void par_sort(void* arg) {
sort_data_t *in = (sort_data_t*) arg;
TYPE* data = in->buffer;
int left = in->left;
int right = in->right;
if (right - left + 1 > HC_GRANULARITY) {
int index = partition(data, left, right);
hclib_start_finish(); {
if (left < index - 1) {
sort_data_t* buf = (sort_data_t*) malloc(sizeof(sort_data_t));
buf->buffer = data;
buf->left = left;
buf->right = index - 1;
{
pragma137_omp_task *new_ctx = (pragma137_omp_task *)malloc(sizeof(pragma137_omp_task));
new_ctx->buf_ptr = &(buf);
new_ctx->index_ptr = &(index);
new_ctx->in_ptr = &(in);
new_ctx->data_ptr = &(data);
new_ctx->left_ptr = &(left);
new_ctx->right_ptr = &(right);
new_ctx->arg_ptr = &(arg);
hclib_async(pragma137_omp_task_hclib_async, new_ctx, NO_FUTURE, ANY_PLACE);
}
}
if (index < right) {
sort_data_t* buf = (sort_data_t*) malloc(sizeof(sort_data_t));
buf->buffer = data;
buf->left = index;
buf->right = right;
{
pragma147_omp_task *new_ctx = (pragma147_omp_task *)malloc(sizeof(pragma147_omp_task));
new_ctx->buf_ptr = &(buf);
new_ctx->index_ptr = &(index);
new_ctx->in_ptr = &(in);
new_ctx->data_ptr = &(data);
new_ctx->left_ptr = &(left);
new_ctx->right_ptr = &(right);
new_ctx->arg_ptr = &(arg);
hclib_async(pragma147_omp_task_hclib_async, new_ctx, NO_FUTURE, ANY_PLACE);
}
}
} ; hclib_end_finish();
}
else {
// quicksort in C library
qsort(data+left, right - left + 1, sizeof(TYPE), compare);
}
free(arg);
}
void fib(void * raw_args) {
FibArgs *args = raw_args;
if (args->n < 2) {
args->res = args->n;
}
else {
FibArgs lhsArgs = { args->n - 1, 0 };
FibArgs rhsArgs = { args->n - 2, 0 };
FINISH {
hclib_async(fib, &lhsArgs, NO_FUTURE, NO_PHASER, ANY_PLACE, NO_PROP);
hclib_async(fib, &rhsArgs, NO_FUTURE, NO_PHASER, ANY_PLACE, NO_PROP);
}
args->res = lhsArgs.res + rhsArgs.res;
}
}
void hclib_async_memset_helper(place_t *pl, void *ptr, int val, size_t nbytes,
hclib_future_t **future_list, void *user_arg,
hclib_promise_t *out_promise) {
gpu_task_t *task = malloc(sizeof(gpu_task_t));
task->t._fp = NULL;
task->t.is_asyncAnyType = 0;
task->t.future_list = NULL;
task->t.args = NULL;
task->t.place = NULL;
hclib_promise_init(out_promise);
task->gpu_type = GPU_MEMSET_TASK;
task->promise_to_put = out_promise;
task->arg_to_put = user_arg;
task->gpu_task_def.memset_task.pl = pl;
task->gpu_task_def.memset_task.ptr = ptr;
task->gpu_task_def.memset_task.val = val;
task->gpu_task_def.memset_task.nbytes = nbytes;
#ifdef VERBOSE
fprintf(stderr, "hclib_async_memset: pl=%p ptr=%p nbytes=%lu\n", pl, ptr,
(unsigned long)nbytes);
#endif
if (future_list) {
hclib_async(async_gpu_task_launcher, task, future_list, NULL, NULL, 0);
} else {
spawn_gpu_task((hclib_task_t *)task);
}
}
void hclib_async_copy_helper(place_t *dst_pl, void *dst, place_t *src_pl,
void *src, size_t nbytes, hclib_future_t **future_list,
void *user_arg, hclib_promise_t *out_promise) {
gpu_task_t *task = malloc(sizeof(gpu_task_t));
task->t._fp = NULL;
task->t.is_asyncAnyType = 0;
task->t.future_list = NULL;
task->t.args = NULL;
task->t.place = NULL;
hclib_promise_init(out_promise);
task->gpu_type = GPU_COMM_TASK;
task->promise_to_put = out_promise;
