+++ /dev/null
-/* Copyright (c) 2006-2021. 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 "simdag_private.hpp"
-#include "simgrid/kernel/routing/NetPoint.hpp"
-#include "src/surf/HostImpl.hpp"
-#include "src/surf/surf_interface.hpp"
-#include <algorithm>
-
-XBT_LOG_NEW_DEFAULT_SUBCATEGORY(sd_task, sd, "Logging specific to SimDag (task)");
-
-namespace simgrid {
-
-template class xbt::Extendable<sd::Task>;
-
-namespace sd {
-
-Task* Task::create(const std::string& name, double amount, void* userdata)
-{
- auto task = new Task();
- task->set_name(name);
- task->set_amount(amount);
- task->set_data(userdata);
- task->allocation_ = new std::vector<sg_host_t>();
- sd_global->initial_tasks.insert(task);
-
- return task;
-}
-
-Task* Task::create_comm_e2e(const std::string& name, double amount, void* userdata)
-{
- auto task = create(name, amount, userdata);
- task->bytes_amount_ = xbt_new0(double, 4);
- task->bytes_amount_[2] = amount;
- task->set_kind(SD_TASK_COMM_E2E);
-
- return task;
-}
-
-Task* Task::create_comp_seq(const std::string& name, double amount, void* userdata)
-{
- auto task = create(name, amount, userdata);
- task->flops_amount_ = xbt_new0(double, 1);
- task->flops_amount_[0] = amount;
- task->set_kind(SD_TASK_COMP_SEQ);
-
- return task;
-}
-
-Task* Task::create_comp_par_amdahl(const std::string& name, double amount, void* userdata, double alpha)
-{
- xbt_assert(alpha < 1. && alpha >= 0., "Invalid parameter: alpha must be in [0.;1.[");
-
- auto task = create(name, amount, userdata);
- task->set_alpha(alpha);
- task->set_kind(SD_TASK_COMP_PAR_AMDAHL);
-
- return task;
-}
-
-Task* Task::create_comm_par_mxn_1d_block(const std::string& name, double amount, void* userdata)
-{
- auto task = create(name, amount, userdata);
- task->set_kind(SD_TASK_COMM_PAR_MXN_1D_BLOCK);
-
- return task;
-}
-
-void Task::distribute_comp_amdahl(unsigned long count)
-{
- xbt_assert(kind_ == SD_TASK_COMP_PAR_AMDAHL,
- "Task %s is not a SD_TASK_COMP_PAR_AMDAHL typed task."
- "Cannot use this function.",
- get_cname());
- flops_amount_ = xbt_new0(double, count);
- for (unsigned long i = 0; i < count; i++)
- flops_amount_[i] = (alpha_ + (1 - alpha_) / count) * amount_;
-}
-
-void Task::build_MxN_1D_block_matrix(unsigned long src_nb, unsigned long dst_nb)
-{
- xbt_assert(kind_ == SD_TASK_COMM_PAR_MXN_1D_BLOCK,
- "Task %s is not a SD_TASK_COMM_PAR_MXN_1D_BLOCK typed task."
- "Cannot use this function.",
- get_cname());
- xbt_free(bytes_amount_);
- bytes_amount_ = xbt_new0(double, allocation_->size() * allocation_->size());
-
- for (unsigned long i = 0; i < src_nb; i++) {
- double src_start = i * amount_ / src_nb;
- double src_end = src_start + amount_ / src_nb;
- for (unsigned long j = 0; j < dst_nb; j++) {
- double dst_start = j * amount_ / dst_nb;
- double dst_end = dst_start + amount_ / dst_nb;
- XBT_VERB("(%lu->%lu): (%.2f, %.2f)-> (%.2f, %.2f)", i, j, src_start, src_end, dst_start, dst_end);
- bytes_amount_[i * (src_nb + dst_nb) + src_nb + j] = 0.0;
- if ((src_end > dst_start) && (dst_end > src_start)) { /* There is something to send */
- bytes_amount_[i * (src_nb + dst_nb) + src_nb + j] = std::min(src_end, dst_end) - std::max(src_start, dst_start);
- XBT_VERB("==> %.2f", bytes_amount_[i * (src_nb + dst_nb) + src_nb + j]);
- }
- }
- }
-}
-
-void Task::dependency_add(Task* task)
-{
- if (this == task)
- throw std::invalid_argument(
- simgrid::xbt::string_printf("Cannot add a dependency between task '%s' and itself", get_cname()));
-
- if (state_ == SD_DONE || state_ == SD_FAILED)
- throw std::invalid_argument(simgrid::xbt::string_printf(
- "Task '%s' must be SD_NOT_SCHEDULED, SD_SCHEDULABLE, SD_SCHEDULED, SD_RUNNABLE, or SD_RUNNING", get_cname()));
