X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/70ed180f3ce3495678d048e9e396ef5eb65a6a99..ad12e936954df10a0395da3b161e820e07e97f52:/src/smpi/smpi_mpi.c

diff --git a/src/smpi/smpi_mpi.c b/src/smpi/smpi_mpi.c
index 5436c0a4cc..4a5cd8d51d 100644
--- a/src/smpi/smpi_mpi.c
+++ b/src/smpi/smpi_mpi.c
@@ -1,4 +1,8 @@
-/* $Id$tag */
+/* Copyright (c) 2007, 2008, 2009, 2010. 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 "private.h"
 #include "smpi_coll_private.h"
@@ -11,12 +15,18 @@ XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_mpi, smpi,
 
 int MPI_Init(int* argc, char*** argv) {
   smpi_process_init(argc, argv);
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
+#ifdef HAVE_TRACING
+  TRACE_smpi_init(smpi_process_index());
+#endif
   return MPI_SUCCESS;
 }
 
 int MPI_Finalize(void) {
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  TRACE_smpi_finalize(smpi_process_index());
+#endif
+  smpi_bench_end(-1, NULL);
   smpi_process_destroy();
   return MPI_SUCCESS;
 }
@@ -31,33 +41,33 @@ int MPI_Init_thread(int* argc, char*** argv, int required, int* provided) {
 int MPI_Query_thread(int* provided) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(provided == NULL) {
     retval = MPI_ERR_ARG;
   } else {
     *provided = MPI_THREAD_MULTIPLE;
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Is_thread_main(int* flag) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(flag == NULL) {
     retval = MPI_ERR_ARG;
   } else {
     *flag = smpi_process_index() == 0;
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Abort(MPI_Comm comm, int errorcode) {
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   smpi_process_destroy();
   // FIXME: should kill all processes in comm instead
   SIMIX_process_kill(SIMIX_process_self());
@@ -67,16 +77,43 @@ int MPI_Abort(MPI_Comm comm, int errorcode) {
 double MPI_Wtime(void) {
   double time;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   time = SIMIX_get_clock();
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return time;
 }
 
