https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108336
--- Comment #2 from Gareth Davies <gareth.davies.ga.code at gmail dot com> --- (In reply to Andrew Pinski from comment #1) > Most likely there is only one seed per program so the random # generator is > using locks. So the order is dependent on which thread calls the function. > > Now having a seed per thread might be a good idea but it does have other > kind of issues. Even initializing it and even Making sure it does not > reinitialized each time a new openmp thread is launched. There is some documentation as to how it's supposed to work here: https://gcc.gnu.org/onlinedocs/gfortran/RANDOM_005fSEED.html It sounds like we should be able to get repeated random numbers - even if their mapping to the threads does vary. Indeed, the results are pretty close to that.