-
- if (task->get_state() == SD_DONE || task->get_state() == SD_FAILED || task->get_state() == SD_RUNNING)
- throw std::invalid_argument(simgrid::xbt::string_printf(
- "Task '%s' must be SD_NOT_SCHEDULED, SD_SCHEDULABLE, SD_SCHEDULED, or SD_RUNNABLE", task->get_cname()));
-
- if (dependency_exist(task))
- throw std::invalid_argument(simgrid::xbt::string_printf(
- "A dependency already exists between task '%s' and task '%s'", get_cname(), task->get_cname()));
-
- successors_.push_back(task);
- task->dependencies_.insert({this});
-
- /* if 'task' was runnable, it goes back to the SD_SCHEDULED state because of the new dependency*/
- if (task->get_state() == SD_RUNNABLE) {
- XBT_DEBUG("SD_task_dependency_add: %s was runnable and becomes scheduled!", task->get_cname());
- task->set_state(SD_SCHEDULED);
- }
-}
-
-bool Task::dependency_exist(Task* task) const
-{
- return (std::find(successors_.begin(), successors_.end(), task) != successors_.end() ||
- dependencies_.find(task) != dependencies_.end());
-}
-
-void Task::dependency_remove(Task* task)
-{
- if (this == task)
- throw std::invalid_argument("Cannot ask to remove itself from successors");
-
- auto p = std::find(successors_.begin(), successors_.end(), task);
- if (p != successors_.end()) {
- successors_.erase(p);
- task->dependencies_.erase({this});
- } else
- throw std::invalid_argument(simgrid::xbt::string_printf(
- "No dependency found between task '%s' and '%s': task '%s' is not a successor of task '%s'", get_cname(),
- task->get_cname(), task->get_cname(), get_cname()));
-
- /* if 'task' was scheduled and dependencies are satisfied, we can make it runnable */
- if (task->has_unsolved_dependencies() == 0 && task->get_state() == SD_SCHEDULED)
- task->set_state(SD_RUNNABLE);
-}
-
-std::set<Task*> Task::get_predecessors() const
-{
- std::set<Task*> res;
- for (const auto& d : dependencies_)
- if (d->get_kind() == SD_TASK_COMP_SEQ || d->get_kind() == SD_TASK_COMP_PAR_AMDAHL)
- res.insert(d);
- return res;
-}
-
-std::set<Task*> Task::get_inputs() const
-{
- std::set<Task*> res;
- for (const auto& d : dependencies_)
- if (d->get_kind() == SD_TASK_COMM_E2E || d->get_kind() == SD_TASK_COMM_PAR_MXN_1D_BLOCK)
- res.insert(d);
- return res;
-}
-
-std::vector<Task*> Task::get_outputs() const
-{
- std::vector<Task*> res;
- for (const auto& d : successors_)
- if (d->get_kind() == SD_TASK_COMM_E2E || d->get_kind() == SD_TASK_COMM_PAR_MXN_1D_BLOCK)
- res.push_back(d);
- return res;
-}
-
-void Task::set_amount(double amount)
-{
- amount_ = amount;
- if (kind_ == SD_TASK_COMP_SEQ)
- flops_amount_[0] = amount;
- if (kind_ == SD_TASK_COMM_E2E) {
- bytes_amount_[2] = amount;
- }
-}
-
-void Task::set_rate(double rate)
-{
- xbt_assert(kind_ == SD_TASK_COMM_E2E, "The rate can be modified for end-to-end communications only.");
- if (state_ < SD_RUNNING) {
- rate_ = rate;
- } else {
- XBT_WARN("Task %p has started. Changing rate is ineffective.", this);
- }
-}
-
-void Task::set_state(e_SD_task_state_t new_state)
-{
- std::set<Task*>::iterator idx;
- XBT_DEBUG("Set state of '%s' to %d", get_cname(), new_state);
- if ((new_state == SD_NOT_SCHEDULED || new_state == SD_SCHEDULABLE) && state_ == SD_FAILED) {
- sd_global->completed_tasks.erase(this);
- sd_global->initial_tasks.insert(this);
- }
-
- if (new_state == SD_SCHEDULED && state_ == SD_RUNNABLE) {
- sd_global->initial_tasks.insert(this);
- sd_global->runnable_tasks.erase(this);
- }
-
- if (new_state == SD_RUNNABLE) {
- idx = sd_global->initial_tasks.find(this);
- if (idx != sd_global->initial_tasks.end()) {
- sd_global->runnable_tasks.insert(*idx);
- sd_global->initial_tasks.erase(idx);
- }
- }
-
- if (new_state == SD_RUNNING)
