This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: K6 patches for review
- To: "David S. Miller" <davem at dm dot cobaltmicro dot com>
- Subject: Re: K6 patches for review
- From: Jeffrey A Law <law at cygnus dot com>
- Date: Sat, 22 Aug 1998 18:23:29 -0600
- cc: jfc at tiac dot net, egcs at cygnus dot com, egcs-patches at cygnus dot com
- Reply-To: law at cygnus dot com
In message <199808221758.KAA06341@dm.cobaltmicro.com>you write:
> From: John Carr <jfc@tiac.net>
> Date: Sat, 22 Aug 1998 09:46:34 EDT
>
> Please comment.
>
> This patch adds AMD K6 optimizations, based on AMD publication 21924B
> for the K6-2. I don't know if the K6 pipeline is significantly
> different. I have not run benchmarks yet.
>
> You will get more profitable results if you use a slotting scheme like
> I did on the UltraSparc using the MD_SCHED macros. Haifa is really
> dumb. Haifa never got quad issue on Ultra before my hacks, I simply
> couldn't describe the chip using the current define_function_unit
> scheme, now for FPU code Haifa gets quad issue often. ADJUST_COST is
> yet another hack that doesn't work at all...
I'd argue with "doesn't work at all...". It's value as diminished
over time as superscalar machines have become popular.
MD_SCHED is the way to go right now. There's serveral things I want
to describe for the PA using MD_SCHED when I get the chance. Presumably
as we get a better feel for the kinds of problems everyone uses MD_SCHED
to solve we'll have a better clue what kinds of things we need to do
in the generic scheduler :-)
jeff