-/* Copyright (c) 2007-2021. The SimGrid Team. All rights reserved. */
+/* Copyright (c) 2007-2022. 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. */
{
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);
} else
simgrid::smpi::colls::ibarrier(comm, request);
- TRACE_smpi_comm_out(rank);
- smpi_bench_begin();
+ TRACE_smpi_comm_out(pid);
return MPI_SUCCESS;
}
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 pid = simgrid::s4u::this_actor::get_pid();
+ 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<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
+
if (comm->size() > 1) {
if (request == MPI_REQUEST_IGNORED)
simgrid::smpi::colls::bcast(buf, count, datatype, root, comm);
}
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return MPI_SUCCESS;
}
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(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;
}
}
- smpi_bench_end();
+ const SmpiBenchGuard suspend_bench;
+
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
- int pid = simgrid::s4u::this_actor::get_pid();
+ aid_t pid = simgrid::s4u::this_actor::get_pid();
TRACE_smpi_comm_in(pid, request == MPI_REQUEST_IGNORED ? "PMPI_Gather" : "PMPI_Igather",
new simgrid::instr::CollTIData(
request == MPI_REQUEST_IGNORED ? "gather" : "igather", root, -1.0,
- real_sendtype->is_replayable() ? real_sendcount : real_sendcount * real_sendtype->size(),
- (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);
request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return MPI_SUCCESS;
}
MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request)
{
CHECK_COMM(9)
+ SET_BUF1(sendbuf)
int rank = comm->rank();
- CHECK_BUFFER(1, sendbuf, sendcount)
if(sendbuf != MPI_IN_PLACE){
CHECK_TYPE(3, sendtype)
CHECK_COUNT(2, sendcount)
}
+ 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_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 (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<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
+
const void* real_sendbuf = sendbuf;
int real_sendcount = sendcount;
MPI_Datatype real_sendtype = sendtype;
real_sendtype = recvtype;
}
- int pid = 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 = std::make_shared<std::vector<int>>();
- if (rank == root) {
- for (int i = 0; i < comm->size(); i++) // copy data to avoid bad free
- trace_recvcounts->push_back(recvcounts[i] * dt_size_recv);
- }
+ 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(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,
root, comm, request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return MPI_SUCCESS;
}
MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request)
{
CHECK_COMM(7)
+ SET_BUF1(sendbuf)
+ SET_BUF2(recvbuf)
int rank = comm->rank();
- CHECK_BUFFER(1, sendbuf, sendcount)
- CHECK_BUFFER(4, recvbuf, recvcount)
CHECK_NOT_IN_PLACE(4, recvbuf)
if(sendbuf != MPI_IN_PLACE){
CHECK_COUNT(2, sendcount)
}
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<char*>(recvbuf) + recvtype->get_extent() * recvcount * comm->rank();
return MPI_ERR_TRUNCATE;
}
- smpi_bench_end();
+ const SmpiBenchGuard suspend_bench;
- int pid = simgrid::s4u::this_actor::get_pid();
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
+
+ aid_t pid = simgrid::s4u::this_actor::get_pid();
TRACE_smpi_comm_in(pid, request == MPI_REQUEST_IGNORED ? "PMPI_Allgather" : "PMPI_Iallggather",
new simgrid::instr::CollTIData(
request == MPI_REQUEST_IGNORED ? "allgather" : "iallgather", -1, -1.0,
- sendtype->is_replayable() ? sendcount : sendcount * sendtype->size(),
- 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);
simgrid::smpi::colls::iallgather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm, request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return MPI_SUCCESS;
}
MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request)
{
CHECK_COMM(8)
+ SET_BUF1(sendbuf)
+ SET_BUF2(recvbuf)
int rank = comm->rank();
- 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)
+ 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;
+
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
- 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 pid = 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 = std::make_shared<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_recvcounts->insert(trace_recvcounts->end(), &recvcounts[0], &recvcounts[comm->size()]);
TRACE_smpi_comm_in(
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);
request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return MPI_SUCCESS;
}
MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request)
{
CHECK_COMM(8)
+ SET_BUF2(recvbuf)
int rank = comm->rank();
if(rank == root){
+ SET_BUF1(sendbuf)
CHECK_NOT_IN_PLACE_ROOT(1, sendbuf)
- CHECK_BUFFER(1, sendbuf, sendcount)
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;
return MPI_ERR_TRUNCATE;
}
- smpi_bench_end();
+ const SmpiBenchGuard suspend_bench;
+
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
- int pid = simgrid::s4u::this_actor::get_pid();
