void cpu_codelet(void *descr[], void *_args)
{
float *block = (float *)STARPU_BLOCK_GET_PTR(descr[0]);
int nx = (int)STARPU_BLOCK_GET_NX(descr[0]);
int ny = (int)STARPU_BLOCK_GET_NY(descr[0]);
int nz = (int)STARPU_BLOCK_GET_NZ(descr[0]);
unsigned ldy = STARPU_BLOCK_GET_LDY(descr[0]);
unsigned ldz = STARPU_BLOCK_GET_LDZ(descr[0]);
float *multiplier = (float *)_args;
int i, j, k;
for(k=0; k<nz ; k++) {
for(j=0; j<ny ; j++) {
for(i=0; i<nx ; i++)
block[(k*ldz)+(j*ldy)+i] *= *multiplier;
}
}
}
void opencl_codelet(void *descr[], void *_args)
{
cl_kernel kernel;
cl_command_queue queue;
cl_event event;
int id, devid, err;
cl_mem block = (cl_mem)STARPU_BLOCK_GET_DEV_HANDLE(descr[0]);
int nx = (int)STARPU_BLOCK_GET_NX(descr[0]);
int ny = (int)STARPU_BLOCK_GET_NY(descr[0]);
int nz = (int)STARPU_BLOCK_GET_NZ(descr[0]);
int ldy = (int)STARPU_BLOCK_GET_LDY(descr[0]);
int ldz = (int) STARPU_BLOCK_GET_LDZ(descr[0]);
float *multiplier = (float *)_args;
id = starpu_worker_get_id();
devid = starpu_worker_get_devid(id);
err = starpu_opencl_load_kernel(&kernel, &queue, &opencl_code, "block", devid);
if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
CHECK_CL_SET_KERNEL_ARG(kernel, 0, sizeof(block), &block);
CHECK_CL_SET_KERNEL_ARG(kernel, 1, sizeof(nx), &nx);
CHECK_CL_SET_KERNEL_ARG(kernel, 2, sizeof(ny), &ny);
CHECK_CL_SET_KERNEL_ARG(kernel, 3, sizeof(nz), &nz);
CHECK_CL_SET_KERNEL_ARG(kernel, 4, sizeof(ldy), &ldy);
CHECK_CL_SET_KERNEL_ARG(kernel, 5, sizeof(ldz), &ldz);
CHECK_CL_SET_KERNEL_ARG(kernel, 6, sizeof(*multiplier), multiplier);
{
size_t global=nx*ny*nz;
err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, NULL, 0, NULL, &event);
if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
}
clFinish(queue);
starpu_opencl_collect_stats(event);
clReleaseEvent(event);
starpu_opencl_release_kernel(kernel);
}
void opencl_codelet(void *descr[], void *_args)
{
cl_kernel kernel;
cl_command_queue queue;
int id, devid, err, n;
float *block = (float *)STARPU_BLOCK_GET_PTR(descr[0]);
int nx = (int)STARPU_BLOCK_GET_NX(descr[0]);
int ny = (int)STARPU_BLOCK_GET_NY(descr[0]);
int nz = (int)STARPU_BLOCK_GET_NZ(descr[0]);
unsigned ldy = STARPU_BLOCK_GET_LDY(descr[0]);
unsigned ldz = STARPU_BLOCK_GET_LDZ(descr[0]);
float *multiplier = (float *)_args;
id = starpu_worker_get_id();
devid = starpu_worker_get_devid(id);
err = starpu_opencl_load_kernel(&kernel, &queue, &opencl_code, "block", devid);
if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
err = 0;
n=0;
err = clSetKernelArg(kernel, 0, sizeof(cl_mem), &block);
err = clSetKernelArg(kernel, 1, sizeof(int), &nx);
err = clSetKernelArg(kernel, 2, sizeof(int), &ny);
err = clSetKernelArg(kernel, 3, sizeof(int), &nz);
err = clSetKernelArg(kernel, 4, sizeof(ldy), &ldy);
err = clSetKernelArg(kernel, 5, sizeof(ldz), &ldz);
err = clSetKernelArg(kernel, 6, sizeof(float), multiplier);
if (err) STARPU_OPENCL_REPORT_ERROR(err);
{
size_t global=nx*ny*nz;
err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, NULL, 0, NULL, NULL);
if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
}
clFinish(queue);
starpu_opencl_release_kernel(kernel);
}