blob: 79955bec4a34877256cdd0db9ff177663c3bdb40 [file] [log] [blame]
// Copyright 2019 The Dawn Authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "tests/perf_tests/DawnPerfTest.h"
#include "common/Assert.h"
#include "common/Constants.h"
#include "common/Math.h"
#include "utils/ComboRenderPipelineDescriptor.h"
#include "utils/WGPUHelpers.h"
namespace {
constexpr unsigned int kNumDraws = 2000;
constexpr uint32_t kTextureSize = 64;
constexpr size_t kUniformSize = 3 * sizeof(float);
constexpr float kVertexData[12] = {
0.0f, 0.5f, 0.0f, 1.0f, -0.5f, -0.5f, 0.0f, 1.0f, 0.5f, -0.5f, 0.0f, 1.0f,
};
constexpr char kVertexShader[] = R"(
[[stage(vertex)]] fn main(
[[location(0)]] pos : vec4<f32>
) -> [[builtin(position)]] vec4<f32> {
return pos;
})";
constexpr char kFragmentShaderA[] = R"(
struct Uniforms {
color : vec3<f32>;
};
[[group(0), binding(0)]] var<uniform> uniforms : Uniforms;
[[stage(fragment)]] fn main() -> [[location(0)]] vec4<f32> {
return vec4<f32>(uniforms.color * (1.0 / 5000.0), 1.0);
})";
constexpr char kFragmentShaderB[] = R"(
struct Constants {
color : vec3<f32>;
};
struct Uniforms {
color : vec3<f32>;
};
[[group(0), binding(0)]] var<uniform> constants : Constants;
[[group(1), binding(0)]] var<uniform> uniforms : Uniforms;
[[stage(fragment)]] fn main() -> [[location(0)]] vec4<f32> {
return vec4<f32>((constants.color + uniforms.color) * (1.0 / 5000.0), 1.0);
})";
enum class Pipeline {
Static, // Keep the same pipeline for all draws.
Redundant, // Use the same pipeline, but redundantly set it.
Dynamic, // Change the pipeline between draws.
};
enum class UniformData {
Static, // Don't update per-draw uniform data.
Dynamic, // Update the per-draw uniform data once per frame.
};
enum class BindGroup {
NoChange, // Use one bind group for all draws.
Redundant, // Use the same bind group, but redundantly set it.
NoReuse, // Create a new bind group every time.
Multiple, // Use multiple static bind groups.
Dynamic, // Use bind groups with dynamic offsets.
};
enum class VertexBuffer {
NoChange, // Use one vertex buffer for all draws.
Multiple, // Use multiple static vertex buffers.
Dynamic, // Switch vertex buffers between draws.
};
enum class RenderBundle {
No, // Record commands in a render pass
Yes, // Record commands in a render bundle
};
struct DrawCallParam {
Pipeline pipelineType;
VertexBuffer vertexBufferType;
BindGroup bindGroupType;
UniformData uniformDataType;
RenderBundle withRenderBundle;
};
using DrawCallParamTuple =
std::tuple<Pipeline, VertexBuffer, BindGroup, UniformData, RenderBundle>;
template <typename T>
unsigned int AssignParam(T& lhs, T rhs) {
lhs = rhs;
return 0u;
}
// This helper function allows creating a DrawCallParam from a list of arguments
// without specifying all of the members. Provided members can be passed once in an arbitrary
// order. Unspecified members default to:
// - Pipeline::Static
// - VertexBuffer::NoChange
// - BindGroup::NoChange
// - UniformData::Static
// - RenderBundle::No
template <typename... Ts>
DrawCallParam MakeParam(Ts... args) {
// Baseline param
DrawCallParamTuple paramTuple{Pipeline::Static, VertexBuffer::NoChange, BindGroup::NoChange,
UniformData::Static, RenderBundle::No};
unsigned int unused[] = {
0, // Avoid making a 0-sized array.
