X-Git-Url: http://bilbo.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/13b40c020d5701ec7ebedfc0f2840ed263d46a92..abf3b21ad3f4a19475431076e6e2b26f76d2e7d6:/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 1c4b10ccfc..7e613a5b63 100644 --- a/src/smpi/bindings/smpi_pmpi_coll.cpp +++ b/src/smpi/bindings/smpi_pmpi_coll.cpp @@ -1,4 +1,4 @@ -/* Copyright (c) 2007-2021. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2007-2023. 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. */ @@ -37,10 +37,10 @@ int PMPI_Ibarrier(MPI_Comm comm, MPI_Request *request) { CHECK_COMM(1) 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", + CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Barrier" : "PMPI_Ibarrier") + const SmpiBenchGuard suspend_bench; + 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); @@ -49,8 +49,7 @@ int PMPI_Ibarrier(MPI_Comm comm, MPI_Request *request) } else simgrid::smpi::colls::ibarrier(comm, request); - TRACE_smpi_comm_out(rank); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return MPI_SUCCESS; } @@ -59,22 +58,28 @@ int PMPI_Bcast(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm c return PMPI_Ibcast(buf, count, datatype, root, comm, MPI_REQUEST_IGNORED); } -int PMPI_Ibcast(void *buf, int count, MPI_Datatype datatype, - int root, MPI_Comm comm, MPI_Request* request) +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) + CHECK_COLLECTIVE(comm, std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Bcast" : "PMPI_Ibcast") + " with root " + + std::to_string(root)) - smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Bcast" : "PMPI_Ibcast", + const SmpiBenchGuard suspend_bench; + 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, + count, 0, simgrid::smpi::Datatype::encode(datatype), "")); + if (simgrid::config::get_value("smpi/barrier-collectives") && + request == MPI_REQUEST_IGNORED) // No barrier in Ibcast + smpi_deployment_startup_barrier(smpi_process()->get_instance_id()); + if (comm->size() > 1) { if (request == MPI_REQUEST_IGNORED) simgrid::smpi::colls::bcast(buf, count, datatype, root, comm); @@ -85,8 +90,7 @@ int PMPI_Ibcast(void *buf, int count, MPI_Datatype datatype, *request = MPI_REQUEST_NULL; } - TRACE_smpi_comm_out(rank); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return MPI_SUCCESS; } @@ -99,23 +103,31 @@ 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) + 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){ + SET_BUF2(recvbuf) + 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) + CHECK_COLLECTIVE(comm, std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Gather" : "PMPI_Igather") + " with root " + + std::to_string(root)) const void* real_sendbuf = sendbuf; int real_sendcount = sendcount; MPI_Datatype real_sendtype = sendtype; - if (comm->rank() == root){ + if (rank == root){ if (sendbuf == MPI_IN_PLACE) { real_sendcount = 0; real_sendtype = recvtype; @@ -125,15 +137,18 @@ int PMPI_Igather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void } } - smpi_bench_end(); + const SmpiBenchGuard suspend_bench; + + if (simgrid::config::get_value("smpi/barrier-collectives") && + request == MPI_REQUEST_IGNORED) // no barrier in Igather + smpi_deployment_startup_barrier(smpi_process()->get_instance_id()); - 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_Gather" : "PMPI_Igather", + 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(), - (comm->rank() != root || recvtype->is_replayable()) ? recvcount : recvcount * recvtype->size(), + real_sendcount, recvcount, 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); @@ -141,8 +156,7 @@ 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); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return MPI_SUCCESS; } @@ -155,49 +169,61 @@ 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) + 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){ + SET_BUF2(recvbuf) + 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) + CHECK_COLLECTIVE(comm, std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Gatherv" : "PMPI_Igatherv") + + " with root " + std::to_string(root)) - 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) } } - smpi_bench_end(); + const SmpiBenchGuard suspend_bench; + + if (simgrid::config::get_value("smpi/barrier-collectives") && + request == MPI_REQUEST_IGNORED) // no barrier in Igatherv + smpi_deployment_startup_barrier(smpi_process()->get_instance_id()); + 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(); - int dt_size_recv = recvtype->is_replayable() ? 