openMP for loop Segmentation fault (core dumped)

Question

I recently started learning openMP and I am trying to parallelize my code for convolution. Once I added #pragma to the for loop which is initializing the image array and there's no data dependency, the code broke and threw Segmentation fault (core dumped). I couldn't figure out what's wrong. Please help!

 // map values from original image to padded image #pragma omp parallel for schedule(static) for (size_t j = 0; j < n * n; j++) { size_t row = (j / n) + padding; size_t col = (j % n) + padding; size_t pos = (n + (padding * 2)) * row + col; padded_image[pos] = image[j]; }

Where is output allocated / declared? Adding (or removing) code, and all of a sudden things break for no apparent reason, has all the earmarks of memory corruption. — PaulMcKenzie
– PaulMcKenzie, Commented Mar 16, 2020 at 20:04
You should provide the inputs for the function. It seems like your segfault happens because you are accessing memory which does not belong to you. — shargors
– shargors, Commented Mar 16, 2020 at 20:11
output is allocated in main() in another file. the code runs in sequential version but breaks in parallel version — lookingglass
– lookingglass, Commented Mar 16, 2020 at 20:40

Jérôme Richard · Accepted Answer · 2020-03-16 20:40:25Z

The computed indices does not match with the allocated size. Indeed, the size of the allocated array padded_image does not depend on n while the accesses to the same in the targeted loop does. Regarding the loop, the size of the array should probably be: (n + (padding * 2)) * (n + (padding * 2)).

Please note that padded_image is not deleted. Moreover, modulus are very slow, please consider using two loops with an OpenMP clause collapse(2).

Collectives™ on Stack Overflow

openMP for loop Segmentation fault (core dumped)

1 Answer 1

Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Related