+int MPI_Address(void *location, MPI_Aint *address) {
+  int retval;
+
+  smpi_bench_end(-1, NULL);
+  if(!address) {
+    retval = MPI_ERR_ARG;
+  } else {
+    *address = (MPI_Aint)location;
+  }
+  smpi_bench_begin(-1, NULL);
+  return retval;
+}
+
+int MPI_Type_free(MPI_Datatype* datatype) {
+  int retval;
+
+  smpi_bench_end(-1, NULL);
+  if(!datatype) {
+    retval = MPI_ERR_ARG;
+  } else {
+    // FIXME: always fail for now
+    retval = MPI_ERR_TYPE;
+  }
+  smpi_bench_begin(-1, NULL);
+  return retval;
+}
+
 int MPI_Type_size(MPI_Datatype datatype, size_t* size) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(datatype == MPI_DATATYPE_NULL) {
     retval = MPI_ERR_TYPE;
   } else if(size == NULL) {
@@ -85,14 +122,14 @@ int MPI_Type_size(MPI_Datatype datatype, size_t* size) {
     *size = smpi_datatype_size(datatype);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Type_get_extent(MPI_Datatype datatype, MPI_Aint* lb, MPI_Aint* extent) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(datatype == MPI_DATATYPE_NULL) {
     retval = MPI_ERR_TYPE;
   } else if(lb == NULL || extent == NULL) {
@@ -100,14 +137,30 @@ int MPI_Type_get_extent(MPI_Datatype datatype, MPI_Aint* lb, MPI_Aint* extent) {
   } else {
     retval = smpi_datatype_extent(datatype, lb, extent);
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
+  return retval;
+}
+
+int MPI_Type_extent(MPI_Datatype datatype, MPI_Aint* extent) {
+  int retval;
+  MPI_Aint dummy;
+
+  smpi_bench_end(-1, NULL);
+  if(datatype == MPI_DATATYPE_NULL) {
+    retval = MPI_ERR_TYPE;
+  } else if(extent == NULL) {
+    retval = MPI_ERR_ARG;
+  } else {
+    retval = smpi_datatype_extent(datatype, &dummy, extent);
+  }
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Type_lb(MPI_Datatype datatype, MPI_Aint* disp) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(datatype == MPI_DATATYPE_NULL) {
     retval = MPI_ERR_TYPE;
   } else if(disp == NULL) {
@@ -116,14 +169,14 @@ int MPI_Type_lb(MPI_Datatype datatype, MPI_Aint* disp) {
     *disp = smpi_datatype_lb(datatype);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Type_ub(MPI_Datatype datatype, MPI_Aint* disp) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(datatype == MPI_DATATYPE_NULL) {
     retval = MPI_ERR_TYPE;
   } else if(disp == NULL) {
@@ -132,28 +185,28 @@ int MPI_Type_ub(MPI_Datatype datatype, MPI_Aint* disp) {
     *disp = smpi_datatype_ub(datatype);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Op_create(MPI_User_function* function, int commute, MPI_Op* op) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(function == NULL || op == NULL) {
     retval = MPI_ERR_ARG;
   } else {
     *op = smpi_op_new(function, commute);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Op_free(MPI_Op* op) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(op == NULL) {
     retval = MPI_ERR_ARG;
   } else if(*op == MPI_OP_NULL) {
@@ -163,14 +216,14 @@ int MPI_Op_free(MPI_Op* op) {
     *op = MPI_OP_NULL;
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Group_free(MPI_Group *group) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(group == NULL) {
     retval = MPI_ERR_ARG;
   } else {
@@ -178,14 +231,14 @@ int MPI_Group_free(MPI_Group *group) {
     *group = MPI_GROUP_NULL;
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Group_size(MPI_Group group, int* size) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(group == MPI_GROUP_NULL) {
     retval = MPI_ERR_GROUP;
   } else if(size == NULL) {
@@ -194,14 +247,14 @@ int MPI_Group_size(MPI_Group group, int* size) {
     *size = smpi_group_size(group);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Group_rank(MPI_Group group, int* rank) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(group == MPI_GROUP_NULL) {
     retval = MPI_ERR_GROUP;
   } else if(rank == NULL) {
@@ -210,14 +263,14 @@ int MPI_Group_rank(MPI_Group group, int* rank) {
     *rank = smpi_group_rank(group, smpi_process_index());
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Group_translate_ranks (MPI_Group group1, int n, int* ranks1, MPI_Group group2, int* ranks2) {
   int retval, i, index;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(group1 == MPI_GROUP_NULL || group2 == MPI_GROUP_NULL) {
     retval = MPI_ERR_GROUP;
   } else {
@@ -227,14 +280,14 @@ int MPI_Group_translate_ranks (MPI_Group group1, int n, int* ranks1, MPI_Group g
     }
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Group_compare(MPI_Group group1, MPI_Group group2, int* result) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(group1 == MPI_GROUP_NULL || group2 == MPI_GROUP_NULL) {
     retval = MPI_ERR_GROUP;
   } else if(result == NULL) {
@@ -243,14 +296,14 @@ int MPI_Group_compare(MPI_Group group1, MPI_Group group2, int* result) {
     *result = smpi_group_compare(group1, group2);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Group_union(MPI_Group group1, MPI_Group group2, MPI_Group* newgroup) {
   int retval, i, proc1, proc2, size, size2;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(group1 == MPI_GROUP_NULL || group2 == MPI_GROUP_NULL) {
     retval = MPI_ERR_GROUP;
   } else if(newgroup == NULL) {
@@ -282,14 +335,14 @@ int MPI_Group_union(MPI_Group group1, MPI_Group group2, MPI_Group* newgroup) {
     smpi_group_use(*newgroup);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Group_intersection(MPI_Group group1, MPI_Group group2, MPI_Group* newgroup) {
    int retval, i, proc1, proc2, size, size2;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(group1 == MPI_GROUP_NULL || group2 == MPI_GROUP_NULL) {
     retval = MPI_ERR_GROUP;
   } else if(newgroup == NULL) {
@@ -320,14 +373,14 @@ int MPI_Group_intersection(MPI_Group group1, MPI_Group group2, MPI_Group* newgro
     smpi_group_use(*newgroup);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Group_difference(MPI_Group group1, MPI_Group group2, MPI_Group* newgroup) {
   int retval, i, proc1, proc2, size, size2;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(group1 == MPI_GROUP_NULL || group2 == MPI_GROUP_NULL) {
     retval = MPI_ERR_GROUP;
   } else if(newgroup == NULL) {