1 : recvtype->size(); + aid_t pid = simgrid::s4u::this_actor::get_pid(); - auto* 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); - } + auto trace_recvcounts = std::make_shared>(); + if (rank == root) + trace_recvcounts->insert(trace_recvcounts->end(), &recvcounts[0], &recvcounts[comm->size()]); + else //this is not significant outside of root, put 0 as we don't know if recvcounts is initialized + trace_recvcounts->insert(trace_recvcounts->end(), comm->size(), 0); - 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(), - nullptr, dt_size_recv, trace_recvcounts, simgrid::smpi::Datatype::encode(real_sendtype), + sendcount, + nullptr, -1, 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, @@ -206,8 +232,7 @@ 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); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return MPI_SUCCESS; } @@ -220,15 +245,20 @@ 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) + CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Allgather" : "PMPI_Iallggather") if (sendbuf == MPI_IN_PLACE) { sendbuf = static_cast(recvbuf) + recvtype->get_extent() * recvcount * comm->rank(); @@ -241,23 +271,21 @@ int PMPI_Iallgather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, v return MPI_ERR_TRUNCATE; } - smpi_bench_end(); + const SmpiBenchGuard suspend_bench; - 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_Allgather" : "PMPI_Iallggather", + 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(), - recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), + sendcount, recvcount, 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(); + TRACE_smpi_comm_out(pid); return MPI_SUCCESS; } @@ -270,39 +298,43 @@ 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) } + CHECK_COLLECTIVE(comm, MPI_REQUEST_IGNORED ? "PMPI_Allgatherv" : "PMPI_Iallgatherv") + + const SmpiBenchGuard suspend_bench; - 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(); + aid_t pid = simgrid::s4u::this_actor::get_pid(); - auto* 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); - } + auto trace_recvcounts = std::make_shared>(); + trace_recvcounts->insert(trace_recvcounts->end(), &recvcounts[0], &recvcounts[comm->size()]); 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), + sendcount, nullptr, + -1, 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); @@ -310,8 +342,7 @@ 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); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return MPI_SUCCESS; } @@ -324,46 +355,56 @@ 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_BUF2(recvbuf) + int rank = comm->rank(); + if(rank == root){ + SET_BUF1(sendbuf) + 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) + CHECK_COLLECTIVE(comm, std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Scatter" : "PMPI_Iscatter") + + " with root " + std::to_string(root)) if (recvbuf == MPI_IN_PLACE) { recvtype = sendtype; recvcount = sendcount; } - if((comm->rank() == root) && (recvtype->size() * recvcount != sendtype->size() * sendcount)){ + 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(); + const SmpiBenchGuard suspend_bench; - int rank = simgrid::s4u::this_actor::get_pid(); + if (simgrid::config::get_value("smpi/barrier-collectives") && + request == MPI_REQUEST_IGNORED) // no barrier in Iscatter + smpi_deployment_startup_barrier(smpi_process()->get_instance_id()); - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Scatter" : "PMPI_Iscatter", + 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(), - recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), + sendcount, recvcount, 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(); + TRACE_smpi_comm_out(pid); return MPI_SUCCESS; } @@ -375,44 +416,55 @@ 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_BUF2(recvbuf) CHECK_COMM(9) + int rank = comm->rank(); if(recvbuf != MPI_IN_PLACE){ - CHECK_BUFFER(4, recvbuf, recvcount) 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) { + SET_BUF1(sendbuf) + CHECK_NOT_IN_PLACE_ROOT(1, sendbuf) 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) } + CHECK_COLLECTIVE(comm, std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Scatterv" : "PMPI_Iscatterv") + + " with root " + std::to_string(root)) - smpi_bench_end(); + const SmpiBenchGuard suspend_bench; - int rank = simgrid::s4u::this_actor::get_pid(); - int dt_size_send = sendtype->is_replayable() ? 1 : sendtype->size(); + if (simgrid::config::get_value("smpi/barrier-collectives") && + request == MPI_REQUEST_IGNORED) // no barrier in Iscatterv + smpi_deployment_startup_barrier(smpi_process()->get_instance_id()); - auto* 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); - } - } + aid_t pid = simgrid::s4u::this_actor::get_pid(); - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Scatterv" : "PMPI_Iscatterv", + auto trace_sendcounts = std::make_shared>(); + if (rank == root) + trace_sendcounts->insert(trace_sendcounts->end(), &sendcounts[0], &sendcounts[comm->size()]); + else //this is not significant outside of root, put 0 as we don't know if sendcounts is initialized + trace_sendcounts->insert(trace_sendcounts->end(), comm->size(), 0); + + + 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(), + request == MPI_REQUEST_IGNORED ? "scatterv" : "iscatterv", root, -1, + trace_sendcounts, recvcount, nullptr, simgrid::smpi::Datatype::encode(sendtype), simgrid::smpi::Datatype::encode(recvtype))); if (request == MPI_REQUEST_IGNORED) @@ -421,8 +473,7 @@ 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); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return MPI_SUCCESS; } @@ -434,43 +485,55 @@ 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) + int rank = comm->rank(); CHECK_TYPE(4, datatype) CHECK_COUNT(3, count) - CHECK_OP(5) + CHECK_BUFFER(1, sendbuf, count, datatype) + if(rank == root){ + SET_BUF2(recvbuf) + CHECK_NOT_IN_PLACE(2, recvbuf) + CHECK_BUFFER(5, recvbuf, count, datatype) + } + CHECK_OP(5, op, datatype) CHECK_ROOT(7) CHECK_REQUEST(8) + CHECK_COLLECTIVE(comm, std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Reduce" : "PMPI_Ireduce") + " with op " + + op->name() + " and root " + std::to_string(root)) - smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); + const SmpiBenchGuard suspend_bench; - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Reduce" : "PMPI_Ireduce", + if (simgrid::config::get_value("smpi/barrier-collectives") && + request == MPI_REQUEST_IGNORED) // no barrier in Ireduce + smpi_deployment_startup_barrier(smpi_process()->get_instance_id()); + + aid_t pid = simgrid::s4u::this_actor::get_pid(); + + 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, + count, 0, 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(); + TRACE_smpi_comm_out(pid); 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(); + const SmpiBenchGuard suspend_bench; op->apply(inbuf, inoutbuf, &count, datatype); - smpi_bench_begin(); return MPI_SUCCESS; } @@ -482,22 +545,29 @@ 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) + CHECK_COLLECTIVE(comm, std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Alleduce" : "PMPI_Iallreduce") + + " with op " + op->name()) + + const SmpiBenchGuard suspend_bench; - smpi_bench_end(); 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, + count, 0, simgrid::smpi::Datatype::encode(datatype), "")); if (request == MPI_REQUEST_IGNORED) @@ -505,8 +575,7 @@ 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); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return MPI_SUCCESS; } @@ -518,22 +587,30 @@ 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) + CHECK_COLLECTIVE(comm, + std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Scan" : "PMPI_Iscan") + " with op " + op->name()) + + const SmpiBenchGuard suspend_bench; - smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); + if (simgrid::config::get_value("smpi/barrier-collectives") && + request == MPI_REQUEST_IGNORED) // no barrier in Iscan + smpi_deployment_startup_barrier(smpi_process()->get_instance_id()); + + 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", - 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_in(pid, request == MPI_REQUEST_IGNORED ? "PMPI_Scan" : "PMPI_Iscan", + new simgrid::instr::CollTIData(request == MPI_REQUEST_IGNORED ? "scan" : "iscan", -1, 0.0, + count, 0, simgrid::smpi::Datatype::encode(datatype), "")); int retval; if (request == MPI_REQUEST_IGNORED) @@ -541,8 +618,7 @@ 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); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return retval; } @@ -553,22 +629,30 @@ 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) + CHECK_COLLECTIVE(comm, std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Exscan" : "PMPI_Iexscan") + " with op " + + op->name()) + + const SmpiBenchGuard suspend_bench; + + if (simgrid::config::get_value("smpi/barrier-collectives") && + request == MPI_REQUEST_IGNORED) // no barrier in Iexscan + smpi_deployment_startup_barrier(smpi_process()->get_instance_id()); - smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); + 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", - 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_in(pid, request == MPI_REQUEST_IGNORED ? "PMPI_Exscan" : "PMPI_Iexscan", + new simgrid::instr::CollTIData(request == MPI_REQUEST_IGNORED ? "exscan" : "iexscan", -1, 0.0, + count, 0, simgrid::smpi::Datatype::encode(datatype), "")); int retval; if (request == MPI_REQUEST_IGNORED) @@ -576,8 +660,7 @@ 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); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return retval; } @@ -589,41 +672,47 @@ 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) } + CHECK_COLLECTIVE(comm, std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Reduce_scatter" : "PMPI_Ireduce_scatter") + + " with op " + op->name()) + + const SmpiBenchGuard suspend_bench; + + aid_t pid = simgrid::s4u::this_actor::get_pid(); + auto trace_recvcounts = std::make_shared>(); + trace_recvcounts->insert(trace_recvcounts->end(), &recvcounts[0], &recvcounts[comm->size()]); - smpi_bench_end(); - int rank = simgrid::s4u::this_actor::get_pid(); - auto* 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::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), "")); + request == MPI_REQUEST_IGNORED ? "reducescatter" : "ireducescatter", -1, -1, nullptr, + -1 , trace_recvcounts, std::to_string(0), simgrid::smpi::Datatype::encode(datatype))); 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(); + TRACE_smpi_comm_out(pid); return MPI_SUCCESS; } @@ -637,25 +726,31 @@ 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) + CHECK_COLLECTIVE( + comm, std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Reduce_scatter_block" : "PMPI_Ireduce_scatter_block") + + " with op " + op->name()) + + const SmpiBenchGuard suspend_bench; - 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(); - auto* trace_recvcounts = new std::vector(recvcount * dt_send_size); // copy data to avoid bad free + aid_t pid = simgrid::s4u::this_actor::get_pid(); + auto