int
main(void)
{
#ifdef STARPU_USE_CPU
int ret;
ret = starpu_init(NULL);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
ncpu = starpu_cpu_worker_get_count();
#ifdef STARPU_USE_CUDA
ncuda = starpu_cuda_worker_get_count();
#endif
#ifdef STARPU_USE_OPENCL
nopencl = starpu_opencl_worker_get_count();
#endif
if (ncpu == 0 || !gpus_available())
{
starpu_shutdown();
return 77;
}
#ifdef STARPU_USE_OPENCL
ret = starpu_opencl_load_opencl_from_file("examples/basic_examples/multiformat_opencl_kernel.cl",
&opencl_program, NULL);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
ret = starpu_opencl_load_opencl_from_file("examples/basic_examples/multiformat_conversion_codelets_opencl_kernel.cl",
&opencl_conversion_program, NULL);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
#endif
init_problem_data();
print_it();
register_data();
create_and_submit_tasks();
unregister_data();
print_it();
#ifdef STARPU_USE_OPENCL
ret = starpu_opencl_unload_opencl(&opencl_program);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_unload_opencl");
starpu_opencl_unload_opencl(&opencl_conversion_program);
#endif
starpu_shutdown();
return check_it();
#else
/* Without the CPU, there is no point in using the multiformat
* interface, so this test is pointless. */
return 77;
#endif
}
int main(int argc, char **argv)
{
int ret;
#ifdef STARPU_QUICK_CHECK
n /= 10;
k /= 16;
#endif
ret = starpu_initialize(NULL, &argc, &argv);
if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
#ifdef STARPU_USE_OPENCL
ret = starpu_opencl_load_opencl_from_file("tests/datawizard/sync_and_notify_data_opencl_codelet.cl", &opencl_code, NULL);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
#endif
starpu_vector_data_register(&v_handle, STARPU_MAIN_RAM, (uintptr_t)v, VECTORSIZE, sizeof(unsigned));
unsigned iter;
for (iter = 0; iter < k; iter++)
{
unsigned ind;
for (ind = 0; ind < n; ind++)
{
/* increment a = v[0] */
struct starpu_codelet cl_inc_a =
{
.cpu_funcs = {cpu_codelet_incA},
#ifdef STARPU_USE_CUDA
.cuda_funcs = {cuda_codelet_incA},
.cuda_flags = {STARPU_CUDA_ASYNC},
#endif
#ifdef STARPU_USE_OPENCL
.opencl_funcs = {opencl_codelet_incA},
.opencl_flags = {STARPU_OPENCL_ASYNC},
int main(int argc, char **argv)
{
int ret;
struct starpu_conf conf;
starpu_conf_init(&conf);
conf.sched_policy_name = "eager";
conf.calibrate = 2;
ret = starpu_initialize(&conf, &argc, &argv);
if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
#ifdef STARPU_USE_OPENCL
ret = starpu_opencl_load_opencl_from_file("tests/perfmodels/opencl_memset_kernel.cl",
&opencl_program, NULL);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
#endif
int slog;
for (slog = START_LOG; slog < END_LOG; slog++)
{
int size = 1 << slog;
test_memset(size);
}
#ifdef STARPU_USE_OPENCL
ret = starpu_opencl_unload_opencl(&opencl_program);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_unload_opencl");
#endif
starpu_shutdown();
return EXIT_SUCCESS;
}
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;
}
int main(int argc, char **argv)
{
unsigned i;
float foo;
starpu_data_handle float_array_handle;
starpu_codelet cl;
starpu_init(NULL);
if (argc == 2) niter = atoi(argv[1]);
foo = 0.0f;
starpu_variable_data_register(&float_array_handle, 0 /* home node */,
(uintptr_t)&foo, sizeof(float));
#ifdef STARPU_USE_OPENCL
starpu_opencl_load_opencl_from_file("examples/basic_examples/variable_kernels_opencl_codelet.cl", &opencl_code);
#endif
cl.where = STARPU_CPU|STARPU_CUDA|STARPU_OPENCL;
cl.cpu_func = cpu_codelet;
#ifdef STARPU_USE_CUDA
cl.cuda_func = cuda_codelet;
#endif
#ifdef STARPU_USE_OPENCL
cl.opencl_func = opencl_codelet;
#endif
cl.nbuffers = 1;
cl.model = NULL;
for (i = 0; i < niter; i++)
{
struct starpu_task *task = starpu_task_create();
int ret;
task->cl = &cl;
task->callback_func = NULL;
task->buffers[0].handle = float_array_handle;
task->buffers[0].mode = STARPU_RW;
