1 /* Copyright (c) 2007-2021. The SimGrid Team. All rights reserved. */
3 /* This program is free software; you can redistribute it and/or modify it
4 * under the terms of the license (GNU LGPL) which comes with this package. */
6 #include "smpi_request.hpp"
10 #include "simgrid/Exception.hpp"
11 #include "simgrid/s4u/Exec.hpp"
12 #include "simgrid/s4u/Mutex.hpp"
13 #include "simgrid/s4u/ConditionVariable.hpp"
14 #include "smpi_comm.hpp"
15 #include "smpi_datatype.hpp"
16 #include "smpi_host.hpp"
17 #include "smpi_op.hpp"
18 #include "src/kernel/activity/CommImpl.hpp"
19 #include "src/mc/mc_replay.hpp"
20 #include "src/smpi/include/smpi_actor.hpp"
25 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_request, smpi, "Logging specific to SMPI (request)");
27 static simgrid::config::Flag<double> smpi_iprobe_sleep(
28 "smpi/iprobe", "Minimum time to inject inside a call to MPI_Iprobe", 1e-4);
29 static simgrid::config::Flag<double> smpi_test_sleep(
30 "smpi/test", "Minimum time to inject inside a call to MPI_Test", 1e-4);
32 std::vector<s_smpi_factor_t> smpi_ois_values;
34 extern void (*smpi_comm_copy_data_callback)(simgrid::kernel::activity::CommImpl*, void*, size_t);
39 Request::Request(const void* buf, int count, MPI_Datatype datatype, int src, int dst, int tag, MPI_Comm comm,
40 unsigned flags, MPI_Op op)
41 : buf_(const_cast<void*>(buf))
43 , size_(datatype->size() * count)
53 if(op != MPI_REPLACE && op != MPI_OP_NULL)
57 detached_sender_ = nullptr;
62 if (flags & MPI_REQ_PERSISTENT)
66 nbc_requests_=nullptr;
76 void Request::unref(MPI_Request* request)
78 xbt_assert(*request != MPI_REQUEST_NULL, "freeing an already free request");
80 (*request)->refcount_--;
81 if ((*request)->refcount_ < 0) {
82 (*request)->print_request("wrong refcount");
83 xbt_die("Whoops, wrong refcount");
85 if ((*request)->refcount_ == 0) {
86 if ((*request)->flags_ & MPI_REQ_GENERALIZED) {
87 ((*request)->generalized_funcs)->free_fn(((*request)->generalized_funcs)->extra_state);
89 Comm::unref((*request)->comm_);
90 Datatype::unref((*request)->old_type_);
92 if ((*request)->op_ != MPI_REPLACE && (*request)->op_ != MPI_OP_NULL)
93 Op::unref(&(*request)->op_);
95 (*request)->print_request("Destroying");
96 F2C::free_f((*request)->f2c_id());
98 *request = MPI_REQUEST_NULL;
100 (*request)->print_request("Decrementing");
104 bool Request::match_common(MPI_Request req, MPI_Request sender, MPI_Request receiver)
106 xbt_assert(sender, "Cannot match against null sender");
107 xbt_assert(receiver, "Cannot match against null receiver");
108 XBT_DEBUG("Trying to match %s of sender src %d against %d, tag %d against %d, id %d against %d",
109 (req == receiver ? "send" : "recv"), sender->src_, receiver->src_, sender->tag_, receiver->tag_,
110 sender->comm_->id(), receiver->comm_->id());
112 if ((receiver->comm_->id() == MPI_UNDEFINED || sender->comm_->id() == MPI_UNDEFINED ||
113 receiver->comm_->id() == sender->comm_->id()) &&
114 ((receiver->src_ == MPI_ANY_SOURCE && (receiver->comm_->group()->rank(sender->src_) != MPI_UNDEFINED)) ||
115 receiver->src_ == sender->src_) &&
116 ((receiver->tag_ == MPI_ANY_TAG && sender->tag_ >= 0) || receiver->tag_ == sender->tag_)) {
117 // we match, we can transfer some values
118 if (receiver->src_ == MPI_ANY_SOURCE)
119 receiver->real_src_ = sender->src_;
120 if (receiver->tag_ == MPI_ANY_TAG)
121 receiver->real_tag_ = sender->tag_;
122 if ((receiver->flags_ & MPI_REQ_PROBE) == 0 ){
123 if (receiver->real_size_ < sender->real_size_){
124 XBT_DEBUG("Truncating message - should not happen: receiver size : %zu < sender size : %zu", receiver->real_size_, sender->real_size_);
125 receiver->truncated_ = true;
126 } else if (receiver->real_size_ > sender->real_size_){
127 receiver->real_size_=sender->real_size_;
130 if (sender->detached_)
131 receiver->detached_sender_ = sender; // tie the sender to the receiver, as it is detached and has to be freed in
133 req->flags_ |= MPI_REQ_MATCHED; // mark as impossible to cancel anymore
134 XBT_DEBUG("match succeeded");
140 void Request::init_buffer(int count){
141 void *old_buf = nullptr;
142 // FIXME Handle the case of a partial shared malloc.
