}\n{}", shader->getInfoLog(),
shader->getInfoDebugLog()); shader->getInfoDebugLog());
LOG_INFO(Render_Vulkan, "Shader Source:\n{}", code);
return {}; return {};
} }

13
src/video_core/renderer_vulkan/vk_stream_buffer.cpp
View file

@ -82,7 +82,7 @@ StreamBuffer::~StreamBuffer() {
device.freeMemory(memory); device.freeMemory(memory);
} }
std::tuple<u8*, u64, bool> StreamBuffer::Map(u64 size, u64 alignment) { std::tuple<u8*, u32, bool> StreamBuffer::Map(u32 size, u64 alignment) {
if (!is_coherent && type == BufferType::Stream) { if (!is_coherent && type == BufferType::Stream) {
size = Common::AlignUp(size, instance.NonCoherentAtomSize()); size = Common::AlignUp(size, instance.NonCoherentAtomSize());
} }
@ -114,7 +114,7 @@ std::tuple<u8*, u64, bool> StreamBuffer::Map(u64 size, u64 alignment) {
return std::make_tuple(mapped + offset, offset, invalidate); return std::make_tuple(mapped + offset, offset, invalidate);
} }
void StreamBuffer::Commit(u64 size) { void StreamBuffer::Commit(u32 size) {
if (!is_coherent && type == BufferType::Stream) { if (!is_coherent && type == BufferType::Stream) {
size = Common::AlignUp(size, instance.NonCoherentAtomSize()); size = Common::AlignUp(size, instance.NonCoherentAtomSize());
} }
@ -200,11 +200,10 @@ void StreamBuffer::CreateBuffers(u64 prefered_size) {
mapped = reinterpret_cast<u8*>(device.mapMemory(memory, 0, VK_WHOLE_SIZE)); mapped = reinterpret_cast<u8*>(device.mapMemory(memory, 0, VK_WHOLE_SIZE));
if (instance.HasDebuggingToolAttached()) { if (instance.HasDebuggingToolAttached()) {
Vulkan::SetObjectName(device, buffer, "StreamBuffer({}): {} KiB {}", BufferTypeName(type), SetObjectName(device, buffer, "StreamBuffer({}): {} KiB {}", BufferTypeName(type),
stream_buffer_size / 1024, vk::to_string(mem_type.propertyFlags)); stream_buffer_size / 1024, vk::to_string(mem_type.propertyFlags));
Vulkan::SetObjectName(device, memory, "StreamBufferMemory({}): {} Kib {}", SetObjectName(device, memory, "StreamBufferMemory({}): {} Kib {}", BufferTypeName(type),
BufferTypeName(type), stream_buffer_size / 1024, stream_buffer_size / 1024, vk::to_string(mem_type.propertyFlags));
vk::to_string(mem_type.propertyFlags));
} }
} }

8
src/video_core/renderer_vulkan/vk_stream_buffer.h
View file

@ -35,10 +35,10 @@ public:
* @param size Size to reserve. * @param size Size to reserve.
* @returns A pair of a raw memory pointer (with offset added), and the buffer offset * @returns A pair of a raw memory pointer (with offset added), and the buffer offset
*/ */
std::tuple<u8*, u64, bool> Map(u64 size, u64 alignment); std::tuple<u8*, u32, bool> Map(u32 size, u64 alignment);
/// Ensures that "size" bytes of memory are available to the GPU, potentially recording a copy. /// Ensures that "size" bytes of memory are available to the GPU, potentially recording a copy.
void Commit(u64 size); void Commit(u32 size);
vk::Buffer Handle() const noexcept { vk::Buffer Handle() const noexcept {
return buffer; return buffer;
@ -70,8 +70,8 @@ private:
vk::BufferUsageFlags usage{}; vk::BufferUsageFlags usage{};
BufferType type; BufferType type;
u64 offset{}; ///< Buffer iterator. u32 offset{}; ///< Buffer iterator.
u64 mapped_size{}; ///< Size reserved for the current copy. u32 mapped_size{}; ///< Size reserved for the current copy.
bool is_coherent{}; ///< True if the buffer is coherent bool is_coherent{}; ///< True if the buffer is coherent
std::vector<Watch> current_watches; ///< Watches recorded in the current iteration. std::vector<Watch> current_watches; ///< Watches recorded in the current iteration.

