// Main function
// *********************************************************************
int main(int argc, char **argv)
{
gp_argc = &argc;
gp_argv = &argv;
shrQAStart(argc, argv);
// Get the NVIDIA platform
ciErrNum = oclGetPlatformID(&cpPlatform);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
shrLog("clGetPlatformID...\n");
// Get the NVIDIA platform
ciErrNum = oclGetPlatformID(&cpPlatform);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
shrLog("clGetPlatformID...\n");
//Get all the devices
cl_uint uiNumDevices = 0; // Number of devices available
cl_uint uiTargetDevice = 0; // Default Device to compute on
cl_uint uiNumComputeUnits; // Number of compute units (SM's on NV GPU)
shrLog("Get the Device info and select Device...\n");
ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_GPU, 0, NULL, &uiNumDevices);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
cdDevices = (cl_device_id *)malloc(uiNumDevices * sizeof(cl_device_id) );
ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_GPU, uiNumDevices, cdDevices, NULL);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
// Get command line device options and config accordingly
shrLog(" # of Devices Available = %u\n", uiNumDevices);
if(shrGetCmdLineArgumentu(argc, (const char**)argv, "device", &uiTargetDevice)== shrTRUE)
{
uiTargetDevice = CLAMP(uiTargetDevice, 0, (uiNumDevices - 1));
}
shrLog(" Using Device %u: ", uiTargetDevice);
oclPrintDevName(LOGBOTH, cdDevices[uiTargetDevice]);
ciErrNum = clGetDeviceInfo(cdDevices[uiTargetDevice], CL_DEVICE_MAX_COMPUTE_UNITS, sizeof(uiNumComputeUnits), &uiNumComputeUnits, NULL);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, NULL);
shrLog("\n # of Compute Units = %u\n", uiNumComputeUnits);
// get command line arg for quick test, if provided
bNoPrompt = shrCheckCmdLineFlag(argc, (const char**)argv, "noprompt");
// start logs
cExecutableName = argv[0];
shrSetLogFileName ("oclDotProduct.txt");
shrLog("%s Starting...\n\n# of float elements per Array \t= %u\n", argv[0], iNumElements);
// set and log Global and Local work size dimensions
szLocalWorkSize = 256;
szGlobalWorkSize = shrRoundUp((int)szLocalWorkSize, iNumElements); // rounded up to the nearest multiple of the LocalWorkSize
shrLog("Global Work Size \t\t= %u\nLocal Work Size \t\t= %u\n# of Work Groups \t\t= %u\n\n",
szGlobalWorkSize, szLocalWorkSize, (szGlobalWorkSize % szLocalWorkSize + szGlobalWorkSize/szLocalWorkSize));
// Allocate and initialize host arrays
shrLog( "Allocate and Init Host Mem...\n");
srcA = (void *)malloc(sizeof(cl_float4) * szGlobalWorkSize);
srcB = (void *)malloc(sizeof(cl_float4) * szGlobalWorkSize);
dst = (void *)malloc(sizeof(cl_float) * szGlobalWorkSize);
Golden = (void *)malloc(sizeof(cl_float) * iNumElements);
shrFillArray((float*)srcA, 4 * iNumElements);
shrFillArray((float*)srcB, 4 * iNumElements);
// Get the NVIDIA platform
ciErrNum = oclGetPlatformID(&cpPlatform);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
// Get a GPU device
ciErrNum = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_GPU, 1, &cdDevices[uiTargetDevice], NULL);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
// Create the context
cxGPUContext = clCreateContext(0, 1, &cdDevices[uiTargetDevice], NULL, NULL, &ciErrNum);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
// Create a command-queue
shrLog("clCreateCommandQueue...\n");
cqCommandQueue = clCreateCommandQueue(cxGPUContext, cdDevices[uiTargetDevice], 0, &ciErrNum);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
// Allocate the OpenCL buffer memory objects for source and result on the device GMEM
shrLog("clCreateBuffer (SrcA, SrcB and Dst in Device GMEM)...\n");
cmDevSrcA = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, sizeof(cl_float) * szGlobalWorkSize * 4, NULL, &ciErrNum);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
