need state to ack sync send retransmits, even after the recvreq is gone..

This commit was SVN r9397.

ompi/mca/pml/dr/pml_dr_comm.c

@@ -34,6 +34,7 @@ static void mca_pml_dr_comm_proc_construct(mca_pml_dr_comm_proc_t* proc)
     OBJ_CONSTRUCT(&proc->unexpected_frags, opal_list_t);
     OBJ_CONSTRUCT(&proc->seq_sends, ompi_seq_tracker_t);
     OBJ_CONSTRUCT(&proc->seq_recvs, ompi_seq_tracker_t);
+    OBJ_CONSTRUCT(&proc->seq_recvs_matched, ompi_seq_tracker_t);
 }
 
 
@@ -45,6 +46,7 @@ static void mca_pml_dr_comm_proc_destruct(mca_pml_dr_comm_proc_t* proc)
     OBJ_DESTRUCT(&proc->unexpected_frags);
     OBJ_DESTRUCT(&proc->seq_sends);
     OBJ_DESTRUCT(&proc->seq_recvs);
+    OBJ_DESTRUCT(&proc->seq_recvs_matched);
 }
 
 

ompi/mca/pml/dr/pml_dr_comm.h

@@ -48,6 +48,7 @@ struct mca_pml_dr_comm_proc_t {
     ompi_proc_t* ompi_proc;        /**< back pointer to ompi_proc_t */
     ompi_seq_tracker_t seq_sends; /**< Tracks the send vfrags that have been acked */ 
     ompi_seq_tracker_t seq_recvs; /**< Tracks the receive vfrags that have been acked */ 
+    ompi_seq_tracker_t seq_recvs_matched; /**< Tracks the received vfrags that have been matched */ 
 };
 typedef struct mca_pml_dr_comm_proc_t mca_pml_dr_comm_proc_t;
 

ompi/mca/pml/dr/pml_dr_recvfrag.c

@@ -157,17 +157,24 @@ void mca_pml_dr_recv_frag_callback(
 
             /* seq_recvs protected by matching lock */
             OPAL_THREAD_LOCK(&comm->matching_lock);
-            if(ompi_seq_tracker_check_duplicate(&proc->seq_recvs, hdr->hdr_common.hdr_vid)) {
+            if(ompi_seq_tracker_check_duplicate(&proc->seq_recvs, hdr->hdr_common.hdr_vid)){ 
                 /* ack only if the vfrag has been matched */
                 mca_pml_dr_recv_request_t* recvreq = 
                     mca_pml_dr_comm_proc_check_matched(proc, hdr->hdr_common.hdr_vid);
                 OPAL_THREAD_UNLOCK(&comm->matching_lock);
                 if(NULL != recvreq) {
-                    OPAL_OUTPUT((0, "%s:%d: acking duplicate matched rendezvous\n", __FILE__, __LINE__));
+                    OPAL_OUTPUT((0, "%s:%d: acking duplicate matched rendezvous from pending matched\n", __FILE__, __LINE__));
                     mca_pml_dr_recv_request_ack(recvreq, &hdr->hdr_common, 
                         hdr->hdr_match.hdr_src_ptr, recvreq->req_bytes_received, 1);
                 } else { 
-                    OPAL_OUTPUT((0, "%s:%d: droping duplicate unmatched rendezvous\n", __FILE__, __LINE__));
+                    if(ompi_seq_tracker_check_duplicate(&proc->seq_recvs_matched, hdr->hdr_common.hdr_vid)) {
+                        OPAL_OUTPUT((0, "%s:%d: acking duplicate matched rendezvous from sequence tracker\n", __FILE__, __LINE__));
+                        mca_pml_dr_recv_request_ack(recvreq, &hdr->hdr_common, 
+                                                    hdr->hdr_match.hdr_src_ptr, recvreq->req_bytes_received, 1);
+                        
+                    } else { 
+                        OPAL_OUTPUT((0, "%s:%d: droping duplicate unmatched rendezvous\n", __FILE__, __LINE__));
+                    }
                 }
             } else {
                 OPAL_THREAD_UNLOCK(&comm->matching_lock);

ompi/mca/pml/dr/pml_dr_recvreq.h

@@ -225,6 +225,7 @@ do {
     (request)->req_recv.req_base.req_ompi.req_status.MPI_SOURCE =                    \
                                                        (hdr)->hdr_common.hdr_src;    \
     opal_list_append(&proc->matched_receives, (opal_list_item_t*)request);           \
+    ompi_seq_tracker_insert(&proc->seq_recvs_matched, (hdr)->hdr_common.hdr_vid);    \
 } while(0)