42
src/video_core/renderer_vulkan/vk_texture_runtime.cpp
View file

@ -11,7 +11,6 @@
#include "video_core/rasterizer_cache/texture_codec.h" #include "video_core/rasterizer_cache/texture_codec.h"
#include "video_core/rasterizer_cache/utils.h" #include "video_core/rasterizer_cache/utils.h"
#include "video_core/renderer_vulkan/pica_to_vk.h" #include "video_core/renderer_vulkan/pica_to_vk.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_instance.h" #include "video_core/renderer_vulkan/vk_instance.h"
#include "video_core/renderer_vulkan/vk_render_manager.h" #include "video_core/renderer_vulkan/vk_render_manager.h"
#include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_scheduler.h"
@ -249,10 +248,10 @@ constexpr u64 DOWNLOAD_BUFFER_SIZE = 16_MiB;
} // Anonymous namespace } // Anonymous namespace
TextureRuntime::TextureRuntime(const Instance& instance, Scheduler& scheduler, TextureRuntime::TextureRuntime(const Instance& instance, Scheduler& scheduler,
RenderManager& render_manager, DescriptorPool& pool, RenderManager& render_manager, DescriptorUpdateQueue& update_queue,
DescriptorSetProvider& texture_provider_, u32 num_swapchain_images_) u32 num_swapchain_images_)
: instance{instance}, scheduler{scheduler}, render_manager{render_manager}, : instance{instance}, scheduler{scheduler}, render_manager{render_manager},
texture_provider{texture_provider_}, blit_helper{instance, scheduler, pool, render_manager}, blit_helper{instance, scheduler, render_manager, update_queue},
upload_buffer{instance, scheduler, vk::BufferUsageFlagBits::eTransferSrc, UPLOAD_BUFFER_SIZE, upload_buffer{instance, scheduler, vk::BufferUsageFlagBits::eTransferSrc, UPLOAD_BUFFER_SIZE,
BufferType::Upload}, BufferType::Upload},
download_buffer{instance, scheduler, download_buffer{instance, scheduler,
@ -268,7 +267,7 @@ VideoCore::StagingData TextureRuntime::FindStaging(u32 size, bool upload) {
const auto [data, offset, invalidate] = buffer.Map(size, 16); const auto [data, offset, invalidate] = buffer.Map(size, 16);
return VideoCore::StagingData{ return VideoCore::StagingData{
.size = size, .size = size,
.offset = static_cast<u32>(offset), .offset = offset,
.mapped = std::span{data, size}, .mapped = std::span{data, size},
}; };
} }
@ -453,7 +452,7 @@ void TextureRuntime::ClearTextureWithRenderpass(Surface& surface,
} }
bool TextureRuntime::CopyTextures(Surface& source, Surface& dest, bool TextureRuntime::CopyTextures(Surface& source, Surface& dest,
const VideoCore::TextureCopy& copy) { std::span<const VideoCore::TextureCopy> copies) {
render_manager.EndRendering(); render_manager.EndRendering();
const RecordParams params = { const RecordParams params = {
@ -466,8 +465,9 @@ bool TextureRuntime::CopyTextures(Surface& source, Surface& dest,
.dst_image = dest.Image(), .dst_image = dest.Image(),
}; };
scheduler.Record([params, copy](vk::CommandBuffer cmdbuf) { boost::container::small_vector<vk::ImageCopy, 2> vk_copies;
const vk::ImageCopy image_copy = { std::ranges::transform(copies, std::back_inserter(vk_copies), [&](const auto& copy) {
return vk::ImageCopy{
.srcSubresource{ .srcSubresource{
.aspectMask = params.aspect, .aspectMask = params.aspect,
.mipLevel = copy.src_level, .mipLevel = copy.src_level,
@ -486,7 +486,9 @@ bool TextureRuntime::CopyTextures(Surface& source, Surface& dest,
0}, 0},
.extent = {copy.extent.width, copy.extent.height, 1}, .extent = {copy.extent.width, copy.extent.height, 1},
}; };
});
scheduler.Record([params, copies = std::move(vk_copies)](vk::CommandBuffer cmdbuf) {
const bool self_copy = params.src_image == params.dst_image; const bool self_copy = params.src_image == params.dst_image;
const vk::ImageLayout new_src_layout = const vk::ImageLayout new_src_layout =
self_copy ? vk::ImageLayout::eGeneral : vk::ImageLayout::eTransferSrcOptimal; self_copy ? vk::ImageLayout::eGeneral : vk::ImageLayout::eTransferSrcOptimal;