cmDevSrcB = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, sizeof(cl_float) * szGlobalWorkSize * 4, NULL, &ciErrNum);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
cmDevDst = clCreateBuffer(cxGPUContext, CL_MEM_WRITE_ONLY, sizeof(cl_float) * szGlobalWorkSize, NULL, &ciErrNum);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
// Read the OpenCL kernel in from source file
shrLog("oclLoadProgSource (%s)...\n", cSourceFile);
cPathAndName = shrFindFilePath(cSourceFile, argv[0]);
oclCheckErrorEX(cPathAndName != NULL, shrTRUE, pCleanup);
cSourceCL = oclLoadProgSource(cPathAndName, "", &szKernelLength);
oclCheckErrorEX(cSourceCL != NULL, shrTRUE, pCleanup);
// Create the program
shrLog("clCreateProgramWithSource...\n");
cpProgram = clCreateProgramWithSource(cxGPUContext, 1, (const char **)&cSourceCL, &szKernelLength, &ciErrNum);
// Build the program with 'mad' Optimization option
#ifdef MAC
char* flags = "-cl-fast-relaxed-math -DMAC";
#else
char* flags = "-cl-fast-relaxed-math";
#endif
shrLog("clBuildProgram...\n");
ciErrNum = clBuildProgram(cpProgram, 0, NULL, NULL, NULL, NULL);
if (ciErrNum != CL_SUCCESS)
{
// write out standard error, Build Log and PTX, then cleanup and exit
shrLogEx(LOGBOTH | ERRORMSG, ciErrNum, STDERROR);
oclLogBuildInfo(cpProgram, oclGetFirstDev(cxGPUContext));
oclLogPtx(cpProgram, oclGetFirstDev(cxGPUContext), "oclDotProduct.ptx");
Cleanup(EXIT_FAILURE);
}
// Create the kernel
shrLog("clCreateKernel (DotProduct)...\n");
ckKernel = clCreateKernel(cpProgram, "DotProduct", &ciErrNum);
// Set the Argument values
shrLog("clSetKernelArg 0 - 3...\n\n");
ciErrNum = clSetKernelArg(ckKernel, 0, sizeof(cl_mem), (void*)&cmDevSrcA);
ciErrNum |= clSetKernelArg(ckKernel, 1, sizeof(cl_mem), (void*)&cmDevSrcB);
ciErrNum |= clSetKernelArg(ckKernel, 2, sizeof(cl_mem), (void*)&cmDevDst);
ciErrNum |= clSetKernelArg(ckKernel, 3, sizeof(cl_int), (void*)&iNumElements);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
// --------------------------------------------------------
// Core sequence... copy input data to GPU, compute, copy results back
// Asynchronous write of data to GPU device
shrLog("clEnqueueWriteBuffer (SrcA and SrcB)...\n");
ciErrNum = clEnqueueWriteBuffer(cqCommandQueue, cmDevSrcA, CL_FALSE, 0, sizeof(cl_float) * szGlobalWorkSize * 4, srcA, 0, NULL, NULL);
ciErrNum |= clEnqueueWriteBuffer(cqCommandQueue, cmDevSrcB, CL_FALSE, 0, sizeof(cl_float) * szGlobalWorkSize * 4, srcB, 0, NULL, NULL);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
// Launch kernel
shrLog("clEnqueueNDRangeKernel (DotProduct)...\n");
ciErrNum = clEnqueueNDRangeKernel(cqCommandQueue, ckKernel, 1, NULL, &szGlobalWorkSize, &szLocalWorkSize, 0, NULL, NULL);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
// Read back results and check accumulated errors
shrLog("clEnqueueReadBuffer (Dst)...\n\n");
ciErrNum = clEnqueueReadBuffer(cqCommandQueue, cmDevDst, CL_TRUE, 0, sizeof(cl_float) * szGlobalWorkSize, dst, 0, NULL, NULL);
oclCheckErrorEX(ciErrNum, CL_SUCCESS, pCleanup);
// Compute and compare results for golden-host and report errors and pass/fail
shrLog("Comparing against Host/C++ computation...\n\n");
DotProductHost ((const float*)srcA, (const float*)srcB, (float*)Golden, iNumElements);
shrBOOL bMatch = shrComparefet((const float*)Golden, (const float*)dst, (unsigned int)iNumElements, 0.0f, 0);
// Cleanup and leave
Cleanup (EXIT_SUCCESS);
}
bool CheckRender::compareBin2BinFloat(const char *src_file, const char *ref_file, unsigned int nelements, const float epsilon, const float threshold)
{
float *src_buffer, *ref_buffer;
FILE *src_fp = NULL, *ref_fp = NULL;
unsigned long error_count = 0;
#ifdef WIN32
fopen_s(&src_fp, src_file, "rb");
#else
src_fp = fopen(src_file, "rb");