143 // This part handles the problem of non-contiguous memory (for the unserialization at the reception)
144 if ((((flags_ & MPI_REQ_RECV) != 0) && ((flags_ & MPI_REQ_ACCUMULATE) != 0)) || (old_type_->flags() & DT_FLAG_DERIVED)) {
145 // This part handles the problem of non-contiguous memory
150 buf_ = xbt_malloc(count*old_type_->size());
151 if ((old_type_->flags() & DT_FLAG_DERIVED) && ((flags_ & MPI_REQ_SEND) != 0)) {
152 old_type_->serialize(old_buf, buf_, count);
159 bool Request::match_recv(void* a, void* b, simgrid::kernel::activity::CommImpl*)
161 auto ref = static_cast<MPI_Request>(a);
162 auto req = static_cast<MPI_Request>(b);
163 return match_common(req, req, ref);
166 bool Request::match_send(void* a, void* b, simgrid::kernel::activity::CommImpl*)
168 auto ref = static_cast<MPI_Request>(a);
169 auto req = static_cast<MPI_Request>(b);
170 return match_common(req, ref, req);
173 void Request::print_request(const char* message) const
175 XBT_VERB("%s request %p [buf = %p, size = %zu, src = %d, dst = %d, tag = %d, flags = %x]",
176 message, this, buf_, size_, src_, dst_, tag_, flags_);
179 /* factories, to hide the internal flags from the caller */
180 MPI_Request Request::bsend_init(const void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
182 return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
183 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
184 MPI_REQ_PERSISTENT | MPI_REQ_SEND | MPI_REQ_PREPARED | MPI_REQ_BSEND);
187 MPI_Request Request::send_init(const void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
189 return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
190 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
191 MPI_REQ_PERSISTENT | MPI_REQ_SEND | MPI_REQ_PREPARED);
194 MPI_Request Request::ssend_init(const void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
196 return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
197 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
198 MPI_REQ_PERSISTENT | MPI_REQ_SSEND | MPI_REQ_SEND | MPI_REQ_PREPARED);
201 MPI_Request Request::isend_init(const void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
203 return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
204 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
205 MPI_REQ_PERSISTENT | MPI_REQ_ISEND | MPI_REQ_SEND | MPI_REQ_PREPARED);
209 MPI_Request Request::rma_send_init(const void *buf, int count, MPI_Datatype datatype, int src, int dst, int tag, MPI_Comm comm,
212 MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
214 request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, comm->group()->actor_pid(src),
215 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
216 MPI_REQ_RMA | MPI_REQ_NON_PERSISTENT | MPI_REQ_ISEND | MPI_REQ_SEND | MPI_REQ_PREPARED);
218 request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, comm->group()->actor_pid(src),
219 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
220 MPI_REQ_RMA | MPI_REQ_NON_PERSISTENT | MPI_REQ_ISEND | MPI_REQ_SEND | MPI_REQ_PREPARED |
227 MPI_Request Request::recv_init(void *buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm)
229 int source = MPI_PROC_NULL;
230 if (src == MPI_ANY_SOURCE)
231 source = MPI_ANY_SOURCE;
232 else if (src != MPI_PROC_NULL)
233 source = comm->group()->actor_pid(src);
234 return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype,
236 simgrid::s4u::this_actor::get_pid(), tag, comm,
237 MPI_REQ_PERSISTENT | MPI_REQ_RECV | MPI_REQ_PREPARED);
240 MPI_Request Request::rma_recv_init(void *buf, int count, MPI_Datatype datatype, int src, int dst, int tag, MPI_Comm comm,
243 MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
244 int source = MPI_PROC_NULL;
245 if (src == MPI_ANY_SOURCE)
246 source = MPI_ANY_SOURCE;
247 else if (src != MPI_PROC_NULL)
248 source = comm->group()->actor_pid(src);
250 request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, source,
251 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
252 MPI_REQ_RMA | MPI_REQ_NON_PERSISTENT | MPI_REQ_RECV | MPI_REQ_PREPARED);
255 new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, source,
256 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
257 MPI_REQ_RMA | MPI_REQ_NON_PERSISTENT | MPI_REQ_RECV | MPI_REQ_PREPARED | MPI_REQ_ACCUMULATE, op);
262 MPI_Request Request::irecv_init(void *buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm)
264 int source = MPI_PROC_NULL;
265 if (src == MPI_ANY_SOURCE)
266 source = MPI_ANY_SOURCE;
267 else if (src != MPI_PROC_NULL)
268 source = comm->group()->actor_pid(src);
269 return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype,
270 source, simgrid::s4u::this_actor::get_pid(), tag, comm,
271 MPI_REQ_PERSISTENT | MPI_REQ_RECV | MPI_REQ_PREPARED);
274 MPI_Request Request::ibsend(const void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
276 MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
277 request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
278 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
279 MPI_REQ_NON_PERSISTENT | MPI_REQ_ISEND | MPI_REQ_SEND | MPI_REQ_BSEND);
280 if(dst != MPI_PROC_NULL)
285 MPI_Request Request::isend(const void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
287 MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
288 request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
289 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
290 MPI_REQ_NON_PERSISTENT | MPI_REQ_ISEND | MPI_REQ_SEND);
291 if(dst != MPI_PROC_NULL)
296 MPI_Request Request::issend(const void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
298 MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
299 request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
300 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
301 MPI_REQ_NON_PERSISTENT | MPI_REQ_ISEND | MPI_REQ_SSEND | MPI_REQ_SEND);
302 if(dst != MPI_PROC_NULL)
308 MPI_Request Request::irecv(void *buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm)
310 MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
311 int source = MPI_PROC_NULL;
312 if (src == MPI_ANY_SOURCE)
313 source = MPI_ANY_SOURCE;
314 else if (src != MPI_PROC_NULL)
315 source = comm->group()->actor_pid(src);
316 request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype,
317 source, simgrid::s4u::this_actor::get_pid(), tag, comm,
318 MPI_REQ_NON_PERSISTENT | MPI_REQ_RECV);
319 if(src != MPI_PROC_NULL)
324 int Request::recv(void *buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm, MPI_Status * status)
326 MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
