How to create an `omp parallel for` with synchronization (`barrier`) of all threads in the middle with OpenMP

Question

Hello,

I have two fonction, do_step_one(i) and do_step_two(i) for i from 0 to N

Currently, I have this (sequential) code ;

for(unsigned int i=0; i<N; i++){
     do_step_one(i);
}

for(unsigned int i=0; i<N; i++){
     do_step_two(i);
}

for any step (one or two), each case of N can be done in any order and in parallel, but any step_two need the end of all the step_one to start (it use step_one results).

I tried the following :

#omp parallel for
for(unsigned int i=0; i<N; i++){
    do_step_one(i);

#omp barrier

    do_step_two(i);
}

But gcc complains

convolve_slices.c:21: warning: barrier region may not be closely nested inside of work-sharing, critical, ordered, master or explicit task region.

What do I missunderstand ? How to solve that issue ?

Thanks.

Answer 1

One problem I see with this code, is that the code does not comply with the spec :)

If you need all do_step_one()'s to end, you'll need something like the following:

#pragma omp parallel for
for(unsigned int i=0; i<N; i++){
     do_step_one(i);
}

#pragma omp parallel for
for(unsigned int i=0; i<N; i++){
     do_step_two(i);
}

The result of this would be a parallelism of the first for, and then a parallelism of the second for.