#endif
if (src_fp == NULL) {
shrLog("compareBin2Bin <float> unable to open src_file: %s\n", src_file);
error_count = 1;
}
char *ref_file_path = shrFindFilePath(ref_file, m_ExecPath);
if (ref_file_path == NULL) {
shrLog("compareBin2Bin <float> unable to find <%s> in <%s>\n", ref_file, m_ExecPath);
shrLog(">>> Check info.xml and [project//data] folder <%s> <<<\n", m_ExecPath);
shrLog("Aborting comparison!\n");
error_count++;
if (src_fp) fclose(src_fp);
if (ref_fp) fclose(ref_fp);
}
else
{
#ifdef WIN32
fopen_s(&ref_fp, ref_file_path, "rb");
#else
ref_fp = fopen(ref_file_path, "rb");
#endif
if (ref_fp == NULL) {
shrLog("compareBin2Bin <float> unable to open ref_file: %s\n", ref_file_path);
error_count = 1;
}
if (src_fp && ref_fp) {
src_buffer = (float *)malloc(nelements*sizeof(float));
ref_buffer = (float *)malloc(nelements*sizeof(float));
fread(src_buffer, nelements, sizeof(float), src_fp);
fread(ref_buffer, nelements, sizeof(float), ref_fp);
shrLog("> compareBin2Bin <float> nelements=%d, epsilon=%.2f, threshold=%.2f\n", nelements, epsilon, threshold);
shrLog(" src_file <%s>\n", src_file);
shrLog(" ref_file <%s>\n", ref_file_path);
if (!shrComparefet( ref_buffer, src_buffer, nelements, epsilon, threshold)) {
error_count++;
}
fclose(src_fp);
fclose(ref_fp);
free(src_buffer);
free(ref_buffer);
} else {
if (src_fp) fclose(src_fp);
if (ref_fp) fclose(ref_fp);
}
}
if (error_count == 0) {
shrLog(" Data Matches\n");
} else {
shrLog(" Data Doesn't Match!!! %d errors...\n", (unsigned int)error_count);
}
return (error_count == 0); // returns true if all pixels pass
}
// Main function
// *********************************************************************
int ymain(int argc, char **argv)
{
shrQAStart(argc, argv);
// get command line arg for quick test, if provided
bNoPrompt = shrCheckCmdLineFlag(argc, (const char**)argv, "noprompt");
// start logs
cExecutableName = argv[0];
shrSetLogFileName ("oclVectorAdd2.txt");
shrLog("%s Starting...\n\n# of float elements per Array \t= %i\n", argv[0], iNumElements);
// set and log Global and Local work size dimensions
szLocalWorkSize = 256;
szGlobalWorkSize = shrRoundUp((int)szLocalWorkSize, iNumElements); // rounded up to the nearest multiple of the LocalWorkSize
shrLog("Global Work Size \t\t= %u\nLocal Work Size \t\t= %u\n# of Work Groups \t\t= %u\n\n",
szGlobalWorkSize, szLocalWorkSize, (szGlobalWorkSize % szLocalWorkSize + szGlobalWorkSize/szLocalWorkSize));
// Allocate and initialize host arrays
shrLog( "Allocate and Init Host Mem...\n");
srcA = (void *)malloc(sizeof(cl_float) * szGlobalWorkSize);
srcB = (void *)malloc(sizeof(cl_float) * szGlobalWorkSize);
dst = (void *)malloc(sizeof(cl_float) * szGlobalWorkSize);
Golden = (void *)malloc(sizeof(cl_float) * iNumElements);
shrFillArray((float*)srcA, iNumElements);
shrFillArray((float*)srcB, iNumElements);
//Get an OpenCL platform
ciErr1 = clGetPlatformIDs(1, &cpPlatform, NULL);
shrLog("clGetPlatformID...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clGetPlatformID, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(argc, argv, EXIT_FAILURE);
}
//Get the devices
ciErr1 = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_GPU, 1, &cdDevice, NULL);
shrLog("clGetDeviceIDs...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clGetDeviceIDs, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(argc, argv, EXIT_FAILURE);
}
//Create the context
cxGPUContext = clCreateContext(0, 1, &cdDevice, NULL, NULL, &ciErr1);
shrLog("clCreateContext...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clCreateContext, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(argc, argv, EXIT_FAILURE);
}
// Create a command-queue
cqCommandQueue = clCreateCommandQueue(cxGPUContext, cdDevice, 0, &ciErr1);