@ -502,7 +504,7 @@ bool TextureRuntime::CopyTextures(Surface& source, Surface& dest,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = params.src_image, .image = params.src_image,
.subresourceRange = MakeSubresourceRange(params.aspect, copy.src_level), .subresourceRange = MakeSubresourceRange(params.aspect, 0, VK_REMAINING_MIP_LEVELS),
}, },
vk::ImageMemoryBarrier{ vk::ImageMemoryBarrier{
.srcAccessMask = params.dst_access, .srcAccessMask = params.dst_access,
@ -512,7 +514,7 @@ bool TextureRuntime::CopyTextures(Surface& source, Surface& dest,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = params.dst_image, .image = params.dst_image,
.subresourceRange = MakeSubresourceRange(params.aspect, copy.dst_level), .subresourceRange = MakeSubresourceRange(params.aspect, 0, VK_REMAINING_MIP_LEVELS),
}, },
}; };
const std::array post_barriers = { const std::array post_barriers = {
@ -524,7 +526,7 @@ bool TextureRuntime::CopyTextures(Surface& source, Surface& dest,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = params.src_image, .image = params.src_image,
.subresourceRange = MakeSubresourceRange(params.aspect, copy.src_level), .subresourceRange = MakeSubresourceRange(params.aspect, 0, VK_REMAINING_MIP_LEVELS),
}, },
vk::ImageMemoryBarrier{ vk::ImageMemoryBarrier{
.srcAccessMask = vk::AccessFlagBits::eTransferWrite, .srcAccessMask = vk::AccessFlagBits::eTransferWrite,
@ -534,7 +536,7 @@ bool TextureRuntime::CopyTextures(Surface& source, Surface& dest,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = params.dst_image, .image = params.dst_image,
.subresourceRange = MakeSubresourceRange(params.aspect, copy.dst_level), .subresourceRange = MakeSubresourceRange(params.aspect, 0, VK_REMAINING_MIP_LEVELS),
}, },
}; };
@ -542,7 +544,7 @@ bool TextureRuntime::CopyTextures(Surface& source, Surface& dest,
vk::DependencyFlagBits::eByRegion, {}, {}, pre_barriers); vk::DependencyFlagBits::eByRegion, {}, {}, pre_barriers);
cmdbuf.copyImage(params.src_image, new_src_layout, params.dst_image, new_dst_layout, cmdbuf.copyImage(params.src_image, new_src_layout, params.dst_image, new_dst_layout,
image_copy); copies);
cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer, params.pipeline_flags, cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer, params.pipeline_flags,
vk::DependencyFlagBits::eByRegion, {}, {}, post_barriers); vk::DependencyFlagBits::eByRegion, {}, {}, post_barriers);
@ -694,13 +696,6 @@ bool TextureRuntime::NeedsConversion(VideoCore::PixelFormat format) const {
traits.aspect != (vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil); traits.aspect != (vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil);
} }
void TextureRuntime::FreeDescriptorSetsWithImage(vk::ImageView image_view) {
texture_provider.FreeWithImage(image_view);
blit_helper.compute_provider.FreeWithImage(image_view);
blit_helper.compute_buffer_provider.FreeWithImage(image_view);
blit_helper.two_textures_provider.FreeWithImage(image_view);
}
Surface::Surface(TextureRuntime& runtime_, const VideoCore::SurfaceParams& params) Surface::Surface(TextureRuntime& runtime_, const VideoCore::SurfaceParams& params)
: SurfaceBase{params}, runtime{&runtime_}, instance{&runtime_.GetInstance()}, : SurfaceBase{params}, runtime{&runtime_}, instance{&runtime_.GetInstance()},
scheduler{&runtime_.GetScheduler()}, traits{instance->GetTraits(pixel_format)} { scheduler{&runtime_.GetScheduler()}, traits{instance->GetTraits(pixel_format)} {
@ -800,9 +795,6 @@ Surface::~Surface() {
return; return;
} }
for (const auto& [alloc, image, image_view] : handles) { for (const auto& [alloc, image, image_view] : handles) {
if (image_view) {
runtime->FreeDescriptorSetsWithImage(*image_view);
}
if (image) { if (image) {
