Files
renderdoc/util/test/demos/d3d12/d3d12_workgroup_zoo.cpp
T

495 lines
12 KiB
C++

/******************************************************************************
* The MIT License (MIT)
*
* Copyright (c) 2019-2025 Baldur Karlsson
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
******************************************************************************/
#include "3rdparty/fmt/core.h"
#include "d3d12_test.h"
RD_TEST(D3D12_Workgroup_Zoo, D3D12GraphicsTest)
{
static constexpr const char *Description =
"Test of behaviour around workgroup operations in shaders.";
const std::string common = R"EOSHADER(
cbuffer rootconsts : register(b0)
{
uint root_test;
uint root_two;
}
uint GetTest() { return root_test; }
#define IsTest(x) (GetTest() == x)
)EOSHADER";
const std::string compCommon = common + R"EOSHADER(
RWStructuredBuffer<float4> outbuf : register(u0);
static uint3 tid;
static uint flatId;
groupshared uint4 gsmUint4[1024];
void SetOutput(float4 val)
{
outbuf[root_test * 1024 + tid.y * GROUP_SIZE_X + tid.x] = val;
}
void Init(float4 val)
{
flatId = tid.z * GROUP_SIZE_X * GROUP_SIZE_Y + tid.y * GROUP_SIZE_X + tid.x;
gsmUint4[flatId].xyz = tid;
gsmUint4[flatId].z = tid.x;
SetOutput(val);
}
)EOSHADER";
const std::string testShader = compCommon + R"EOSHADER(
float4 funcD(uint id)
{
return WaveActiveSum(id/2).xxxx;
}
float4 nestedFunc(uint id)
{
float4 ret = funcD(id/3);
ret.w = WaveActiveSum(id);
return ret;
}
float4 funcA(uint id)
{
return nestedFunc(id*2);
}
float4 funcB(uint id)
{
return nestedFunc(id*4);
}
float4 funcTest(uint id)
{
if ((id % 2) == 0)
{
return 0.xxxx;
}
else
{
float value = WaveActiveSum(id);
if (id < 10)
{
return value.xxxx;
}
value += WaveActiveSum(id/2);
return value.xxxx;
}
}
float4 ComplexPartialReconvergence(uint id)
{
float4 result = 0.0.xxxx;
// Loops with different number of iterations per thread
for (uint i = id; i < 23; i++)
{
result.x += WaveActiveSum(id);
}
if ((result.x < 5) || (id > 20))
{
result.y += WaveActiveSum(id);
if (id < 25)
result.z += WaveActiveSum(id);
}
else if (result.x < 10)
{
result.y += WaveActiveSum(id);
if (result.x > 5)
result.z += WaveActiveSum(id);
}
result.w += WaveActiveSum(id);
return result;
}
[numthreads(GROUP_SIZE_X, GROUP_SIZE_Y, GROUP_SIZE_Z)]
void main(uint3 inTid : SV_GroupThreadID)
{
tid = inTid;
float4 testResult = 0.0f.xxxx;
Init(testResult);
uint id = flatId;
if(IsTest(0))
{
testResult.x = id;
AllMemoryBarrierWithGroupSync();
}
else if(IsTest(1))
{
testResult.x = WaveActiveSum(id);
AllMemoryBarrierWithGroupSync();
}
else if(IsTest(2))
{
// Diverged threads which reconverge
if (id < 10)
{
// active threads 0-9
testResult.x = WaveActiveSum(id);
if ((id % 2) == 0)
testResult.y = WaveActiveSum(id);
else
testResult.y = WaveActiveSum(id);
testResult.x += WaveActiveSum(id);
}
else
{
// active threads 10...
