[Openmp-commits] [PATCH] D13827: Avoid bad conversion for __kmp_sys_max_nth

Dimitry Andric via Openmp-commits openmp-commits at lists.llvm.org
Mon Oct 19 03:47:05 PDT 2015

dim added a comment.

In http://reviews.llvm.org/D13827#269992, @jcownie wrote:

> If you choose this approach, 32Ki seems too small. A current maxed-out SGI is already at 256 sockets which could get you to 256*18*2 = 9216 threads, so over 1/4 of the way to your 32Ki.

This review is not about that limit, but about the case where PTHREAD_THREADS_MAX does not fit into an int, which ranges from -2^31 .. 2^31-1 on almost all architectures these days.

That said, I'm not sure where the default 32Ki value came from originally.  If some system does not define PTHREAD_THREADS_MAX, it seems pretty safe to assume a not too high limit?  Or maybe even just `#error` out with a message similar to "please define PTHREAD_THREADS_MAX for your system".


More information about the Openmp-commits mailing list