-/* Copyright (c) 2007-2017. 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. */
#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);
+ static const void* smpi_get_in_place_buf(const void* inplacebuf, const void* otherbuf,std::unique_ptr<unsigned char[]>& 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;
+ }
+}
/* PMPI User level calls */
-extern "C" { // Obviously, the C MPI interface should use the C linkage
-int PMPI_Bcast(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm)
+int PMPI_Barrier(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_ARG;
- } else {
- int rank = simgrid::s4u::Actor::self()->getPid();
- TRACE_smpi_comm_in(rank, __FUNCTION__,
- new simgrid::instr::CollTIData("bcast", root, -1.0,
- datatype->is_replayable() ? count : count * datatype->size(), -1,
- encode_datatype(datatype), ""));
- if (comm->size() > 1)
- simgrid::smpi::Colls::bcast(buf, count, datatype, root, comm);
- retval = MPI_SUCCESS;
-
- TRACE_smpi_comm_out(rank);
- }
- smpi_bench_begin();
- return retval;
+ return PMPI_Ibarrier(comm, MPI_REQUEST_IGNORED);
}
-int PMPI_Barrier(MPI_Comm comm)
+int PMPI_Ibarrier(MPI_Comm comm, MPI_Request *request)
{
- int retval = 0;
+ CHECK_COMM(1)
+ CHECK_REQUEST(2)
smpi_bench_end();
-
- if (comm == MPI_COMM_NULL) {
- retval = MPI_ERR_COMM;
- } else {
- int rank = simgrid::s4u::Actor::self()->getPid();
- TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::NoOpTIData("barrier"));
-
- simgrid::smpi::Colls::barrier(comm);
-
- //Barrier can be used to synchronize RMA calls. Finish all requests from comm before.
+ 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_Gather(void *sendbuf, int sendcount, MPI_Datatype sendtype,void *recvbuf, int recvcount, MPI_Datatype recvtype,
- int root, MPI_Comm comm)
+int PMPI_Bcast(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm)
{
- int retval = 0;
-
- 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;
- } else {
-
- char* sendtmpbuf = static_cast<char*>(sendbuf);
- int sendtmpcount = sendcount;
- MPI_Datatype sendtmptype = sendtype;
- if( (comm->rank() == root) && (sendbuf == MPI_IN_PLACE )) {
- sendtmpcount=0;
- sendtmptype=recvtype;
- }
- int rank = simgrid::s4u::Actor::self()->getPid();
-
- TRACE_smpi_comm_in(rank, __FUNCTION__,
- 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(),
- encode_datatype(sendtmptype), encode_datatype(recvtype)));
-
- simgrid::smpi::Colls::gather(sendtmpbuf, sendtmpcount, sendtmptype, recvbuf, recvcount, recvtype, root, comm);
-
- retval = MPI_SUCCESS;
- TRACE_smpi_comm_out(rank);
- }
-
- smpi_bench_begin();
- return retval;
+ return PMPI_Ibcast(buf, count, datatype, root, comm, MPI_REQUEST_IGNORED);
}
-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_Ibcast(void *buf, int count, MPI_Datatype datatype,
+ int root, MPI_Comm comm, MPI_Request* request)
{
- int retval = 0;
+ CHECK_COMM(5)
+ CHECK_BUFFER(1, buf, count)
+ CHECK_COUNT(2, count)
+ CHECK_TYPE(3, datatype)
+ CHECK_ROOT(4)
+ CHECK_REQUEST(6)
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)){
- retval = MPI_ERR_COUNT;
- } else if (recvcounts == nullptr || displs == nullptr) {
- retval = MPI_ERR_ARG;
+ 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<char*>(sendbuf);
- int sendtmpcount = sendcount;
- MPI_Datatype sendtmptype = sendtype;
- if( (comm->rank() == root) && (sendbuf == MPI_IN_PLACE )) {
- sendtmpcount=0;
- sendtmptype=recvtype;
- }
-
- int rank = simgrid::s4u::Actor::self()->getPid();
- int dt_size_recv = recvtype->is_replayable() ? 1 : recvtype->size();