- sd_global->runnable_tasks.erase(this);
-
- if (new_state == SD_DONE || new_state == SD_FAILED) {
- sd_global->completed_tasks.insert(this);
- start_time_ = surf_action_->get_start_time();
- if (new_state == SD_DONE) {
- finish_time_ = surf_action_->get_finish_time();
-#if SIMGRID_HAVE_JEDULE
- jedule_log_sd_event(this);
-#endif
- } else
- finish_time_ = simgrid_get_clock();
- surf_action_->unref();
- surf_action_ = nullptr;
- allocation_->clear();
- }
-
- state_ = new_state;
-
- if (watch_points_ & new_state) {
- XBT_VERB("Watch point reached with task '%s'!", get_cname());
- sd_global->watch_point_reached = true;
- unwatch(new_state); /* remove the watch point */
- }
-}
-
-double Task::get_remaining_amount() const
-{
- if (surf_action_)
- return surf_action_->get_remains();
- else
- return (state_ == SD_DONE) ? 0 : amount_;
-}
-
-double Task::get_start_time() const
-{
- return surf_action_ ? surf_action_->get_start_time() : start_time_;
-}
-
-double Task::get_finish_time() const
-{
- if (surf_action_) /* should never happen as actions are destroyed right after their completion */
- return surf_action_->get_finish_time();
- else
- return finish_time_;
-}
-
-void Task::set_sender_side_allocation(unsigned long count, const std::vector<s4u::Host*>* sender)
-{
- for (unsigned long i = 0; i < count; i++)
- allocation_->push_back(sender->at(i));
-}
-
-void Task::set_receiver_side_allocation(unsigned long count, const std::vector<s4u::Host*>* receiver)
-{
- for (unsigned long i = 0; i < count; i++)
- allocation_->insert(allocation_->begin() + i, receiver->at(i));
-}
-
-void Task::watch(e_SD_task_state_t state)
-{
- if (state & SD_NOT_SCHEDULED)
- throw std::invalid_argument("Cannot add a watch point for state SD_NOT_SCHEDULED");
-
- watch_points_ = watch_points_ | state;
-}
-
-void Task::unwatch(e_SD_task_state_t state)
-{
- xbt_assert(state != SD_NOT_SCHEDULED, "SimDag error: Cannot have a watch point for state SD_NOT_SCHEDULED");
- watch_points_ = watch_points_ & ~state;
-}
-
-void Task::dump() const
-{
- XBT_INFO("Displaying task %s", get_cname());
- if (state_ == SD_RUNNABLE)
- XBT_INFO(" - state: runnable");
- else if (state_ < SD_RUNNABLE)
- XBT_INFO(" - state: %s not runnable", __get_state_name(state_));
- else
- XBT_INFO(" - state: not runnable %s", __get_state_name(state_));
-
- if (kind_ != 0) {
- switch (kind_) {
- case SD_TASK_COMM_E2E:
- XBT_INFO(" - kind: end-to-end communication");
- break;
- case SD_TASK_COMP_SEQ:
- XBT_INFO(" - kind: sequential computation");
- break;
- case SD_TASK_COMP_PAR_AMDAHL:
- XBT_INFO(" - kind: parallel computation following Amdahl's law");
- break;
- case SD_TASK_COMM_PAR_MXN_1D_BLOCK:
- XBT_INFO(" - kind: MxN data redistribution assuming 1D block distribution");
- break;
- default:
- XBT_INFO(" - (unknown kind %d)", kind_);
- }
- }
-
- XBT_INFO(" - amount: %.0f", amount_);
- if (kind_ == SD_TASK_COMP_PAR_AMDAHL)
- XBT_INFO(" - alpha: %.2f", alpha_);
- XBT_INFO(" - Dependencies to satisfy: %lu", has_unsolved_dependencies());
- if (has_unsolved_dependencies() > 0) {
- XBT_INFO(" - pre-dependencies:");
- for (auto const& it : dependencies_)
- XBT_INFO(" %s", it->get_cname());
- }
- if (is_waited_by() > 0) {
- XBT_INFO(" - post-dependencies:");
-
- for (auto const& it : successors_)
- XBT_INFO(" %s", it->get_cname());
- }
-}
-
-void Task::released_by(Task* pred)
-{
- dependencies_.erase(pred);
- XBT_DEBUG("Release dependency on %s: %lu remain(s). Becomes schedulable if %zu=0", get_cname(),
- has_unsolved_dependencies(), get_predecessors().size());
-
- if (state_ == SD_NOT_SCHEDULED && get_predecessors().empty())
- set_state(SD_SCHEDULABLE);
-
- if (state_ == SD_SCHEDULED && has_unsolved_dependencies() == 0)
