build opencl kernel failure

Question

When I use clBuildProgram in my OpenCl code it fails with the error code -11 without any log information.

Here is what my code looks like:

  ret = clBuildProgram(program, 1, &device_id, NULL, NULL, NULL);

  if (ret != CL_SUCCESS)
    {
        size_t len;
        char buffer[2048];
    cl_build_status bldstatus;
    printf("\nError %d: Failed to build program executable [ %s ]\n",ret,get_error_string(ret));
        ret = clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_STATUS, sizeof(bldstatus), (void *)&bldstatus, &len);
        printf("Build Status %d: %s\n",ret,get_error_string(ret));
    printf("INFO: %s\n", get_error_string(bldstatus));
        ret = clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_OPTIONS, sizeof(buffer), buffer, &len);
        printf("Build Options %d: %s\n",ret,get_error_string(ret));
    printf("INFO: %s\n", buffer);   
        ret = clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, sizeof(buffer), buffer, &len);    
    printf("Build Log %d: %s\n",ret,get_error_string(ret));
    printf("%s\n", buffer);
    (void)infoinfo(platform_id,device_id);
    exit(1);
    }

Here is what the output looks like:

Error -11: Failed to build program executable [ CL_BUILD_PROGRAM_FAILURE ]
Build Status 0: CL_SUCCESS
INFO: CL_DEVICE_NOT_AVAILABLE
Build Options 0: CL_SUCCESS
INFO: 
Build Log -30: CL_INVALID_VALUE


CL_PLATFORM_NAME : NVIDIA CUDA
CL_PLATFORM_VERSION : OpenCL 1.1 CUDA 4.2.1
Device name : Tesla K20m
Driver version : 319.32
Global Memory (MB) : 4799
Global Memory Cache (KB) : 208
Local Memory (KB) : 48
Max clock (MHz) : 705
Max Work Group Size : 1024
Number of parallel compute cores : 13
Is the device available : yes

So, just by chance can you guys see any mistake or something weird in the lines above ?

Thank you,

Éric.

Answer 1

It looks like you might be using clGetProgramBuildInfo() wrong. The first time it is called, it should should be telling the size of the buffer it needs on the host:

clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, NULL, NULL, &len);

then on the host allocate the memory:

char *log = new char[len] //or whatever you use

then recall clGetProgramBuildInfo():

clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, len, log, null);