int PMPI_Waitany(int count, MPI_Request requests[], int *index, MPI_Status * status)
{
if (index == nullptr)
return MPI_ERR_ARG;
if (count <= 0)
return MPI_SUCCESS;
smpi_bench_end();
//for tracing, save the handles which might get overriden before we can use the helper on it
std::vector<MPI_Request> savedreqs(requests, requests + count);
for (MPI_Request& req : savedreqs) {
if (req != MPI_REQUEST_NULL && not(req->flags() & MPI_REQ_FINISHED))
req->ref();
else
req = MPI_REQUEST_NULL;
}
int rank_traced = simgrid::s4u::this_actor::get_pid(); // FIXME: In PMPI_Wait, we check if the comm is null?
TRACE_smpi_comm_in(rank_traced, __func__, new simgrid::instr::CpuTIData("waitAny", static_cast<double>(count)));
*index = simgrid::smpi::Request::waitany(count, requests, status);
if(*index!=MPI_UNDEFINED){
trace_smpi_recv_helper(&savedreqs[*index], status);
TRACE_smpi_comm_out(rank_traced);
}
for (MPI_Request& req : savedreqs)
if (req != MPI_REQUEST_NULL)
simgrid::smpi::Request::unref(&req);
smpi_bench_begin();
return MPI_SUCCESS;
}
int PMPI_Test(MPI_Request * request, int *flag, MPI_Status * status)
{
int retval = 0;
smpi_bench_end();
if (request == nullptr || flag == nullptr) {
retval = MPI_ERR_ARG;
} else if (*request == MPI_REQUEST_NULL) {
if (status != MPI_STATUS_IGNORE){
*flag= true;
simgrid::smpi::Status::empty(status);
}
retval = MPI_SUCCESS;
} else {
int my_proc_id = ((*request)->comm() != MPI_COMM_NULL) ? simgrid::s4u::this_actor::get_pid() : -1;
TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::NoOpTIData("test"));
*flag = simgrid::smpi::Request::test(request,status);
TRACE_smpi_comm_out(my_proc_id);
retval = MPI_SUCCESS;
}
smpi_bench_begin();
return retval;
}
int PMPI_Compare_and_swap(const void* origin_addr, void* compare_addr, void* result_addr, MPI_Datatype datatype,
int target_rank, MPI_Aint target_disp, MPI_Win win)
{
int retval = 0;
smpi_bench_end();
if (win == MPI_WIN_NULL) {
retval = MPI_ERR_WIN;
} else if (target_rank == MPI_PROC_NULL) {
retval = MPI_SUCCESS;
} else if (target_rank <0){
retval = MPI_ERR_RANK;
} else if (win->dynamic()==0 && target_disp <0){
//in case of dynamic window, target_disp can be mistakenly seen as negative, as it is an address
retval = MPI_ERR_ARG;
} else if (origin_addr==nullptr || result_addr==nullptr || compare_addr==nullptr){
retval = MPI_ERR_COUNT;
} else if ((datatype == MPI_DATATYPE_NULL) || (not datatype->is_valid())) {
retval = MPI_ERR_TYPE;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
MPI_Group group;
win->get_group(&group);
TRACE_smpi_comm_in(my_proc_id, __func__,
new simgrid::instr::Pt2PtTIData("Compare_and_swap", target_rank,
datatype->is_replayable() ? 1 : datatype->size(),
simgrid::smpi::Datatype::encode(datatype)));
retval = win->compare_and_swap(origin_addr, compare_addr, result_addr, datatype, target_rank, target_disp);
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
/*
* This function starts a request returned by init functions such as
* MPI_Send_init(), MPI_Ssend_init (see above), and friends.
* They should already have performed sanity checks.
