The-Homebrew-Cloud/skyline
2
0
Fork 0
mirror of https://github.com/skyline-emu/skyline.git synced 2024-11-04 23:55:08 +01:00
Code Issues Packages Projects Releases Wiki Activity

Optimize descriptor set writes using push descriptors

Browse Source 
We can use inline push descriptors for writing to descriptor rather than allocating a descriptor set for a one time write and freeing it as this is rather inefficient while an inline push descriptor generally ends up being a direct `memcpy` on the driver side designed for this use-case.
This commit is contained in:
PixelyIon 2022-04-24 13:45:09 +05:30
parent 5adafbff04
commit 50a8b69f7b
3 changed files with 15 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
app/src/main/cpp/skyline/gpu/interconnect/graphics_context.h
View File

@ -1170,6 +1170,7 @@ namespace skyline::gpu::interconnect {
std::move(shaderModules),
std::move(shaderStages),
vk::raii::DescriptorSetLayout(gpu.vkDevice, vk::DescriptorSetLayoutCreateInfo{
.flags = gpu.traits.supportsPushDescriptors ? vk::DescriptorSetLayoutCreateFlagBits::ePushDescriptorKHR : vk::DescriptorSetLayoutCreateFlags{},
.pBindings = layoutBindings.data(),
.bindingCount = static_cast<u32>(layoutBindings.size()),
}),
@ -2847,7 +2848,7 @@ namespace skyline::gpu::interconnect {
auto fenceStorage{std::make_shared<FenceStorage>(std::move(descriptorSet))};
// Submit Draw
executor.AddSubpass([=, &vkDevice = gpu.vkDevice, shaderModules = programState.shaderModules, shaderStages = programState.shaderStages, inputAssemblyState = inputAssemblyState, multiViewport = gpu.traits.supportsMultipleViewports, viewports = viewports, scissors = scissors, rasterizerState = rasterizerState, multisampleState = multisampleState, depthState = depthState, blendState = blendState, drawStorage = std::move(drawStorage), fenceStorage = std::move(fenceStorage), supportsVertexAttributeDivisor = gpu.traits.supportsVertexAttributeDivisor, pipelineCache = *pipelineCache](vk::raii::CommandBuffer &commandBuffer, const std::shared_ptr<FenceCycle> &cycle, GPU &, vk::RenderPass renderPass, u32 subpassIndex) mutable {
executor.AddSubpass([=, &vkDevice = gpu.vkDevice, shaderModules = programState.shaderModules, shaderStages = programState.shaderStages, inputAssemblyState = inputAssemblyState, multiViewport = gpu.traits.supportsMultipleViewports, viewports = viewports, scissors = scissors, rasterizerState = rasterizerState, multisampleState = multisampleState, depthState = depthState, blendState = blendState, drawStorage = std::move(drawStorage), fenceStorage = std::move(fenceStorage), supportsVertexAttributeDivisor = gpu.traits.supportsVertexAttributeDivisor, supportsPushDescriptors = gpu.traits.supportsPushDescriptors, pipelineCache = *pipelineCache](vk::raii::CommandBuffer &commandBuffer, const std::shared_ptr<FenceCycle> &cycle, GPU &, vk::RenderPass renderPass, u32 subpassIndex) mutable {
vk::StructureChain<vk::PipelineVertexInputStateCreateInfo, vk::PipelineVertexInputDivisorStateCreateInfoEXT> vertexState{
vk::PipelineVertexInputStateCreateInfo{
.pVertexBindingDescriptions = vertexBindingDescriptions.data(),
@ -2909,8 +2910,15 @@ namespace skyline::gpu::interconnect {
}
}
vkDevice.updateDescriptorSets(**drawStorage->descriptorSetWrites, nullptr);
commandBuffer.bindDescriptorSets(vk::PipelineBindPoint::eGraphics, *drawStorage->pipelineLayout, 0, fenceStorage->descriptorSet, nullptr);
if (supportsPushDescriptors) {
commandBuffer.pushDescriptorSetKHR(vk::PipelineBindPoint::eGraphics, *drawStorage->pipelineLayout, 0, **drawStorage->descriptorSetWrites);
} else {
auto descriptorSet{gpu.descriptor.AllocateSet(*drawStorage->descriptorSetLayout)};
for (auto &descriptorSetWrite : **drawStorage->descriptorSetWrites)
descriptorSetWrite.dstSet = descriptorSet;
vkDevice.updateDescriptorSets(**drawStorage->descriptorSetWrites, nullptr);
commandBuffer.bindDescriptorSets(vk::PipelineBindPoint::eGraphics, *drawStorage->pipelineLayout, 0, descriptorSet, nullptr);
}
if constexpr (IsIndexed) {
commandBuffer.bindIndexBuffer(boundIndexBuffer->handle, boundIndexBuffer->offset, boundIndexBuffer->type);