327 request = irecv(buf, count, datatype, src, tag, comm);
328 int retval = wait(&request,status);
333 void Request::bsend(const void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
335 MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
336 request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
337 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
338 MPI_REQ_NON_PERSISTENT | MPI_REQ_SEND | MPI_REQ_BSEND);
340 if(dst != MPI_PROC_NULL)
342 wait(&request, MPI_STATUS_IGNORE);
346 void Request::send(const void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
348 MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
349 request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
350 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
351 MPI_REQ_NON_PERSISTENT | MPI_REQ_SEND);
352 if(dst != MPI_PROC_NULL)
354 wait(&request, MPI_STATUS_IGNORE);
358 void Request::ssend(const void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
360 MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
361 request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
362 dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL, tag, comm,
363 MPI_REQ_NON_PERSISTENT | MPI_REQ_SSEND | MPI_REQ_SEND);
365 if(dst != MPI_PROC_NULL)
367 wait(&request,MPI_STATUS_IGNORE);
371 void Request::sendrecv(const void *sendbuf, int sendcount, MPI_Datatype sendtype,int dst, int sendtag,
372 void *recvbuf, int recvcount, MPI_Datatype recvtype, int src, int recvtag,
373 MPI_Comm comm, MPI_Status * status)
375 int source = MPI_PROC_NULL;
376 if (src == MPI_ANY_SOURCE)
377 source = MPI_ANY_SOURCE;
378 else if (src != MPI_PROC_NULL)
379 source = comm->group()->actor_pid(src);
380 int destination = dst != MPI_PROC_NULL ? comm->group()->actor_pid(dst) : MPI_PROC_NULL;
382 std::array<MPI_Request, 2> requests;
383 std::array<MPI_Status, 2> stats;
384 int myid = simgrid::s4u::this_actor::get_pid();
385 if ((destination == myid) && (source == myid)) {
386 Datatype::copy(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype);
387 if (status != MPI_STATUS_IGNORE) {
388 status->MPI_SOURCE = source;
389 status->MPI_TAG = recvtag;
390 status->MPI_ERROR = MPI_SUCCESS;
391 status->count = sendcount * sendtype->size();
395 requests[0] = isend_init(sendbuf, sendcount, sendtype, dst, sendtag, comm);
396 requests[1] = irecv_init(recvbuf, recvcount, recvtype, src, recvtag, comm);
397 startall(2, requests.data());
398 waitall(2, requests.data(), stats.data());
401 if(status != MPI_STATUS_IGNORE) {
402 // Copy receive status
407 void Request::start()
409 s4u::Mailbox* mailbox;
411 xbt_assert(action_ == nullptr, "Cannot (re-)start unfinished communication");
412 //reinitialize temporary buffer for persistent requests
413 if(real_size_ > 0 && flags_ & MPI_REQ_FINISHED){
415 init_buffer(real_size_/old_type_->size());
417 flags_ &= ~MPI_REQ_PREPARED;
418 flags_ &= ~MPI_REQ_FINISHED;
421 // we make a copy here, as the size is modified by simix, and we may reuse the request in another receive later
423 if ((flags_ & MPI_REQ_RECV) != 0) {
424 this->print_request("New recv");
426 simgrid::smpi::ActorExt* process = smpi_process_remote(simgrid::s4u::Actor::by_pid(dst_));
428 simgrid::s4u::MutexPtr mut = process->mailboxes_mutex();
429 if (smpi_cfg_async_small_thresh() != 0 || (flags_ & MPI_REQ_RMA) != 0)
432 if (smpi_cfg_async_small_thresh() == 0 && (flags_ & MPI_REQ_RMA) == 0) {
433 mailbox = process->mailbox();
434 } else if (((flags_ & MPI_REQ_RMA) != 0) || static_cast<int>(size_) < smpi_cfg_async_small_thresh()) {
435 //We have to check both mailboxes (because SSEND messages are sent to the large mbox).
436 //begin with the more appropriate one : the small one.
437 mailbox = process->mailbox_small();
438 XBT_DEBUG("Is there a corresponding send already posted in the small mailbox %s (in case of SSEND)?",
439 mailbox->get_cname());
440 simgrid::kernel::activity::ActivityImplPtr action = mailbox->iprobe(0, &match_recv, static_cast<void*>(this));
442 if (action == nullptr) {
443 mailbox = process->mailbox();
444 XBT_DEBUG("No, nothing in the small mailbox test the other one : %s", mailbox->get_cname());
445 action = mailbox->iprobe(0, &match_recv, static_cast<void*>(this));
446 if (action == nullptr) {
447 XBT_DEBUG("Still nothing, switch back to the small mailbox : %s", mailbox->get_cname());
448 mailbox = process->mailbox_small();
451 XBT_DEBUG("yes there was something for us in the large mailbox");
454 mailbox = process->mailbox_small();
455 XBT_DEBUG("Is there a corresponding send already posted the small mailbox?");
456 simgrid::kernel::activity::ActivityImplPtr action = mailbox->iprobe(0, &match_recv, static_cast<void*>(this));
458 if (action == nullptr) {
459 XBT_DEBUG("No, nothing in the permanent receive mailbox");
460 mailbox = process->mailbox();
462 XBT_DEBUG("yes there was something for us in the small mailbox");
466 action_ = simcall_comm_irecv(
467 process->get_actor()->get_impl(), mailbox->get_impl(), buf_, &real_size_, &match_recv,
468 process->replaying() ? &smpi_comm_null_copy_buffer_callback : smpi_comm_copy_data_callback, this, -1.0);
469 XBT_DEBUG("recv simcall posted");
471 if (smpi_cfg_async_small_thresh() != 0 || (flags_ & MPI_REQ_RMA) != 0)
473 } else { /* the RECV flag was not set, so this is a send */
474 const simgrid::smpi::ActorExt* process = smpi_process_remote(simgrid::s4u::Actor::by_pid(dst_));
475 xbt_assert(process, "Actor pid=%d is gone??", dst_);
476 if (TRACE_smpi_view_internals())
477 TRACE_smpi_send(src_, src_, dst_, tag_, size_);
478 this->print_request("New send");
481 if ((flags_ & MPI_REQ_SSEND) == 0 &&
482 ((flags_ & MPI_REQ_RMA) != 0 || (flags_ & MPI_REQ_BSEND) != 0 ||
483 static_cast<int>(size_) < smpi_cfg_detached_send_thresh())) {
484 void *oldbuf = nullptr;
486 XBT_DEBUG("Send request %p is detached", this);
488 if (not(old_type_->flags() & DT_FLAG_DERIVED)) {
490 if (not process->replaying() && oldbuf != nullptr && size_ != 0) {
491 if ((smpi_cfg_privatization() != SmpiPrivStrategies::NONE) &&
492 (static_cast<char*>(buf_) >= smpi_data_exe_start) &&
493 (static_cast<char*>(buf_) < smpi_data_exe_start + smpi_data_exe_size)) {
494 XBT_DEBUG("Privatization : We are sending from a zone inside global memory. Switch data segment ");
495 smpi_switch_data_segment(simgrid::s4u::Actor::by_pid(src_));
497 //we need this temporary buffer even for bsend, as it will be released in the copy callback and we don't have a way to differentiate it
498 //so actually ... don't use manually attached buffer space.
