Open MPI logo

Hardware Locality Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Hardware Locality Development mailing list

Subject: [hwloc-devel] dynamic cpuset_t?
From: Samuel Thibault (samuel.thibault_at_[hidden])
Date: 2009-09-23 17:51:30


One thing we weren't so sure about while designing libtopology was
whether to use fixed-size cpuset_t or dynamic-size cpuset_t. The
advantage of the former is simplicity of course. The currently defined
size is 1024 cpus and can easily be configured if one needs more. We
could think that 1024 is plenty for now and we'll see later. But if
later we realize that we have to change the API in order to be able to
use dynamic-size cpuset_t (e.g. copy & destroy functions at least), that
will be a burden. Also, dynamic-size cpuset_t is actually more efficient
for small boxes for most operations, as the bitmask will be smaller.

While we are at libtopology->hwloc API transition and now that a few
people are subscribed to hwloc-devel, what do people think about it?
Maybe we should at least make sure that the API doesn't hinder potential
for dynamic support?