5
app/src/main/cpp/skyline/gpu/trait_manager.cpp
View File

@ -35,6 +35,7 @@ namespace skyline::gpu {
EXT_SET("VK_EXT_custom_border_color", hasCustomBorderColorExt);
EXT_SET("VK_EXT_provoking_vertex", hasProvokingVertexExt);
EXT_SET("VK_EXT_vertex_attribute_divisor", hasVertexAttributeDivisorExt);
EXT_SET("VK_KHR_push_descriptor", supportsPushDescriptors);
EXT_SET("VK_EXT_global_priority", supportsGlobalPriority);
EXT_SET("VK_EXT_shader_viewport_index_layer", supportsShaderViewportIndexLayer);
EXT_SET("VK_KHR_spirv_1_4", supportsSpirv14);
@ -131,8 +132,8 @@ namespace skyline::gpu {
std::string TraitManager::Summary() {
return fmt::format(
"\n* Supports U8 Indices: {}\n* Supports Sampler Mirror Clamp To Edge: {}\n* Supports Sampler Reduction Mode: {}\n* Supports Custom Border Color (Without Format): {}\n* Supports Last Provoking Vertex: {}\n* Supports Logical Operations: {}\n* Supports Vertex Attribute Divisor: {}\n* Supports Vertex Attribute Zero Divisor: {}\n* Supports Global Priority: {}\n* Supports Multiple Viewports: {}\n* Supports Shader Viewport Index: {}\n* Supports SPIR-V 1.4: {}\n* Supports Shader Invocation Demotion: {}\n* Supports 16-bit FP: {}\n* Supports 8-bit Integers: {}\n* Supports 16-bit Integers: {}\n* Supports 64-bit Integers: {}\n* Supports Atomic 64-bit Integers: {}\n* Supports Floating Point Behavior Control: {}\n* Supports Image Read Without Format: {}\n* Supports List Primitive Topology Restart: {}\n* Supports Patch List Primitive Topology Restart: {}\n* Supports Subgroup Vote: {}\n* Subgroup Size: {}",
supportsUint8Indices, supportsSamplerMirrorClampToEdge, supportsSamplerReductionMode, supportsCustomBorderColor, supportsLastProvokingVertex, supportsLogicOp, supportsVertexAttributeDivisor, supportsVertexAttributeZeroDivisor, supportsGlobalPriority, supportsMultipleViewports, supportsShaderViewportIndexLayer, supportsSpirv14, supportsShaderDemoteToHelper, supportsFloat16, supportsInt8, supportsInt16, supportsInt64, supportsAtomicInt64, supportsFloatControls, supportsImageReadWithoutFormat, supportsTopologyListRestart, supportsTopologyPatchListRestart, supportsSubgroupVote, subgroupSize
"\n* Supports U8 Indices: {}\n* Supports Sampler Mirror Clamp To Edge: {}\n* Supports Sampler Reduction Mode: {}\n* Supports Custom Border Color (Without Format): {}\n* Supports Last Provoking Vertex: {}\n* Supports Logical Operations: {}\n* Supports Vertex Attribute Divisor: {}\n* Supports Vertex Attribute Zero Divisor: {}\n* Supports Push Descriptors: {}\n* Supports Global Priority: {}\n* Supports Multiple Viewports: {}\n* Supports Shader Viewport Index: {}\n* Supports SPIR-V 1.4: {}\n* Supports Shader Invocation Demotion: {}\n* Supports 16-bit FP: {}\n* Supports 8-bit Integers: {}\n* Supports 16-bit Integers: {}\n* Supports 64-bit Integers: {}\n* Supports Atomic 64-bit Integers: {}\n* Supports Floating Point Behavior Control: {}\n* Supports Image Read Without Format: {}\n* Supports List Primitive Topology Restart: {}\n* Supports Patch List Primitive Topology Restart: {}\n* Supports Subgroup Vote: {}\n* Subgroup Size: {}",
supportsUint8Indices, supportsSamplerMirrorClampToEdge, supportsSamplerReductionMode, supportsCustomBorderColor, supportsLastProvokingVertex, supportsLogicOp, supportsVertexAttributeDivisor, supportsVertexAttributeZeroDivisor, supportsPushDescriptors, supportsGlobalPriority, supportsMultipleViewports, supportsShaderViewportIndexLayer, supportsSpirv14, supportsShaderDemoteToHelper, supportsFloat16, supportsInt8, supportsInt16, supportsInt64, supportsAtomicInt64, supportsFloatControls, supportsImageReadWithoutFormat, supportsTopologyListRestart, supportsTopologyPatchListRestart, supportsSubgroupVote, subgroupSize
);
}

1
app/src/main/cpp/skyline/gpu/trait_manager.h
View File

@ -20,6 +20,7 @@ namespace skyline::gpu {
bool supportsLogicOp{}; //!< If the device supports framebuffer logical operations during blending
bool supportsVertexAttributeDivisor{}; //!< If the device supports a divisor for instance-rate vertex attributes (with VK_EXT_vertex_attribute_divisor)
bool supportsVertexAttributeZeroDivisor{}; //!< If the device supports a zero divisor for instance-rate vertex attributes (with VK_EXT_vertex_attribute_divisor)
bool supportsPushDescriptors{}; //!< If the device supports push descriptors (with VK_KHR_push_descriptor)
bool supportsGlobalPriority{}; //!< If the device supports global priorities for queues (with VK_EXT_global_priority)
bool supportsMultipleViewports{}; //!< If the device supports more than one viewport
bool supportsShaderViewportIndexLayer{}; //!< If the device supports retrieving the viewport index in shaders (with VK_EXT_shader_viewport_index_layer)