Linux Thread synchronization------barriers

A barrier is a synchronization mechanism that coordinates the parallel work of multiple threads in Linux, allowing each thread to wait until all the cooperating threads reach a point and then continue to execute from that point, Pthread_join is a barrier but allows only one thread to wait, Pthread_ Barrier allows any number of threads to wait!

Pthread_barrier_init (pthread_barrier_t * barrier, attribute, unsigned int barrier count value);

Pthread_barrier_wait (pthread_barrier_t * barrier): Call the Count value in each thread and block the current thread here until the count value is set at the initialization value;

Thread Barrier usage Examples:

1, if a 8 million of the array to sort, in a thread to complete the need for a long time, if the use of a thread barrier, 8 threads to the 1 million number of heap sorting, and then in the main thread of the merge merge operation, time-consuming will save 6 times times!

2, the following example is the initialization of an array of 8 million elements, respectively, using 1, 4, 8 threads to complete, found that the use of 8 threads is the shortest time.

/* Program function: Create a dynamic array, assign an initial value to the array, create a num_thread thread, initialize an array of length size, pass (thread barrier synchronization) and print the entire array value when all threads have finished processing. Compares the time spent by a time count.  When using 8 threads, sort took 0.0958 seconds when four threads are used thread-1249961072 done job.thread-1260450928 do job.thread-1239471216 done job.thread-1270944880 done Job.sort took 0.1104 seconds when taking a thread thread-1239983216 done Job.sort took 0.2567 seconds*/#in Clude <pthread.h> #include <stdlib.h> #include <stdio.h> #define SIZE 8000000L//Array length # define Num_ Thread 8//number of threads # define SIZE_PER (Size/num_thread)//The array length to be handled by each thread pthread_barrier_t barrier;//define barrier int *a;/* thread handler for each thread * /void * Thr_fun (void *arg) {Long n = (long) Arg;long i;for (i=n;i<n+size_per;i++) {a[i] = i;} printf ("Thread%d done job.\n", pthread_self ());p thread_barrier_wait (&barrier); return ((void *) 1);} int main () {pthread_t tid;struct timeval start,end;long long startusec,endusec;double elapsed;int i;a = (int *) malloc (SIZE  *sizeof (int)); Dynamically allocated array pthread_barrier_init (&AMP;BARRIER,NULL,NUM_THREAD+1);//Initialize the thread barrier count to the number of child threads plus the main thread gettimeofday (&start,null);//Get Start time for (i=0;i<num_thread;i++) {pthread_create (&tid,null,thr_fun, (void *) (I*size_per) );//Create Child thread}pthread_barrier_wait (&barrier);//wait for all child threads to finish Gettimeofday (&end,null);//Get End time for (i=0;i<  size;i++)//print array contents//printf ("%d", A[i]); startusec = start.tv_sec * 1000000 + start.tv_usec;endusec = end.tv_sec * 1000000 + end.tv_usec;elapsed = (double) (endusec-startusec)/1000000.0;//the time spent calculating processing printf ("Sort took%.4f seconds\n", elapsed); return 0;}

Linux Thread synchronization------barriers