Renamed all of the sw_event code to spc for consistency. Fixed a bug where multithreaded runs would fail because the spc initialization happened before opal initialization. Made a better way to keep track of timer-based counters. Made the output in MPI_Finalize make more sense in cases where some counters are turned off. Added SPC_FINI() to MPI_Finalize and made some coding style/formatting changes.

Author: David Eberius

configure.ac

@@ -292,15 +292,15 @@ AC_ARG_ENABLE(spc,
                    [Enable software-based performance counters capability (default: disabled)]))
 if test "$enable_spc" = "yes"; then
     AC_MSG_RESULT([yes])
-    SOFTWARE_EVENTS_ENABLE=1
+    SPC_ENABLE=1
 else
     AC_MSG_RESULT([no])
-    SOFTWARE_EVENTS_ENABLE=0
+    SPC_ENABLE=0
 fi
-AC_DEFINE_UNQUOTED([SOFTWARE_EVENTS_ENABLE],
-                   [$SOFTWARE_EVENTS_ENABLE],
+AC_DEFINE_UNQUOTED([SPC_ENABLE],
+                   [$SPC_ENABLE],
                    [If the software-based performance counters capability should be enabled.])
-AM_CONDITIONAL(SOFTWARE_EVENTS_ENABLE, test "$SOFTWARE_EVENTS_ENABLE" = "1")
+AM_CONDITIONAL(SPC_ENABLE, test "$SPC_ENABLE" = "1")
 
 AS_IF([test "$enable_spc" != "no"], [project_spc_amc=true], [project_spc_amc=false])
 

ompi/mca/pml/ob1/pml_ob1.c

@@ -36,7 +36,7 @@
 #include "opal_stdint.h"
 #include "opal/mca/btl/btl.h"
 #include "opal/mca/btl/base/base.h"
-#include "ompi/runtime/ompi_software_events.h"
+#include "ompi/runtime/ompi_spc.h"
 
 #include "ompi/mca/pml/pml.h"
 #include "ompi/mca/pml/base/base.h"
@@ -196,7 +196,9 @@ int mca_pml_ob1_add_comm(ompi_communicator_t* comm)
     mca_pml_ob1_recv_frag_t *frag, *next_frag;
     mca_pml_ob1_comm_proc_t* pml_proc;
     mca_pml_ob1_match_hdr_t* hdr;
-    opal_timer_t usecs = 0;
+#if SPC_ENABLE == 1
+    opal_timer_t timer = 0;
+#endif
 
     if (NULL == pml_comm) {
         return OMPI_ERR_OUT_OF_RESOURCE;
@@ -266,7 +268,7 @@ int mca_pml_ob1_add_comm(ompi_communicator_t* comm)
              * situation as the cant_match is only checked when a new fragment is received from
              * the network.
              */
-            SW_EVENT_TIMER_START(OMPI_OOS_MATCH_TIME, &usecs);
+            SPC_TIMER_START(OMPI_OOS_MATCH_TIME, &timer);
             OPAL_LIST_FOREACH(frag, &pml_proc->frags_cant_match, mca_pml_ob1_recv_frag_t) {
                 hdr = &frag->hdr.hdr_match;
                 /* If the message has the next expected seq from that proc...  */
@@ -276,7 +278,7 @@ int mca_pml_ob1_add_comm(ompi_communicator_t* comm)
                 opal_list_remove_item(&pml_proc->frags_cant_match, (opal_list_item_t*)frag);
                 goto add_fragment_to_unexpected;
             }
-            SW_EVENT_TIMER_STOP(OMPI_OOS_MATCH_TIME, &usecs);
+            SPC_TIMER_STOP(OMPI_OOS_MATCH_TIME, &timer);
         } else {
             opal_list_append( &pml_proc->frags_cant_match, (opal_list_item_t*)frag );
         }

ompi/mca/pml/ob1/pml_ob1_isend.c

@@ -28,7 +28,7 @@
 #include "pml_ob1_sendreq.h"
 #include "pml_ob1_recvreq.h"
 #include "ompi/peruse/peruse-internal.h"
-#include "ompi/runtime/ompi_software_events.h"
+#include "ompi/runtime/ompi_spc.h"
 
 /**
  * Single usage request. As we allow recursive calls (as an
@@ -121,13 +121,8 @@ static inline int mca_pml_ob1_send_inline (const void *buf, size_t count,
                              size, MCA_BTL_NO_ORDER, MCA_BTL_DES_FLAGS_PRIORITY | MCA_BTL_DES_FLAGS_BTL_OWNERSHIP,
                              MCA_PML_OB1_HDR_TYPE_MATCH, NULL);
 