task->arg_to_put = user_arg;
task->gpu_task_def.comm_task.src_pl = src_pl;
task->gpu_task_def.comm_task.dst_pl = dst_pl;
task->gpu_task_def.comm_task.src = src;
task->gpu_task_def.comm_task.dst = dst;
task->gpu_task_def.comm_task.nbytes = nbytes;
#ifdef VERBOSE
fprintf(stderr, "hclib_async_copy: dst_pl=%p dst=%p src_pl=%p src=%p "
"nbytes=%lu future_list=%p\n", dst_pl, dst, src_pl, src,
(unsigned long)nbytes, future_list);
#endif
if (future_list) {
hclib_async(async_gpu_task_launcher, task, future_list, NULL, NULL, 0);
} else {
spawn_gpu_task((hclib_task_t *)task);
}
}
void fib_ddt(void * raw_args) {
FibDDtArgs *args = raw_args;
if (args->n < 2) {
args->resval = args->n;
hclib_promise_put(args->res, args);
}
else {
FibDDtArgs *lhsArgs = setup_fib_ddt_args(args->n - 1);
FibDDtArgs *rhsArgs = setup_fib_ddt_args(args->n - 2);
args->subres[0] = lhsArgs->res;
args->subres[1] = rhsArgs->res;
// sub-computation asyncs
hclib_async(fib_ddt, lhsArgs, NO_FUTURE, NO_PHASER, ANY_PLACE, MY_ESCAPE_PROP);
hclib_async(fib_ddt, rhsArgs, NO_FUTURE, NO_PHASER, ANY_PLACE, MY_ESCAPE_PROP);
// async-await for sub-results
hclib_async(fib_ddt_res, args, ps2fs(args->subres), NO_PHASER, ANY_PLACE, MY_ESCAPE_PROP);
}
}
hclib_future_t *hclib_async_future(future_fct_t fp, void *arg,
hclib_future_t **futures, const int nfutures,
hclib_locale_t *locale) {
future_args_wrapper *wrapper = malloc(sizeof(future_args_wrapper));
hclib_promise_init(&wrapper->event);
wrapper->fp = fp;
wrapper->actual_in = arg;
hclib_async(future_caller, wrapper, futures, nfutures, locale);
return hclib_get_future_for_promise(&wrapper->event);
}
void spawn_async(volatile int * indices, int i) {
if (i < NB_ASYNC) {
hclib_start_finish();
indices[i] = i;
hclib_async(async_fct, (void*) (indices+i), NO_FUTURE, NO_PHASER,
ANY_PLACE, NO_PROP);
spawn_async(indices, i+1);
hclib_end_finish();
assert_done(i, i+1);
}
}
long long fib (int n)
{
long long x, y;
if (n < 2) return n;
{
pragma44_omp_task *new_ctx = (pragma44_omp_task *)malloc(sizeof(pragma44_omp_task));
new_ctx->x_ptr = &(x);
new_ctx->y_ptr = &(y);
new_ctx->n = n;
hclib_async(pragma44_omp_task_hclib_async, new_ctx, NO_FUTURE, ANY_PLACE);
} ;
{
pragma46_omp_task *new_ctx = (pragma46_omp_task *)malloc(sizeof(pragma46_omp_task));
new_ctx->x_ptr = &(x);
new_ctx->y_ptr = &(y);
new_ctx->n = n;
hclib_async(pragma46_omp_task_hclib_async, new_ctx, NO_FUTURE, ANY_PLACE);
} ;
hclib_end_finish(); hclib_start_finish(); ;
return x + y;
}
void hclib_launch(generic_frame_ptr fct_ptr, void *arg, const char **deps,
int ndeps) {
unsigned long long start_time = 0;
unsigned long long end_time;
const int instrument = (getenv("HCLIB_INSTRUMENT") != NULL);
hclib_init(deps, ndeps, instrument);
if (profile_launch_body) {
start_time = current_time_ns();
}
hclib_async(fct_ptr, arg, NULL, 0, hclib_get_closest_locale());
hclib_finalize(instrument);
if (profile_launch_body) {
end_time = current_time_ns();
printf("\nHCLIB TIME %llu ns\n", end_time - start_time);
}
}
void taskMain(void *raw_args) {
char **argv = raw_args;
const int n = atoi(argv[1]);
const int doDDT = argv[2] && atoi(argv[2]);