ret = starpu_task_submit(task, NULL);
if (STARPU_UNLIKELY(ret == -ENODEV))
{
fprintf(stderr, "No worker may execute this task\n");
exit(0);
}
}
starpu_task_wait_for_all();
/* update the array in RAM */
starpu_data_acquire(float_array_handle, STARPU_R);
fprintf(stderr, "variable -> %f\n", foo);
starpu_data_release(float_array_handle);
starpu_shutdown();
return 0;
}
int main(int argc, char **argv)
{
int ret;
ret = starpu_initialize(NULL, &argc, &argv);
if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
#ifdef STARPU_USE_OPENCL
ret = starpu_opencl_load_opencl_from_file("tests/datawizard/scratch_opencl_kernel.cl",
&opencl_program, NULL);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
#endif
A = (unsigned *) calloc(VECTORSIZE, sizeof(unsigned));
starpu_vector_data_register(&A_handle, STARPU_MAIN_RAM, (uintptr_t)A, VECTORSIZE, sizeof(unsigned));
starpu_vector_data_register(&B_handle, -1, (uintptr_t)NULL, VECTORSIZE, sizeof(unsigned));
unsigned loop;
for (loop = 0; loop < NLOOPS; loop++)
{
struct starpu_task *task_f = starpu_task_create();
task_f->cl = &cl_f;
task_f->handles[0] = A_handle;
task_f->handles[1] = B_handle;
ret = starpu_task_submit(task_f);
if (ret == -ENODEV) goto enodev;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_submit");
}
ret = starpu_task_wait_for_all();
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_wait_for_all");
starpu_data_unregister(A_handle);
starpu_data_unregister(B_handle);
#ifdef STARPU_USE_OPENCL
ret = starpu_opencl_unload_opencl(&opencl_program);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_unload_opencl");
#endif
starpu_shutdown();
/* Check result */
unsigned i;
ret = EXIT_SUCCESS;
for (i = 0; i < VECTORSIZE; i++)
{
if (A[i] != NLOOPS)
{
FPRINTF(stderr, "Error: Incorrect value A[%u] = %u != %d\n", i, A[i], NLOOPS);
ret = EXIT_FAILURE;
break;
}
}
free(A);
STARPU_RETURN(ret);
enodev:
starpu_data_unregister(A_handle);
starpu_data_unregister(B_handle);
#ifdef STARPU_USE_OPENCL
ret = starpu_opencl_unload_opencl(&opencl_program);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_unload_opencl");
#endif
starpu_shutdown();
/* 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 */
fprintf(stderr, "WARNING: No one can execute this task\n");
return STARPU_TEST_SKIPPED;
}
int main(int argc, char **argv)
{
unsigned *foo;
starpu_data_handle_t handle;
int ret;
unsigned n, i, size;
ret = starpu_initialize(NULL, &argc, &argv);
if (ret == -ENODEV) return STARPU_TEST_SKIPPED;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
#ifdef STARPU_USE_OPENCL
ret = starpu_opencl_load_opencl_from_file("tests/datawizard/scal_opencl.cl", &opencl_program, NULL);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
#endif
n = starpu_worker_get_count();
if (n == 1)
{
starpu_shutdown();
return STARPU_TEST_SKIPPED;
}
size = 10 * n;
foo = (unsigned *) calloc(size, sizeof(*foo));
for (i = 0; i < size; i++)
foo[i] = i;
starpu_vector_data_register(&handle, STARPU_MAIN_RAM, (uintptr_t)foo, size, sizeof(*foo));
/* Broadcast the data to force in-place partitioning */
for (i = 0; i < n; i++)
starpu_data_prefetch_on_node(handle, starpu_worker_get_memory_node(i), 0);
struct starpu_data_filter f =
{
.filter_func = starpu_vector_filter_block,
.nchildren = n,
};
starpu_data_partition(handle, &f);
for (i = 0; i < f.nchildren; i++)
{
struct starpu_task *task = starpu_task_create();
task->handles[0] = starpu_data_get_sub_data(handle, 1, i);
task->cl = &scal_codelet;
task->execute_on_a_specific_worker = 1;
task->workerid = i;
ret = starpu_task_submit(task);
if (ret == -ENODEV) goto enodev;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_submit");
}
ret = starpu_task_wait_for_all();
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_wait_for_all");
starpu_data_unpartition(handle, STARPU_MAIN_RAM);
starpu_data_unregister(handle);
