X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/c6095437348b2a76e6cd24e3e98c81ecd9b8164a..ea74f5d95928a521a588737e81f1de94eef25d19:/src/s4u/s4u_Comm.cpp diff --git a/src/s4u/s4u_Comm.cpp b/src/s4u/s4u_Comm.cpp index 8548847e08..aed48fdd8b 100644 --- a/src/s4u/s4u_Comm.cpp +++ b/src/s4u/s4u_Comm.cpp @@ -1,56 +1,112 @@ -/* Copyright (c) 2006-2021. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2006-2022. 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 "src/msg/msg_private.hpp" -#include "xbt/log.h" - -#include "simgrid/Exception.hpp" -#include "simgrid/s4u/Comm.hpp" -#include "simgrid/s4u/Mailbox.hpp" +//#include "src/msg/msg_private.hpp" +//#include "xbt/log.h" +#include #include +#include +#include +#include + +#include "src/kernel/activity/CommImpl.hpp" +#include "src/kernel/actor/ActorImpl.hpp" XBT_LOG_NEW_DEFAULT_SUBCATEGORY(s4u_comm, s4u_activity, "S4U asynchronous communications"); namespace simgrid { namespace s4u { -xbt::signal Comm::on_start; +xbt::signal Comm::on_send; +xbt::signal Comm::on_recv; xbt::signal Comm::on_completion; Comm::~Comm() { if (state_ == State::STARTED && not detached_ && (pimpl_ == nullptr || pimpl_->state_ == kernel::activity::State::RUNNING)) { - XBT_INFO("Comm %p freed before its completion. Detached: %d, State: %d", this, detached_, (int)state_); + XBT_INFO("Comm %p freed before its completion. Did you forget to detach it? (state: %s)", this, get_state_str()); if (pimpl_ != nullptr) - XBT_INFO("pimpl_->state: %d", static_cast(pimpl_->state_)); + XBT_INFO("pimpl_->state: %s", pimpl_->get_state_str()); else XBT_INFO("pimpl_ is null"); xbt_backtrace_display_current(); } } -int Comm::wait_any_for(const std::vector* comms, double timeout) +ssize_t Comm::wait_any_for(const std::vector& comms, double timeout) { - std::vector rcomms(comms->size()); - std::transform(begin(*comms), end(*comms), begin(rcomms), + std::vector rcomms(comms.size()); + std::transform(begin(comms), end(comms), begin(rcomms), [](const CommPtr& comm) { return static_cast(comm->pimpl_.get()); }); - int changed_pos = simcall_comm_waitany(rcomms.data(), rcomms.size(), timeout); + ssize_t changed_pos; + try { + changed_pos = simcall_comm_waitany(rcomms.data(), rcomms.size(), timeout); + } catch (const NetworkFailureException& e) { + for (auto c : comms) { + if (c->pimpl_->state_ == kernel::activity::State::FAILED) { + c->complete(State::FAILED); + } + } + e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode."); + } if (changed_pos != -1) - comms->at(changed_pos)->release_dependencies(); + comms.at(changed_pos)->complete(State::FINISHED); return changed_pos; } -void Comm::wait_all(const std::vector* comms) +void Comm::wait_all(const std::vector& comms) { // TODO: this should be a simcall or something - // TODO: we are missing a version with timeout - for (CommPtr comm : *comms) + for (auto& comm : comms) comm->wait(); } +size_t Comm::wait_all_for(const std::vector& comms, double timeout) +{ + if (timeout < 0.0) { + wait_all(comms); + return comms.size(); + } + + double deadline = Engine::get_clock() + timeout; + std::vector waited_comm(1, nullptr); + for (size_t i = 0; i < comms.size(); i++) { + double wait_timeout = std::max(0.0, deadline - Engine::get_clock()); + waited_comm[0] = comms[i]; + // Using wait_any_for() here (and not wait_for) because we don't want comms to be invalidated on timeout + if (wait_any_for(waited_comm, wait_timeout) == -1) { + XBT_DEBUG("Timeout (%g): i = %zu", wait_timeout, i); + return i; + } + } + return comms.size(); +} + +CommPtr Comm::set_source(Host* from) +{ + xbt_assert(state_ == State::INITED || state_ == State::STARTING, + "Cannot change the source of a Comm once it's started (state: %s)", to_c_str(state_)); + from_ = from; + // Setting 'from_' may allow to start the activity, let's try + vetoable_start(); + + return this; +} + +CommPtr Comm::set_destination(Host* to) +{ + xbt_assert(state_ == State::INITED || state_ == State::STARTING, + "Cannot change the destination of a Comm once it's started (state: %s)", to_c_str(state_)); + to_ = to; + // Setting 'to_' may allow to start the activity, let's try + vetoable_start(); + + return this; +} + CommPtr Comm::set_rate(double rate) { xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)", @@ -114,22 +170,40 @@ CommPtr Comm::set_dst_data(void** buff, size_t size) dst_buff_size_ = size; return this; } +CommPtr Comm::set_payload_size(uint64_t bytes) +{ + Activity::set_remaining(bytes); + return this; +} -CommPtr Comm::sendto_init(Host* from, Host* to) +CommPtr Comm::sendto_init() { CommPtr res(new Comm()); + res->sender_ = kernel::actor::ActorImpl::self(); + return res; +} + +CommPtr Comm::sendto_init(Host* from, Host* to) +{ + auto res = Comm::sendto_init(); res->from_ = from; res->to_ = to; return res; } -CommPtr Comm::sendto_async(Host* from, Host* to, double simulated_size_in_bytes) + +CommPtr Comm::sendto_async(Host* from, Host* to, uint64_t simulated_size_in_bytes) { - auto res = Comm::sendto_init(from, to); - res->set_remaining(simulated_size_in_bytes)->start(); + auto res = Comm::sendto_init(from, to)->set_payload_size(simulated_size_in_bytes); + res->vetoable_start(); return res; } +void Comm::sendto(Host* from, Host* to, uint64_t simulated_size_in_bytes) +{ + sendto_async(from, to, simulated_size_in_bytes)->wait(); +} + Comm* Comm::start() { xbt_assert(get_state() == State::INITED || get_state() == State::STARTING, @@ -139,18 +213,18 @@ Comm* Comm::start() xbt_assert(src_buff_ == nullptr && dst_buff_ == nullptr, "Direct host-to-host communications cannot carry any data."); pimpl_ = kernel::actor::simcall([this] { - auto res = new kernel::activity::CommImpl(this->from_, this->to_, this->get_remaining()); + kernel::activity::CommImplPtr res(new kernel::activity::CommImpl(this->from_, this->to_, this->get_remaining())); res->start(); return res; }); } else if (src_buff_ != nullptr) { // Sender side - on_start(*this, true /* is_sender*/); + on_send(*this); pimpl_ = simcall_comm_isend(sender_, mailbox_->get_impl(), remains_, rate_, src_buff_, src_buff_size_, match_fun_, clean_fun_, copy_data_function_, get_user_data(), detached_); } else if (dst_buff_ != nullptr) { // Receiver side xbt_assert(not detached_, "Receive cannot be detached"); - on_start(*this, false /*is_sender*/); + on_recv(*this); pimpl_ = simcall_comm_irecv(receiver_, mailbox_->get_impl(), dst_buff_, &dst_buff_size_, match_fun_, copy_data_function_, get_user_data(), rate_); @@ -161,16 +235,15 @@ Comm* Comm::start() if (suspended_) pimpl_->suspend(); + if (not detached_) { + pimpl_->set_iface(this); + pimpl_->set_actor(sender_); + } + state_ = State::STARTED; return this; } -/** @brief Block the calling actor until the communication is finished */ -Comm* Comm::wait() -{ - return this->wait_for(-1); -} - /** @brief Block the calling actor until the communication is finished, or until timeout * * On timeout, an exception is thrown and the communication is invalidated. @@ -182,29 +255,32 @@ Comm* Comm::wait_for(double timeout) switch (state_) { case State::FINISHED: break; + case State::FAILED: + throw NetworkFailureException(XBT_THROW_POINT, "Cannot wait for a failed communication"); case State::INITED: case State::STARTING: // It's not started yet. Do it in one simcall if it's a regular communication if (from_ != nullptr || to_ != nullptr) { - return start()->wait_for(timeout); // In the case of host2host comm, do it in two simcalls + return vetoable_start()->wait_for(timeout); // In the case of host2host comm, do it in two simcalls } else if (src_buff_ != nullptr) { - on_start(*this, true /*is_sender*/); + on_send(*this); simcall_comm_send(sender_, mailbox_->get_impl(), remains_, rate_, src_buff_, src_buff_size_, match_fun_, copy_data_function_, get_user_data(), timeout); } else { // Receiver - on_start(*this, false /*is_sender*/); + on_recv(*this); simcall_comm_recv(receiver_, mailbox_->get_impl(), dst_buff_, &dst_buff_size_, match_fun_, copy_data_function_, get_user_data(), timeout, rate_); } - state_ = State::FINISHED; - this->release_dependencies(); break; case State::STARTED: - simcall_comm_wait(get_impl(), timeout); - state_ = State::FINISHED; - this->release_dependencies(); + try { + simcall_comm_wait(get_impl(), timeout); + } catch (const NetworkFailureException& e) { + complete(State::FAILED); + e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode."); + } break; case State::CANCELED: @@ -213,18 +289,18 @@ Comm* Comm::wait_for(double timeout) default: THROW_IMPOSSIBLE; } - on_completion(*this); + complete(State::FINISHED); return