Paris7/work-stealing-scheduler
Archived
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

each structure now have his own mutex, finally taking advantage of the work stealing

Browse source
This commit is contained in:
Mylloon 2024-04-24 23:00:25 +02:00
parent c0b12bfa50
commit f190c2f034
Signed by: Anri
GPG key ID: A82D63DFF8D1317F
1 changed files with 126 additions and 121 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

247
src/sched-ws.c
View file

@ -4,21 +4,37 @@
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
/* Tâche */
struct task_info {
void *closure;
taskfunc f;
};
struct scheduler {
/* Structure de chaque thread */
struct worker {
/* Premier élément du deque (dernier ajouter) */
int *bottom;
int bottom;
/* Variable de conditions pour reveillé les threads au besoin */
/* Mutex qui protège cette structure */
pthread_mutex_t mutex;
/* Deque de tâches */
struct task_info *tasks;
/* Thread */
pthread_t thread;
/* Dernier élément du deque (premier ajouter) */
int top;
};
/* Scheduler partagé */
struct scheduler {
/* Condition threads dormant */
pthread_cond_t cond;
/* Mutex qui protège les piles */
/* Mutex qui protège cette structure */
pthread_mutex_t mutex;
/* Nombre de threads instanciés */
@ -27,17 +43,11 @@ struct scheduler {
/* Compteur des threads dormants */
int nthsleep;
/* Taille des piles */
/* Taille deque */
int qlen;
/* Piles de tâches */
struct task_info **tasks;
/* Liste des threads */
pthread_t *threads;
/* Dernier élément du deque (premier ajouter) */
int *top;
/* Liste de workers par threads */
struct worker *workers;
};
/* Lance une tâche de la pile */
@ -46,18 +56,16 @@ void *sched_worker(void *);
/* Nettoie les opérations effectuées par l'initialisation de l'ordonnanceur */
int sched_init_cleanup(struct scheduler, int);
/* Récupère l'index du thread courant */
/* Récupère l'index du thread courant
*
* Assume que le mutex de l'ordonnanceur est verrouillé */
int current_thread(struct scheduler *);
int
sched_init(int nthreads, int qlen, taskfunc f, void *closure)
{
static struct scheduler sched;
sched.bottom = NULL;
sched.tasks = NULL;
sched.threads = NULL;
sched.top = NULL;
sched.workers = NULL;
if(qlen <= 0) {
fprintf(stderr, "qlen must be greater than 0\n");
@ -73,7 +81,11 @@ sched_init(int nthreads, int qlen, taskfunc f, void *closure)
}
sched.nthreads = 0;
sched.nthsleep = 0;
// Initialisation variable de condition
if(pthread_cond_init(&sched.cond, NULL) != 0) {
fprintf(stderr, "Can't init condition variable\n");
return sched_init_cleanup(sched, -1);
}
// Initialisation du mutex
if(pthread_mutex_init(&sched.mutex, NULL) != 0) {
@ -81,34 +93,28 @@ sched_init(int nthreads, int qlen, taskfunc f, void *closure)
return sched_init_cleanup(sched, -1);
}
// Initialisation variable de condition
if(pthread_cond_init(&sched.cond, NULL) != 0) {
fprintf(stderr, "Can't init varcond\n");
return sched_init_cleanup(sched, -1);
}
sched.nthsleep = 0;
// Initialisation du curseur suivant l'état de la pile de chaque processus
if(!(sched.bottom = malloc(nthreads * sizeof(int)))) {
perror("Cursor bottom stack");
return sched_init_cleanup(sched, -1);
}
if(!(sched.top = malloc(nthreads * sizeof(int)))) {
perror("Cursor top stack");
return sched_init_cleanup(sched, -1);
// Initialize workers
if(!(sched.workers = malloc(nthreads * sizeof(struct worker)))) {
perror("Workers");
return -1;
}