vmaDestroyImage(instance->GetAllocator(), image, alloc); vmaDestroyImage(instance->GetAllocator(), image, alloc);
} }
@ -904,7 +896,7 @@ void Surface::UploadCustom(const VideoCore::Material* material, u32 level) {
const Common::Rectangle rect{0U, height, width, 0U}; const Common::Rectangle rect{0U, height, width, 0U};
const auto upload = [&](u32 index, VideoCore::CustomTexture* texture) { const auto upload = [&](u32 index, VideoCore::CustomTexture* texture) {
const u64 custom_size = texture->data.size(); const u32 custom_size = static_cast<u32>(texture->data.size());
const RecordParams params = { const RecordParams params = {
.aspect = vk::ImageAspectFlagBits::eColor, .aspect = vk::ImageAspectFlagBits::eColor,
.pipeline_flags = PipelineStageFlags(), .pipeline_flags = PipelineStageFlags(),
@ -1518,7 +1510,7 @@ Sampler::Sampler(TextureRuntime& runtime, const VideoCore::SamplerParams& params
instance.IsCustomBorderColorSupported() && (params.wrap_s == TextureConfig::ClampToBorder || instance.IsCustomBorderColorSupported() && (params.wrap_s == TextureConfig::ClampToBorder ||
params.wrap_t == TextureConfig::ClampToBorder); params.wrap_t == TextureConfig::ClampToBorder);
const Common::Vec4f color = PicaToVK::ColorRGBA8(params.border_color); const auto color = PicaToVK::ColorRGBA8(params.border_color);
const vk::SamplerCustomBorderColorCreateInfoEXT border_color_info = { const vk::SamplerCustomBorderColorCreateInfoEXT border_color_info = {
.customBorderColor = MakeClearColorValue(color), .customBorderColor = MakeClearColorValue(color),
.format = vk::Format::eUndefined, .format = vk::Format::eUndefined,

18
src/video_core/renderer_vulkan/vk_texture_runtime.h
View file

@ -23,9 +23,8 @@ namespace Vulkan {
class Instance; class Instance;
class RenderManager; class RenderManager;
class DescriptorPool;
class DescriptorSetProvider;
class Surface; class Surface;
class DescriptorUpdateQueue;
struct Handle { struct Handle {
VmaAllocation alloc; VmaAllocation alloc;
@ -42,8 +41,8 @@ class TextureRuntime {
public: public:
explicit TextureRuntime(const Instance& instance, Scheduler& scheduler, explicit TextureRuntime(const Instance& instance, Scheduler& scheduler,
RenderManager& render_manager, DescriptorPool& pool, RenderManager& render_manager, DescriptorUpdateQueue& update_queue,
DescriptorSetProvider& texture_provider, u32 num_swapchain_images); u32 num_swapchain_images);
~TextureRuntime(); ~TextureRuntime();
const Instance& GetInstance() const { const Instance& GetInstance() const {
@ -74,7 +73,12 @@ public:
bool ClearTexture(Surface& surface, const VideoCore::TextureClear& clear); bool ClearTexture(Surface& surface, const VideoCore::TextureClear& clear);
/// Copies a rectangle of src_tex to another rectange of dst_rect /// Copies a rectangle of src_tex to another rectange of dst_rect
bool CopyTextures(Surface& source, Surface& dest, const VideoCore::TextureCopy& copy); bool CopyTextures(Surface& source, Surface& dest,
std::span<const VideoCore::TextureCopy> copies);
bool CopyTextures(Surface& source, Surface& dest, const VideoCore::TextureCopy& copy) {
return CopyTextures(source, dest, std::array{copy});
}
/// Blits a rectangle of src_tex to another rectange of dst_rect /// Blits a rectangle of src_tex to another rectange of dst_rect
bool BlitTextures(Surface& surface, Surface& dest, const VideoCore::TextureBlit& blit); bool BlitTextures(Surface& surface, Surface& dest, const VideoCore::TextureBlit& blit);
@ -85,9 +89,6 @@ public:
/// Returns true if the provided pixel format needs convertion /// Returns true if the provided pixel format needs convertion
bool NeedsConversion(VideoCore::PixelFormat format) const; bool NeedsConversion(VideoCore::PixelFormat format) const;
/// Removes any descriptor sets that contain the provided image view.