499 buf = xbt_malloc(size_);
500 memcpy(buf,oldbuf,size_);
501 XBT_DEBUG("buf %p copied into %p",oldbuf,buf);
506 //if we are giving back the control to the user without waiting for completion, we have to inject timings
507 double sleeptime = 0.0;
508 if (detached_ || ((flags_ & (MPI_REQ_ISEND | MPI_REQ_SSEND)) != 0)) { // issend should be treated as isend
509 // isend and send timings may be different
510 sleeptime = ((flags_ & MPI_REQ_ISEND) != 0)
511 ? simgrid::s4u::Actor::self()->get_host()->extension<simgrid::smpi::Host>()->oisend(size_)
512 : simgrid::s4u::Actor::self()->get_host()->extension<simgrid::smpi::Host>()->osend(size_);
516 simgrid::s4u::this_actor::sleep_for(sleeptime);
517 XBT_DEBUG("sending size of %zu : sleep %f ", size_, sleeptime);
520 simgrid::s4u::MutexPtr mut = process->mailboxes_mutex();
522 if (smpi_cfg_async_small_thresh() != 0 || (flags_ & MPI_REQ_RMA) != 0)
525 if (not(smpi_cfg_async_small_thresh() != 0 || (flags_ & MPI_REQ_RMA) != 0)) {
526 mailbox = process->mailbox();
527 } else if (((flags_ & MPI_REQ_RMA) != 0) || static_cast<int>(size_) < smpi_cfg_async_small_thresh()) { // eager mode
528 mailbox = process->mailbox();
529 XBT_DEBUG("Is there a corresponding recv already posted in the large mailbox %s?", mailbox->get_cname());
530 simgrid::kernel::activity::ActivityImplPtr action = mailbox->iprobe(1, &match_send, static_cast<void*>(this));
531 if (action == nullptr) {
532 if ((flags_ & MPI_REQ_SSEND) == 0) {
533 mailbox = process->mailbox_small();
534 XBT_DEBUG("No, nothing in the large mailbox, message is to be sent on the small one %s",
535 mailbox->get_cname());
537 mailbox = process->mailbox_small();
538 XBT_DEBUG("SSEND : Is there a corresponding recv already posted in the small mailbox %s?",
539 mailbox->get_cname());
540 action = mailbox->iprobe(1, &match_send, static_cast<void*>(this));
541 if (action == nullptr) {
542 XBT_DEBUG("No, we are first, send to large mailbox");
543 mailbox = process->mailbox();
547 XBT_DEBUG("Yes there was something for us in the large mailbox");
550 mailbox = process->mailbox();
551 XBT_DEBUG("Send request %p is in the large mailbox %s (buf: %p)", this, mailbox->get_cname(), buf_);
554 size_t payload_size_ = size_ + 16;//MPI enveloppe size (tag+dest+communicator)
555 action_ = simcall_comm_isend(
556 simgrid::kernel::actor::ActorImpl::by_pid(src_), mailbox->get_impl(), payload_size_, -1.0, buf, real_size_,
558 &xbt_free_f, // how to free the userdata if a detached send fails
559 process->replaying() ? &smpi_comm_null_copy_buffer_callback : smpi_comm_copy_data_callback, this,
560 // detach if msg size < eager/rdv switch limit
562 XBT_DEBUG("send simcall posted");
564 /* FIXME: detached sends are not traceable (action_ == nullptr) */
565 if (action_ != nullptr) {
566 boost::static_pointer_cast<kernel::activity::CommImpl>(action_)->set_tracing_category(
567 smpi_process()->get_tracing_category());
570 if (smpi_cfg_async_small_thresh() != 0 || ((flags_ & MPI_REQ_RMA) != 0))
575 void Request::startall(int count, MPI_Request * requests)
577 if(requests== nullptr)
580 for(int i = 0; i < count; i++) {
581 if(requests[i]->src_ != MPI_PROC_NULL && requests[i]->dst_ != MPI_PROC_NULL)
582 requests[i]->start();
586 void Request::cancel()
588 this->flags_ |= MPI_REQ_CANCELLED;
589 if (this->action_ != nullptr)
590 (boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(this->action_))->cancel();
593 int Request::test(MPI_Request * request, MPI_Status * status, int* flag) {
594 // assume that *request is not MPI_REQUEST_NULL (filtered in PMPI_Test or testall before)
595 // to avoid deadlocks if used as a break condition, such as
596 // while (MPI_Test(request, flag, status) && flag) dostuff...
597 // because the time will not normally advance when only calls to MPI_Test are made -> deadlock
598 // multiplier to the sleeptime, to increase speed of execution, each failed test will increase it
599 xbt_assert(*request != MPI_REQUEST_NULL);
601 static int nsleeps = 1;
602 int ret = MPI_SUCCESS;
604 // Are we testing a request meant for non blocking collectives ?
605 // If so, test all the subrequests.
