Update copyright lines with new year.

[simgrid.git] / src / smpi / colls / smpi_nbc_impl.cpp

diff --git a/src/smpi/colls/smpi_nbc_impl.cpp b/src/smpi/colls/smpi_nbc_impl.cpp
index 4193b13..4a174df 100644 (file)
--- a/src/smpi/colls/smpi_nbc_impl.cpp
+++ b/src/smpi/colls/smpi_nbc_impl.cpp
@@ -1,6 +1,6 @@
 /* Asynchronous parts of the basic collective algorithms, meant to be used both for the naive default implementation, but also for non blocking collectives */
 
-/* Copyright (c) 2009-2019. The SimGrid Team. All rights reserved.          */
+/* Copyright (c) 2009-2020. 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. */
@@ -11,44 +11,46 @@
 namespace simgrid{
 namespace smpi{
 
-
-int Colls::ibarrier(MPI_Comm comm, MPI_Request* request)
+int colls::ibarrier(MPI_Comm comm, MPI_Request* request, int external)
 {
   int size = comm->size();
   int rank = comm->rank();
+  int system_tag=COLL_TAG_BARRIER-external;
   (*request) = new Request( nullptr, 0, MPI_BYTE,
-                         rank,rank, COLL_TAG_BARRIER, comm, MPI_REQ_PERSISTENT);
+                         rank,rank, system_tag, comm, MPI_REQ_PERSISTENT);
   if (rank > 0) {
     MPI_Request* requests = new MPI_Request[2];
     requests[0] = Request::isend (nullptr, 0, MPI_BYTE, 0,
-                             COLL_TAG_BARRIER,
+                             system_tag,
                              comm);
     requests[1] = Request::irecv (nullptr, 0, MPI_BYTE, 0,
-                             COLL_TAG_BARRIER,
+                             system_tag,
                              comm);
     (*request)->set_nbc_requests(requests, 2);
   }
   else {
     MPI_Request* requests = new MPI_Request[(size - 1) * 2];
     for (int i = 1; i < 2 * size - 1; i += 2) {
-      requests[i - 1] = Request::irecv(nullptr, 0, MPI_BYTE, MPI_ANY_SOURCE, COLL_TAG_BARRIER, comm);
-      requests[i]     = Request::isend(nullptr, 0, MPI_BYTE, (i + 1) / 2, COLL_TAG_BARRIER, comm);
+      requests[i - 1] = Request::irecv(nullptr, 0, MPI_BYTE, MPI_ANY_SOURCE, system_tag, comm);
+      requests[i]     = Request::isend(nullptr, 0, MPI_BYTE, (i + 1) / 2, system_tag, comm);
     }
     (*request)->set_nbc_requests(requests, 2*(size-1));
   }
   return MPI_SUCCESS;
 }
 
-int Colls::ibcast(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm, MPI_Request* request)
+int colls::ibcast(void* buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm, MPI_Request* request,
+                  int external)
 {
   int size = comm->size();
   int rank = comm->rank();
+  int system_tag=COLL_TAG_BCAST-external;
   (*request) = new Request( nullptr, 0, MPI_BYTE,
-                         rank,rank, COLL_TAG_BCAST, comm, MPI_REQ_PERSISTENT);
+                         rank,rank, system_tag, comm, MPI_REQ_PERSISTENT);
   if (rank != root) {
     MPI_Request* requests = new MPI_Request[1];
     requests[0] = Request::irecv (buf, count, datatype, root,
-                             COLL_TAG_BCAST,
+                             system_tag,
                              comm);
     (*request)->set_nbc_requests(requests, 1);
   }
@@ -58,7 +60,7 @@ int Colls::ibcast(void *buf, int count, MPI_Datatype datatype, int root, MPI_Com
     for (int i = 0; i < size; i++) {
       if(i!=root){
         requests[n] = Request::isend(buf, count, datatype, i,
-                                 COLL_TAG_BCAST,
+                                 system_tag,
                                  comm
                                  );
         n++;
@@ -69,11 +71,11 @@ int Colls::ibcast(void *buf, int count, MPI_Datatype datatype, int root, MPI_Com
   return MPI_SUCCESS;
 }
 
-int Colls::iallgather(const void *sendbuf, int sendcount, MPI_Datatype sendtype,
-                        void *recvbuf,int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request)
+int colls::iallgather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
+                      MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request, int external)
 {
 
-  const int system_tag = COLL_TAG_ALLGATHER;
+  const int system_tag = COLL_TAG_ALLGATHER-external;
   MPI_Aint lb = 0;
   MPI_Aint recvext = 0;
 
@@ -103,10 +105,10 @@ int Colls::iallgather(const void *sendbuf, int sendcount, MPI_Datatype sendtype,
   return MPI_SUCCESS;
 }
 
-int Colls::iscatter(const void *sendbuf, int sendcount, MPI_Datatype sendtype,
-                      void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request)
+int colls::iscatter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
+                    MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request, int external)
 {
-  const int system_tag = COLL_TAG_SCATTER;
+  const int system_tag = COLL_TAG_SCATTER-external;
   MPI_Aint lb = 0;
   MPI_Aint sendext = 0;
 