void FreeDescriptorSetsWithImage(vk::ImageView image_view);
private: private:
/// Clears a partial texture rect using a clear rectangle /// Clears a partial texture rect using a clear rectangle
void ClearTextureWithRenderpass(Surface& surface, const VideoCore::TextureClear& clear); void ClearTextureWithRenderpass(Surface& surface, const VideoCore::TextureClear& clear);
@ -96,7 +97,6 @@ private:
const Instance& instance; const Instance& instance;
Scheduler& scheduler; Scheduler& scheduler;
RenderManager& render_manager; RenderManager& render_manager;
DescriptorSetProvider& texture_provider;
BlitHelper blit_helper; BlitHelper blit_helper;
StreamBuffer upload_buffer; StreamBuffer upload_buffer;
StreamBuffer download_buffer; StreamBuffer download_buffer;

182
src/video_core/shader/generator/glsl_fs_shader_gen.cpp
View file

@ -106,7 +106,11 @@ FragmentModule::FragmentModule(const FSConfig& config_, const Profile& profile_)
out.reserve(RESERVE_SIZE); out.reserve(RESERVE_SIZE);
DefineExtensions(); DefineExtensions();
DefineInterface(); DefineInterface();
DefineBindings(); if (profile.is_vulkan) {
DefineBindingsVK();
} else {
DefineBindingsGL();
}
DefineHelpers(); DefineHelpers();
DefineShadowHelpers(); DefineShadowHelpers();
DefineLightingHelpers(); DefineLightingHelpers();
@ -1272,7 +1276,43 @@ void FragmentModule::DefineInterface() {
out += "layout (location = 0) out vec4 color;\n\n"; out += "layout (location = 0) out vec4 color;\n\n";
} }
void FragmentModule::DefineBindings() { void FragmentModule::DefineBindingsVK() {
// Uniform and texture buffers
out += FSUniformBlockDef;
out += "layout(set = 0, binding = 3) uniform samplerBuffer texture_buffer_lut_lf;\n";
out += "layout(set = 0, binding = 4) uniform samplerBuffer texture_buffer_lut_rg;\n";
out += "layout(set = 0, binding = 5) uniform samplerBuffer texture_buffer_lut_rgba;\n\n";
// Texture samplers
const auto texture_type = config.texture.texture0_type.Value();
const auto sampler_tex0 = [&] {
switch (texture_type) {
case TextureType::Shadow2D:
case TextureType::ShadowCube:
return "usampler2D";
case TextureType::TextureCube:
return "samplerCube";
default:
return "sampler2D";
}
}();
for (u32 i = 0; i < 3; i++) {
const auto sampler = i == 0 ? sampler_tex0 : "sampler2D";
const auto num_descriptors = i == 0 && texture_type == TextureType::ShadowCube ? "[6]" : "";
out += fmt::format("layout(set = 1, binding = {0}) uniform {1} tex{0}{2};\n", i, sampler,
num_descriptors);
}
// Utility textures
if (config.framebuffer.shadow_rendering) {
out += "layout(set = 2, binding = 0, r32ui) uniform uimage2D shadow_buffer;\n\n";
}
if (config.user.use_custom_normal) {
out += "layout(set = 2, binding = 1) uniform sampler2D tex_normal;\n";
}
}
void FragmentModule::DefineBindingsGL() {
// Uniform and texture buffers // Uniform and texture buffers
out += FSUniformBlockDef; out += FSUniformBlockDef;
out += "layout(binding = 3) uniform samplerBuffer texture_buffer_lut_lf;\n"; out += "layout(binding = 3) uniform samplerBuffer texture_buffer_lut_lf;\n";
@ -1280,33 +1320,32 @@ void FragmentModule::DefineBindings() {
out += "layout(binding = 5) uniform samplerBuffer texture_buffer_lut_rgba;\n\n"; out += "layout(binding = 5) uniform samplerBuffer texture_buffer_lut_rgba;\n\n";
// Texture samplers // Texture samplers
const auto texunit_set = profile.is_vulkan ? "set = 1, " : "";