-
- std::vector<int>* trace_recvcounts = new std::vector<int>;
- 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, __FUNCTION__,
- new simgrid::instr::VarCollTIData(
- "gatherV", root,
- sendtmptype->is_replayable() ? sendtmpcount : sendtmpcount * sendtmptype->size(), nullptr,
- dt_size_recv, trace_recvcounts, encode_datatype(sendtmptype), encode_datatype(recvtype)));
-
- retval = simgrid::smpi::Colls::gatherv(sendtmpbuf, sendtmpcount, sendtmptype, recvbuf, recvcounts, displs, recvtype, root, comm);
- 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_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_Allgather(void *sendbuf, int sendcount, MPI_Datatype sendtype,
- void *recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm)
+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 = MPI_SUCCESS;
+ CHECK_COMM(8)
+ if(sendbuf != MPI_IN_PLACE){
+ CHECK_BUFFER(1,sendbuf, sendcount)
+ CHECK_COUNT(2, sendcount)
+ CHECK_TYPE(3, sendtype)
+ }
+ if(comm->rank() == root){
+ CHECK_TYPE(6, recvtype)
+ CHECK_COUNT(5, recvcount)
+ CHECK_BUFFER(4, recvbuf, recvcount)
+ }
+ CHECK_ROOT(7)
+ CHECK_REQUEST(9)
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<char*>(recvbuf)+recvtype->get_extent()*recvcount*comm->rank();
- sendcount=recvcount;
- sendtype=recvtype;
- }
- int rank = simgrid::s4u::Actor::self()->getPid();
-
- TRACE_smpi_comm_in(rank, __FUNCTION__,
- new simgrid::instr::CollTIData("allGather", -1, -1.0,
- sendtype->is_replayable() ? sendcount : sendcount * sendtype->size(),
- recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(),
- encode_datatype(sendtype), encode_datatype(recvtype)));
-
- simgrid::smpi::Colls::allgather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm);
- TRACE_smpi_comm_out(rank);
+ 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 retval;
+ return MPI_SUCCESS;
+}
+
+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 PMPI_Allgatherv(void *sendbuf, int sendcount, MPI_Datatype sendtype,
- void *recvbuf, int *recvcounts, int *displs, MPI_Datatype recvtype, MPI_Comm comm)
+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)
{
- int retval = 0;
+ CHECK_COMM(9)
+ CHECK_BUFFER(1, sendbuf, sendcount)
+ if(sendbuf != MPI_IN_PLACE){
+ CHECK_TYPE(3, sendtype)
+ CHECK_COUNT(2, sendcount)
+ }
+ if(comm->rank() == root){
+ CHECK_TYPE(6, recvtype)
+ CHECK_NULL(5, MPI_ERR_COUNT, recvcounts)
+ CHECK_NULL(6, MPI_ERR_ARG, displs)
+ }
+ CHECK_ROOT(8)
+ CHECK_REQUEST(10)
- smpi_bench_end();
+ if (comm->rank() == root){
+ for (int i = 0; i < comm->size(); i++) {
+ CHECK_COUNT(5, recvcounts[i])
+ CHECK_BUFFER(4,recvbuf,recvcounts[i])
+ }
+ }
- 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 {
+ 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(sendbuf == MPI_IN_PLACE) {
- sendbuf=static_cast<char*>(recvbuf)+recvtype->get_extent()*displs[comm->rank()];
- sendcount=recvcounts[comm->rank()];
- sendtype=recvtype;
- }
- int rank = simgrid::s4u::Actor::self()->getPid();
- int dt_size_recv = recvtype->is_replayable() ? 1 : recvtype->size();
+ int rank = simgrid::s4u::this_actor::get_pid();
+ int dt_size_recv = recvtype->is_replayable() ? 1 : recvtype->size();
- std::vector<int>* trace_recvcounts = new std::vector<int>;
+ std::vector<int>* trace_recvcounts = new std::vector<int>();
+ 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, __FUNCTION__,
- new simgrid::instr::VarCollTIData(
- "allGatherV", -1, sendtype->is_replayable() ? sendcount : sendcount * sendtype->size(), nullptr,
