mirror/tracy
1
0
Fork 0
mirror of https://github.com/wolfpld/tracy.git synced 2024-11-22 22:44:34 +00:00
Code Issues Packages Projects Releases Wiki Activity

Make OpenCL example async

Browse Source
This commit is contained in:
Cloud Han 2022-01-13 12:58:57 +08:00
parent d5d0a8676e
commit f6894c8d6c
1 changed files with 4 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
examples/OpenCLVectorAdd/OpenCLVectorAdd.cpp
View File

@ -123,7 +123,7 @@ int main()
ZoneScopedN("Write Buffer A"); ZoneScopedN("Write Buffer A");
TracyCLZoneS(tracyCLCtx, "Write BufferA", 5); TracyCLZoneS(tracyCLCtx, "Write BufferA", 5);
CL_ASSERT(clEnqueueWriteBuffer(commandQueue, bufferA, CL_TRUE, 0, N * sizeof(float), hostA.data(), 0, nullptr, &writeBufferAEvent)); CL_ASSERT(clEnqueueWriteBuffer(commandQueue, bufferA, CL_FALSE, 0, N * sizeof(float), hostA.data(), 0, nullptr, &writeBufferAEvent));
TracyCLZoneSetEvent(writeBufferAEvent); TracyCLZoneSetEvent(writeBufferAEvent);
} }
@ -131,13 +131,13 @@ int main()
ZoneScopedN("Write Buffer B"); ZoneScopedN("Write Buffer B");
TracyCLZone(tracyCLCtx, "Write BufferB"); TracyCLZone(tracyCLCtx, "Write BufferB");
CL_ASSERT(clEnqueueWriteBuffer(commandQueue, bufferB, CL_TRUE, 0, N * sizeof(float), hostB.data(), 0, nullptr, &writeBufferBEvent)); CL_ASSERT(clEnqueueWriteBuffer(commandQueue, bufferB, CL_FALSE, 0, N * sizeof(float), hostB.data(), 0, nullptr, &writeBufferBEvent));
TracyCLZoneSetEvent(writeBufferBEvent); TracyCLZoneSetEvent(writeBufferBEvent);
} }
} }
for (int i = 0; i < 10; ++i) for (int i = 0; i < 100; ++i)
{ {
int n_value = static_cast<int>(N); int n_value = static_cast<int>(N);
ZoneScopedN("VectorAdd Kernel Launch"); ZoneScopedN("VectorAdd Kernel Launch");
@ -150,15 +150,8 @@ int main()
cl_event vectorAddKernelEvent; cl_event vectorAddKernelEvent;
CL_ASSERT(clEnqueueNDRangeKernel(commandQueue, vectorAddKernel, 1, nullptr, &N, nullptr, 0, nullptr, &vectorAddKernelEvent)); CL_ASSERT(clEnqueueNDRangeKernel(commandQueue, vectorAddKernel, 1, nullptr, &N, nullptr, 0, nullptr, &vectorAddKernelEvent));
CL_ASSERT(clWaitForEvents(1, &vectorAddKernelEvent));
TracyCLZoneSetEvent(vectorAddKernelEvent); TracyCLZoneSetEvent(vectorAddKernelEvent);
std::cout << "VectorAdd Kernel Enqueued" << std::endl;
cl_ulong kernelStartTime, kernelEndTime;
CL_ASSERT(clGetEventProfilingInfo(vectorAddKernelEvent, CL_PROFILING_COMMAND_START, sizeof(cl_ulong), &kernelStartTime, nullptr));
CL_ASSERT(clGetEventProfilingInfo(vectorAddKernelEvent, CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &kernelEndTime, nullptr));
std::cout << "VectorAdd Kernel Elapsed: " << ((kernelEndTime - kernelStartTime) / 1000) << " us" << std::endl;
} }
{ {