const auto texture_type = config.texture.texture0_type.Value(); const auto texture_type = config.texture.texture0_type.Value();
for (u32 i = 0; i < 3; i++) { for (u32 i = 0; i < 3; i++) {
const auto sampler = const auto sampler =
i == 0 && texture_type == TextureType::TextureCube ? "samplerCube" : "sampler2D"; i == 0 && texture_type == TextureType::TextureCube ? "samplerCube" : "sampler2D";
out += out += fmt::format("layout(binding = {0}) uniform {1} tex{0};\n", i, sampler);
fmt::format("layout({0}binding = {1}) uniform {2} tex{1};\n", texunit_set, i, sampler);
} }
if (config.user.use_custom_normal && !profile.is_vulkan) { // Utility textures
if (config.user.use_custom_normal) {
out += "layout(binding = 6) uniform sampler2D tex_normal;\n"; out += "layout(binding = 6) uniform sampler2D tex_normal;\n";
} }
if (use_blend_fallback && !profile.is_vulkan) { if (use_blend_fallback) {
out += "layout(location = 7) uniform sampler2D tex_color;\n"; out += "layout(location = 7) uniform sampler2D tex_color;\n";
} }
// Storage images // Shadow textures
static constexpr std::array postfixes = {"px", "nx", "py", "ny", "pz", "nz"}; if (texture_type == TextureType::Shadow2D || texture_type == TextureType::ShadowCube) {
const auto shadow_set = profile.is_vulkan ? "set = 2, " : ""; static constexpr std::array postfixes = {"px", "nx", "py", "ny", "pz", "nz"};
for (u32 i = 0; i < postfixes.size(); i++) { for (u32 i = 0; i < postfixes.size(); i++) {
out += fmt::format( out += fmt::format(
"layout({}binding = {}, r32ui) uniform readonly uimage2D shadow_texture_{};\n", "layout(binding = {}, r32ui) uniform readonly uimage2D shadow_texture_{};\n", i,
shadow_set, i, postfixes[i]); postfixes[i]);
}
} }
if (config.framebuffer.shadow_rendering) { if (config.framebuffer.shadow_rendering) {
out += fmt::format("layout({}binding = 6, r32ui) uniform uimage2D shadow_buffer;\n\n", out += "layout(binding = 6, r32ui) uniform uimage2D shadow_buffer;\n\n";
shadow_set);
} }
} }
@ -1414,19 +1453,48 @@ float mix2(vec4 s, vec2 a) {
)"; )";
if (config.texture.texture0_type == TexturingRegs::TextureConfig::Shadow2D) { if (config.texture.texture0_type == TexturingRegs::TextureConfig::Shadow2D) {
out += R"( if (profile.is_vulkan) {
out += R"(
float SampleShadow2D(ivec2 uv, uint z) { float SampleShadow2D(ivec2 uv, uint z) {
if (any(bvec4( lessThan(uv, ivec2(0)), greaterThanEqual(uv, imageSize(shadow_texture_px)) ))) if (any(bvec4(lessThan(uv, ivec2(0)), greaterThanEqual(uv, textureSize(tex0, 0)))))
return 1.0;
return CompareShadow(texelFetch(tex0, uv, 0).x, z);
}
vec4 shadowTexture(vec2 uv, float w) {
)";
if (!config.texture.shadow_texture_orthographic) {
out += "uv /= w;";
}
out += R"(
uint z = uint(max(0, int(min(abs(w), 1.0) * float(0xFFFFFF)) - shadow_texture_bias));
vec2 coord = vec2(textureSize(tex0, 0)) * uv - vec2(0.5);
vec2 coord_floor = floor(coord);
vec2 f = coord - coord_floor;
ivec2 i = ivec2(coord_floor);
vec4 s = vec4(
SampleShadow2D(i , z),
SampleShadow2D(i + ivec2(1, 0), z),
SampleShadow2D(i + ivec2(0, 1), z),
SampleShadow2D(i + ivec2(1, 1), z));
return vec4(mix2(s, f));
}
)";
} else {
out += R"(
float SampleShadow2D(ivec2 uv, uint z) {
if (any(bvec4(lessThan(uv, ivec2(0)), greaterThanEqual(uv, imageSize(shadow_texture_px)))))
return 1.0; return 1.0;
return CompareShadow(imageLoad(shadow_texture_px, uv).x, z); return CompareShadow(imageLoad(shadow_texture_px, uv).x, z);
} }
vec4 shadowTexture(vec2 uv, float w) { vec4 shadowTexture(vec2 uv, float w) {