starpu_shutdown();
ret = EXIT_SUCCESS;
for (i = 0; i < size; i++)
{
if (foo[i] != i*2)
{
FPRINTF(stderr,"value %u is %u instead of %u\n", i, foo[i], 2*i);
ret = EXIT_FAILURE;
}
}
return ret;
enodev:
starpu_data_unregister(handle);
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 */
starpu_shutdown();
return STARPU_TEST_SKIPPED;
}
int main(int argc, char **argv)
{
int ret, exit_value = 0;
/* Initialize StarPU */
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/axpy/axpy_opencl_kernel.cl",
&opencl_program, NULL);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
#endif
starpu_cublas_init();
/* This is equivalent to
vec_a = malloc(N*sizeof(TYPE));
vec_b = malloc(N*sizeof(TYPE));
*/
starpu_malloc((void **)&_vec_x, N*sizeof(TYPE));
assert(_vec_x);
starpu_malloc((void **)&_vec_y, N*sizeof(TYPE));
assert(_vec_y);
unsigned i;
for (i = 0; i < N; i++)
{
_vec_x[i] = 1.0f; /*(TYPE)starpu_drand48(); */
_vec_y[i] = 4.0f; /*(TYPE)starpu_drand48(); */
}
FPRINTF(stderr, "BEFORE x[0] = %2.2f\n", _vec_x[0]);
FPRINTF(stderr, "BEFORE y[0] = %2.2f\n", _vec_y[0]);
/* Declare the data to StarPU */
starpu_vector_data_register(&_handle_x, STARPU_MAIN_RAM, (uintptr_t)_vec_x, N, sizeof(TYPE));
starpu_vector_data_register(&_handle_y, STARPU_MAIN_RAM, (uintptr_t)_vec_y, N, sizeof(TYPE));
/* Divide the vector into blocks */
struct starpu_data_filter block_filter =
{
.filter_func = starpu_vector_filter_block,
.nchildren = NBLOCKS
};
starpu_data_partition(_handle_x, &block_filter);
starpu_data_partition(_handle_y, &block_filter);
double start;
double end;
start = starpu_timing_now();
unsigned b;
for (b = 0; b < NBLOCKS; b++)
{
struct starpu_task *task = starpu_task_create();
task->cl = &axpy_cl;
task->cl_arg = &_alpha;
task->cl_arg_size = sizeof(_alpha);
task->handles[0] = starpu_data_get_sub_data(_handle_x, 1, b);
task->handles[1] = starpu_data_get_sub_data(_handle_y, 1, b);
task->tag_id = b;
ret = starpu_task_submit(task);
if (ret == -ENODEV)
{
exit_value = 77;
goto enodev;
}
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_submit");
}
starpu_task_wait_for_all();
enodev:
starpu_data_unpartition(_handle_x, STARPU_MAIN_RAM);
starpu_data_unpartition(_handle_y, STARPU_MAIN_RAM);
starpu_data_unregister(_handle_x);
starpu_data_unregister(_handle_y);
end = starpu_timing_now();
double timing = end - start;
FPRINTF(stderr, "timing -> %2.2f us %2.2f MB/s\n", timing, 3*N*sizeof(TYPE)/timing);
FPRINTF(stderr, "AFTER y[0] = %2.2f (ALPHA = %2.2f)\n", _vec_y[0], _alpha);
if (exit_value != 77)
exit_value = check();
starpu_free((void *)_vec_x);
starpu_free((void *)_vec_y);
#ifdef STARPU_USE_OPENCL
ret = starpu_opencl_unload_opencl(&opencl_program);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_unload_opencl");
#endif
/* Stop StarPU */
starpu_shutdown();
return exit_value;
}
int main(int argc, char **argv)
{
/* We consider a vector of float that is initialized just as any of C
* data */
float vector[NX];
unsigned i;
for (i = 0; i < NX; i++)
vector[i] = 1.0f;
fprintf(stderr, "BEFORE : First element was %f\n", vector[0]);
/* Initialize StarPU with default configuration */
int ret = starpu_init(NULL);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
#ifdef STARPU_USE_OPENCL
starpu_opencl_load_opencl_from_file("vector_scal_opencl_kernel.cl", &programs, NULL);
#endif
/* Tell StaPU to associate the "vector" vector with the "vector_handle"
* identifier. When a task needs to access a piece of data, it should
* refer to the handle that is associated to it.
* In the case of the "vector" data interface:
* - the first argument of the registration method is a pointer to the
* handle that should describe the data
* - the second argument is the memory node where the data (ie. "vector")
* resides initially: STARPU_MAIN_RAM stands for an address in main memory, as
* opposed to an adress on a GPU for instance.
