X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/bc93975e89c1b7d29f7247e4700fca9277895e2f..9d1a9a4b871895531b7e70f313691ef75dc47a96:/src/smpi/bindings/smpi_pmpi_coll.cpp diff --git a/src/smpi/bindings/smpi_pmpi_coll.cpp b/src/smpi/bindings/smpi_pmpi_coll.cpp index ce321238a3..e4a607654d 100644 --- a/src/smpi/bindings/smpi_pmpi_coll.cpp +++ b/src/smpi/bindings/smpi_pmpi_coll.cpp @@ -1,4 +1,4 @@ -/* Copyright (c) 2007-2018. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2007-2019. 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. */ @@ -6,672 +6,926 @@ #include "private.hpp" #include "smpi_coll.hpp" #include "smpi_comm.hpp" +#include "smpi_request.hpp" #include "smpi_datatype_derived.hpp" #include "smpi_op.hpp" -#include "smpi_process.hpp" +#include "src/smpi/include/smpi_actor.hpp" XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(smpi_pmpi); -/* PMPI User level calls */ - -int PMPI_Bcast(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm) -{ - int retval = 0; - - smpi_bench_end(); +#define CHECK_ARGS(test, errcode, ...) \ + if (test) { \ + XBT_WARN(__VA_ARGS__); \ + return (errcode); \ + } - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if (not datatype->is_valid()) { - retval = MPI_ERR_ARG; - } else { - int rank = simgrid::s4u::this_actor::get_pid(); - TRACE_smpi_comm_in(rank, __func__, - new simgrid::instr::CollTIData("bcast", root, -1.0, - datatype->is_replayable() ? count : count * datatype->size(), -1, - simgrid::smpi::Datatype::encode(datatype), "")); - if (comm->size() > 1) - simgrid::smpi::Colls::bcast(buf, count, datatype, root, comm); - retval = MPI_SUCCESS; - - TRACE_smpi_comm_out(rank); + static const void* smpi_get_in_place_buf(const void* inplacebuf, const void* otherbuf,std::unique_ptr& tmp_sendbuf, int count, MPI_Datatype datatype){ + if (inplacebuf == MPI_IN_PLACE) { + tmp_sendbuf.reset(new unsigned char[count * datatype->get_extent()]); + simgrid::smpi::Datatype::copy(otherbuf, count, datatype, tmp_sendbuf.get(), count, datatype); + return tmp_sendbuf.get(); + }else{ + return inplacebuf; } - smpi_bench_begin(); - return retval; } +/* PMPI User level calls */ int PMPI_Barrier(MPI_Comm comm) { - int retval = 0; - - smpi_bench_end(); - - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else { - int rank = simgrid::s4u::this_actor::get_pid(); - TRACE_smpi_comm_in(rank, __func__, new simgrid::instr::NoOpTIData("barrier")); + return PMPI_Ibarrier(comm, MPI_REQUEST_IGNORED); +} - simgrid::smpi::Colls::barrier(comm); +int PMPI_Ibarrier(MPI_Comm comm, MPI_Request *request) +{ + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if (request == nullptr) + return MPI_ERR_ARG; - //Barrier can be used to synchronize RMA calls. Finish all requests from comm before. + smpi_bench_end(); + int rank = simgrid::s4u::this_actor::get_pid(); + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Barrier" : "PMPI_Ibarrier", + new simgrid::instr::NoOpTIData(request == MPI_REQUEST_IGNORED ? "barrier" : "ibarrier")); + if (request == MPI_REQUEST_IGNORED) { + simgrid::smpi::colls::barrier(comm); + // Barrier can be used to synchronize RMA calls. Finish all requests from comm before. comm->finish_rma_calls(); + } else + simgrid::smpi::colls::ibarrier(comm, request); - retval = MPI_SUCCESS; - - TRACE_smpi_comm_out(rank); - } - + TRACE_smpi_comm_out(rank); smpi_bench_begin(); - return retval; + return MPI_SUCCESS; +} + +int PMPI_Bcast(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm) +{ + return PMPI_Ibcast(buf, count, datatype, root, comm, MPI_REQUEST_IGNORED); } -int PMPI_Gather(void *sendbuf, int sendcount, MPI_Datatype sendtype,void *recvbuf, int recvcount, MPI_Datatype recvtype, - int root, MPI_Comm comm) +int PMPI_Ibcast(void *buf, int count, MPI_Datatype datatype, + int root, MPI_Comm comm, MPI_Request* request) { - int retval = 0; + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if (buf == nullptr && count > 0) + return MPI_ERR_BUFFER; + if (datatype == MPI_DATATYPE_NULL || not datatype->is_valid()) + return MPI_ERR_TYPE; + if (count < 0) + return MPI_ERR_COUNT; + if (root < 0 || root >= comm->size()) + return MPI_ERR_ROOT; + if (request == nullptr) + return MPI_ERR_ARG; smpi_bench_end(); - - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if ((( sendbuf != MPI_IN_PLACE) && (sendtype == MPI_DATATYPE_NULL)) || - ((comm->rank() == root) && (recvtype == MPI_DATATYPE_NULL))){ - retval = MPI_ERR_TYPE; - } else if ((( sendbuf != MPI_IN_PLACE) && (sendcount <0)) || ((comm->rank() == root) && (recvcount <0))){ - retval = MPI_ERR_COUNT; + int rank = simgrid::s4u::this_actor::get_pid(); + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Bcast" : "PMPI_Ibcast", + new simgrid::instr::CollTIData(request == MPI_REQUEST_IGNORED ? "bcast" : "ibcast", root, -1.0, + datatype->is_replayable() ? count : count * datatype->size(), -1, + simgrid::smpi::Datatype::encode(datatype), "")); + if (comm->size() > 1) { + if (request == MPI_REQUEST_IGNORED) + simgrid::smpi::colls::bcast(buf, count, datatype, root, comm); + else + simgrid::smpi::colls::ibcast(buf, count, datatype, root, comm, request); } else { - - char* sendtmpbuf = static_cast(sendbuf); - int sendtmpcount = sendcount; - MPI_Datatype sendtmptype = sendtype; - if( (comm->rank() == root) && (sendbuf == MPI_IN_PLACE )) { - sendtmpcount=0; - sendtmptype=recvtype; - } - int rank = simgrid::s4u::this_actor::get_pid(); - - TRACE_smpi_comm_in( - rank, __func__, - new simgrid::instr::CollTIData( - "gather", root, -1.0, sendtmptype->is_replayable() ? sendtmpcount : sendtmpcount * sendtmptype->size(), - (comm->rank() != root || recvtype->is_replayable()) ? recvcount : recvcount * recvtype->size(), - simgrid::smpi::Datatype::encode(sendtmptype), simgrid::smpi::Datatype::encode(recvtype))); - - simgrid::smpi::Colls::gather(sendtmpbuf, sendtmpcount, sendtmptype, recvbuf, recvcount, recvtype, root, comm); - - retval = MPI_SUCCESS; - TRACE_smpi_comm_out(rank); + if (request != MPI_REQUEST_IGNORED) + *request = MPI_REQUEST_NULL; } + TRACE_smpi_comm_out(rank); smpi_bench_begin(); - return retval; + return MPI_SUCCESS; } -int PMPI_Gatherv(void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int *recvcounts, int *displs, - MPI_Datatype recvtype, int root, MPI_Comm comm) +int PMPI_Gather(const void *sendbuf, int sendcount, MPI_Datatype sendtype,void *recvbuf, int recvcount, MPI_Datatype recvtype, + int root, MPI_Comm comm){ + return PMPI_Igather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm, MPI_REQUEST_IGNORED); +} + +int PMPI_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 retval = 0; + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if ((sendbuf == nullptr && sendcount > 0) || ((comm->rank() == root) && recvbuf == nullptr && recvcount > 0)) + return MPI_ERR_BUFFER; + if (((sendbuf != MPI_IN_PLACE && sendcount > 0) && (sendtype == MPI_DATATYPE_NULL)) || + ((comm->rank() == root) && (recvtype == MPI_DATATYPE_NULL))) + return MPI_ERR_TYPE; + if (((sendbuf != MPI_IN_PLACE) && (sendcount < 0)) || ((comm->rank() == root) && (recvcount < 0))) + return MPI_ERR_COUNT; + if (root < 0 || root >= comm->size()) + return MPI_ERR_ROOT; + if (request == nullptr) + return MPI_ERR_ARG; smpi_bench_end(); + const void* real_sendbuf = sendbuf; + int real_sendcount = sendcount; + MPI_Datatype real_sendtype = sendtype; + if ((comm->rank() == root) && (sendbuf == MPI_IN_PLACE)) { + real_sendcount = 0; + real_sendtype = recvtype; + } + int rank = simgrid::s4u::this_actor::get_pid(); + + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Gather" : "PMPI_Igather", + new simgrid::instr::CollTIData( + request == MPI_REQUEST_IGNORED ? "gather" : "igather", root, -1.0, + real_sendtype->is_replayable() ? real_sendcount : real_sendcount * real_sendtype->size(), + (comm->rank() != root || recvtype->is_replayable()) ? recvcount : recvcount * recvtype->size(), + simgrid::smpi::Datatype::encode(real_sendtype), simgrid::smpi::Datatype::encode(recvtype))); + if (request == MPI_REQUEST_IGNORED) + simgrid::smpi::colls::gather(real_sendbuf, real_sendcount, real_sendtype, recvbuf, recvcount, recvtype, root, comm); + else + simgrid::smpi::colls::igather(real_sendbuf, real_sendcount, real_sendtype, recvbuf, recvcount, recvtype, root, comm, + request); + + TRACE_smpi_comm_out(rank); + smpi_bench_begin(); + return MPI_SUCCESS; +} - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if ((( sendbuf != MPI_IN_PLACE) && (sendtype == MPI_DATATYPE_NULL)) || - ((comm->rank() == root) && (recvtype == MPI_DATATYPE_NULL))){ - retval = MPI_ERR_TYPE; - } else if (( sendbuf != MPI_IN_PLACE) && (sendcount <0)){ - retval = MPI_ERR_COUNT; - } else if ((comm->rank() == root) && (recvcounts == nullptr || displs == nullptr)) { - retval = MPI_ERR_ARG; - } else { - char* sendtmpbuf = static_cast(sendbuf); - int sendtmpcount = sendcount; - MPI_Datatype sendtmptype = sendtype; - if( (comm->rank() == root) && (sendbuf == MPI_IN_PLACE )) { - sendtmpcount=0; - sendtmptype=recvtype; - } +int PMPI_Gatherv(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, const int *recvcounts, const int *displs, + MPI_Datatype recvtype, int root, MPI_Comm comm){ + return PMPI_Igatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, root, comm, MPI_REQUEST_IGNORED); +} - int rank = simgrid::s4u::this_actor::get_pid(); - int dt_size_recv = recvtype->is_replayable() ? 1 : recvtype->size(); +int PMPI_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) +{ + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if ((sendbuf == nullptr && sendcount > 0) || ((comm->rank() == root) && recvbuf == nullptr)) + return MPI_ERR_BUFFER; + if (((sendbuf != MPI_IN_PLACE) && (sendtype == MPI_DATATYPE_NULL)) || + ((comm->rank() == root) && (recvtype == MPI_DATATYPE_NULL))) + return MPI_ERR_TYPE; + if ((sendbuf != MPI_IN_PLACE) && (sendcount < 0)) + return MPI_ERR_COUNT; + if ((comm->rank() == root) && (recvcounts == nullptr || displs == nullptr)) + return MPI_ERR_ARG; + if (root < 0 || root >= comm->size()) + return MPI_ERR_ROOT; + if (request == nullptr) + return MPI_ERR_ARG; + + for (int i = 0; i < comm->size(); i++) { + if ((comm->rank() == root) && (recvcounts[i] < 0)) + return MPI_ERR_COUNT; + } - std::vector* trace_recvcounts = new std::vector; - if (comm->rank() == root) { - for (int i = 0; i < comm->size(); i++) // copy data to avoid bad free - trace_recvcounts->push_back(recvcounts[i] * dt_size_recv); - } + smpi_bench_end(); + const void* real_sendbuf = sendbuf; + int real_sendcount = sendcount; + MPI_Datatype real_sendtype = sendtype; + if ((comm->rank() == root) && (sendbuf == MPI_IN_PLACE)) { + real_sendcount = 0; + real_sendtype = recvtype; + } - TRACE_smpi_comm_in(rank, __func__, - new simgrid::instr::VarCollTIData( - "gatherV", root, - sendtmptype->is_replayable() ? sendtmpcount : sendtmpcount * sendtmptype->size(), nullptr, - dt_size_recv, trace_recvcounts, simgrid::smpi::Datatype::encode(sendtmptype), - simgrid::smpi::Datatype::encode(recvtype))); + int rank = simgrid::s4u::this_actor::get_pid(); + int dt_size_recv = recvtype->is_replayable() ? 1 : recvtype->size(); - retval = simgrid::smpi::Colls::gatherv(sendtmpbuf, sendtmpcount, sendtmptype, recvbuf, recvcounts, displs, recvtype, root, comm); - TRACE_smpi_comm_out(rank); + std::vector* trace_recvcounts = new std::vector; + if (comm->rank() == root) { + for (int i = 0; i < comm->size(); i++) // copy data to avoid bad free + trace_recvcounts->push_back(recvcounts[i] * dt_size_recv); } + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Gatherv" : "PMPI_Igatherv", + new simgrid::instr::VarCollTIData( + request == MPI_REQUEST_IGNORED ? "gatherv" : "igatherv", root, + real_sendtype->is_replayable() ? real_sendcount : real_sendcount * real_sendtype->size(), + nullptr, dt_size_recv, trace_recvcounts, simgrid::smpi::Datatype::encode(real_sendtype), + simgrid::smpi::Datatype::encode(recvtype))); + if (request == MPI_REQUEST_IGNORED) + simgrid::smpi::colls::gatherv(real_sendbuf, real_sendcount, real_sendtype, recvbuf, recvcounts, displs, recvtype, + root, comm); + else + simgrid::smpi::colls::igatherv(real_sendbuf, real_sendcount, real_sendtype, recvbuf, recvcounts, displs, recvtype, + root, comm, request); + + TRACE_smpi_comm_out(rank); smpi_bench_begin(); - return retval; + return MPI_SUCCESS; +} + +int PMPI_Allgather(const void *sendbuf, int sendcount, MPI_Datatype sendtype, + void *recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm){ + return PMPI_Iallgather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm, MPI_REQUEST_IGNORED); } -int PMPI_Allgather(void *sendbuf, int sendcount, MPI_Datatype sendtype, - void *recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm) +int PMPI_Iallgather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, + MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request) { - int retval = MPI_SUCCESS; + CHECK_ARGS(comm == MPI_COMM_NULL, MPI_ERR_COMM, "Iallgather: the communicator cannot be MPI_COMM_NULL"); + CHECK_ARGS(recvbuf == nullptr && recvcount > 0, MPI_ERR_BUFFER, "Iallgather: param 4 recvbuf cannot be NULL"); + CHECK_ARGS(sendbuf == nullptr && sendcount > 0, MPI_ERR_BUFFER, + "Iallgather: param 1 sendbuf cannot be NULL when sendcound > 0"); + CHECK_ARGS((sendbuf != MPI_IN_PLACE) && (sendtype == MPI_DATATYPE_NULL), MPI_ERR_TYPE, + "Iallgather: param 3 sendtype cannot be MPI_DATATYPE_NULL when sendbuff is not MPI_IN_PLACE"); + CHECK_ARGS(recvtype == MPI_DATATYPE_NULL, MPI_ERR_TYPE, "Iallgather: param 6 recvtype cannot be MPI_DATATYPE_NULL"); + CHECK_ARGS(recvcount < 0, MPI_ERR_COUNT, "Iallgather: param 5 recvcount cannot be negative"); + CHECK_ARGS((sendbuf != MPI_IN_PLACE) && (sendcount < 0), MPI_ERR_COUNT, + "Iallgather: param 2 sendcount cannot be negative when sendbuf is not MPI_IN_PLACE"); + CHECK_ARGS(request == nullptr, MPI_ERR_ARG, "Iallgather: param 8 request cannot be NULL"); smpi_bench_end(); - - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if ((( sendbuf != MPI_IN_PLACE) && (sendtype == MPI_DATATYPE_NULL)) || - (recvtype == MPI_DATATYPE_NULL)){ - retval = MPI_ERR_TYPE; - } else if ((( sendbuf != MPI_IN_PLACE) && (sendcount <0)) || - (recvcount <0)){ - retval = MPI_ERR_COUNT; - } else { - if(sendbuf == MPI_IN_PLACE) { - sendbuf=static_cast(recvbuf)+recvtype->get_extent()*recvcount*comm->rank(); - sendcount=recvcount; - sendtype=recvtype; - } - int rank = simgrid::s4u::this_actor::get_pid(); - - TRACE_smpi_comm_in(rank, __func__, - new simgrid::instr::CollTIData( - "allGather", -1, -1.0, sendtype->is_replayable() ? sendcount : sendcount * sendtype->size(), - recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), - simgrid::smpi::Datatype::encode(sendtype), simgrid::smpi::Datatype::encode(recvtype))); - - simgrid::smpi::Colls::allgather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm); - TRACE_smpi_comm_out(rank); + if (sendbuf == MPI_IN_PLACE) { + sendbuf = static_cast(recvbuf) + recvtype->get_extent() * recvcount * comm->rank(); + sendcount = recvcount; + sendtype = recvtype; } + int rank = simgrid::s4u::this_actor::get_pid(); + + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Allgather" : "PMPI_Iallggather", + new simgrid::instr::CollTIData( + request == MPI_REQUEST_IGNORED ? "allgather" : "iallgather", -1, -1.0, + sendtype->is_replayable() ? sendcount : sendcount * sendtype->size(), + recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), + simgrid::smpi::Datatype::encode(sendtype), simgrid::smpi::Datatype::encode(recvtype))); + if (request == MPI_REQUEST_IGNORED) + simgrid::smpi::colls::allgather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm); + else + simgrid::smpi::colls::iallgather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm, request); + + TRACE_smpi_comm_out(rank); smpi_bench_begin(); - return retval; + return MPI_SUCCESS; } -int PMPI_Allgatherv(void *sendbuf, int sendcount, MPI_Datatype sendtype, - void *recvbuf, int *recvcounts, int *displs, MPI_Datatype recvtype, MPI_Comm comm) +int PMPI_Allgatherv(const void *sendbuf, int sendcount, MPI_Datatype sendtype, + void *recvbuf, const int *recvcounts, const int *displs, MPI_Datatype recvtype, MPI_Comm comm){ + return PMPI_Iallgatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, comm, MPI_REQUEST_IGNORED); +} + +int PMPI_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 retval = 0; + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if (sendbuf == nullptr && sendcount > 0) + return MPI_ERR_BUFFER; + if (((sendbuf != MPI_IN_PLACE) && (sendtype == MPI_DATATYPE_NULL)) || (recvtype == MPI_DATATYPE_NULL)) + return MPI_ERR_TYPE; + if ((sendbuf != MPI_IN_PLACE) && (sendcount < 0)) + return MPI_ERR_COUNT; + if (recvcounts == nullptr || displs == nullptr) + return MPI_ERR_ARG; + if (request == nullptr) + return MPI_ERR_ARG; + + for (int i = 0; i < comm->size(); i++) { + if (recvcounts[i] < 0) + return MPI_ERR_COUNT; + else if (recvcounts[i] > 0 && recvbuf == nullptr) + return MPI_ERR_BUFFER; + } smpi_bench_end(); + if (sendbuf == MPI_IN_PLACE) { + sendbuf = static_cast(recvbuf) + recvtype->get_extent() * displs[comm->rank()]; + sendcount = recvcounts[comm->rank()]; + sendtype = recvtype; + } + int rank = simgrid::s4u::this_actor::get_pid(); + int dt_size_recv = recvtype->is_replayable() ? 1 : recvtype->size(); - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if (((sendbuf != MPI_IN_PLACE) && (sendtype == MPI_DATATYPE_NULL)) || (recvtype == MPI_DATATYPE_NULL)) { - retval = MPI_ERR_TYPE; - } else if (( sendbuf != MPI_IN_PLACE) && (sendcount <0)){ - retval = MPI_ERR_COUNT; - } else if (recvcounts == nullptr || displs == nullptr) { - retval = MPI_ERR_ARG; - } else { - - if(sendbuf == MPI_IN_PLACE) { - sendbuf=static_cast(recvbuf)+recvtype->get_extent()*displs[comm->rank()]; - sendcount=recvcounts[comm->rank()]; - sendtype=recvtype; - } - int rank = simgrid::s4u::this_actor::get_pid(); - int dt_size_recv = recvtype->is_replayable() ? 1 : recvtype->size(); - - std::vector* trace_recvcounts = new std::vector; - for (int i = 0; i < comm->size(); i++) // copy data to avoid bad free - trace_recvcounts->push_back(recvcounts[i] * dt_size_recv); - - TRACE_smpi_comm_in(rank, __func__, - new simgrid::instr::VarCollTIData( - "allGatherV", -1, sendtype->is_replayable() ? sendcount : sendcount * sendtype->size(), - nullptr, dt_size_recv, trace_recvcounts, simgrid::smpi::Datatype::encode(sendtype), - simgrid::smpi::Datatype::encode(recvtype))); - - simgrid::smpi::Colls::allgatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, comm); - retval = MPI_SUCCESS; - TRACE_smpi_comm_out(rank); + std::vector* trace_recvcounts = new std::vector; + for (int i = 0; i < comm->size(); i++) { // copy data to avoid bad free + trace_recvcounts->push_back(recvcounts[i] * dt_size_recv); } + TRACE_smpi_comm_in( + rank, request == MPI_REQUEST_IGNORED ? "PMPI_Allgatherv" : "PMPI_Iallgatherv", + new simgrid::instr::VarCollTIData(request == MPI_REQUEST_IGNORED ? "allgatherv" : "iallgatherv", -1, + sendtype->is_replayable() ? sendcount : sendcount * sendtype->size(), nullptr, + dt_size_recv, trace_recvcounts, simgrid::smpi::Datatype::encode(sendtype), + simgrid::smpi::Datatype::encode(recvtype))); + if (request == MPI_REQUEST_IGNORED) + simgrid::smpi::colls::allgatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, comm); + else + simgrid::smpi::colls::iallgatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, comm, + request); + + TRACE_smpi_comm_out(rank); smpi_bench_begin(); - return retval; + return MPI_SUCCESS; +} + +int PMPI_Scatter(const void *sendbuf, int sendcount, MPI_Datatype sendtype, + void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm){ + return PMPI_Iscatter(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm, MPI_REQUEST_IGNORED); } -int PMPI_Scatter(void *sendbuf, int sendcount, MPI_Datatype sendtype, - void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) +int PMPI_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 retval = 0; + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if (((comm->rank() == root) && (sendtype == MPI_DATATYPE_NULL || not sendtype->is_valid())) || + ((recvbuf != MPI_IN_PLACE) && (recvtype == MPI_DATATYPE_NULL || not recvtype->is_valid()))) + return MPI_ERR_TYPE; + if (((comm->rank() == root) && (sendcount < 0)) || ((recvbuf != MPI_IN_PLACE) && (recvcount < 0))) + return MPI_ERR_COUNT; + if ((sendbuf == recvbuf) || ((comm->rank() == root) && sendcount > 0 && (sendbuf == nullptr)) || + (recvcount > 0 && recvbuf == nullptr)) + return MPI_ERR_BUFFER; + if (root < 0 || root >= comm->size()) + return MPI_ERR_ROOT; + if (request == nullptr) + return MPI_ERR_ARG; smpi_bench_end(); - - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if (((comm->rank() == root) && (not sendtype->is_valid())) || - ((recvbuf != MPI_IN_PLACE) && (not recvtype->is_valid()))) { - retval = MPI_ERR_TYPE; - } else if ((sendbuf == recvbuf) || - ((comm->rank()==root) && sendcount>0 && (sendbuf == nullptr))){ - retval = MPI_ERR_BUFFER; - }else { - - if (recvbuf == MPI_IN_PLACE) { - recvtype = sendtype; - recvcount = sendcount; - } - int rank = simgrid::s4u::this_actor::get_pid(); - - TRACE_smpi_comm_in( - rank, __func__, - new simgrid::instr::CollTIData( - "scatter", root, -1.0, - (comm->rank() != root || sendtype->is_replayable()) ? sendcount : sendcount * sendtype->size(), - recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), - simgrid::smpi::Datatype::encode(sendtype), simgrid::smpi::Datatype::encode(recvtype))); - - simgrid::smpi::Colls::scatter(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm); - retval = MPI_SUCCESS; - TRACE_smpi_comm_out(rank); + if (recvbuf == MPI_IN_PLACE) { + recvtype = sendtype; + recvcount = sendcount; } - + int rank = simgrid::s4u::this_actor::get_pid(); + + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Scatter" : "PMPI_Iscatter", + new simgrid::instr::CollTIData( + request == MPI_REQUEST_IGNORED ? "scatter" : "iscatter", root, -1.0, + (comm->rank() != root || sendtype->is_replayable()) ? sendcount : sendcount * sendtype->size(), + recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), + simgrid::smpi::Datatype::encode(sendtype), simgrid::smpi::Datatype::encode(recvtype))); + if (request == MPI_REQUEST_IGNORED) + simgrid::smpi::colls::scatter(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm); + else + simgrid::smpi::colls::iscatter(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm, request); + + TRACE_smpi_comm_out(rank); smpi_bench_begin(); - return retval; + return MPI_SUCCESS; } -int PMPI_Scatterv(void *sendbuf, int *sendcounts, int *displs, - MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) -{ - int retval = 0; - - smpi_bench_end(); +int PMPI_Scatterv(const void *sendbuf, const int *sendcounts, const int *displs, + MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm){ + return PMPI_Iscatterv(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root, comm, MPI_REQUEST_IGNORED); +} - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if (sendcounts == nullptr || displs == nullptr) { - retval = MPI_ERR_ARG; - } else if (((comm->rank() == root) && (sendtype == MPI_DATATYPE_NULL)) || - ((recvbuf != MPI_IN_PLACE) && (recvtype == MPI_DATATYPE_NULL))) { - retval = MPI_ERR_TYPE; - } else { +int PMPI_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) +{ + CHECK_ARGS(comm == MPI_COMM_NULL, MPI_ERR_COMM, "Iscatterv: the communicator cannot be MPI_COMM_NULL"); + CHECK_ARGS((comm->rank() == root) && (sendcounts == nullptr), MPI_ERR_ARG, + "Iscatterv: param 2 sendcounts cannot be NULL on the root rank"); + CHECK_ARGS((comm->rank() == root) && (displs == nullptr), MPI_ERR_ARG, + "Iscatterv: param 3 displs cannot be NULL on the root rank"); + CHECK_ARGS((comm->rank() == root) && (sendtype == MPI_DATATYPE_NULL), MPI_ERR_TYPE, + "Iscatterv: The sendtype cannot be NULL on the root rank"); + CHECK_ARGS((recvbuf != MPI_IN_PLACE) && (recvtype == MPI_DATATYPE_NULL), MPI_ERR_TYPE, + "Iscatterv: the recvtype cannot be NULL when not receiving in place"); + CHECK_ARGS(request == nullptr, MPI_ERR_ARG, "Iscatterv: param 10 request cannot be NULL"); + CHECK_ARGS(recvbuf != MPI_IN_PLACE && recvcount < 0, MPI_ERR_COUNT, + "Iscatterv: When not receiving in place, the recvcount cannot be negative"); + CHECK_ARGS(root < 0, MPI_ERR_ROOT, "Iscatterv: root cannot be negative"); + CHECK_ARGS(root >= comm->size(), MPI_ERR_ROOT, "Iscatterv: root (=%d) is larger than communicator size (=%d)", root, + comm->size()); + + if (comm->rank() == root) { if (recvbuf == MPI_IN_PLACE) { recvtype = sendtype; recvcount = sendcounts[comm->rank()]; } - int rank = simgrid::s4u::this_actor::get_pid(); - int dt_size_send = sendtype->is_replayable() ? 1 : sendtype->size(); - - std::vector* trace_sendcounts = new std::vector; - if (comm->rank() == root) { - for (int i = 0; i < comm->size(); i++) // copy data to avoid bad free - trace_sendcounts->push_back(sendcounts[i] * dt_size_send); - } + for (int i = 0; i < comm->size(); i++) + CHECK_ARGS(sendcounts[i] < 0, MPI_ERR_COUNT, "Iscatterv: sendcounts[%d]=%d but this cannot be negative", i, + sendcounts[i]); + } - TRACE_smpi_comm_in(rank, __func__, - new simgrid::instr::VarCollTIData( - "scatterV", root, dt_size_send, trace_sendcounts, - recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), nullptr, - simgrid::smpi::Datatype::encode(sendtype), simgrid::smpi::Datatype::encode(recvtype))); + smpi_bench_end(); - retval = simgrid::smpi::Colls::scatterv(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root, comm); + int rank = simgrid::s4u::this_actor::get_pid(); + int dt_size_send = sendtype->is_replayable() ? 1 : sendtype->size(); - TRACE_smpi_comm_out(rank); + std::vector* trace_sendcounts = new std::vector; + if (comm->rank() == root) { + for (int i = 0; i < comm->size(); i++) { // copy data to avoid bad free + trace_sendcounts->push_back(sendcounts[i] * dt_size_send); + } } + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Scatterv" : "PMPI_Iscatterv", + new simgrid::instr::VarCollTIData( + request == MPI_REQUEST_IGNORED ? "scatterv" : "iscatterv", root, dt_size_send, + trace_sendcounts, recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), + nullptr, simgrid::smpi::Datatype::encode(sendtype), + simgrid::smpi::Datatype::encode(recvtype))); + if (request == MPI_REQUEST_IGNORED) + simgrid::smpi::colls::scatterv(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root, comm); + else + simgrid::smpi::colls::iscatterv(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root, comm, + request); + + TRACE_smpi_comm_out(rank); smpi_bench_begin(); - return retval; + return MPI_SUCCESS; } -int PMPI_Reduce(void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm) +int PMPI_Reduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm) { - int retval = 0; - - smpi_bench_end(); - - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if (not datatype->is_valid() || op == MPI_OP_NULL) { - retval = MPI_ERR_ARG; - } else { - int rank = simgrid::s4u::this_actor::get_pid(); - - TRACE_smpi_comm_in(rank, __func__, - new simgrid::instr::CollTIData("reduce", root, 0, - datatype->is_replayable() ? count : count * datatype->size(), -1, - simgrid::smpi::Datatype::encode(datatype), "")); - - simgrid::smpi::Colls::reduce(sendbuf, recvbuf, count, datatype, op, root, comm); - - retval = MPI_SUCCESS; - TRACE_smpi_comm_out(rank); - } - - smpi_bench_begin(); - return retval; + return PMPI_Ireduce(sendbuf, recvbuf, count, datatype, op, root, comm, MPI_REQUEST_IGNORED); } -int PMPI_Reduce_local(void *inbuf, void *inoutbuf, int count, MPI_Datatype