Function initCUDA

cpp/0_Introduction/simpleTextureDrv/simpleTextureDrv.cpp:311–352

/////////////////////////////////////////////////////////////////////////// This initializes CUDA, and loads the *.ptx CUDA module containing the kernel function. After the module is loaded, cuModuleGetFunction retrieves the CUDA function pointer "cuFunction" ///////////////////////////////////////////////////////////////////////////

//! retrieves the CUDA function pointer "cuFunction"
////////////////////////////////////////////////////////////////////////////////
static CUresult initCUDA(int argc, char **argv, CUfunction *transform)
{
    CUfunction        cuFunction = 0;
    int               major = 0, minor = 0, devID = 0;
    char              deviceName[100];
    string            module_path;
    CUctxCreateParams ctxCreateParams = {};

    cuDevice = findCudaDeviceDRV(argc, (const char **)argv);

    // get compute capabilities and the devicename
    checkCudaErrors(cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
    checkCudaErrors(cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
    checkCudaErrors(cuDeviceGetName(deviceName, sizeof(deviceName), cuDevice));
    printf("> GPU Device has SM %d.%d compute capability\n", major, minor);

    checkCudaErrors(cuCtxCreate(&cuContext, &ctxCreateParams, 0, cuDevice));

    // first search for the module_path before we try to load the results
    std::ostringstream fatbin;

    if (!findFatbinPath(FATBIN_FILE, module_path, argv, fatbin)) {
        exit(EXIT_FAILURE);
    }
    else {
        printf("> initCUDA loading module: <%s>\n", module_path.c_str());
    }

    if (!fatbin.str().size()) {
        printf("fatbin file empty. exiting..\n");
        exit(EXIT_FAILURE);
    }

    // Create module from binary file (FATBIN)
    checkCudaErrors(cuModuleLoadData(&cuModule, fatbin.str().c_str()));

    checkCudaErrors(cuModuleGetFunction(&cuFunction, cuModule, "transformKernel"));

    *transform = cuFunction;

    return CUDA_SUCCESS;
}