@@ -143,10 +145,10 @@ int Colls::iscatter(const void *sendbuf, int sendcount, MPI_Datatype sendtype,
   return MPI_SUCCESS;
 }
 
-int Colls::iallgatherv(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf,
-                         const int *recvcounts, const int *displs, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request)
+int colls::iallgatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int* recvcounts,
+                       const int* displs, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request, int external)
 {
-  const int system_tag = COLL_TAG_ALLGATHERV;
+  const int system_tag = COLL_TAG_ALLGATHERV-external;
   MPI_Aint lb = 0;
   MPI_Aint recvext = 0;
 
@@ -177,8 +179,10 @@ int Colls::iallgatherv(const void *sendbuf, int sendcount, MPI_Datatype sendtype
   return MPI_SUCCESS;
 }
 
-int Colls::ialltoall( const void *sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request){
-  int system_tag   = COLL_TAG_ALLTOALL;
+int colls::ialltoall(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
+                     MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request, int external)
+{
+  int system_tag   = COLL_TAG_ALLTOALL-external;
   MPI_Aint lb      = 0;
   MPI_Aint sendext = 0;
   MPI_Aint recvext = 0;
@@ -220,9 +224,11 @@ int Colls::ialltoall( const void *sendbuf, int sendcount, MPI_Datatype sendtype,
   return MPI_SUCCESS;
 }
 
-int Colls::ialltoallv(const void *sendbuf, const int *sendcounts, const int *senddisps, MPI_Datatype sendtype,
-                              void *recvbuf, const int *recvcounts, const int *recvdisps, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request *request){
-  const int system_tag = COLL_TAG_ALLTOALLV;
+int colls::ialltoallv(const void* sendbuf, const int* sendcounts, const int* senddisps, MPI_Datatype sendtype,
+                      void* recvbuf, const int* recvcounts, const int* recvdisps, MPI_Datatype recvtype, MPI_Comm comm,
+                      MPI_Request* request, int external)
+{
+  const int system_tag = COLL_TAG_ALLTOALLV-external;
   MPI_Aint lb = 0;
   MPI_Aint sendext = 0;
   MPI_Aint recvext = 0;
@@ -268,9 +274,11 @@ int Colls::ialltoallv(const void *sendbuf, const int *sendcounts, const int *sen
   return err;
 }
 
-int Colls::ialltoallw(const void *sendbuf, const int *sendcounts, const int *senddisps, const MPI_Datatype* sendtypes,
-                              void *recvbuf, const int *recvcounts, const int *recvdisps, const MPI_Datatype* recvtypes, MPI_Comm comm, MPI_Request *request){
-  const int system_tag = COLL_TAG_ALLTOALLV;
+int colls::ialltoallw(const void* sendbuf, const int* sendcounts, const int* senddisps, const MPI_Datatype* sendtypes,
+                      void* recvbuf, const int* recvcounts, const int* recvdisps, const MPI_Datatype* recvtypes,
+                      MPI_Comm comm, MPI_Request* request, int external)
+{
+  const int system_tag = COLL_TAG_ALLTOALLW-external;
 
   /* Initialize. */
   int rank = comm->rank();
@@ -311,10 +319,10 @@ int Colls::ialltoallw(const void *sendbuf, const int *sendcounts, const int *sen
   return err;
 }
 
-int Colls::igather(const void *sendbuf, int sendcount, MPI_Datatype sendtype,
-                     void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request)
+int colls::igather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
+                   MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request, int external)
 {
-  const int system_tag = COLL_TAG_GATHER;
+  const int system_tag = COLL_TAG_GATHER-external;
   MPI_Aint lb = 0;
   MPI_Aint recvext = 0;
 
@@ -349,10 +357,11 @@ int Colls::igather(const void *sendbuf, int sendcount, MPI_Datatype sendtype,
   return MPI_SUCCESS;
 }
 
-int Colls::igatherv(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, const int *recvcounts, const int *displs,
-                      MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request)
+int colls::igatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int* recvcounts,
+                    const int* displs, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request,
+                    int external)
 {
-  int system_tag = COLL_TAG_GATHERV;
+  int system_tag = COLL_TAG_GATHERV-external;
   MPI_Aint lb = 0;
   MPI_Aint recvext = 0;
   
@@ -386,10 +395,11 @@ int Colls::igatherv(const void *sendbuf, int sendcount, MPI_Datatype sendtype, v
   }
   return MPI_SUCCESS;
 }
-int Colls::iscatterv(const void *sendbuf, const int *sendcounts, const int *displs, MPI_Datatype sendtype, void *recvbuf, int recvcount,
-                       MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request *request)
+int colls::iscatterv(const void* sendbuf, const int* sendcounts, const int* displs, MPI_Datatype sendtype,
+                     void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request,
+                     int external)
 {
-  int system_tag = COLL_TAG_SCATTERV;
+  int system_tag = COLL_TAG_SCATTERV-external;
   MPI_Aint lb = 0;
   MPI_Aint sendext = 0;
 
