Add Reduce SMP collective from MVAPICH2

[simgrid.git] / src / smpi / colls / reduce-ompi.c

diff --git a/src/smpi/colls/reduce-ompi.c b/src/smpi/colls/reduce-ompi.c
index c5d767c..e9e3803 100644 (file)
--- a/src/smpi/colls/reduce-ompi.c
+++ b/src/smpi/colls/reduce-ompi.c
@@ -1,3 +1,9 @@
+/* Copyright (c) 2013-2014. The SimGrid Team.
+ * All rights reserved.                                                     */
+
+/* This program is free software; you can redistribute it and/or modify it
+ * under the terms of the license (GNU LGPL) which comes with this package. */
+
 /*
  * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
@@ -9,16 +15,12 @@
  *                         University of Stuttgart.  All rights reserved.
  * Copyright (c) 2004-2005 The Regents of the University of California.
  *                         All rights reserved.
- * $COPYRIGHT$
  *
  * Additional copyrights may follow
- *
- * $HEADER$
  */
 
 #include "colls_private.h"
 #include "coll_tuned_topo.h"
-#define MCA_COLL_BASE_TAG_REDUCE 555
 
 
 
@@ -92,8 +94,8 @@ int smpi_coll_tuned_ompi_reduce_generic( void* sendbuf, void* recvbuf, int origi
            sendbuf to the accumbuf, in order to simplfy the loops */
         if (!smpi_op_is_commute(op)) {
             smpi_datatype_copy(
-                                                (char*)accumbuf, original_count, datatype,
-                                                (char*)sendtmpbuf, original_count, datatype);
+                                                (char*)sendtmpbuf, original_count, datatype,
+                                                (char*)accumbuf, original_count, datatype);
         }
         /* Allocate two buffers for incoming segments */
         real_segment_size = true_extent + (count_by_segment - 1) * extent;
@@ -151,7 +153,7 @@ int smpi_coll_tuned_ompi_reduce_generic( void* sendbuf, void* recvbuf, int origi
 
                     reqs[inbi]=smpi_mpi_irecv(local_recvbuf, recvcount, datatype,
                                              tree->tree_next[i], 
-                                             MCA_COLL_BASE_TAG_REDUCE, comm
+                                             COLL_TAG_REDUCE, comm
                                              );
                 }
                 /* wait for previous req to complete, if any.
@@ -195,7 +197,7 @@ int smpi_coll_tuned_ompi_reduce_generic( void* sendbuf, void* recvbuf, int origi
                         /* send combined/accumulated data to parent */
                         smpi_mpi_send( accumulator, prevcount, 
                                                   datatype, tree->tree_prev, 
-                                                  MCA_COLL_BASE_TAG_REDUCE,
+                                                  COLL_TAG_REDUCE,
                                                   comm);
                     }
 
@@ -242,7 +244,7 @@ int smpi_coll_tuned_ompi_reduce_generic( void* sendbuf, void* recvbuf, int origi
                                          segindex * segment_increment,
                                          count_by_segment, datatype,
                                          tree->tree_prev, 
-                                         MCA_COLL_BASE_TAG_REDUCE,
+                                         COLL_TAG_REDUCE,
                                          comm) ;
                 segindex++;
                 original_count -= count_by_segment;
@@ -270,7 +272,7 @@ int smpi_coll_tuned_ompi_reduce_generic( void* sendbuf, void* recvbuf, int origi
                                           segindex * segment_increment,
                                           count_by_segment, datatype,
                                           tree->tree_prev, 
-                                          MCA_COLL_BASE_TAG_REDUCE,
+                                          COLL_TAG_REDUCE,
                                           comm);
                 original_count -= count_by_segment;
             }
@@ -288,7 +290,7 @@ int smpi_coll_tuned_ompi_reduce_generic( void* sendbuf, void* recvbuf, int origi
                                           segindex * segment_increment, 
                                           count_by_segment, datatype, 
                                           tree->tree_prev, 
-                                          MCA_COLL_BASE_TAG_REDUCE, 
+                                          COLL_TAG_REDUCE,
                                           comm );
                 creq = (creq + 1) % max_outstanding_reqs;
                 segindex++;
@@ -303,6 +305,7 @@ int smpi_coll_tuned_ompi_reduce_generic( void* sendbuf, void* recvbuf, int origi
             free(sreq);
         }
     }
+    free(tree);
     return MPI_SUCCESS;
 