)"; )";
if (!config.texture.shadow_texture_orthographic) { if (!config.texture.shadow_texture_orthographic) {
out += "uv /= w;"; out += "uv /= w;";
} }
out += R"( out += R"(
uint z = uint(max(0, int(min(abs(w), 1.0) * float(0xFFFFFF)) - shadow_texture_bias)); uint z = uint(max(0, int(min(abs(w), 1.0) * float(0xFFFFFF)) - shadow_texture_bias));
vec2 coord = vec2(imageSize(shadow_texture_px)) * uv - vec2(0.5); vec2 coord = vec2(imageSize(shadow_texture_px)) * uv - vec2(0.5);
vec2 coord_floor = floor(coord); vec2 coord_floor = floor(coord);
@ -1440,8 +1508,75 @@ vec4 shadowTexture(vec2 uv, float w) {
return vec4(mix2(s, f)); return vec4(mix2(s, f));
} }
)"; )";
}
} else if (config.texture.texture0_type == TexturingRegs::TextureConfig::ShadowCube) { } else if (config.texture.texture0_type == TexturingRegs::TextureConfig::ShadowCube) {
out += R"( if (profile.is_vulkan) {
out += R"(
uvec4 SampleShadowCube(int face, ivec2 i00, ivec2 i10, ivec2 i01, ivec2 i11) {
return uvec4(
texelFetch(tex0[face], i00, 0).r,
texelFetch(tex0[face], i10, 0).r,
texelFetch(tex0[face], i01, 0).r,
texelFetch(tex0[face], i11, 0).r);
}
vec4 shadowTextureCube(vec2 uv, float w) {
ivec2 size = textureSize(tex0[0], 0);
vec3 c = vec3(uv, w);
vec3 a = abs(c);
if (a.x > a.y && a.x > a.z) {
w = a.x;
uv = -c.zy;
if (c.x < 0.0) uv.x = -uv.x;
} else if (a.y > a.z) {
w = a.y;
uv = c.xz;
if (c.y < 0.0) uv.y = -uv.y;
} else {
w = a.z;
uv = -c.xy;
if (c.z > 0.0) uv.x = -uv.x;
}
uint z = uint(max(0, int(min(w, 1.0) * float(0xFFFFFF)) - shadow_texture_bias));
vec2 coord = vec2(size) * (uv / w * vec2(0.5) + vec2(0.5)) - vec2(0.5);
vec2 coord_floor = floor(coord);
vec2 f = coord - coord_floor;
ivec2 i00 = ivec2(coord_floor);
ivec2 i10 = i00 + ivec2(1, 0);
ivec2 i01 = i00 + ivec2(0, 1);
ivec2 i11 = i00 + ivec2(1, 1);
ivec2 cmin = ivec2(0), cmax = size - ivec2(1, 1);
i00 = clamp(i00, cmin, cmax);
i10 = clamp(i10, cmin, cmax);
i01 = clamp(i01, cmin, cmax);
i11 = clamp(i11, cmin, cmax);
uvec4 pixels;
if (a.x > a.y && a.x > a.z) {
if (c.x > 0.0)
pixels = SampleShadowCube(0, i00, i10, i01, i11);
else
pixels = SampleShadowCube(1, i00, i10, i01, i11);
} else if (a.y > a.z) {
if (c.y > 0.0)
pixels = SampleShadowCube(2, i00, i10, i01, i11);
else
pixels = SampleShadowCube(3, i00, i10, i01, i11);
} else {
if (c.z > 0.0)
pixels = SampleShadowCube(4, i00, i10, i01, i11);
else
pixels = SampleShadowCube(5, i00, i10, i01, i11);
}
vec4 s = vec4(
CompareShadow(pixels.x, z),
CompareShadow(pixels.y, z),
CompareShadow(pixels.z, z),
CompareShadow(pixels.w, z));
return vec4(mix2(s, f));
}
)";
} else {
out += R"(
vec4 shadowTextureCube(vec2 uv, float w) { vec4 shadowTextureCube(vec2 uv, float w) {
ivec2 size = imageSize(shadow_texture_px); ivec2 size = imageSize(shadow_texture_px);
vec3 c = vec3(uv, w); vec3 c = vec3(uv, w);
@ -1523,6 +1658,7 @@ vec4 shadowTextureCube(vec2 uv, float w) {
return vec4(mix2(s, f)); return vec4(mix2(s, f));
} }
)"; )";
}
} }
} }
} }

3
src/video_core/shader/generator/glsl_fs_shader_gen.h
View file

@ -74,7 +74,8 @@ private:
void DefineExtensions(); void DefineExtensions();
void DefineInterface(); void DefineInterface();
void DefineBindings(); void DefineBindingsVK();
void DefineBindingsGL();
void DefineHelpers(); void DefineHelpers();
void DefineLightingHelpers(); void DefineLightingHelpers();
void DefineShadowHelpers(); void DefineShadowHelpers();