@@ -356,14 +409,14 @@ int MPI_Group_difference(MPI_Group group1, MPI_Group group2, MPI_Group* newgroup
     smpi_group_use(*newgroup);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Group_incl(MPI_Group group, int n, int* ranks, MPI_Group* newgroup) {
   int retval, i, index;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(group == MPI_GROUP_NULL) {
     retval = MPI_ERR_GROUP;
   } else if(newgroup == NULL) {
@@ -383,14 +436,14 @@ int MPI_Group_incl(MPI_Group group, int n, int* ranks, MPI_Group* newgroup) {
     smpi_group_use(*newgroup);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Group_excl(MPI_Group group, int n, int* ranks, MPI_Group* newgroup) {
   int retval, i, size, rank, index;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(group == MPI_GROUP_NULL) {
     retval = MPI_ERR_GROUP;
   } else if(newgroup == NULL) {
@@ -420,14 +473,14 @@ int MPI_Group_excl(MPI_Group group, int n, int* ranks, MPI_Group* newgroup) {
     smpi_group_use(*newgroup);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Group_range_incl(MPI_Group group, int n, int ranges[][3], MPI_Group* newgroup) {
   int retval, i, j, rank, size, index;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(group == MPI_GROUP_NULL) {
     retval = MPI_ERR_GROUP;
   } else if(newgroup == NULL) {
@@ -463,14 +516,14 @@ int MPI_Group_range_incl(MPI_Group group, int n, int ranges[][3], MPI_Group* new
     smpi_group_use(*newgroup);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Group_range_excl(MPI_Group group, int n, int ranges[][3], MPI_Group* newgroup) {
   int retval, i, newrank, rank, size, index, add;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(group == MPI_GROUP_NULL) {
     retval = MPI_ERR_GROUP;
   } else if(newgroup == NULL) {
@@ -514,28 +567,28 @@ int MPI_Group_range_excl(MPI_Group group, int n, int ranges[][3], MPI_Group* new
     smpi_group_use(*newgroup);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Comm_rank(MPI_Comm comm, int* rank) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else {
     *rank = smpi_comm_rank(comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Comm_size(MPI_Comm comm, int* size) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(size == NULL) {
@@ -544,14 +597,14 @@ int MPI_Comm_size(MPI_Comm comm, int* size) {
     *size = smpi_comm_size(comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Comm_group(MPI_Comm comm, MPI_Group* group) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(group == NULL) {
@@ -560,14 +613,14 @@ int MPI_Comm_group(MPI_Comm comm, MPI_Group* group) {
     *group = smpi_comm_group(comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Comm_compare(MPI_Comm comm1, MPI_Comm comm2, int* result) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(comm1 == MPI_COMM_NULL || comm2 == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(result == NULL) {
@@ -583,14 +636,14 @@ int MPI_Comm_compare(MPI_Comm comm1, MPI_Comm comm2, int* result) {
     }
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Comm_dup(MPI_Comm comm, MPI_Comm* newcomm) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(newcomm == NULL) {
@@ -599,14 +652,14 @@ int MPI_Comm_dup(MPI_Comm comm, MPI_Comm* newcomm) {
     *newcomm = smpi_comm_new(smpi_comm_group(comm));
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Comm_create(MPI_Comm comm, MPI_Group group, MPI_Comm* newcomm) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(group == MPI_GROUP_NULL) {
@@ -617,14 +670,14 @@ int MPI_Comm_create(MPI_Comm comm, MPI_Group group, MPI_Comm* newcomm) {
     *newcomm = smpi_comm_new(group);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Comm_free(MPI_Comm* comm) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL);
   if(comm == NULL) {
     retval = MPI_ERR_ARG;
   } else if(*comm == MPI_COMM_NULL) {
@@ -634,57 +687,216 @@ int MPI_Comm_free(MPI_Comm* comm) {
     *comm = MPI_COMM_NULL;
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
+  return retval;
+}
+
+int MPI_Send_init(void* buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm, MPI_Request* request) {
+  int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+  smpi_bench_end(rank, "Send_init");
+  if(request == NULL) {
+    retval = MPI_ERR_ARG;
+  } else if (comm == MPI_COMM_NULL) {
+    retval = MPI_ERR_COMM;
+  } else {
+    *request = smpi_mpi_send_init(buf, count, datatype, dst, tag, comm);
+    retval = MPI_SUCCESS;
+  }
+  smpi_bench_begin(rank, "Send_init");
+  return retval;
+}
+
+int MPI_Recv_init(void* buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm, MPI_Request* request) {
+  int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+  smpi_bench_end(rank, "Recv_init");
+  if(request == NULL) {
+    retval = MPI_ERR_ARG;
+  } else if (comm == MPI_COMM_NULL) {
+    retval = MPI_ERR_COMM;
+  } else {
+    *request = smpi_mpi_recv_init(buf, count, datatype, src, tag, comm);
+    retval = MPI_SUCCESS;
+  }
+  smpi_bench_begin(rank, "Recv_init");
+  return retval;
+}
+
+int MPI_Start(MPI_Request* request) {
+  int retval;
+  MPI_Comm comm = (*request)->comm;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+  smpi_bench_end(rank, "Start");
+  if(request == NULL) {
+    retval = MPI_ERR_ARG;
+  } else {
+    smpi_mpi_start(*request);
+    retval = MPI_SUCCESS;
+  }
+  smpi_bench_begin(rank, "Start");
+  return retval;
+}
+
+int MPI_Startall(int count, MPI_Request* requests) {
+  int retval;
+  MPI_Comm comm = count > 0 && requests ? requests[0]->comm : MPI_COMM_NULL;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+  smpi_bench_end(rank, "Startall");
+  if(requests == NULL) {
+    retval = MPI_ERR_ARG;
+  } else {
+    smpi_mpi_startall(count, requests);
+    retval = MPI_SUCCESS;
+  }
+  smpi_bench_begin(rank, "Startall");
+  return retval;
+}
+
+int MPI_Request_free(MPI_Request* request) {
+  int retval;
+  MPI_Comm comm = (*request)->comm;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+  smpi_bench_end(rank, "Request_free");
+  if(request == NULL) {
+    retval = MPI_ERR_ARG;
+  } else {
+    smpi_mpi_request_free(request);
+    retval = MPI_SUCCESS;
+  }
+  smpi_bench_begin(rank, "Request_free");
   return retval;
 }
 
