X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/4ebc53f4ea63c0ec84680e5484ad9c9a81c85ff3..49a75a876cd9b9284722a63f1b3a02d1283630e1:/src/simix/smx_network.cpp diff --git a/src/simix/smx_network.cpp b/src/simix/smx_network.cpp index b2327ed30d..54219f1999 100644 --- a/src/simix/smx_network.cpp +++ b/src/simix/smx_network.cpp @@ -1,77 +1,21 @@ -/* Copyright (c) 2009-2018. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2009-2019. The SimGrid Team. All rights reserved. */ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ -#include - -#include - -#include "src/kernel/activity/CommImpl.hpp" -#include - -#include "simgrid/s4u/Host.hpp" - #include "mc/mc.h" -#include "simgrid/s4u/Activity.hpp" -#include "simgrid/s4u/Mailbox.hpp" +#include "simgrid/Exception.hpp" +#include "src/kernel/activity/MailboxImpl.hpp" #include "src/mc/mc_replay.hpp" #include "src/simix/smx_private.hpp" #include "src/surf/cpu_interface.hpp" -#include "src/surf/surf_interface.hpp" - #include "src/surf/network_interface.hpp" +#include + XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix, "SIMIX network-related synchronization"); static void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall); -static void SIMIX_comm_copy_data(smx_activity_t comm); -static void SIMIX_comm_start(simgrid::kernel::activity::CommImplPtr synchro); - -/** - * \brief Checks if there is a communication activity queued in a deque matching our needs - * \param deque where to search into - * \param type The type of communication we are looking for (comm_send, comm_recv) - * \param match_fun the function to apply - * \param this_user_data additional parameter to the match_fun - * \param my_synchro what to compare against - * \param remove_matching whether or not to clean the found object from the queue - * \return The communication activity if found, nullptr otherwise - */ -static simgrid::kernel::activity::CommImplPtr -_find_matching_comm(boost::circular_buffer_space_optimized* deque, e_smx_comm_type_t type, - int (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*), void* this_user_data, - simgrid::kernel::activity::CommImplPtr my_synchro, bool remove_matching) -{ - void* other_user_data = nullptr; - - for(auto it = deque->begin(); it != deque->end(); it++){ - simgrid::kernel::activity::CommImplPtr comm = - boost::dynamic_pointer_cast(std::move(*it)); - - if (comm->type == SIMIX_COMM_SEND) { - other_user_data = comm->src_data; - } else if (comm->type == SIMIX_COMM_RECEIVE) { - other_user_data = comm->dst_data; - } - if (comm->type == type && (match_fun == nullptr || match_fun(this_user_data, other_user_data, comm.get())) && - (not comm->match_fun || comm->match_fun(other_user_data, this_user_data, my_synchro.get()))) { - XBT_DEBUG("Found a matching communication synchro %p", comm.get()); - if (remove_matching) - deque->erase(it); -#if SIMGRID_HAVE_MC - comm->mbox_cpy = comm->mbox; -#endif - comm->mbox = nullptr; - return comm; - } - XBT_DEBUG("Sorry, communication synchro %p does not match our needs:" - " its type is %d but we are looking for a comm of type %d (or maybe the filtering didn't match)", - comm.get(), (int)comm->type, (int)type); - } - XBT_DEBUG("No matching communication synchro found"); - return nullptr; -} /******************************************************************************/ /* Communication synchros */ @@ -88,6 +32,7 @@ XBT_PRIVATE void simcall_HANDLER_comm_send(smx_simcall_t simcall, smx_actor_t sr SIMCALL_SET_MC_VALUE(simcall, 0); simcall_HANDLER_comm_wait(simcall, comm, timeout); } + XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_isend( smx_simcall_t /*simcall*/, smx_actor_t src_proc, smx_mailbox_t mbox, double task_size, double rate, void* src_buff, size_t src_buff_size, int (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*), @@ -98,24 +43,25 @@ XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_isend( XBT_DEBUG("send from mailbox %p", mbox); /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */ - simgrid::kernel::activity::CommImplPtr this_comm = - simgrid::kernel::activity::CommImplPtr(new simgrid::kernel::activity::CommImpl(SIMIX_COMM_SEND)); + simgrid::kernel::activity::CommImplPtr this_comm = simgrid::kernel::activity::CommImplPtr( + new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::SEND)); /* Look for communication synchro matching our needs. We also provide a description of * ourself so that the other side also gets a chance of choosing if it wants to match with us. * * If it is not found then push our communication into the rendez-vous point */ simgrid::kernel::activity::CommImplPtr other_comm = - _find_matching_comm(&mbox->comm_queue, SIMIX_COMM_RECEIVE, match_fun, data, this_comm, /*remove_matching*/ true); + mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::RECEIVE, match_fun, data, this_comm, + /*done*/ false, /*remove_matching*/ true); if (not other_comm) { other_comm = std::move(this_comm); - if (mbox->permanent_receiver != nullptr) { + if (mbox->permanent_receiver_ != nullptr) { //this mailbox is for small messages, which have to be sent right now other_comm->state_ = SIMIX_READY; - other_comm->dst_proc=mbox->permanent_receiver.get(); - mbox->done_comm_queue.push_back(other_comm); + other_comm->dst_actor_ = mbox->permanent_receiver_.get(); + mbox->done_comm_queue_.push_back(other_comm); XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get()); }else{ @@ -125,7 +71,7 @@ XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_isend( XBT_DEBUG("Receive already pushed"); other_comm->state_ = SIMIX_READY; - other_comm->type = SIMIX_COMM_READY; + other_comm->type = simgrid::kernel::activity::CommImpl::Type::READY; } src_proc->comms.push_back(other_comm); @@ -137,12 +83,12 @@ XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_isend( } /* Setup the communication synchro */ - other_comm->src_proc = src_proc; - other_comm->task_size = task_size; - other_comm->rate = rate; - other_comm->src_buff = src_buff; - other_comm->src_buff_size = src_buff_size; - other_comm->src_data = data; + other_comm->src_actor_ = src_proc; + other_comm->task_size_ = task_size; + other_comm->rate_ = rate; + other_comm->src_buff_ = src_buff; + other_comm->src_buff_size_ = src_buff_size; + other_comm->src_data_ = data; other_comm->match_fun = match_fun; other_comm->copy_data_fun = copy_data_fun; @@ -153,7 +99,8 @@ XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_isend( return (detached ? nullptr : other_comm); } - SIMIX_comm_start(other_comm); + other_comm->start(); + return (detached ? nullptr : other_comm); } @@ -163,7 +110,8 @@ XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t re void (*copy_data_fun)(smx_activity_t, void*, size_t), void* data, double timeout, double rate) { - smx_activity_t comm = SIMIX_comm_irecv(receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, data, rate); + smx_activity_t comm = simcall_HANDLER_comm_irecv(simcall, receiver, mbox, dst_buff, dst_buff_size, match_fun, + copy_data_fun, data, rate); SIMCALL_SET_MC_VALUE(simcall, 0); simcall_HANDLER_comm_wait(simcall, comm, timeout); } @@ -174,37 +122,29 @@ XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, void (*copy_data_fun)(smx_activity_t, void*, size_t), void* data, double rate) { - return SIMIX_comm_irecv(receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, data, rate); -} - -smx_activity_t -SIMIX_comm_irecv(smx_actor_t dst_proc, smx_mailbox_t mbox, void* dst_buff, size_t* dst_buff_size, - int (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*), - void (*copy_data_fun)(smx_activity_t, void*, size_t), // used