-/* Copyright (c) 2013-2014. The SimGrid Team.
+/* Copyright (c) 2013-2021. 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"
+#include "../colls_private.hpp"
-int smpi_coll_tuned_allgather_SMP_NTS(void *sbuf, int scount,
- MPI_Datatype stype, void *rbuf,
- int rcount, MPI_Datatype rtype,
- MPI_Comm comm)
+namespace simgrid{
+namespace smpi{
+
+
+int allgather__SMP_NTS(const void *sbuf, int scount,
+ MPI_Datatype stype, void *rbuf,
+ int rcount, MPI_Datatype rtype,
+ MPI_Comm comm)
{
int src, dst, comm_size, rank;
comm_size = comm->size();
int num_core_in_current_smp = num_core;
if(comm_size%num_core)
- THROWF(arg_error,0, "allgather SMP NTS algorithm can't be used with non multiple of NUM_CORE=%d number of processes ! ", num_core);
+ throw std::invalid_argument(xbt::string_printf(
+ "allgather SMP NTS algorithm can't be used with non multiple of NUM_CORE=%d number of processes!", num_core));
/* for too small number of processes, use default implementation */
if (comm_size <= num_core) {
- XBT_WARN("MPI_allgather_SMP_NTS use default MPI_allgather.");
- smpi_mpi_allgather(sbuf, scount, stype, rbuf, rcount, rtype, comm);
- return MPI_SUCCESS;
+ XBT_INFO("MPI_allgather_SMP_NTS: comm_size <= num_core, use default MPI_allgather.");
+ allgather__default(sbuf, scount, stype, rbuf, rcount, rtype, comm);
+ return MPI_SUCCESS;
}
// the last SMP node may have fewer number of running processes than all others
}
- // INTER-SMP-ALLGATHER
+ // INTER-SMP-ALLGATHER
// Every root of each SMP node post INTER-Sendrecv, then do INTRA-Bcast for each receiving message
// Use logical ring algorithm
// root of each SMP
if (intra_rank == 0) {
- MPI_Request *rrequest_array = xbt_new(MPI_Request, inter_comm_size - 1);
- MPI_Request *srequest_array = xbt_new(MPI_Request, inter_comm_size - 1);
+ auto* rrequest_array = new MPI_Request[inter_comm_size - 1];
+ auto* srequest_array = new MPI_Request[inter_comm_size - 1];
src = ((inter_rank - 1 + inter_comm_size) % inter_comm_size) * num_core;
dst = ((inter_rank + 1) % inter_comm_size) * num_core;
}
Request::waitall(inter_comm_size - 1, srequest_array, MPI_STATUSES_IGNORE);
- xbt_free(rrequest_array);
- xbt_free(srequest_array);
+ delete[] rrequest_array;
+ delete[] srequest_array;
}
// last rank of each SMP
else if (intra_rank == (num_core_in_current_smp - 1)) {
return MPI_SUCCESS;
}
+
+
+}
+}