*/
int PMPI_Start(MPI_Request * request)
{
int retval = 0;
smpi_bench_end();
if (request == nullptr || *request == MPI_REQUEST_NULL) {
retval = MPI_ERR_REQUEST;
} else {
MPI_Request req = *request;
int my_proc_id = (req->comm() != MPI_COMM_NULL) ? simgrid::s4u::this_actor::get_pid() : -1;
TRACE_smpi_comm_in(my_proc_id, __func__,
new simgrid::instr::Pt2PtTIData("Start", req->dst(),
req->size(),
req->tag(),
simgrid::smpi::Datatype::encode(req->type())));
if (not TRACE_smpi_view_internals() && req->flags() & MPI_REQ_SEND)
TRACE_smpi_send(my_proc_id, my_proc_id, getPid(req->comm(), req->dst()), req->tag(), req->size());
req->start();
if (not TRACE_smpi_view_internals() && req->flags() & MPI_REQ_RECV)
TRACE_smpi_recv(getPid(req->comm(), req->src()), my_proc_id, req->tag());
retval = MPI_SUCCESS;
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
int PMPI_Ssend(const void* buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm) {
int retval = 0;
smpi_bench_end();
if (comm == MPI_COMM_NULL) {
retval = MPI_ERR_COMM;
} else if (dst == MPI_PROC_NULL) {
retval = MPI_SUCCESS;
} else if (dst >= comm->group()->size() || dst <0){
retval = MPI_ERR_RANK;
} else if ((count < 0) || (buf==nullptr && count > 0)) {
retval = MPI_ERR_COUNT;
} else if (datatype==MPI_DATATYPE_NULL || not datatype->is_valid()) {
retval = MPI_ERR_TYPE;
} else if(tag<0 && tag != MPI_ANY_TAG){
retval = MPI_ERR_TAG;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
int dst_traced = getPid(comm, dst);
TRACE_smpi_comm_in(my_proc_id, __func__,
new simgrid::instr::Pt2PtTIData("Ssend", dst,
datatype->is_replayable() ? count : count * datatype->size(),
tag, simgrid::smpi::Datatype::encode(datatype)));
TRACE_smpi_send(my_proc_id, my_proc_id, dst_traced, tag, count * datatype->size());
simgrid::smpi::Request::ssend(buf, count, datatype, dst, tag, comm);
retval = MPI_SUCCESS;
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
int PMPI_Waitall(int count, MPI_Request requests[], MPI_Status status[])
{
smpi_bench_end();
//for tracing, save the handles which might get overriden before we can use the helper on it
std::vector<MPI_Request> savedreqs(requests, requests + count);
for (MPI_Request& req : savedreqs) {
if (req != MPI_REQUEST_NULL && not(req->flags() & MPI_REQ_FINISHED))
req->ref();
else
req = MPI_REQUEST_NULL;
}
int rank_traced = simgrid::s4u::this_actor::get_pid(); // FIXME: In PMPI_Wait, we check if the comm is null?
TRACE_smpi_comm_in(rank_traced, __func__, new simgrid::instr::CpuTIData("waitall", static_cast<double>(count)));
int retval = simgrid::smpi::Request::waitall(count, requests, status);
for (int i = 0; i < count; i++) {
trace_smpi_recv_helper(&savedreqs[i], status!=MPI_STATUSES_IGNORE ? &status[i]: MPI_STATUS_IGNORE);
}
TRACE_smpi_comm_out(rank_traced);
for (MPI_Request& req : savedreqs)
if (req != MPI_REQUEST_NULL)
simgrid::smpi::Request::unref(&req);
smpi_bench_begin();
return retval;
}
int PMPI_Sendrecv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, int dst, int sendtag, void* recvbuf,
int recvcount, MPI_Datatype recvtype, int src, int recvtag, MPI_Comm comm, MPI_Status* status)
{
int retval = 0;
smpi_bench_end();
if (comm == MPI_COMM_NULL) {
retval = MPI_ERR_COMM;
} else if (not sendtype->is_valid() || not recvtype->is_valid()) {
retval = MPI_ERR_TYPE;
} else if (src == MPI_PROC_NULL) {
if(status!=MPI_STATUS_IGNORE){
simgrid::smpi::Status::empty(status);
status->MPI_SOURCE = MPI_PROC_NULL;
}
if(dst != MPI_PROC_NULL)
simgrid::smpi::Request::send(sendbuf, sendcount, sendtype, dst, sendtag, comm);
retval = MPI_SUCCESS;
}else if (dst == MPI_PROC_NULL){
simgrid::smpi::Request::recv(recvbuf, recvcount, recvtype, src, recvtag, comm, status);
retval = MPI_SUCCESS;
}else if (dst >= comm->group()->size() || dst <0 ||
(src!=MPI_ANY_SOURCE && (src >= comm->group()->size() || src <0))){
retval = MPI_ERR_RANK;
} else if ((sendcount < 0 || recvcount<0) ||