AssignParam(std::get<Ts>(paramTuple), args)...,
};
DAWN_UNUSED(unused);
return DrawCallParam{
std::get<Pipeline>(paramTuple), std::get<VertexBuffer>(paramTuple),
std::get<BindGroup>(paramTuple), std::get<UniformData>(paramTuple),
std::get<RenderBundle>(paramTuple),
};
}
struct DrawCallParamForTest : AdapterTestParam {
DrawCallParamForTest(const AdapterTestParam& backendParam, DrawCallParam param)
: AdapterTestParam(backendParam), param(param) {
}
DrawCallParam param;
};
std::ostream& operator<<(std::ostream& ostream, const DrawCallParamForTest& testParams) {
ostream << static_cast<const AdapterTestParam&>(testParams);
const DrawCallParam& param = testParams.param;
switch (param.pipelineType) {
case Pipeline::Static:
break;
case Pipeline::Redundant:
ostream << "_RedundantPipeline";
break;
case Pipeline::Dynamic:
ostream << "_DynamicPipeline";
break;
}
switch (param.vertexBufferType) {
case VertexBuffer::NoChange:
break;
case VertexBuffer::Multiple:
ostream << "_MultipleVertexBuffers";
break;
case VertexBuffer::Dynamic:
ostream << "_DynamicVertexBuffer";
}
switch (param.bindGroupType) {
case BindGroup::NoChange:
break;
case BindGroup::Redundant:
ostream << "_RedundantBindGroups";
break;
case BindGroup::NoReuse:
ostream << "_NoReuseBindGroups";
break;
case BindGroup::Multiple:
ostream << "_MultipleBindGroups";
break;
case BindGroup::Dynamic:
ostream << "_DynamicBindGroup";
break;
}
switch (param.uniformDataType) {
case UniformData::Static:
break;
case UniformData::Dynamic:
ostream << "_DynamicData";
break;
}
switch (param.withRenderBundle) {
case RenderBundle::No:
break;
case RenderBundle::Yes:
ostream << "_RenderBundle";
break;
}
return ostream;
}
} // anonymous namespace
// DrawCallPerf is an uber-benchmark with supports many parameterizations.
// The specific parameterizations we care about are explicitly instantiated at the bottom
// of this test file.
// DrawCallPerf tests drawing a simple triangle with many ways of encoding commands,
// binding, and uploading data to the GPU. The rationale for this is the following:
// - Static/Multiple/Dynamic vertex buffers: Tests switching buffer bindings. This has
// a state tracking cost as well as a GPU driver cost.
// - Static/Multiple/Dynamic bind groups: Same rationale as vertex buffers
// - Static/Dynamic pipelines: In addition to a change to GPU state, changing the pipeline
// layout incurs additional state tracking costs in Dawn.
// - With/Without render bundles: All of the above can have lower validation costs if
// precomputed in a render bundle.
// - Static/Dynamic data: Updating data for each draw is a common use case. It also tests
// the efficiency of resource transitions.
class DrawCallPerf : public DawnPerfTestWithParams<DrawCallParamForTest> {
public:
DrawCallPerf() : DawnPerfTestWithParams(kNumDraws, 3) {
}
~DrawCallPerf() override = default;
void SetUp() override;
protected:
DrawCallParam GetParam() const {
return DawnPerfTestWithParams::GetParam().param;
}
template <typename Encoder>
void RecordRenderCommands(Encoder encoder);
private:
void Step() override;
// One large dynamic vertex buffer, or multiple separate vertex buffers.
wgpu::Buffer mVertexBuffers[kNumDraws];
size_t mAlignedVertexDataSize;
std::vector<float> mUniformBufferData;
// One large dynamic uniform buffer, or multiple separate uniform buffers.
wgpu::Buffer mUniformBuffers[kNumDraws];
wgpu::BindGroupLayout mUniformBindGroupLayout;
// One dynamic bind group or multiple bind groups.
wgpu::BindGroup mUniformBindGroups[kNumDraws];
size_t mAlignedUniformSize;
size_t mNumUniformFloats;
wgpu::BindGroupLayout mConstantBindGroupLayout;
wgpu::BindGroup mConstantBindGroup;
// If the pipeline is static, only the first is used.
