X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/53b88680c91bba523830793ce69bb8a0fab54382..fb5dcd3589886a5f3845bbdd77fab5edbb82a842:/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 62b46a87d6..b292a42f09 100644 --- a/src/smpi/bindings/smpi_pmpi_coll.cpp +++ b/src/smpi/bindings/smpi_pmpi_coll.cpp @@ -1,4 +1,4 @@ -/* Copyright (c) 2007-2019. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2007-2021. 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. */ @@ -11,45 +11,17 @@ #include "smpi_op.hpp" #include "src/smpi/include/smpi_actor.hpp" +#include + XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(smpi_pmpi); -#define CHECK_ARGS(test, errcode, ...) \ - if (test) { \ - XBT_WARN(__VA_ARGS__); \ - return (errcode); \ - } - -#define CHECK_COMM(num)\ - CHECK_ARGS(comm == MPI_COMM_NULL, MPI_ERR_COMM,\ - "%s: param %d communicator cannot be MPI_COMM_NULL", __func__, num); -#define CHECK_REQUEST(num)\ - CHECK_ARGS(request == nullptr, MPI_ERR_ARG,\ - "%s: param %d request cannot be NULL",__func__, num); -#define CHECK_BUFFER(num,buf,count)\ - CHECK_ARGS(buf == nullptr && count > 0, MPI_ERR_BUFFER,\ - "%s: param %d %s cannot be NULL if %s > 0",__func__, num, #buf, #count); -#define CHECK_COUNT(num,count)\ - CHECK_ARGS(count < 0, MPI_ERR_COUNT,\ - "%s: param %d %s cannot be negative", __func__, num, #count); -#define CHECK_TYPE(num, datatype)\ - CHECK_ARGS((datatype == MPI_DATATYPE_NULL|| not datatype->is_valid()), MPI_ERR_TYPE,\ - "%s: param %d %s cannot be MPI_DATATYPE_NULL or invalid", __func__, num, #datatype); -#define CHECK_OP(num)\ - CHECK_ARGS(op == MPI_OP_NULL, MPI_ERR_OP,\ - "%s: param %d op cannot be MPI_OP_NULL or invalid", __func__, num); -#define CHECK_ROOT(num)\ - CHECK_ARGS((root < 0 || root >= comm->size()), MPI_ERR_ROOT,\ - "%s: param %d root (=%d) cannot be negative or larger than communicator size (=%d)", __func__, num, root,\ - comm->size()); -#define CHECK_NULL(num,err,buf)\ - CHECK_ARGS(buf == nullptr, err,\ - "%s: param %d %s cannot be NULL", __func__, num, #buf); - - static const void* smpi_get_in_place_buf(const void* inplacebuf, const void* otherbuf,std::unique_ptr& tmp_sendbuf, int count, MPI_Datatype datatype){ +static const void* smpi_get_in_place_buf(const void* inplacebuf, const void* otherbuf, + std::vector& 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(); + tmp_sendbuf.resize(count * datatype->get_extent()); + simgrid::smpi::Datatype::copy(otherbuf, count, datatype, tmp_sendbuf.data(), count, datatype); + return tmp_sendbuf.data(); }else{ return inplacebuf; } @@ -67,8 +39,8 @@ int PMPI_Ibarrier(MPI_Comm comm, MPI_Request *request) CHECK_REQUEST(2) smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Barrier" : "PMPI_Ibarrier", + aid_t pid = simgrid::s4u::this_actor::get_pid(); + TRACE_smpi_comm_in(pid, 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); @@ -77,7 +49,7 @@ int PMPI_Ibarrier(MPI_Comm comm, MPI_Request *request) } else simgrid::smpi::colls::ibarrier(comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return MPI_SUCCESS; } @@ -90,16 +62,17 @@ int PMPI_Bcast(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm c int PMPI_Ibcast(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm, MPI_Request* request) { + SET_BUF1(buf) CHECK_COMM(5) - CHECK_BUFFER(1, buf, count) CHECK_COUNT(2, count) CHECK_TYPE(3, datatype) + CHECK_BUFFER(1, buf, count, datatype) CHECK_ROOT(4) CHECK_REQUEST(6) smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Bcast" : "PMPI_Ibcast", + aid_t pid = simgrid::s4u::this_actor::get_pid(); + TRACE_smpi_comm_in(pid, 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), "")); @@ -113,7 +86,7 @@ int PMPI_Ibcast(void *buf, int count, MPI_Datatype datatype, *request = MPI_REQUEST_NULL; } - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return