possible major performance regression

[Prakash Punnoor]

Dear devs,

I and some other users who have been trying some glibc snapshots (see
http://forums.gentoo.org/viewtopic-t-435659-postdays-0-postorder-asc-start-25.html )
are experiencing major performance problems on x86 and x86_64 with a sat 
solver. I don't know whether we do something wrong (it is not vanilla 
snapshot, as it incorporates several patches) or it is something in glibc.

(It is not compiler dependant, but strangely enough with dynamically linked 
exes - if compiled on machine with stable glibc, the speed is ok on machine 
with snapshot glibc. Even if I compile the source to an object on the stable 
glibc machine, and link on the snapshot machine, speed is ok. But vice-versa, 
speed is slow. So a problem in the headers?)

Perhaps you just want to check yourself. I uploaded sources and properly 
working static binaries here: http://punnoor.de/various/boehm-sat.tar.bz2

Just run runme10.sh for a quick or runme.sh for a thourough test. The last 
lines of each run will show you cumulative run-times (so lower is better). In 
the runme10.sh test I get about 20 sec run-time with stable glibc and about 
30 secs with the snapshot.

I am not subscribed, so please cc me.

Thanks for your time,
-- 
(°=                 =°)
//\ Prakash Punnoor /\\
V_/                 \_V
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 191 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/libc-alpha/attachments/20060304/381eef7a/attachment.sig>