[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: pbuilder running :) [ppro trolls read this]



On Thu, Jan 23, 2003 at 11:23:44PM +0100, Eduard Bloch wrote:
> * Jesus Climent [Thu, Jan 23 2003, 09:27:31PM]:
> 
> > > I'm also running it with pentium-builder and compiling with pentium-pro
> > > optimizations. So to all the i686 trolls, give me a list of the packages
> > > that you want me to benchmark against the regular i386 ones. I'll
> > > run all benchmarks on an Athlon 650 with 512M once I finish compiling
> > > Sid. :)
> > 
> > Compiling with PPro optimisations and running the test under an AMD
> > might bring the opposite results than expected. Please, if you plan to
> > run some benchmarking related to i686 optims, use a i686.
> 
> Does not K7 fullfil the i686 specification? So you mean a real 
> Pentium >=Pro machine.

A K7 doesn't have the same performance profile. On the other hand, the
gcc support for i386 subarchs is generally very poor, and has been
wrong on numerous occasions before now, so it's entirely possible that
the optimal mode is not the one with the same name as the box you run
it on.

Any such benchmark is only valid for the specific combination of
compiler flags and hardware used. To get a *good* impression of what
effect compiler flags have, you really need to tabulate hardware
against flags, for all interesting combinations of flags that you can
think of (preferably selected based on some understanding of what they
do, rather than the usual randomness that you see from people doing
this sort of thing), and fill in the table.

To the best of my knowledge, nobody has ever attempted this on a
non-trivial scale. The process should probably be automated.

-- 
  .''`.  ** Debian GNU/Linux ** | Andrew Suffield
 : :' :  http://www.debian.org/ | Dept. of Computing,
 `. `'                          | Imperial College,
   `-             -><-          | London, UK

Attachment: pgpJjzktWv9ZY.pgp
Description: PGP signature


Reply to: