You need to run gentoo.
I got a good website for you: http://funroll-loops.org/
On May 31, 2005, at 7:49 AM, Markus Kolb wrote:
Peter Hessler wrote on Mon, May 30, 2005 at 14:39:12 -0700:
Because gcc generates broken code for -march=i486. The above link
*is*
an informative answer, just not what you were looking for.
Is it a bug in gcc version implemented in OpenBSD or why I've never
seen
this in any other architecture optimized compilations?
What problem are you trying to solve with -march=i386 vs -march=i386?
Here's a hint: it doesn't matter.
Well, to say it doesn't matter is a bit ignorance. I am pretty sure
that
an optimised compilation is measureable and feelable faster on
restricted hardware.
Mostly it works well but the adaptec aha isa driver breaks the loading
when optimised compiled.
By the way parts of obsd kernel is compiled for i486 by default and
future GCC releases will compile i486 code without any compiler
options
by default.
For example SUSE compiled their older SUSE Linux for i486, later for
i586 without problems. I have no problems with other BSDs.
I don't talk about any speed optimising O flags and so on. It is
simple
processor instruction optimised code.
The ser port is compiled by default with a really stupid march option
for athlon:
gcc -fPIC -DPIC -O9 -funroll-loops -Wcast-align -Wall
-minline-all-stringops -malign-double -falign-loops -march=athlon
-fno-stack-protector -DNAME='"ser"' -DVERSION='"0.8.10"' -
DARCH='"i386"'
-DOS='"openbsd"' -DCOMPILER='"gcc 3.3"' -D __CPU_i386
-DCFG_DIR='"/etc/"' -DPKG_MALLOC -DSHM_MEM -DSHM_MMAP -
DADAPTIVE_WAI T
-DADAPTIVE_WAIT_LOOPS=1024 -DDNS_IP_HACK -DUSE_IPV6 -DDBG_QM_MALLOC
-DFAST_LO CK -DHAVE_GETHOSTBYNAME2 -DHAVE_UNION_SEMUN
-DHAVE_SOCKADDR_SA_LEN -DDLSYM_PREFI X='"_"' -I/include -c xjab_jcon.c
-o xjab_jcon.o
So it has to be just luck that it works everywhere.
Simple to say if something doesn't work, it is the fault of anyoneelse
isn't good style.
So maybe there are a few people who knows about and want to have a
technical talk.
Markus
