mirror of
https://github.com/KytyPS5/KytyPS5.git
synced 2026-10-27 16:02:37 +00:00
pipeline: create pipelines from GraphicsPipelineKey and shader modules
This commit is contained in:
nmzik committed
1 parent
4b651f2ba7
commit
bfc377627a
13 files changed
+594
-695
No files matched your search
@@ -13,7 +13,6 @@
|
||||
#include "graphics/host_gpu/renderer/renderContext.h"
|
||||
#include "graphics/host_gpu/renderer/renderTarget.h"
|
||||
#include "graphics/host_gpu/vulkanCommon.h"
|
||||
#include "graphics/shader/recompiler/BufferFormat.h"
|
||||
#include "graphics/shader/recompiler/ir/ShaderIR.h"
|
||||
#include "graphics/shader/rectListShader.h"
|
||||
#include "graphics/shader/shader.h"
|
||||
@@ -21,124 +20,11 @@
|
||||
#include <algorithm>
|
||||
#include <array>
|
||||
#include <bit>
|
||||
#include <limits>
|
||||
#include <span>
|
||||
#include <vector>
|
||||
|
||||
namespace Libs::Graphics {
|
||||
|
||||
// IDK: maybe we can remove it?
|
||||
constexpr uint8_t kTemporaryVertexAttribFormat113 =
|
||||
static_cast<uint8_t>(Prospero::VertexAttribFormat::k16_16SInt);
|
||||
constexpr uint32_t kTemporaryPs5BufferFormat121 = 121u;
|
||||
|
||||
static bool NarrowInputFormat(vk::Format& format, uint32_t& size, uint32_t used_components) {
|
||||
if (used_components == 0 || used_components >= size) {
|
||||
return false;
|
||||
}
|
||||
|
||||
switch (format) {
|
||||
case vk::Format::eR32G32B32A32Sfloat:
|
||||
switch (used_components) {
|
||||
case 1: format = vk::Format::eR32Sfloat; break;
|
||||
case 2: format = vk::Format::eR32G32Sfloat; break;
|
||||
case 3: format = vk::Format::eR32G32B32Sfloat; break;
|
||||
default: return false;
|
||||
}
|
||||
size = used_components;
|
||||
return true;
|
||||
case vk::Format::eR32G32B32Sfloat:
|
||||
switch (used_components) {
|
||||
case 1: format = vk::Format::eR32Sfloat; break;
|
||||
case 2: format = vk::Format::eR32G32Sfloat; break;
|
||||
default: return false;
|
||||
}
|
||||
size = used_components;
|
||||
return true;
|
||||
case vk::Format::eR16G16B16A16Sfloat:
|
||||
switch (used_components) {
|
||||
case 1: format = vk::Format::eR16Sfloat; break;
|
||||
case 2: format = vk::Format::eR16G16Sfloat; break;
|
||||
default: return false;
|
||||
}
|
||||
size = used_components;
|
||||
return true;
|
||||
case vk::Format::eR8G8B8A8Unorm:
|
||||
switch (used_components) {
|
||||
case 1: format = vk::Format::eR8Unorm; break;
|
||||
case 2: format = vk::Format::eR8G8Unorm; break;
|
||||
default: return false;
|
||||
}
|
||||
size = used_components;
|
||||
return true;
|
||||
case vk::Format::eR8G8B8A8Snorm:
|
||||
if (used_components != 2) {
|
||||
return false;
|
||||
}
|
||||
format = vk::Format::eR8G8Snorm;
|
||||
size = 2;
|
||||
return true;
|
||||
case vk::Format::eR8G8B8A8Uint:
|
||||
switch (used_components) {
|
||||
case 1: format = vk::Format::eR8Uint; break;
|
||||
case 2: format = vk::Format::eR8G8Uint; break;
|
||||
default: return false;
|
||||
}
|
||||
size = used_components;
|
||||
return true;
|
||||
default: break;
|
||||
}
|
||||
|
||||
return false;
|
||||
}
|
||||
|
||||
static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format, uint32_t& size,
|
||||
uint32_t used_components) {