datatype, MPI_Op op){ - int retval = 0; +int PMPI_Ireduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm, MPI_Request* request) +{ + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if ((sendbuf == nullptr && count > 0) || ((comm->rank() == root) && recvbuf == nullptr)) + return MPI_ERR_BUFFER; + if (datatype == MPI_DATATYPE_NULL || not datatype->is_valid()) + return MPI_ERR_TYPE; + if (op == MPI_OP_NULL) + return MPI_ERR_OP; + if (request == nullptr) + return MPI_ERR_ARG; + if (root < 0 || root >= comm->size()) + return MPI_ERR_ROOT; + if (count < 0) + return MPI_ERR_COUNT; smpi_bench_end(); - if (not datatype->is_valid() || op == MPI_OP_NULL) { - retval = MPI_ERR_ARG; - } else { - op->apply(inbuf, inoutbuf, &count, datatype); - retval = MPI_SUCCESS; - } + int rank = simgrid::s4u::this_actor::get_pid(); + + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Reduce" : "PMPI_Ireduce", + new simgrid::instr::CollTIData(request == MPI_REQUEST_IGNORED ? "reduce" : "ireduce", root, 0, + datatype->is_replayable() ? count : count * datatype->size(), -1, + simgrid::smpi::Datatype::encode(datatype), "")); + if (request == MPI_REQUEST_IGNORED) + simgrid::smpi::colls::reduce(sendbuf, recvbuf, count, datatype, op, root, comm); + else + simgrid::smpi::colls::ireduce(sendbuf, recvbuf, count, datatype, op, root, comm, request); + + TRACE_smpi_comm_out(rank); smpi_bench_begin(); - return retval; + return MPI_SUCCESS; } -int PMPI_Allreduce(void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) +int PMPI_Reduce_local(const void* inbuf, void* inoutbuf, int count, MPI_Datatype datatype, MPI_Op op) { - int retval = 0; + if (datatype == MPI_DATATYPE_NULL || not datatype->is_valid()) + return MPI_ERR_TYPE; + if (op == MPI_OP_NULL) + return MPI_ERR_OP; + if (count < 0) + return MPI_ERR_COUNT; smpi_bench_end(); + op->apply(inbuf, inoutbuf, &count, datatype); + smpi_bench_begin(); + return MPI_SUCCESS; +} - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if (not datatype->is_valid()) { - retval = MPI_ERR_TYPE; - } else if (op == MPI_OP_NULL) { - retval = MPI_ERR_OP; - } else { +int PMPI_Allreduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) +{ + return PMPI_Iallreduce(sendbuf, recvbuf, count, datatype, op, comm, MPI_REQUEST_IGNORED); +} - char* sendtmpbuf = static_cast(sendbuf); - if( sendbuf == MPI_IN_PLACE ) { - sendtmpbuf = static_cast(xbt_malloc(count*datatype->get_extent())); - simgrid::smpi::Datatype::copy(recvbuf, count, datatype,sendtmpbuf, count, datatype); - } - int rank = simgrid::s4u::this_actor::get_pid(); +int PMPI_Iallreduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request) +{ + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if ((sendbuf == nullptr && count > 0) || (recvbuf == nullptr)) + return MPI_ERR_BUFFER; + if (datatype == MPI_DATATYPE_NULL || not datatype->is_valid()) + return MPI_ERR_TYPE; + if (count < 0) + return MPI_ERR_COUNT; + if (op == MPI_OP_NULL) + return MPI_ERR_OP; + if (request == nullptr) + return MPI_ERR_ARG; - TRACE_smpi_comm_in(rank, __func__, - new simgrid::instr::CollTIData("allreduce", -1, 0, - datatype->is_replayable() ? count : count * datatype->size(), -1, - simgrid::smpi::Datatype::encode(datatype), "")); + smpi_bench_end(); + std::unique_ptr tmp_sendbuf; + const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, count, datatype); - simgrid::smpi::Colls::allreduce(sendtmpbuf, recvbuf, count, datatype, op, comm); + int rank = simgrid::s4u::this_actor::get_pid(); - if( sendbuf == MPI_IN_PLACE ) - xbt_free(sendtmpbuf); + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Allreduce" : "PMPI_Iallreduce", + new simgrid::instr::CollTIData(request == MPI_REQUEST_IGNORED ? "allreduce" : "iallreduce", -1, 0, + datatype->is_replayable() ? count : count * datatype->size(), -1, + simgrid::smpi::Datatype::encode(datatype), "")); - retval = MPI_SUCCESS; - TRACE_smpi_comm_out(rank); - } + if (request == MPI_REQUEST_IGNORED) + simgrid::smpi::colls::allreduce(real_sendbuf, recvbuf, count, datatype, op, comm); + else + simgrid::smpi::colls::iallreduce(real_sendbuf, recvbuf, count, datatype, op, comm, request); + TRACE_smpi_comm_out(rank); smpi_bench_begin(); - return retval; + return MPI_SUCCESS; } -int PMPI_Scan(void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) +int PMPI_Scan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) { - int retval = 0; + return PMPI_Iscan(sendbuf, recvbuf, count, datatype, op, comm, MPI_REQUEST_IGNORED); +} - smpi_bench_end(); +int PMPI_Iscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request* request) +{ + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if (datatype == MPI_DATATYPE_NULL || not datatype->is_valid()) + return MPI_ERR_TYPE; + if (op == MPI_OP_NULL) + return MPI_ERR_OP; + if (request == nullptr) + return MPI_ERR_ARG; + if (count < 0) + return MPI_ERR_COUNT; + if (sendbuf == nullptr || recvbuf == nullptr) + return MPI_ERR_BUFFER; - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if (not datatype->is_valid()) { - retval = MPI_ERR_TYPE; - } else if (op == MPI_OP_NULL) { - retval = MPI_ERR_OP; - } else { - int rank = simgrid::s4u::this_actor::get_pid(); - void* sendtmpbuf = sendbuf; - if (sendbuf == MPI_IN_PLACE) { - sendtmpbuf = static_cast(xbt_malloc(count * datatype->size())); - memcpy(sendtmpbuf, recvbuf, count * datatype->size()); - } - TRACE_smpi_comm_in(rank, __func__, new simgrid::instr::Pt2PtTIData( - "scan", -1, datatype->is_replayable() ? count : count * datatype->size(), - simgrid::smpi::Datatype::encode(datatype))); + smpi_bench_end(); + int rank = simgrid::s4u::this_actor::get_pid(); + std::unique_ptr tmp_sendbuf; + const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, count, datatype);; - retval = simgrid::smpi::Colls::scan(sendtmpbuf, recvbuf, count, datatype, op, comm); + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Scan" : "PMPI_Iscan", + new simgrid::instr::Pt2PtTIData(request == MPI_REQUEST_IGNORED ? "scan" : "iscan", -1, + datatype->is_replayable() ? count : count * datatype->size(), + simgrid::smpi::Datatype::encode(datatype))); - TRACE_smpi_comm_out(rank); - if (sendbuf == MPI_IN_PLACE) - xbt_free(sendtmpbuf); - } + int retval; + if (request == MPI_REQUEST_IGNORED) + retval = simgrid::smpi::colls::scan(real_sendbuf, recvbuf, count, datatype, op, comm); + else + retval = simgrid::smpi::colls::iscan(real_sendbuf, recvbuf, count, datatype, op, comm, request); + TRACE_smpi_comm_out(rank); smpi_bench_begin(); return retval; } -int PMPI_Exscan(void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm){ - int retval = 0; - - smpi_bench_end(); +int PMPI_Exscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) +{ + return PMPI_Iexscan(sendbuf, recvbuf, count, datatype, op, comm, MPI_REQUEST_IGNORED); +} - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if (not datatype->is_valid()) { - retval = MPI_ERR_TYPE; - } else