X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/482cc3463d7b94d6c39816c23f0d1e972a289728..954676b700e711f38ec4d286d33d5427d3f4ca46:/src/smpi/colls/alltoall-pair.c

diff --git a/src/smpi/colls/alltoall-pair.c b/src/smpi/colls/alltoall-pair.c
index 83658e02c0..8a0a76e15a 100644
--- a/src/smpi/colls/alltoall-pair.c
+++ b/src/smpi/colls/alltoall-pair.c
@@ -1,4 +1,11 @@
-#include "smpi/mpi.h"
+/* 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"
+
 /*****************************************************************************
 
  * Function: alltoall_pair
@@ -20,76 +27,72 @@
  * Auther: Ahmad Faraj
 
  ****************************************************************************/
-/*
-int
-alltoall_pair(void * send_buff, int send_count, MPI_Datatype send_type,
-	      void * recv_buff, int recv_count, MPI_Datatype recv_type,
-	      MPI_Comm comm)
+
+int smpi_coll_tuned_alltoall_pair_rma(void *send_buff, int send_count, MPI_Datatype send_type,
+                  void *recv_buff, int recv_count, MPI_Datatype recv_type,
+                  MPI_Comm comm)
 {
 
   MPI_Aint send_chunk, recv_chunk;
-  MPI_Status s;
   MPI_Win win;
   int assert = 0;
-  int i, src, dst, rank, num_procs;
-  int tag = 1, success = 1, failure = 0, pof2 = 1;
-
-  char * send_ptr = (char *) send_buff;
-  char * recv_ptr = (char *) recv_buff;
-  
-  MPI_Comm_rank(comm, &rank);
-  MPI_Comm_size(comm, &num_procs);
-  MPI_Type_extent(send_type, &send_chunk);
-  MPI_Type_extent(recv_type, &recv_chunk);
-
-  MPI_Win_create(recv_buff, num_procs*recv_chunk*send_count,recv_chunk,0,
-		 comm, &win);
+  int i, dst, rank, num_procs;
+
+  char *send_ptr = (char *) send_buff;
+
+  rank = smpi_comm_rank(comm);
+  num_procs = smpi_comm_size(comm);
+  send_chunk = smpi_datatype_get_extent(send_type);
+  recv_chunk = smpi_datatype_get_extent(recv_type);
+
+  win=smpi_mpi_win_create(recv_buff, num_procs * recv_chunk * send_count, recv_chunk, 0,
+                 comm);
   send_chunk *= send_count;
-  recv_chunk *= recv_count;  
-
-  MPI_Win_fence(assert, win);
-  for (i = 0; i < num_procs; i++)
-    {
-      src = dst = rank ^ i;
-      MPI_Put(send_ptr + dst * send_chunk, send_count, send_type, dst,
-	      rank*send_chunk, send_count, send_type, win);
-    }
-  MPI_Win_fence (assert, win);
-  MPI_Win_free(&win);
+  recv_chunk *= recv_count;
+
+  smpi_mpi_win_fence(assert, win);
+  for (i = 0; i < num_procs; i++) {
+    dst = rank ^ i;
+    smpi_mpi_put(send_ptr + dst * send_chunk, send_count, send_type, dst,
+            rank /* send_chunk*/, send_count, send_type, win);
+  }
+  smpi_mpi_win_fence(assert, win);
+  smpi_mpi_win_free(&win);
   return 0;
 }
-*/
 
-int
-smpi_coll_tuned_alltoall_pair(void * send_buff, int send_count, MPI_Datatype send_type,
-	      void * recv_buff, int recv_count, MPI_Datatype recv_type,
-	      MPI_Comm comm)
+
+int smpi_coll_tuned_alltoall_pair(void *send_buff, int send_count,
+                                  MPI_Datatype send_type,
+                                  void *recv_buff, int recv_count,
+                                  MPI_Datatype recv_type, MPI_Comm comm)
 {
 
   MPI_Aint send_chunk, recv_chunk;
   MPI_Status s;
   int i, src, dst, rank, num_procs;
-  int tag = 1, success = 1;
+  int tag = COLL_TAG_ALLTOALL;
+  char *send_ptr = (char *) send_buff;
+  char *recv_ptr = (char *) recv_buff;
 
-  char * send_ptr = (char *) send_buff;
-  char * recv_ptr = (char *) recv_buff;
-  
-  MPI_Comm_rank(comm, &rank);
-  MPI_Comm_size(comm, &num_procs);
-  MPI_Type_extent(send_type, &send_chunk);
-  MPI_Type_extent(recv_type, &recv_chunk);
+  rank = smpi_comm_rank(comm);
+  num_procs = smpi_comm_size(comm);
+
+  if((num_procs&(num_procs-1)))
+    THROWF(arg_error,0, "alltoall pair algorithm can't be used with non power of two number of processes ! ");
+
+  send_chunk = smpi_datatype_get_extent(send_type);
+  recv_chunk = smpi_datatype_get_extent(recv_type);
 
   send_chunk *= send_count;
-  recv_chunk *= recv_count;  
+  recv_chunk *= recv_count;
 
-  for (i = 0; i < num_procs; i++)
-    {
-      src = dst = rank ^ i;     
-      MPI_Sendrecv(send_ptr + dst * send_chunk, send_count, send_type, dst,
-		   tag, recv_ptr + src * recv_chunk, recv_count, recv_type,
-		   src, tag, comm, &s);
-    }
+  for (i = 0; i < num_procs; i++) {
+    src = dst = rank ^ i;
+    smpi_mpi_sendrecv(send_ptr + dst * send_chunk, send_count, send_type, dst, tag,
+		 recv_ptr + src * recv_chunk, recv_count, recv_type, src, tag,
+		 comm, &s);
+  }
 
-  return success;
+  return MPI_SUCCESS;
 }
-