|
||||
const auto fmt = res.Format();
|
||||
const auto raw_format = res.RawFormat();
|
||||
if (raw_format == kTemporaryVertexAttribFormat113) {
|
||||
static bool logged_113 = false;
|
||||
if (!logged_113) {
|
||||
LOGF("InputFormat: temporary: accepting invalid PS5 buffer format 113 as "
|
||||
"vk::Format::eR32G32B32A32Sfloat\n");
|
||||
logged_113 = true;
|
||||
}
|
||||
format = vk::Format::eR32G32B32A32Sfloat;
|
||||
size = 4;
|
||||
if (NarrowInputFormat(format, size, used_components)) {
|
||||
LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", raw_format,
|
||||
vk::to_string(format).c_str(), used_components);
|
||||
}
|
||||
return;
|
||||
}
|
||||
if (raw_format == kTemporaryPs5BufferFormat121) {
|
||||
static bool logged_121 = false;
|
||||
if (!logged_121) {
|
||||
LOGF("InputFormat: accepting PS5 buffer format 121 as vk::Format::eR16G16Sfloat\n");
|
||||
logged_121 = true;
|
||||
}
|
||||
format = vk::Format::eR16G16Sfloat;
|
||||
size = 2;
|
||||
return;
|
||||
}
|
||||
|
||||
format = VulkanFormat(fmt);
|
||||
size = ShaderRecompiler::Format::GetFormatInfo(fmt).component_count;
|
||||
if (format == vk::Format::eUndefined || size == 0) {
|
||||
EXIT("unknown vertex format: fmt = %u\n", raw_format);
|
||||
}
|
||||
|
||||
if (NarrowInputFormat(format, size, used_components)) {
|
||||
static std::atomic<uint64_t> log_count = 0;
|
||||
auto log_id = log_count.fetch_add(1);
|
||||
if (log_id < 32) {
|
||||
LOGF("VertexInput: narrowed vertex format to %" PRIu32
|
||||
" component(s) for shader fetch\n",
|
||||
used_components);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
static vk::BlendFactor GetBlendFactor(uint32_t factor) {
|
||||
switch (static_cast<Prospero::BlendFactor>(factor)) {
|
||||
case Prospero::BlendFactor::kZero: return vk::BlendFactor::eZero;
|
||||
@@ -242,30 +128,27 @@ static void CreateDescriptorLayout(GraphicContext& graphics, PipelineCache::Pipe
|
||||
|
||||
// NOLINTNEXTLINE(readability-function-cognitive-complexity)
|
||||
void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
|
||||
const PipelineRenderingState& rendering,
|
||||
const PipelineVertexInputState& vertex_input,
|
||||
const DrawRenderState& state,
|
||||
const PipelineStaticParameters& static_params,
|
||||
vk::PipelineCache driver_cache) {
|
||||
const auto vertex_info = state.VertexStages();
|
||||
const auto& vs_input_info = vertex_info.front();
|
||||
const auto* ps_input_info = state.ps_active ? &state.ps_input_info : nullptr;
|
||||
const auto& vertex_program = *vs_input_info.stage.handle;
|
||||
const auto* pixel_program = state.ps_input_info.stage.handle;
|
||||
const bool tessellation = vertex_info.size() == 3;
|
||||
const bool ps_active = state.ps_active;
|
||||
const bool with_depth = rendering.depth_format != vk::Format::eUndefined ||
|
||||
rendering.stencil_format != vk::Format::eUndefined;
|
||||
const bool mesh = vs_input_info.stage.program->stage == ShaderType::Mesh;
|
||||
const GraphicsPipelineKey& key,
|
||||
const std::array<const ShaderModule*, 4>& stages,
|
||||
vk::PipelineCache driver_cache) {
|
||||
const auto& rendering = key.rendering;
|
||||
const auto& vertex_input = key.vertex_input;
|
||||