if (op == MPI_OP_NULL) { - retval = MPI_ERR_OP; - } else { - int rank = simgrid::s4u::this_actor::get_pid(); - void* sendtmpbuf = sendbuf; - if (sendbuf == MPI_IN_PLACE) { - sendtmpbuf = static_cast(xbt_malloc(count * datatype->size())); - memcpy(sendtmpbuf, recvbuf, count * datatype->size()); - } +int PMPI_Iexscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request* request){ + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if (not datatype->is_valid()) + return MPI_ERR_TYPE; + if (op == MPI_OP_NULL) + return MPI_ERR_OP; + if (request == nullptr) + return MPI_ERR_ARG; + if (count < 0) + return MPI_ERR_COUNT; + if (sendbuf == nullptr || recvbuf == nullptr) + return MPI_ERR_BUFFER; - TRACE_smpi_comm_in(rank, __func__, new simgrid::instr::Pt2PtTIData( - "exscan", -1, datatype->is_replayable() ? count : count * datatype->size(), - simgrid::smpi::Datatype::encode(datatype))); + smpi_bench_end(); + int rank = simgrid::s4u::this_actor::get_pid(); + std::unique_ptr tmp_sendbuf; + const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, count, datatype);; - retval = simgrid::smpi::Colls::exscan(sendtmpbuf, recvbuf, count, datatype, op, comm); + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Exscan" : "PMPI_Iexscan", + new simgrid::instr::Pt2PtTIData(request == MPI_REQUEST_IGNORED ? "exscan" : "iexscan", -1, + datatype->is_replayable() ? count : count * datatype->size(), + simgrid::smpi::Datatype::encode(datatype))); - TRACE_smpi_comm_out(rank); - if (sendbuf == MPI_IN_PLACE) - xbt_free(sendtmpbuf); - } + int retval; + if (request == MPI_REQUEST_IGNORED) + retval = simgrid::smpi::colls::exscan(real_sendbuf, recvbuf, count, datatype, op, comm); + else + retval = simgrid::smpi::colls::iexscan(real_sendbuf, recvbuf, count, datatype, op, comm, request); + TRACE_smpi_comm_out(rank); smpi_bench_begin(); return retval; } -int PMPI_Reduce_scatter(void *sendbuf, void *recvbuf, int *recvcounts, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) +int PMPI_Reduce_scatter(const void *sendbuf, void *recvbuf, const int *recvcounts, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) { - int retval = 0; - smpi_bench_end(); - - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if (not datatype->is_valid()) { - retval = MPI_ERR_TYPE; - } else if (op == MPI_OP_NULL) { - retval = MPI_ERR_OP; - } else if (recvcounts == nullptr) { - retval = MPI_ERR_ARG; - } else { - int rank = simgrid::s4u::this_actor::get_pid(); - std::vector* trace_recvcounts = new std::vector; - int dt_send_size = datatype->is_replayable() ? 1 : datatype->size(); - int totalcount = 0; - - for (int i = 0; i < comm->size(); i++) { // copy data to avoid bad free - trace_recvcounts->push_back(recvcounts[i] * dt_send_size); - totalcount += recvcounts[i]; - } + return PMPI_Ireduce_scatter(sendbuf, recvbuf, recvcounts, datatype, op, comm, MPI_REQUEST_IGNORED); +} - void* sendtmpbuf = sendbuf; - if (sendbuf == MPI_IN_PLACE) { - sendtmpbuf = static_cast(xbt_malloc(totalcount * datatype->size())); - memcpy(sendtmpbuf, recvbuf, totalcount * datatype->size()); - } +int PMPI_Ireduce_scatter(const void *sendbuf, void *recvbuf, const int *recvcounts, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request) +{ + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if ((sendbuf == nullptr) || (recvbuf == nullptr)) + return MPI_ERR_BUFFER; + if (datatype == MPI_DATATYPE_NULL || not datatype->is_valid()) + return MPI_ERR_TYPE; + if (op == MPI_OP_NULL) + return MPI_ERR_OP; + if (recvcounts == nullptr) + return MPI_ERR_ARG; + if (request == nullptr) + return MPI_ERR_ARG; + + for (int i = 0; i < comm->size(); i++) { // copy data to avoid bad free + if (recvcounts[i] < 0) + return MPI_ERR_COUNT; + } - TRACE_smpi_comm_in(rank, __func__, new simgrid::instr::VarCollTIData( - "reduceScatter", -1, dt_send_size, nullptr, -1, trace_recvcounts, - simgrid::smpi::Datatype::encode(datatype), "")); + smpi_bench_end(); + int rank = simgrid::s4u::this_actor::get_pid(); + std::vector* trace_recvcounts = new std::vector; + int dt_send_size = datatype->is_replayable() ? 1 : datatype->size(); + int totalcount = 0; + + for (int i = 0; i < comm->size(); i++) { // copy data to avoid bad free + trace_recvcounts->push_back(recvcounts[i] * dt_send_size); + totalcount += recvcounts[i]; + } + std::unique_ptr tmp_sendbuf; + const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, totalcount, datatype);; - simgrid::smpi::Colls::reduce_scatter(sendtmpbuf, recvbuf, recvcounts, datatype, op, comm); - retval = MPI_SUCCESS; - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Reduce_scatter" : "PMPI_Ireduce_scatter", + new simgrid::instr::VarCollTIData( + request == MPI_REQUEST_IGNORED ? "reducescatter" : "ireducescatter", -1, dt_send_size, nullptr, + -1, trace_recvcounts, simgrid::smpi::Datatype::encode(datatype), "")); - if (sendbuf == MPI_IN_PLACE) - xbt_free(sendtmpbuf); - } + if (request == MPI_REQUEST_IGNORED) + simgrid::smpi::colls::reduce_scatter(real_sendbuf, recvbuf, recvcounts, datatype, op, comm); + else + simgrid::smpi::colls::ireduce_scatter(real_sendbuf, recvbuf, recvcounts, datatype, op, comm, request); + TRACE_smpi_comm_out(rank); smpi_bench_begin(); - return retval; + return MPI_SUCCESS; } -int PMPI_Reduce_scatter_block(void *sendbuf, void *recvbuf, int recvcount, +int PMPI_Reduce_scatter_block(const void *sendbuf, void *recvbuf, int recvcount, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) { - int retval; - smpi_bench_end(); - - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if (not datatype->is_valid()) { - retval = MPI_ERR_TYPE; - } else if (op == MPI_OP_NULL) { - retval = MPI_ERR_OP; - } else if (recvcount < 0) { - retval = MPI_ERR_ARG; - } else { - int count = comm->size(); - - int rank = simgrid::s4u::this_actor::get_pid(); - int dt_send_size = datatype->is_replayable() ? 1 : datatype->size(); - std::vector* trace_recvcounts = new std::vector(recvcount * dt_send_size); // copy data to avoid bad free + return PMPI_Ireduce_scatter_block(sendbuf, recvbuf, recvcount, datatype, op, comm, MPI_REQUEST_IGNORED); +} - void* sendtmpbuf = sendbuf; - if (sendbuf == MPI_IN_PLACE) { - sendtmpbuf = static_cast(xbt_malloc(recvcount * count * datatype->size())); - memcpy(sendtmpbuf, recvbuf, recvcount * count * datatype->size()); - } +int PMPI_Ireduce_scatter_block(const void* sendbuf, void* recvbuf, int recvcount, MPI_Datatype datatype, MPI_Op op, + MPI_Comm comm, MPI_Request* request) +{ + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if (not datatype->is_valid()) + return MPI_ERR_TYPE; + if (op == MPI_OP_NULL) + return MPI_ERR_OP; + if (recvcount < 0) + return MPI_ERR_ARG; + if (request == nullptr) + return MPI_ERR_ARG; - TRACE_smpi_comm_in(rank, __func__, - new simgrid::instr::VarCollTIData("reduceScatter", -1, 0, nullptr, -1, trace_recvcounts, - simgrid::smpi::Datatype::encode(datatype), "")); + smpi_bench_end(); + int count = comm->size(); + + int rank = simgrid::s4u::this_actor::get_pid(); + int dt_send_size = datatype->is_replayable() ? 