(sendbuf==nullptr && sendcount > 0) || (recvbuf==nullptr && recvcount>0)) {
retval = MPI_ERR_COUNT;
} else if((sendtag<0 && sendtag != MPI_ANY_TAG)||(recvtag<0 && recvtag != MPI_ANY_TAG)){
retval = MPI_ERR_TAG;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
int dst_traced = getPid(comm, dst);
int src_traced = getPid(comm, src);
// FIXME: Hack the way to trace this one
std::vector<int>* dst_hack = new std::vector<int>;
std::vector<int>* src_hack = new std::vector<int>;
dst_hack->push_back(dst_traced);
src_hack->push_back(src_traced);
TRACE_smpi_comm_in(my_proc_id, __func__,
new simgrid::instr::VarCollTIData(
"sendRecv", -1, sendtype->is_replayable() ? sendcount : sendcount * sendtype->size(),
dst_hack, recvtype->is_replayable() ? recvcount : recvcount * recvtype->size(), src_hack,
simgrid::smpi::Datatype::encode(sendtype), simgrid::smpi::Datatype::encode(recvtype)));
TRACE_smpi_send(my_proc_id, my_proc_id, dst_traced, sendtag, sendcount * sendtype->size());
simgrid::smpi::Request::sendrecv(sendbuf, sendcount, sendtype, dst, sendtag, recvbuf, recvcount, recvtype, src,
recvtag, comm, status);
retval = MPI_SUCCESS;
TRACE_smpi_recv(src_traced, my_proc_id, recvtag);
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
int PMPI_Win_flush_local_all(MPI_Win win){
int retval = 0;
smpi_bench_end();
if (win == MPI_WIN_NULL) {
retval = MPI_ERR_WIN;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::NoOpTIData("Win_flush_local_all"));
retval = win->flush_local_all();
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
int PMPI_Win_unlock(int rank, MPI_Win win){
int retval = 0;
smpi_bench_end();
if (win == MPI_WIN_NULL) {
retval = MPI_ERR_WIN;
} else if (rank == MPI_PROC_NULL){
retval = MPI_SUCCESS;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::NoOpTIData("Win_unlock"));
retval = win->unlock(rank);
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
int PMPI_Win_start(MPI_Group group, int assert, MPI_Win win){
int retval = 0;
smpi_bench_end();
if (win == MPI_WIN_NULL) {
retval = MPI_ERR_WIN;
} else if (group==MPI_GROUP_NULL){
retval = MPI_ERR_GROUP;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::NoOpTIData("Win_start"));
retval = win->start(group,assert);
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
int PMPI_Testsome(int incount, MPI_Request requests[], int* outcount, int* indices, MPI_Status status[])
{
int retval = 0;
smpi_bench_end();
if (outcount == nullptr) {
retval = MPI_ERR_ARG;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::NoOpTIData("testsome"));
retval = simgrid::smpi::Request::testsome(incount, requests, outcount, indices, status);
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
int PMPI_Testall(int count, MPI_Request* requests, int* flag, MPI_Status* statuses)
{
int retval = 0;
smpi_bench_end();
if (flag == nullptr) {
retval = MPI_ERR_ARG;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::NoOpTIData("testall"));
retval = simgrid::smpi::Request::testall(count, requests, flag, statuses);
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
int PMPI_Recv(void *buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm, MPI_Status * status)
{
int retval = 0;
smpi_bench_end();
if (comm == MPI_COMM_NULL) {
retval = MPI_ERR_COMM;
} else if (src == MPI_PROC_NULL) {
if(status != MPI_STATUS_IGNORE){
simgrid::smpi::Status::empty(status);
status->MPI_SOURCE = MPI_PROC_NULL;
}
retval = MPI_SUCCESS;
} else if (src!=MPI_ANY_SOURCE && (src >= comm->group()->size() || src <0)){
retval = MPI_ERR_RANK;
} else if ((count < 0) || (buf==nullptr && count > 0)) {
retval = MPI_ERR_COUNT;
} else if (datatype==MPI_DATATYPE_NULL || not datatype->is_valid()) {
retval = MPI_ERR_TYPE;
} else if(tag<0 && tag != MPI_ANY_TAG){
retval = MPI_ERR_TAG;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
TRACE_smpi_comm_in(my_proc_id, __func__,
new simgrid::instr::Pt2PtTIData("recv", src,
datatype->is_replayable() ? count : count * datatype->size(),
