1 /* Copyright (c) 2007-2022. 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 <simgrid/modelchecker.h>
7 #include "smpi_win.hpp"
10 #include "smpi_coll.hpp"
11 #include "smpi_comm.hpp"
12 #include "smpi_datatype.hpp"
13 #include "smpi_info.hpp"
14 #include "smpi_keyvals.hpp"
15 #include "smpi_request.hpp"
16 #include "src/smpi/include/smpi_actor.hpp"
17 #include "src/mc/mc_replay.hpp"
21 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_rma, smpi, "Logging specific to SMPI (RMA operations)");
23 #define CHECK_RMA_REMOTE_WIN(fun, win)\
24 if(target_count*target_datatype->get_extent()>win->size_){\
25 XBT_WARN("%s: Trying to move %zd, which exceeds the window size on target process %d : %zd - Bailing out.",\
26 fun, target_count*target_datatype->get_extent(), target_rank, win->size_);\
27 simgrid::smpi::utils::set_current_buffer(1,"win_base",win->base_);\
28 return MPI_ERR_RMA_RANGE;\
31 #define CHECK_WIN_LOCKED(win) \
32 if (opened_ == 0) { /*check that post/start has been done*/ \
33 bool locked = std::any_of(begin(win->lockers_), end(win->lockers_), [this](int it) { return it == this->rank_; }); \
40 std::unordered_map<int, smpi_key_elem> Win::keyvals_;
41 int Win::keyval_id_=0;
43 Win::Win(void* base, MPI_Aint size, int disp_unit, MPI_Info info, MPI_Comm comm, bool allocated, bool dynamic)
46 , disp_unit_(disp_unit)
49 , connected_wins_(comm->size())
51 , allocated_(allocated)
54 XBT_DEBUG("Creating window");
55 if(info!=MPI_INFO_NULL)
57 connected_wins_[rank_] = this;
59 comm->add_rma_win(this);
62 colls::allgather(&connected_wins_[rank_], sizeof(MPI_Win), MPI_BYTE, connected_wins_.data(), sizeof(MPI_Win),
64 if (MC_is_active() || MC_record_replay_is_active()){
66 bar_ = new s4u::Barrier(comm->size());
68 colls::bcast(&bar_, sizeof(s4u::Barrier*), MPI_BYTE, 0, comm);
77 //As per the standard, perform a barrier to ensure every async comm is finished
78 if (MC_is_active() || MC_record_replay_is_active())
81 colls::barrier(comm_);
84 if (info_ != MPI_INFO_NULL)
85 simgrid::smpi::Info::unref(info_);
86 if (errhandler_ != MPI_ERRHANDLER_NULL)
87 simgrid::smpi::Errhandler::unref(errhandler_);
89 comm_->remove_rma_win(this);
91 colls::barrier(comm_);
100 F2C::free_f(this->f2c_id());
104 int Win::attach(void* /*base*/, MPI_Aint size)
106 if (not(base_ == MPI_BOTTOM || base_ == nullptr))
108 base_ = nullptr; // actually the address will be given in the RMA calls, as being the disp.
113 int Win::detach(const void* /*base*/)
120 void Win::get_name(char* name, int* length) const
122 *length = static_cast<int>(name_.length());
123 if (not name_.empty()) {
124 name_.copy(name, *length);
125 name[*length] = '\0';
129 void Win::get_group(MPI_Group* group){
130 if(comm_ != MPI_COMM_NULL){
131 *group = comm_->group();
133 *group = MPI_GROUP_NULL;
142 int Win::rank() const
147 MPI_Comm Win::comm() const
152 MPI_Aint Win::size() const
157 void* Win::base() const
162 int Win::disp_unit() const
167 bool Win::dynamic() const
172 void Win::set_info(MPI_Info info)
174 if (info_ != MPI_INFO_NULL)
175 simgrid::smpi::Info::unref(info_);
177 if (info_ != MPI_INFO_NULL)
181 void Win::set_name(const char* name){
185 int Win::fence(int assert)
187 XBT_DEBUG("Entering fence");
189 if (not (assert & MPI_MODE_NOPRECEDE)) {
190 // This is not the first fence => finalize what came before
191 if (MC_is_active() || MC_record_replay_is_active())
194 colls::barrier(comm_);
199 if (assert & MPI_MODE_NOSUCCEED) // there should be no ops after this one, tell we are closed.