-    if(OPAL_LIKELY(rc == OPAL_SUCCESS)){
-        if(tag >= 0){
-            SW_EVENT_RECORD(OMPI_BYTES_SENT_USER, size);
-        }
-        else{
-            SW_EVENT_RECORD(OMPI_BYTES_SENT_MPI, size);
-        }
+    if(OPAL_LIKELY(rc == OPAL_SUCCESS)) {
+        SPC_USER_OR_MPI(tag, (long long)size, OMPI_BYTES_SENT_USER, OMPI_BYTES_SENT_MPI);
     }
 
     if (count > 0) {

ompi/mca/pml/ob1/pml_ob1_recvfrag.c

@@ -39,7 +39,7 @@
 #include "ompi/mca/pml/pml.h"
 #include "ompi/peruse/peruse-internal.h"
 #include "ompi/memchecker.h"
-#include "ompi/runtime/ompi_software_events.h"
+#include "ompi/runtime/ompi_spc.h"
 
 #include "pml_ob1.h"
 #include "pml_ob1_comm.h"
@@ -180,20 +180,19 @@ match_one(mca_btl_base_module_t *btl,
 static inline mca_pml_ob1_recv_frag_t* check_cantmatch_for_match(mca_pml_ob1_comm_proc_t *proc)
 {
     mca_pml_ob1_recv_frag_t *frag = NULL;
-#if SOFTWARE_EVENTS_ENABLE == 1
+#if SPC_ENABLE == 1
     opal_timer_t timer = 0;
 #endif
-
-    SW_EVENT_TIMER_START(OMPI_OOS_MATCH_TIME, &timer);
+    SPC_TIMER_START(OMPI_OOS_MATCH_TIME, &timer);
 
     frag = (mca_pml_ob1_recv_frag_t*)opal_list_get_first(&proc->frags_cant_match);
     if( (opal_list_get_end(&proc->frags_cant_match) != (opal_list_item_t*)frag) &&
         (frag->hdr.hdr_match.hdr_seq == proc->expected_sequence) ) {
         opal_list_remove_item(&proc->frags_cant_match, (opal_list_item_t*)frag);
-        SW_EVENT_TIMER_STOP(OMPI_OOS_MATCH_TIME, &timer);
+        SPC_TIMER_STOP(OMPI_OOS_MATCH_TIME, &timer);
         return frag;
     }
-    SW_EVENT_TIMER_STOP(OMPI_OOS_MATCH_TIME, &timer);
+    SPC_TIMER_STOP(OMPI_OOS_MATCH_TIME, &timer);
     return NULL;
 }
 
@@ -263,7 +262,7 @@ void mca_pml_ob1_recv_frag_callback_match(mca_btl_base_module_t* btl,
         if(OPAL_UNLIKELY(((uint16_t) hdr->hdr_seq) != ((uint16_t) proc->expected_sequence))) {
             append_frag_to_ordered_list(&proc->frags_cant_match, btl,
                                         hdr, segments, num_segments, NULL);
-            SW_EVENT_RECORD(OMPI_OUT_OF_SEQUENCE, 1);
+            SPC_RECORD(OMPI_OUT_OF_SEQUENCE, 1);
             OB1_MATCHING_UNLOCK(&comm->matching_lock);
             return;
         }
@@ -324,9 +323,8 @@ void mca_pml_ob1_recv_frag_callback_match(mca_btl_base_module_t* btl,
                                    &iov_count,
                                    &bytes_received );
             match->req_bytes_received = bytes_received;
-
-            SW_EVENT_USER_OR_MPI(match->req_recv.req_base.req_tag, (long long)bytes_received, OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
-
+            SPC_USER_OR_MPI(match->req_recv.req_base.req_tag, (long long)bytes_received,
+                            OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
             /*
              *  Unpacking finished, make the user buffer unaccessable again.
              */
@@ -651,10 +649,10 @@ match_one(mca_btl_base_module_t *btl,
           mca_pml_ob1_comm_proc_t *proc,
           mca_pml_ob1_recv_frag_t* frag)
 {
-#if SOFTWARE_EVENTS_ENABLE == 1
+#if SPC_ENABLE == 1
     opal_timer_t timer = 0;
 #endif
-    SW_EVENT_TIMER_START(OMPI_MATCH_TIME, &timer);
+    SPC_TIMER_START(OMPI_MATCH_TIME, &timer);
 
     mca_pml_ob1_recv_request_t *match;
     mca_pml_ob1_comm_t *comm = (mca_pml_ob1_comm_t *)comm_ptr->c_pml_comm;
@@ -693,28 +691,23 @@ match_one(mca_btl_base_module_t *btl,
                                                        num_segments);
                 /* this frag is already processed, so we want to break out
                    of the loop and not end up back on the unexpected queue. */
-                SW_EVENT_TIMER_STOP(OMPI_MATCH_TIME, &timer);
-
+                SPC_TIMER_STOP(OMPI_MATCH_TIME, &timer);
                 return NULL;
             }
 