 int MPI_Irecv(void* buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm, MPI_Request* request) {
   int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  int src_traced = smpi_group_rank(smpi_comm_group(comm), src);
+  TRACE_smpi_ptp_in (rank, src_traced, rank, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Irecv");
   if(request == NULL) {
     retval = MPI_ERR_ARG;
+  } else if (comm == MPI_COMM_NULL) {
+    retval = MPI_ERR_COMM;
   } else {
     *request = smpi_mpi_irecv(buf, count, datatype, src, tag, comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Irecv");
+#ifdef HAVE_TRACING
+  TRACE_smpi_ptp_out (rank, src_traced, rank, __FUNCTION__);
+  (*request)->recv = 1;
+#endif
   return retval;
 }
 
 int MPI_Isend(void* buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm, MPI_Request* request) {
   int retval;
-
-  smpi_bench_end();
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+#ifdef HAVE_TRACING
+  int dst_traced = smpi_group_rank(smpi_comm_group(comm), dst);
+  TRACE_smpi_ptp_in (rank, rank, dst_traced, __FUNCTION__);
+  TRACE_smpi_send (rank, rank, dst_traced);
+#endif
+  smpi_bench_end(rank, "Isend");
   if(request == NULL) {
     retval = MPI_ERR_ARG;
+  } else if (comm == MPI_COMM_NULL) {
+    retval = MPI_ERR_COMM;
   } else {
     *request = smpi_mpi_isend(buf, count, datatype, dst, tag, comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Isend");
+#ifdef HAVE_TRACING
+  TRACE_smpi_ptp_out (rank, rank, dst_traced, __FUNCTION__);
+  (*request)->send = 1;
+#endif
   return retval;
 }
 
 int MPI_Recv(void* buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm, MPI_Status* status) {
-  smpi_bench_end();
-  smpi_mpi_recv(buf, count, datatype, src, tag, comm, status);
-  smpi_bench_begin();
-  return MPI_SUCCESS;
+  int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+#ifdef HAVE_TRACING
+  int src_traced = smpi_group_rank(smpi_comm_group(comm), src);
+  TRACE_smpi_ptp_in (rank, src_traced, rank, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Recv");
+  if (comm == MPI_COMM_NULL) {
+    retval = MPI_ERR_COMM;
+  } else {
+    smpi_mpi_recv(buf, count, datatype, src, tag, comm, status);
+    retval = MPI_SUCCESS;
+  }
+  smpi_bench_begin(rank, "Recv");
+#ifdef HAVE_TRACING
+  TRACE_smpi_ptp_out (rank, src_traced, rank, __FUNCTION__);
+  TRACE_smpi_recv (rank, src_traced, rank);
+#endif
+  return retval;
 }
 
 int MPI_Send(void* buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm) {
-  smpi_bench_end();
-  smpi_mpi_send(buf, count, datatype, dst, tag, comm);
-  smpi_bench_begin();
-  return MPI_SUCCESS;
+  int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+#ifdef HAVE_TRACING
+  int dst_traced = smpi_group_rank(smpi_comm_group(comm), dst);
+  TRACE_smpi_ptp_in (rank, rank, dst_traced, __FUNCTION__);
+  TRACE_smpi_send (rank, rank, dst_traced);
+#endif
+  smpi_bench_end(rank, "Send");
+  if (comm == MPI_COMM_NULL) {
+    retval = MPI_ERR_COMM;
+  } else {
+    smpi_mpi_send(buf, count, datatype, dst, tag, comm);
+    retval = MPI_SUCCESS;
+  }
+  smpi_bench_begin(rank, "Send");
+#ifdef HAVE_TRACING
+  TRACE_smpi_ptp_out (rank, rank, dst_traced, __FUNCTION__);
+#endif
+  return retval;
 }
 
