some of the action checking code didn't work right, so I removed it.

[simgrid.git] / src / smpi / smpi_sender.c

diff --git a/src/smpi/smpi_sender.c b/src/smpi/smpi_sender.c
index 41ff7a1..26f5125 100644 (file)
--- a/src/smpi/smpi_sender.c
+++ b/src/smpi/smpi_sender.c
@@ -1,14 +1,16 @@
 #include "private.h"
 
+XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_sender, smpi, "Logging specific to SMPI (sender)");
+
 int smpi_sender(int argc, char **argv)
 {
        smx_process_t self;
        smx_host_t shost;
-       int rank;
+
+       int index;
 
        xbt_fifo_t request_queue;
        smx_mutex_t request_queue_mutex;
-       int size;
 
        int running_hosts_count;
 
@@ -16,12 +18,13 @@ int smpi_sender(int argc, char **argv)
 
        smx_host_t dhost;
 
-       char communicate[] = "communicate";
        smx_action_t action;
 
+       e_surf_action_state_t state;
+
        smpi_received_message_t message;
 
-       int drank;
+       int dindex;
 
        smx_process_t receiver_process;
 
@@ -30,27 +33,27 @@ int smpi_sender(int argc, char **argv)
 
        // make sure root is done before own initialization
        SIMIX_mutex_lock(smpi_global->start_stop_mutex);
-       if (!smpi_global->root_ready) {
+       while (!smpi_global->root_ready) {
                SIMIX_cond_wait(smpi_global->start_stop_cond, smpi_global->start_stop_mutex);
        }
        SIMIX_mutex_unlock(smpi_global->start_stop_mutex);
 
-       rank = smpi_mpi_comm_rank(smpi_mpi_global->mpi_comm_world, shost);
-       size = smpi_mpi_comm_size(smpi_mpi_global->mpi_comm_world);
+       index = smpi_host_index();
 
-       request_queue       = smpi_global->pending_send_request_queues[rank];
-       request_queue_mutex = smpi_global->pending_send_request_queues_mutexes[rank];
+       request_queue       = smpi_global->pending_send_request_queues[index];
+       request_queue_mutex = smpi_global->pending_send_request_queues_mutexes[index];
 
-       smpi_global->sender_processes[rank] = self;
+       smpi_global->sender_processes[index] = self;
 
        // wait for all nodes to signal initializatin complete
        SIMIX_mutex_lock(smpi_global->start_stop_mutex);
        smpi_global->ready_process_count++;
-       if (smpi_global->ready_process_count < 3 * size) {
-               SIMIX_cond_wait(smpi_global->start_stop_cond, smpi_global->start_stop_mutex);
-       } else {
+       if (smpi_global->ready_process_count >= 3 * smpi_global->host_count) {
                SIMIX_cond_broadcast(smpi_global->start_stop_cond);
        }
+       while (smpi_global->ready_process_count < 3 * smpi_global->host_count) {
+               SIMIX_cond_wait(smpi_global->start_stop_cond, smpi_global->start_stop_mutex);
+       }
        SIMIX_mutex_unlock(smpi_global->start_stop_mutex);
 
        do {
@@ -63,38 +66,56 @@ int smpi_sender(int argc, char **argv)
                        SIMIX_process_suspend(self);
                } else {
 
-                       message       = xbt_mallocator_get(smpi_global->message_mallocator);
+                       message          = xbt_mallocator_get(smpi_global->message_mallocator);
 
                        SIMIX_mutex_lock(request->mutex);
 
-                       message->comm = request->comm;
-                       message->src  = request->src;
-                       message->dst  = request->dst;
-                       message->tag  = request->tag;
-                       message->buf  = xbt_malloc(request->datatype->size * request->count);
+                       message->comm    = request->comm;
+                       message->src     = request->comm->index_to_rank_map[index];
+                       message->tag     = request->tag;
+                       message->data    = request->data;
+                       message->buf     = xbt_malloc(request->datatype->size * request->count);
                        memcpy(message->buf, request->buf, request->datatype->size * request->count);
 
-                       dhost = request->comm->hosts[request->dst];
-                       drank = smpi_mpi_comm_rank(smpi_mpi_global->mpi_comm_world, dhost);
+                       dindex = request->comm->rank_to_index_map[request->dst];
+                       dhost  = smpi_global->hosts[dindex];
 
-                       SIMIX_mutex_lock(smpi_global->received_message_queues_mutexes[drank]);
-                       xbt_fifo_push(smpi_global->received_message_queues[drank], message);
-                       SIMIX_mutex_unlock(smpi_global->received_message_queues_mutexes[drank]);
+                       message->forward = (request->forward - 1) / 2;
+                       request->forward = request->forward / 2;
 
-                       request->completed = 1;
+                       if (0 < request->forward) {
+                               request->dst = (request->dst + message->forward + 1) % request->comm->size;
+                               SIMIX_mutex_lock(request_queue_mutex);
+                               xbt_fifo_push(request_queue, request);
+                               SIMIX_mutex_unlock(request_queue_mutex);
+                       } else {
+                               request->completed = 1;
+                       }
 
-                       action = SIMIX_action_communicate(shost, dhost, communicate, request->datatype->size * request->count, -1.0);
+                       action = SIMIX_action_communicate(shost, dhost, "communication", request->datatype->size * request->count, -1.0);
 
                        SIMIX_register_action_to_condition(action, request->cond);
-                       SIMIX_cond_wait(request->cond, request->mutex);
-                       SIMIX_unregister_action_to_condition(action, request->cond);
 
-                       SIMIX_mutex_unlock(request->mutex);
+                       for (
+                               state  = SIMIX_action_get_state(action);
+                               state == SURF_ACTION_READY ||
+                               state == SURF_ACTION_RUNNING;
+                               state  = SIMIX_action_get_state(action)
+                       ) {
+                               SIMIX_cond_wait(request->cond, request->mutex);
+                       }
+
+                       SIMIX_mutex_lock(smpi_global->received_message_queues_mutexes[dindex]);
+                       xbt_fifo_push(smpi_global->received_message_queues[dindex], message);
+                       SIMIX_mutex_unlock(smpi_global->received_message_queues_mutexes[dindex]);
 
+                       SIMIX_unregister_action_to_condition(action, request->cond);
                        SIMIX_action_destroy(action);
 
+                       SIMIX_mutex_unlock(request->mutex);
+
                        // wake up receiver if necessary
-                       receiver_process = smpi_global->receiver_processes[drank];
+                       receiver_process = smpi_global->receiver_processes[dindex];
                        if (SIMIX_process_is_suspended(receiver_process)) {
                                SIMIX_process_resume(receiver_process);
                        }