OpenCL Deconstruction Fails

[Malte_P_]

Hi there,

it seems that there is a bug in the openCL implementation. With that Code:

[cpp]
for (int run = 0; run < 10; run++) {   
cl_int iError;   
cl_uint uiPlatformCount;   
clGetPlatformIDs(0, nullptr, &uiPlatformCount);   
cl_platform_id* pPlatformIDs = new cl_platform_id[uiPlatformCount];   
clGetPlatformIDs(uiPlatformCount, pPlatformIDs, nullptr);   
for (int i = 0; i < (int)uiPlatformCount; i++)    {      
  cl_platform_id pID = pPlatformIDs;      
  char cPlatformInfo[1024];      
  iError = clGetPlatformInfo(pID, CL_PLATFORM_NAME, sizeof(cPlatformInfo), cPlatformInfo, nullptr);      
  std::wcout << "  Plattform Name:\t" << cPlatformInfo <<  std::endl;      
  cl_uint uiDeviceCount;      
  iError = clGetDeviceIDs(pID, CL_DEVICE_TYPE_ALL, 0, nullptr, &uiDeviceCount);      
  cl_device_id* pDeviceIDs = new cl_device_id[uiDeviceCount];      
  iError = clGetDeviceIDs(pID, CL_DEVICE_TYPE_ALL, uiDeviceCount, pDeviceIDs, &uiDeviceCount);      
  cl_context ctx;      
  cl_context_properties pContextProperties[3] = {CL_CONTEXT_PLATFORM, (cl_context_properties)pID, 0};      
  ctx = clCreateContext(pContextProperties, uiDeviceCount, pDeviceIDs, nullptr, nullptr, &iError);       
  for (int j = 0; j < (int)uiDeviceCount; j++)      
  {         
    cl_device_id dID = pDeviceIDs;         
    cl_command_queue queue = clCreateCommandQueue(ctx, dID, CL_QUEUE_PROFILING_ENABLE, &iError);         
    clReleaseCommandQueue(queue);       
  }
  clReleaseContext(ctx);   
}
}
[/cpp]

Deconstruction with NVidia platform works but won't work with intel platform. But when you use the intel platform by compiling a Kernel and release it again, it works.

[cpp]
for (int run = 0; run < 10; run++) {
   cl_int iError; cl_uint uiPlatformCount;
   clGetPlatformIDs(0, nullptr, &uiPlatformCount);
   cl_platform_id* pPlatformIDs = new cl_platform_id[uiPlatformCount];
   clGetPlatformIDs(uiPlatformCount, pPlatformIDs, nullptr);
   for (int i = 0; i < (int)uiPlatformCount; i++) {
     cl_platform_id pID = pPlatformIDs;
    char cPlatformInfo[1024]; 
    iError = clGetPlatformInfo(pID, CL_PLATFORM_NAME, sizeof(cPlatformInfo), cPlatformInfo, nullptr);
    std::wcout << "  Plattform Name:\t" << cPlatformInfo <<  std::endl;  
    char *sIntel = "Intel";
    if(strstr(cPlatformInfo, sIntel)) {
      //continue;
    }
    cl_uint uiDeviceCount;
     iError = clGetDeviceIDs(pID, CL_DEVICE_TYPE_ALL, 0, nullptr, &uiDeviceCount);
     cl_device_id* pDeviceIDs = new cl_device_id[uiDeviceCount];
    iError = clGetDeviceIDs(pID, CL_DEVICE_TYPE_ALL, uiDeviceCount, pDeviceIDs, &uiDeviceCount);
     cl_context ctx;
    cl_context_properties pContextProperties[3] = {CL_CONTEXT_PLATFORM, (cl_context_properties)pID, 0};
    ctx = clCreateContext(pContextProperties, uiDeviceCount, pDeviceIDs, nullptr, nullptr, &iError);
    CString sKernel = _T("__kernel void crunch(float f) { float x,y,cs,sn,xx,yy; x = 1.0f; y = 0.0f; cs = cos(2.0f); sn = sin(2.0f); for (int i=0;i<40000;i++){ xx = cs * x - sn * y; yy = cs * y + sn * x; x = xx; y = yy; }  } ");
    size_t szSourceSize = (size_t) sKernel.GetLength();
    CStringA saCode = sKernel;
    const char* cpKernel = saCode;
    cl_program pclProgram = clCreateProgramWithSource(ctx, 1, &cpKernel, &szSourceSize, &iError);
    char *pOptions = "-Werror";
    iError = clBuildProgram(pclProgram, uiDeviceCount, pDeviceIDs, /*nullptr*/pOptions, nullptr, nullptr);
    const char* kernelName = "crunch";
    cl_kernel pclKernel = clCreateKernel(pclProgram, kernelName, &iError);
    for (int j = 0; j < (int)uiDeviceCount; j++) {
      cl_device_id dID = pDeviceIDs;
      cl_command_queue queue = clCreateCommandQueue(ctx, dID, CL_QUEUE_PROFILING_ENABLE, &iError);
      cl_kernel pclKernel = clCreateKernel(pclProgram, kernelName, &iError);
      float f = 1.f; iError = clSetKernelArg(pclKernel, 0, sizeof(cl_mem), &f);
      size_t lws[1] = {256};
      size_t gws[1] = {256};
      iError = clEnqueueNDRangeKernel(queue, pclKernel, 1, nullptr, gws, lws, 0,nullptr, nullptr);
      clReleaseKernel(pclKernel); clReleaseProgram(pclProgram);
      /*float pfTemp[4] = {0};
       cl_mem pBuffer = clCreateBuffer(ctx, CL_MEM_READ_WRITE, 4 * sizeof(float), &pfTemp[0], &iError);
      clReleaseMemObject(pBuffer);*/
      iError = clReleaseCommandQueue(queue);
    }
     iError = clReleaseContext(ctx); 
  }
} 
[/cpp]

[Malte_P_]

Forgot to say that i use windows 7 on an x990 with OpenCL 1.1 Build 31360.31441 in Visual studio

[Chuck_De_Sylva]

We were not able to reproduce the issue you are having. Can you share your system configuration? Also what SDK version and driver version are you using?

[Malte_P_]

Deinstalled and reinstalled the SDK and it worked fine. Don't have a clue, what went wrong. Thanks for your reply