 int MPI_Sendrecv(void* sendbuf, int sendcount, MPI_Datatype sendtype, int dst, int sendtag, void* recvbuf, int recvcount, MPI_Datatype recvtype, int src, int recvtag, MPI_Comm comm, MPI_Status* status) {
-  smpi_bench_end();
-  smpi_mpi_sendrecv(sendbuf, sendcount, sendtype, dst, sendtag, recvbuf, recvcount, recvtype, src, recvtag, comm, status);
-  smpi_bench_begin();
-  return MPI_SUCCESS;
+  int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+#ifdef HAVE_TRACING
+  int dst_traced = smpi_group_rank(smpi_comm_group(comm), dst);
+  int src_traced = smpi_group_rank(smpi_comm_group(comm), src);
+  TRACE_smpi_ptp_in (rank, src_traced, dst_traced, __FUNCTION__);
+  TRACE_smpi_send (rank, rank, dst_traced);
+  TRACE_smpi_send (rank, src_traced, rank);
+#endif
+  smpi_bench_end(rank, "Sendrecv");
+  if (comm == MPI_COMM_NULL) {
+    retval = MPI_ERR_COMM;
+  } else if (sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) {
+    retval = MPI_ERR_TYPE;
+  } else {
+    smpi_mpi_sendrecv(sendbuf, sendcount, sendtype, dst, sendtag, recvbuf, recvcount, recvtype, src, recvtag, comm, status);
+    retval = MPI_SUCCESS;
+  }
+  smpi_bench_begin(rank, "Sendrecv");
+#ifdef HAVE_TRACING
+  TRACE_smpi_ptp_out (rank, src_traced, dst_traced, __FUNCTION__);
+  TRACE_smpi_recv (rank, rank, dst_traced);
+  TRACE_smpi_recv (rank, src_traced, rank);
+#endif
+  return retval;
 }
 
 int MPI_Sendrecv_replace(void* buf, int count, MPI_Datatype datatype, int dst, int sendtag, int src, int recvtag, MPI_Comm comm, MPI_Status* status) {
@@ -702,8 +914,11 @@ int MPI_Sendrecv_replace(void* buf, int count, MPI_Datatype datatype, int dst, i
 
 int MPI_Test(MPI_Request* request, int* flag, MPI_Status* status) {
   int retval;
+  int rank = request && (*request)->comm != MPI_COMM_NULL
+             ? smpi_comm_rank((*request)->comm)
+             : -1;
 
-  smpi_bench_end();
+  smpi_bench_end(rank, "Test");
   if(request == NULL || flag == NULL) {
     retval = MPI_ERR_ARG;
   } else if(*request == MPI_REQUEST_NULL) {
@@ -712,28 +927,38 @@ int MPI_Test(MPI_Request* request, int* flag, MPI_Status* status) {
     *flag = smpi_mpi_test(request, status);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Test");
   return retval;
 }
 
 int MPI_Testany(int count, MPI_Request requests[], int* index, int* flag, MPI_Status* status) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL); //FIXME
   if(index == NULL || flag == NULL) {
     retval = MPI_ERR_ARG;
   } else {
     *flag = smpi_mpi_testany(count, requests, index, status);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Wait(MPI_Request* request, MPI_Status* status) {
   int retval;
-
-  smpi_bench_end();
+  int rank = request && (*request)->comm != MPI_COMM_NULL
+             ? smpi_comm_rank((*request)->comm)
+             : -1;
+
+#ifdef HAVE_TRACING
+  MPI_Group group = smpi_comm_group((*request)->comm);
+  int src_traced = smpi_group_rank (group , (*request)->src);
+  int dst_traced = smpi_group_rank (group , (*request)->dst);
+  int is_wait_for_receive = (*request)->recv;
+  TRACE_smpi_ptp_in (rank, src_traced, dst_traced, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Wait");
   if(request == NULL) {
     retval = MPI_ERR_ARG;
   } else if(*request == MPI_REQUEST_NULL) {
@@ -742,77 +967,197 @@ int MPI_Wait(MPI_Request* request, MPI_Status* status) {
     smpi_mpi_wait(request, status);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Wait");
+#ifdef HAVE_TRACING
+  TRACE_smpi_ptp_out (rank, src_traced, dst_traced, __FUNCTION__);
+  if (is_wait_for_receive){
+    TRACE_smpi_recv (rank, src_traced, dst_traced);
+  }
+#endif
   return retval;
 }
 
