bool load_kernels(bool experimental)
{
/* check if cuda init succeeded */
if(cuContext == 0)
return false;
/* check if GPU is supported with current feature set */
if(!support_device(experimental))
return false;
/* get kernel */
string cubin = compile_kernel();
if(cubin == "")
return false;
/* open module */
cuda_push_context();
CUresult result = cuModuleLoad(&cuModule, cubin.c_str());
if(cuda_error_(result, "cuModuleLoad"))
cuda_error_message(string_printf("Failed loading CUDA kernel %s.", cubin.c_str()));
cuda_pop_context();
return (result == CUDA_SUCCESS);
}
bool load_kernels(bool experimental)
{
/* verify if device was initialized */
if(!device_initialized) {
fprintf(stderr, "OpenCL: failed to initialize device.\n");
return false;
}
/* verify we have right opencl version */
if(!opencl_version_check())
return false;
/* md5 hash to detect changes */
string kernel_path = path_get("kernel");
string kernel_md5 = path_files_md5_hash(kernel_path);
string device_md5 = device_md5_hash();
/* try to use cache binary */
string clbin = string_printf("cycles_kernel_%s_%s.clbin", device_md5.c_str(), kernel_md5.c_str());;
clbin = path_user_get(path_join("cache", clbin));
if(path_exists(clbin)) {
/* if exists already, try use it */
if(!load_binary(kernel_path, clbin))
return false;
}
else {
/* compile kernel */
if(!compile_kernel(kernel_path, kernel_md5))
return false;
/* save binary for reuse */
save_binary(clbin);
}
/* find kernels */
ckPathTraceKernel = clCreateKernel(cpProgram, "kernel_ocl_path_trace", &ciErr);
if(opencl_error(ciErr))
return false;
ckFilmConvertKernel = clCreateKernel(cpProgram, "kernel_ocl_tonemap", &ciErr);
if(opencl_error(ciErr))
return false;
return true;
}