MPI_SUCCESS; } @@ -127,30 +100,43 @@ int PMPI_Igather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request) { CHECK_COMM(8) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) + int rank = comm->rank(); if(sendbuf != MPI_IN_PLACE){ - CHECK_BUFFER(1,sendbuf, sendcount) CHECK_COUNT(2, sendcount) CHECK_TYPE(3, sendtype) + CHECK_BUFFER(1,sendbuf, sendcount, sendtype) } - if(comm->rank() == root){ + if(rank == root){ + CHECK_NOT_IN_PLACE_ROOT(4, recvbuf) CHECK_TYPE(6, recvtype) CHECK_COUNT(5, recvcount) - CHECK_BUFFER(4, recvbuf, recvcount) + CHECK_BUFFER(4, recvbuf, recvcount, recvtype) + } else { + CHECK_NOT_IN_PLACE_ROOT(1, sendbuf) } CHECK_ROOT(7) CHECK_REQUEST(9) - 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; + if (rank == root){ + if (sendbuf == MPI_IN_PLACE) { + real_sendcount = 0; + real_sendtype = recvtype; + } else if(recvtype->size() * recvcount != sendtype->size() * sendcount){ + XBT_WARN("MPI_(I)Gather : received size at root differs from sent size : %zu vs %zu", recvtype->size() * recvcount , sendtype->size() * sendcount); + return MPI_ERR_TRUNCATE; + } } - int rank = simgrid::s4u::this_actor::get_pid(); - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Gather" : "PMPI_Igather", + smpi_bench_end(); + + aid_t pid = simgrid::s4u::this_actor::get_pid(); + + TRACE_smpi_comm_in(pid, 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(), @@ -162,7 +148,7 @@ int PMPI_Igather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void simgrid::smpi::colls::igather(real_sendbuf, real_sendcount, real_sendtype, recvbuf, recvcount, recvtype, root, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return MPI_SUCCESS; } @@ -176,23 +162,29 @@ int PMPI_Igatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, voi MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request) { CHECK_COMM(9) - CHECK_BUFFER(1, sendbuf, sendcount) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) + int rank = comm->rank(); if(sendbuf != MPI_IN_PLACE){ CHECK_TYPE(3, sendtype) CHECK_COUNT(2, sendcount) } - if(comm->rank() == root){ + CHECK_BUFFER(1, sendbuf, sendcount, sendtype) + if(rank == root){ + CHECK_NOT_IN_PLACE_ROOT(4, recvbuf) CHECK_TYPE(6, recvtype) CHECK_NULL(5, MPI_ERR_COUNT, recvcounts) CHECK_NULL(6, MPI_ERR_ARG, displs) + } else { + CHECK_NOT_IN_PLACE_ROOT(1, sendbuf) } CHECK_ROOT(8) CHECK_REQUEST(10) - if (comm->rank() == root){ + if (rank == root){ for (int i = 0; i < comm->size(); i++) { CHECK_COUNT(5, recvcounts[i]) - CHECK_BUFFER(4,recvbuf,recvcounts[i]) + CHECK_BUFFER(4,recvbuf,recvcounts[i], recvtype) } } @@ -200,21 +192,21 @@ int PMPI_Igatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, voi const void* real_sendbuf = sendbuf; int real_sendcount = sendcount; MPI_Datatype real_sendtype = sendtype; - if ((comm->rank() == root) && (sendbuf == MPI_IN_PLACE)) { + if ((rank == root) && (sendbuf == MPI_IN_PLACE)) { real_sendcount = 0; real_sendtype = recvtype; } - int rank = simgrid::s4u::this_actor::get_pid(); + aid_t pid = simgrid::s4u::this_actor::get_pid(); int dt_size_recv = recvtype->is_replayable() ? 1 : recvtype->size(); - std::vector* trace_recvcounts = new std::vector; - if (comm->rank() == root) { + auto trace_recvcounts = std::make_shared>(); + if (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", + TRACE_smpi_comm_in(pid, 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(), @@ -227,7 +219,7 @@ int PMPI_Igatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, voi simgrid::smpi::colls::igatherv(real_sendbuf, real_sendcount, real_sendtype, recvbuf, recvcounts, displs, recvtype, root, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return MPI_SUCCESS; } @@ -241,25 +233,36 @@ int PMPI_Iallgather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, v MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request) { CHECK_COMM(7) - CHECK_BUFFER(1, sendbuf, sendcount) - CHECK_BUFFER(4, recvbuf, recvcount) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) + int rank = comm->rank(); + CHECK_NOT_IN_PLACE(4, recvbuf) if(sendbuf != MPI_IN_PLACE){ CHECK_COUNT(2, sendcount) CHECK_TYPE(3, sendtype) } CHECK_TYPE(6, recvtype) CHECK_COUNT(5, recvcount) + CHECK_BUFFER(1, sendbuf, sendcount, sendtype) + CHECK_BUFFER(4, recvbuf, recvcount, recvtype) CHECK_REQUEST(8) - smpi_bench_end(); 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", + if(recvtype->size() * recvcount != sendtype->size() * sendcount){ + XBT_WARN("MPI_(I)Allgather : received size from each process differs from sent size : %zu vs %zu", recvtype->size() * recvcount, sendtype->size() * sendcount); + return MPI_ERR_TRUNCATE; + } + + smpi_bench_end(); + + aid_t pid = simgrid::s4u::this_actor::get_pid(); + + TRACE_smpi_comm_in(pid, 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(), @@ -270,7 +273,7 @@ int PMPI_Iallgather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, v else simgrid::smpi::colls::iallgather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return MPI_SUCCESS; } @@ -284,18 +287,23 @@ int PMPI_Iallgatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request) { CHECK_COMM(8) - CHECK_BUFFER(1, sendbuf, sendcount) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) + int rank = comm->rank(); if(sendbuf != MPI_IN_PLACE) CHECK_TYPE(3, sendtype) CHECK_TYPE(6, recvtype) CHECK_NULL(5, MPI_ERR_COUNT, recvcounts) CHECK_NULL(6, MPI_ERR_ARG, displs) - if(sendbuf != MPI_IN_PLACE) + if(sendbuf != MPI_IN_PLACE){ CHECK_COUNT(2, sendcount) + CHECK_BUFFER(1, sendbuf, sendcount, sendtype) + } CHECK_REQUEST(9) + CHECK_NOT_IN_PLACE(4, recvbuf) for (int i = 0; i < comm->size(); i++) { CHECK_COUNT(5, recvcounts[i]) - CHECK_BUFFER(4, recvbuf, recvcounts[i]) + CHECK_BUFFER(4, recvbuf, recvcounts[i], recvtype) } smpi_bench_end(); @@ -304,16 +312,16 @@ int PMPI_Iallgatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, sendcount = recvcounts[comm->rank()]; sendtype = recvtype; } - int rank = simgrid::s4u::this_actor::get_pid(); + aid_t pid = simgrid::s4u::this_actor::get_pid(); int dt_size_recv = recvtype->is_replayable() ? 1 : recvtype->size(); - std::vector* trace_recvcounts = new std::vector; + auto trace_recvcounts = std::make_shared>(); 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", + pid, 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), @@ -324,7 +332,7 @@ int PMPI_Iallgatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, simgrid::smpi::colls::iallgatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return MPI_SUCCESS; } @@ -338,30 +346,43 @@ int PMPI_Iscatter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, voi MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request) { CHECK_COMM(8) - if(comm->rank() == root){ - CHECK_BUFFER(1, sendbuf, sendcount) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) + int rank = comm->rank(); + if(rank == root){ + CHECK_NOT_IN_PLACE_ROOT(1, sendbuf) CHECK_COUNT(2, sendcount) CHECK_TYPE(3, sendtype) + CHECK_BUFFER(1, sendbuf, sendcount, sendtype) + } else { + CHECK_NOT_IN_PLACE_ROOT(4, recvbuf) } if(recvbuf != MPI_IN_PLACE){ - CHECK_BUFFER(4, recvbuf, recvcount) CHECK_COUNT(5, recvcount) CHECK_TYPE(6, recvtype) + CHECK_BUFFER(4, recvbuf, recvcount, recvtype) } CHECK_ROOT(8) CHECK_REQUEST(9) - smpi_bench_end(); 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", + if((rank == root) && (recvtype->size() * recvcount != sendtype->size() * sendcount)){ + XBT_WARN("MPI_(I)Scatter : sent size to each process differs from receive size"); + return MPI_ERR_TRUNCATE; + } + + smpi_bench_end(); + + aid_t pid = simgrid::s4u::this_actor::get_pid(); + + TRACE_smpi_comm_in(pid, 