* - the third argument is the adress of the vector in RAM
* - the fourth argument is the number of elements in the vector
* - the fifth argument is the size of each element.
*/
starpu_data_handle_t vector_handle;
starpu_vector_data_register(&vector_handle, STARPU_MAIN_RAM, (uintptr_t)vector,
NX, sizeof(vector[0]));
float factor = 3.14;
/* create a synchronous task: any call to starpu_task_submit will block
* until it is terminated */
struct starpu_task *task = starpu_task_create();
task->synchronous = 1;
task->cl = &cl;
/* the codelet manipulates one buffer in RW mode */
task->handles[0] = vector_handle;
/* an argument is passed to the codelet, beware that this is a
* READ-ONLY buffer and that the codelet may be given a pointer to a
* COPY of the argument */
task->cl_arg = &factor;
task->cl_arg_size = sizeof(factor);
/* execute the task on any eligible computational ressource */
ret = starpu_task_submit(task);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_submit");
/* StarPU does not need to manipulate the array anymore so we can stop
* monitoring it */
starpu_data_unregister(vector_handle);
#ifdef STARPU_USE_OPENCL
starpu_opencl_unload_opencl(&programs);
#endif
/* terminate StarPU, no task can be submitted after */
starpu_shutdown();
fprintf(stderr, "AFTER First element is %f\n", vector[0]);
return 0;
}
void cpu_to_opencl_opencl_func(void *buffers[], void *args)
{
STARPU_SKIP_IF_VALGRIND;
(void) args;
int id, devid, ret;
cl_int err;
cl_kernel kernel;
cl_command_queue queue;
cl_event event;
unsigned n = STARPU_MULTIFORMAT_GET_NX(buffers[0]);
cl_mem src = (cl_mem) STARPU_MULTIFORMAT_GET_CPU_PTR(buffers[0]);
cl_mem dst = (cl_mem) STARPU_MULTIFORMAT_GET_OPENCL_PTR(buffers[0]);
id = starpu_worker_get_id();
devid = starpu_worker_get_devid(id);
ret = starpu_opencl_load_opencl_from_file(KERNEL_LOCATION,
&opencl_conversion_program,
NULL);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
err = starpu_opencl_load_kernel(&kernel,
&queue,
&opencl_conversion_program,
"cpu_to_opencl_opencl",
devid);
if (err != CL_SUCCESS)
STARPU_OPENCL_REPORT_ERROR(err);
err = clSetKernelArg(kernel, 0, sizeof(src), &src);
if (err != CL_SUCCESS)
STARPU_OPENCL_REPORT_ERROR(err);
err = clSetKernelArg(kernel, 1, sizeof(dst), &dst);
if (err != CL_SUCCESS)
STARPU_OPENCL_REPORT_ERROR(err);
err = clSetKernelArg(kernel, 2, sizeof(n), &n);
if (err != CL_SUCCESS)
STARPU_OPENCL_REPORT_ERROR(err);
{
size_t global=n;
size_t local;
size_t s;
cl_device_id device;
starpu_opencl_get_device(devid, &device);
err = clGetKernelWorkGroupInfo (kernel,
device,
CL_KERNEL_WORK_GROUP_SIZE,
sizeof(local),
&local,
&s);
if (err != CL_SUCCESS)
STARPU_OPENCL_REPORT_ERROR(err);
if (local > global)
local = global;
err = clEnqueueNDRangeKernel(queue,
kernel,
1,
NULL,
&global,
&local,
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);
ret = starpu_opencl_unload_opencl(&opencl_conversion_program);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_unload_opencl");
}
int main(int argc, char **argv)
{
int ret = 0;
starpu_data_handle_t handle1;
starpu_data_handle_t handle2;
double real = 45.0;
double imaginary = 12.0;
double copy_real = 78.0;
double copy_imaginary = 78.0;
int compare;
int *compare_ptr = &compare;
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/interface/complex_kernels.cl",
&opencl_program, NULL);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
#endif
starpu_complex_data_register(&handle1, STARPU_MAIN_RAM, &real, &imaginary, 1);
starpu_complex_data_register(&handle2, STARPU_MAIN_RAM, ©_real, ©_imaginary, 1);
ret = starpu_task_insert(&cl_display, STARPU_VALUE, "handle1", strlen("handle1")+1, STARPU_R, handle1, 0);
if (ret == -ENODEV) goto end;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_insert");