202 if (MC_is_active() || MC_record_replay_is_active())
205 colls::barrier(comm_);
206 XBT_DEBUG("Leaving fence");
211 int Win::put(const void *origin_addr, int origin_count, MPI_Datatype origin_datatype, int target_rank,
212 MPI_Aint target_disp, int target_count, MPI_Datatype target_datatype, MPI_Request* request)
214 //get receiver pointer
215 Win* recv_win = connected_wins_[target_rank];
217 CHECK_WIN_LOCKED(recv_win)
218 CHECK_RMA_REMOTE_WIN("MPI_Put", recv_win)
220 void* recv_addr = static_cast<char*>(recv_win->base_) + target_disp * recv_win->disp_unit_;
222 if (target_rank != rank_) { // This is not for myself, so we need to send messages
223 XBT_DEBUG("Entering MPI_Put to remote rank %d", target_rank);
224 // prepare send_request
226 Request::rma_send_init(origin_addr, origin_count, origin_datatype, rank_, target_rank, SMPI_RMA_TAG + 1, comm_,
229 //prepare receiver request
230 MPI_Request rreq = Request::rma_recv_init(recv_addr, target_count, target_datatype, rank_, target_rank,
231 SMPI_RMA_TAG + 1, recv_win->comm_, MPI_OP_NULL);
236 if(request!=nullptr){
240 requests_.push_back(sreq);
244 //push request to receiver's win
245 recv_win->mut_->lock();
246 recv_win->requests_.push_back(rreq);
248 recv_win->mut_->unlock();
250 XBT_DEBUG("Entering MPI_Put from myself to myself, rank %d", target_rank);
251 Datatype::copy(origin_addr, origin_count, origin_datatype, recv_addr, target_count, target_datatype);
253 *request = MPI_REQUEST_NULL;
259 int Win::get( void *origin_addr, int origin_count, MPI_Datatype origin_datatype, int target_rank,
260 MPI_Aint target_disp, int target_count, MPI_Datatype target_datatype, MPI_Request* request)
263 Win* send_win = connected_wins_[target_rank];
265 CHECK_WIN_LOCKED(send_win)
266 CHECK_RMA_REMOTE_WIN("MPI_Get", send_win)
268 const void* send_addr = static_cast<void*>(static_cast<char*>(send_win->base_) + target_disp * send_win->disp_unit_);
269 XBT_DEBUG("Entering MPI_Get from %d", target_rank);
271 if (target_rank != rank_) {
272 //prepare send_request
273 MPI_Request sreq = Request::rma_send_init(send_addr, target_count, target_datatype, target_rank, rank_,
274 SMPI_RMA_TAG + 2, send_win->comm_, MPI_OP_NULL);
276 //prepare receiver request
277 MPI_Request rreq = Request::rma_recv_init(origin_addr, origin_count, origin_datatype, target_rank, rank_,
278 SMPI_RMA_TAG + 2, comm_, MPI_OP_NULL);
280 //start the send, with another process than us as sender.
282 // push request to sender's win
283 send_win->mut_->lock();
284 send_win->requests_.push_back(sreq);
285 send_win->mut_->unlock();
290 if(request!=nullptr){
294 requests_.push_back(rreq);
298 Datatype::copy(send_addr, target_count, target_datatype, origin_addr, origin_count, origin_datatype);
300 *request=MPI_REQUEST_NULL;
305 int Win::accumulate(const void *origin_addr, int origin_count, MPI_Datatype origin_datatype, int target_rank,
306 MPI_Aint target_disp, int target_count, MPI_Datatype target_datatype, MPI_Op op, MPI_Request* request)
308 XBT_DEBUG("Entering MPI_Win_Accumulate");
309 //get receiver pointer
310 Win* recv_win = connected_wins_[target_rank];
312 //FIXME: local version
313 CHECK_WIN_LOCKED(recv_win)
314 CHECK_RMA_REMOTE_WIN("MPI_Accumulate", recv_win)
316 void* recv_addr = static_cast<char*>(recv_win->base_) + target_disp * recv_win->disp_unit_;
317 XBT_DEBUG("Entering MPI_Accumulate to %d", target_rank);
318 // As the tag will be used for ordering of the operations, subtract count from it (to avoid collisions with other
319 // SMPI tags, SMPI_RMA_TAG is set below all the other ones we use)
320 // prepare send_request
322 MPI_Request sreq = Request::rma_send_init(origin_addr, origin_count, origin_datatype, rank_, target_rank,
323 SMPI_RMA_TAG - 3 - count_, comm_, op);
325 // prepare receiver request
326 MPI_Request rreq = Request::rma_recv_init(recv_addr, target_count, target_datatype, rank_, target_rank,
327 SMPI_RMA_TAG - 3 - count_, recv_win->comm_, op);
333 // push request to receiver's win
334 recv_win->mut_->lock();
335 recv_win->requests_.push_back(rreq);
337 recv_win->mut_->unlock();
339 if (request != nullptr) {
343 requests_.push_back(sreq);
347 // FIXME: The current implementation fails to ensure the correct ordering of the accumulate requests. The following
348 // 'flush' is a workaround to fix that.
