static void ws_add_workers(unsigned sched_ctx_id, int *workerids,unsigned nworkers)
{
struct _starpu_work_stealing_data *ws = (struct _starpu_work_stealing_data*)starpu_sched_ctx_get_policy_data(sched_ctx_id);
unsigned i;
int workerid;
for (i = 0; i < nworkers; i++)
{
workerid = workerids[i];
starpu_sched_ctx_worker_shares_tasks_lists(workerid, sched_ctx_id);
ws->queue_array[workerid] = _starpu_create_deque();
/* Tell helgrid that we are fine with getting outdated values,
* this is just an estimation */
STARPU_HG_DISABLE_CHECKING(ws->queue_array[workerid]->njobs);
/**
* The first WS_POP_TASK will increase NPROCESSED though no task was actually performed yet,
* we need to initialize it at -1.
*/
ws->queue_array[workerid]->nprocessed = -1;
ws->queue_array[workerid]->njobs = 0;
}
}
static void _starpu_data_partition(starpu_data_handle_t initial_handle, starpu_data_handle_t *childrenp, unsigned nparts, struct starpu_data_filter *f, int inherit_state)
{
unsigned i;
unsigned node;
/* first take care to properly lock the data header */
_starpu_spin_lock(&initial_handle->header_lock);
initial_handle->nplans++;
STARPU_ASSERT_MSG(nparts > 0, "Partitioning data %p in 0 piece does not make sense", initial_handle);
/* allocate the children */
if (inherit_state)
{
initial_handle->children = (struct _starpu_data_state *) calloc(nparts, sizeof(struct _starpu_data_state));
STARPU_ASSERT(initial_handle->children);
/* this handle now has children */
initial_handle->nchildren = nparts;
}
unsigned nworkers = starpu_worker_get_count();
for (node = 0; node < STARPU_MAXNODES; node++)
{
if (initial_handle->per_node[node].state != STARPU_INVALID)
break;
}
if (node == STARPU_MAXNODES)
{
/* This is lazy allocation, allocate it now in main RAM, so as
* to have somewhere to gather pieces later */
/* FIXME: mark as unevictable! */
int ret = _starpu_allocate_memory_on_node(initial_handle, &initial_handle->per_node[STARPU_MAIN_RAM], 0);
#ifdef STARPU_DEVEL
#warning we should reclaim memory if allocation failed
#endif
STARPU_ASSERT(!ret);
}
for (i = 0; i < nparts; i++)
{
starpu_data_handle_t child;
if (inherit_state)
child = &initial_handle->children[i];
else
child = childrenp[i];
STARPU_ASSERT(child);
struct starpu_data_interface_ops *ops;
/* each child may have his own interface type */
/* what's this child's interface ? */
if (f->get_child_ops)
ops = f->get_child_ops(f, i);
else
ops = initial_handle->ops;
_starpu_data_handle_init(child, ops, initial_handle->mf_node);
child->nchildren = 0;
child->nplans = 0;
child->switch_cl = NULL;
child->partitioned = 0;
child->readonly = 0;
child->mpi_data = initial_handle->mpi_data;
child->root_handle = initial_handle->root_handle;
child->father_handle = initial_handle;
child->sibling_index = i;
child->depth = initial_handle->depth + 1;
child->is_not_important = initial_handle->is_not_important;
child->wt_mask = initial_handle->wt_mask;
child->home_node = initial_handle->home_node;
child->is_readonly = initial_handle->is_readonly;
/* initialize the chunk lock */
_starpu_data_requester_list_init(&child->req_list);
_starpu_data_requester_list_init(&child->reduction_req_list);
child->reduction_tmp_handles = NULL;
child->write_invalidation_req = NULL;
child->refcnt = 0;
child->unlocking_reqs = 0;
child->busy_count = 0;
child->busy_waiting = 0;
STARPU_PTHREAD_MUTEX_INIT(&child->busy_mutex, NULL);
STARPU_PTHREAD_COND_INIT(&child->busy_cond, NULL);
child->reduction_refcnt = 0;
_starpu_spin_init(&child->header_lock);
child->sequential_consistency = initial_handle->sequential_consistency;