testResult.x = WaveActiveSum(id);
}
testResult.y = WaveActiveSum(id);
AllMemoryBarrierWithGroupSync();
}
else if(IsTest(3))
{
// Converged threads calling a function
testResult = funcTest(id);
testResult.y = WaveActiveSum(id);
AllMemoryBarrierWithGroupSync();
}
else if(IsTest(4))
{
// Converged threads calling a function which has a nested function call in it
testResult = nestedFunc(id);
testResult.y = WaveActiveSum(id);
AllMemoryBarrierWithGroupSync();
}
else if(IsTest(5))
{
// Diverged threads calling the same function
if (id < 10)
{
testResult = funcD(id);
}
else
{
testResult = funcD(id);
}
testResult.y = WaveActiveSum(id);
AllMemoryBarrierWithGroupSync();
}
else if(IsTest(6))
{
// Diverged threads calling the same function which has a nested function call in it
if (id < 10)
{
testResult = funcA(id);
}
else
{
testResult = funcB(id);
}
testResult.y = WaveActiveSum(id);
AllMemoryBarrierWithGroupSync();
}
else if(IsTest(7))
{
// Diverged threads which early exit
if (id < 10)
{
testResult.x = WaveActiveSum(id+10);
SetOutput(testResult);
return;
}
testResult.x = WaveActiveSum(id);
}
else if(IsTest(8))
{
// Loops with different number of iterations per thread
for (uint i = 0; i < id; i++)
{
testResult.x += WaveActiveSum(id);
}
AllMemoryBarrierWithGroupSync();
}
else if(IsTest(9))
{
// Query functions : unit tests
testResult.x = float(WaveGetLaneCount());
testResult.y = float(WaveGetLaneIndex());
testResult.z = float(WaveIsFirstLane());
AllMemoryBarrierWithGroupSync();
}
else if(IsTest(10))
{
testResult = ComplexPartialReconvergence(id);
AllMemoryBarrierWithGroupSync();
}
SetOutput(testResult);
}
)EOSHADER";
const std::string perfShader = compCommon + R"EOSHADER(
[numthreads(GROUP_SIZE_X, GROUP_SIZE_Y, GROUP_SIZE_Z)]
void main(uint3 inTid : SV_DispatchThreadID)
{
tid = inTid;
float4 testResult = 0.0f.xxxx;
Init(testResult);
uint id = flatId;
// TEST CASES:
// 0: GPU math : loops 100
// 1: CPU math : loops 100
// 2: GPU math : loops 200
// 3: CPU math : loops 200
// 4: GPU math : loops 400
// 5: CPU math : loops 400
// 6: GPU math : loops 5000
// 7: CPU math : loops 5000
bool useCpu = GetTest() & 0x1;
uint count = 0;
{
uint temp = GetTest() >> 1;
if(temp == 0)
count = 100U;
if(temp == 1)
count = 200U;
if(temp == 2)
count = 400U;
if(temp == 3)
count = 5000U;
}
if(useCpu)
{
for (uint i = 0; i < count; ++i)
{
gsmUint4[id].x += i;
gsmUint4[id].y += i * i;
testResult.x = testResult.x * testResult.x;
testResult.x += dot(gsmUint4[id], gsmUint4[id]);
}
}
else
{
for (uint i = 0; i < count; ++i)
{
gsmUint4[id].x += i;
gsmUint4[id].y += i * i;
testResult.x = pow(testResult.x, float(root_two));
testResult.x += dot(gsmUint4[id], gsmUint4[id]);
}
}
AllMemoryBarrierWithGroupSync();
SetOutput(testResult);
}
)EOSHADER";
void Prepare(int argc, char **argv)
{
D3D12GraphicsTest::Prepare(argc, argv);
if(opts1.WaveLaneCountMax < 16)
Avail = "Subgroup size is less than 16";
bool supportSM60 = (m_HighestShaderModel >= D3D_SHADER_MODEL_6_0) && m_DXILSupport;
if(!supportSM60)
Avail = "SM 6.0 not supported";
}
int main()
{
// initialise, create window, create device, etc
if(!Init())
return 3;
ID3D12RootSignaturePtr sig = MakeSig({constParam(D3D12_SHADER_VISIBILITY_ALL, 0, 0, 2),
uavParam(D3D12_SHADER_VISIBILITY_ALL, 0, 0)});
int32_t numCompTests = 0;
size_t pos = 0;
while(pos != std::string::npos)
{
pos = testShader.find("IsTest(", pos);
