Question

I am trying to use a zip iterator (from either boost or thrust library) with openmp. My goal is to have a for_each loop that will process things in parallel through openmp.

Using the example from the boost documentation:

#pragma omp parallel for
std::for_each(
    boost::make_zip_iterator(
    boost::make_tuple(beg1, beg2)
),
    boost::make_zip_iterator(
    boost::make_tuple(end1, end2)
),
    zip_func()
);

However, this doesn’t appear to run in parallel.

Suggestions?

Answer 1

What you have to do is either use omp parallel for followed by a a for loop, use gcc __gnu_parallel::for_each as was previously suggested or explicitly partition your iteration range (notice the absence of for in the pragma):

#pragma omp parallel
{
int tid = omp_get_thread_num();
int n   = omp_get_num_threads();
auto nbeg1, nbeg2, nend1, nend2  = ... // create n chunks of work, each thread takes one chunk
std::for_each(
  boost::make_zip_iterator(boost::make_tuple(nbeg1, nbeg2)),
  boost::make_zip_iterator(boost::make_tuple(nend1, nend2)),
  zip_func() );
}