const long fn = fib_iter(n);
const long fnp1 = fib_iter(n+1);
long answer;
double t_start, t_end;
// ASYNC-FINISH version
if (!doDDT) {
printf("async/finish version\n");
t_start = get_seconds();
FibArgs args = { n, 0 };
FINISH {
hclib_async(fib, &args, NO_FUTURE, NO_PHASER, ANY_PLACE, NO_PROP);
}
t_end = get_seconds();
answer = args.res;
//printf("asyncs = %ld\tfins=%ld\n", 2*fnp1-1, fnp1);
}
void nqueens(int n, int j, char *a, int *solutions, int depth)
{
int *csols;
int i;
if (n == j) {
/* good solution, count it */
*solutions = 1;
return;
}
*solutions = 0;
csols = (int *)malloc(n*sizeof(int));
memset(csols,0,n*sizeof(int));
/* try each possible position for queen <j> */
for (i = 0; i < n; i++) {
{
pragma126_omp_task *new_ctx = (pragma126_omp_task *)malloc(sizeof(pragma126_omp_task));
new_ctx->csols = csols;
new_ctx->i = i;
new_ctx->n = n;
new_ctx->j = j;
new_ctx->a = a;
new_ctx->solutions = solutions;
new_ctx->depth = depth;
hclib_async(pragma126_omp_task_hclib_async, new_ctx, NO_FUTURE, ANY_PLACE);
}
}
hclib_end_finish(); hclib_start_finish(); ;
for ( i = 0; i < n; i++) *solutions += csols[i];
free(csols);
}
void sweep (int nx, int ny, double dx, double dy, double *f_,
int itold, int itnew, double *u_, double *unew_, int block_size)
{
int it;
int block_x, block_y;
if (block_size == 0)
block_size = nx;
int max_blocks_x = (nx / block_size);
int max_blocks_y = (ny / block_size);
hclib_start_finish(); {
for (it = itold + 1; it <= itnew; it++) {
// Save the current estimate.
for (block_x = 0; block_x < max_blocks_x; block_x++) {
for (block_y = 0; block_y < max_blocks_y; block_y++) {
{
pragma26_omp_task *new_ctx = (pragma26_omp_task *)malloc(sizeof(pragma26_omp_task));
new_ctx->it_ptr = &(it);
new_ctx->block_x = block_x;
new_ctx->block_y = block_y;
new_ctx->max_blocks_x_ptr = &(max_blocks_x);
new_ctx->max_blocks_y_ptr = &(max_blocks_y);
new_ctx->nx_ptr = &(nx);
new_ctx->ny_ptr = &(ny);
new_ctx->dx_ptr = &(dx);
new_ctx->dy_ptr = &(dy);
new_ctx->f__ptr = &(f_);
new_ctx->itold_ptr = &(itold);
new_ctx->itnew_ptr = &(itnew);
new_ctx->u__ptr = &(u_);
new_ctx->unew__ptr = &(unew_);
new_ctx->block_size_ptr = &(block_size);
hclib_async(pragma26_omp_task_hclib_async, new_ctx, NO_FUTURE, ANY_PLACE);
} ;
}
}
hclib_end_finish(); hclib_start_finish(); ;
// Compute a new estimate.
for (block_x = 0; block_x < max_blocks_x; block_x++) {
for (block_y = 0; block_y < max_blocks_y; block_y++) {
{
pragma36_omp_task *new_ctx = (pragma36_omp_task *)malloc(sizeof(pragma36_omp_task));
new_ctx->block_x = block_x;
new_ctx->block_y = block_y;
new_ctx->nx_ptr = &(nx);
new_ctx->ny_ptr = &(ny);
new_ctx->dx_ptr = &(dx);
new_ctx->dy_ptr = &(dy);
new_ctx->f__ptr = &(f_);
new_ctx->u__ptr = &(u_);
new_ctx->unew__ptr = &(unew_);
new_ctx->block_size_ptr = &(block_size);
hclib_async(pragma36_omp_task_hclib_async, new_ctx, NO_FUTURE, ANY_PLACE);
} ;
}
}
hclib_end_finish(); hclib_start_finish(); ;
}
} ; hclib_end_finish();
}
void shmem_task_nbi (void (*body)(void *), void *user_data, shmem_future_t **optional_future)
{
hclib_async(body, user_data, optional_future, NULL, NULL, 0);
}