This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: -mtune=generic for i386 backend
- From: Jan Hubicka <hubicka at ucw dot cz>
- To: Andi Kleen <ak at suse dot de>
- Cc: Jan Hubicka <jh at suse dot cz>, gcc-patches at gcc dot gnu dot org
- Date: Wed, 18 Jan 2006 08:46:26 +0100
- Subject: Re: -mtune=generic for i386 backend
- References: <20060117232507.GK26136@kam.mff.cuni.cz.suse.lists.egcs-patches> <p731wz61b9k.fsf@verdi.suse.de>
>
> 20k what? Bigger code size? If yes measured how? Did you do any
Yes, it is 20k bigger code. It is measured by building SPECint
with -mtune=generic versus -mtune=i686/k8 (in 32bit/64bit mode).
This is somewhat unforutnate fact I want to play with a bit as followup,
but it naturaly comes with fact that our instruction choice is limited
to common ground of the two CPUs and K8 alignment+move based passing
conventions are themselves introducing pretty noticeable hit.
I didn't included relative growth, since I used statically linked
binaries (so I test proper fortran library as well).
Just for quick idea,
from other runs I get the dynamic sizes of SPEC binaries 5MB (with
debugging), so it translates as sub 1% (modulo debugging).
> further measurements on how code size is impact from this?
Not yet, but I hope to get into it.
Honza
>
> Thanks,
> -Andi (looking forward to using this switch in the future)