+ 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,
- (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);
simgrid::smpi::colls::iscatter(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm, request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return MPI_SUCCESS;
}
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_NOT_IN_PLACE_ROOT(1, sendbuf)
- 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 (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;
} 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 pid = simgrid::s4u::this_actor::get_pid();
- int dt_size_send = sendtype->is_replayable() ? 1 : sendtype->size();
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
+
+ aid_t pid = simgrid::s4u::this_actor::get_pid();
auto trace_sendcounts = std::make_shared<std::vector<int>>();
- if (rank == root) {
- for (int i = 0; i < comm->size(); i++) { // copy data to avoid bad free
- trace_sendcounts->push_back(sendcounts[i] * dt_size_send);
- }
- }
+ 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)
request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return MPI_SUCCESS;
}
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)
+ SET_BUF1(sendbuf)
int rank = comm->rank();
- CHECK_BUFFER(1, sendbuf, count)
+ CHECK_TYPE(4, datatype)
+ CHECK_COUNT(3, count)
+ CHECK_BUFFER(1, sendbuf, count, datatype)
if(rank == root){
+ SET_BUF2(recvbuf)
CHECK_NOT_IN_PLACE(2, recvbuf)
- CHECK_BUFFER(5, recvbuf, count)
+ CHECK_BUFFER(5, recvbuf, count, datatype)
}
- CHECK_TYPE(4, datatype)
- CHECK_COUNT(3, count)
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))
+
+ const SmpiBenchGuard suspend_bench;
- smpi_bench_end();
- int pid = simgrid::s4u::this_actor::get_pid();
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
+
+ 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,
+ new simgrid::instr::CollTIData(request == MPI_REQUEST_IGNORED ? "reduce" : "ireduce", root, 0,
+ count, 0,
simgrid::smpi::Datatype::encode(datatype), ""));
if (request == MPI_REQUEST_IGNORED)
simgrid::smpi::colls::reduce(sendbuf, recvbuf, count, datatype, op, root, comm);
simgrid::smpi::colls::ireduce(sendbuf, recvbuf, count, datatype, op, root, comm, request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return MPI_SUCCESS;
}
int PMPI_Reduce_local(const void* inbuf, void* inoutbuf, int count, MPI_Datatype datatype, MPI_Op op)
{
- CHECK_BUFFER(1, inbuf, count)
- CHECK_BUFFER(2, inoutbuf, count)
+ SET_BUF1(inbuf)
+ SET_BUF2(inoutbuf)
CHECK_TYPE(4, datatype)
CHECK_COUNT(3, count)
+ CHECK_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;
}
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)
+ SET_BUF1(sendbuf)
+ SET_BUF2(recvbuf)
int rank = comm->rank();
- CHECK_BUFFER(1, sendbuf, count)
- CHECK_BUFFER(2, recvbuf, count)
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, op, datatype)
+ CHECK_COLLECTIVE(comm, std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Alleduce" : "PMPI_Iallreduce") +
+ " with op " + op->name())
+
+ const SmpiBenchGuard suspend_bench;
+
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
- smpi_bench_end();
std::vector<unsigned char> tmp_sendbuf;
const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, count, datatype);
- int pid = simgrid::s4u::this_actor::get_pid();
+ aid_t pid = simgrid::s4u::this_actor::get_pid();
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)
simgrid::smpi::colls::iallreduce(real_sendbuf, recvbuf, count, datatype, op, comm, request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return MPI_SUCCESS;
}
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, 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 pid = simgrid::s4u::this_actor::get_pid();
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
+
+ aid_t pid = simgrid::s4u::this_actor::get_pid();
std::vector<unsigned char> tmp_sendbuf;
const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, count, datatype);
TRACE_smpi_comm_in(pid, request == MPI_REQUEST_IGNORED ? "PMPI_Scan" : "PMPI_Iscan",
- new simgrid::instr::Pt2PtTIData(request == MPI_REQUEST_IGNORED ? "scan" : "iscan", -1,
- datatype->is_replayable() ? count : count * datatype->size(),
- simgrid::smpi::Datatype::encode(datatype)));
+ 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)
retval = simgrid::smpi::colls::iscan(real_sendbuf, recvbuf, count, datatype, op, comm, request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return retval;
}
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, op, datatype)
+ CHECK_COLLECTIVE(comm, std::string(request == MPI_REQUEST_IGNORED ? "PMPI_Exscan" : "PMPI_Iexscan") + " with op " +
+ op->name())
+
+ const SmpiBenchGuard suspend_bench;
- smpi_bench_end();
- int pid = simgrid::s4u::this_actor::get_pid();
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
+
+ aid_t pid = simgrid::s4u::this_actor::get_pid();
std::vector<unsigned char> tmp_sendbuf;
const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, count, datatype);
TRACE_smpi_comm_in(pid, request == MPI_REQUEST_IGNORED ? "PMPI_Exscan" : "PMPI_Iexscan",