STARPU_PTHREAD_MUTEX_INIT(&child->sequential_consistency_mutex, NULL);
child->last_submitted_mode = STARPU_R;
child->last_sync_task = NULL;
child->last_submitted_accessors.task = NULL;
child->last_submitted_accessors.next = &child->last_submitted_accessors;
child->last_submitted_accessors.prev = &child->last_submitted_accessors;
child->post_sync_tasks = NULL;
/* Tell helgrind that the race in _starpu_unlock_post_sync_tasks is fine */
STARPU_HG_DISABLE_CHECKING(child->post_sync_tasks_cnt);
child->post_sync_tasks_cnt = 0;
/* The methods used for reduction are propagated to the
* children. */
child->redux_cl = initial_handle->redux_cl;
child->init_cl = initial_handle->init_cl;
#ifdef STARPU_USE_FXT
child->last_submitted_ghost_sync_id_is_valid = 0;
child->last_submitted_ghost_sync_id = 0;
child->last_submitted_ghost_accessors_id = NULL;
#endif
if (_starpu_global_arbiter)
/* Just for testing purpose */
starpu_data_assign_arbiter(child, _starpu_global_arbiter);
else
child->arbiter = NULL;
_starpu_data_requester_list_init(&child->arbitered_req_list);
for (node = 0; node < STARPU_MAXNODES; node++)
{
struct _starpu_data_replicate *initial_replicate;
struct _starpu_data_replicate *child_replicate;
initial_replicate = &initial_handle->per_node[node];
child_replicate = &child->per_node[node];
if (inherit_state)
child_replicate->state = initial_replicate->state;
else
child_replicate->state = STARPU_INVALID;
if (inherit_state || !initial_replicate->automatically_allocated)
child_replicate->allocated = initial_replicate->allocated;
else
child_replicate->allocated = 0;
/* Do not allow memory reclaiming within the child for parent bits */
child_replicate->automatically_allocated = 0;
child_replicate->refcnt = 0;
child_replicate->memory_node = node;
child_replicate->relaxed_coherency = 0;
if (inherit_state)
child_replicate->initialized = initial_replicate->initialized;
else
child_replicate->initialized = 0;
/* update the interface */
void *initial_interface = starpu_data_get_interface_on_node(initial_handle, node);
void *child_interface = starpu_data_get_interface_on_node(child, node);
STARPU_ASSERT_MSG(!(!inherit_state && child_replicate->automatically_allocated && child_replicate->allocated), "partition planning is currently not supported when handle has some automatically allocated buffers");
f->filter_func(initial_interface, child_interface, f, i, nparts);
}
unsigned worker;
for (worker = 0; worker < nworkers; worker++)
{
struct _starpu_data_replicate *child_replicate;
child_replicate = &child->per_worker[worker];
child_replicate->state = STARPU_INVALID;
child_replicate->allocated = 0;
child_replicate->automatically_allocated = 0;
child_replicate->refcnt = 0;
child_replicate->memory_node = starpu_worker_get_memory_node(worker);
child_replicate->requested = 0;
for (node = 0; node < STARPU_MAXNODES; node++)
{
child_replicate->request[node] = NULL;
}
child_replicate->relaxed_coherency = 1;
child_replicate->initialized = 0;
/* duplicate the content of the interface on node 0 */
memcpy(child_replicate->data_interface, child->per_node[0].data_interface, child->ops->interface_size);
}
/* We compute the size and the footprint of the child once and
* store it in the handle */
child->footprint = _starpu_compute_data_footprint(child);
void *ptr;
ptr = starpu_data_handle_to_pointer(child, STARPU_MAIN_RAM);
if (ptr != NULL)
_starpu_data_register_ram_pointer(child, ptr);
}
/* now let the header */
_starpu_spin_unlock(&initial_handle->header_lock);
}
