[SunRescue] Code Optimization -- was RE: GCC code generation sucks....
Greg A. Woods
rescue at sunhelp.org
Mon Apr 30 02:28:24 CDT 2001
[ On Sunday, April 29, 2001 at 22:13:56 (-0700), Chris Byrne wrote: ]
> Subject: [SunRescue] Code Optimization -- was RE: GCC code generation sucks....
>
> Why? Because no good work has gone into optimizing GCC for new x86 platforms
> since they wrote the 586/686 extensions. Hell 90% of the x86 code out there
> is still compiled with 386 or 486 optimization.
Not to mention that some folks are even still fixing VAX code generation
in GCC too! This just in from NetBSD's source-changes list:
Module Name: gnusrc
Committed By: matt
Date: Mon Apr 30 06:19:59 UTC 2001
Modified Files:
gnusrc/gnu/dist/toolchain/gcc/config/vax: vax.c vax.h vax.md
Log Message:
Fix two code generation problems. The first is to fix how negative const
int values are added in DI (if the the constant is <0 & > -2*1024*1024 then
use a decl/subl and sbwc $0 instead of addl/adwc). The second fix is to
disallow register elimination peepholes when that register is used by
other operands in the other instructions in the peephole.
--
Greg A. Woods
+1 416 218-0098 VE3TCP <gwoods at acm.org> <woods at robohack.ca>
Planix, Inc. <woods at planix.com>; Secrets of the Weird <woods at weird.com>
More information about the rescue
mailing list