- set_state(SD_RUNNABLE);
-
- if (state_ == SD_RUNNABLE && not sd_global->watch_point_reached)
- run();
-}
-
-void Task::produced_by(Task* pred)
-{
- if (state_ == SD_RUNNABLE)
- return;
-
- start_time_ = pred->get_finish_time();
- dependencies_.erase(pred);
- if (state_ == SD_SCHEDULED)
- set_state(SD_RUNNABLE);
- else
- set_state(SD_SCHEDULABLE);
-
- Task* comm_dst = *(successors_.begin());
- if (comm_dst->get_state() == SD_NOT_SCHEDULED && comm_dst->get_predecessors().empty()) {
- XBT_DEBUG("%s is a transfer, %s may be ready now if %zu=0", get_cname(), comm_dst->get_cname(),
- comm_dst->get_predecessors().size());
- comm_dst->set_state(SD_SCHEDULABLE);
- }
- if (state_ == SD_RUNNABLE && not sd_global->watch_point_reached)
- run();
-}
-
-void Task::do_schedule()
-{
- if (state_ > SD_SCHEDULABLE)
- throw std::invalid_argument(simgrid::xbt::string_printf("Task '%s' has already been scheduled", get_cname()));
-
- if (has_unsolved_dependencies() == 0)
- set_state(SD_RUNNABLE);
- else
- set_state(SD_SCHEDULED);
-}
-
-void Task::schedule(const std::vector<s4u::Host*>& hosts, const double* flops_amount, const double* bytes_amount,
- double rate)
-{
- unsigned long host_count = hosts.size();
- rate_ = rate;
-
- if (flops_amount) {
- flops_amount_ = static_cast<double*>(xbt_realloc(flops_amount_, sizeof(double) * host_count));
- memcpy(flops_amount_, flops_amount, sizeof(double) * host_count);
- } else {
- xbt_free(flops_amount_);
- flops_amount_ = nullptr;
- }
-
- unsigned long communication_nb = host_count * host_count;
- if (bytes_amount) {
- bytes_amount_ = static_cast<double*>(xbt_realloc(bytes_amount_, sizeof(double) * communication_nb));
- memcpy(bytes_amount_, bytes_amount, sizeof(double) * communication_nb);
- } else {
- xbt_free(bytes_amount_);
- bytes_amount_ = nullptr;
- }
-
- for (const auto& h : hosts)
- allocation_->push_back(h);
-
- do_schedule();
-}
-
-void Task::schedulev(const std::vector<s4u::Host*>& hosts)
-{
- xbt_assert(kind_ == SD_TASK_COMP_SEQ || kind_ == SD_TASK_COMP_PAR_AMDAHL,
- "Task %s is not typed. Cannot automatically schedule it.", get_cname());
-
- for (unsigned long i = 0; i < hosts.size(); i++)
- allocation_->push_back(hosts[i]);
-
- XBT_VERB("Schedule computation task %s on %zu host(s)", get_cname(), allocation_->size());
-
- if (kind_ == SD_TASK_COMP_SEQ) {
- if (not flops_amount_) { /*This task has failed and is rescheduled. Reset the flops_amount*/
- flops_amount_ = xbt_new0(double, 1);
- flops_amount_[0] = amount_;
- }
- XBT_VERB("It costs %.f flops", flops_amount_[0]);
- }
-
- if (kind_ == SD_TASK_COMP_PAR_AMDAHL) {
- distribute_comp_amdahl(hosts.size());
- XBT_VERB("%.f flops will be distributed following Amdahl's Law", flops_amount_[0]);
- }
-
- do_schedule();
-
- /* Iterate over all inputs and outputs to say where I am located (and start them if runnable) */
- for (auto const& input : get_inputs()) {
- unsigned long src_nb = input->get_allocation_size();
- unsigned long dst_nb = hosts.size();
- if (src_nb == 0)
- XBT_VERB("Sender side of '%s' not scheduled. Set receiver side to '%s''s allocation", input->get_cname(),
- get_cname());
- input->set_sender_side_allocation(dst_nb, allocation_);
-
- if (input->get_allocation_size() > allocation_->size()) {
- if (kind_ == SD_TASK_COMP_PAR_AMDAHL)
- input->build_MxN_1D_block_matrix(src_nb, dst_nb);
-
- input->do_schedule();
- XBT_VERB("Auto-Schedule Communication task '%s'. Send %.f bytes from %lu hosts to %lu hosts.", input->get_cname(),
- input->get_amount(), src_nb, dst_nb);
- }
- }
-
- for (auto const& output : get_outputs()) {
- unsigned long src_nb = hosts.size();
- unsigned long dst_nb = output->get_allocation_size();