606 if ((*request)->nbc_requests_size_>0){
607 ret = testall((*request)->nbc_requests_size_, (*request)->nbc_requests_, flag, MPI_STATUSES_IGNORE);
609 delete[] (*request)->nbc_requests_;
610 (*request)->nbc_requests_size_=0;
616 if(smpi_test_sleep > 0)
617 simgrid::s4u::this_actor::sleep_for(nsleeps * smpi_test_sleep);
619 Status::empty(status);
621 if (((*request)->flags_ & (MPI_REQ_PREPARED | MPI_REQ_FINISHED)) == 0) {
622 if ((*request)->action_ != nullptr && ((*request)->flags_ & MPI_REQ_CANCELLED) == 0){
624 *flag = simcall_comm_test((*request)->action_.get());
625 } catch (const Exception&) {
630 if (((*request)->flags_ & MPI_REQ_GENERALIZED) && !((*request)->flags_ & MPI_REQ_COMPLETE))
633 finish_wait(request, status); // may invalidate *request
634 if (*request != MPI_REQUEST_NULL && ((*request)->flags_ & MPI_REQ_GENERALIZED)){
635 MPI_Status tmp_status;
636 MPI_Status* mystatus;
637 if (status == MPI_STATUS_IGNORE) {
638 mystatus = &tmp_status;
639 Status::empty(mystatus);
643 ret = ((*request)->generalized_funcs)->query_fn(((*request)->generalized_funcs)->extra_state, mystatus);
645 nsleeps=1;//reset the number of sleeps we will do next time
646 if (*request != MPI_REQUEST_NULL && ((*request)->flags_ & MPI_REQ_PERSISTENT) == 0)
647 *request = MPI_REQUEST_NULL;
648 } else if (smpi_cfg_grow_injected_times()) {
655 int Request::testsome(int incount, MPI_Request requests[], int *count, int *indices, MPI_Status status[])
661 MPI_Status *pstat = status == MPI_STATUSES_IGNORE ? MPI_STATUS_IGNORE : &stat;
664 for (int i = 0; i < incount; i++) {
665 if (requests[i] != MPI_REQUEST_NULL && not (requests[i]->flags_ & MPI_REQ_FINISHED)) {
666 int ret = test(&requests[i], pstat, &flag);
671 if (status != MPI_STATUSES_IGNORE)
672 status[*count] = *pstat;
674 if ((requests[i] != MPI_REQUEST_NULL) && (requests[i]->flags_ & MPI_REQ_NON_PERSISTENT))
675 requests[i] = MPI_REQUEST_NULL;
681 if(count_dead==incount)*count=MPI_UNDEFINED;
683 return MPI_ERR_IN_STATUS;
688 int Request::testany(int count, MPI_Request requests[], int *index, int* flag, MPI_Status * status)
690 std::vector<simgrid::kernel::activity::CommImpl*> comms;
691 comms.reserve(count);
695 int ret = MPI_SUCCESS;
696 *index = MPI_UNDEFINED;
698 std::vector<int> map; /** Maps all matching comms back to their location in requests **/
699 for(i = 0; i < count; i++) {
700 if ((requests[i] != MPI_REQUEST_NULL) && requests[i]->action_ && not(requests[i]->flags_ & MPI_REQ_PREPARED)) {
701 comms.push_back(static_cast<simgrid::kernel::activity::CommImpl*>(requests[i]->action_.get()));
705 if (not map.empty()) {
706 //multiplier to the sleeptime, to increase speed of execution, each failed testany will increase it
707 static int nsleeps = 1;
708 if(smpi_test_sleep > 0)
709 simgrid::s4u::this_actor::sleep_for(nsleeps * smpi_test_sleep);
711 i = simcall_comm_testany(comms.data(), comms.size()); // The i-th element in comms matches!
712 } catch (const Exception&) {
713 XBT_DEBUG("Exception in testany");
717 if (i != -1) { // -1 is not MPI_UNDEFINED but a SIMIX return code. (nothing matches)
719 if (requests[*index] != MPI_REQUEST_NULL &&
720 (requests[*index]->flags_ & MPI_REQ_GENERALIZED)
721 && !(requests[*index]->flags_ & MPI_REQ_COMPLETE)) {
724 finish_wait(&requests[*index],status);
725 if (requests[*index] != MPI_REQUEST_NULL && (requests[*index]->flags_ & MPI_REQ_GENERALIZED)){
726 MPI_Status tmp_status;
727 MPI_Status* mystatus;
728 if (status == MPI_STATUS_IGNORE) {
729 mystatus = &tmp_status;
730 Status::empty(mystatus);
734 ret=(requests[*index]->generalized_funcs)->query_fn((requests[*index]->generalized_funcs)->extra_state, mystatus);
737 if (requests[*index] != MPI_REQUEST_NULL && (requests[*index]->flags_ & MPI_REQ_NON_PERSISTENT))
738 requests[*index] = MPI_REQUEST_NULL;
739 XBT_DEBUG("Testany - returning with index %d", *index);
747 XBT_DEBUG("Testany on inactive handles, returning flag=1 but empty status");
748 //all requests are null or inactive, return true
750 *index = MPI_UNDEFINED;
751 Status::empty(status);
757 int Request::testall(int count, MPI_Request requests[], int* outflag, MPI_Status status[])
760 MPI_Status *pstat = status == MPI_STATUSES_IGNORE ? MPI_STATUS_IGNORE : &stat;
764 for(int i=0; i<count; i++){
765 if (requests[i] != MPI_REQUEST_NULL && not(requests[i]->flags_ & MPI_REQ_PREPARED)) {
766 int ret = test(&requests[i], pstat, &flag);
769 requests[i]=MPI_REQUEST_NULL;
773 if (ret != MPI_SUCCESS)
776 Status::empty(pstat);
778 if(status != MPI_STATUSES_IGNORE) {
783 return MPI_ERR_IN_STATUS;
788 void Request::probe(int source, int tag, MPI_Comm comm, MPI_Status* status){
790 //FIXME find another way to avoid busy waiting ?
791 // the issue here is that we have to wait on a nonexistent comm
793 iprobe(source, tag, comm, &flag, status);
794 XBT_DEBUG("Busy Waiting on probing : %d", flag);
798 void Request::iprobe(int source, int tag, MPI_Comm comm, int* flag, MPI_Status* status){
799 // to avoid deadlock, we have to sleep some time here, or the timer won't advance and we will only do iprobe simcalls
800 // especially when used as a break condition, such as while (MPI_Iprobe(...)) dostuff...
801 // nsleeps is a multiplier to the sleeptime, to increase speed of execution, each failed iprobe will increase it
802 // This can speed up the execution of certain applications by an order of magnitude, such as HPL
803 static int nsleeps = 1;
804 double speed = s4u::this_actor::get_host()->get_speed();
805 double maxrate = smpi_cfg_iprobe_cpu_usage();
807 new Request(nullptr, 0, MPI_CHAR, source == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : comm->group()->actor_pid(source),
808 simgrid::s4u::this_actor::get_pid(), tag, comm, MPI_REQ_PERSISTENT | MPI_REQ_RECV | MPI_REQ_PROBE);
809 if (smpi_iprobe_sleep > 0) {
810 /** Compute the number of flops we will sleep **/
811 s4u::this_actor::exec_init(/*nsleeps: See comment above */ nsleeps *
812 /*(seconds * flop/s -> total flops)*/ smpi_iprobe_sleep * speed * maxrate)
814 /* Not the entire CPU can be used when iprobing: This is important for
815 * the energy consumption caused by polling with iprobes.
