OpenMP Segfault | 易学教程

问题

I am trying to add OpenMP parallelization to a working code (just to a single for loop), however I cannot get rid of a segmentation fault. The problem arises from this line:

pos += sprintf(com + pos, "%d ", i);

com is a character array, and I tried defining it as char com[255] or char *com = malloc(255*sizeof(char)), both inside and before the for loop. I added private(com) to #pragma omp parallel for directive when I defined com before the loop. I also tried initializing it and using firstprivate. (pos is an integer, initialized to 0)

When I do not add -fopenmp everything works fine, but with -fopenmp it gives segfault. What am I missing?

回答1:

The segmentation fault comes from multiple threads updating the value of pos at the same time, therefore setting it to some value that turns com + pos into a pointer that points beyond or before the allocated memory for com. The proper way to parallelise such a loop would be to concatenate the values in private strings and then concatenate the private strings in an ordered fashion:

char com[255];
int pos = 0;

#pragma omp parallel
{
   char mycom[255];
   int mypos = 0;

   #pragma omp for schedule(static) nowait
   for (int i = 0; i < N; i++)
      mypos += sprintf(mycom + mypos, "%d ", i);

   // Concatenate the strings in an ordered fashion
   #pragma omp for schedule(static) ordered
   for (int i = 0; i < omp_get_num_threads(); i++)
   {
      #pragma omp ordered
      pos += sprintf(com + pos, "%s", mycom);
   }
}

The ordered construct ensures proper synchronisation so one does not need critical. The use of schedule(static) is important in order to guarantee each thread processes a single contiguous section of the iteration space.

来源：https://stackoverflow.com/questions/29247593/openmp-segfault

标签

char

malloc

openmp