to copy data if not default one - void* data, double rate) -{ - simgrid::kernel::activity::CommImplPtr this_synchro = - simgrid::kernel::activity::CommImplPtr(new simgrid::kernel::activity::CommImpl(SIMIX_COMM_RECEIVE)); + simgrid::kernel::activity::CommImplPtr this_synchro = simgrid::kernel::activity::CommImplPtr( + new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::RECEIVE)); XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get()); simgrid::kernel::activity::CommImplPtr other_comm; //communication already done, get it inside the list of completed comms - if (mbox->permanent_receiver != nullptr && not mbox->done_comm_queue.empty()) { + if (mbox->permanent_receiver_ != nullptr && not mbox->done_comm_queue_.empty()) { XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication"); //find a match in the list of already received comms - other_comm = _find_matching_comm(&mbox->done_comm_queue, SIMIX_COMM_SEND, match_fun, data, this_synchro, - /*remove_matching*/ true); + other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data, + this_synchro, /*done*/ true, + /*remove_matching*/ true); //if not found, assume the receiver came first, register it to the mailbox in the classical way if (not other_comm) { XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request into list"); other_comm = std::move(this_synchro); mbox->push(other_comm); } else { - if (other_comm->surfAction_ && other_comm->remains() < 1e-12) { + if (other_comm->surf_action_ && other_comm->remains() < 1e-12) { XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get()); other_comm->state_ = SIMIX_DONE; - other_comm->type = SIMIX_COMM_DONE; + other_comm->type = simgrid::kernel::activity::CommImpl::Type::DONE; other_comm->mbox = nullptr; } } @@ -215,30 +155,31 @@ SIMIX_comm_irecv(smx_actor_t dst_proc, smx_mailbox_t mbox, void* dst_buff, size_ * ourself so that the other side also gets a chance of choosing if it wants to match with us. * * If it is not found then push our communication into the rendez-vous point */ - other_comm = _find_matching_comm(&mbox->comm_queue, SIMIX_COMM_SEND, match_fun, data, this_synchro, - /*remove_matching*/ true); + other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data, + this_synchro, /*done*/ false, + /*remove_matching*/ true); if (other_comm == nullptr) { - XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->comm_queue.size()); + XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->comm_queue_.size()); other_comm = std::move(this_synchro); mbox->push(other_comm); } else { XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get()); other_comm->state_ = SIMIX_READY; - other_comm->type = SIMIX_COMM_READY; + other_comm->type = simgrid::kernel::activity::CommImpl::Type::READY; } - dst_proc->comms.push_back(other_comm); + receiver->comms.push_back(other_comm); } /* Setup communication synchro */ - other_comm->dst_proc = dst_proc; - other_comm->dst_buff = dst_buff; - other_comm->dst_buff_size = dst_buff_size; - other_comm->dst_data = data; + other_comm->dst_actor_ = receiver; + other_comm->dst_buff_ = dst_buff; + other_comm->dst_buff_size_ = dst_buff_size; + other_comm->dst_data_ = data; - if (rate > -1.0 && (other_comm->rate < 0.0 || rate < other_comm->rate)) - other_comm->rate = rate; + if (rate > -1.0 && (other_comm->rate_ < 0.0 || rate < other_comm->rate_)) + other_comm->rate_ = rate; other_comm->match_fun = match_fun; other_comm->copy_data_fun = copy_data_fun; @@ -247,45 +188,10 @@ SIMIX_comm_irecv(smx_actor_t dst_proc, smx_mailbox_t mbox, void* dst_buff, size_ other_comm->state_ = SIMIX_RUNNING; return other_comm; } - - SIMIX_comm_start(other_comm); + other_comm->start(); return other_comm; } -smx_activity_t simcall_HANDLER_comm_iprobe(smx_simcall_t