 int MPI_Waitany(int count, MPI_Request requests[], int* index, MPI_Status* status) {
   int retval;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  //save requests information for tracing
+  int i;
+  xbt_dynar_t srcs = xbt_dynar_new (sizeof(int), xbt_free);
+  xbt_dynar_t dsts = xbt_dynar_new (sizeof(int), xbt_free);
+  xbt_dynar_t recvs = xbt_dynar_new (sizeof(int), xbt_free);
+  for (i = 0; i < count; i++){
+    MPI_Request req = requests[i]; //already received requests are no longer valid
+    if (req){
+      int *asrc = xbt_new(int, 1);
+      int *adst = xbt_new(int, 1);
+      int *arecv = xbt_new(int, 1);
+      *asrc = req->src;
+      *adst = req->dst;
+      *arecv = req->recv;
+      xbt_dynar_insert_at (srcs, i, asrc);
+      xbt_dynar_insert_at (dsts, i, adst);
+      xbt_dynar_insert_at (recvs, i, arecv);
+    }else{
+      int *t = xbt_new(int, 1);
+      xbt_dynar_insert_at (srcs, i, t);
+      xbt_dynar_insert_at (dsts, i, t);
+      xbt_dynar_insert_at (recvs, i, t);
+    }
+  }
+
+  //search for a suitable request to give the rank of current mpi proc
+  MPI_Request req = NULL;
+  for (i = 0; i < count && req == NULL; i++) {
+    req = requests[i];
+  }
+  MPI_Comm comm = (req)->comm;
+  int rank_traced = smpi_comm_rank(comm);
+  TRACE_smpi_ptp_in (rank_traced, -1, -1, __FUNCTION__);
+#endif
+  smpi_bench_end(-1, NULL); //FIXME
   if(index == NULL) {
     retval = MPI_ERR_ARG;
   } else {
     *index = smpi_mpi_waitany(count, requests, status);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
+#ifdef HAVE_TRACING
+  int src_traced, dst_traced, is_wait_for_receive;
+  xbt_dynar_get_cpy (srcs, *index, &src_traced);
+  xbt_dynar_get_cpy (dsts, *index, &dst_traced);
+  xbt_dynar_get_cpy (recvs, *index, &is_wait_for_receive);
+  if (is_wait_for_receive){
+    TRACE_smpi_recv (rank_traced, src_traced, dst_traced);
+  }
+  TRACE_smpi_ptp_out (rank_traced, src_traced, dst_traced, __FUNCTION__);
+  //clean-up of dynars
+  xbt_free (srcs);
+  xbt_free (dsts);
+  xbt_free (recvs);
+#endif
   return retval;
 }
 
 int MPI_Waitall(int count, MPI_Request requests[],  MPI_Status status[]) {
-  smpi_bench_end();
+
+#ifdef HAVE_TRACING
+  //save information from requests
+  int i;
+  xbt_dynar_t srcs = xbt_dynar_new (sizeof(int), xbt_free);
+  xbt_dynar_t dsts = xbt_dynar_new (sizeof(int), xbt_free);
+  xbt_dynar_t recvs = xbt_dynar_new (sizeof(int), xbt_free);
+  for (i = 0; i < count; i++){
+    MPI_Request req = requests[i]; //all req should be valid in Waitall
+    int *asrc = xbt_new(int, 1);
+    int *adst = xbt_new(int, 1);
+    int *arecv = xbt_new(int, 1);
+    *asrc = req->src;
+    *adst = req->dst;
+    *arecv = req->recv;
+    xbt_dynar_insert_at (srcs, i, asrc);
+    xbt_dynar_insert_at (dsts, i, adst);
+    xbt_dynar_insert_at (recvs, i, arecv);
+  }
+
+//  find my rank inside one of MPI_Comm's of the requests
+  MPI_Request req = NULL;
+  for (i = 0; i < count && req == NULL; i++) {
+    req = requests[i];
+  }
+  MPI_Comm comm = (req)->comm;
+  int rank_traced = smpi_comm_rank(comm);
+  TRACE_smpi_ptp_in (rank_traced, -1, -1, __FUNCTION__);
+#endif
+  smpi_bench_end(-1, NULL); //FIXME
   smpi_mpi_waitall(count, requests, status);
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
+#ifdef HAVE_TRACING
+  for (i = 0; i < count; i++){
+    int src_traced, dst_traced, is_wait_for_receive;
+    xbt_dynar_get_cpy (srcs, i, &src_traced);
+    xbt_dynar_get_cpy (dsts, i, &dst_traced);
+    xbt_dynar_get_cpy (recvs, i, &is_wait_for_receive);
+    if (is_wait_for_receive){
+      TRACE_smpi_recv (rank_traced, src_traced, dst_traced);
+    }
+  }
+  TRACE_smpi_ptp_out (rank_traced, -1, -1, __FUNCTION__);
+  //clean-up of dynars
+  xbt_free (srcs);
+  xbt_free (dsts);
+  xbt_free (recvs);
+#endif
   return MPI_SUCCESS;
 }
 
 int MPI_Waitsome(int incount, MPI_Request requests[], int* outcount, int* indices, MPI_Status status[]) {
   int retval;
 
