This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: SPEC regression in perl
- From: Wolfgang Bangerth <bangerth at ices dot utexas dot edu>
- To: Jeffrey A Law <law at redhat dot com>
- Cc: Andrew Pinski <pinskia at physics dot uc dot edu>, gcc List <gcc at gcc dot gnu dot org>, mmitchel at gcc dot gnu dot org
- Date: Tue, 17 Aug 2004 09:24:24 -0500
- Subject: Re: SPEC regression in perl
Jeff,
> Perl is notoriously flakey -- to the point of running the same binary
> with the same inputs varying by 10% without rhyme or reason. In fact,
> I've found it's the most useless of all the SPEC tests because of the
> inability to get consistent results out of the bloody thing.
>
> I wouldn't mind looking at a specific problem if you can find one, but
> I'm not likely to go on what is likely a wild goose chase inside perl.
While your claim about variability is clearly supported by the charts in the
tester, I think you can't decline looking at the problem that easily.
Averaged over about 10 runs, the results for base tests in perlbmk are about
20% worse starting at the date Andrew mentioned, and the drop is about four
times as large as the variability.
It would be nice if you could spend a little more time explaining why you
think that you need not look into the problem pointed out.
Thanks
Wolfgang
-------------------------------------------------------------------------
Wolfgang Bangerth email: bangerth@ices.utexas.edu
www: http://www.ices.utexas.edu/~bangerth/