This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH 4/4] Add make autoprofiledbootstrap
- From: Andi Kleen <ak at linux dot intel dot com>
- To: Bernd Schmidt <bschmidt at redhat dot com>
- Cc: Andi Kleen <andi at firstfloor dot org>, gcc-patches at gcc dot gnu dot org
- Date: Wed, 27 Apr 2016 09:27:24 -0700
- Subject: Re: [PATCH 4/4] Add make autoprofiledbootstrap
- Authentication-results: sourceware.org; auth=none
- References: <1459140266-6902-1-git-send-email-andi at firstfloor dot org> <1459140266-6902-2-git-send-email-andi at firstfloor dot org> <1459140266-6902-3-git-send-email-andi at firstfloor dot org> <1459140266-6902-4-git-send-email-andi at firstfloor dot org> <5720DC69 dot 60402 at redhat dot com>
On Wed, Apr 27, 2016 at 05:36:09PM +0200, Bernd Schmidt wrote:
> On 03/28/2016 06:44 AM, Andi Kleen wrote:
> >From: Andi Kleen <ak@linux.intel.com>
> >
> >Add support for profiledbootstrap with autofdo. Will be useful
> >to get better testing coverage of autofdo.
>
> Is this the only purpose? I'll admit this is the patch I like least out of
> the series.
It's the main purpose, since we already have profiled feedback.
bootstrap is one of the best tests we have and autofdo badly needs it.
It's also a useful benchmark, so if autofdo gets improved it can catch up more
with explicit profiling.
> >The autofdo'ed compiler is ~7% faster on insn-recog.i (vs ~11% for
> >profiledfeedback), and ~4% faster for tramp3d-v4 (vs 10% for
> >profiledfeedback)
>
> So it seems like we get worse results than with a feature we already have,
> so I don't quite see the value.
I hope as it gets tuned that will improve. Also autofdo is so much
easier to use on other software that the main usage elsewhere may well be autofdo,
not explicit instrumentation.
-Andi
--
ak@linux.intel.com -- Speaking for myself only