Subject: Re: Yet another optimization vs. assembler problem
To: Martin Husemann <martin@rumolt.teuto.de>
From: Sean Doran <smd@ebone.net>
List: current-users
Date: 09/09/1999 00:08:24
mk.conf:
USE_EGCS=1
USE_EGCS_HAIFA=1
So far i386 -current kernel, userland and X are happy with:
for kernel:
-O3 -mpentium --mfancy-math-387 -fsched-interblock \
-fsched-spec -fsched-spec-load -fbranch-count-reg
... -mpentiumpro ...
for userland, same but:
-march=i686
-march=pentium
with the caveat that the "static"s have to be excised from
lib/csu/i386_elf/crt0.c and the bootblocks have to be
recompiled with less optimization (no optimization at all
works fine).
-current on an Alpha box is happy with:
-O3 -mcpu=21064 -mmemory-latency=main -fsched-interblock \
-fsched-spec -fsched-spec-load -fbranch-count-reg
for both the kernel and userland.
Incidentally, feeding "A>>HR" into malloc(3) seems to make my
boxes a little snappier feeling. I have no quantitative
measurements, though, so it could just be my imagination. -:)
(I haven't hit anything that's triggered the "A" feature
yet, except for a small script that set a resource limit too low...)
Sean.