Exemplo n.º 1
0
int main(int argc, char **argv)
{
	int ret;

	/* Not supported yet */
	if (starpu_get_env_number_default("STARPU_GLOBAL_ARBITER", 0) > 0)
		return 77;

	ret = starpu_init(NULL);
	if (ret == -ENODEV)
		return 77;
	STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");

#ifdef STARPU_USE_OPENCL
	ret = starpu_opencl_load_opencl_from_file("examples/reductions/dot_product_opencl_kernels.cl",
						  &_opencl_program, NULL);
	STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
#endif

#ifdef STARPU_USE_CUDA
	/* cublasSdot has synchronization issues when using a non-blocking stream */
	cublasGetVersion(&cublas_version);
	if (cublas_version >= 7050)
		starpu_cublas_init();
#endif

	unsigned long nelems = _nblocks*_entries_per_block;
	size_t size = nelems*sizeof(float);

	_x = (float *) malloc(size);
	_y = (float *) malloc(size);

	_x_handles = (starpu_data_handle_t *) calloc(_nblocks, sizeof(starpu_data_handle_t));
	_y_handles = (starpu_data_handle_t *) calloc(_nblocks, sizeof(starpu_data_handle_t));

	assert(_x && _y);

        starpu_srand48(0);

	DOT_TYPE reference_dot = 0.0;

	unsigned long i;
	for (i = 0; i < nelems; i++)
	{
		_x[i] = (float)starpu_drand48();
		_y[i] = (float)starpu_drand48();

		reference_dot += (DOT_TYPE)_x[i]*(DOT_TYPE)_y[i];
	}

	unsigned block;
	for (block = 0; block < _nblocks; block++)
	{
		starpu_vector_data_register(&_x_handles[block], STARPU_MAIN_RAM,
			(uintptr_t)&_x[_entries_per_block*block], _entries_per_block, sizeof(float));
		starpu_vector_data_register(&_y_handles[block], STARPU_MAIN_RAM,
			(uintptr_t)&_y[_entries_per_block*block], _entries_per_block, sizeof(float));
	}

	starpu_variable_data_register(&_dot_handle, STARPU_MAIN_RAM, (uintptr_t)&_dot, sizeof(DOT_TYPE));

	/*
	 *	Compute dot product with StarPU
	 */
	starpu_data_set_reduction_methods(_dot_handle, &redux_codelet, &init_codelet);

	for (block = 0; block < _nblocks; block++)
	{
		struct starpu_task *task = starpu_task_create();

		task->cl = &dot_codelet;
		task->destroy = 1;

		task->handles[0] = _x_handles[block];
		task->handles[1] = _y_handles[block];
		task->handles[2] = _dot_handle;

		ret = starpu_task_submit(task);
		if (ret == -ENODEV) goto enodev;
		STARPU_ASSERT(!ret);
	}

	for (block = 0; block < _nblocks; block++)
	{
		starpu_data_unregister(_x_handles[block]);
		starpu_data_unregister(_y_handles[block]);
	}
	starpu_data_unregister(_dot_handle);

	FPRINTF(stderr, "Reference : %e vs. %e (Delta %e)\n", reference_dot, _dot, reference_dot - _dot);

#ifdef STARPU_USE_CUDA
	if (cublas_version >= 7050)
		starpu_cublas_shutdown();
#endif

#ifdef STARPU_USE_OPENCL
        ret = starpu_opencl_unload_opencl(&_opencl_program);
        STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_unload_opencl");
#endif
	starpu_shutdown();

	free(_x);
	free(_y);
	free(_x_handles);
	free(_y_handles);

	if (fabs(reference_dot - _dot) < reference_dot * 1e-6)
		return EXIT_SUCCESS;
	else
		return EXIT_FAILURE;

enodev:
	fprintf(stderr, "WARNING: No one can execute this task\n");
	/* yes, we do not perform the computation but we did detect that no one
 	 * could perform the kernel, so this is not an error from StarPU */
	return 77;
}
Exemplo n.º 2
0
int main(int argc, char **argv)
{
	unsigned long i;
	int ret;

	/* Not supported yet */
	if (starpu_get_env_number_default("STARPU_GLOBAL_ARBITER", 0) > 0)
		return 77;

	ret = starpu_init(NULL);
	if (ret == -ENODEV)
		return 77;
	STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");

	unsigned long nelems = _nblocks*_entries_per_bock;
	size_t size = nelems*sizeof(TYPE);

	_x = (TYPE *) malloc(size);
	_x_handles = (starpu_data_handle_t *) calloc(_nblocks, sizeof(starpu_data_handle_t));

	assert(_x && _x_handles);

	/* Initialize the vector with random values */
        starpu_srand48(0);
	for (i = 0; i < nelems; i++)
		_x[i] = (TYPE)starpu_drand48();

	unsigned block;
	for (block = 0; block < _nblocks; block++)
	{
		uintptr_t block_start = (uintptr_t)&_x[_entries_per_bock*block];
		starpu_vector_data_register(&_x_handles[block], STARPU_MAIN_RAM, block_start,
					    _entries_per_bock, sizeof(TYPE));
	}

	/* Initialize current min */
	_minmax[0] = TYPE_MAX;

	/* Initialize current max */
	_minmax[1] = TYPE_MIN;

	starpu_variable_data_register(&_minmax_handle, STARPU_MAIN_RAM, (uintptr_t)_minmax, 2*sizeof(TYPE));

	/* Set the methods to define neutral elements and to perform the reduction operation */
	starpu_data_set_reduction_methods(_minmax_handle, &minmax_redux_codelet, &minmax_init_codelet);

	for (block = 0; block < _nblocks; block++)
	{
		struct starpu_task *task = starpu_task_create();

		task->cl = &minmax_codelet;

		task->handles[0] = _x_handles[block];
		task->handles[1] = _minmax_handle;

		ret = starpu_task_submit(task);
		if (ret)
		{
			STARPU_ASSERT(ret == -ENODEV);
			FPRINTF(stderr, "This test can only run on CPUs, but there are no CPU workers (this is not a bug).\n");
			return 77;
		}
	}

	for (block = 0; block < _nblocks; block++)
	{
		starpu_data_unregister(_x_handles[block]);
	}
	starpu_data_unregister(_minmax_handle);

	FPRINTF(stderr, "Min : %e\n", _minmax[0]);
	FPRINTF(stderr, "Max : %e\n", _minmax[1]);

	STARPU_ASSERT(_minmax[0] <= _minmax[1]);

	free(_x);
	free(_x_handles);
	starpu_shutdown();

	return 0;
}