- if (dst_nb == 0)
- XBT_VERB("Receiver side of '%s' not scheduled. Set sender side to '%s''s allocation", output->get_cname(),
- get_cname());
- output->set_receiver_side_allocation(src_nb, allocation_);
-
- if (output->get_allocation_size() > allocation_->size()) {
- if (kind_ == SD_TASK_COMP_PAR_AMDAHL)
- output->build_MxN_1D_block_matrix(src_nb, dst_nb);
-
- output->do_schedule();
- XBT_VERB("Auto-Schedule Communication task %s. Send %.f bytes from %lu hosts to %lu hosts.", output->get_cname(),
- output->get_amount(), src_nb, dst_nb);
- }
- }
-}
-
-void Task::unschedule()
-{
- if (state_ == SD_NOT_SCHEDULED || state_ == SD_SCHEDULABLE)
- throw std::invalid_argument(xbt::string_printf(
- "Task %s: the state must be SD_SCHEDULED, SD_RUNNABLE, SD_RUNNING or SD_FAILED", get_cname()));
-
- if (state_ == SD_SCHEDULED || state_ == SD_RUNNABLE) /* if the task is scheduled or runnable */ {
- allocation_->clear();
- if (kind_ == SD_TASK_COMP_PAR_AMDAHL || kind_ == SD_TASK_COMM_PAR_MXN_1D_BLOCK) {
- /* Don't free scheduling data for typed tasks */
- xbt_free(flops_amount_);
- xbt_free(bytes_amount_);
- bytes_amount_ = nullptr;
- flops_amount_ = nullptr;
- }
- }
-
- if (state_ == SD_RUNNING)
- /* the task should become SD_FAILED */
- surf_action_->cancel();
- else {
- if (has_unsolved_dependencies() == 0)
- set_state(SD_SCHEDULABLE);
- else
- set_state(SD_NOT_SCHEDULED);
- }
- start_time_ = -1.0;
-}
-
-void Task::run()
-{
- xbt_assert(state_ == SD_RUNNABLE, "Task '%s' is not runnable! Task state: %d", get_cname(), (int)state_);
- xbt_assert(not allocation_->empty(), "Task '%s': host_list is empty!", get_cname());
-
- XBT_VERB("Executing task '%s'", get_cname());
-
- /* Beware! The scheduling data are now used by the surf action directly! no copy was done */
- auto host_model = allocation_->front()->get_netpoint()->get_englobing_zone()->get_host_model();
- surf_action_ = host_model->execute_parallel(*allocation_, flops_amount_, bytes_amount_, rate_);
-
- surf_action_->set_data(this);
-
- XBT_DEBUG("surf_action = %p", surf_action_);
-
- set_state(SD_RUNNING);
- sd_global->return_set.insert(this);
-}
-
-void Task::destroy()
-{
- XBT_DEBUG("Destroying task %s...", get_cname());
-
- /* First Remove all dependencies associated with the task. */
- while (not dependencies_.empty())
- (*(dependencies_.begin()))->dependency_remove(this);
- while (not successors_.empty())
- this->dependency_remove(successors_.front());
-
- if (state_ == SD_SCHEDULED || state_ == SD_RUNNABLE) {
- xbt_free(flops_amount_);
- xbt_free(bytes_amount_);
- bytes_amount_ = nullptr;
- flops_amount_ = nullptr;
- }
-
- xbt_free(flops_amount_);
- xbt_free(bytes_amount_);
-
- delete allocation_;
-
- if (surf_action_ != nullptr)
- surf_action_->unref();
-
- XBT_DEBUG("Task destroyed.");
- delete this;
-}
-} // namespace sd
-} // namespace simgrid
-
-/* **************************** Public C interface *************************** */
-
-/**
- * @brief Creates a new task.
- *
- * @param name the name of the task (can be @c nullptr)
- * @param data the user data you want to associate with the task (can be @c nullptr)
- * @param amount amount of the task
- * @return the new task
- * @see SD_task_destroy()
- */
-SD_task_t SD_task_create(const char* name, void* data, double amount)
-{
- return simgrid::sd::Task::create(name, amount, data);
-}
-
-/** @brief create an end-to-end communication task that can then be auto-scheduled
- *
- * Auto-scheduling mean that the task can be used with SD_task_schedulev(). This allows one to specify the task costs at
- * creation, and decouple them from the scheduling process where you just specify which resource should deliver the
- * mandatory power.
