Merge pull request #7579 from devreal/progress-returns-v4.0.x

Harmonize return values of progress callbacks (v4.0.x)

ompi/communicator/comm_request.c

@@ -100,6 +100,7 @@ static int ompi_comm_request_progress (void)
 {
     ompi_comm_request_t *request, *next;
     static int32_t progressing = 0;
+    int completed = 0;
 
     /* don't allow re-entry */
     if (opal_atomic_swap_32 (&progressing, 1)) {
@@ -121,6 +122,7 @@ static int ompi_comm_request_progress (void)
                 if( REQUEST_COMPLETE(subreq) ) {
                     ompi_request_free (&subreq);
                     request_item->subreq_count--;
+                    completed++;
                 } else {
                     item_complete = false;
                     break;
@@ -156,7 +158,7 @@ static int ompi_comm_request_progress (void)
     opal_mutex_unlock (&ompi_comm_request_mutex);
     progressing = 0;
 
-    return 1;
+    return completed;
 }
 
 void ompi_comm_request_start (ompi_comm_request_t *request)

ompi/mca/coll/libnbc/coll_libnbc_component.c

@@ -307,6 +307,7 @@ ompi_coll_libnbc_progress(void)
 {
     ompi_coll_libnbc_request_t* request, *next;
     int res;
+    int completed = 0;
 
     if (0 == opal_list_get_size (&mca_coll_libnbc_component.active_requests)) {
         /* no requests -- nothing to do. do not grab a lock */
@@ -344,6 +345,7 @@ ompi_coll_libnbc_progress(void)
                 if(!request->super.super.req_persistent || !REQUEST_COMPLETE(&request->super.super)) {
             	    ompi_request_complete(&request->super.super, true);
                 }
+                completed++;
             }
             OPAL_THREAD_LOCK(&mca_coll_libnbc_component.lock);
         }
@@ -351,7 +353,7 @@ ompi_coll_libnbc_progress(void)
     }
     OPAL_THREAD_UNLOCK(&mca_coll_libnbc_component.lock);
 
-    return 0;
+    return completed;
 }
 
 

ompi/mca/mtl/psm2/mtl_psm2.c

@@ -403,7 +403,7 @@ int ompi_mtl_psm2_progress( void ) {
     mca_mtl_psm2_request_t* mtl_psm2_request;
     psm2_mq_status2_t psm2_status;
     psm2_mq_req_t req;
-    int completed = 1;
+    int completed = 0;
 
     do {
         OPAL_THREAD_LOCK(&mtl_psm2_mq_mutex);
@@ -469,5 +469,5 @@ int ompi_mtl_psm2_progress( void ) {
     opal_show_help("help-mtl-psm2.txt",
 		   "error polling network", true,
 		   psm2_error_get_string(err));
-    return 1;
+    return OMPI_ERROR;
 }

ompi/mca/osc/pt2pt/osc_pt2pt_component.c

@@ -153,6 +153,7 @@ static int component_register (void)
 
 static int component_progress (void)
 {
+    int completed = 0;
     int pending_count = opal_list_get_size (&mca_osc_pt2pt_component.pending_operations);
     int recv_count = opal_list_get_size (&mca_osc_pt2pt_component.pending_receives);
     ompi_osc_pt2pt_pending_t *pending, *next;
@@ -167,6 +168,7 @@ static int component_progress (void)
             }
 
             (void) ompi_osc_pt2pt_process_receive (recv);
+            completed++;
         }
     }
 
@@ -194,12 +196,13 @@ static int component_progress (void)
             if (OMPI_SUCCESS == ret) {
                 opal_list_remove_item (&mca_osc_pt2pt_component.pending_operations, &pending->super);
                 OBJ_RELEASE(pending);
+                completed++;
             }
         }
         OPAL_THREAD_UNLOCK(&mca_osc_pt2pt_component.pending_operations_lock);
     }
 
-    return 1;
+    return completed;
 }
 
 static int

ompi/mca/pml/ucx/pml_ucx.c

@@ -512,8 +512,7 @@ int mca_pml_ucx_enable(bool enable)
 
 int mca_pml_ucx_progress(void)
 {
-    ucp_worker_progress(ompi_pml_ucx.ucp_worker);
-    return OMPI_SUCCESS;
+    return ucp_worker_progress(ompi_pml_ucx.ucp_worker);
 }
 
 int mca_pml_ucx_add_comm(struct ompi_communicator_t* comm)

ompi/request/grequest.c

@@ -202,14 +202,13 @@ int ompi_grequest_invoke_query(ompi_request_t *request,
     int rc = OMPI_SUCCESS;
     ompi_grequest_t *g = (ompi_grequest_t*) request;
 