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(), + (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) @@ -369,7 +390,7 @@ int PMPI_Iscatter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, voi else simgrid::smpi::colls::iscatter(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return MPI_SUCCESS; } @@ -382,41 +403,47 @@ int PMPI_Scatterv(const void *sendbuf, const int *sendcounts, const int *displs, 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) { + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) CHECK_COMM(9) + int rank = comm->rank(); if(recvbuf != MPI_IN_PLACE){ - CHECK_BUFFER(4, recvbuf, recvcount) + CHECK_NOT_IN_PLACE_ROOT(1, sendbuf) CHECK_COUNT(5, recvcount) CHECK_TYPE(7, recvtype) + CHECK_BUFFER(4, recvbuf, recvcount, recvtype) } CHECK_ROOT(9) CHECK_REQUEST(10) - if (comm->rank() == root) { + if (rank == root) { CHECK_NULL(2, MPI_ERR_COUNT, sendcounts) CHECK_NULL(3, MPI_ERR_ARG, displs) CHECK_TYPE(4, sendtype) for (int i = 0; i < comm->size(); i++){ - CHECK_BUFFER(1, sendbuf, sendcounts[i]) CHECK_COUNT(2, sendcounts[i]) + CHECK_BUFFER(1, sendbuf, sendcounts[i], sendtype) } if (recvbuf == MPI_IN_PLACE) { recvtype = sendtype; - recvcount = sendcounts[comm->rank()]; + recvcount = sendcounts[rank]; } + } else { + CHECK_NOT_IN_PLACE_ROOT(4, recvbuf) } smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); + aid_t pid = 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) { + auto trace_sendcounts = std::make_shared>(); + if (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", + TRACE_smpi_comm_in(pid, 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(), @@ -428,7 +455,7 @@ int PMPI_Iscatterv(const void* sendbuf, const int* sendcounts, const int* displs simgrid::smpi::colls::iscatterv(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return MPI_SUCCESS; } @@ -441,20 +468,25 @@ int PMPI_Reduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype data int PMPI_Ireduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm, MPI_Request* request) { CHECK_COMM(7) - CHECK_BUFFER(1, sendbuf, count) - if(comm->rank() == root) - CHECK_BUFFER(5, recvbuf, count) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) + int rank = comm->rank(); CHECK_TYPE(4, datatype) CHECK_COUNT(3, count) - CHECK_OP(5) + CHECK_BUFFER(1, sendbuf, count, datatype) + if(rank == root){ + CHECK_NOT_IN_PLACE(2, recvbuf) + CHECK_BUFFER(5, recvbuf, count, datatype) + } + CHECK_OP(5, op, datatype) CHECK_ROOT(7) CHECK_REQUEST(8) smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); + aid_t pid = 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, + TRACE_smpi_comm_in(pid, 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) @@ -462,18 +494,20 @@ int PMPI_Ireduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype dat else simgrid::smpi::colls::ireduce(sendbuf, recvbuf, count, datatype, op, root, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return MPI_SUCCESS; } int PMPI_Reduce_local(const void* inbuf, void* inoutbuf, int count, MPI_Datatype datatype, MPI_Op op) { - CHECK_BUFFER(1, inbuf, count) - CHECK_BUFFER(2, inoutbuf, count) + SET_BUF1(inbuf) + SET_BUF2(inoutbuf) CHECK_TYPE(4, datatype) CHECK_COUNT(3, count) - CHECK_OP(5) + CHECK_BUFFER(1, inbuf, count, datatype) + CHECK_BUFFER(2, inoutbuf, count, datatype) + CHECK_OP(5, op, datatype) smpi_bench_end(); op->apply(inbuf, inoutbuf, &count, datatype); @@ -488,22 +522,25 @@ int PMPI_Allreduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype d int PMPI_Iallreduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request) { - CHECK_COMM(6) - CHECK_BUFFER(1, sendbuf, count) - CHECK_BUFFER(2, recvbuf, count) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) + int rank = comm->rank(); + CHECK_NOT_IN_PLACE(2, recvbuf) CHECK_TYPE(4, datatype) + CHECK_OP(5, op, datatype) CHECK_COUNT(3, count) + CHECK_BUFFER(1, sendbuf, count, datatype) + CHECK_BUFFER(2, recvbuf, count, datatype) CHECK_REQUEST(7) - CHECK_OP(5) smpi_bench_end(); - std::unique_ptr tmp_sendbuf; + std::vector tmp_sendbuf; const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, count, datatype); - int rank = simgrid::s4u::this_actor::get_pid(); + aid_t pid = simgrid::s4u::this_actor::get_pid(); - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Allreduce" : "PMPI_Iallreduce", + TRACE_smpi_comm_in(pid, 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), "")); @@ -513,7 +550,7 @@ int PMPI_Iallreduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype else simgrid::smpi::colls::iallreduce(real_sendbuf, recvbuf, count, datatype, op, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return MPI_SUCCESS; } @@ -526,19 +563,21 @@ int PMPI_Scan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype dataty int PMPI_Iscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request* request) { CHECK_COMM(6) - CHECK_BUFFER(1,sendbuf,count) - CHECK_BUFFER(2,recvbuf,count) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) CHECK_TYPE(4, datatype) CHECK_COUNT(3, count) + CHECK_BUFFER(1,sendbuf,count, datatype) + CHECK_BUFFER(2,recvbuf,count, datatype) CHECK_REQUEST(7) - CHECK_OP(5) + CHECK_OP(5, op, datatype) smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); - std::unique_ptr tmp_sendbuf; + aid_t pid = simgrid::s4u::this_actor::get_pid(); + std::vector tmp_sendbuf; const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, count, datatype); - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Scan" : "PMPI_Iscan", + TRACE_smpi_comm_in(pid, 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))); @@ -549,7 +588,7 @@ int PMPI_Iscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datat else retval = simgrid::smpi::colls::iscan(real_sendbuf, recvbuf, count, datatype, op, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return retval; } @@ -561,19 +600,21 @@ int PMPI_Exscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype data int PMPI_Iexscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request* request){ CHECK_COMM(6) - CHECK_BUFFER(1, sendbuf, count) - CHECK_BUFFER(2, recvbuf, count) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) CHECK_TYPE(4, datatype) CHECK_COUNT(3, count) + CHECK_BUFFER(1, sendbuf, count, datatype) + CHECK_BUFFER(2, recvbuf, count, datatype) CHECK_REQUEST(7) - CHECK_OP(5) + CHECK_OP(5, op, datatype) smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); - std::unique_ptr tmp_sendbuf; + aid_t pid = simgrid::s4u::this_actor::get_pid(); + std::vector tmp_sendbuf; const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, count, datatype); - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Exscan" : "PMPI_Iexscan", + TRACE_smpi_comm_in(pid, 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))); @@ -584,7 +625,7 @@ int PMPI_Iexscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype dat else retval = simgrid::smpi::colls::iexscan(real_sendbuf, recvbuf, count, datatype, op, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return retval; } @@ -597,19 +638,23 @@ int PMPI_Reduce_scatter(const void *sendbuf, void *recvbuf, const int *recvcount int PMPI_Ireduce_scatter(const void *sendbuf, void *recvbuf, const int *recvcounts, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, MPI_Request *request) { CHECK_COMM(6) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) + int rank = comm->rank(); + CHECK_NOT_IN_PLACE(2, recvbuf) CHECK_TYPE(4, datatype) CHECK_NULL(3, MPI_ERR_COUNT, recvcounts) CHECK_REQUEST(7) - CHECK_OP(5) + CHECK_OP(5, op, datatype) for (int i = 0; i < comm->size(); i++) { CHECK_COUNT(3, recvcounts[i]) - CHECK_BUFFER(1, sendbuf, recvcounts[i]) - CHECK_BUFFER(2, recvbuf, recvcounts[i]) + CHECK_BUFFER(1, sendbuf, recvcounts[i], datatype) + CHECK_BUFFER(2, recvbuf, recvcounts[i], datatype) } smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); - std::vector* trace_recvcounts = new std::vector; + aid_t pid = simgrid::s4u::this_actor::get_pid(); + auto trace_recvcounts = std::make_shared>(); int dt_send_size = datatype->is_replayable() ? 