350 XBT_DEBUG("Leaving MPI_Win_Accumulate");
354 int Win::get_accumulate(const void* origin_addr, int origin_count, MPI_Datatype origin_datatype, void* result_addr,
355 int result_count, MPI_Datatype result_datatype, int target_rank, MPI_Aint target_disp,
356 int target_count, MPI_Datatype target_datatype, MPI_Op op, MPI_Request*)
359 const Win* send_win = connected_wins_[target_rank];
361 CHECK_WIN_LOCKED(send_win)
362 CHECK_RMA_REMOTE_WIN("MPI_Get_Accumulate", send_win)
364 XBT_DEBUG("Entering MPI_Get_accumulate from %d", target_rank);
365 //need to be sure ops are correctly ordered, so finish request here ? slow.
366 MPI_Request req = MPI_REQUEST_NULL;
367 send_win->atomic_mut_->lock();
368 get(result_addr, result_count, result_datatype, target_rank,
369 target_disp, target_count, target_datatype, &req);
370 if (req != MPI_REQUEST_NULL)
371 Request::wait(&req, MPI_STATUS_IGNORE);
373 accumulate(origin_addr, origin_count, origin_datatype, target_rank,
374 target_disp, target_count, target_datatype, op, &req);
375 if (req != MPI_REQUEST_NULL)
376 Request::wait(&req, MPI_STATUS_IGNORE);
377 send_win->atomic_mut_->unlock();
381 int Win::compare_and_swap(const void* origin_addr, const void* compare_addr, void* result_addr, MPI_Datatype datatype,
382 int target_rank, MPI_Aint target_disp)
385 const Win* send_win = connected_wins_[target_rank];
387 CHECK_WIN_LOCKED(send_win)
389 XBT_DEBUG("Entering MPI_Compare_and_swap with %d", target_rank);
390 MPI_Request req = MPI_REQUEST_NULL;
391 send_win->atomic_mut_->lock();
392 get(result_addr, 1, datatype, target_rank,
393 target_disp, 1, datatype, &req);
394 if (req != MPI_REQUEST_NULL)
395 Request::wait(&req, MPI_STATUS_IGNORE);
396 if (not memcmp(result_addr, compare_addr, datatype->get_extent())) {
397 put(origin_addr, 1, datatype, target_rank,
398 target_disp, 1, datatype);
400 send_win->atomic_mut_->unlock();
404 int Win::start(MPI_Group group, int /*assert*/)
406 /* From MPI forum advices
407 The call to MPI_WIN_COMPLETE does not return until the put call has completed at the origin; and the target window
408 will be accessed by the put operation only after the call to MPI_WIN_START has matched a call to MPI_WIN_POST by
409 the target process. This still leaves much choice to implementors. The call to MPI_WIN_START can block until the
410 matching call to MPI_WIN_POST occurs at all target processes. One can also have implementations where the call to
411 MPI_WIN_START is nonblocking, but the call to MPI_PUT blocks until the matching call to MPI_WIN_POST occurred; or
412 implementations where the first two calls are nonblocking, but the call to MPI_WIN_COMPLETE blocks until the call
413 to MPI_WIN_POST occurred; or even implementations where all three calls can complete before any target process
414 called MPI_WIN_POST --- the data put must be buffered, in this last case, so as to allow the put to complete at the
415 origin ahead of its completion at the target. However, once the call to MPI_WIN_POST is issued, the sequence above
416 must complete, without further dependencies. */
418 //naive, blocking implementation.
419 XBT_DEBUG("Entering MPI_Win_Start");
420 std::vector<MPI_Request> reqs;
421 for (int i = 0; i < group->size(); i++) {
422 int src = comm_->group()->rank(group->actor(i));
423 xbt_assert(src != MPI_UNDEFINED);
425 reqs.emplace_back(Request::irecv_init(nullptr, 0, MPI_CHAR, src, SMPI_RMA_TAG + 4, comm_));
427 int size = static_cast<int>(reqs.size());
429 Request::startall(size, reqs.data());
430 Request::waitall(size, reqs.data(), MPI_STATUSES_IGNORE);
431 for (auto& req : reqs)
432 Request::unref(&req);
436 opened_++; // we're open for business !
