[parsec-users] Per thread affinity in each benchmark

avadh patel avadh4all at gmail.com
Fri Apr 13 10:24:42 EDT 2012


Hi,

I recently saw email that you are planning to release 3.0 version.
I have one patch that might help lot of people that uses Parsec with Marss
like simulators.
This patch adds 'per-thread-affinity' to each benchmark.  In current
release when
'hooks' is used, all threads are given a range of CPUs for affinity which
introduce lot of
run-by-run variation in simulation runs.

This patch try to fix each thread to a CPU instead of range of CPUs which
reduces
simulation run-by-run variation by eliminating thread-scheduling randomness.

To enable per thread affinity, add -DENABLE_THREAD_AFFINITY to CFLAGS and
CXXFLAGS in .bldconf files.  Also set 'PARSEC_CPU_NUM' environment variable
to number of CPUs you want to allocate for all threads of benchmark.

Thanks,
Avadh
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/parsec-users/attachments/20120413/63590e3d/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: parsec_affinity.patch
Type: application/octet-stream
Size: 19916 bytes
Desc: not available
URL: <http://lists.cs.princeton.edu/pipermail/parsec-users/attachments/20120413/63590e3d/attachment-0001.obj>


More information about the parsec-users mailing list