Barrier

Tags: comp-arch

Is a synchronisation primitive for threads. A barriers blocks the execution of one or more threads until all the threads reach the barrier. You can imagine this as a guard that prevents any thread that reaches it from continuing until all the other threads catch up.

The barrier is given a count when its initialised for the maximum number of threads it can block and then is called by each thread to block execution at a given point. You can think of this process as a reverse semaphore. Instead of blocking when the count drops to 0, you block until the count drops to 0.

#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <pthread.h>
#include <time.h>

#define THREAD_COUNT 4

pthread_barrier_t mybarrier;

void* thread_function(void *id_ptr) {
  int thread_id = *(int*)id_ptr;
  int wait_sec = 1 + rand() % 5;
  printf("thread %d: Wait for %d seconds.\n", thread_id, wait_sec);
  sleep(wait_sec);
  printf("thread %d: I'm ready...\n", thread_id);

  pthread_barrier_wait(&mybarrier);

  printf("thread %d: going!\n", thread_id);
  return NULL;
}


int main() {
  int i;
  pthread_t ids[THREAD_COUNT];
  int short_ids[THREAD_COUNT];

  srand(time(NULL));
  pthread_barrier_init(&mybarrier, NULL, THREAD_COUNT + 1);

  for (i=0; i < THREAD_COUNT; i++) {
    short_ids[i] = i;
    pthread_create(&ids[i], NULL, thread_function, &short_ids[i]);
  }

  printf("main() is ready.\n");

  pthread_barrier_wait(&mybarrier);

  printf("main() is going!\n");

  for (i=0; i < THREAD_COUNT; i++) {
    pthread_join(ids[i], NULL);
  }

  pthread_barrier_destroy(&mybarrier);

  return 0;
}

Code Snippet 1: An example application using the pthread barrier implementation.