ret = starpu_task_insert(&cl_display, STARPU_VALUE, "handle2", strlen("handle2")+1, STARPU_R, handle2, 0);
if (ret == -ENODEV) goto end;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_insert");
ret = starpu_task_insert(&cl_compare,
STARPU_R, handle1,
STARPU_R, handle2,
STARPU_VALUE, &compare_ptr, sizeof(compare_ptr),
0);
if (ret == -ENODEV) goto end;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_insert");
starpu_task_wait_for_all();
if (compare != 0)
{
FPRINTF(stderr, "Complex numbers should NOT be similar\n");
goto end;
}
ret = starpu_task_insert(&cl_copy,
STARPU_R, handle1,
STARPU_W, handle2,
0);
if (ret == -ENODEV) goto end;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_insert");
ret = starpu_task_insert(&cl_display, STARPU_VALUE, "handle1", strlen("handle1")+1, STARPU_R, handle1, 0);
if (ret == -ENODEV) goto end;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_insert");
ret = starpu_task_insert(&cl_display, STARPU_VALUE, "handle2", strlen("handle2")+1, STARPU_R, handle2, 0);
if (ret == -ENODEV) goto end;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_insert");
ret = starpu_task_insert(&cl_compare,
STARPU_R, handle1,
STARPU_R, handle2,
STARPU_VALUE, &compare_ptr, sizeof(compare_ptr),
0);
if (ret == -ENODEV) goto end;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_insert");
starpu_task_wait_for_all();
if (compare != 1)
{
FPRINTF(stderr, "Complex numbers should be similar\n");
}
end:
#ifdef STARPU_USE_OPENCL
{
int ret2 = starpu_opencl_unload_opencl(&opencl_program);
STARPU_CHECK_RETURN_VALUE(ret2, "starpu_opencl_unload_opencl");
}
#endif
starpu_data_unregister(handle1);
starpu_data_unregister(handle2);
starpu_shutdown();
if (ret == -ENODEV) return 77; else return !compare;
}
void test_variable_opencl_func(void *buffers[], void *args)
{
STARPU_SKIP_IF_VALGRIND;
int id, devid, ret;
int factor = *(int *) args;
cl_int err;
cl_kernel kernel;
cl_command_queue queue;
cl_event event;
ret = starpu_opencl_load_opencl_from_file(KERNEL_LOCATION, &opencl_program, NULL);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
cl_mem val = (cl_mem)STARPU_VARIABLE_GET_PTR(buffers[0]);
cl_context context;
id = starpu_worker_get_id();
devid = starpu_worker_get_devid(id);
starpu_opencl_get_context(devid, &context);
cl_mem fail = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR,
sizeof(int), &variable_config.copy_failed, &err);
if (err != CL_SUCCESS)
STARPU_OPENCL_REPORT_ERROR(err);
err = starpu_opencl_load_kernel(&kernel,
&queue,
&opencl_program,
"variable_opencl",
devid);
if (err != CL_SUCCESS)
STARPU_OPENCL_REPORT_ERROR(err);
err = clSetKernelArg(kernel, 0, sizeof(val), &val);
if (err != CL_SUCCESS)
STARPU_OPENCL_REPORT_ERROR(err);
err = clSetKernelArg(kernel, 1, sizeof(fail), &fail);
if (err)
STARPU_OPENCL_REPORT_ERROR(err);
err = clSetKernelArg(kernel, 2, sizeof(factor), &factor);
if (err)
STARPU_OPENCL_REPORT_ERROR(err);
{
size_t global = 1;
size_t local;
size_t s;
cl_device_id device;
starpu_opencl_get_device(devid, &device);
err = clGetKernelWorkGroupInfo (kernel,
device,
CL_KERNEL_WORK_GROUP_SIZE,
sizeof(local),
&local,
&s);
if (err != CL_SUCCESS)
STARPU_OPENCL_REPORT_ERROR(err);
if (local > global)
local = global;
err = clEnqueueNDRangeKernel(queue,
kernel,
1,
NULL,
&global,
&local,
0,
NULL,
&event);
if (err != CL_SUCCESS)
STARPU_OPENCL_REPORT_ERROR(err);
}
err = clEnqueueReadBuffer(queue,
fail,
CL_TRUE,
0,
sizeof(int),
&variable_config.copy_failed,
0,
NULL,
NULL);
if (err != CL_SUCCESS)
STARPU_OPENCL_REPORT_ERROR(err);
clFinish(queue);
starpu_opencl_collect_stats(event);
clReleaseEvent(event);
starpu_opencl_release_kernel(kernel);
ret = starpu_opencl_unload_opencl(&opencl_program);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_unload_opencl");
return;
}