simcall, smx_mailbox_t mbox, int type, - simix_match_func_t match_fun, void* data) -{ - return SIMIX_comm_iprobe(simcall->issuer, mbox, type, match_fun, data); -} - -smx_activity_t SIMIX_comm_iprobe(smx_actor_t dst_proc, smx_mailbox_t mbox, int type, simix_match_func_t match_fun, - void* data) -{ - XBT_DEBUG("iprobe from %p %p", mbox, &mbox->comm_queue); - simgrid::kernel::activity::CommImplPtr this_comm; - int smx_type; - if(type == 1){ - this_comm = simgrid::kernel::activity::CommImplPtr(new simgrid::kernel::activity::CommImpl(SIMIX_COMM_SEND)); - smx_type = SIMIX_COMM_RECEIVE; - } else{ - this_comm = simgrid::kernel::activity::CommImplPtr(new simgrid::kernel::activity::CommImpl(SIMIX_COMM_RECEIVE)); - smx_type = SIMIX_COMM_SEND; - } - smx_activity_t other_synchro=nullptr; - if (mbox->permanent_receiver != nullptr && not mbox->done_comm_queue.empty()) { - XBT_DEBUG("first check in the permanent recv mailbox, to see if we already got something"); - other_synchro = _find_matching_comm(&mbox->done_comm_queue, - (e_smx_comm_type_t) smx_type, match_fun, data, this_comm,/*remove_matching*/false); - } - if (not other_synchro) { - XBT_DEBUG("check if we have more luck in the normal mailbox"); - other_synchro = _find_matching_comm(&mbox->comm_queue, - (e_smx_comm_type_t) smx_type, match_fun, data, this_comm,/*remove_matching*/false); - } - - return other_synchro; -} - void simcall_HANDLER_comm_wait(smx_simcall_t simcall, smx_activity_t synchro, double timeout) { /* Associate this simcall to the wait synchro */ @@ -306,7 +212,7 @@ void simcall_HANDLER_comm_wait(smx_simcall_t simcall, smx_activity_t synchro, do simgrid::kernel::activity::CommImplPtr comm = boost::static_pointer_cast(synchro); - if (comm->src_proc == simcall->issuer) + if (comm->src_actor_ == simcall->issuer) comm->state_ = SIMIX_SRC_TIMEOUT; else comm->state_ = SIMIX_DST_TIMEOUT; @@ -320,17 +226,16 @@ void simcall_HANDLER_comm_wait(smx_simcall_t simcall, smx_activity_t synchro, do /* otherwise set up a waiting timeout on the right side */ if (synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING) { SIMIX_comm_finish(synchro); - } else { /* we need a surf sleep action even when there is no timeout, otherwise surf won't tell us when the host - fails */ - simgrid::kernel::resource::Action* sleep = simcall->issuer->host_->pimpl_cpu->sleep(timeout); + } else { /* we need a sleep action (even when there is no timeout) to be notified of host failures */ + simgrid::kernel::resource::Action* sleep = simcall->issuer->get_host()->pimpl_cpu->sleep(timeout); sleep->set_data(synchro.get()); simgrid::kernel::activity::CommImplPtr comm = boost::static_pointer_cast(synchro); - if (simcall->issuer == comm->src_proc) - comm->src_timeout = sleep; + if (simcall->issuer == comm->src_actor_) + comm->src_timeout_ = sleep; else - comm->dst_timeout = sleep; + comm->dst_timeout_ = sleep; } } @@ -342,7 +247,7 @@ void simcall_HANDLER_comm_test(smx_simcall_t simcall, smx_activity_t synchro) int res; if (MC_is_active() || MC_record_replay_is_active()){ - res = comm->src_proc && comm->dst_proc; + res = comm->src_actor_ && comm->dst_actor_; if (res) synchro->state_ = SIMIX_DONE; } else { @@ -408,7 +313,7 @@ void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, xbt_dynar_t synchros, d if (timeout < 0.0){ simcall->timer = NULL; } else { - simcall->timer = SIMIX_timer_set(SIMIX_get_clock() + timeout, [simcall]() { + simcall->timer = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall]() { SIMIX_waitany_remove_simcall_from_actions(simcall); simcall_comm_waitany__set__result(simcall, -1); SIMIX_simcall_answer(simcall); @@ -447,52 +352,8 @@ void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall) } /** - * \brief Starts the simulation of a communication synchro. - * \param comm