tag, simgrid::smpi::Datatype::encode(datatype)));
simgrid::smpi::Request::recv(buf, count, datatype, src, tag, comm, status);
retval = MPI_SUCCESS;
// the src may not have been known at the beginning of the recv (MPI_ANY_SOURCE)
int src_traced=0;
if (status != MPI_STATUS_IGNORE)
src_traced = getPid(comm, status->MPI_SOURCE);
else
src_traced = getPid(comm, src);
if (not TRACE_smpi_view_internals()) {
TRACE_smpi_recv(src_traced, my_proc_id, tag);
}
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
int PMPI_Rget_accumulate(const void *origin_addr, int origin_count, MPI_Datatype origin_datatype, void *result_addr,
int result_count, MPI_Datatype result_datatype, int target_rank, MPI_Aint target_disp, int target_count,
MPI_Datatype target_datatype, MPI_Op op, MPI_Win win, MPI_Request* request){
int retval = 0;
smpi_bench_end();
if (win == MPI_WIN_NULL) {
retval = MPI_ERR_WIN;
} else if (target_rank == MPI_PROC_NULL) {
*request = MPI_REQUEST_NULL;
retval = MPI_SUCCESS;
} else if (target_rank <0){
retval = MPI_ERR_RANK;
} else if (win->dynamic()==0 && target_disp <0){
//in case of dynamic window, target_disp can be mistakenly seen as negative, as it is an address
retval = MPI_ERR_ARG;
} else if ((origin_count < 0 || target_count < 0 || result_count <0) ||
(origin_addr==nullptr && origin_count > 0 && op != MPI_NO_OP) ||
(result_addr==nullptr && result_count > 0)){
retval = MPI_ERR_COUNT;
} else if (((target_datatype == MPI_DATATYPE_NULL) || (result_datatype == MPI_DATATYPE_NULL)) ||
(((origin_datatype != MPI_DATATYPE_NULL) && (not origin_datatype->is_valid())) || (not target_datatype->is_valid()) || (not result_datatype->is_valid()))) {
retval = MPI_ERR_TYPE;
} else if (op == MPI_OP_NULL) {
retval = MPI_ERR_OP;
} else if(request == nullptr){
retval = MPI_ERR_REQUEST;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
MPI_Group group;
win->get_group(&group);
TRACE_smpi_comm_in(my_proc_id, __func__,
new simgrid::instr::Pt2PtTIData(
"Rget_accumulate", target_rank,
target_datatype->is_replayable() ? target_count : target_count * target_datatype->size(),
simgrid::smpi::Datatype::encode(target_datatype)));
retval = win->get_accumulate( origin_addr, origin_count, origin_datatype, result_addr,
result_count, result_datatype, target_rank, target_disp,
target_count, target_datatype, op, request);
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
int PMPI_Win_lock(int lock_type, int rank, int assert, MPI_Win win){
int retval = 0;
smpi_bench_end();
if (win == MPI_WIN_NULL) {
retval = MPI_ERR_WIN;
} else if (lock_type != MPI_LOCK_EXCLUSIVE &&
lock_type != MPI_LOCK_SHARED) {
retval = MPI_ERR_LOCKTYPE;
} else if (rank == MPI_PROC_NULL){
retval = MPI_SUCCESS;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::NoOpTIData("Win_lock"));
retval = win->lock(lock_type,rank,assert);
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
int PMPI_Irecv(void *buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm, MPI_Request * request)
{
int retval = 0;
smpi_bench_end();
if (request == nullptr) {
retval = MPI_ERR_ARG;
} else if (comm == MPI_COMM_NULL) {
retval = MPI_ERR_COMM;
} else if (src == MPI_PROC_NULL) {
*request = MPI_REQUEST_NULL;
retval = MPI_SUCCESS;
} else if (src!=MPI_ANY_SOURCE && (src >= comm->group()->size() || src <0)){
retval = MPI_ERR_RANK;
} else if ((count < 0) || (buf==nullptr && count > 0)) {
retval = MPI_ERR_COUNT;
} else if (datatype==MPI_DATATYPE_NULL || not datatype->is_valid()) {
retval = MPI_ERR_TYPE;
} else if(tag<0 && tag != MPI_ANY_TAG){
retval = MPI_ERR_TAG;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
TRACE_smpi_comm_in(my_proc_id, __func__,
new simgrid::instr::Pt2PtTIData("irecv", src,
datatype->is_replayable() ? count : count * datatype->size(),
tag, simgrid::smpi::Datatype::encode(datatype)));
*request = simgrid::smpi::Request::irecv(buf, count, datatype, src, tag, comm);
retval = MPI_SUCCESS;
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