-  smpi_bench_end();
+  smpi_bench_end(-1, NULL); //FIXME
   if(outcount == NULL || indices == NULL) {
     retval = MPI_ERR_ARG;
   } else {
     *outcount = smpi_mpi_waitsome(incount, requests, indices, status);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
 
 int MPI_Bcast(void* buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm) {
   int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  int root_traced = smpi_group_rank(smpi_comm_group(comm), root);
+  TRACE_smpi_collective_in (rank, root_traced, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Bcast");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else {
     smpi_mpi_bcast(buf, count, datatype, root, comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Bcast");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, root_traced, __FUNCTION__);
+#endif
   return retval;
 }
 
 int MPI_Barrier(MPI_Comm comm) {
   int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_in (rank, -1, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Barrier");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else {
     smpi_mpi_barrier(comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Barrier");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, -1, __FUNCTION__);
+#endif
   return retval;
 }
 
 int MPI_Gather(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) {
   int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  int root_traced = smpi_group_rank(smpi_comm_group(comm), root);
+  TRACE_smpi_collective_in (rank, root_traced, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Gather");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) {
@@ -821,14 +1166,22 @@ int MPI_Gather(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbu
     smpi_mpi_gather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Gather");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, root_traced, __FUNCTION__);
+#endif
   return retval;
 }
 
 int MPI_Gatherv(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int* recvcounts, int* displs, MPI_Datatype recvtype, int root, MPI_Comm comm) {
   int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  int root_traced = smpi_group_rank(smpi_comm_group(comm), root);
+  TRACE_smpi_collective_in (rank, root_traced, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Gatherv");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) {
@@ -839,14 +1192,21 @@ int MPI_Gatherv(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvb
     smpi_mpi_gatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, root, comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Gatherv");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, root_traced, __FUNCTION__);
+#endif
   return retval;
 }
 
 int MPI_Allgather(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm) {
   int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_in (rank, -1, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Allgather");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) {
@@ -855,14 +1215,21 @@ int MPI_Allgather(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* rec
     smpi_mpi_allgather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Allgather");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, -1, __FUNCTION__);
+#endif
   return retval;
 }
 
 int MPI_Allgatherv(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int* recvcounts, int* displs, MPI_Datatype recvtype, MPI_Comm comm) {
   int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_in (rank, -1, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Allgatherv");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) {
@@ -873,14 +1240,22 @@ int MPI_Allgatherv(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* re
     smpi_mpi_allgatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Allgatherv");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, -1, __FUNCTION__);
+#endif
   return retval;
 }
 
 int MPI_Scatter(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) {
   int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  int root_traced = smpi_group_rank(smpi_comm_group(comm), root);
+  TRACE_smpi_collective_in (rank, root_traced, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Scatter");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) {
@@ -889,14 +1264,22 @@ int MPI_Scatter(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvb
     smpi_mpi_scatter(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Scatter");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, root_traced, __FUNCTION__);
+#endif
   return retval;
 }
 
 int MPI_Scatterv(void* sendbuf, int* sendcounts, int* displs, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) {
   int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  int root_traced = smpi_group_rank(smpi_comm_group(comm), root);
+  TRACE_smpi_collective_in (rank, root_traced, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Scatterv");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) {
@@ -907,14 +1290,22 @@ int MPI_Scatterv(void* sendbuf, int* sendcounts, int* displs, MPI_Datatype sendt
     smpi_mpi_scatterv(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root, comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Scatterv");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, root_traced, __FUNCTION__);
+#endif
   return retval;
 }
 
 int MPI_Reduce(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm) {
   int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  int root_traced = smpi_group_rank(smpi_comm_group(comm), root);
+  TRACE_smpi_collective_in (rank, root_traced, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Reduce");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(datatype == MPI_DATATYPE_NULL || op == MPI_OP_NULL) {
@@ -923,14 +1314,21 @@ int MPI_Reduce(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, M
     smpi_mpi_reduce(sendbuf, recvbuf, count, datatype, op, root, comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Reduce");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, root_traced, __FUNCTION__);
+#endif
   return retval;
 }
 