816 * Note also that the number of flops that was
817 * computed above contains a maxrate factor and is hence reduced (maxrate < 1)
819 ->set_bound(maxrate*speed)
823 // behave like a receive, but don't do it
824 s4u::Mailbox* mailbox;
826 request->print_request("New iprobe");
827 // We have to test both mailboxes as we don't know if we will receive one or another
828 if (smpi_cfg_async_small_thresh() > 0) {
829 mailbox = smpi_process()->mailbox_small();
830 XBT_DEBUG("Trying to probe the perm recv mailbox");
831 request->action_ = mailbox->iprobe(0, &match_recv, static_cast<void*>(request));
834 if (request->action_ == nullptr){
835 mailbox = smpi_process()->mailbox();
836 XBT_DEBUG("trying to probe the other mailbox");
837 request->action_ = mailbox->iprobe(0, &match_recv, static_cast<void*>(request));
840 if (request->action_ != nullptr){
841 kernel::activity::CommImplPtr sync_comm = boost::static_pointer_cast<kernel::activity::CommImpl>(request->action_);
842 const Request* req = static_cast<MPI_Request>(sync_comm->src_data_);
844 if (status != MPI_STATUS_IGNORE && (req->flags_ & MPI_REQ_PREPARED) == 0) {
845 status->MPI_SOURCE = comm->group()->rank(req->src_);
846 status->MPI_TAG = req->tag_;
847 status->MPI_ERROR = MPI_SUCCESS;
848 status->count = req->real_size_;
850 nsleeps = 1;//reset the number of sleeps we will do next time
854 if (smpi_cfg_grow_injected_times())
858 xbt_assert(request == MPI_REQUEST_NULL);
861 void Request::finish_wait(MPI_Request* request, MPI_Status * status)
863 MPI_Request req = *request;
864 Status::empty(status);
865 if((req->flags_ & MPI_REQ_CANCELLED) != 0 && (req->flags_ & MPI_REQ_MATCHED) == 0) {
866 if (status!=MPI_STATUS_IGNORE)
868 if(req->detached_sender_ != nullptr)
869 unref(&(req->detached_sender_));
874 if ((req->flags_ & (MPI_REQ_PREPARED | MPI_REQ_GENERALIZED | MPI_REQ_FINISHED)) == 0) {
875 if (status != MPI_STATUS_IGNORE) {
876 if (req->src_== MPI_PROC_NULL || req->dst_== MPI_PROC_NULL){
877 Status::empty(status);
878 status->MPI_SOURCE = MPI_PROC_NULL;
880 int src = req->src_ == MPI_ANY_SOURCE ? req->real_src_ : req->src_;
881 status->MPI_SOURCE = req->comm_->group()->rank(src);
882 status->MPI_TAG = req->tag_ == MPI_ANY_TAG ? req->real_tag_ : req->tag_;
883 status->MPI_ERROR = req->truncated_ ? MPI_ERR_TRUNCATE : MPI_SUCCESS;
885 // this handles the case were size in receive differs from size in send
886 status->count = req->real_size_;
888 //detached send will be finished at the other end
889 if (not(req->detached_ && ((req->flags_ & MPI_REQ_SEND) != 0))) {
890 req->print_request("Finishing");
891 MPI_Datatype datatype = req->old_type_;
893 // FIXME Handle the case of a partial shared malloc.
894 if (((req->flags_ & MPI_REQ_ACCUMULATE) != 0) ||
895 (datatype->flags() & DT_FLAG_DERIVED)) { // && (not smpi_is_shared(req->old_buf_))){
896 if (not smpi_process()->replaying() && smpi_cfg_privatization() != SmpiPrivStrategies::NONE &&
897 static_cast<char*>(req->old_buf_) >= smpi_data_exe_start &&
898 static_cast<char*>(req->old_buf_) < smpi_data_exe_start + smpi_data_exe_size) {
899 XBT_VERB("Privatization : We are unserializing to a zone in global memory Switch data segment ");
900 smpi_switch_data_segment(simgrid::s4u::Actor::self());
903 if(datatype->flags() & DT_FLAG_DERIVED){
904 // This part handles the problem of non-contiguous memory the unserialization at the reception
905 if ((req->flags_ & MPI_REQ_RECV) && datatype->size() != 0)
906 datatype->unserialize(req->buf_, req->old_buf_, req->real_size_/datatype->size() , req->op_);
909 } else if (req->flags_ & MPI_REQ_RECV) { // apply op on contiguous buffer for accumulate
910 if (datatype->size() != 0) {
911 int n = req->real_size_ / datatype->size();
912 req->op_->apply(req->buf_, req->old_buf_, &n, datatype);
921 if (TRACE_smpi_view_internals() && ((req->flags_ & MPI_REQ_RECV) != 0)) {
922 int rank = simgrid::s4u::this_actor::get_pid();
923 int src_traced = (req->src_ == MPI_ANY_SOURCE ? req->real_src_ : req->src_);
924 TRACE_smpi_recv(src_traced, rank,req->tag_);
926 if(req->detached_sender_ != nullptr){
927 //integrate pseudo-timing for buffering of small messages, do not bother to execute the simcall if 0
929 simgrid::s4u::Actor::self()->get_host()->extension<simgrid::smpi::Host>()->orecv(req->real_size());
930 if (sleeptime > 0.0) {
931 simgrid::s4u::this_actor::sleep_for(sleeptime);
932 XBT_DEBUG("receiving size of %zu : sleep %f ", req->real_size_, sleeptime);
934 unref(&(req->detached_sender_));
936 if (req->flags_ & MPI_REQ_PERSISTENT)
937 req->action_ = nullptr;
938 req->flags_ |= MPI_REQ_FINISHED;
940 if (req->truncated_) {
941 char error_string[MPI_MAX_ERROR_STRING];
943 PMPI_Error_string(MPI_ERR_TRUNCATE, error_string, &error_size);
944 MPI_Errhandler err = (req->comm_) ? (req->comm_)->errhandler() : MPI_ERRHANDLER_NULL;
945 if (err == MPI_ERRHANDLER_NULL || err == MPI_ERRORS_RETURN)
946 XBT_WARN("recv - returned %.*s instead of MPI_SUCCESS", error_size, error_string);
947 else if (err == MPI_ERRORS_ARE_FATAL)
948 xbt_die("recv - returned %.*s instead of MPI_SUCCESS", error_size, error_string);
950 err->call((req->comm_), MPI_ERR_TRUNCATE);
951 if (err != MPI_ERRHANDLER_NULL)
952 simgrid::smpi::Errhandler::unref(err);
953 MC_assert(not MC_is_active()); /* Only fail in MC mode */
955 if(req->src_ != MPI_PROC_NULL && req->dst_ != MPI_PROC_NULL)
959 int Request::wait(MPI_Request * request, MPI_Status * status)
961 // assume that *request is not MPI_REQUEST_NULL (filtered in PMPI_Wait before)
962 xbt_assert(*request != MPI_REQUEST_NULL);
966 if((*request)->src_ == MPI_PROC_NULL || (*request)->dst_ == MPI_PROC_NULL){
967 if (status != MPI_STATUS_IGNORE) {
968 Status::empty(status);
969 status->MPI_SOURCE = MPI_PROC_NULL;
971 (*request)=MPI_REQUEST_NULL;
974 // Are we waiting on a request meant for non blocking collectives ?