- *
- * A end-to-end communication must be scheduled on 2 hosts, and the amount specified at creation is sent from hosts[0]
- * to hosts[1].
- */
-SD_task_t SD_task_create_comm_e2e(const char* name, void* data, double amount)
-{
- return simgrid::sd::Task::create_comm_e2e(name, amount, data);
-}
-
-/** @brief create a sequential computation task that can then be auto-scheduled
- *
- * Auto-scheduling mean that the task can be used with SD_task_schedulev(). This allows one to specify the task costs at
- * creation, and decouple them from the scheduling process where you just specify which resource should deliver the
- * mandatory power.
- *
- * A sequential computation must be scheduled on 1 host, and the amount specified at creation to be run on hosts[0].
- *
- * @param name the name of the task (can be @c nullptr)
- * @param data the user data you want to associate with the task (can be @c nullptr)
- * @param flops_amount amount of compute work to be done by the task
- * @return the new SD_TASK_COMP_SEQ typed task
- */
-SD_task_t SD_task_create_comp_seq(const char* name, void* data, double flops_amount)
-{
- return simgrid::sd::Task::create_comp_seq(name, flops_amount, data);
-}
-
-/** @brief create a parallel computation task that can then be auto-scheduled
- *
- * Auto-scheduling mean that the task can be used with SD_task_schedulev(). This allows one to specify the task costs at
- * creation, and decouple them from the scheduling process where you just specify which resource should deliver the
- * mandatory power.
- *
- * A parallel computation can be scheduled on any number of host.
- * The underlying speedup model is Amdahl's law.
- * To be auto-scheduled, @see SD_task_distribute_comp_amdahl has to be called first.
- * @param name the name of the task (can be @c nullptr)
- * @param data the user data you want to associate with the task (can be @c nullptr)
- * @param flops_amount amount of compute work to be done by the task
- * @param alpha purely serial fraction of the work to be done (in [0.;1.[)
- * @return the new task
- */
-SD_task_t SD_task_create_comp_par_amdahl(const char* name, void* data, double flops_amount, double alpha)
-{
- return simgrid::sd::Task::create_comp_par_amdahl(name, flops_amount, data, alpha);
-}
-
-/** @brief create a complex data redistribution task that can then be auto-scheduled
- *
- * Auto-scheduling mean that the task can be used with SD_task_schedulev().
- * This allows one to specify the task costs at creation, and decouple them from the scheduling process where you just
- * specify which resource should communicate.
- *
- * A data redistribution can be scheduled on any number of host.
- * The assumed distribution is a 1D block distribution. Each host owns the same share of the @see amount.
- * To be auto-scheduled, @see SD_task_distribute_comm_mxn_1d_block has to be called first.
- * @param name the name of the task (can be @c nullptr)
- * @param data the user data you want to associate with the task (can be @c nullptr)
- * @param amount amount of data to redistribute by the task
- * @return the new task
- */
-SD_task_t SD_task_create_comm_par_mxn_1d_block(const char* name, void* data, double amount)
-{
- return simgrid::sd::Task::create_comm_par_mxn_1d_block(name, amount, data);
-}
-
-/**
- * @brief Destroys a task.
- *
- * The user data (if any) should have been destroyed first.
- *
- * @param task the task you want to destroy
- * @see SD_task_create()
- */
-void SD_task_destroy(SD_task_t task)
-{
- task->destroy();
-}
-
-/** @brief Returns the user data of a task */
-void* SD_task_get_data(const_SD_task_t task)
-{
- return task->get_data();
-}
-
-/** @brief Sets the user data of a task
- * The new data can be @c nullptr. The old data should have been freed first, if it was not @c nullptr.
