This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Pretty good gcc-3.1 on sparc32-linux
- From: "Ulrich Weigand" <Ulrich dot Weigand at de dot ibm dot com>
- To: hp at bitrange dot com, christian at j-son dot org
- Cc: gcc at gcc dot gnu dot org
- Date: Thu, 18 Apr 2002 21:56:59 +0200
- Subject: Re: Pretty good gcc-3.1 on sparc32-linux
- Sensitivity:
Hans-Peter Nilsson wrote:
>On Thu, 18 Apr 2002, Christian Jönsson wrote:
>
>> On Thu, Apr 18, 2002 at 10:52:54AM -0700, Richard Henderson wrote:
>> > On Thu, Apr 18, 2002 at 06:29:43PM +0200, Christian Jönsson wrote:
>> > > Now, what could be lurking behind the time outs I get in the gcc test
>> > > 20001226-1? Optimisation perhaps takes too long time to do or somehow
>> > > gets stuck...
>> >
>> > Takes too long in delayed branch scheduling, IIRC.
>>
>> Well, that could of course very well be the case. Do you suggest that
>> I try something special to find something out for all of us or should
>> I just drop the issue?
>
>It's not just dbr, though that might contribute. I see it off
>and on. Your mileage depends on load and speed of the build
>machine. For example, I last saw it cross-compiling to mmix,
>mn10300 and sh, and only the last one has delay-slots. It's
>mentioned from time to time; see archives.
I've profiled it on s390 once, and it appears that >90% of its
run time are spent in the #ifdef ENABLE_CHECKING section at the
beginning of subst_reloads ...
You could try just disabling these checks and see whether it
makes a difference.
As it didn't seem particularly important, I haven't tracked it
down any further (and nowadays for some reason the test doesn't
timeout anymore on s390 anyway).
Mit freundlichen Gruessen / Best Regards
Ulrich Weigand
--
Dr. Ulrich Weigand
Linux for S/390 Design & Development
IBM Deutschland Entwicklung GmbH, Schoenaicher Str. 220, 71032 Boeblingen
Phone: +49-7031/16-3727 --- Email: Ulrich.Weigand@de.ibm.com