- dt_size_recv, trace_recvcounts, encode_datatype(sendtype), encode_datatype(recvtype)));
-
- simgrid::smpi::Colls::allgatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, comm);
- retval = MPI_SUCCESS;
- TRACE_smpi_comm_out(rank);
}
+ 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_Scatter(void *sendbuf, int sendcount, MPI_Datatype sendtype,
- void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm)
+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_Iallgather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
+ MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request)
{
- int retval = 0;
+ CHECK_COMM(7)
+ CHECK_BUFFER(1, sendbuf, sendcount)
+ CHECK_BUFFER(4, recvbuf, recvcount)
+ if(sendbuf != MPI_IN_PLACE){
+ CHECK_COUNT(2, sendcount)
+ CHECK_TYPE(3, sendtype)
+ }
+ CHECK_TYPE(6, recvtype)
+ CHECK_COUNT(5, recvcount)
+ CHECK_REQUEST(8)
smpi_bench_end();
+ if (sendbuf == MPI_IN_PLACE) {
+ sendbuf = static_cast<char*>(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 MPI_SUCCESS;
+}
- 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 {
+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);
+}
- if (recvbuf == MPI_IN_PLACE) {
- recvtype = sendtype;
- recvcount = sendcount;
- }
- int rank = simgrid::s4u::Actor::self()->getPid();
+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)
+{
+ CHECK_COMM(8)
+ CHECK_BUFFER(1, sendbuf, sendcount)
+ 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)
+ CHECK_COUNT(2, sendcount)
+ CHECK_REQUEST(9)
+ for (int i = 0; i < comm->size(); i++) {
+ CHECK_COUNT(5, recvcounts[i])
+ CHECK_BUFFER(4, recvbuf, recvcounts[i])
+ }
- TRACE_smpi_comm_in(rank, __FUNCTION__,
- 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(), encode_datatype(sendtype),
- encode_datatype(recvtype)));
+ smpi_bench_end();
+ if (sendbuf == MPI_IN_PLACE) {
+ sendbuf = static_cast<char*>(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();
- simgrid::smpi::Colls::scatter(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm);
- retval = MPI_SUCCESS;
- TRACE_smpi_comm_out(rank);
+ std::vector<int>* trace_recvcounts = new std::vector<int>();
+ 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_Scatterv(void *sendbuf, int *sendcounts, int *displs,
- 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;
+ CHECK_COMM(8)
+ if(comm->rank() == root){
+ CHECK_BUFFER(1, sendbuf, sendcount)
+ CHECK_COUNT(2, sendcount)
+ CHECK_TYPE(3, sendtype)
+ }
+ if(recvbuf != MPI_IN_PLACE){
+ CHECK_BUFFER(4, recvbuf, recvcount)
+ CHECK_COUNT(5, recvcount)
+ CHECK_TYPE(6, 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",
+ 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 MPI_SUCCESS;
+}
- 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_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);
+}
+
+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_COMM(9)
+ if(recvbuf != MPI_IN_PLACE){
+ CHECK_BUFFER(4, recvbuf, recvcount)
+ CHECK_COUNT(5, recvcount)
+ CHECK_TYPE(7, recvtype)
+ }
+ CHECK_ROOT(9)
+ CHECK_REQUEST(10)
+ if (comm->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])
+ }
if (recvbuf == MPI_IN_PLACE) {
recvtype = sendtype;
recvcount = sendcounts[comm->rank()];
}
- int rank = simgrid::s4u::Actor::self()->getPid();
- int dt_size_send = sendtype->is_replayable() ? 1 : sendtype->size();
-
- std::vector<int>* trace_sendcounts = new std::vector<int>;
- 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, __FUNCTION__, new simgrid::instr::VarCollTIData(
- "scatterV", root, dt_size_send, trace_sendcounts,
- recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), nullptr,