- */
-void SD_task_set_data(SD_task_t task, void* data)
-{
- task->set_data(data);
-}
-
-void SD_task_set_rate(SD_task_t task, double rate)
-{
- task->set_rate(rate);
-}
-
-/**
- * @brief Returns the state of a task
- *
- * @param task a task
- * @return the current @ref e_SD_task_state_t "state" of this task:
- * #SD_NOT_SCHEDULED, #SD_SCHEDULED, #SD_RUNNABLE, #SD_RUNNING, #SD_DONE or #SD_FAILED
- * @see e_SD_task_state_t
- */
-e_SD_task_state_t SD_task_get_state(const_SD_task_t task)
-{
- return task->get_state();
-}
-
-const char* SD_task_get_name(const_SD_task_t task)
-{
- return task->get_cname();
-}
-
-void SD_task_set_name(SD_task_t task, const char* name)
-{
- task->set_name(name);
-}
-
-/** @brief Returns the parents of a task ina dynar */
-xbt_dynar_t SD_task_get_parents(const_SD_task_t task)
-{
- xbt_dynar_t parents = xbt_dynar_new(sizeof(SD_task_t), nullptr);
-
- for (auto const& it : task->get_dependencies())
- xbt_dynar_push(parents, &it);
-
- return parents;
-}
-
-/** @brief Returns the children of a task in a dynar */
-xbt_dynar_t SD_task_get_children(const_SD_task_t task)
-{
- xbt_dynar_t children = xbt_dynar_new(sizeof(SD_task_t), nullptr);
-
- for (auto const& it : task->get_successors())
- xbt_dynar_push(children, &it);
-
- return children;
-}
-
-double SD_task_get_start_time(const_SD_task_t task)
-{
- return task->get_start_time();
-}
-
-double SD_task_get_finish_time(const_SD_task_t task)
-{
- return task->get_finish_time();
-}
-
-void SD_task_distribute_comp_amdahl(SD_task_t task, int count)
-{
- task->distribute_comp_amdahl(count);
-}
-
-void SD_task_build_MxN_1D_block_matrix(SD_task_t task, int src_nb, int dst_nb)
-{
- task->build_MxN_1D_block_matrix(src_nb, dst_nb);
-}
-
-/**
- * @brief Returns the number of workstations involved in a task
- *
- * Only call this on already scheduled tasks!
- * @param task a task
- */
-int SD_task_get_workstation_count(const_SD_task_t task)
-{
- return static_cast<int>(task->get_allocation_size());
-}
-
-/**
- * @brief Returns the list of workstations involved in a task
- *
- * Only call this on already scheduled tasks!
- * @param task a task
- */
-sg_host_t* SD_task_get_workstation_list(const_SD_task_t task)
-{
- return task->get_allocation()->data();
-}
-
-/**
- * @brief Returns the total amount of work contained in a task
- *
- * @param task a task
- * @return the total amount of work (computation or data transfer) for this task
- * @see SD_task_get_remaining_amount()
- */
-double SD_task_get_amount(const_SD_task_t task)
-{
- return task->get_amount();
-}
-
-void SD_task_set_amount(SD_task_t task, double amount)
-{
- task->set_amount(amount);
-}
-
-double SD_task_get_remaining_amount(const_SD_task_t task)
-{
- return task->get_remaining_amount();
-}
-
-e_SD_task_kind_t SD_task_get_kind(const_SD_task_t task)
-{
- return task->get_kind();
-}
-
-void SD_task_dump(const_SD_task_t task)
-{
- task->dump();
-}
-
-void SD_task_dependency_add(SD_task_t src, SD_task_t dst)
-{
- XBT_DEBUG("SD_task_dependency_add: src = %s, dst = %s", src->get_cname(), dst->get_cname());
- src->dependency_add(dst);
-}
-void SD_task_dependency_remove(SD_task_t src, SD_task_t dst)
-{
- XBT_DEBUG("SD_task_dependency_remove: src = %s, dst = %s", src->get_cname(), dst->get_cname());
- src->dependency_remove(dst);
-}
-
-/**
- * @brief Indicates whether there is a dependency between two tasks.
- * If src is nullptr, checks whether dst has any pre-dependency.
- * If dst is nullptr, checks whether src has any post-dependency.
- */
-int SD_task_dependency_exists(const_SD_task_t src, SD_task_t dst)
-{
- xbt_assert(src != nullptr || dst != nullptr, "Invalid parameter: both src and dst are nullptr");
-
- if (src)
- if (dst)
- return src->dependency_exist(dst);
- else
- return static_cast<int>(src->is_waited_by());
- else
- return static_cast<int>(dst->has_unsolved_dependencies());
-}
-
-void SD_task_watch(SD_task_t task, e_SD_task_state_t state)
-{
- task->watch(state);
-}
-
-void SD_task_unwatch(SD_task_t task, e_SD_task_state_t state)
-{
- task->unwatch(state);
-}
-
-/** @brief Dumps the task in dotty formalism into the FILE* passed as second argument */
-void SD_task_dotty(const_SD_task_t task, void* out)
-{
- auto* fout = static_cast<FILE*>(out);
- fprintf(fout, " T%p [label=\"%.20s\"", task, task->get_cname());
- switch (task->get_kind()) {
- case SD_TASK_COMM_E2E:
- case SD_TASK_COMM_PAR_MXN_1D_BLOCK:
- fprintf(fout, ", shape=box");
- break;
- case SD_TASK_COMP_SEQ:
- case SD_TASK_COMP_PAR_AMDAHL:
- fprintf(fout, ", shape=circle");
- break;
- default:
- xbt_die("Unknown task type!");
- }
- fprintf(fout, "];\n");
- for (auto const& it : task->get_dependencies())
- fprintf(fout, " T%p -> T%p;\n", it, task);
-}
-
-/**
- * @brief Returns an approximative estimation of the execution time of a task.