const auto& static_params = key.static_params;
|
||||
const auto& vertex = *stages[0];
|
||||
const auto* pixel = stages[3];
|
||||
const bool with_depth = rendering.depth_format != vk::Format::eUndefined ||
|
||||
rendering.stencil_format != vk::Format::eUndefined;
|
||||
const bool mesh = vertex.info.stage == ShaderType::Mesh;
|
||||
EXIT_NOT_IMPLEMENTED(mesh && !graphics.mesh_shader_enabled);
|
||||
const bool rect_list =
|
||||
!mesh && !tessellation && static_params.topology == vk::PrimitiveTopology::ePatchList;
|
||||
const bool rect_list = static_params.patch_control_points != 0 && stages[1] == nullptr;
|
||||
|
||||
vk::ShaderModule tess_control_shader_module = nullptr;
|
||||
vk::ShaderModule tess_eval_shader_module = nullptr;
|
||||
|
||||
if (rect_list) {
|
||||
const auto shaders = BuildRectListShaders(vs_input_info, ps_input_info);
|
||||
std::span<const RectListParameter> parameters;
|
||||
if (pixel != nullptr) parameters = pixel->rect_parameters;
|
||||
const auto shaders = BuildRectListShaders(parameters, vertex.info.param_export_mask);
|
||||
tess_control_shader_module = CompileSPV(shaders.control, graphics.device);
|
||||
if (graphics_debug_dump_enabled()) {
|
||||
LOGF("PipelineTrace: vkCreateShaderModule RectList TCS done module=%p\n",
|
||||
@@ -284,11 +167,12 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
|
||||
|
||||
vk::PipelineShaderStageCreateInfo shader_stages[4] {};
|
||||
uint32_t shader_stage_count = 0;
|
||||
for (uint32_t i = 0; i < vertex_info.size(); i++) {
|
||||
shader_stages[shader_stage_count++] = {.stage =
|
||||
NativeShaderStage(vertex_info[i].logical_stage),
|
||||
.module = vertex_info[i].stage.handle->module,
|
||||
.pName = "main"};
|
||||
for (const auto* stage: stages) {
|
||||
if (stage != nullptr) {
|
||||
shader_stages[shader_stage_count++] = {.stage = NativeShaderStage(stage->info.stage),
|
||||
.module = stage->module,
|
||||
.pName = "main"};
|
||||
}
|
||||
}
|
||||
if (rect_list) {
|
||||
shader_stages[shader_stage_count++] = {.stage =
|
||||
@@ -300,11 +184,6 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
|
||||
.module = tess_eval_shader_module,
|
||||
.pName = "main"};
|
||||
}
|
||||
if (ps_active) {
|
||||
shader_stages[shader_stage_count++] = {.stage = vk::ShaderStageFlagBits::eFragment,
|
||||
.module = pixel_program->module,
|
||||
.pName = "main"};
|
||||
}
|
||||
|
||||
vk::VertexInputAttributeDescription input_attr[ShaderVertexInputInfo::RES_MAX] {};
|
||||
vk::VertexInputBindingDescription input_desc[ShaderVertexInputInfo::RES_MAX] {};
|
||||
@@ -320,46 +199,7 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
|
||||
input_attr[index].binding = vertex_input.attributes[index].binding;
|
||||
input_attr[index].location = index;
|
||||
input_attr[index].offset = vertex_input.attributes[index].offset;
|
||||
|
||||
uint32_t attr_size = 4;
|
||||
const auto registers_num = vs_input_info.resources_dst[index].registers_num;
|
||||
const auto compiled_components =
|
||||
vs_input_info.stage.program->info.vertex_fetch_components[index];
|
||||
const auto used_components =
|
||||
compiled_components > 0 ? static_cast<int>(compiled_components) : registers_num;
|
||||