shrLog("clCreateCommandQueue...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clCreateCommandQueue, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(argc, argv, EXIT_FAILURE);
}
// Allocate the OpenCL buffer memory objects for source and result on the device GMEM
cmDevSrcA = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, sizeof(cl_float) * szGlobalWorkSize, NULL, &ciErr1);
cmDevSrcB = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, sizeof(cl_float) * szGlobalWorkSize, NULL, &ciErr2);
ciErr1 |= ciErr2;
cmDevDst = clCreateBuffer(cxGPUContext, CL_MEM_WRITE_ONLY, sizeof(cl_float) * szGlobalWorkSize, NULL, &ciErr2);
ciErr1 |= ciErr2;
shrLog("clCreateBuffer...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clCreateBuffer, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(argc, argv, EXIT_FAILURE);
}
// Read the OpenCL kernel in from source file
shrLog("oclLoadProgSource (%s)...\n", cSourceFile);
cPathAndName = shrFindFilePath(cSourceFile, argv[0]);
shrLog("Looking for: %s in Path: %s\n", cSourceFile, argv[0]);
cSourceCL = oclLoadProgSource(cPathAndName, "", &szKernelLength);
// Create the program
cpProgram = clCreateProgramWithSource(cxGPUContext, 1, (const char **)&cSourceCL, &szKernelLength, &ciErr1);
shrLog("clCreateProgramWithSource...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clCreateProgramWithSource, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(argc, argv, EXIT_FAILURE);
}
// Build the program with 'mad' Optimization option
#ifdef MAC
char* flags = "-cl-fast-relaxed-math -DMAC";
#else
char* flags = "-cl-fast-relaxed-math";
#endif
ciErr1 = clBuildProgram(cpProgram, 0, NULL, NULL, NULL, NULL);
shrLog("clBuildProgram...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clBuildProgram, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(argc, argv, EXIT_FAILURE);
}
// Create the kernel
ckKernel = clCreateKernel(cpProgram, "VectorAdd", &ciErr1);
shrLog("clCreateKernel (VectorAdd)...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clCreateKernel, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(argc, argv, EXIT_FAILURE);
}
// Set the Argument values
ciErr1 = clSetKernelArg(ckKernel, 0, sizeof(cl_mem), (void*)&cmDevSrcA);
ciErr1 |= clSetKernelArg(ckKernel, 1, sizeof(cl_mem), (void*)&cmDevSrcB);
ciErr1 |= clSetKernelArg(ckKernel, 2, sizeof(cl_mem), (void*)&cmDevDst);
ciErr1 |= clSetKernelArg(ckKernel, 3, sizeof(cl_int), (void*)&iNumElements);
shrLog("clSetKernelArg 0 - 3...\n\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clSetKernelArg, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(argc, argv, EXIT_FAILURE);
}
// --------------------------------------------------------
// Start Core sequence... copy input data to GPU, compute, copy results back
// Asynchronous write of data to GPU device
ciErr1 = clEnqueueWriteBuffer(cqCommandQueue, cmDevSrcA, CL_FALSE, 0, sizeof(cl_float) * szGlobalWorkSize, srcA, 0, NULL, NULL);
ciErr1 |= clEnqueueWriteBuffer(cqCommandQueue, cmDevSrcB, CL_FALSE, 0, sizeof(cl_float) * szGlobalWorkSize, srcB, 0, NULL, NULL);
shrLog("clEnqueueWriteBuffer (SrcA and SrcB)...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clEnqueueWriteBuffer, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(argc, argv, EXIT_FAILURE);
}
// Launch kernel
ciErr1 = clEnqueueNDRangeKernel(cqCommandQueue, ckKernel, 1, NULL, &szGlobalWorkSize, &szLocalWorkSize, 0, NULL, NULL);
shrLog("clEnqueueNDRangeKernel (VectorAdd)...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clEnqueueNDRangeKernel, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(argc, argv, EXIT_FAILURE);
}
// Synchronous/blocking read of results, and check accumulated errors
ciErr1 = clEnqueueReadBuffer(cqCommandQueue, cmDevDst, CL_TRUE, 0, sizeof(cl_float) * szGlobalWorkSize, dst, 0, NULL, NULL);
