Bidouille

mpi/src/mpi_failure_tolerance/starpu_mpi_checkpoint.c

@@ -221,7 +221,10 @@ int starpu_mpi_submit_checkpoint_template(starpu_mpi_checkpoint_template_t cp_te
 					arg->type = STARPU_R;
 					arg->count = item->count;
 					arg->msg.checkpoint_id = cp_template->cp_id;
+					arg->msg.klm=42;
 					arg->msg.checkpoint_instance = current_instance;
+					if (arg->msg.checkpoint_instance>3)
+						fprintf(stderr, "arg->msg.checkpoint_instance:%d\n", arg->msg.checkpoint_instance);
 					_starpu_mpi_isend_cache_aware(handle, item->backupped_by, starpu_mpi_data_get_tag(handle), MPI_COMM_WORLD, 1, 0, prio,
 					                              &_send_cp_internal_data_cb, (void*)arg, 1, &arg->cache_flag);
 					// the callbacks need to post ack recv. The cache one needs to release the handle.
@@ -241,7 +244,10 @@ int starpu_mpi_submit_checkpoint_template(starpu_mpi_checkpoint_template_t cp_te
 					arg->type = STARPU_R;
 					arg->count = item->count;
 					arg->msg.checkpoint_id = cp_template->cp_id;
+					arg->msg.klm=42;
 					arg->msg.checkpoint_instance = current_instance;
+					if (arg->msg.checkpoint_instance>3)
+						fprintf(stderr, "arg->msg.checkpoint_instance:%d\n", arg->msg.checkpoint_instance);
 					_starpu_mpi_irecv_cache_aware(handle, starpu_mpi_data_get_rank(handle), starpu_mpi_data_get_tag(handle), MPI_COMM_WORLD, 1, 0,
 												  NULL, NULL, 1, 0, 1, &arg->cache_flag);
 					// The callback needs to do nothing. The cached one must release the handle.

mpi/src/starpu_mpi.c

@@ -181,11 +181,11 @@ struct _starpu_mpi_req* _starpu_mpi_isend_cache_aware(starpu_data_handle_t data_
 	int already_sent = starpu_mpi_cached_send_set(data_handle, dest);
 	if (already_sent == 0)
 	{
+		*cache_flag = 0;
 		if (data_tag == -1)
 			_STARPU_ERROR("StarPU needs to be told the MPI tag of this data, using starpu_mpi_data_register\n");
 		_STARPU_MPI_DEBUG(1, "Send data %p to %d\n", data_handle, dest);
 		req = _starpu_mpi_isend_common(data_handle, dest, data_tag, comm, detached, sync, prio, callback, _arg, sequential_consistency);
-		*cache_flag = 0;
 	}
 	else
 	{