1 : datatype->size(); + std::vector* trace_recvcounts = new std::vector(recvcount * dt_send_size); // copy data to avoid bad free + std::unique_ptr tmp_sendbuf; + const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, recvcount * count, datatype); + + TRACE_smpi_comm_in( + rank, request == MPI_REQUEST_IGNORED ? "PMPI_Reduce_scatter_block" : "PMPI_Ireduce_scatter_block", + new simgrid::instr::VarCollTIData(request == MPI_REQUEST_IGNORED ? "reducescatter" : "ireducescatter", -1, 0, + nullptr, -1, trace_recvcounts, simgrid::smpi::Datatype::encode(datatype), "")); + + int* recvcounts = new int[count]; + for (int i = 0; i < count; i++) + recvcounts[i] = recvcount; + if (request == MPI_REQUEST_IGNORED) + simgrid::smpi::colls::reduce_scatter(real_sendbuf, recvbuf, recvcounts, datatype, op, comm); + else + simgrid::smpi::colls::ireduce_scatter(real_sendbuf, recvbuf, recvcounts, datatype, op, comm, request); + delete[] recvcounts; + + TRACE_smpi_comm_out(rank); + smpi_bench_begin(); + return MPI_SUCCESS; +} - int* recvcounts = new int[count]; - for (int i = 0; i < count; i++) - recvcounts[i] = recvcount; - simgrid::smpi::Colls::reduce_scatter(sendtmpbuf, recvbuf, recvcounts, datatype, op, comm); - delete[] recvcounts; - retval = MPI_SUCCESS; +int PMPI_Alltoall(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, + MPI_Datatype recvtype, MPI_Comm comm){ + return PMPI_Ialltoall(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm, MPI_REQUEST_IGNORED); +} - TRACE_smpi_comm_out(rank); +int PMPI_Ialltoall(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, + MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request) +{ + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if ((sendbuf == nullptr && sendcount > 0) || (recvbuf == nullptr && recvcount > 0)) + return MPI_ERR_BUFFER; + if ((sendbuf != MPI_IN_PLACE && sendtype == MPI_DATATYPE_NULL) || recvtype == MPI_DATATYPE_NULL) + return MPI_ERR_TYPE; + if ((sendbuf != MPI_IN_PLACE && sendcount < 0) || recvcount < 0) + return MPI_ERR_COUNT; + if (request == nullptr) + return MPI_ERR_ARG; - if (sendbuf == MPI_IN_PLACE) - xbt_free(sendtmpbuf); + smpi_bench_end(); + int rank = simgrid::s4u::this_actor::get_pid(); + int real_sendcount = sendcount; + MPI_Datatype real_sendtype = sendtype; + + std::unique_ptr tmp_sendbuf; + const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, recvcount * comm->size(), recvtype); + + if (sendbuf == MPI_IN_PLACE) { + real_sendcount = recvcount; + real_sendtype = recvtype; } + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Alltoall" : "PMPI_Ialltoall", + new simgrid::instr::CollTIData( + request == MPI_REQUEST_IGNORED ? "alltoall" : "ialltoall", -1, -1.0, + real_sendtype->is_replayable() ? real_sendcount : real_sendcount * real_sendtype->size(), + recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), + simgrid::smpi::Datatype::encode(real_sendtype), simgrid::smpi::Datatype::encode(recvtype))); + int retval; + if (request == MPI_REQUEST_IGNORED) + retval = + simgrid::smpi::colls::alltoall(real_sendbuf, real_sendcount, real_sendtype, recvbuf, recvcount, recvtype, comm); + else + retval = simgrid::smpi::colls::ialltoall(real_sendbuf, real_sendcount, real_sendtype, recvbuf, recvcount, recvtype, + comm, request); + + TRACE_smpi_comm_out(rank); smpi_bench_begin(); return retval; } -int PMPI_Alltoall(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, - MPI_Datatype recvtype, MPI_Comm comm) +int PMPI_Alltoallv(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) { - int retval = 0; - smpi_bench_end(); + return PMPI_Ialltoallv(sendbuf, sendcounts, senddisps, sendtype, recvbuf, recvcounts, recvdisps, recvtype, comm, MPI_REQUEST_IGNORED); +} - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if ((sendbuf != MPI_IN_PLACE && sendtype == MPI_DATATYPE_NULL) || recvtype == MPI_DATATYPE_NULL) { - retval = MPI_ERR_TYPE; - } else { - int rank = simgrid::s4u::this_actor::get_pid(); - void* sendtmpbuf = static_cast(sendbuf); - int sendtmpcount = sendcount; - MPI_Datatype sendtmptype = sendtype; - if (sendbuf == MPI_IN_PLACE) { - sendtmpbuf = static_cast(xbt_malloc(recvcount * comm->size() * recvtype->size())); - memcpy(sendtmpbuf, recvbuf, recvcount * comm->size() * recvtype->size()); - sendtmpcount = recvcount; - sendtmptype = recvtype; - } +int PMPI_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) +{ + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if (sendbuf == nullptr || recvbuf == nullptr) + return MPI_ERR_BUFFER; + if ((sendbuf != MPI_IN_PLACE && sendtype == MPI_DATATYPE_NULL) || recvtype == MPI_DATATYPE_NULL) + return MPI_ERR_TYPE; + if ((sendbuf != MPI_IN_PLACE && (sendcounts == nullptr || senddisps == nullptr)) || recvcounts == nullptr || + recvdisps == nullptr) + return MPI_ERR_ARG; + if (request == nullptr) + return MPI_ERR_ARG; + + int rank = simgrid::s4u::this_actor::get_pid(); + int size = comm->size(); + for (int i = 0; i < size; i++) { + if (recvcounts[i] < 0 || (sendbuf != MPI_IN_PLACE && sendcounts[i] < 0)) + return MPI_ERR_COUNT; + } - TRACE_smpi_comm_in(rank, __func__, - new simgrid::instr::CollTIData( - "allToAll", -1, -1.0, - sendtmptype->is_replayable() ? sendtmpcount : sendtmpcount * sendtmptype->size(), - recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), - simgrid::smpi::Datatype::encode(sendtmptype), simgrid::smpi::Datatype::encode(recvtype))); + smpi_bench_end(); + int send_size = 0; + int recv_size = 0; + std::vector* trace_sendcounts = new std::vector; + std::vector* trace_recvcounts = new std::vector; + int dt_size_recv = recvtype->size(); + + const int* real_sendcounts = sendcounts; + const int* real_senddisps = senddisps; + MPI_Datatype real_sendtype = sendtype; + int maxsize = 0; + for (int i = 0; i < size; i++) { // copy data to avoid bad free + recv_size += recvcounts[i] * dt_size_recv; + trace_recvcounts->push_back(recvcounts[i] * dt_size_recv); + if (((recvdisps[i] + recvcounts[i]) * dt_size_recv) > maxsize) + maxsize = (recvdisps[i] + recvcounts[i]) * dt_size_recv; + } - retval = simgrid::smpi::Colls::alltoall(sendtmpbuf, sendtmpcount, sendtmptype, recvbuf, recvcount, recvtype, comm); + std::unique_ptr tmp_sendbuf; + std::unique_ptr tmp_sendcounts; + std::unique_ptr tmp_senddisps; + const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, maxsize, MPI_CHAR); + if (sendbuf == MPI_IN_PLACE) { + tmp_sendcounts.reset(new int[size]); + std::copy(recvcounts, recvcounts + size, tmp_sendcounts.get()); + real_sendcounts = tmp_sendcounts.get(); + tmp_senddisps.reset(new int[size]); + std::copy(recvdisps, recvdisps + size, tmp_senddisps.get()); + real_senddisps = tmp_senddisps.get(); + real_sendtype = recvtype; + } - TRACE_smpi_comm_out(rank); + int dt_size_send = real_sendtype->size(); - if (sendbuf == MPI_IN_PLACE) - xbt_free(sendtmpbuf); + for (int i = 0; i < size; i++) { // copy data to avoid bad free + send_size += real_sendcounts[i] * dt_size_send; + trace_sendcounts->push_back(real_sendcounts[i] * dt_size_send); } + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Alltoallv" : "PMPI_Ialltoallv", + new simgrid::instr::VarCollTIData(request == MPI_REQUEST_IGNORED ? "alltoallv" : "ialltoallv", -1, + send_size, trace_sendcounts, recv_size, trace_recvcounts, + simgrid::smpi::Datatype::encode(real_sendtype), + simgrid::smpi::Datatype::encode(recvtype))); + + int retval; + if (request == MPI_REQUEST_IGNORED) + retval = simgrid::smpi::colls::alltoallv(real_sendbuf, real_sendcounts, real_senddisps, real_sendtype, recvbuf, + recvcounts, recvdisps, recvtype, comm); + else + retval = simgrid::smpi::colls::ialltoallv(real_sendbuf, real_sendcounts, real_senddisps, real_sendtype, recvbuf, + recvcounts, recvdisps, recvtype, comm, request); + + TRACE_smpi_comm_out(rank); smpi_bench_begin(); return retval; } -int PMPI_Alltoallv(void* sendbuf, int* sendcounts, int* senddisps, MPI_Datatype sendtype, void* recvbuf, - int* recvcounts, int* recvdisps, MPI_Datatype recvtype, MPI_Comm comm) +int PMPI_Alltoallw(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) { - int retval = 0; - - smpi_bench_end(); - - if (comm == MPI_COMM_NULL) { - retval = MPI_ERR_COMM; - } else if (sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) { - retval = MPI_ERR_TYPE; - } else if ((sendbuf != MPI_IN_PLACE && (sendcounts == nullptr || senddisps == nullptr)) || recvcounts == nullptr || - recvdisps == nullptr) { - retval = MPI_ERR_ARG; - } else { - int rank = simgrid::s4u::this_actor::get_pid(); - int size = comm->size(); - int send_size = 0; - int recv_size = 0; - std::vector* trace_sendcounts = new std::vector; - std::vector* trace_recvcounts = new std::vector; - int dt_size_recv = recvtype->size(); - - void* sendtmpbuf = static_cast(sendbuf); - int* sendtmpcounts = sendcounts; - int* sendtmpdisps = senddisps; - MPI_Datatype sendtmptype = sendtype; - int maxsize = 0; - for (int i = 0; i < size; i++) { // copy data to avoid bad free - recv_size += recvcounts[i] * dt_size_recv; - trace_recvcounts->push_back(recvcounts[i] * dt_size_recv); - if (((recvdisps[i] + recvcounts[i]) * dt_size_recv) > maxsize) - maxsize = (recvdisps[i] + recvcounts[i]) * dt_size_recv; - } - - if (sendbuf == MPI_IN_PLACE) { - sendtmpbuf = static_cast(xbt_malloc(maxsize)); - memcpy(sendtmpbuf, recvbuf, maxsize); - sendtmpcounts = static_cast(xbt_malloc(size * sizeof(int))); - memcpy(sendtmpcounts, recvcounts, size * sizeof(int)); - sendtmpdisps = static_cast(xbt_malloc(size * sizeof(int))); - memcpy(sendtmpdisps, recvdisps, size * sizeof(int)); - sendtmptype = recvtype; - } + return PMPI_Ialltoallw(sendbuf, sendcounts, senddisps, sendtypes, recvbuf, recvcounts, recvdisps, recvtypes, comm, MPI_REQUEST_IGNORED); +} - int dt_size_send = sendtmptype->size(); +int PMPI_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) +{ + if (comm == MPI_COMM_NULL) + return MPI_ERR_COMM; + if (sendbuf == nullptr || recvbuf == nullptr) + return MPI_ERR_BUFFER; + if ((sendbuf != MPI_IN_PLACE && sendtypes == nullptr) || recvtypes == nullptr) + return MPI_ERR_TYPE; + if ((sendbuf != MPI_IN_PLACE && (sendcounts == nullptr || senddisps == nullptr)) || recvcounts == nullptr || + recvdisps == nullptr) + return MPI_ERR_ARG; + if (request == nullptr) + return MPI_ERR_ARG; - for (int i = 0; i < size; i++) { // copy data to avoid bad free - send_size += sendtmpcounts[i] * dt_size_send; - trace_sendcounts->push_back(sendtmpcounts[i] * dt_size_send); + smpi_bench_end(); + int rank = simgrid::s4u::this_actor::get_pid(); + int size = comm->size(); + for (int i = 0; i < size; i++) { + if (recvcounts[i] < 0 || (sendbuf != MPI_IN_PLACE && sendcounts[i] < 0)) + return MPI_ERR_COUNT; + } + int send_size = 0; + int recv_size = 0; + std::vector* trace_sendcounts = new std::vector; + std::vector* trace_recvcounts = new std::vector; + + const int* real_sendcounts = sendcounts; + const int* real_senddisps = senddisps; + const MPI_Datatype* real_sendtypes = sendtypes; + unsigned long maxsize = 0; + for (int i = 0; i < size; i++) { // copy data to avoid bad free + if (recvtypes[i] == MPI_DATATYPE_NULL) { + delete trace_recvcounts; + delete trace_sendcounts; + return MPI_ERR_TYPE; } + recv_size += recvcounts[i] * recvtypes[i]->size(); + trace_recvcounts->push_back(recvcounts[i] * recvtypes[i]->size()); + if ((recvdisps[i] + (recvcounts[i] * recvtypes[i]->size())) > maxsize) + maxsize = recvdisps[i] + (recvcounts[i] * recvtypes[i]->size()); + } - TRACE_smpi_comm_in(rank, __func__, - new simgrid::instr::VarCollTIData("allToAllV", -1, send_size, trace_sendcounts, recv_size, - trace_recvcounts, simgrid::smpi::Datatype::encode(sendtype), - simgrid::smpi::Datatype::encode(recvtype))); - - retval = simgrid::smpi::Colls::alltoallv(sendtmpbuf, sendtmpcounts, sendtmpdisps, sendtmptype, recvbuf, recvcounts, - recvdisps, recvtype, comm); - TRACE_smpi_comm_out(rank); + std::unique_ptr tmp_sendbuf; + std::unique_ptr tmp_sendcounts; + std::unique_ptr tmp_senddisps; + std::unique_ptr tmp_sendtypes; + const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, maxsize, MPI_CHAR); + if (sendbuf == MPI_IN_PLACE) { + tmp_sendcounts.reset(new int[size]); + std::copy(recvcounts, recvcounts + size, tmp_sendcounts.get()); + real_sendcounts = tmp_sendcounts.get(); + tmp_senddisps.reset(new int[size]); + std::copy(recvdisps, recvdisps + size, tmp_senddisps.get()); + real_senddisps = tmp_senddisps.get(); + tmp_sendtypes.reset(new MPI_Datatype[size]); + std::copy(recvtypes, recvtypes + size, tmp_sendtypes.get()); + real_sendtypes = tmp_sendtypes.get(); + } - if (sendbuf == MPI_IN_PLACE) { - xbt_free(sendtmpbuf); - xbt_free(sendtmpcounts); - xbt_free(sendtmpdisps); - } + for (int i = 0; i < size; i++) { // copy data to avoid bad free + send_size += real_sendcounts[i] * real_sendtypes[i]->size(); + trace_sendcounts->push_back(real_sendcounts[i] * real_sendtypes[i]->size()); } + TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Alltoallw" : "PMPI_Ialltoallw", + new simgrid::instr::VarCollTIData(request == MPI_REQUEST_IGNORED ? "alltoallv" : "ialltoallv", -1, + send_size, trace_sendcounts, recv_size, trace_recvcounts, + simgrid::smpi::Datatype::encode(real_sendtypes[0]), + simgrid::smpi::Datatype::encode(recvtypes[0]))); + + int retval; + if (request == MPI_REQUEST_IGNORED) + retval = simgrid::smpi::colls::alltoallw(real_sendbuf, real_sendcounts, real_senddisps, real_sendtypes, recvbuf, + recvcounts, recvdisps, recvtypes, comm); + else + retval = simgrid::smpi::colls::ialltoallw(real_sendbuf, real_sendcounts, real_senddisps, real_sendtypes, recvbuf, + recvcounts, recvdisps, recvtypes, comm, request); + + TRACE_smpi_comm_out(rank); smpi_bench_begin(); return retval; }