// Otherwise, the test alternates between two pipelines for each draw.
wgpu::RenderPipeline mPipelines[2];
wgpu::TextureView mColorAttachment;
wgpu::TextureView mDepthStencilAttachment;
wgpu::RenderBundle mRenderBundle;
};
void DrawCallPerf::SetUp() {
DawnPerfTestWithParams::SetUp();
// Compute aligned uniform / vertex data sizes.
mAlignedUniformSize =
Align(kUniformSize, GetSupportedLimits().limits.minUniformBufferOffsetAlignment);
mAlignedVertexDataSize = Align(sizeof(kVertexData), 4);
// Initialize uniform buffer data.
mNumUniformFloats = mAlignedUniformSize / sizeof(float);
mUniformBufferData = std::vector<float>(kNumDraws * mNumUniformFloats, 0.0);
// Create the color / depth stencil attachments.
{
wgpu::TextureDescriptor descriptor = {};
descriptor.dimension = wgpu::TextureDimension::e2D;
descriptor.size.width = kTextureSize;
descriptor.size.height = kTextureSize;
descriptor.size.depthOrArrayLayers = 1;
descriptor.usage = wgpu::TextureUsage::RenderAttachment;
descriptor.format = wgpu::TextureFormat::RGBA8Unorm;
mColorAttachment = device.CreateTexture(&descriptor).CreateView();
descriptor.format = wgpu::TextureFormat::Depth24PlusStencil8;
mDepthStencilAttachment = device.CreateTexture(&descriptor).CreateView();
}
// Create vertex buffer(s)
switch (GetParam().vertexBufferType) {
case VertexBuffer::NoChange:
mVertexBuffers[0] = utils::CreateBufferFromData(
device, kVertexData, sizeof(kVertexData), wgpu::BufferUsage::Vertex);
break;
case VertexBuffer::Multiple: {
for (uint32_t i = 0; i < kNumDraws; ++i) {
mVertexBuffers[i] = utils::CreateBufferFromData(
device, kVertexData, sizeof(kVertexData), wgpu::BufferUsage::Vertex);
}
break;
}
case VertexBuffer::Dynamic: {
std::vector<char> data(mAlignedVertexDataSize * kNumDraws);
for (uint32_t i = 0; i < kNumDraws; ++i) {
memcpy(data.data() + mAlignedVertexDataSize * i, kVertexData, sizeof(kVertexData));
}
mVertexBuffers[0] = utils::CreateBufferFromData(device, data.data(), data.size(),
wgpu::BufferUsage::Vertex);
break;
}
}
// Create the bind group layout.
switch (GetParam().bindGroupType) {
case BindGroup::NoChange:
case BindGroup::Redundant:
case BindGroup::NoReuse:
case BindGroup::Multiple:
mUniformBindGroupLayout = utils::MakeBindGroupLayout(
device,
{
{0, wgpu::ShaderStage::Fragment, wgpu::BufferBindingType::Uniform, false},
});
break;
case BindGroup::Dynamic:
mUniformBindGroupLayout = utils::MakeBindGroupLayout(
device,
{
{0, wgpu::ShaderStage::Fragment, wgpu::BufferBindingType::Uniform, true},
});
break;
default:
UNREACHABLE();
break;
}
// Setup the base render pipeline descriptor.
utils::ComboRenderPipelineDescriptor renderPipelineDesc;
renderPipelineDesc.vertex.bufferCount = 1;
renderPipelineDesc.cBuffers[0].arrayStride = 4 * sizeof(float);
renderPipelineDesc.cBuffers[0].attributeCount = 1;
renderPipelineDesc.cAttributes[0].format = wgpu::VertexFormat::Float32x4;
renderPipelineDesc.EnableDepthStencil(wgpu::TextureFormat::Depth24PlusStencil8);
renderPipelineDesc.cTargets[0].format = wgpu::TextureFormat::RGBA8Unorm;
// Create the pipeline layout for the first pipeline.