- encode_datatype(sendtype), encode_datatype(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<int>* trace_sendcounts = new std::vector<int>();
+ 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::Actor::self()->getPid();
-
- TRACE_smpi_comm_in(rank, __FUNCTION__,
- new simgrid::instr::CollTIData("reduce", root, 0,
- datatype->is_replayable() ? count : count * datatype->size(), -1,
- encode_datatype(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)
+{
+ CHECK_COMM(7)
+ CHECK_BUFFER(1, sendbuf, count)
+ if(comm->rank() == root)
+ CHECK_BUFFER(5, recvbuf, count)
+ CHECK_TYPE(4, datatype)
+ CHECK_COUNT(3, count)
+ CHECK_OP(5)
+ CHECK_ROOT(7)
+ CHECK_REQUEST(8)
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;
+ CHECK_BUFFER(1, inbuf, count)
+ CHECK_BUFFER(2, inoutbuf, count)
+ CHECK_TYPE(4, datatype)
+ CHECK_COUNT(3, count)
+ CHECK_OP(5)
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<char*>(sendbuf);
- if( sendbuf == MPI_IN_PLACE ) {
- sendtmpbuf = static_cast<char*>(xbt_malloc(count*datatype->get_extent()));
- simgrid::smpi::Datatype::copy(recvbuf, count, datatype,sendtmpbuf, count, datatype);
- }
- int rank = simgrid::s4u::Actor::self()->getPid();
+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)
+ CHECK_TYPE(4, datatype)
+ CHECK_COUNT(3, count)
+ CHECK_REQUEST(7)
+ CHECK_OP(5)
- TRACE_smpi_comm_in(rank, __FUNCTION__,
- new simgrid::instr::CollTIData("allReduce", -1, 0,
- datatype->is_replayable() ? count : count * datatype->size(), -1,
- encode_datatype(datatype), ""));
+ smpi_bench_end();
+ std::unique_ptr<unsigned char[]> 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;
-
- smpi_bench_end();
+ return PMPI_Iscan(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::Actor::self()->getPid();
+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)
+ CHECK_TYPE(4, datatype)
+ CHECK_COUNT(3, count)
+ CHECK_REQUEST(7)
+ CHECK_OP(5)
- TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::Pt2PtTIData(
- "scan", -1, datatype->is_replayable() ? count : count * datatype->size(),
- encode_datatype(datatype)));
+ smpi_bench_end();
+ int rank = simgrid::s4u::this_actor::get_pid();
+ std::unique_ptr<unsigned char[]> tmp_sendbuf;
+ const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, count, datatype);
- retval = simgrid::smpi::Colls::scan(sendbuf, 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);
- }
+ 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::Actor::self()->getPid();
- void* sendtmpbuf = sendbuf;
- if (sendbuf == MPI_IN_PLACE) {
- sendtmpbuf = static_cast<void*>(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){
+ CHECK_COMM(6)
+ CHECK_BUFFER(1, sendbuf, count)
+ CHECK_BUFFER(2, recvbuf, count)
+ CHECK_TYPE(4, datatype)
+ CHECK_COUNT(3, count)
+ CHECK_REQUEST(7)
+ CHECK_OP(5)
- TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::Pt2PtTIData(
- "exscan", -1, datatype->is_replayable() ? count : count * datatype->size(),
- encode_datatype(datatype)));
+ smpi_bench_end();
+ int rank = simgrid::s4u::this_actor::get_pid();
+ std::unique_ptr<unsigned char[]> 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::Actor::self()->getPid();
- std::vector<int>* trace_recvcounts = new std::vector<int>;
- 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<void*>(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)
+{
+ CHECK_COMM(6)
+ CHECK_TYPE(4, datatype)
+ CHECK_NULL(3, MPI_ERR_COUNT, recvcounts)
+ CHECK_REQUEST(7)
+ CHECK_OP(5)
+ 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])