this; } -int Comm::test_any(const std::vector* comms) +ssize_t Comm::test_any(const std::vector& comms) { - std::vector rcomms(comms->size()); - std::transform(begin(*comms), end(*comms), begin(rcomms), + std::vector rcomms(comms.size()); + std::transform(begin(comms), end(comms), begin(rcomms), [](const CommPtr& comm) { return static_cast(comm->pimpl_.get()); }); - int changed_pos = simcall_comm_testany(rcomms.data(), rcomms.size()); + ssize_t changed_pos = simcall_comm_testany(rcomms.data(), rcomms.size()); if (changed_pos != -1) - comms->at(changed_pos)->release_dependencies(); + comms.at(changed_pos)->complete(State::FINISHED); return changed_pos; } @@ -238,30 +314,19 @@ Comm* Comm::detach() return this; } -Comm* Comm::cancel() -{ - kernel::actor::simcall([this] { - if (pimpl_) - boost::static_pointer_cast(pimpl_)->cancel(); - }); - state_ = State::CANCELED; - return this; -} - -bool Comm::test() +bool Comm::test() // TODO: merge with Activity::test, once modernized { xbt_assert(state_ == State::INITED || state_ == State::STARTED || state_ == State::STARTING || - state_ == State::FINISHED); + state_ == State::CANCELED || state_ == State::FINISHED); - if (state_ == State::FINISHED) + if (state_ == State::CANCELED || state_ == State::FINISHED) return true; if (state_ == State::INITED || state_ == State::STARTING) this->vetoable_start(); if (simcall_comm_test(get_impl())) { - state_ = State::FINISHED; - this->release_dependencies(); + complete(State::FINISHED); return true; } return false; @@ -280,6 +345,28 @@ Actor* Comm::get_sender() const return sender ? sender->get_ciface() : nullptr; } +CommPtr Comm::set_copy_data_callback(void (*callback)(kernel::activity::CommImpl*, void*, size_t)) +{ + copy_data_function_ = callback; + return this; +} +void Comm::copy_buffer_callback(kernel::activity::CommImpl* comm, void* buff, size_t buff_size) +{ + XBT_DEBUG("Copy the data over"); + 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; + } +} + +void Comm::copy_pointer_callback(kernel::activity::CommImpl* comm, void* buff, size_t buff_size) +{ + xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size); + *(void**)(comm->dst_buff_) = buff; +} + } // namespace s4u } // namespace simgrid /* **************************** Public C interface *************************** */ @@ -302,19 +389,7 @@ int sg_comm_test(sg_comm_t comm) sg_error_t sg_comm_wait(sg_comm_t comm) { - sg_error_t status = SG_OK; - - simgrid::s4u::CommPtr s4u_comm(comm, false); - try { - s4u_comm->wait_for(-1); - } catch (const simgrid::TimeoutException&) { - status = SG_ERROR_TIMEOUT; - } catch (const simgrid::CancelException&) { - status = SG_ERROR_CANCELED; - } catch (const simgrid::NetworkFailureException&) { - status = SG_ERROR_NETWORK; - } - return status; + return sg_comm_wait_for(comm, -1); } sg_error_t sg_comm_wait_for(sg_comm_t comm, double timeout) @@ -335,28 +410,36 @@ sg_error_t sg_comm_wait_for(sg_comm_t comm, double timeout) } void sg_comm_wait_all(sg_comm_t* comms, size_t count) +{ + sg_comm_wait_all_for(comms, count, -1); +} + +size_t sg_comm_wait_all_for(sg_comm_t* comms, size_t count, double timeout) { std::vector s4u_comms; - for (unsigned int i = 0; i < count; i++) + for (size_t i = 0; i < count; i++) s4u_comms.emplace_back(comms[i], false); - simgrid::s4u::Comm::wait_all(&s4u_comms); + size_t pos = simgrid::s4u::Comm::wait_all_for(s4u_comms, timeout); + for (size_t i = pos; i < count; i++) + s4u_comms[i]->add_ref(); + return pos; } -int sg_comm_wait_any(sg_comm_t* comms, size_t count) +ssize_t sg_comm_wait_any(sg_comm_t* comms, size_t count) { return sg_comm_wait_any_for(comms, count, -1); } -int sg_comm_wait_any_for(sg_comm_t* comms, size_t count, double timeout) +ssize_t sg_comm_wait_any_for(sg_comm_t* comms, size_t count, double timeout) { std::vector s4u_comms; - for (unsigned int i = 0; i < count; i++) + for (size_t i = 0; i < count; i++) s4u_comms.emplace_back(comms[i], false); - int pos = simgrid::s4u::Comm::wait_any_for(&s4u_comms, timeout); - for (unsigned i = 0; i < count; i++) { - if (pos != -1 && static_cast(pos) != i) + ssize_t pos = simgrid::s4u::Comm::wait_any_for(s4u_comms, timeout); + for (size_t i = 0; i < count; i++) { + if (pos != -1 && static_cast(pos) != i) s4u_comms[i]->add_ref(); } return pos;