the communication that will be started - */ -static inline void SIMIX_comm_start(simgrid::kernel::activity::CommImplPtr comm) -{ - /* If both the sender and the receiver are already there, start the communication */ - if (comm->state_ == SIMIX_READY) { - - simgrid::s4u::Host* sender = comm->src_proc->host_; - simgrid::s4u::Host* receiver = comm->dst_proc->host_; - - comm->surfAction_ = surf_network_model->communicate(sender, receiver, comm->task_size, comm->rate); - comm->surfAction_->set_data(comm.get()); - comm->state_ = SIMIX_RUNNING; - - XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p)", comm.get(), sender->get_cname(), - receiver->get_cname(), comm->surfAction_); - - /* If a link is failed, detect it immediately */ - if (comm->surfAction_->get_state() == simgrid::kernel::resource::Action::State::FAILED) { - XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", sender->get_cname(), - receiver->get_cname()); - comm->state_ = SIMIX_LINK_FAILURE; - comm->cleanupSurf(); - } - - /* If any of the process is suspended, create the synchro but stop its execution, - it will be restarted when the sender process resume */ - if (comm->src_proc->is_suspended() || comm->dst_proc->is_suspended()) { - if (comm->src_proc->is_suspended()) - XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the " - "communication", - comm->src_proc->get_cname(), comm->src_proc->host_->get_cname()); - else - XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the " - "communication", - comm->dst_proc->get_cname(), comm->dst_proc->host_->get_cname()); - - comm->surfAction_->suspend(); - } - } -} - -/** - * \brief Answers the SIMIX simcalls associated to a communication synchro. - * \param synchro a finished communication synchro + * @brief Answers the SIMIX simcalls associated to a communication synchro. + * @param synchro a finished communication synchro */ void SIMIX_comm_finish(smx_activity_t synchro) { @@ -512,7 +373,7 @@ void SIMIX_comm_finish(smx_activity_t synchro) if (simcall->call == SIMCALL_COMM_WAITANY) { SIMIX_waitany_remove_simcall_from_actions(simcall); if (simcall->timer) { - SIMIX_timer_remove(simcall->timer); + simcall->timer->remove(); simcall->timer = nullptr; } if (not MC_is_active() && not MC_record_replay_is_active()) @@ -528,60 +389,68 @@ void SIMIX_comm_finish(smx_activity_t synchro) /* Check out for errors */ - if (simcall->issuer->host_->is_off()) { - simcall->issuer->context_->iwannadie = 1; - SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed"); + if (not simcall->issuer->get_host()->is_on()) { + simcall->issuer->context_->iwannadie = true; + simcall->issuer->exception_ = + std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed")); } else { switch (comm->state_) { case SIMIX_DONE: XBT_DEBUG("Communication %p complete!", synchro.get()); - SIMIX_comm_copy_data(synchro); + comm->copy_data(); break; case SIMIX_SRC_TIMEOUT: - SMX_EXCEPTION(simcall->issuer, timeout_error, 0, "Communication timeouted because of sender"); + simcall->issuer->exception_ = std::make_exception_ptr( + simgrid::TimeoutError(XBT_THROW_POINT, "Communication timeouted because of the sender")); break; case SIMIX_DST_TIMEOUT: - SMX_EXCEPTION(simcall->issuer, timeout_error, 0, "Communication timeouted because of receiver"); + simcall->issuer->exception_ = std::make_exception_ptr( + simgrid::TimeoutError(XBT_THROW_POINT, "Communication timeouted because of the receiver")); break; case SIMIX_SRC_HOST_FAILURE: - if (simcall->issuer == comm->src_proc) - simcall->issuer->context_->iwannadie = 1; + if (simcall->issuer == comm->src_actor_) + simcall->issuer->context_->iwannadie = true; else - SMX_EXCEPTION(simcall->issuer, network_error, 0, "Remote peer failed"); + simcall->issuer->exception_ = + std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); break; case