X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/4485d715fa3a090c1dd38def83ab8239feebecb3..1c75860aa49dbde3f251678a7d00a2e27660efa8:/src/kernel/activity/ExecImpl.cpp diff --git a/src/kernel/activity/ExecImpl.cpp b/src/kernel/activity/ExecImpl.cpp index cac429ac5f..cfdf73e8e1 100644 --- a/src/kernel/activity/ExecImpl.cpp +++ b/src/kernel/activity/ExecImpl.cpp @@ -3,95 +3,186 @@ /* 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 "src/kernel/activity/ExecImpl.hpp" +#include "simgrid/Exception.hpp" #include "simgrid/modelchecker.h" #include "src/mc/mc_replay.hpp" - -#include "src/kernel/activity/ExecImpl.hpp" -#include "src/simix/smx_host_private.hpp" -#include "src/surf/surf_interface.hpp" +#include "src/surf/HostImpl.hpp" #include "src/surf/cpu_interface.hpp" +#include "src/surf/surf_interface.hpp" #include "simgrid/s4u/Host.hpp" +#include + XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process); -namespace simgrid { -namespace kernel { -namespace activity { -ExecImpl::ExecImpl(std::string name, std::string tracing_category, resource::Action* timeout_detector, s4u::Host* host) - : ActivityImpl(name), host_(host), timeout_detector_(timeout_detector) +void simcall_HANDLER_execution_wait(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro) { - this->state_ = SIMIX_RUNNING; - this->set_category(tracing_category); + XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state_); - if (timeout_detector != nullptr) - timeout_detector_->set_data(this); + /* Associate this simcall to the synchro */ + synchro->register_simcall(simcall); - XBT_DEBUG("Create exec %p", this); + /* set surf's synchro */ + if (MC_is_active() || MC_record_replay_is_active()) { + synchro->state_ = SIMIX_DONE; + synchro->finish(); + return; + } + + /* If the synchro is already finished then perform the error handling */ + if (synchro->state_ != SIMIX_RUNNING) + synchro->finish(); } +void simcall_HANDLER_execution_test(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro) +{ + bool res = (synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING); + if (res) { + synchro->simcalls_.push_back(simcall); + synchro->finish(); + } else { + simcall->issuer->simcall_answer(); + } + simcall_execution_test__set__result(simcall, res); +} + +void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[], + size_t count, double timeout) +{ + if (timeout < 0.0) { + simcall->timeout_cb = nullptr; + } else { + simcall->timeout_cb = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall, execs, count]() { + for (size_t i = 0; i < count; i++) { + // Remove the first occurence of simcall: + auto* exec = execs[i]; + auto j = boost::range::find(exec->simcalls_, simcall); + if (j != exec->simcalls_.end()) + exec->simcalls_.erase(j); + } + simcall_execution_waitany_for__set__result(simcall, -1); + simcall->issuer->simcall_answer(); + }); + } + + for (size_t i = 0; i < count; i++) { + /* associate this simcall to the the synchro */ + auto* exec = execs[i]; + exec->simcalls_.push_back(simcall); + + /* see if the synchro is already finished */ + if (exec->state_ != SIMIX_WAITING && exec->state_ != SIMIX_RUNNING) { + exec->finish(); + break; + } + } +} + +namespace simgrid { +namespace kernel { +namespace activity { + ExecImpl::~ExecImpl() { - if (surf_action_) - surf_action_->unref(); if (timeout_detector_) timeout_detector_->unref(); XBT_DEBUG("Destroy exec %p", this); } -ExecImpl* ExecImpl::start(double flops_amount, double priority, double bound) +ExecImpl& ExecImpl::set_host(s4u::Host* host) { - if (not MC_is_active() && not MC_record_replay_is_active()) { - surf_action_ = host_->pimpl_cpu->execution_start(flops_amount); - surf_action_->set_data(this); - surf_action_->set_priority(priority); - if (bound > 0) - surf_action_->set_bound(bound); + if (not hosts_.empty()) + hosts_.clear(); + hosts_.push_back(host); + return *this; +} + +ExecImpl& ExecImpl::set_hosts(const std::vector& hosts) +{ + hosts_ = hosts; + return *this; +} + +ExecImpl& ExecImpl::set_timeout(double timeout) +{ + if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) { + timeout_detector_ = hosts_.front()->pimpl_cpu->sleep(timeout); + timeout_detector_->set_activity(this); } + return *this; +} - XBT_DEBUG("Create execute synchro %p: %s", this, name_.c_str()); - ExecImpl::on_creation(this); - return this; +ExecImpl& ExecImpl::set_flops_amount(double flops_amount) +{ + if (not flops_amounts_.empty()) + flops_amounts_.clear(); + flops_amounts_.push_back(flops_amount); + return *this; } -void ExecImpl::cancel() +ExecImpl& ExecImpl::set_flops_amounts(const std::vector& flops_amounts) { - XBT_VERB("This exec %p is canceled", this); - if (surf_action_ != nullptr) - surf_action_->cancel(); + flops_amounts_ = flops_amounts; + return *this; } -double ExecImpl::get_remaining() +ExecImpl& ExecImpl::set_bytes_amounts(const std::vector& bytes_amounts) { - xbt_assert(host_ != nullptr, "Calling remains() on a parallel execution is not allowed. " - "We would need to return a vector instead of a scalar. " - "Did you mean remainingRatio() instead?"); - return surf_action_ ? surf_action_->get_remains() : 0; + bytes_amounts_ = bytes_amounts; + + return *this; } -double ExecImpl::get_remaining_ratio() +ExecImpl* ExecImpl::start() { - if (host_ == - nullptr) // parallel task: their remain is already between 0 and 1 (see comment in ExecImpl::get_remaining()) - return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains(); - else // Actually compute the ratio for sequential tasks - return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost(); + state_ = SIMIX_RUNNING; + if (not MC_is_active() && not MC_record_replay_is_active()) { + if (hosts_.size() == 1) { + surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front()); + surf_action_->set_sharing_penalty(sharing_penalty_); + surf_action_->set_category(get_tracing_category()); + + if (bound_ > 0) + surf_action_->set_bound(bound_); + } else { + surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1); + } + surf_action_->set_activity(this); + } + + XBT_DEBUG("Create execute synchro %p: %s", this, get_cname()); + ExecImpl::on_creation(*this); + return this; } -void ExecImpl::set_bound(double bound) +double ExecImpl::get_seq_remaining_ratio() { - if (surf_action_) - surf_action_->set_bound(bound); + return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost(); } -void ExecImpl::set_priority(double priority) + +double ExecImpl::get_par_remaining_ratio() +{ + // parallel task: their remain is already between 0 and 1 + return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains(); +} + +ExecImpl& ExecImpl::set_bound(double bound) { - if (surf_action_) - surf_action_->set_priority(priority); + bound_ = bound; + return *this; +} + +ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty) +{ + sharing_penalty_ = sharing_penalty; + return *this; } void ExecImpl::post() { - if (host_ && not host_->is_on()) { /* FIXME: handle resource failure for parallel tasks too */ + if (hosts_.size() == 1 && not hosts_.front()->is_on()) { /* FIXME: handle resource failure for parallel tasks too */ /* If the host running the synchro failed, notice it. This way, the asking * process can be killed if it runs on that host itself */ state_ = SIMIX_FAILED; @@ -104,12 +195,10 @@ void ExecImpl::post() state_ = SIMIX_DONE; } - on_completion(this); + on_completion(*this); + + clean_action(); - if (surf_action_) { - surf_action_->unref(); - surf_action_ = nullptr; - } if (timeout_detector_) { timeout_detector_->unref(); timeout_detector_ = nullptr; @@ -117,37 +206,113 @@ void ExecImpl::post() /* If there are simcalls associated with the synchro, then answer them */ if (not simcalls_.empty()) - SIMIX_execution_finish(this); + finish(); +} + +void ExecImpl::finish() +{ + while (not simcalls_.empty()) { + smx_simcall_t simcall = simcalls_.front(); + simcalls_.pop_front(); + + /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany + * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the + * simcall */ + + if (simcall->call == SIMCALL_NONE) // FIXME: maybe a better way to handle this case + continue; // if process handling comm is killed + if (simcall->call == SIMCALL_EXECUTION_WAITANY_FOR) { + simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall); + size_t count = simcall_execution_waitany_for__get__count(simcall); + + for (size_t i = 0; i < count; i++) { + // Remove the first occurence of simcall: + auto* exec = execs[i]; + auto j = boost::range::find(exec->simcalls_, simcall); + if (j != exec->simcalls_.end()) + exec->simcalls_.erase(j); + + if (simcall->timeout_cb) { + simcall->timeout_cb->remove(); + simcall->timeout_cb = nullptr; + } + } + + if (not MC_is_active() && not MC_record_replay_is_active()) { + ExecImpl** element = std::find(execs, execs + count, this); + int rank = (element != execs + count) ? element - execs : -1; + simcall_execution_waitany_for__set__result(simcall, rank); + } + } + + switch (state_) { + + case SIMIX_DONE: + /* do nothing, synchro done */ + XBT_DEBUG("ExecImpl::finish(): execution successful"); + break; + + case SIMIX_FAILED: + XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer->get_host()->get_cname()); + simcall->issuer->context_->iwannadie = true; + if (simcall->issuer->get_host()->is_on()) + simcall->issuer->exception_ = + std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed")); + /* else, the actor will be killed with no possibility to survive */ + break; + + case SIMIX_CANCELED: + XBT_DEBUG("ExecImpl::finish(): execution canceled"); + simcall->issuer->exception_ = + std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled")); + break; + + case SIMIX_TIMEOUT: + XBT_DEBUG("ExecImpl::finish(): execution timeouted"); + simcall->issuer->exception_ = std::make_exception_ptr(simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted")); + break; + + default: + xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast(state_)); + } + + simcall->issuer->waiting_synchro = nullptr; + /* Fail the process if the host is down */ + if (simcall->issuer->get_host()->is_on()) + simcall->issuer->simcall_answer(); + else + simcall->issuer->context_->iwannadie = true; + } } -ActivityImpl* ExecImpl::migrate(simgrid::s4u::Host* to) +ActivityImpl* ExecImpl::migrate(s4u::Host* to) { if (not MC_is_active() && not MC_record_replay_is_active()) { resource::Action* old_action = this->surf_action_; resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost()); new_action->set_remains(old_action->get_remains()); - new_action->set_data(this); - new_action->set_priority(old_action->get_priority()); + new_action->set_activity(this); + new_action->set_sharing_penalty(old_action->get_sharing_penalty()); // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling. // I hope that the user did not provide any. - old_action->set_data(nullptr); + old_action->set_activity(nullptr); old_action->cancel(); old_action->unref(); this->surf_action_ = new_action; } - on_migration(this, to); + on_migration(*this, to); return this; } /************* * Callbacks * *************/ -xbt::signal ExecImpl::on_creation; -xbt::signal ExecImpl::on_completion; -xbt::signal ExecImpl::on_migration; +xbt::signal ExecImpl::on_creation; +xbt::signal ExecImpl::on_completion; +xbt::signal ExecImpl::on_migration; } // namespace activity } // namespace kernel