975 // If so, wait for all the subrequests.
976 if ((*request)->nbc_requests_size_>0){
977 ret = waitall((*request)->nbc_requests_size_, (*request)->nbc_requests_, MPI_STATUSES_IGNORE);
978 for (int i = 0; i < (*request)->nbc_requests_size_; i++) {
979 if((*request)->buf_!=nullptr && (*request)->nbc_requests_[i]!=MPI_REQUEST_NULL){//reduce case
980 void * buf=(*request)->nbc_requests_[i]->buf_;
981 if((*request)->old_type_->flags() & DT_FLAG_DERIVED)
982 buf=(*request)->nbc_requests_[i]->old_buf_;
983 if((*request)->nbc_requests_[i]->flags_ & MPI_REQ_RECV ){
984 if((*request)->op_!=MPI_OP_NULL){
985 int count=(*request)->size_/ (*request)->old_type_->size();
986 (*request)->op_->apply(buf, (*request)->buf_, &count, (*request)->old_type_);
988 smpi_free_tmp_buffer(static_cast<unsigned char*>(buf));
991 if((*request)->nbc_requests_[i]!=MPI_REQUEST_NULL)
992 Request::unref(&((*request)->nbc_requests_[i]));
994 delete[] (*request)->nbc_requests_;
995 (*request)->nbc_requests_size_=0;
997 (*request)=MPI_REQUEST_NULL;
1001 (*request)->print_request("Waiting");
1002 if ((*request)->flags_ & (MPI_REQ_PREPARED | MPI_REQ_FINISHED)) {
1003 Status::empty(status);
1007 if ((*request)->action_ != nullptr){
1009 // this is not a detached send
1010 simcall_comm_wait((*request)->action_.get(), -1.0);
1011 } catch (const Exception&) {
1012 XBT_VERB("Request cancelled");
1016 if ((*request)->flags_ & MPI_REQ_GENERALIZED) {
1017 if(!((*request)->flags_ & MPI_REQ_COMPLETE)){
1018 ((*request)->generalized_funcs)->mutex->lock();
1019 ((*request)->generalized_funcs)->cond->wait(((*request)->generalized_funcs)->mutex);
1020 ((*request)->generalized_funcs)->mutex->unlock();
1022 MPI_Status tmp_status;
1023 MPI_Status* mystatus;
1024 if (status == MPI_STATUS_IGNORE) {
1025 mystatus = &tmp_status;
1026 Status::empty(mystatus);
1030 ret = ((*request)->generalized_funcs)->query_fn(((*request)->generalized_funcs)->extra_state, mystatus);
1033 if ((*request)->truncated_)
1034 ret = MPI_ERR_TRUNCATE;
1036 finish_wait(request, status); // may invalidate *request
1037 if (*request != MPI_REQUEST_NULL && (((*request)->flags_ & MPI_REQ_NON_PERSISTENT) != 0))
1038 *request = MPI_REQUEST_NULL;
1042 int Request::waitany(int count, MPI_Request requests[], MPI_Status * status)
1044 int index = MPI_UNDEFINED;
1047 // Wait for a request to complete
1048 std::vector<simgrid::kernel::activity::CommImpl*> comms;
1049 std::vector<int> map;
1050 XBT_DEBUG("Wait for one of %d", count);
1051 for(int i = 0; i < count; i++) {
1052 if (requests[i] != MPI_REQUEST_NULL && not(requests[i]->flags_ & MPI_REQ_PREPARED) &&
1053 not(requests[i]->flags_ & MPI_REQ_FINISHED)) {
1054 if (requests[i]->action_ != nullptr) {
1055 XBT_DEBUG("Waiting any %p ", requests[i]);
1056 comms.push_back(static_cast<simgrid::kernel::activity::CommImpl*>(requests[i]->action_.get()));
1059 // This is a finished detached request, let's return this one
1060 comms.clear(); // don't do the waitany call afterwards
1062 finish_wait(&requests[i], status); // cleanup if refcount = 0
1063 if (requests[i] != MPI_REQUEST_NULL && (requests[i]->flags_ & MPI_REQ_NON_PERSISTENT))
1064 requests[i] = MPI_REQUEST_NULL; // set to null
1069 if (not comms.empty()) {
1070 XBT_DEBUG("Enter waitany for %zu comms", comms.size());
1073 i = simcall_comm_waitany(comms.data(), comms.size(), -1);
1074 } catch (const Exception&) {
1075 XBT_INFO("request cancelled");
1079 // not MPI_UNDEFINED, as this is a simix return code
1082 //in case of an accumulate, we have to wait the end of all requests to apply the operation, ordered correctly.