 int MPI_Allreduce(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) {
   int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_in (rank, -1, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Allreduce");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(datatype == MPI_DATATYPE_NULL) {
@@ -941,15 +1339,47 @@ int MPI_Allreduce(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype
     smpi_mpi_allreduce(sendbuf, recvbuf, count, datatype, op, comm);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Allreduce");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, -1, __FUNCTION__);
+#endif
+  return retval;
+}
+
+int MPI_Scan(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) {
+  int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
+
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_in (rank, -1, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Scan");
+  if(comm == MPI_COMM_NULL) {
+    retval = MPI_ERR_COMM;
+  } else if(datatype == MPI_DATATYPE_NULL) {
+    retval = MPI_ERR_TYPE;
+  } else if(op == MPI_OP_NULL) {
+    retval = MPI_ERR_OP;
+  } else {
+    smpi_mpi_scan(sendbuf, recvbuf, count, datatype, op, comm);
+    retval = MPI_SUCCESS;
+  }
+  smpi_bench_begin(rank, "Scan");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, -1, __FUNCTION__);
+#endif
   return retval;
 }
 
 int MPI_Reduce_scatter(void* sendbuf, void* recvbuf, int* recvcounts, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) {
-  int retval, i, rank, size, count;
+  int retval, i, size, count;
   int* displs;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_in (rank, -1, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Reduce_scatter");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(datatype == MPI_DATATYPE_NULL) {
@@ -961,7 +1391,6 @@ int MPI_Reduce_scatter(void* sendbuf, void* recvbuf, int* recvcounts, MPI_Dataty
   } else {
     /* arbitrarily choose root as rank 0 */
     /* TODO: faster direct implementation ? */
-    rank = smpi_comm_rank(comm);
     size = smpi_comm_size(comm);
     count = 0;
     displs = xbt_new(int, size);
@@ -974,21 +1403,21 @@ int MPI_Reduce_scatter(void* sendbuf, void* recvbuf, int* recvcounts, MPI_Dataty
     xbt_free(displs);
     retval = MPI_SUCCESS;
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Reduce_scatter");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, -1, __FUNCTION__);
+#endif
   return retval;
 }
 
-/**
- * MPI_Alltoall user entry point
- * 
- * Uses the logic of OpenMPI (upto 1.2.7 or greater) for the optimizations
- * ompi/mca/coll/tuned/coll_tuned_module.c
- **/
-
 int MPI_Alltoall(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm) {
   int retval, size, sendsize;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_in (rank, -1, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Alltoall");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) {
@@ -1004,14 +1433,21 @@ int MPI_Alltoall(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recv
       retval = smpi_coll_tuned_alltoall_pairwise(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm);
     }
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Alltoall");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, -1, __FUNCTION__);
+#endif
   return retval;
 }
 
 int MPI_Alltoallv(void* sendbuf, int* sendcounts, int* senddisps, MPI_Datatype sendtype, void* recvbuf, int *recvcounts, int* recvdisps, MPI_Datatype recvtype, MPI_Comm comm) {
   int retval;
+  int rank = comm != MPI_COMM_NULL ? smpi_comm_rank(comm) : -1;
 
-  smpi_bench_end();
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_in (rank, -1, __FUNCTION__);
+#endif
+  smpi_bench_end(rank, "Alltoallv");
   if(comm == MPI_COMM_NULL) {
     retval = MPI_ERR_COMM;
   } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) {
@@ -1019,8 +1455,23 @@ int MPI_Alltoallv(void* sendbuf, int* sendcounts, int* senddisps, MPI_Datatype s
   } else if(sendcounts == NULL || senddisps == NULL || recvcounts == NULL || recvdisps == NULL) {
     retval = MPI_ERR_ARG;
   } else {
-    retval = smpi_coll_basic_alltoallv(sendbuf, sendcounts, senddisps, sendtype, recvbuf, recvcounts, recvdisps, recvtype, comm); 
+    retval = smpi_coll_basic_alltoallv(sendbuf, sendcounts, senddisps, sendtype, recvbuf, recvcounts, recvdisps, recvtype, comm);
   }
-  smpi_bench_begin();
+  smpi_bench_begin(rank, "Alltoallv");
+#ifdef HAVE_TRACING
+  TRACE_smpi_collective_out (rank, -1, __FUNCTION__);
+#endif
+  return retval;
+}
+
+
+int MPI_Get_processor_name( char *name, int *resultlen ) {
+  int retval = MPI_SUCCESS;
+  smpi_bench_end(-1, NULL);
+  strncpy( name , SIMIX_host_get_name(SIMIX_host_self()), MPI_MAX_PROCESSOR_NAME-1);
+  *resultlen= strlen(name) > MPI_MAX_PROCESSOR_NAME ? MPI_MAX_PROCESSOR_NAME : strlen(name);
+
+  smpi_bench_begin(-1, NULL);
   return retval;
 }
+