437 XBT_DEBUG("Leaving MPI_Win_Start");
441 int Win::post(MPI_Group group, int /*assert*/)
443 //let's make a synchronous send here
444 XBT_DEBUG("Entering MPI_Win_Post");
445 std::vector<MPI_Request> reqs;
446 for (int i = 0; i < group->size(); i++) {
447 int dst = comm_->group()->rank(group->actor(i));
448 xbt_assert(dst != MPI_UNDEFINED);
450 reqs.emplace_back(Request::send_init(nullptr, 0, MPI_CHAR, dst, SMPI_RMA_TAG + 4, comm_));
452 int size = static_cast<int>(reqs.size());
454 Request::startall(size, reqs.data());
455 Request::waitall(size, reqs.data(), MPI_STATUSES_IGNORE);
456 for (auto& req : reqs)
457 Request::unref(&req);
461 opened_++; // we're open for business !
462 XBT_DEBUG("Leaving MPI_Win_Post");
467 xbt_assert(opened_ != 0, "Complete called on already opened MPI_Win");
469 XBT_DEBUG("Entering MPI_Win_Complete");
470 std::vector<MPI_Request> reqs;
471 for (int i = 0; i < dst_group_->size(); i++) {
472 int dst = comm_->group()->rank(dst_group_->actor(i));
473 xbt_assert(dst != MPI_UNDEFINED);
475 reqs.emplace_back(Request::send_init(nullptr, 0, MPI_CHAR, dst, SMPI_RMA_TAG + 5, comm_));
477 int size = static_cast<int>(reqs.size());
479 XBT_DEBUG("Win_complete - Sending sync messages to %d processes", size);
480 Request::startall(size, reqs.data());
481 Request::waitall(size, reqs.data(), MPI_STATUSES_IGNORE);
482 for (auto& req : reqs)
483 Request::unref(&req);
487 opened_--; //we're closed for business !
488 Group::unref(dst_group_);
489 dst_group_ = MPI_GROUP_NULL;
494 //naive, blocking implementation.
495 XBT_DEBUG("Entering MPI_Win_Wait");
496 std::vector<MPI_Request> reqs;
497 for (int i = 0; i < src_group_->size(); i++) {
498 int src = comm_->group()->rank(src_group_->actor(i));
499 xbt_assert(src != MPI_UNDEFINED);
501 reqs.emplace_back(Request::irecv_init(nullptr, 0, MPI_CHAR, src, SMPI_RMA_TAG + 5, comm_));
503 int size = static_cast<int>(reqs.size());
505 XBT_DEBUG("Win_wait - Receiving sync messages from %d processes", size);
506 Request::startall(size, reqs.data());
507 Request::waitall(size, reqs.data(), MPI_STATUSES_IGNORE);
508 for (auto& req : reqs)
509 Request::unref(&req);
513 opened_--; //we're closed for business !
514 Group::unref(src_group_);
515 src_group_ = MPI_GROUP_NULL;
519 int Win::lock(int lock_type, int rank, int /*assert*/)
521 MPI_Win target_win = connected_wins_[rank];
523 if ((lock_type == MPI_LOCK_EXCLUSIVE && target_win->mode_ != MPI_LOCK_SHARED)|| target_win->mode_ == MPI_LOCK_EXCLUSIVE){
524 target_win->lock_mut_->lock();
525 target_win->mode_+= lock_type;//add the lock_type to differentiate case when we are switching from EXCLUSIVE to SHARED (no release needed in the unlock)
526 if(lock_type == MPI_LOCK_SHARED){//the window used to be exclusive, it's now shared.