1 : datatype->size(); int totalcount = 0; @@ -617,10 +662,10 @@ int PMPI_Ireduce_scatter(const void *sendbuf, void *recvbuf, const int *recvcoun trace_recvcounts->push_back(recvcounts[i] * dt_send_size); totalcount += recvcounts[i]; } - std::unique_ptr tmp_sendbuf; + std::vector tmp_sendbuf; const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, totalcount, datatype); - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Reduce_scatter" : "PMPI_Ireduce_scatter", + TRACE_smpi_comm_in(pid, 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), "")); @@ -630,7 +675,7 @@ int PMPI_Ireduce_scatter(const void *sendbuf, void *recvbuf, const int *recvcoun else simgrid::smpi::colls::ireduce_scatter(real_sendbuf, recvbuf, recvcounts, datatype, op, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return MPI_SUCCESS; } @@ -645,37 +690,38 @@ int PMPI_Ireduce_scatter_block(const void* sendbuf, void* recvbuf, int recvcount MPI_Comm comm, MPI_Request* request) { CHECK_COMM(6) - CHECK_BUFFER(1, sendbuf, recvcount) - CHECK_BUFFER(2, recvbuf, recvcount) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) CHECK_TYPE(4, datatype) CHECK_COUNT(3, recvcount) + CHECK_BUFFER(1, sendbuf, recvcount, datatype) + CHECK_BUFFER(2, recvbuf, recvcount, datatype) CHECK_REQUEST(7) - CHECK_OP(5) + CHECK_OP(5, op, datatype) smpi_bench_end(); int count = comm->size(); - int rank = simgrid::s4u::this_actor::get_pid(); + aid_t pid = 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; + auto trace_recvcounts = std::make_shared>(recvcount * dt_send_size); // copy data to avoid bad free + std::vector 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", + pid, 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]; + std::vector recvcounts(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); + simgrid::smpi::colls::reduce_scatter(real_sendbuf, recvbuf, recvcounts.data(), datatype, op, comm); else - simgrid::smpi::colls::ireduce_scatter(real_sendbuf, recvbuf, recvcounts, datatype, op, comm, request); - delete[] recvcounts; + simgrid::smpi::colls::ireduce_scatter(real_sendbuf, recvbuf, recvcounts.data(), datatype, op, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return MPI_SUCCESS; } @@ -689,23 +735,24 @@ int PMPI_Ialltoall(const void* sendbuf, int sendcount, MPI_Datatype sendtype, vo MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request) { CHECK_COMM(7) - CHECK_BUFFER(1, sendbuf, sendcount) - CHECK_BUFFER(4, recvbuf, recvcount) - if(sendbuf != MPI_IN_PLACE) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) + if(sendbuf != MPI_IN_PLACE){ CHECK_TYPE(3, sendtype) + CHECK_COUNT(2, sendcount) + CHECK_BUFFER(1, sendbuf, sendcount, sendtype) + } CHECK_TYPE(6, recvtype) CHECK_COUNT(5, recvcount) - if(sendbuf != MPI_IN_PLACE) - CHECK_COUNT(2, sendcount) CHECK_COUNT(5, recvcount) + CHECK_BUFFER(4, recvbuf, recvcount, recvtype) CHECK_REQUEST(8) - smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); + aid_t pid = simgrid::s4u::this_actor::get_pid(); int real_sendcount = sendcount; MPI_Datatype real_sendtype = sendtype; - - std::unique_ptr tmp_sendbuf; + + std::vector tmp_sendbuf; const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, recvcount * comm->size(), recvtype); if (sendbuf == MPI_IN_PLACE) { @@ -713,7 +760,14 @@ int PMPI_Ialltoall(const void* sendbuf, int sendcount, MPI_Datatype sendtype, vo real_sendtype = recvtype; } - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Alltoall" : "PMPI_Ialltoall", + if(recvtype->size() * recvcount != real_sendtype->size() * real_sendcount){ + XBT_WARN("MPI_(I)Alltoall : receive size from each process differs from sent size : %zu vs %zu", recvtype->size() * recvcount, real_sendtype->size() * real_sendcount); + return MPI_ERR_TRUNCATE; + } + + smpi_bench_end(); + + TRACE_smpi_comm_in(pid, 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(), @@ -727,7 +781,7 @@ int PMPI_Ialltoall(const void* sendbuf, int sendcount, MPI_Datatype sendtype, vo retval = simgrid::smpi::colls::ialltoall(real_sendbuf, real_sendcount, real_sendtype, recvbuf, recvcount, recvtype, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return retval; } @@ -742,6 +796,8 @@ int PMPI_Ialltoallv(const void* sendbuf, const int* sendcounts, const int* sendd const int* recvcounts, const int* recvdispls, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request) { CHECK_COMM(9) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) if(sendbuf != MPI_IN_PLACE){ CHECK_NULL(2, MPI_ERR_COUNT, sendcounts) CHECK_NULL(3, MPI_ERR_ARG, senddispls) @@ -752,22 +808,22 @@ int PMPI_Ialltoallv(const void* sendbuf, const int* sendcounts, const int* sendd CHECK_NULL(7, MPI_ERR_ARG, recvdispls) CHECK_REQUEST(10) - int rank = simgrid::s4u::this_actor::get_pid(); + aid_t pid = simgrid::s4u::this_actor::get_pid(); int size = comm->size(); for (int i = 0; i < size; i++) { if(sendbuf != MPI_IN_PLACE){ - CHECK_BUFFER(1, sendbuf, sendcounts[i]) + CHECK_BUFFER(1, sendbuf, sendcounts[i], sendtype) CHECK_COUNT(2, sendcounts[i]) } - CHECK_BUFFER(5, recvbuf, recvcounts[i]) + CHECK_BUFFER(5, recvbuf, recvcounts[i], recvtype) CHECK_COUNT(6, recvcounts[i]) } 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; + auto trace_sendcounts = std::make_shared>(); + auto trace_recvcounts = std::make_shared>(); int dt_size_recv = recvtype->size(); const int* real_sendcounts = sendcounts; @@ -781,20 +837,24 @@ int PMPI_Ialltoallv(const void* sendbuf, const int* sendcounts, const int* sendd maxsize = (recvdispls[i] + recvcounts[i]) * dt_size_recv; } - std::unique_ptr tmp_sendbuf; - std::unique_ptr tmp_sendcounts; - std::unique_ptr tmp_senddispls; + std::vector tmp_sendbuf; + std::vector tmp_sendcounts; + std::vector tmp_senddispls; 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_senddispls.reset(new int[size]); - std::copy(recvdispls, recvdispls + size, tmp_senddispls.get()); - real_senddispls = tmp_senddispls.get(); + tmp_sendcounts.assign(recvcounts, recvcounts + size); + real_sendcounts = tmp_sendcounts.data(); + tmp_senddispls.assign(recvdispls, recvdispls + size); + real_senddispls = tmp_senddispls.data(); real_sendtype = recvtype; } + if(recvtype->size() * recvcounts[comm->rank()] != real_sendtype->size() * real_sendcounts[comm->rank()]){ + XBT_WARN("MPI_(I)Alltoallv : receive size from me differs from sent size to me : %zu vs %zu", recvtype->size() * recvcounts[comm->rank()], real_sendtype->size() * real_sendcounts[comm->rank()]); + smpi_bench_begin(); + return MPI_ERR_TRUNCATE; + } + int dt_size_send = real_sendtype->size(); for (int i = 0; i < size; i++) { // copy data to avoid bad free @@ -802,7 +862,7 @@ int PMPI_Ialltoallv(const void* sendbuf, const int* sendcounts, const int* sendd trace_sendcounts->push_back(real_sendcounts[i] * dt_size_send); } - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Alltoallv" : "PMPI_Ialltoallv", + TRACE_smpi_comm_in(pid, 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), @@ -816,7 +876,7 @@ int PMPI_Ialltoallv(const void* sendbuf, const int* sendcounts, const int* sendd retval = simgrid::smpi::colls::ialltoallv(real_sendbuf, real_sendcounts, real_senddispls, real_sendtype, recvbuf, recvcounts, recvdispls, recvtype, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return