#include #include extern struct starpu_opencl_codelet codelet; void scal_opencl_func(void *buffers[], void *_args) @{ float *factor = (float *)_args; int id, devid, err; cl_kernel kernel; cl_command_queue queue; /* length of the vector */ unsigned n = STARPU_GET_VECTOR_NX(buffers[0]); /* local copy of the vector pointer */ float *val = (float *)STARPU_GET_VECTOR_PTR(buffers[0]); id = starpu_worker_get_id(); devid = starpu_worker_get_devid(id); err = starpu_opencl_load_kernel(&kernel, &queue, &codelet, "vector_mult_opencl", devid); if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err); err = 0; err = clSetKernelArg(kernel, 0, sizeof(cl_mem), &val); err = clSetKernelArg(kernel, 1, sizeof(int), &n); err |= clSetKernelArg(kernel, 2, sizeof(float), (void*)factor); if (err) STARPU_OPENCL_REPORT_ERROR(err); @{ size_t global=n; size_t local=n; err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 0, NULL, NULL); if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err); @} clFinish(queue); starpu_opencl_release_kernel(kernel); @}