1083 if ((requests[index] == MPI_REQUEST_NULL) ||
1084 (not((requests[index]->flags_ & MPI_REQ_ACCUMULATE) && (requests[index]->flags_ & MPI_REQ_RECV)))) {
1085 finish_wait(&requests[index],status);
1086 if (requests[index] != MPI_REQUEST_NULL && (requests[index]->flags_ & MPI_REQ_NON_PERSISTENT))
1087 requests[index] = MPI_REQUEST_NULL;
1093 if (index==MPI_UNDEFINED)
1094 Status::empty(status);
1099 static int sort_accumulates(const Request* a, const Request* b)
1101 return (a->tag() > b->tag());
1104 int Request::waitall(int count, MPI_Request requests[], MPI_Status status[])
1106 std::vector<MPI_Request> accumulates;
1109 MPI_Status *pstat = (status == MPI_STATUSES_IGNORE ? MPI_STATUS_IGNORE : &stat);
1110 int retvalue = MPI_SUCCESS;
1111 //tag invalid requests in the set
1112 if (status != MPI_STATUSES_IGNORE) {
1113 for (int c = 0; c < count; c++) {
1114 if (requests[c] == MPI_REQUEST_NULL || requests[c]->dst_ == MPI_PROC_NULL ||
1115 (requests[c]->flags_ & MPI_REQ_PREPARED)) {
1116 Status::empty(&status[c]);
1117 } else if (requests[c]->src_ == MPI_PROC_NULL) {
1118 Status::empty(&status[c]);
1119 status[c].MPI_SOURCE = MPI_PROC_NULL;
1123 for (int c = 0; c < count; c++) {
1124 if (MC_is_active() || MC_record_replay_is_active()) {
1125 wait(&requests[c],pstat);
1128 index = waitany(count, requests, pstat);
1130 if (index == MPI_UNDEFINED)
1133 if (requests[index] != MPI_REQUEST_NULL && (requests[index]->flags_ & MPI_REQ_RECV) &&
1134 (requests[index]->flags_ & MPI_REQ_ACCUMULATE))
1135 accumulates.push_back(requests[index]);
1136 if (requests[index] != MPI_REQUEST_NULL && (requests[index]->flags_ & MPI_REQ_NON_PERSISTENT))
1137 requests[index] = MPI_REQUEST_NULL;
1139 if (status != MPI_STATUSES_IGNORE) {
1140 status[index] = *pstat;
1141 if (status[index].MPI_ERROR == MPI_ERR_TRUNCATE)
1142 retvalue = MPI_ERR_IN_STATUS;
1146 if (not accumulates.empty()) {
1147 std::sort(accumulates.begin(), accumulates.end(), sort_accumulates);
1148 for (auto& req : accumulates) {
1149 finish_wait(&req, status);
1156 int Request::waitsome(int incount, MPI_Request requests[], int *indices, MPI_Status status[])
1162 MPI_Status *pstat = status == MPI_STATUSES_IGNORE ? MPI_STATUS_IGNORE : &stat;
1163 index = waitany(incount, requests, pstat);
1164 if(index==MPI_UNDEFINED) return MPI_UNDEFINED;
1165 if(status != MPI_STATUSES_IGNORE) {
1166 status[count] = *pstat;
1168 indices[count] = index;
1170 for (int i = 0; i < incount; i++) {
1171 if (i!=index && requests[i] != MPI_REQUEST_NULL
1172 && not(requests[i]->flags_ & MPI_REQ_FINISHED)) {
1173 test(&requests[i], pstat,&flag);
1176 if(status != MPI_STATUSES_IGNORE) {
1177 status[count] = *pstat;
1179 if (requests[i] != MPI_REQUEST_NULL && (requests[i]->flags_ & MPI_REQ_NON_PERSISTENT))
1180 requests[i]=MPI_REQUEST_NULL;
1188 MPI_Request Request::f2c(int id)
1190 if(id==MPI_FORTRAN_REQUEST_NULL)
1191 return MPI_REQUEST_NULL;
1192 return static_cast<MPI_Request>(F2C::lookup()->at(id));
1195 void Request::free_f(int id)
1197 if (id != MPI_FORTRAN_REQUEST_NULL) {
1198 F2C::lookup()->erase(id);
1202 int Request::get_status(const Request* req, int* flag, MPI_Status* status)
1206 if(req != MPI_REQUEST_NULL && req->action_ != nullptr) {
1207 req->iprobe(req->comm_->group()->rank(req->src_), req->tag_, req->comm_, flag, status);
1211 if (req != MPI_REQUEST_NULL &&
1212 (req->flags_ & MPI_REQ_GENERALIZED)
1213 && !(req->flags_ & MPI_REQ_COMPLETE)) {
1219 if(req != MPI_REQUEST_NULL &&
1220 status != MPI_STATUS_IGNORE) {
1221 int src = req->src_ == MPI_ANY_SOURCE ? req->real_src_ : req->src_;
1222 status->MPI_SOURCE = req->comm_->group()->rank(src);
1223 status->MPI_TAG = req->tag_ == MPI_ANY_TAG ? req->real_tag_ : req->tag_;
1224 status->MPI_ERROR = req->truncated_ ? MPI_ERR_TRUNCATE : MPI_SUCCESS;
1225 status->count = req->real_size_;
1230 int Request::grequest_start(MPI_Grequest_query_function* query_fn, MPI_Grequest_free_function* free_fn,
1231 MPI_Grequest_cancel_function* cancel_fn, void* extra_state, MPI_Request* request)
1233 *request = new Request();
1234 (*request)->flags_ |= MPI_REQ_GENERALIZED;
1235 (*request)->flags_ |= MPI_REQ_PERSISTENT;
1236 (*request)->refcount_ = 1;
1237 ((*request)->generalized_funcs) = std::make_unique<smpi_mpi_generalized_request_funcs_t>();
1238 ((*request)->generalized_funcs)->query_fn=query_fn;
1239 ((*request)->generalized_funcs)->free_fn=free_fn;
1240 ((*request)->generalized_funcs)->cancel_fn=cancel_fn;
1241 ((*request)->generalized_funcs)->extra_state=extra_state;
1242 ((*request)->generalized_funcs)->cond = simgrid::s4u::ConditionVariable::create();
1243 ((*request)->generalized_funcs)->mutex = simgrid::s4u::Mutex::create();
1247 int Request::grequest_complete(MPI_Request request)
1249 if ((!(request->flags_ & MPI_REQ_GENERALIZED)) || request->generalized_funcs->mutex == nullptr)
1250 return MPI_ERR_REQUEST;
1251 request->generalized_funcs->mutex->lock();
1252 request->flags_ |= MPI_REQ_COMPLETE; // in case wait would be called after complete
1253 request->generalized_funcs->cond->notify_one();
1254 request->generalized_funcs->mutex->unlock();
1258 void Request::set_nbc_requests(MPI_Request* reqs, int size){
1259 nbc_requests_size_ = size;
1261 nbc_requests_ = reqs;
1264 nbc_requests_ = nullptr;
1268 int Request::get_nbc_requests_size() const
1270 return nbc_requests_size_;
1273 MPI_Request* Request::get_nbc_requests() const
1275 return nbc_requests_;