SIMIX_DST_HOST_FAILURE: - if (simcall->issuer == comm->dst_proc) - simcall->issuer->context_->iwannadie = 1; + if (simcall->issuer == comm->dst_actor_) + simcall->issuer->context_->iwannadie = true; else - SMX_EXCEPTION(simcall->issuer, network_error, 0, "Remote peer failed"); + simcall->issuer->exception_ = + std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); break; case SIMIX_LINK_FAILURE: XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) " "detached:%d", - synchro.get(), comm->src_proc ? comm->src_proc->host_->get_cname() : nullptr, - comm->dst_proc ? comm->dst_proc->host_->get_cname() : nullptr, simcall->issuer->get_cname(), - simcall->issuer, comm->detached); - if (comm->src_proc == simcall->issuer) { + synchro.get(), comm->src_actor_ ? comm->src_actor_->get_host()->get_cname() : nullptr, + comm->dst_actor_ ? comm->dst_actor_->get_host()->get_cname() : nullptr, + simcall->issuer->get_cname(), simcall->issuer, comm->detached); + if (comm->src_actor_ == simcall->issuer) { XBT_DEBUG("I'm source"); - } else if (comm->dst_proc == simcall->issuer) { + } else if (comm->dst_actor_ == simcall->issuer) { XBT_DEBUG("I'm dest"); } else { XBT_DEBUG("I'm neither source nor dest"); } - SMX_EXCEPTION(simcall->issuer, network_error, 0, "Link failure"); + simcall->issuer->throw_exception( + std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Link failure"))); break; case SIMIX_CANCELED: - if (simcall->issuer == comm->dst_proc) - SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Communication canceled by the sender"); + if (simcall->issuer == comm->dst_actor_) + simcall->issuer->exception_ = std::make_exception_ptr( + simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the sender")); else - SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Communication canceled by the receiver"); + simcall->issuer->exception_ = std::make_exception_ptr( + simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the receiver")); break; default: @@ -590,123 +459,70 @@ void SIMIX_comm_finish(smx_activity_t synchro) } /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */ - if (simcall->issuer->exception) { + if (simcall->issuer->exception_ && + (simcall->call == SIMCALL_COMM_WAITANY || simcall->call == SIMCALL_COMM_TESTANY)) { + // First retrieve the rank of our failing synchro + int rank = -1; + if (simcall->call == SIMCALL_COMM_WAITANY) { + rank = xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &synchro); + } else if (simcall->call == SIMCALL_COMM_TESTANY) { + rank = -1; + auto* comms = simcall_comm_testany__get__comms(simcall); + auto count = simcall_comm_testany__get__count(simcall); + auto element = std::find(comms, comms + count, synchro); + if (element == comms + count) + rank = -1; + else + rank = element - comms; + } + // In order to modify the exception we have to rethrow it: try { - std::rethrow_exception(simcall->issuer->exception); - } - catch(xbt_ex& e) { - if (simcall->call == SIMCALL_COMM_WAITANY) { - e.value = xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &synchro); - } - else if (simcall->call == SIMCALL_COMM_TESTANY) { - e.value = -1; - auto* comms = simcall_comm_testany__get__comms(simcall); - auto count = simcall_comm_testany__get__count(simcall); - auto element = std::find(comms, comms + count, synchro); - if (element == comms + count) - e.value = -1; - else - e.value = element - comms; - } - simcall->issuer->exception = std::make_exception_ptr(e); + std::rethrow_exception(simcall->issuer->exception_); + } catch (simgrid::TimeoutError& e) { + e.value = rank; + simcall->issuer->exception_ = std::make_exception_ptr(e); + } catch (simgrid::NetworkFailureException& e) { + e.value = rank; + simcall->issuer->exception_ = std::make_exception_ptr(e); + } catch (simgrid::CancelException& e) { + e.value = rank; + simcall->issuer->exception_ = std::make_exception_ptr(e); } - catch(...) { - // Nothing to do - } - } - - if (simcall->issuer->host_->is_off()) { - simcall->issuer->context_->iwannadie = 1; } simcall->issuer->waiting_synchro = nullptr; simcall->issuer->comms.remove(synchro); if(comm->detached){ - if(simcall->issuer == comm->src_proc){ - if(comm->dst_proc) - comm->dst_proc->comms.remove(synchro); - } - else if(simcall->issuer == comm->dst_proc){ - if(comm->src_proc) - comm->src_proc->comms.remove(synchro); + if (simcall->issuer == comm->src_actor_) { + if (comm->dst_actor_) + comm->dst_actor_->comms.remove(synchro); + } else if (simcall->issuer == comm->dst_actor_) { + if (comm->src_actor_) + comm->src_actor_->comms.remove(synchro); } else{ - comm->dst_proc->comms.remove(synchro); - comm->src_proc->comms.remove(synchro); + comm->dst_actor_->comms.remove(synchro); + comm->src_actor_->comms.remove(synchro); } } - SIMIX_simcall_answer(simcall); + if (simcall->issuer->get_host()->is_on()) + SIMIX_simcall_answer(simcall); + else + simcall->issuer->context_->iwannadie = true; } } -/******************************************************************************/ -/* SIMIX_comm_copy_data callbacks */ -/******************************************************************************/ -static void (*SIMIX_comm_copy_data_callback) (smx_activity_t, void*, size_t) = &SIMIX_comm_copy_pointer_callback; - -void SIMIX_comm_set_copy_data_callback(void (*callback) (smx_activity_t, void*, size_t)) -{ - SIMIX_comm_copy_data_callback = callback; -} - -void SIMIX_comm_copy_pointer_callback(smx_activity_t synchro, void* buff, size_t buff_size) -{ - simgrid::kernel::activity::CommImplPtr comm = - boost::static_pointer_cast(synchro); - - xbt_assert((buff_size == sizeof(void *)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size); - *(void **) (comm->dst_buff) = buff; -} - void SIMIX_comm_copy_buffer_callback(smx_activity_t synchro, void* buff, size_t buff_size) { simgrid::kernel::activity::CommImplPtr comm = boost::static_pointer_cast(synchro); XBT_DEBUG("Copy the data over"); - memcpy(comm->dst_buff, buff, buff_size); + memcpy(comm->dst_buff_, buff, buff_size); if (comm->detached) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the original buffer available to the application ASAP xbt_free(buff); - comm->src_buff = nullptr; - } -} - -/** - * @brief Copy the communication data from the sender's buffer to the receiver's one - * @param synchro The communication - */ -void SIMIX_comm_copy_data(smx_activity_t synchro) -{ - simgrid::kernel::activity::CommImplPtr comm = - boost::static_pointer_cast(synchro); - - size_t buff_size = comm->src_buff_size; - /* If there is no data to copy then return */ - if (not comm->src_buff || not comm->dst_buff || comm->copied) - return; - - XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", comm.get(), - comm->src_proc ? comm->src_proc->host_->get_cname() : "a finished process", comm->src_buff, - comm->dst_proc ? comm->dst_proc->host_->get_cname() : "a finished process", comm->dst_buff, buff_size); - - /* Copy at most dst_buff_size bytes of the message to receiver's buffer */ - if (comm->dst_buff_size) - buff_size = std::min(buff_size, *(comm->dst_buff_size)); - - /* Update the receiver's buffer size to the copied amount */ - if (comm->dst_buff_size) - *comm->dst_buff_size = buff_size; - - if (buff_size > 0){ - if(comm->copy_data_fun) - comm->copy_data_fun (comm, comm->src_buff, buff_size); - else - SIMIX_comm_copy_data_callback (comm, comm->src_buff, buff_size); + comm->src_buff_ = nullptr; } - - /* Set the copied flag so we copy data only once */ - /* (this function might be called from both communication ends) */ - comm->copied = 1; }