527 target_win->lock_mut_->unlock();
529 } else if (not(target_win->mode_ == MPI_LOCK_SHARED && lock_type == MPI_LOCK_EXCLUSIVE))
530 target_win->mode_ += lock_type; // don't set to exclusive if it's already shared
532 target_win->lockers_.push_back(rank_);
538 int Win::lock_all(int assert){
539 int retval = MPI_SUCCESS;
540 for (int i = 0; i < comm_->size(); i++) {
541 int ret = this->lock(MPI_LOCK_SHARED, i, assert);
542 if (ret != MPI_SUCCESS)
548 int Win::unlock(int rank){
549 MPI_Win target_win = connected_wins_[rank];
550 int target_mode = target_win->mode_;
551 target_win->mode_= 0;
552 target_win->lockers_.remove(rank_);
553 if (target_mode==MPI_LOCK_EXCLUSIVE){
554 target_win->lock_mut_->unlock();
561 int Win::unlock_all(){
562 int retval = MPI_SUCCESS;
563 for (int i = 0; i < comm_->size(); i++) {
564 int ret = this->unlock(i);
565 if (ret != MPI_SUCCESS)
571 int Win::flush(int rank){
572 int finished = finish_comms(rank);
573 XBT_DEBUG("Win_flush on local %d for remote %d - Finished %d RMA calls", rank_, rank, finished);
575 finished = connected_wins_[rank]->finish_comms(rank_);
576 XBT_DEBUG("Win_flush on remote %d for local %d - Finished %d RMA calls", rank, rank_, finished);
581 int Win::flush_local(int rank){
582 int finished = finish_comms(rank);
583 XBT_DEBUG("Win_flush_local on local %d for remote %d - Finished %d RMA calls", rank_, rank, finished);
587 int Win::flush_all(){
588 int finished = finish_comms();
589 XBT_DEBUG("Win_flush_all on local %d - Finished %d RMA calls", rank_, finished);
590 for (int i = 0; i < comm_->size(); i++) {
592 finished = connected_wins_[i]->finish_comms(rank_);
593 XBT_DEBUG("Win_flush_all on remote %d for local %d - Finished %d RMA calls", i, rank_, finished);
599 int Win::flush_local_all(){
600 int finished = finish_comms();
601 XBT_DEBUG("Win_flush_local_all on local %d - Finished %d RMA calls", rank_, finished);
605 Win* Win::f2c(int id){
606 return static_cast<Win*>(F2C::f2c(id));
609 int Win::finish_comms(){
610 // This (simulated) mutex ensures that no process pushes to the vector of requests during the waitall.
611 // Without this, the vector could get redimensioned when another process pushes.
612 // This would result in the array used by Request::waitall() to be invalidated.
613 // Another solution would be to copy the data and cleanup the vector *before* Request::waitall
615 //Finish own requests
616 int size = static_cast<int>(requests_.size());
618 MPI_Request* treqs = requests_.data();
619 Request::waitall(size, treqs, MPI_STATUSES_IGNORE);
626 int Win::finish_comms(int rank){
627 // See comment about the mutex in finish_comms() above
629 // Finish own requests
630 // Let's see if we're either the destination or the sender of this request
631 // because we only wait for requests that we are responsible for.
632 // Also use the process id here since the request itself returns from src()
633 // and dst() the process id, NOT the rank (which only exists in the context of a communicator).
634 aid_t proc_id = comm_->group()->actor(rank);
635 auto it = std::stable_partition(begin(requests_), end(requests_), [proc_id](const MPI_Request& req) {
636 return (req == MPI_REQUEST_NULL || (req->src() != proc_id && req->dst() != proc_id));
638 std::vector<MPI_Request> myreqqs(it, end(requests_));
639 requests_.erase(it, end(requests_));
640 int size = static_cast<int>(myreqqs.size());
642 MPI_Request* treqs = myreqqs.data();
643 Request::waitall(size, treqs, MPI_STATUSES_IGNORE);
650 int Win::shared_query(int rank, MPI_Aint* size, int* disp_unit, void* baseptr) const
652 const Win* target_win = rank != MPI_PROC_NULL ? connected_wins_[rank] : nullptr;
653 for (int i = 0; not target_win && i < comm_->size(); i++) {
654 if (connected_wins_[i]->size_ > 0)
655 target_win = connected_wins_[i];
658 *size = target_win->size_;
659 *disp_unit = target_win->disp_unit_;
660 *static_cast<void**>(baseptr) = target_win->base_;
663 *static_cast<void**>(baseptr) = nullptr;
668 MPI_Errhandler Win::errhandler()
670 if (errhandler_ != MPI_ERRHANDLER_NULL)
675 void Win::set_errhandler(MPI_Errhandler errhandler)
677 if (errhandler_ != MPI_ERRHANDLER_NULL)
678 simgrid::smpi::Errhandler::unref(errhandler_);
679 errhandler_ = errhandler;
680 if (errhandler_ != MPI_ERRHANDLER_NULL)
684 } // namespace simgrid