  error_hndl:  /* error handler */
@@ -521,8 +524,8 @@ int smpi_coll_tuned_reduce_ompi_in_order_binary( void *sendbuf, void *recvbuf,
                 return MPI_ERR_INTERN;
             }
             smpi_datatype_copy (
-                                                (char*)tmpbuf, count, datatype,
-                                                (char*)recvbuf, count, datatype);
+                                                (char*)recvbuf, count, datatype,
+                                                (char*)tmpbuf, count, datatype);
             use_this_sendbuf = tmpbuf;
         } else if (io_root == rank) {
             tmpbuf = (char *) malloc(text + (count - 1) * ext);
@@ -545,7 +548,7 @@ int smpi_coll_tuned_reduce_ompi_in_order_binary( void *sendbuf, void *recvbuf,
         if (root == rank) {
             /* Receive result from rank io_root to recvbuf */
             smpi_mpi_recv(recvbuf, count, datatype, io_root,
-                                    MCA_COLL_BASE_TAG_REDUCE, comm,
+                                    COLL_TAG_REDUCE, comm,
                                     MPI_STATUS_IGNORE);
             if (MPI_IN_PLACE == sendbuf) {
                 free(use_this_sendbuf);
@@ -554,7 +557,7 @@ int smpi_coll_tuned_reduce_ompi_in_order_binary( void *sendbuf, void *recvbuf,
         } else if (io_root == rank) {
             /* Send result from use_this_recvbuf to root */
             smpi_mpi_send(use_this_recvbuf, count, datatype, root,
-                                    MCA_COLL_BASE_TAG_REDUCE, 
+                                    COLL_TAG_REDUCE,
                                     comm);
             free(use_this_recvbuf);
         }
@@ -610,9 +613,9 @@ smpi_coll_tuned_reduce_ompi_basic_linear(void *sbuf, void *rbuf, int count,
 
     if (rank != root) {
         smpi_mpi_send(sbuf, count, dtype, root,
-                                MCA_COLL_BASE_TAG_REDUCE,
+                                COLL_TAG_REDUCE,
                                 comm);
-        return -1;
+        return MPI_SUCCESS;
     }
 
     /* see discussion in ompi_coll_basic_reduce_lin_intra about 
@@ -639,11 +642,10 @@ smpi_coll_tuned_reduce_ompi_basic_linear(void *sbuf, void *rbuf, int count,
     /* Initialize the receive buffer. */
 
     if (rank == (size - 1)) {
-        smpi_datatype_copy((char*)rbuf, count, dtype,
-                                                  (char*)sbuf, count, dtype);
+        smpi_datatype_copy((char*)sbuf, count, dtype,(char*)rbuf, count, dtype);
     } else {
         smpi_mpi_recv(rbuf, count, dtype, size - 1,
-                                MCA_COLL_BASE_TAG_REDUCE, comm,
+                                COLL_TAG_REDUCE, comm,
                                 MPI_STATUS_IGNORE);
     }
 
@@ -654,7 +656,7 @@ smpi_coll_tuned_reduce_ompi_basic_linear(void *sbuf, void *rbuf, int count,
             inbuf = (char*)sbuf;
         } else {
             smpi_mpi_recv(pml_buffer, count, dtype, i,
-                                    MCA_COLL_BASE_TAG_REDUCE, comm,
+                                    COLL_TAG_REDUCE, comm,
                                     MPI_STATUS_IGNORE);
             inbuf = pml_buffer;
         }
@@ -664,8 +666,8 @@ smpi_coll_tuned_reduce_ompi_basic_linear(void *sbuf, void *rbuf, int count,
     }
 
     if (NULL != inplace_temp) {
-        smpi_datatype_copy((char*)sbuf, count, dtype,
-                                                  inplace_temp,count , dtype);
+        smpi_datatype_copy(inplace_temp, count, dtype,(char*)sbuf
+                                                  ,count , dtype);
         free(inplace_temp);
     }
     if (NULL != free_buffer) {