Merge pull request #1725 from hjelmn/request_fixes

ompi/request: fix bugs in MPI_Wait_some and MPI_Wait_any

ompi/request/req_wait.c

@@ -1,3 +1,4 @@
+/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
 /*
  * Copyright (c) 2004-2010 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
@@ -12,6 +13,8 @@
  * Copyright (c) 2006-2008 Cisco Systems, Inc.  All rights reserved.
  * Copyright (c) 2010-2012 Oracle and/or its affiliates.  All rights reserved.
  * Copyright (c) 2012      Oak Ridge National Labs.  All rights reserved.
+ * Copyright (c) 2016      Los Alamos National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -85,7 +88,7 @@ int ompi_request_default_wait_any(size_t count,
                                   int *index,
                                   ompi_status_public_t * status)
 {
-    size_t i = 0, completed = count, num_requests_null_inactive = 0;
+    size_t completed = count, num_requests_null_inactive = 0;
     int rc = OMPI_SUCCESS;
     ompi_request_t **rptr=NULL;
     ompi_request_t *request=NULL;
@@ -95,7 +98,7 @@ int ompi_request_default_wait_any(size_t count,
     
     rptr = requests;
     num_requests_null_inactive = 0;
-    for (i = 0; i < count; i++, rptr++) {
+    for (size_t i = 0; i < count; i++, rptr++) {
         request = *rptr;
 
         /* Sanity test */
@@ -111,6 +114,7 @@ int ompi_request_default_wait_any(size_t count,
             num_requests_null_inactive++;
             continue;
         }
+
         if(!OPAL_ATOMIC_CMPSET_PTR(&request->req_complete, REQUEST_PENDING, &sync)) {
             assert(REQUEST_COMPLETE(request));
             wait_sync_update( &sync, 1, request->req_status.MPI_ERROR);
@@ -118,11 +122,21 @@ int ompi_request_default_wait_any(size_t count,
             break;
         }
     }
+
+    if(num_requests_null_inactive == count) {
+        *index = MPI_UNDEFINED;
+        if (MPI_STATUS_IGNORE != status) {
+            *status = ompi_status_empty;
+        }
+        WAIT_SYNC_RELEASE(&sync);
+        return rc;
+    }
+
     SYNC_WAIT(&sync);
     
     /* recheck the complete status and clean up the sync primitives */
     rptr = requests;
-    for(i = 0; i < completed; i++, rptr++) {
+    for(size_t i = 0, pending_count = completed; i < pending_count ; i++, rptr++) {
         request = *rptr;
 
         if( request->req_state == OMPI_REQUEST_INACTIVE ) {
@@ -134,14 +148,14 @@ int ompi_request_default_wait_any(size_t count,
          * marked as REQUEST_COMPLETE.
          */
         OPAL_ATOMIC_CMPSET_PTR(&request->req_complete, &sync, REQUEST_PENDING);
+        if (REQUEST_COMPLETE(request)) {
+            completed = i;
+        }
     }
 
-    if(num_requests_null_inactive == count) {
-        *index = MPI_UNDEFINED;
-        if (MPI_STATUS_IGNORE != status) {
-            *status = ompi_status_empty;
-        }
-    } else {
+    rptr = requests + completed;
+    request = *rptr;
+
     assert( REQUEST_COMPLETE(request) );
     /* Per note above, we have to call gen request query_fn even
        if STATUS_IGNORE was provided */
@@ -166,12 +180,11 @@ int ompi_request_default_wait_any(size_t count,
         rc = ompi_request_free(rptr);
     }
     *index = completed;
-    }
 
 #if OPAL_ENABLE_FT_CR == 1
     if( opal_cr_is_enabled) {
         rptr = requests;
-        for (i = 0; i < count; i++, rptr++) {
+        for (size_t i = 0; i < count; i++, rptr++) {
             request = *rptr;
             if( REQUEST_COMPLETE(request) ) {
                 OMPI_CRCP_REQUEST_COMPLETE(request);
@@ -352,7 +365,7 @@ int ompi_request_default_wait_some(size_t count,
                                    int * indices,
                                    ompi_status_public_t * statuses)
 {
-    size_t i, num_requests_null_inactive=0, num_requests_done=0;
+    size_t num_requests_null_inactive=0, num_requests_done=0;
     int rc = MPI_SUCCESS;
     ompi_request_t **rptr = NULL;
     ompi_request_t *request = NULL;
@@ -369,7 +382,7 @@ int ompi_request_default_wait_some(size_t count,
     rptr = requests;
     num_requests_null_inactive = 0;
     num_requests_done = 0;
-    for (i = 0; i < count; i++, rptr++) {
+    for (size_t i = 0; i < count; i++, rptr++) {
         request = *rptr;
         /*
          * Check for null or completed persistent request.
@@ -385,6 +398,12 @@ int ompi_request_default_wait_some(size_t count,
             num_requests_done++;
         }
     }
+    if(num_requests_null_inactive == count) {
+        *outcount = MPI_UNDEFINED;
+        WAIT_SYNC_RELEASE(&sync);
+        return rc;
+    }
+
     if( 0 != num_requests_done ) {
         /* As we only expect one trigger update the sync with count 1 */
         wait_sync_update(&sync, 1, request->req_status.MPI_ERROR);
@@ -403,16 +422,12 @@ int ompi_request_default_wait_some(size_t count,
     }
 #endif
 
-    if(num_requests_null_inactive == count) {
-        *outcount = MPI_UNDEFINED;
-    } else {
-
     /* Do the final counting and */
     /* Clean up the synchronization primitives */
 
     rptr = requests;
     num_requests_done = 0;
-        for (i = 0; i < count; i++, rptr++) {
+    for (size_t i = 0; i < count; i++, rptr++) {
         request = *rptr;
 
         if( request->req_state == OMPI_REQUEST_INACTIVE ) {
@@ -430,7 +445,7 @@ int ompi_request_default_wait_some(size_t count,
 
     *outcount = num_requests_done;
 
-        for (i = 0; i < num_requests_done; i++) {
+    for (size_t i = 0; i < num_requests_done; i++) {
         request = requests[indices[i]];
         assert( REQUEST_COMPLETE(request) );
         /* Per note above, we have to call gen request query_fn even
@@ -459,6 +474,6 @@ int ompi_request_default_wait_some(size_t count,
             }
         }
     }
-    }
+
     return rc;
 }