for(int i = 0; i < nthreads; ++i) {
sched.bottom[i] = 0;
sched.top[i] = 0;
}
sched.workers[i].bottom = 0;
sched.workers[i].top = 0;
// Allocation mémoire pour la pile de chaque processus
if(!(sched.tasks = malloc(nthreads * sizeof(struct task_info *)))) {
perror("Deque list");
return sched_init_cleanup(sched, -1);
}
for(int i = 0; i < nthreads; ++i) {
if(!(sched.tasks[i] = malloc(qlen * sizeof(struct task_info)))) {
fprintf(stderr, "Deque for thread %d: %s\n", i, strerror(errno));
// Initialisation mutex
if(pthread_mutex_init(&sched.workers[i].mutex, NULL) != 0) {
fprintf(stderr, "Can't init mutex %d\n", i);
return sched_init_cleanup(sched, -1);
}
// Allocation mémoire deque
if(!(sched.workers[i].tasks =
malloc(sched.qlen * sizeof(struct task_info)))) {
fprintf(stderr, "Thread %d: ", i);
perror("Deque list");
return sched_init_cleanup(sched, -1);
}
}
@ -116,45 +122,44 @@ sched_init(int nthreads, int qlen, taskfunc f, void *closure)
// Initialise l'aléatoire
srand(time(NULL));
// Créer les threads
if(!(sched.threads = malloc(nthreads * sizeof(pthread_t)))) {
perror("Threads");
return sched_init_cleanup(sched, -1);
}
// Ajoute la tâche initiale
if(sched_spawn(f, closure, &sched) < 0) {
fprintf(stderr, "Can't create the initial task\n");
return sched_init_cleanup(sched, -1);
}
// Démarre les threads
// Création des threads
for(int i = 0; i < nthreads; ++i) {
if(pthread_create(&sched.threads[i], NULL, sched_worker, &sched) != 0) {
fprintf(stderr, "Can't create the thread %d\n", i);
pthread_mutex_lock(&sched.mutex);
if(pthread_create(&sched.workers[i].thread, NULL, sched_worker,
(void *)&sched) != 0) {
fprintf(stderr, "Can't create thread %d\n", i);
if(i > 0) {
fprintf(stderr, ", cancelling already created threads...\n");
for(int j = 0; j < i; ++j) {
if(pthread_cancel(sched.threads[j]) != 0) {
fprintf(stderr, "Can't cancel the thread %d\n", j);
}
}
} else {
fprintf(stderr, "\n");
// Annule les threads déjà créer
for(int j = 0; j < i; ++j) {
pthread_cancel(sched.workers[j].thread);
}
return sched_init_cleanup(sched, -1);
}
pthread_mutex_lock(&sched.mutex);
sched.nthreads++;
pthread_mutex_unlock(&sched.mutex);
}
// Ajoute la tâche initiale
if(sched_spawn(f, closure, &sched) < 0) {
fprintf(stderr, "Can't queue the initial task\n");
return sched_init_cleanup(sched, -1);
}
// Attend la fin des threads
for(int i = 0; i < nthreads; ++i) {
if((pthread_join(sched.threads[i], NULL) != 0)) {
if((pthread_join(sched.workers[i].thread, NULL) != 0)) {
fprintf(stderr, "Can't wait the thread %d\n", i);
// Quelque chose s'est mal passé, on annule les threads en cours
for(int j = 0; j < nthreads; ++j) {
if(j != i) {
pthread_cancel(sched.workers[j].thread);
}
}
return sched_init_cleanup(sched, -1);
}
}
@ -165,35 +170,20 @@ sched_init(int nthreads, int qlen, taskfunc f, void *closure)
int
sched_init_cleanup(struct scheduler s, int ret_code)
{
pthread_mutex_destroy(&s.mutex);
pthread_cond_destroy(&s.cond);
if(s.tasks) {
pthread_mutex_destroy(&s.mutex);
if(s.workers) {
for(int i = 0; i < s.nthreads; ++i) {
if(s.tasks[i]) {
free(s.tasks[i]);
s.tasks[i] = NULL;
}
pthread_mutex_destroy(&s.workers[i].mutex);
free(s.workers[i].tasks);
s.workers[i].tasks = NULL;
}
free(s.tasks);
s.tasks = NULL;
}
if(s.threads) {
free(s.threads);
s.threads = NULL;
}
if(s.bottom) {
free(s.bottom);
s.bottom = NULL;
}
if(s.top) {
free(s.top);
s.top = NULL;
free(s.workers);
