Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2005-08-15 10:45:58


Is anyone having problems compiling the "linux" paffinity component?

It turns out that there are *3* different implementations of the Linux
system call sched_setaffinity() (and 3 different corresponding
prototypes -- #@$@#$%!!!). The differences seem to be loosely
categorized as:

1. implementation that was back-ported to 2.4 kernels (the one that is
listed in sched_setaffinity(2))
2. implementation in early 2.6 kernels
3. implementation in later 2.6 kernels

(although there seem to be some exceptions to the above categorization)

Plus, at least some 2.4 kernels seem to have broken aspects. Or is it
glibc? Who knows, who cares: in some places, the CPU_ZERO() macro
won't compile.

The default code path is for case #3. I have fixes for all cases ready
to commit, but they involve changes to that component's configure.m4,
which means making everyone re-run autogen.sh.

Unless someone is actively running into compilation problems now, I was
planning until waiting until tonight to commit (observing the "try not
to cause people to re-run autogen.sh during the US work day" rule).

-- 
{+} Jeff Squyres
{+} The Open MPI Project
{+} http://www.open-mpi.org/