@@ -426,14 +436,14 @@ int Colls::iscatterv(const void *sendbuf, const int *sendcounts, const int *disp
   return MPI_SUCCESS;
 }
 
-int Colls::ireduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root,
-                     MPI_Comm comm, MPI_Request* request)
+int colls::ireduce(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root,
+                   MPI_Comm comm, MPI_Request* request, int external)
 {
-  const int system_tag = COLL_TAG_REDUCE;
+  const int system_tag = COLL_TAG_REDUCE-external;
   MPI_Aint lb = 0;
   MPI_Aint dataext = 0;
 
-  void* sendtmpbuf = const_cast<void *>(sendbuf);
+  const void* real_sendbuf = sendbuf;
 
   int rank = comm->rank();
   int size = comm->size();
@@ -441,9 +451,11 @@ int Colls::ireduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype d
   if (size <= 0)
     return MPI_ERR_COMM;
 
+  unsigned char* tmp_sendbuf = nullptr;
   if( sendbuf == MPI_IN_PLACE ) {
-    sendtmpbuf = static_cast<void *>(smpi_get_tmp_sendbuffer(count*datatype->get_extent()));
-    Datatype::copy(recvbuf, count, datatype,sendtmpbuf, count, datatype);
+    tmp_sendbuf = smpi_get_tmp_sendbuffer(count * datatype->get_extent());
+    Datatype::copy(recvbuf, count, datatype, tmp_sendbuf, count, datatype);
+    real_sendbuf = tmp_sendbuf;
   }
 
   if(rank == root){
@@ -457,13 +469,13 @@ int Colls::ireduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype d
   if(rank != root) {
     // Send buffer to root
     MPI_Request* requests = new MPI_Request[1];
-    requests[0]=Request::isend(sendtmpbuf, count, datatype, root, system_tag, comm);
+    requests[0]           = Request::isend(real_sendbuf, count, datatype, root, system_tag, comm);
     (*request)->set_nbc_requests(requests, 1);
   } else {
     datatype->extent(&lb, &dataext);
     // Local copy from root
-    if (sendtmpbuf != nullptr && recvbuf != nullptr)
-      Datatype::copy(sendtmpbuf, count, datatype, recvbuf, count, datatype);
+    if (real_sendbuf != nullptr && recvbuf != nullptr)
+      Datatype::copy(real_sendbuf, count, datatype, recvbuf, count, datatype);
     // Receive buffers from senders
     MPI_Request *requests = new MPI_Request[size - 1];
     int index = 0;
@@ -479,16 +491,16 @@ int Colls::ireduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype d
     (*request)->set_nbc_requests(requests, size - 1);
   }    
   if( sendbuf == MPI_IN_PLACE ) {
-    smpi_free_tmp_buffer(sendtmpbuf);
+    smpi_free_tmp_buffer(tmp_sendbuf);
   }
   return MPI_SUCCESS;
 }
 
-int Colls::iallreduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype,
-                      MPI_Op op, MPI_Comm comm, MPI_Request* request)
+int colls::iallreduce(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm,
+                      MPI_Request* request, int external)
 {
 
-  const int system_tag = COLL_TAG_ALLREDUCE;
+  const int system_tag = COLL_TAG_ALLREDUCE-external;
   MPI_Aint lb = 0;
   MPI_Aint dataext = 0;
 
@@ -517,9 +529,10 @@ int Colls::iallreduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatyp
   return MPI_SUCCESS;
 }
 
-int Colls::iscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request* request)
+int colls::iscan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm,
+                 MPI_Request* request, int external)
 {
-  int system_tag = -888;
+  int system_tag = -888-external;
   MPI_Aint lb      = 0;
   MPI_Aint dataext = 0;
 
@@ -549,9 +562,10 @@ int Colls::iscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype dat
   return MPI_SUCCESS;
 }
 
-int Colls::iexscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request* request)
+int colls::iexscan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm,
+                   MPI_Request* request, int external)
 {
-  int system_tag = -888;
+  int system_tag = -888-external;
   MPI_Aint lb         = 0;
   MPI_Aint dataext    = 0;
   int rank = comm->rank();
@@ -579,10 +593,11 @@ int Colls::iexscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype d
   return MPI_SUCCESS;
 }
 
-int Colls::ireduce_scatter(const void *sendbuf, void *recvbuf, const int *recvcounts, MPI_Datatype datatype, MPI_Op op,
-                             MPI_Comm comm, MPI_Request* request){
-//Version where each process performs the reduce for its own part. Alltoall pattern for comms.
-  const int system_tag = COLL_TAG_REDUCE_SCATTER;
+int colls::ireduce_scatter(const void* sendbuf, void* recvbuf, const int* recvcounts, MPI_Datatype datatype, MPI_Op op,
+                           MPI_Comm comm, MPI_Request* request, int external)
+{
+  // Version where each process performs the reduce for its own part. Alltoall pattern for comms.
+  const int system_tag = COLL_TAG_REDUCE_SCATTER-external;
   MPI_Aint lb = 0;
   MPI_Aint dataext = 0;