shrLog("clEnqueueReadBuffer (Dst)...\n\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clEnqueueReadBuffer, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(argc, argv, EXIT_FAILURE);
}
//--------------------------------------------------------
// Compute and compare results for golden-host and report errors and pass/fail
shrLog("Comparing against Host/C++ computation...\n\n");
VectorAddHost ((const float*)srcA, (const float*)srcB, (float*)Golden, iNumElements);
shrBOOL bMatch = shrComparefet((const float*)Golden, (const float*)dst, (unsigned int)iNumElements, 0.0f, 0);
// Cleanup and leave
Cleanup (argc, argv, (bMatch == shrTRUE) ? EXIT_SUCCESS : EXIT_FAILURE);
}
// Main function
// *********************************************************************
int main(int argc, char **argv)
{
//////////////////////////////////////////////////////////////////////////
unsigned int count = iNumElements;
int k = 8;
unsigned int random_seed, random_seed2;
srand( (unsigned)time( NULL ) );
random_seed = rand();
random_seed2 = rand();
//////////////////////////////////////////////////////////////////////////
// get command line arg for quick test, if provided
bNoPrompt = shrCheckCmdLineFlag(argc, (const char**)argv, "noprompt");
// start logs
shrSetLogFileName ("oclVectorAdd.txt");
shrLog("%s Starting...\n\n# of float elements per Array \t= %i\n", argv[0], iNumElements);
// set and log Global and Local work size dimensions
szLocalWorkSize = 256;
szGlobalWorkSize = shrRoundUp((int)szLocalWorkSize, iNumElements); // rounded up to the nearest multiple of the LocalWorkSize
shrLog("Global Work Size \t\t= %u\nLocal Work Size \t\t= %u\n# of Work Groups \t\t= %u\n\n",
szGlobalWorkSize, szLocalWorkSize, (szGlobalWorkSize % szLocalWorkSize + szGlobalWorkSize/szLocalWorkSize));
// Allocate and initialize host arrays
shrLog( "Allocate and Init Host Mem...\n");
srcA = (void *)malloc(sizeof(cl_float) * szGlobalWorkSize);
srcB = (void *)malloc(sizeof(cl_float) * szGlobalWorkSize);
dst = (void *)malloc(sizeof(cl_float) * szGlobalWorkSize);
Golden = (void *)malloc(sizeof(cl_float) * iNumElements);
shrFillArray((float*)srcA, iNumElements);
shrFillArray((float*)srcB, iNumElements);
//////////////////////////////////////////////////////////////////////////
float *scalar_value = new float[count];
float *gradient_magnitude = new float[count];
float *second_derivative_magnitude = new float[count];
unsigned char *label_ptr = new unsigned char[count];
shrFillArray(scalar_value, count);
shrFillArray(gradient_magnitude, count);
shrFillArray(second_derivative_magnitude, count);
//////////////////////////////////////////////////////////////////////////
//Get an OpenCL platform
ciErr1 = clGetPlatformIDs(1, &cpPlatform, NULL);
shrLog("clGetPlatformID...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clGetPlatformID, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(EXIT_FAILURE);
}
//Get the devices
ciErr1 = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_GPU, 1, &cdDevice, NULL);
shrLog("clGetDeviceIDs...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clGetDeviceIDs, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(EXIT_FAILURE);
}
//Create the context
cxGPUContext = clCreateContext(0, 1, &cdDevice, NULL, NULL, &ciErr1);
shrLog("clCreateContext...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clCreateContext, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(EXIT_FAILURE);
}
// Create a command-queue
cqCommandQueue = clCreateCommandQueue(cxGPUContext, cdDevice, 0, &ciErr1);
shrLog("clCreateCommandQueue...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clCreateCommandQueue, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(EXIT_FAILURE);
}
// Allocate the OpenCL buffer memory objects for source and result on the device GMEM
cmDevSrcA = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, sizeof(cl_float) * szGlobalWorkSize, NULL, &ciErr1);