if(pos == std::string::npos)
break;
pos += sizeof("IsTest(") - 1;
numCompTests = std::max(numCompTests, atoi(testShader.c_str() + pos) + 1);
}
const int32_t countPerfTests = 8;
struct CompSize
{
int x, y, z;
};
CompSize compsizes[] = {
{70, 1, 1},
};
std::string comppipe_name[ARRAY_COUNT(compsizes)];
ID3D12PipelineStatePtr testPipes[ARRAY_COUNT(compsizes)];
ID3D12PipelineStatePtr perfPipes[ARRAY_COUNT(compsizes)];
std::string defines;
for(int i = 0; i < ARRAY_COUNT(compsizes); i++)
{
std::string sizedefine;
sizedefine =
fmt::format("#define GROUP_SIZE_X {}\n#define GROUP_SIZE_Y {}\n#define GROUP_SIZE_Z {}",
compsizes[i].x, compsizes[i].y, compsizes[i].z);
comppipe_name[i] = fmt::format("{}x{}x{}", compsizes[i].x, compsizes[i].y, compsizes[i].z);
testPipes[i] =
MakePSO().RootSig(sig).CS(Compile(defines + sizedefine + testShader, "main", "cs_6_0"));
testPipes[i]->SetName(UTF82Wide(comppipe_name[i]).c_str());
perfPipes[i] =
MakePSO().RootSig(sig).CS(Compile(defines + sizedefine + perfShader, "main", "cs_6_0"));
perfPipes[i]->SetName(UTF82Wide(comppipe_name[i]).c_str());
}
ID3D12ResourcePtr bufOut = MakeBuffer().Size(sizeof(Vec4f) * 1024 * numCompTests).UAV();
D3D12ViewCreator uavView =
MakeUAV(bufOut).Format(DXGI_FORMAT_R32_UINT).NumElements(4 * 1024 * numCompTests);
D3D12_CPU_DESCRIPTOR_HANDLE uavcpu = uavView.CreateClearCPU(10);
D3D12_GPU_DESCRIPTOR_HANDLE uavgpu = uavView.CreateGPU(10);
bufOut->SetName(L"bufOut");
while(Running())
{
ID3D12GraphicsCommandListPtr cmd = GetCommandBuffer();
Reset(cmd);
cmd->SetDescriptorHeaps(1, &m_CBVUAVSRV.GetInterfacePtr());
ID3D12ResourcePtr bb = StartUsingBackbuffer(cmd, D3D12_RESOURCE_STATE_RENDER_TARGET);
ClearRenderTargetView(cmd, BBRTV, {0.2f, 0.2f, 0.2f, 1.0f});
pushMarker(cmd, "Compute Tests");
for(size_t p = 0; p < ARRAY_COUNT(testPipes); p++)
{
ResourceBarrier(cmd);
UINT zero[4] = {};
cmd->ClearUnorderedAccessViewUint(uavgpu, uavcpu, bufOut, zero, 0, NULL);
ResourceBarrier(cmd);
pushMarker(cmd, comppipe_name[p]);
cmd->SetPipelineState(testPipes[p]);
cmd->SetComputeRootSignature(sig);
cmd->SetComputeRootUnorderedAccessView(1, bufOut->GetGPUVirtualAddress());
for(int i = 0; i < numCompTests; i++)
{
cmd->SetComputeRoot32BitConstant(0, i, 0);
cmd->SetComputeRoot32BitConstant(0, i + 1, 1);
cmd->Dispatch(2, 1, 1);
}
popMarker(cmd);
}
popMarker(cmd);
pushMarker(cmd, "Perf Tests");
for(size_t p = 0; p < ARRAY_COUNT(testPipes); p++)
{
ResourceBarrier(cmd);
UINT zero[4] = {};
cmd->ClearUnorderedAccessViewUint(uavgpu, uavcpu, bufOut, zero, 0, NULL);
ResourceBarrier(cmd);
pushMarker(cmd, comppipe_name[p]);
cmd->SetPipelineState(perfPipes[p]);
cmd->SetComputeRootSignature(sig);
cmd->SetComputeRootUnorderedAccessView(1, bufOut->GetGPUVirtualAddress());
for(int i = 0; i < countPerfTests; ++i)
{
bool useCpu = (i & 0x1);
int count = 0;
{
int temp = i >> 1;
if(temp == 0)
count = 100U;
if(temp == 1)
count = 200U;
if(temp == 2)
count = 400U;
if(temp == 3)
count = 5000U;
}
std::string perfTestName =
fmt::format("{} Iterations {} Math", count, useCpu ? "CPU" : "GPU");
pushMarker(cmd, perfTestName);
int two = 2;
cmd->SetComputeRoot32BitConstant(0, i, 0);
cmd->SetComputeRoot32BitConstant(0, two, 1);
cmd->Dispatch(2, 1, 1);
popMarker(cmd);
}
popMarker(cmd);
}
popMarker(cmd);
FinishUsingBackbuffer(cmd, D3D12_RESOURCE_STATE_RENDER_TARGET);
cmd->Close();
SubmitAndPresent({cmd});
}
return 0;
}
};
REGISTER_TEST();