retval; } @@ -831,6 +891,8 @@ int PMPI_Ialltoallw(const void* sendbuf, const int* sendcounts, const int* sendd const int* recvcounts, const int* recvdispls, const MPI_Datatype* recvtypes, MPI_Comm comm, MPI_Request* request) { CHECK_COMM(9) + SET_BUF1(sendbuf) + SET_BUF2(recvbuf) if(sendbuf != MPI_IN_PLACE){ CHECK_NULL(2, MPI_ERR_COUNT, sendcounts) CHECK_NULL(3, MPI_ERR_ARG, senddispls) @@ -840,57 +902,58 @@ int PMPI_Ialltoallw(const void* sendbuf, const int* sendcounts, const int* sendd CHECK_NULL(7, MPI_ERR_ARG, recvdispls) CHECK_NULL(8, MPI_ERR_TYPE, recvtypes) CHECK_REQUEST(10) - int rank = simgrid::s4u::this_actor::get_pid(); + aid_t pid = simgrid::s4u::this_actor::get_pid(); int size = comm->size(); for (int i = 0; i < size; i++) { if(sendbuf != MPI_IN_PLACE){ - CHECK_BUFFER(1, sendbuf, sendcounts[i]) CHECK_COUNT(2, sendcounts[i]) CHECK_TYPE(4, sendtypes[i]) + CHECK_BUFFER(1, sendbuf, sendcounts[i], sendtypes[i]) } - CHECK_BUFFER(5, recvbuf, recvcounts[i]) CHECK_COUNT(6, recvcounts[i]) CHECK_TYPE(8, recvtypes[i]) + CHECK_BUFFER(5, recvbuf, recvcounts[i], recvtypes[i]) } 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; + auto trace_sendcounts = std::make_shared>(); + auto trace_recvcounts = std::make_shared>(); const int* real_sendcounts = sendcounts; const int* real_senddispls = senddispls; 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; + if (recvtypes[i] == MPI_DATATYPE_NULL) return MPI_ERR_TYPE; - } recv_size += recvcounts[i] * recvtypes[i]->size(); trace_recvcounts->push_back(recvcounts[i] * recvtypes[i]->size()); if ((recvdispls[i] + (recvcounts[i] * recvtypes[i]->size())) > maxsize) maxsize = recvdispls[i] + (recvcounts[i] * recvtypes[i]->size()); } - std::unique_ptr tmp_sendbuf; - std::unique_ptr tmp_sendcounts; - std::unique_ptr tmp_senddispls; - std::unique_ptr tmp_sendtypes; + std::vector tmp_sendbuf; + std::vector tmp_sendcounts; + std::vector tmp_senddispls; + std::vector 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_senddispls.reset(new int[size]); - std::copy(recvdispls, recvdispls + size, tmp_senddispls.get()); - real_senddispls = tmp_senddispls.get(); - tmp_sendtypes.reset(new MPI_Datatype[size]); - std::copy(recvtypes, recvtypes + size, tmp_sendtypes.get()); - real_sendtypes = tmp_sendtypes.get(); + tmp_sendcounts.assign(recvcounts, recvcounts + size); + real_sendcounts = tmp_sendcounts.data(); + tmp_senddispls.assign(recvdispls, recvdispls + size); + real_senddispls = tmp_senddispls.data(); + tmp_sendtypes.assign(recvtypes, recvtypes + size); + real_sendtypes = tmp_sendtypes.data(); + } + + + if(recvtypes[comm->rank()]->size() * recvcounts[comm->rank()] != real_sendtypes[comm->rank()]->size() * real_sendcounts[comm->rank()]){ + XBT_WARN("MPI_(I)Alltoallw : receive size from me differs from sent size to me : %zu vs %zu", recvtypes[comm->rank()]->size() * recvcounts[comm->rank()], real_sendtypes[comm->rank()]->size() * real_sendcounts[comm->rank()]); + smpi_bench_begin(); + return MPI_ERR_TRUNCATE; } for (int i = 0; i < size; i++) { // copy data to avoid bad free @@ -898,7 +961,7 @@ int PMPI_Ialltoallw(const void* sendbuf, const int* sendcounts, const int* sendd trace_sendcounts->push_back(real_sendcounts[i] * real_sendtypes[i]->size()); } - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Alltoallw" : "PMPI_Ialltoallw", + TRACE_smpi_comm_in(pid, 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]), @@ -912,7 +975,7 @@ int PMPI_Ialltoallw(const void* sendbuf, const int* sendcounts, const int* sendd retval = simgrid::smpi::colls::ialltoallw(real_sendbuf, real_sendcounts, real_senddispls, real_sendtypes, recvbuf, recvcounts, recvdispls, recvtypes, comm, request); - TRACE_smpi_comm_out(rank); + TRACE_smpi_comm_out(pid); smpi_bench_begin(); return retval; }