- *
- * The estimation is very approximative because the value returned is the time the task would take if it was executed
- * now and if it was the only task.
- *
- * @param host_count number of hosts on which the task would be executed
- * @param host_list the hosts on which the task would be executed
- * @param flops_amount computation amount for each host(i.e., an array of host_count doubles)
- * @param bytes_amount communication amount between each pair of hosts (i.e., a matrix of host_count*host_count doubles)
- * @see SD_schedule()
- */
-double SD_task_get_execution_time(const_SD_task_t /*task*/, int host_count, const sg_host_t* host_list,
- const double* flops_amount, const double* bytes_amount)
-{
- xbt_assert(host_count > 0, "Invalid parameter");
- double max_time = 0.0;
-
- /* the task execution time is the maximum execution time of the parallel tasks */
- for (int i = 0; i < host_count; i++) {
- double time = 0.0;
- if (flops_amount != nullptr)
- time = flops_amount[i] / host_list[i]->get_speed();
-
- if (bytes_amount != nullptr)
- for (int j = 0; j < host_count; j++)
- if (bytes_amount[i * host_count + j] != 0)
- time += (sg_host_get_route_latency(host_list[i], host_list[j]) +
- bytes_amount[i * host_count + j] / sg_host_get_route_bandwidth(host_list[i], host_list[j]));
-
- if (time > max_time)
- max_time = time;
- }
- return max_time;
-}
-
-/**
- * @brief Schedules a task
- *
- * The task state must be #SD_NOT_SCHEDULED.
- * Once scheduled, a task is executed as soon as possible in @see SD_simulate, i.e. when its dependencies are satisfied.
- *
- * @param task the task you want to schedule
- * @param host_count number of hosts on which the task will be executed
- * @param host_list the hosts on which the task will be executed
- * @param flops_amount computation amount for each hosts (i.e., an array of host_count doubles)
- * @param bytes_amount communication amount between each pair of hosts (i.e., a matrix of host_count*host_count doubles)
- * @param rate task execution speed rate
- * @see SD_task_unschedule()
- */
-void SD_task_schedule(SD_task_t task, int host_count, const sg_host_t* host_list, const double* flops_amount,
- const double* bytes_amount, double rate)
-{
- xbt_assert(host_count > 0, "host_count must be positive");
- std::vector<sg_host_t> hosts(host_count);
-
- for (int i = 0; i < host_count; i++)
- hosts[i] = host_list[i];
-
- task->schedule(hosts, flops_amount, bytes_amount, rate);
-}
-
-void SD_task_unschedule(SD_task_t task)
-{
- task->unschedule();
-}
-
-/** @brief Auto-schedules a task.
- *
- * Auto-scheduling mean that the task can be used with SD_task_schedulev(). This allows one to specify the task costs at
- * creation, and decouple them from the scheduling process where you just specify which resource should deliver the
- * mandatory power.
- *
- * To be auto-schedulable, a task must be a typed computation SD_TASK_COMP_SEQ or SD_TASK_COMP_PAR_AMDAHL.
- */
-void SD_task_schedulev(SD_task_t task, int count, const sg_host_t* host_list)
-{
- std::vector<sg_host_t> list(count);
- for (int i = 0; i < count; i++)
- list[i] = host_list[i];
- task->schedulev(list);
-}
-
-/** @brief autoschedule a task on a list of hosts
- *
- * This function is similar to SD_task_schedulev(), but takes the list of hosts to schedule onto as separate parameters.
- * It builds a proper vector of hosts and then call SD_task_schedulev()
- */
-void SD_task_schedulel(SD_task_t task, int count, ...)
-{
- va_list ap;
- std::vector<sg_host_t> list(count);
- va_start(ap, count);
- for (int i = 0; i < count; i++)
- list[i] = va_arg(ap, sg_host_t);
-
- va_end(ap);
- task->schedulev(list);
-}