cmDevSrcB = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, sizeof(cl_float) * szGlobalWorkSize, NULL, &ciErr2);
ciErr1 |= ciErr2;
cmDevDst = clCreateBuffer(cxGPUContext, CL_MEM_WRITE_ONLY, sizeof(cl_float) * szGlobalWorkSize, NULL, &ciErr2);
ciErr1 |= ciErr2;
//////////////////////////////////////////////////////////////////////////
cmDevSrc_scalar_value = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, sizeof(cl_float) * szGlobalWorkSize, NULL, &ciErr1);
cmDevSrc_gradient_magnitude = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, sizeof(cl_float) * szGlobalWorkSize, NULL, &ciErr2);
ciErr1 |= ciErr2;
cmDevSrc_second_derivative_magnitude = clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, sizeof(cl_float) * szGlobalWorkSize, NULL, &ciErr2);
ciErr1 |= ciErr2;
cmDevDst_label_ptr = clCreateBuffer(cxGPUContext, CL_MEM_READ_WRITE, sizeof(cl_float) * szGlobalWorkSize, NULL, &ciErr2);
ciErr1 |= ciErr2;
//////////////////////////////////////////////////////////////////////////
shrLog("clCreateBuffer...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clCreateBuffer, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(EXIT_FAILURE);
}
// Read the OpenCL kernel in from source file
shrLog("oclLoadProgSource (%s)...\n", cSourceFile);
cPathAndName = shrFindFilePath(cSourceFile, argv[0]);
cSourceCL = oclLoadProgSource(cPathAndName, "", &szKernelLength);
printf("%s\n%s\n", cSourceFile, cPathAndName);
// Create the program
cpProgram = clCreateProgramWithSource(cxGPUContext, 1, (const char **)&cSourceCL, &szKernelLength, &ciErr1);
shrLog("clCreateProgramWithSource...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clCreateProgramWithSource, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(EXIT_FAILURE);
}
// Build the program with 'mad' Optimization option
#ifdef MAC
char* flags = "-cl-fast-relaxed-math -DMAC";
#else
char* flags = "-cl-fast-relaxed-math";
#endif
ciErr1 = clBuildProgram(cpProgram, 0, NULL, NULL, NULL, NULL);
shrLog("clBuildProgram...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clBuildProgram, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(EXIT_FAILURE);
}
// Create the kernel
ckKernel = clCreateKernel(cpProgram, "k_means", &ciErr1);
shrLog("clCreateKernel (VectorAdd)...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clCreateKernel, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(EXIT_FAILURE);
}
// Set the Argument values
//ciErr1 = clSetKernelArg(ckKernel, 0, sizeof(cl_mem), (void*)&cmDevSrcA);
//ciErr1 |= clSetKernelArg(ckKernel, 1, sizeof(cl_mem), (void*)&cmDevSrcB);
//ciErr1 |= clSetKernelArg(ckKernel, 2, sizeof(cl_mem), (void*)&cmDevDst);
//ciErr1 |= clSetKernelArg(ckKernel, 3, sizeof(cl_int), (void*)&iNumElements);
//////////////////////////////////////////////////////////////////////////
// __global const float *scalar_value, __global const float *gradient_magnitude, __global const float *second_derivative_magnitude, __global unsigned char *label_ptr, __global const unsigned int count, __global const int k, __global const unsigned int random_seed, __global const unsigned int random_seed2
ciErr1 = clSetKernelArg(ckKernel, 0, sizeof(cl_mem), (void*)&cmDevSrc_scalar_value);
ciErr1 |= clSetKernelArg(ckKernel, 1, sizeof(cl_mem), (void*)&cmDevSrc_gradient_magnitude);
ciErr1 |= clSetKernelArg(ckKernel, 2, sizeof(cl_mem), (void*)&cmDevSrc_second_derivative_magnitude);
ciErr1 |= clSetKernelArg(ckKernel, 3, sizeof(cl_mem), (void*)&cmDevDst_label_ptr);
ciErr1 |= clSetKernelArg(ckKernel, 4, sizeof(cl_uint), (void*)&count);
ciErr1 |= clSetKernelArg(ckKernel, 5, sizeof(cl_uint), (void*)&k);
ciErr1 |= clSetKernelArg(ckKernel, 6, sizeof(cl_uint), (void*)&random_seed);
ciErr1 |= clSetKernelArg(ckKernel, 7, sizeof(cl_uint), (void*)&random_seed2);