GetInputFormat(vs_input_info.resources[index], input_attr[index].format, attr_size,
|
||||
static_cast<uint32_t>(used_components));
|
||||
|
||||
if (graphics_debug_dump_enabled()) {
|
||||
static std::atomic_uint log_count = 0;
|
||||
const auto log_id = log_count.fetch_add(1, std::memory_order_relaxed);
|
||||
if (log_id < 128) {
|
||||
LOGF("VertexInputState[%u]: attr=%u binding=%u offset=%u stride=%u fmt=%d "
|
||||
"src_fmt=%u dst=v%u regs=%u"
|
||||
" fetched_components=%u attr_size=%u swizzle=%u,%u,%u,%u\n",
|
||||
log_id, index, input_attr[index].binding, input_attr[index].offset,
|
||||
input_desc[input_attr[index].binding].stride,
|
||||
static_cast<int>(input_attr[index].format),
|
||||
static_cast<uint32_t>(vs_input_info.resources[index].Format()),
|
||||
static_cast<uint32_t>(vs_input_info.resources_dst[index].register_start),
|
||||
static_cast<uint32_t>(registers_num), static_cast<uint32_t>(used_components),
|
||||
attr_size, static_cast<uint32_t>(vs_input_info.resources[index].DstSelX()),
|
||||
static_cast<uint32_t>(vs_input_info.resources[index].DstSelY()),
|
||||
static_cast<uint32_t>(vs_input_info.resources[index].DstSelZ()),
|
||||
static_cast<uint32_t>(vs_input_info.resources[index].DstSelW()));
|
||||
}
|
||||
}
|
||||
|
||||
if (vs_input_info.resources[index].OutOfBounds() != 0) {
|
||||
static bool logged = false;
|
||||
if (!logged) {
|
||||
LOGF("VertexInput: temporary: accepting PS5 out-of-bounds behavior %" PRIu8 "\n",
|
||||
vs_input_info.resources[index].OutOfBounds());
|
||||
logged = true;
|
||||
}
|
||||
}
|
||||
|
||||
EXIT_IF(registers_num < 1 || registers_num > 4);
|
||||
input_attr[index].format = vertex.vertex_formats[index];
|
||||
}
|
||||
|
||||
vk::PipelineVertexInputStateCreateInfo vertex_input_info {};
|
||||
@@ -467,16 +307,13 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
|
||||
|
||||
std::vector<vk::DescriptorSetLayoutBinding> descriptor_bindings;
|
||||
vk::ShaderStageFlags graphics_stages = vk::ShaderStageFlagBits::eFragment;
|
||||
for (const auto& stage: vertex_info) {
|
||||
const auto native_stage = NativeShaderStage(stage.logical_stage);
|
||||
AddLayoutBindings(descriptor_bindings, *stage.stage.program, native_stage);
|
||||
pipeline.uses_image_heap |= stage.stage.program->info.uses_image_heap;
|
||||
graphics_stages |= native_stage;
|
||||
}
|
||||
if (ps_active) {
|
||||
AddLayoutBindings(descriptor_bindings, *ps_input_info->stage.program,
|
||||
vk::ShaderStageFlagBits::eFragment);
|
||||
pipeline.uses_image_heap |= ps_input_info->stage.program->info.uses_image_heap;
|
||||
for (const auto* stage: stages) {
|
||||
if (stage != nullptr) {
|
||||
const auto native_stage = NativeShaderStage(stage->info.stage);
|
||||
AddLayoutBindings(descriptor_bindings, stage->info, native_stage);
|
||||
pipeline.uses_image_heap |= stage->info.info.uses_image_heap;
|
||||
graphics_stages |= native_stage;
|
||||
}
|
||||
}
|
||||
CreateDescriptorLayout(graphics, pipeline, descriptor_bindings);
|
||||
const vk::PushConstantRange push_constants {graphics_stages, 0,
|
||||
@@ -494,7 +331,7 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
|
||||
if (graphics_debug_dump_enabled()) {
|
||||