s.workers = NULL;
}
return ret_code;
@ -204,14 +194,11 @@ current_thread(struct scheduler *s)
{
pthread_t current = pthread_self();
pthread_mutex_lock(&s->mutex);
for(int i = 0; i < s->nthreads; ++i) {
if(pthread_equal(s->threads[i], current)) {
pthread_mutex_unlock(&s->mutex);
if(pthread_equal(s->workers[i].thread, current)) {
return i;
}
}
pthread_mutex_unlock(&s->mutex);
return -1;
}
@ -220,24 +207,28 @@ int
sched_spawn(taskfunc f, void *closure, struct scheduler *s)
{
int th;
pthread_mutex_lock(&s->mutex);
if((th = current_thread(s)) < 0) {
th = 0;
}
pthread_mutex_unlock(&s->mutex);
pthread_mutex_lock(&s->mutex);
pthread_mutex_lock(&s->workers[th].mutex);
int next = (s->bottom[th] + 1) % s->qlen;
if(next == s->top[th]) {
pthread_mutex_unlock(&s->mutex);
int next = (s->workers[th].bottom + 1) % s->qlen;
if(next == s->workers[th].top) {
pthread_mutex_unlock(&s->workers[th].mutex);
fprintf(stderr, "Stack is full\n");
errno = EAGAIN;
return -1;
}
s->tasks[th][s->bottom[th]] = (struct task_info){closure, f};
s->bottom[th] = next;
s->workers[th].tasks[s->workers[th].bottom] =
(struct task_info){closure, f};
s->workers[th].bottom = next;
pthread_mutex_unlock(&s->mutex);
pthread_mutex_unlock(&s->workers[th].mutex);
return 0;
}
@ -249,55 +240,69 @@ sched_worker(void *arg)
// Récupère le processus courant (index tableau)
int curr_th;
pthread_mutex_lock(&s->mutex);
while((curr_th = current_thread(s)) < 0);
pthread_mutex_unlock(&s->mutex);
struct task_info task;
int found;
while(1) {
found = 0;
pthread_mutex_lock(&s->mutex);
pthread_mutex_lock(&s->workers[curr_th].mutex);
if(s->top[curr_th] != s->bottom[curr_th]) {
if(s->workers[curr_th].top != s->workers[curr_th].bottom) {
found = 1;
s->bottom[curr_th] = (s->bottom[curr_th] - 1 + s->qlen) % s->qlen;
task = s->tasks[curr_th][s->bottom[curr_th]];
s->workers[curr_th].bottom =
(s->workers[curr_th].bottom - 1 + s->qlen) % s->qlen;
task = s->workers[curr_th].tasks[s->workers[curr_th].bottom];
}
pthread_mutex_unlock(&s->workers[curr_th].mutex);
if(!found) {
// Vol car aucune tâche trouvée
for(int i = 0, k = rand() % (s->nthreads + 1), target;
i < s->nthreads; ++i) {
target = (i + k) % s->nthreads;
pthread_mutex_lock(&s->mutex);
int nthreads = s->nthreads;
pthread_mutex_unlock(&s->mutex);
if(s->top[target] != s->bottom[target]) {
for(int i = 0, k = rand() % (nthreads + 1), target; i < nthreads;
++i) {
target = (i + k) % nthreads;
pthread_mutex_lock(&s->workers[target].mutex);
if(s->workers[target].top != s->workers[target].bottom) {
// Tâche trouvée
found = 1;
s->bottom[target] =
(s->bottom[target] - 1 + s->qlen) % s->qlen;
task = s->tasks[target][s->bottom[target]];
s->workers[target].bottom =
(s->workers[target].bottom - 1 + s->qlen) % s->qlen;
task = s->workers[target].tasks[s->workers[target].bottom];
pthread_mutex_unlock(&s->workers[target].mutex);
break;
}
pthread_mutex_unlock(&s->workers[target].mutex);
}
// Aucune tâche à faire
if(!found) {
pthread_mutex_lock(&s->mutex);
s->nthsleep++;
// Ne partir que si tout le monde dort
if(s->nthsleep >= s->nthreads) {
pthread_cond_broadcast(&s->cond);
pthread_mutex_unlock(&s->mutex);
break;
}
pthread_cond_wait(&s->cond, &s->mutex);
s->nthsleep--;
pthread_mutex_unlock(&s->mutex);
continue;
}
}
pthread_cond_broadcast(&s->cond);
pthread_mutex_unlock(&s->mutex);
pthread_cond_signal(&s->cond);
// Exécute la tâche
task.f(task.closure, s);