- new simgrid::instr::Pt2PtTIData(request == MPI_REQUEST_IGNORED ? "exscan" : "iexscan", -1,
- datatype->is_replayable() ? count : count * datatype->size(),
- simgrid::smpi::Datatype::encode(datatype)));
+ 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)
retval = simgrid::smpi::colls::iexscan(real_sendbuf, recvbuf, count, datatype, op, comm, request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return retval;
}
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_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;
- smpi_bench_end();
- int pid = simgrid::s4u::this_actor::get_pid();
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
+
+ aid_t pid = simgrid::s4u::this_actor::get_pid();
auto trace_recvcounts = std::make_shared<std::vector<int>>();
- int dt_send_size = datatype->is_replayable() ? 1 : datatype->size();
+ trace_recvcounts->insert(trace_recvcounts->end(), &recvcounts[0], &recvcounts[comm->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<unsigned char> tmp_sendbuf;
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);
simgrid::smpi::colls::ireduce_scatter(real_sendbuf, recvbuf, recvcounts, datatype, op, comm, request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return MPI_SUCCESS;
}
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, 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;
+
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
- smpi_bench_end();
int count = comm->size();
- int pid = simgrid::s4u::this_actor::get_pid();
- int dt_send_size = datatype->is_replayable() ? 1 : datatype->size();
- auto trace_recvcounts = std::make_shared<std::vector<int>>(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<std::vector<int>>(recvcount);
+
std::vector<unsigned char> tmp_sendbuf;
const void* real_sendbuf = smpi_get_in_place_buf(sendbuf, recvbuf, tmp_sendbuf, recvcount * count, datatype);
TRACE_smpi_comm_in(
pid, request == MPI_REQUEST_IGNORED ? "PMPI_Reduce_scatter_block" : "PMPI_Ireduce_scatter_block",
- new simgrid::instr::VarCollTIData(request == MPI_REQUEST_IGNORED ? "reducescatter" : "ireducescatter", -1, 0,
+ new simgrid::instr::VarCollTIData(request == MPI_REQUEST_IGNORED ? "reducescatter" : "ireducescatter", -1, -1,
nullptr, -1, trace_recvcounts, simgrid::smpi::Datatype::encode(datatype), ""));
std::vector<int> recvcounts(count);
simgrid::smpi::colls::ireduce_scatter(real_sendbuf, recvbuf, recvcounts.data(), datatype, op, comm, request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return MPI_SUCCESS;
}
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 pid = simgrid::s4u::this_actor::get_pid();
+ aid_t pid = simgrid::s4u::this_actor::get_pid();
int real_sendcount = sendcount;
MPI_Datatype real_sendtype = sendtype;
return MPI_ERR_TRUNCATE;
}
- smpi_bench_end();
+ const SmpiBenchGuard suspend_bench;
+
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
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)
comm, request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return retval;
}
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)
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 pid = 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;
+
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
+
int send_size = 0;
int recv_size = 0;
auto trace_sendcounts = std::make_shared<std::vector<int>>();
auto trace_recvcounts = std::make_shared<std::vector<int>>();
+ 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<unsigned char> tmp_sendbuf;
std::vector<int> tmp_sendcounts;
std::vector<int> 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();
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();
-
- 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_sendcounts->insert(trace_sendcounts->end(), &real_sendcounts[0], &real_sendcounts[size]);
TRACE_smpi_comm_in(pid, request == MPI_REQUEST_IGNORED ? "PMPI_Alltoallv" : "PMPI_Ialltoallv",
new simgrid::instr::VarCollTIData(request == MPI_REQUEST_IGNORED ? "alltoallv" : "ialltoallv", -1,
recvcounts, recvdispls, recvtype, comm, request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return retval;
}
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)
CHECK_NULL(7, MPI_ERR_ARG, recvdispls)
CHECK_NULL(8, MPI_ERR_TYPE, recvtypes)
CHECK_REQUEST(10)
- int pid = 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;
+
+ if(simgrid::config::get_value<bool>("smpi/colls-inject-barrier"))
+ simgrid::smpi::colls::barrier(comm);
int send_size = 0;
int recv_size = 0;
auto trace_sendcounts = std::make_shared<std::vector<int>>();
auto trace_recvcounts = std::make_shared<std::vector<int>>();
+ 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)
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());
}
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(pid, request == MPI_REQUEST_IGNORED ? "PMPI_Alltoallw" : "PMPI_Ialltoallw",
recvcounts, recvdispls, recvtypes, comm, request);
TRACE_smpi_comm_out(pid);
- smpi_bench_begin();
return retval;
}