LOGF("PipelineTrace: vkCreatePipelineLayout begin VS=%" PRIu64 " PS=%" PRIu64
|
||||
" set_layouts=1 push_constants=%" PRIu32 "\n",
|
||||
vertex_program.id, ps_active ? pixel_program->id : 0, 1u);
|
||||
key.stage_ids[0], key.stage_ids[3], 1u);
|
||||
}
|
||||
auto result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
|
||||
&pipeline.pipeline_layout);
|
||||
@@ -504,8 +341,6 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
|
||||
}
|
||||
EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
|
||||
|
||||
EXIT_NOT_IMPLEMENTED(pipeline.pipeline_layout == nullptr);
|
||||
|
||||
vk::PipelineDepthStencilStateCreateInfo depth_stencil_info {};
|
||||
|
||||
std::vector<vk::DynamicState> dynamic_states {
|
||||
@@ -551,9 +386,9 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
|
||||
pipeline_info.pVertexInputState = mesh ? nullptr : &vertex_input_info;
|
||||
pipeline_info.pInputAssemblyState = mesh ? nullptr : &input_assembly;
|
||||
vk::PipelineTessellationStateCreateInfo tessellation_state {};
|
||||
tessellation_state.patchControlPoints =
|
||||
tessellation ? vs_input_info.tess.input_control_points : 3u;
|
||||
pipeline_info.pTessellationState = (rect_list || tessellation) ? &tessellation_state : nullptr;
|
||||
tessellation_state.patchControlPoints = static_params.patch_control_points;
|
||||
pipeline_info.pTessellationState =
|
||||
static_params.patch_control_points != 0 ? &tessellation_state : nullptr;
|
||||
pipeline_info.pViewportState = &viewport_state;
|
||||
pipeline_info.pRasterizationState = &rasterizer;
|
||||
pipeline_info.pMultisampleState = &multisampling;
|
||||
@@ -569,7 +404,7 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
|
||||
LOGF("PipelineTrace: vkCreateGraphicsPipelines begin VS=%" PRIu64 " PS=%" PRIu64
|
||||
" topology=%" PRIu32 " color_mask=0x%08" PRIx32
|
||||
" depth=%s blend=%s dyn_states=%" PRIu32 "\n",
|
||||
vertex_program.id, ps_active ? pixel_program->id : 0,
|
||||
key.stage_ids[0], key.stage_ids[3],
|
||||
static_cast<uint32_t>(static_params.topology), static_params.color_mask[0],
|
||||
(with_depth ? "true" : "false"), (static_params.blend_enable[0] ? "true" : "false"),
|
||||
dynamic_state.dynamicStateCount);
|
||||
@@ -582,29 +417,18 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
|
||||
}
|
||||
EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
|
||||
|
||||
EXIT_NOT_IMPLEMENTED(pipeline.pipeline == nullptr);
|
||||
|
||||
if (tess_control_shader_module != nullptr) {
|
||||
graphics.device.destroyShaderModule(tess_control_shader_module, nullptr);
|
||||
}
|
||||
if (tess_eval_shader_module != nullptr) {
|
||||
graphics.device.destroyShaderModule(tess_eval_shader_module, nullptr);
|
||||
}
|
||||
graphics.device.destroyShaderModule(tess_control_shader_module, nullptr);
|
||||
graphics.device.destroyShaderModule(tess_eval_shader_module, nullptr);
|
||||
}
|
||||
|
||||
// NOLINTNEXTLINE(readability-function-cognitive-complexity)
|
||||
void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
|
||||
const ShaderComputeInputInfo& input_info,
|
||||
vk::ShaderModule compute_module, vk::PipelineCache driver_cache) {
|
||||
EXIT_IF(compute_module == nullptr);
|
||||
|
||||
const ShaderModule& module, vk::PipelineCache driver_cache) {