if (retval != MPI_SUCCESS && request != nullptr)
*request = MPI_REQUEST_NULL;
return retval;
}
int PMPI_Put(const void *origin_addr, int origin_count, MPI_Datatype origin_datatype, int target_rank,
MPI_Aint target_disp, int target_count, MPI_Datatype target_datatype, MPI_Win win){
int retval = 0;
smpi_bench_end();
if (win == MPI_WIN_NULL) {
retval = MPI_ERR_WIN;
} else if (target_rank == MPI_PROC_NULL) {
retval = MPI_SUCCESS;
} else if (target_rank <0){
retval = MPI_ERR_RANK;
} else if (win->dynamic()==0 && target_disp <0){
//in case of dynamic window, target_disp can be mistakenly seen as negative, as it is an address
retval = MPI_ERR_ARG;
} else if ((origin_count < 0 || target_count < 0) ||
(origin_addr==nullptr && origin_count > 0)){
retval = MPI_ERR_COUNT;
} else if (((origin_datatype == MPI_DATATYPE_NULL) || (target_datatype == MPI_DATATYPE_NULL)) ||
((not origin_datatype->is_valid()) || (not target_datatype->is_valid()))) {
retval = MPI_ERR_TYPE;
} else {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
MPI_Group group;
win->get_group(&group);
int dst_traced = group->actor(target_rank)->get_pid();
TRACE_smpi_comm_in(my_proc_id, __func__,
new simgrid::instr::Pt2PtTIData("Put", target_rank, origin_datatype->is_replayable()
? origin_count
: origin_count * origin_datatype->size(),
simgrid::smpi::Datatype::encode(origin_datatype)));
TRACE_smpi_send(my_proc_id, my_proc_id, dst_traced, SMPI_RMA_TAG, origin_count * origin_datatype->size());
retval = win->put( origin_addr, origin_count, origin_datatype, target_rank, target_disp, target_count,
target_datatype);
TRACE_smpi_comm_out(my_proc_id);
}
smpi_bench_begin();
return retval;
}
int PMPI_Wait(MPI_Request * request, MPI_Status * status)
{
int retval = 0;
smpi_bench_end();
simgrid::smpi::Status::empty(status);
if (request == nullptr) {
retval = MPI_ERR_ARG;
} else if (*request == MPI_REQUEST_NULL) {
retval = MPI_SUCCESS;
} else {
//for tracing, save the handle which might get overriden before we can use the helper on it
MPI_Request savedreq = *request;
if (savedreq != MPI_REQUEST_NULL && not(savedreq->flags() & MPI_REQ_FINISHED)
&& not(savedreq->flags() & MPI_REQ_GENERALIZED))
savedreq->ref();//don't erase te handle in Request::wait, we'll need it later
else
savedreq = MPI_REQUEST_NULL;
int my_proc_id = (*request)->comm() != MPI_COMM_NULL
? simgrid::s4u::this_actor::get_pid()
: -1; // TODO: cheinrich: Check if this correct or if it should be MPI_UNDEFINED
TRACE_smpi_comm_in(my_proc_id, __func__,
new simgrid::instr::WaitTIData((*request)->src(), (*request)->dst(), (*request)->tag()));
retval = simgrid::smpi::Request::wait(request, status);
//the src may not have been known at the beginning of the recv (MPI_ANY_SOURCE)
TRACE_smpi_comm_out(my_proc_id);
trace_smpi_recv_helper(&savedreq, status);
if (savedreq != MPI_REQUEST_NULL)
simgrid::smpi::Request::unref(&savedreq);
}
smpi_bench_begin();
return retval;
}
int PMPI_Startall(int count, MPI_Request * requests)
{
int retval;
smpi_bench_end();
if (requests == nullptr) {
retval = MPI_ERR_ARG;
} else {
retval = MPI_SUCCESS;
for (int i = 0; i < count; i++) {
if(requests[i] == MPI_REQUEST_NULL) {
retval = MPI_ERR_REQUEST;
}
}
if(retval != MPI_ERR_REQUEST) {
int my_proc_id = simgrid::s4u::this_actor::get_pid();
TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::NoOpTIData("Startall"));
MPI_Request req = MPI_REQUEST_NULL;
if (not TRACE_smpi_view_internals())
for (int i = 0; i < count; i++) {
req = requests[i];
if (req->flags() & MPI_REQ_SEND)
TRACE_smpi_send(my_proc_id, my_proc_id, getPid(req->comm(), req->dst()), req->tag(), req->size());
}
simgrid::smpi::Request::startall(count, requests);
if (not TRACE_smpi_view_internals())
for (int i = 0; i < count; i++) {
req = requests[i];
if (req->flags() & MPI_REQ_RECV)
TRACE_smpi_recv(getPid(req->comm(), req->src()), my_proc_id, req->tag());
}
TRACE_smpi_comm_out(my_proc_id);
}
}
smpi_bench_begin();
return retval;
}