X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/e7c0c67af63b3979a597a66e5e1c8b0435fc6e19..1687df79d61a9418bba830bbd0ab7de16e457090:/src/smpi/colls/reduce-scatter-gather.c

diff --git a/src/smpi/colls/reduce-scatter-gather.c b/src/smpi/colls/reduce-scatter-gather.c
index 4dbcbf1327..38db76025b 100644
--- a/src/smpi/colls/reduce-scatter-gather.c
+++ b/src/smpi/colls/reduce-scatter-gather.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. */
+
 #include "colls_private.h"
 
 /*
@@ -10,12 +16,12 @@ int smpi_coll_tuned_reduce_scatter_gather(void *sendbuf, void *recvbuf,
                                           MPI_Op op, int root, MPI_Comm comm)
 {
   MPI_Status status;
-  int comm_size, rank, type_size, pof2, rem, newrank;
+  int comm_size, rank, pof2, rem, newrank;
   int mask, *cnts, *disps, i, j, send_idx = 0;
   int recv_idx, last_idx = 0, newdst;
   int dst, send_cnt, recv_cnt, newroot, newdst_tree_root;
   int newroot_tree_root, new_count;
-  int tag = 4321;
+  int tag = COLL_TAG_REDUCE,temporary_buffer=0;
   void *send_ptr, *recv_ptr, *tmp_buf;
 
   cnts = NULL;
@@ -27,10 +33,16 @@ int smpi_coll_tuned_reduce_scatter_gather(void *sendbuf, void *recvbuf,
     return 0;
   rank = smpi_comm_rank(comm);
   comm_size = smpi_comm_size(comm);
+  
 
-  extent = smpi_datatype_get_extent(datatype);
-  type_size = smpi_datatype_size(datatype);
 
+  extent = smpi_datatype_get_extent(datatype);
+  /* If I'm not the root, then my recvbuf may not be valid, therefore
+  I have to allocate a temporary one */
+  if (rank != root && !recvbuf) {
+    temporary_buffer=1;
+    recvbuf = (void *)smpi_get_tmp_recvbuffer(count * extent);
+  }
   /* find nearest power-of-two less than or equal to comm_size */
   pof2 = 1;
   while (pof2 <= comm_size)
@@ -39,10 +51,10 @@ int smpi_coll_tuned_reduce_scatter_gather(void *sendbuf, void *recvbuf,
 
   if (count < comm_size) {
     new_count = comm_size;
-    send_ptr = (void *) xbt_malloc(new_count * extent);
-    recv_ptr = (void *) xbt_malloc(new_count * extent);
-    tmp_buf = (void *) xbt_malloc(new_count * extent);
-    memcpy(send_ptr, sendbuf, extent * new_count);
+    send_ptr = (void *) smpi_get_tmp_sendbuffer(new_count * extent);
+    recv_ptr = (void *) smpi_get_tmp_recvbuffer(new_count * extent);
+    tmp_buf = (void *) smpi_get_tmp_sendbuffer(new_count * extent);
+    memcpy(send_ptr, sendbuf != MPI_IN_PLACE ? sendbuf : recvbuf, extent * count);
 
     //if ((rank != root))
     smpi_mpi_sendrecv(send_ptr, new_count, datatype, rank, tag,
@@ -56,7 +68,7 @@ int smpi_coll_tuned_reduce_scatter_gather(void *sendbuf, void *recvbuf,
         newrank = -1;
       } else {
         smpi_mpi_recv(tmp_buf, count, datatype, rank + 1, tag, comm, &status);
-        star_reduction(op, tmp_buf, recv_ptr, &new_count, &datatype);
+        smpi_op_apply(op, tmp_buf, recv_ptr, &new_count, &datatype);
         newrank = rank / 2;
       }
     } else                      /* rank >= 2*rem */
@@ -109,7 +121,7 @@ int smpi_coll_tuned_reduce_scatter_gather(void *sendbuf, void *recvbuf,
         /* tmp_buf contains data received in this step.
            recvbuf contains data accumulated so far */
 
-        star_reduction(op, (char *) tmp_buf + disps[recv_idx] * extent,
+        smpi_op_apply(op, (char *) tmp_buf + disps[recv_idx] * extent,
                        (char *) recv_ptr + disps[recv_idx] * extent,
                        &recv_cnt, &datatype);
 
@@ -212,16 +224,16 @@ int smpi_coll_tuned_reduce_scatter_gather(void *sendbuf, void *recvbuf,
       }
     }
     memcpy(recvbuf, recv_ptr, extent * count);
-    free(send_ptr);
-    free(recv_ptr);
+    smpi_free_tmp_buffer(send_ptr);
+    smpi_free_tmp_buffer(recv_ptr);
   }
 
 
-  else if (count >= comm_size) {
-    tmp_buf = (void *) xbt_malloc(count * extent);
+  else /* (count >= comm_size) */ {
+    tmp_buf = (void *) smpi_get_tmp_sendbuffer(count * extent);
 
     //if ((rank != root))
-    smpi_mpi_sendrecv(sendbuf, count, datatype, rank, tag,
+    smpi_mpi_sendrecv(sendbuf != MPI_IN_PLACE ? sendbuf : recvbuf, count, datatype, rank, tag,
                  recvbuf, count, datatype, rank, tag, comm, &status);
 
     rem = comm_size - pof2;
@@ -233,7 +245,7 @@ int smpi_coll_tuned_reduce_scatter_gather(void *sendbuf, void *recvbuf,
 
       else {
         smpi_mpi_recv(tmp_buf, count, datatype, rank + 1, tag, comm, &status);
-        star_reduction(op, tmp_buf, recvbuf, &count, &datatype);
+        smpi_op_apply(op, tmp_buf, recvbuf, &count, &datatype);
         newrank = rank / 2;
       }
     } else                      /* rank >= 2*rem */
@@ -286,7 +298,7 @@ int smpi_coll_tuned_reduce_scatter_gather(void *sendbuf, void *recvbuf,
         /* tmp_buf contains data received in this step.
            recvbuf contains data accumulated so far */
 
-        star_reduction(op, (char *) tmp_buf + disps[recv_idx] * extent,
+        smpi_op_apply(op, (char *) tmp_buf + disps[recv_idx] * extent,
                        (char *) recvbuf + disps[recv_idx] * extent,
                        &recv_cnt, &datatype);
 
@@ -388,6 +400,9 @@ int smpi_coll_tuned_reduce_scatter_gather(void *sendbuf, void *recvbuf,
       }
     }
   }
+  if (tmp_buf)
+    smpi_free_tmp_buffer(tmp_buf);
+  if(temporary_buffer==1) smpi_free_tmp_buffer(recvbuf);
   if (cnts)
     free(cnts);
   if (disps)