|
||||
vk::PipelineShaderStageCreateInfo comp_shader_stage_info {};
|
||||
vk::PipelineShaderStageRequiredSubgroupSizeCreateInfo comp_subgroup_size {};
|
||||
comp_shader_stage_info.stage = vk::ShaderStageFlagBits::eCompute;
|
||||
comp_shader_stage_info.module = compute_module;
|
||||
comp_shader_stage_info.module = module.module;
|
||||
comp_shader_stage_info.pName = "main";
|
||||
EXIT_IF(!input_info.stage);
|
||||
const auto wave_size = input_info.stage.program->wave_size;
|
||||
const auto wave_size = module.info.wave_size;
|
||||
if (graphics.compute_subgroup_size_control_enabled &&
|
||||
wave_size >= graphics.min_subgroup_size && wave_size <= graphics.max_subgroup_size) {
|
||||
comp_subgroup_size.requiredSubgroupSize = wave_size;
|
||||
@@ -612,9 +436,8 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
|
||||
}
|
||||
|
||||
std::vector<vk::DescriptorSetLayoutBinding> descriptor_bindings;
|
||||
AddLayoutBindings(descriptor_bindings, *input_info.stage.program,
|
||||
vk::ShaderStageFlagBits::eCompute);
|
||||
pipeline.uses_image_heap = input_info.stage.program->info.uses_image_heap;
|
||||
AddLayoutBindings(descriptor_bindings, module.info, vk::ShaderStageFlagBits::eCompute);
|
||||
pipeline.uses_image_heap = module.info.info.uses_image_heap;
|
||||
CreateDescriptorLayout(graphics, pipeline, descriptor_bindings);
|
||||
const vk::PushConstantRange push_constants {vk::ShaderStageFlagBits::eCompute, 0,
|
||||
ShaderRecompiler::IR::NativePushConstantSize};
|
||||
@@ -628,18 +451,12 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
|
||||
|
||||
EXIT_IF(pipeline.pipeline_layout != nullptr);
|
||||
|
||||
LOGF("PipelineTrace: vkCreatePipelineLayout CS begin set_layouts=1 push_constants=%u\n",
|
||||
1u);
|
||||
auto result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
|
||||
&pipeline.pipeline_layout);
|
||||
LOGF("PipelineTrace: vkCreatePipelineLayout CS done result=%s layout=%p\n",
|
||||
vk::to_string(result).c_str(), static_cast<void*>(pipeline.pipeline_layout));
|
||||
EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
|
||||
|
||||
EXIT_NOT_IMPLEMENTED(pipeline.pipeline_layout == nullptr);
|
||||
|
||||
vk::ComputePipelineCreateInfo info {};
|
||||
if (input_info.stage.program->info.uses_gds) {
|
||||
if (module.info.info.uses_gds) {
|
||||
info.flags = vk::PipelineCreateFlagBits::eDispatchBase;
|
||||
}
|
||||
info.stage = comp_shader_stage_info;
|
||||
@@ -648,15 +465,9 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& p
|
||||
|
||||
EXIT_IF(pipeline.pipeline != nullptr);
|
||||
|
||||
LOGF("PipelineTrace: vkCreateComputePipelines begin layout=%p\n",
|
||||
static_cast<void*>(pipeline.pipeline_layout));
|
||||
result = graphics.device.createComputePipelines(driver_cache, 1, &info, nullptr,
|
||||
&pipeline.pipeline);
|
||||
LOGF("PipelineTrace: vkCreateComputePipelines done result=%s pipeline=%p\n",
|
||||
vk::to_string(result).c_str(), static_cast<void*>(pipeline.pipeline));
|
||||
EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
|
||||
|
||||
EXIT_NOT_IMPLEMENTED(pipeline.pipeline == nullptr);
|
||||
}
|
||||
|
||||
} // namespace Libs::Graphics
|
||||
Reference in new issue
Block a user