+ }
- TRACE_smpi_comm_in(rank, __FUNCTION__,
- new simgrid::instr::VarCollTIData("reduceScatter", -1, dt_send_size, nullptr, -1,
- trace_recvcounts, encode_datatype(datatype), ""));
+ smpi_bench_end();
+ int rank = simgrid::s4u::this_actor::get_pid();
+ std::vector<int>* trace_recvcounts = new std::vector<int>();
+ 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<unsigned char[]> 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::Actor::self()->getPid();
- int dt_send_size = datatype->is_replayable() ? 1 : datatype->size();
- std::vector<int>* trace_recvcounts = new std::vector<int>(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<void*>(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)
+{
+ CHECK_COMM(6)
+ CHECK_BUFFER(1, sendbuf, recvcount)
+ CHECK_BUFFER(2, recvbuf, recvcount)
+ CHECK_TYPE(4, datatype)
+ CHECK_COUNT(3, recvcount)
+ CHECK_REQUEST(7)
+ CHECK_OP(5)
- TRACE_smpi_comm_in(rank, __FUNCTION__,
- new simgrid::instr::VarCollTIData("reduceScatter", -1, 0, nullptr, -1, trace_recvcounts,
- encode_datatype(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<int>* trace_recvcounts = new std::vector<int>(recvcount * dt_send_size); // copy data to avoid bad free
+ std::unique_ptr<unsigned char[]> 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)
+{
+ CHECK_COMM(7)
+ CHECK_BUFFER(1, sendbuf, sendcount)
+ CHECK_BUFFER(4, recvbuf, recvcount)
+ if(sendbuf != MPI_IN_PLACE)
+ CHECK_TYPE(3, sendtype)
+ CHECK_TYPE(6, recvtype)
+ CHECK_COUNT(5, recvcount)
+ if(sendbuf != MPI_IN_PLACE)
+ CHECK_COUNT(2, sendcount)
+ CHECK_COUNT(5, recvcount)
+ CHECK_REQUEST(8)
- 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<unsigned char[]> 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* senddispls, MPI_Datatype sendtype, void* recvbuf,
+ const int* recvcounts, const int* recvdispls, MPI_Datatype recvtype, MPI_Comm comm)
{
- int retval = 0;
- smpi_bench_end();
+ return PMPI_Ialltoallv(sendbuf, sendcounts, senddispls, sendtype, recvbuf, recvcounts, recvdispls, 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::Actor::self()->getPid();
- void* sendtmpbuf = static_cast<char*>(sendbuf);
- int sendtmpcount = sendcount;
- MPI_Datatype sendtmptype = sendtype;
- if (sendbuf == MPI_IN_PLACE) {
- sendtmpbuf = static_cast<void*>(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* senddispls, MPI_Datatype sendtype, void* recvbuf,
+ const int* recvcounts, const int* recvdispls, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request)
+{
+ CHECK_COMM(9)
+ if(sendbuf != MPI_IN_PLACE){
+ CHECK_NULL(2, MPI_ERR_COUNT, sendcounts)
+ CHECK_NULL(3, MPI_ERR_ARG, senddispls)
+ CHECK_TYPE(4, sendtype)
+ }
+ CHECK_TYPE(8, recvtype)
+ CHECK_NULL(6, MPI_ERR_COUNT, recvcounts)
+ CHECK_NULL(7, MPI_ERR_ARG, recvdispls)
+ CHECK_REQUEST(10)
+
+ int rank = 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_BUFFER(5, recvbuf, recvcounts[i])
+ CHECK_COUNT(6, recvcounts[i])
+ }
- TRACE_smpi_comm_in(
- rank, __FUNCTION__,
- new simgrid::instr::CollTIData("allToAll", -1, -1.0,
- sendtmptype->is_replayable() ? sendtmpcount : sendtmpcount * sendtmptype->size(),
- recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(),
- encode_datatype(sendtmptype), encode_datatype(recvtype)));
+ smpi_bench_end();
+ int send_size = 0;
+ int recv_size = 0;
+ std::vector<int>* trace_sendcounts = new std::vector<int>();
+ std::vector<int>* trace_recvcounts = new std::vector<int>();
+ 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;
+ }