trace_recvcounts = std::make_shared>(recvcount); + 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", - new simgrid::instr::VarCollTIData(request == MPI_REQUEST_IGNORED ? "reducescatter" : "ireducescatter", -1, 0, + pid, request == MPI_REQUEST_IGNORED ? "PMPI_Reduce_scatter_block" : "PMPI_Ireduce_scatter_block", + new simgrid::instr::VarCollTIData(request == MPI_REQUEST_IGNORED ? "reducescatter" : "ireducescatter", -1, -1, nullptr, -1, trace_recvcounts, simgrid::smpi::Datatype::encode(datatype), "")); std::vector recvcounts(count); @@ -666,8 +761,7 @@ int PMPI_Ireduce_scatter_block(const void* sendbuf, void* recvbuf, int recvcount else simgrid::smpi::colls::ireduce_scatter(real_sendbuf, recvbuf, recvcounts.data(), datatype, op, comm, request); - TRACE_smpi_comm_out(rank); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return MPI_SUCCESS; } @@ -680,18 +774,21 @@ 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) + CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Alltoall" : "PMPI_Ialltoall") - 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; @@ -708,13 +805,12 @@ int PMPI_Ialltoall(const void* sendbuf, int sendcount, MPI_Datatype sendtype, vo return MPI_ERR_TRUNCATE; } - smpi_bench_end(); + const SmpiBenchGuard suspend_bench; - TRACE_smpi_comm_in(rank, request == MPI_REQUEST_IGNORED ? "PMPI_Alltoall" : "PMPI_Ialltoall", + 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(), - recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), + real_sendcount, recvcount, simgrid::smpi::Datatype::encode(real_sendtype), simgrid::smpi::Datatype::encode(recvtype))); int retval; if (request == MPI_REQUEST_IGNORED) @@ -724,8 +820,7 @@ 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); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return retval; } @@ -739,6 +834,8 @@ 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) @@ -748,40 +845,38 @@ int PMPI_Ialltoallv(const void* sendbuf, const int* sendcounts, const int* sendd CHECK_NULL(6, MPI_ERR_COUNT, recvcounts) CHECK_NULL(7, MPI_ERR_ARG, recvdispls) CHECK_REQUEST(10) + CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Alltoallv" : "PMPI_Ialltoallv") - 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(); + const SmpiBenchGuard suspend_bench; + int send_size = 0; int recv_size = 0; - auto* trace_sendcounts = new std::vector(); - auto* trace_recvcounts = new std::vector(); + auto trace_sendcounts = std::make_shared>(); + auto trace_recvcounts = std::make_shared>(); + trace_recvcounts->insert(trace_recvcounts->end(), &recvcounts[0], &recvcounts[size]); + int dt_size_recv = recvtype->size(); const int* real_sendcounts = sendcounts; const int* real_senddispls = senddispls; 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 (((recvdispls[i] + recvcounts[i]) * dt_size_recv) > maxsize) - maxsize = (recvdispls[i] + recvcounts[i]) * dt_size_recv; - } - 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); + const void* real_sendbuf; + if (sendbuf == MPI_IN_PLACE) { tmp_sendcounts.assign(recvcounts, recvcounts + size); real_sendcounts = tmp_sendcounts.data(); @@ -790,20 +885,22 @@ int PMPI_Ialltoallv(const void* sendbuf, const int* sendcounts, const int* sendd real_sendtype = recvtype; } + for (int i = 0; i < size; i++) { // copy data to avoid bad free + send_size += real_sendcounts[i] ; + recv_size += recvcounts[i]; + if (((recvdispls[i] + recvcounts[i]) * dt_size_recv) > maxsize) + maxsize = (recvdispls[i] + recvcounts[i]) * dt_size_recv; + } + real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, maxsize, MPI_CHAR); + 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(); + trace_sendcounts->insert(trace_sendcounts->end(), &real_sendcounts[0], &real_sendcounts[size]); - 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", + 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), @@ -817,8 +914,7 @@ 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); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return retval; } @@ -832,6 +928,8 @@ 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) @@ -841,38 +939,37 @@ 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]) } + CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Alltoallw" : "PMPI_Ialltoallw") - smpi_bench_end(); + const SmpiBenchGuard suspend_bench; int send_size = 0; int recv_size = 0; - auto* trace_sendcounts = new std::vector(); - auto* trace_recvcounts = new std::vector(); + auto trace_sendcounts = std::make_shared>(); + auto trace_recvcounts = std::make_shared>(); + trace_recvcounts->insert(trace_recvcounts->end(), &recvcounts[0], &recvcounts[size]); 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()); } @@ -894,16 +991,15 @@ int PMPI_Ialltoallw(const void* sendbuf, const int* sendcounts, const int* sendd 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; } + trace_sendcounts->insert(trace_sendcounts->end(), &real_sendcounts[0], &real_sendcounts[size]); 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", + 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]), @@ -917,7 +1013,6 @@ 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); - smpi_bench_begin(); + TRACE_smpi_comm_out(pid); return retval; }