wgpu::PipelineLayoutDescriptor pipelineLayoutDesc = {};
pipelineLayoutDesc.bindGroupLayouts = &mUniformBindGroupLayout;
pipelineLayoutDesc.bindGroupLayoutCount = 1;
wgpu::PipelineLayout pipelineLayout = device.CreatePipelineLayout(&pipelineLayoutDesc);
// Create the shaders for the first pipeline.
wgpu::ShaderModule vsModule = utils::CreateShaderModule(device, kVertexShader);
wgpu::ShaderModule fsModule = utils::CreateShaderModule(device, kFragmentShaderA);
// Create the first pipeline.
renderPipelineDesc.layout = pipelineLayout;
renderPipelineDesc.vertex.module = vsModule;
renderPipelineDesc.cFragment.module = fsModule;
mPipelines[0] = device.CreateRenderPipeline(&renderPipelineDesc);
// If the test is using a dynamic pipeline, create the second pipeline.
if (GetParam().pipelineType == Pipeline::Dynamic) {
// Create another bind group layout. The data for this binding point will be the same for
// all draws.
mConstantBindGroupLayout = utils::MakeBindGroupLayout(
device, {
{0, wgpu::ShaderStage::Fragment, wgpu::BufferBindingType::Uniform, false},
});
// Create the pipeline layout.
wgpu::BindGroupLayout bindGroupLayouts[2] = {
mConstantBindGroupLayout,
mUniformBindGroupLayout,
};
pipelineLayoutDesc.bindGroupLayouts = bindGroupLayouts,
pipelineLayoutDesc.bindGroupLayoutCount = 2;
wgpu::PipelineLayout pipelineLayout = device.CreatePipelineLayout(&pipelineLayoutDesc);
// Create the fragment shader module. This shader matches the pipeline layout described
// above.
wgpu::ShaderModule fsModule = utils::CreateShaderModule(device, kFragmentShaderB);
// Create the pipeline.
renderPipelineDesc.layout = pipelineLayout;
renderPipelineDesc.cFragment.module = fsModule;
mPipelines[1] = device.CreateRenderPipeline(&renderPipelineDesc);
// Create the buffer and bind group to bind to the constant bind group layout slot.
constexpr float kConstantData[] = {0.01, 0.02, 0.03};
wgpu::Buffer constantBuffer = utils::CreateBufferFromData(
device, kConstantData, sizeof(kConstantData), wgpu::BufferUsage::Uniform);
mConstantBindGroup = utils::MakeBindGroup(device, mConstantBindGroupLayout,
{{0, constantBuffer, 0, sizeof(kConstantData)}});
}
// Create the buffers and bind groups for the per-draw uniform data.
switch (GetParam().bindGroupType) {
case BindGroup::NoChange:
case BindGroup::Redundant:
mUniformBuffers[0] = utils::CreateBufferFromData(
device, mUniformBufferData.data(), 3 * sizeof(float), wgpu::BufferUsage::Uniform);
mUniformBindGroups[0] = utils::MakeBindGroup(
device, mUniformBindGroupLayout, {{0, mUniformBuffers[0], 0, kUniformSize}});
break;
case BindGroup::NoReuse:
for (uint32_t i = 0; i < kNumDraws; ++i) {
mUniformBuffers[i] = utils::CreateBufferFromData(
device, mUniformBufferData.data() + i * mNumUniformFloats, 3 * sizeof(float),
wgpu::BufferUsage::Uniform);
}
// Bind groups are created on-the-fly.
break;
case BindGroup::Multiple:
for (uint32_t i = 0; i < kNumDraws; ++i) {
mUniformBuffers[i] = utils::CreateBufferFromData(
device, mUniformBufferData.data() + i * mNumUniformFloats, 3 * sizeof(float),
wgpu::BufferUsage::Uniform);
mUniformBindGroups[i] = utils::MakeBindGroup(
device, mUniformBindGroupLayout, {{0, mUniformBuffers[i], 0, kUniformSize}});
}
break;
case BindGroup::Dynamic:
mUniformBuffers[0] = utils::CreateBufferFromData(
device, mUniformBufferData.data(), mUniformBufferData.size() * sizeof(float),
wgpu::BufferUsage::Uniform);
mUniformBindGroups[0] = utils::MakeBindGroup(
device, mUniformBindGroupLayout, {{0, mUniformBuffers[0], 0, kUniformSize}});
break;
default:
UNREACHABLE();
break;
}
// If using render bundles, record the render commands now.