- retval = simgrid::smpi::Colls::alltoall(sendtmpbuf, sendtmpcount, sendtmptype, recvbuf, recvcount, recvtype, comm);
+ std::unique_ptr<unsigned char[]> tmp_sendbuf;
+ std::unique_ptr<int[]> tmp_sendcounts;
+ std::unique_ptr<int[]> 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();
+ 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_senddispls, real_sendtype, recvbuf,
+ recvcounts, recvdispls, recvtype, comm);
+ else
+ 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();
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* senddispls, const MPI_Datatype* sendtypes, void* recvbuf,
+ const int* recvcounts, const int* recvdispls, 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::Actor::self()->getPid();
- int size = comm->size();
- int send_size = 0;
- int recv_size = 0;
- std::vector<int>* trace_sendcounts = new std::vector<int>;
- std::vector<int>* trace_recvcounts = new std::vector<int>;
- int dt_size_recv = recvtype->size();
-
- void* sendtmpbuf = static_cast<char*>(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;
- }
+ return PMPI_Ialltoallw(sendbuf, sendcounts, senddispls, sendtypes, recvbuf, recvcounts, recvdispls, recvtypes, comm, MPI_REQUEST_IGNORED);
+}
- if (sendbuf == MPI_IN_PLACE) {
- sendtmpbuf = static_cast<void*>(xbt_malloc(maxsize));
- memcpy(sendtmpbuf, recvbuf, maxsize);
- sendtmpcounts = static_cast<int*>(xbt_malloc(size * sizeof(int)));
- memcpy(sendtmpcounts, recvcounts, size * sizeof(int));
- sendtmpdisps = static_cast<int*>(xbt_malloc(size * sizeof(int)));
- memcpy(sendtmpdisps, recvdisps, size * sizeof(int));
- sendtmptype = recvtype;
+int PMPI_Ialltoallw(const void* sendbuf, const int* sendcounts, const int* senddispls, const MPI_Datatype* sendtypes, void* recvbuf,
+ const int* recvcounts, const int* recvdispls, const MPI_Datatype* recvtypes, MPI_Comm comm, MPI_Request* request)
+{
+ CHECK_COMM(9)
+ if(sendbuf != MPI_IN_PLACE){
+ CHECK_NULL(2, MPI_ERR_COUNT, sendcounts)
+ CHECK_NULL(3, MPI_ERR_ARG, senddispls)
+ CHECK_NULL(4, MPI_ERR_TYPE, sendtypes)
+ }
+ CHECK_NULL(6, MPI_ERR_COUNT, recvcounts)
+ 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();
+ 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(5, recvbuf, recvcounts[i])
+ CHECK_COUNT(6, recvcounts[i])
+ CHECK_TYPE(8, recvtypes[i])
+ }
- int dt_size_send = sendtmptype->size();
+ smpi_bench_end();
- 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);
+ int send_size = 0;
+ int recv_size = 0;
+ std::vector<int>* trace_sendcounts = new std::vector<int>();
+ std::vector<int>* trace_recvcounts = new std::vector<int>();
+
+ 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;
+ 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());
+ }
- TRACE_smpi_comm_in(rank, __FUNCTION__, new simgrid::instr::VarCollTIData(
- "allToAllV", -1, send_size, trace_sendcounts, recv_size,
- trace_recvcounts, encode_datatype(sendtype), encode_datatype(recvtype)));
-
- retval = simgrid::smpi::Colls::alltoallv(sendtmpbuf, sendtmpcounts, sendtmpdisps, sendtmptype, recvbuf, recvcounts,
- recvdisps, recvtype, comm);
- TRACE_smpi_comm_out(rank);
+ std::unique_ptr<unsigned char[]> tmp_sendbuf;
+ std::unique_ptr<int[]> tmp_sendcounts;
+ std::unique_ptr<int[]> tmp_senddispls;
+ std::unique_ptr<MPI_Datatype[]> 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();
+ }
- 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_senddispls, real_sendtypes, recvbuf,
+ recvcounts, recvdispls, recvtypes, comm);
+ else
+ 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();
return retval;
}
-
-}