This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: branch predictions
- To: Mark Mitchell <mark at codesourcery dot com>
- Subject: Re: branch predictions
- From: Jason Eckhardt <jle at cygnus dot com>
- Date: Thu, 4 May 2000 08:00:08 -0700 (PDT)
- cc: law at cygnus dot com, gcc at gcc dot gnu dot org
On Thu, 4 May 2000, Mark Mitchell wrote:
> >>>>> "Jason" == Jason Eckhardt <jle@cygnus.com> writes:
>
> Jason> this is something that just fell through the cracks. The
> Jason> Ball paper discusses heuristics similar to #1,#2.
>
> I withdraw my nervousness about this if their is good academic work
> supporting the idea. Thanks for referencing this!
>
Actually, your concerns are warranted. These are still purely heuristic,
static predictions that could be wrong for a given program. Ball & Larus
did get reasonable results empirically over their benchmarks, but
many times real profiling feedback is better. This is why bb-reorder can
work transparently with either. If profiling feedback is available, it
will be used. Otherwise, we fallback to static predictions.
Jason.