if (GetParam().withRenderBundle == RenderBundle::Yes) {
wgpu::RenderBundleEncoderDescriptor descriptor = {};
descriptor.colorFormatsCount = 1;
descriptor.colorFormats = &renderPipelineDesc.cTargets[0].format;
descriptor.depthStencilFormat = wgpu::TextureFormat::Depth24PlusStencil8;
wgpu::RenderBundleEncoder encoder = device.CreateRenderBundleEncoder(&descriptor);
RecordRenderCommands(encoder);
mRenderBundle = encoder.Finish();
}
}
template <typename Encoder>
void DrawCallPerf::RecordRenderCommands(Encoder pass) {
uint32_t uniformBindGroupIndex = 0;
if (GetParam().pipelineType == Pipeline::Static) {
// Static pipeline can be set now.
pass.SetPipeline(mPipelines[0]);
}
if (GetParam().vertexBufferType == VertexBuffer::NoChange) {
// Static vertex buffer can be set now.
pass.SetVertexBuffer(0, mVertexBuffers[0]);
}
if (GetParam().bindGroupType == BindGroup::NoChange) {
// Incompatible. Can't change pipeline without changing bind groups.
ASSERT(GetParam().pipelineType == Pipeline::Static);
// Static bind group can be set now.
pass.SetBindGroup(uniformBindGroupIndex, mUniformBindGroups[0]);
}
for (unsigned int i = 0; i < kNumDraws; ++i) {
switch (GetParam().pipelineType) {
case Pipeline::Static:
break;
case Pipeline::Redundant:
pass.SetPipeline(mPipelines[0]);
break;
case Pipeline::Dynamic: {
// If the pipeline is dynamic, ping pong between two pipelines.
pass.SetPipeline(mPipelines[i % 2]);
// The pipelines have different layouts so we change the binding index here.
uniformBindGroupIndex = i % 2;
if (uniformBindGroupIndex == 1) {
// Because of the pipeline layout change, we need to rebind bind group index 0.
pass.SetBindGroup(0, mConstantBindGroup);
}
break;
}
}
// Set the vertex buffer, if it changes.
switch (GetParam().vertexBufferType) {
case VertexBuffer::NoChange:
break;
case VertexBuffer::Multiple:
pass.SetVertexBuffer(0, mVertexBuffers[i]);
break;
case VertexBuffer::Dynamic:
pass.SetVertexBuffer(0, mVertexBuffers[0], i * mAlignedVertexDataSize);
break;
}
// Set the bind group, if it changes.
switch (GetParam().bindGroupType) {
case BindGroup::NoChange:
break;
case BindGroup::Redundant:
pass.SetBindGroup(uniformBindGroupIndex, mUniformBindGroups[0]);
break;
case BindGroup::NoReuse: {
wgpu::BindGroup bindGroup = utils::MakeBindGroup(
device, mUniformBindGroupLayout, {{0, mUniformBuffers[i], 0, kUniformSize}});
pass.SetBindGroup(uniformBindGroupIndex, bindGroup);
break;
}
case BindGroup::Multiple:
pass.SetBindGroup(uniformBindGroupIndex, mUniformBindGroups[i]);
break;
case BindGroup::Dynamic: {
uint32_t dynamicOffset = static_cast<uint32_t>(i * mAlignedUniformSize);
pass.SetBindGroup(uniformBindGroupIndex, mUniformBindGroups[0], 1, &dynamicOffset);
break;
}
default:
UNREACHABLE();
break;
}
pass.Draw(3);
}
}
void DrawCallPerf::Step() {
if (GetParam().uniformDataType == UniformData::Dynamic) {
// Update uniform data if it's dynamic.