             PERUSE_TRACE_COMM_EVENT(PERUSE_COMM_MSG_MATCH_POSTED_REQ,
                                     &(match->req_recv.req_base), PERUSE_RECV);
-            SW_EVENT_TIMER_STOP(OMPI_MATCH_TIME, &timer);
-
+            SPC_TIMER_STOP(OMPI_MATCH_TIME, &timer);
             return match;
         }
 
         /* if no match found, place on unexpected queue */
         append_frag_to_list(&proc->unexpected_frags, btl, hdr, segments,
                             num_segments, frag);
-
-        SW_EVENT_RECORD(OMPI_UNEXPECTED, 1);
-
+        SPC_RECORD(OMPI_UNEXPECTED, 1);
         PERUSE_TRACE_MSG_EVENT(PERUSE_COMM_MSG_INSERT_IN_UNEX_Q, comm_ptr,
                                hdr->hdr_src, hdr->hdr_tag, PERUSE_RECV);
-        SW_EVENT_TIMER_STOP(OMPI_MATCH_TIME, &timer);
-
+        SPC_TIMER_STOP(OMPI_MATCH_TIME, &timer);
         return NULL;
     } while(true);
 }

ompi/mca/pml/ob1/pml_ob1_recvreq.c

@@ -29,7 +29,7 @@
 
 #include "opal/mca/mpool/mpool.h"
 #include "opal/util/arch.h"
-#include "ompi/runtime/ompi_software_events.h"
+#include "ompi/runtime/ompi_spc.h"
 #include "ompi/mca/pml/pml.h"
 #include "ompi/mca/bml/bml.h"
 #include "pml_ob1_comm.h"
@@ -199,10 +199,9 @@ static void mca_pml_ob1_put_completion (mca_pml_ob1_rdma_frag_t *frag, int64_t r
     if (OPAL_LIKELY(0 < rdma_size)) {
 
         /* check completion status */
-
         OPAL_THREAD_ADD_FETCH_SIZE_T(&recvreq->req_bytes_received, rdma_size);
-        SW_EVENT_USER_OR_MPI(recvreq->req_recv.req_base.req_tag, (long long)rdma_size, OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
-
+        SPC_USER_OR_MPI(recvreq->req_recv.req_base.req_tag, (long long)rdma_size,
+                        OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
         if (recv_request_pml_complete_check(recvreq) == false &&
             recvreq->req_rdma_offset < recvreq->req_send_offset) {
             /* schedule additional rdma operations */
@@ -246,8 +245,7 @@ int mca_pml_ob1_recv_request_ack_send_btl(
     des->des_cbfunc = mca_pml_ob1_recv_ctl_completion;
 
     rc = mca_bml_base_send(bml_btl, des, MCA_PML_OB1_HDR_TYPE_ACK);
-    SW_EVENT_RECORD(OMPI_BYTES_RECEIVED_MPI, (long long)size);
-
+    SPC_RECORD(OMPI_BYTES_RECEIVED_MPI, (long long)size);
     if( OPAL_LIKELY( rc >= 0 ) ) {
         return OMPI_SUCCESS;
     }
@@ -380,7 +378,8 @@ static void mca_pml_ob1_rget_completion (mca_btl_base_module_t* btl, struct mca_
     } else {
         /* is receive request complete */
         OPAL_THREAD_ADD_FETCH_SIZE_T(&recvreq->req_bytes_received, frag->rdma_length);
-        SW_EVENT_USER_OR_MPI(recvreq->req_recv.req_base.req_tag, (long long)frag->rdma_length, OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
+        SPC_USER_OR_MPI(recvreq->req_recv.req_base.req_tag, (long long)frag->rdma_length,
+                        OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
         /* TODO: re-add order */
         mca_pml_ob1_send_fin (recvreq->req_recv.req_base.req_proc,
                               bml_btl, frag->rdma_hdr.hdr_rget.hdr_frag,
@@ -436,10 +435,8 @@ static int mca_pml_ob1_recv_request_put_frag (mca_pml_ob1_rdma_frag_t *frag)
 