//////////////////////////////////////////////////////////////////////////
shrLog("clSetKernelArg 0 - 3...\n\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clSetKernelArg, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(EXIT_FAILURE);
}
// --------------------------------------------------------
// Start Core sequence... copy input data to GPU, compute, copy results back
// Asynchronous write of data to GPU device
//ciErr1 = clEnqueueWriteBuffer(cqCommandQueue, cmDevSrcA, CL_FALSE, 0, sizeof(cl_float) * szGlobalWorkSize, srcA, 0, NULL, NULL);
//ciErr1 |= clEnqueueWriteBuffer(cqCommandQueue, cmDevSrcB, CL_FALSE, 0, sizeof(cl_float) * szGlobalWorkSize, srcB, 0, NULL, NULL);
//////////////////////////////////////////////////////////////////////////
ciErr1 = clEnqueueWriteBuffer(cqCommandQueue, cmDevSrc_scalar_value, CL_FALSE, 0, sizeof(cl_float) * szGlobalWorkSize, scalar_value, 0, NULL, NULL);
ciErr1 |= clEnqueueWriteBuffer(cqCommandQueue, cmDevSrc_gradient_magnitude, CL_FALSE, 0, sizeof(cl_float) * szGlobalWorkSize, gradient_magnitude, 0, NULL, NULL);
ciErr1 |= clEnqueueWriteBuffer(cqCommandQueue, cmDevSrc_second_derivative_magnitude, CL_FALSE, 0, sizeof(cl_float) * szGlobalWorkSize, second_derivative_magnitude, 0, NULL, NULL);
//////////////////////////////////////////////////////////////////////////
shrLog("clEnqueueWriteBuffer (SrcA and SrcB)...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clEnqueueWriteBuffer, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(EXIT_FAILURE);
}
// Launch kernel
ciErr1 = clEnqueueNDRangeKernel(cqCommandQueue, ckKernel, 1, NULL, &szGlobalWorkSize, &szLocalWorkSize, 0, NULL, NULL);
shrLog("clEnqueueNDRangeKernel (VectorAdd)...\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clEnqueueNDRangeKernel, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(EXIT_FAILURE);
}
// Synchronous/blocking read of results, and check accumulated errors
//ciErr1 = clEnqueueReadBuffer(cqCommandQueue, cmDevDst, CL_TRUE, 0, sizeof(cl_float) * szGlobalWorkSize, dst, 0, NULL, NULL);
//////////////////////////////////////////////////////////////////////////
ciErr1 = clEnqueueReadBuffer(cqCommandQueue, cmDevDst_label_ptr, CL_TRUE, 0, sizeof(cl_float) * szGlobalWorkSize, label_ptr, 0, NULL, NULL);
//////////////////////////////////////////////////////////////////////////
shrLog("clEnqueueReadBuffer (Dst)...\n\n");
if (ciErr1 != CL_SUCCESS)
{
shrLog("Error in clEnqueueReadBuffer, Line %u in file %s !!!\n\n", __LINE__, __FILE__);
Cleanup(EXIT_FAILURE);
}
//--------------------------------------------------------
// Compute and compare results for golden-host and report errors and pass/fail
shrLog("Comparing against Host/C++ computation...\n\n");
VectorAddHost ((const float*)srcA, (const float*)srcB, (float*)Golden, iNumElements);
shrBOOL bMatch = shrComparefet((const float*)Golden, (const float*)dst, (unsigned int)iNumElements, 0.0f, 0);
shrLog("%s\n\n", (bMatch == shrTRUE) ? "PASSED" : "FAILED");
//////////////////////////////////////////////////////////////////////////
//float *a = (float *)srcA;
//float *b = (float *)srcB;
//float *c = (float *)dst;
//float *d = (float *)Golden;
//for (int i=0; i<iNumElements; i++)
//{
// printf("%f+%f=%f=%f\t", a[i], b[i], c[i], a[i]+b[i]);
// printf("%s\n", (a[i]+b[i]==c[i]?"equal":"not equal"));
//}
//for (int i=0; i<iNumElements; i++)
//{
// printf("%f\n", ((float *)dst)[i]);
//}
//////////////////////////////////////////////////////////////////////////
// Cleanup and leave
Cleanup (EXIT_SUCCESS);
//////////////////////////////////////////////////////////////////////////
delete [] scalar_value;
delete [] gradient_magnitude;
delete [] second_derivative_magnitude;
delete [] label_ptr;
//////////////////////////////////////////////////////////////////////////
}
