wu-arch/kernels
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

VectorHypot Additions

Browse Source
This commit is contained in:
Lyons, Ethan Tyler
2019-11-25 00:49:03 -05:00
committed by GitHub Enterprise
parent a970867b7b
commit e391c0a2c5
1 changed files with 11 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
benchmarks/opencl/VectorHypot/main.cc
View File

@@ -292,28 +292,28 @@ int main(int argc, const char **argv)
shrLog("oclLoadProgSource (%s)...\n", cSourceFile); shrLog("oclLoadProgSource (%s)...\n", cSourceFile);
// Create the program object // Create the program object
cpProgram = clCreateProgramWithSource(cxGPUContext, 1, (const char **)&cSourceCL, &szKernelLength, &ciErrNum); //cpProgram = clCreateProgramWithSource(cxGPUContext, 1, (const char **)&cSourceCL, &szKernelLength, &ciErrNum);
//oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup); //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
shrLog("clCreateProgramWithSource...\n"); shrLog("clCreateProgramWithSource...\n");
cl_program program =
clCreateProgramWithBuiltInKernels(context, 1, &device_id, "VectorHypot", NULL);
// Build the program for the target device // Build the program for the target device
clFinish(cqCommandQueue[0]); clFinish(cqCommandQueue[0]);
shrDeltaT(0); shrDeltaT(0);
ciErrNum = clBuildProgram(cpProgram, uiNumDevsUsed, &cdDevices[uiTargetDevice], "-cl-fast-relaxed-math", NULL, NULL); ciErrNum = clBuildProgram(program, uiNumDevsUsed, &cdDevices[uiTargetDevice], "-cl-fast-relaxed-math", NULL, NULL);
shrLog("clBuildProgram..."); shrLog("clBuildProgram...");
if (ciErrNum != CL_SUCCESS) if (ciErrNum != CL_SUCCESS)
{ {
// write out standard error, Build Log and PTX, then cleanup and exit // write out standard error, Build Log and PTX, then cleanup and exit
shrLogEx(LOGBOTH | ERRORMSG, (double)ciErrNum, STDERROR); shrLogEx(LOGBOTH | ERRORMSG, (double)ciErrNum, STDERROR);
oclLogBuildInfo(cpProgram, oclGetFirstDev(cxGPUContext)); oclLogBuildInfo(program, oclGetFirstDev(cxGPUContext));
oclLogPtx(cpProgram, oclGetFirstDev(cxGPUContext), "VectorHypot.ptx"); oclLogPtx(program, oclGetFirstDev(cxGPUContext), "VectorHypot.ptx");
Cleanup(EXIT_FAILURE); Cleanup(EXIT_FAILURE);
} }
dBuildTime = shrDeltaT(0); dBuildTime = shrDeltaT(0);
// Ethan - Kernel Addition // Ethan - Kernel Addition
cl_program program =
clCreateProgramWithBuiltInKernels(context, 1, &device_id, "VectorHypot", NULL);
if (program == NULL) { if (program == NULL) {
std::cerr << "Failed to write program binary" << std::endl; std::cerr << "Failed to write program binary" << std::endl;
Cleanup(context, queue, program, kernel, memObjects); Cleanup(context, queue, program, kernel, memObjects);
@@ -323,9 +323,9 @@ int main(int argc, const char **argv)
} }
// Create the kernel // Create the kernel
ckKernel[0] = clCreateKernel(cl_Program, "VectorHypot", &ciErrNum); ckKernel[0] = clCreateKernel(program, "VectorHypot", &ciErrNum);
//oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup); //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
ckKernel[1] = clCreateKernel(cl_Program, "VectorHypot", &ciErrNum); ckKernel[1] = clCreateKernel(program, "VectorHypot", &ciErrNum);
//oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup); //oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
shrLog("clCreateKernel (ckKernel[2])...\n"); shrLog("clCreateKernel (ckKernel[2])...\n");
@@ -652,7 +652,7 @@ void Cleanup (int iExitCode)
if(Golden)free(Golden); if(Golden)free(Golden);
if(ckKernel[0])clReleaseKernel(ckKernel[0]); if(ckKernel[0])clReleaseKernel(ckKernel[0]);
if(ckKernel[1])clReleaseKernel(ckKernel[1]); if(ckKernel[1])clReleaseKernel(ckKernel[1]);
if(cpProgram)clReleaseProgram(cpProgram); if(program)clReleaseProgram(program);
if(fSourceA)clEnqueueUnmapMemObject(cqCommandQueue[0], cmPinnedSrcA, (void*)fSourceA, 0, NULL, NULL); if(fSourceA)clEnqueueUnmapMemObject(cqCommandQueue[0], cmPinnedSrcA, (void*)fSourceA, 0, NULL, NULL);
if(fSourceB)clEnqueueUnmapMemObject(cqCommandQueue[0], cmPinnedSrcB, (void*)fSourceB, 0, NULL, NULL); if(fSourceB)clEnqueueUnmapMemObject(cqCommandQueue[0], cmPinnedSrcB, (void*)fSourceB, 0, NULL, NULL);
if(fResult)clEnqueueUnmapMemObject(cqCommandQueue[0], cmPinnedResult, (void*)fResult, 0, NULL, NULL); if(fResult)clEnqueueUnmapMemObject(cqCommandQueue[0], cmPinnedResult, (void*)fResult, 0, NULL, NULL);