     /* send rdma request to peer */
     rc = mca_bml_base_send (bml_btl, ctl, MCA_PML_OB1_HDR_TYPE_PUT);
-
     /* Increment counter for bytes_put even though they probably haven't all been received yet */
-    SW_EVENT_RECORD(OMPI_BYTES_PUT, frag->rdma_length);
-
+    SPC_RECORD(OMPI_BYTES_PUT, frag->rdma_length);
     if (OPAL_UNLIKELY(rc < 0)) {
         mca_bml_base_free (bml_btl, ctl);
         return rc;
@@ -481,10 +478,8 @@ int mca_pml_ob1_recv_request_get_frag (mca_pml_ob1_rdma_frag_t *frag)
     rc = mca_bml_base_get (bml_btl, frag->local_address, frag->remote_address, local_handle,
                            (mca_btl_base_registration_handle_t *) frag->remote_handle, frag->rdma_length,
                            0, MCA_BTL_NO_ORDER, mca_pml_ob1_rget_completion, frag);
-
     /* Increment counter for bytes_get even though they probably haven't all been received yet */
-    SW_EVENT_RECORD(OMPI_BYTES_GET, frag->rdma_length);
-
+    SPC_RECORD(OMPI_BYTES_GET, frag->rdma_length);
     if( OPAL_UNLIKELY(OMPI_SUCCESS > rc) ) {
         return mca_pml_ob1_recv_request_get_frag_failed (frag, OMPI_ERR_OUT_OF_RESOURCE);
     }
@@ -540,8 +535,8 @@ void mca_pml_ob1_recv_request_progress_frag( mca_pml_ob1_recv_request_t* recvreq
                );
 
     OPAL_THREAD_ADD_FETCH_SIZE_T(&recvreq->req_bytes_received, bytes_received);
-    SW_EVENT_USER_OR_MPI(recvreq->req_recv.req_base.req_tag, (long long)bytes_received, OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
-
+    SPC_USER_OR_MPI(recvreq->req_recv.req_base.req_tag, (long long)bytes_received,
+                    OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
     /* check completion status */
     if(recv_request_pml_complete_check(recvreq) == false &&
             recvreq->req_rdma_offset < recvreq->req_send_offset) {
@@ -619,8 +614,8 @@ void mca_pml_ob1_recv_request_frag_copy_finished( mca_btl_base_module_t* btl,
     des->des_cbfunc(NULL, NULL, des, 0);
 
     OPAL_THREAD_ADD_FETCH_SIZE_T(&recvreq->req_bytes_received, bytes_received);
-    SW_EVENT_USER_OR_MPI(recvreq->req_recv.req_base.req_tag, (long long)bytes_received, OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
-
+    SPC_USER_OR_MPI(recvreq->req_recv.req_base.req_tag, (long long)bytes_received,
+                    OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
     /* check completion status */
     if(recv_request_pml_complete_check(recvreq) == false &&
             recvreq->req_rdma_offset < recvreq->req_send_offset) {
@@ -834,7 +829,8 @@ void mca_pml_ob1_recv_request_progress_rndv( mca_pml_ob1_recv_request_t* recvreq
                                    recvreq->req_recv.req_base.req_datatype);
                    );
         OPAL_THREAD_ADD_FETCH_SIZE_T(&recvreq->req_bytes_received, bytes_received);
-        SW_EVENT_USER_OR_MPI(recvreq->req_recv.req_base.req_tag, (long long)bytes_received, OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
+        SPC_USER_OR_MPI(recvreq->req_recv.req_base.req_tag, (long long)bytes_received,
+                        OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
     }
     /* check completion status */
     if(recv_request_pml_complete_check(recvreq) == false &&
@@ -905,9 +901,8 @@ void mca_pml_ob1_recv_request_progress_match( mca_pml_ob1_recv_request_t* recvre
      * for this request.
      */
     recvreq->req_bytes_received += bytes_received;
-
-    SW_EVENT_USER_OR_MPI(recvreq->req_recv.req_base.req_tag, (long long)bytes_received, OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
-
+    SPC_USER_OR_MPI(recvreq->req_recv.req_base.req_tag, (long long)bytes_received,
+                    OMPI_BYTES_RECEIVED_USER, OMPI_BYTES_RECEIVED_MPI);
     recv_request_pml_complete(recvreq);
 }