std::fill(mUniformBufferData.begin(), mUniformBufferData.end(),
mUniformBufferData[0] + 1.0);
switch (GetParam().bindGroupType) {
case BindGroup::NoChange:
case BindGroup::Redundant:
queue.WriteBuffer(mUniformBuffers[0], 0, mUniformBufferData.data(),
3 * sizeof(float));
break;
case BindGroup::NoReuse:
case BindGroup::Multiple:
for (uint32_t i = 0; i < kNumDraws; ++i) {
queue.WriteBuffer(mUniformBuffers[i], 0,
mUniformBufferData.data() + i * mNumUniformFloats,
3 * sizeof(float));
}
break;
case BindGroup::Dynamic:
queue.WriteBuffer(mUniformBuffers[0], 0, mUniformBufferData.data(),
mUniformBufferData.size() * sizeof(float));
break;
}
}
wgpu::CommandEncoder commands = device.CreateCommandEncoder();
utils::ComboRenderPassDescriptor renderPass({mColorAttachment}, mDepthStencilAttachment);
wgpu::RenderPassEncoder pass = commands.BeginRenderPass(&renderPass);
switch (GetParam().withRenderBundle) {
case RenderBundle::No:
RecordRenderCommands(pass);
break;
case RenderBundle::Yes:
pass.ExecuteBundles(1, &mRenderBundle);
break;
default:
UNREACHABLE();
break;
}
pass.EndPass();
wgpu::CommandBuffer commandBuffer = commands.Finish();
queue.Submit(1, &commandBuffer);
}
TEST_P(DrawCallPerf, Run) {
RunTest();
}
DAWN_INSTANTIATE_TEST_P(
DrawCallPerf,
{D3D12Backend(), MetalBackend(), OpenGLBackend(), VulkanBackend(),
VulkanBackend({"skip_validation"})},
{
// Baseline
MakeParam(),
// Change vertex buffer binding
MakeParam(VertexBuffer::Multiple), // Multiple vertex buffers
MakeParam(VertexBuffer::Dynamic), // Dynamic vertex buffer
// Change bind group binding
MakeParam(BindGroup::Multiple), // Multiple bind groups
MakeParam(BindGroup::Dynamic), // Dynamic bind groups
MakeParam(BindGroup::NoReuse), // New bind group per-draw
// Redundantly set pipeline / bind groups
MakeParam(Pipeline::Redundant, BindGroup::Redundant),
// Switch the pipeline every draw to test state tracking and updates to binding points
MakeParam(Pipeline::Dynamic,
BindGroup::Multiple), // Multiple bind groups w/ dynamic pipeline
MakeParam(Pipeline::Dynamic,
BindGroup::Dynamic), // Dynamic bind groups w/ dynamic pipeline
// ----------- Render Bundles -----------
// Command validation / state tracking can be futher optimized / precomputed.
// Use render bundles with varying vertex buffer binding
MakeParam(VertexBuffer::Multiple,
RenderBundle::Yes), // Multiple vertex buffers w/ render bundle
MakeParam(VertexBuffer::Dynamic,
RenderBundle::Yes), // Dynamic vertex buffer w/ render bundle
// Use render bundles with varying bind group binding
MakeParam(BindGroup::Multiple, RenderBundle::Yes), // Multiple bind groups w/ render bundle
MakeParam(BindGroup::Dynamic, RenderBundle::Yes), // Dynamic bind groups w/ render bundle
// Use render bundles with dynamic pipeline
MakeParam(Pipeline::Dynamic,
BindGroup::Multiple,
RenderBundle::Yes), // Multiple bind groups w/ dynamic pipeline w/ render bundle
MakeParam(Pipeline::Dynamic,
BindGroup::Dynamic,
RenderBundle::Yes), // Dynamic bind groups w/ dynamic pipeline w/ render bundle
// ----------- Render Bundles (end)-------
// Update per-draw data in the bind group(s). This will cause resource transitions between
// updating and drawing.
MakeParam(BindGroup::Multiple,
UniformData::Dynamic), // Update per-draw data: Multiple bind groups
MakeParam(BindGroup::Dynamic,
UniformData::Dynamic), // Update per-draw data: Dynamic bind groups
});