-    /* MPI-2:8.2 does not say what to do with the return value from
-       the query function (i.e., the int return value from the C
-       function or the ierr argument from the Fortran function).
-       Making the command decision here to ignore it.  If the handler
-       wants to pass an error back, it should set it in the MPI_ERROR
-       field in the status (which is always kept, regardless if the
-       top-level function was invoked with MPI_STATUS[ES]_IGNORE or
-       not).  */
+    /* MPI-3 mandates that the return value from the query function
+     * (i.e., the int return value from the C function or the ierr
+     * argument from the Fortran function) must be returned to the
+     * user. Thus, if the return of the query function is not MPI_SUCCESS
+     * we will update the MPI_ERROR field. Otherwise, the MPI_ERROR
+     * field is untouched (or left to the discretion of the query function).
+     */
     if (NULL != g->greq_query.c_query) {
         if (g->greq_funcs_are_c) {
             rc = g->greq_query.c_query(g->greq_state, status);
@@ -221,7 +220,9 @@ int ompi_grequest_invoke_query(ompi_request_t *request,
             rc = OMPI_FINT_2_INT(ierr);
         }
     }
-
+    if( MPI_SUCCESS != rc ) {
+        status->MPI_ERROR = rc;
+    }
     return rc;
 }
 

ompi/request/grequestx.c

@@ -34,6 +34,7 @@ static opal_mutex_t lock;
 
 static int grequestx_progress(void) {
     ompi_grequest_t *request, *next;
+    int completed = 0;
 
     OPAL_THREAD_LOCK(&lock);
     if (!in_progress) {
@@ -43,18 +44,17 @@ static int grequestx_progress(void) {
             MPI_Status status;
             OPAL_THREAD_UNLOCK(&lock);
             request->greq_poll.c_poll(request->greq_state, &status);
-            if (REQUEST_COMPLETE(&request->greq_base)) {
-                OPAL_THREAD_LOCK(&lock);
-                opal_list_remove_item(&requests, &request->greq_base.super.super);
-                OPAL_THREAD_UNLOCK(&lock);
-            }
             OPAL_THREAD_LOCK(&lock);
+            if (REQUEST_COMPLETE(&request->greq_base)) {
+                opal_list_remove_item(&requests, &request->greq_base.super.super);
+                completed++;
+            }
         }
         in_progress = false;
     }
     OPAL_THREAD_UNLOCK(&lock);
 
-    return OMPI_SUCCESS;
+    return completed;
 }
 
 int ompi_grequestx_start(

opal/mca/btl/uct/btl_uct_component.c

@@ -561,6 +561,7 @@ static int mca_btl_uct_tl_progress (mca_btl_uct_tl_t *tl, int starting_index)
 static int mca_btl_uct_component_progress_pending (mca_btl_uct_module_t *uct_btl)
 {
     mca_btl_uct_base_frag_t *frag, *next;
+    int completed = 0;
     size_t count;
 
     if (0 == (count = opal_list_get_size (&uct_btl->pending_frags))) {
@@ -577,11 +578,13 @@ static int mca_btl_uct_component_progress_pending (mca_btl_uct_module_t *uct_btl
 
         if (OPAL_SUCCESS > mca_btl_uct_send_frag (uct_btl, frag, false)) {
             opal_list_prepend (&uct_btl->pending_frags, (opal_list_item_t *) frag);
+        } else {
+            completed++;
         }
     }
     OPAL_THREAD_UNLOCK(&uct_btl->lock);
 
-    return OPAL_SUCCESS;
+    return completed;
 }
 
 /**

oshmem/mca/spml/ucx/spml_ucx_component.c

@@ -187,20 +187,21 @@ static int mca_spml_ucx_component_register(void)
 
 int spml_ucx_ctx_progress(void)
 {
-    int i;
+    int i, completed = 0;
     for (i = 0; i < mca_spml_ucx.active_array.ctxs_count; i++) {
-        ucp_worker_progress(mca_spml_ucx.active_array.ctxs[i]->ucp_worker[0]);
+        completed += ucp_worker_progress(mca_spml_ucx.active_array.ctxs[i]->ucp_worker[0]);
     }
-    return 1;
+    return completed;
 }
 
 int spml_ucx_default_progress(void)
 {
     unsigned int i=0;
+    int completed = 0;
     for (i = 0; i < mca_spml_ucx.ucp_workers; i++) {
-        ucp_worker_progress(mca_spml_ucx_ctx_default.ucp_worker[i]);
+        completed += ucp_worker_progress(mca_spml_ucx_ctx_default.ucp_worker[i]);
     }
-    return 1;
+    return completed;
 }
 
 int spml_ucx_progress_aux_ctx(void)