Optimization

> On 4/12/07, prdcomp <prdcompuol.com.br> wrote:
> >
><snip>... the following flags: "-O2 -pipe
-march=athlon-xp". Those are >rather common options,
but the number of unexpected (and quite expected by >now)
libmudflap failures raised from the traditional 6 to 320!
> 
> That doesn't seem right. Those are perfectly normal
flags, except
> maybe -march.

It seems -march can indeed bring some trouble... I've seen
it somewhere in the lists...

> In fact, gcc by default uses -O2. Are you pretty
> confident that you built everything the same way?

Pretty sure! :(

Actually, built it in two athlon-xp at the same time (&
same flags) with very similar results: in one, 320
unexpected failures in libmudflap; in the other, 314
unexpected failures in libmudflap... In both using
lfs-livecd-6.2-5.

>I've never heard of
> mudflap totally falling over like that, but there are
issues with it
> timing out some tests. Do the errors say that it's
timing out?

It seems (please correct me if I'm wrong) that it would be
"normal" for libmudflap to have 6 unexpected
failures... But I got 320. I'm gonna try something different
now: I'll go "-march=686 -O3 -mmmx -3dnow -msse
-mfpmath=sse -fomit-frame-pointer -pipe" and check the
results.

Regarding the timing out, I've got one of the log files
opened in front of me now. Will reproduce part of it - it
indeed mentions "program timed out", but don't
know what caused it, its relation with the failures etc:

=== libmudflap Summary ===

# of expected passes              974
# of unexpected failures          314
make[4]: *** [check-DEJAGNU] Error 1
make[4]: Leaving directory
`/sources/gcc-build/i686-pc-linux-gnu/libmudflap/testsuite`
make[3]: *** [check-am] Error 2
make[3]: Target 'check' not remade because of errors.
WARNING: program timed out.
FAIL: libmudflap.cth/pass39-frag.c (-O2) execution test
FAIL: libmudflap.cth/pass39-frag.c (-O2) output pattern
test
WARNING: program timed out.
FAIL: libmudflap.cth/pass39-frag.c (-O2) (rerun 1) execution
test
FAIL: libmudflap.cth/pass39-frag.c (-O2) (rerun 1) output
pattern test
WARNING: program timed out.
...<MANY SIMILAR MESSAGES>...
FAIL: libmudflap.cth/pass37-frag.c (-O2) (rerun 19)
execution test
FAIL: libmudflap.cth/pass37-frag.c (-O2) (rerun 19) output
pattern test
WARNING: program timed out.
...<MANY SIMILAR MESSAGES>...
FAIL: libmudflap.cth/pass40-frag.c (-O3) execution test
FAIL: libmudflap.cth/pass40-frag.c (-O3) output pattern
test

=== libmudflap Summary ===

# of expected passes              974
# of unexpected failures          314

And later it does show the same messages again.

Any tip on what may be happening?

> > Actually, I think the real question is: is it
worth trying to optimize >a LFS (initial) system? For
BLFS, I tend to believe it is. For LFS itself, >not that
sure.
> 
><snip>
> 
> 
> Just shooting from the hip here. I've never gone crazy
optimizing or
> done any kind of benchmarking.

Thanks for the help!

Paulo


-- 
http://linuxfromscratch.org/mailman/listinfo/lfs-support

FAQ: http://w
ww.linuxfromscratch.org/lfs/faq.html
Unsubscribe: See the above information page

On 4/13/07, prdcomp <prdcompuol.com.br> wrote:
>
> Regarding the timing out, I've got one of the log files
opened in front of me now. Will reproduce part of it - it
indeed mentions "program timed out", but don't
know what caused it, its relation with the failures etc:
>
> === libmudflap Summary ===
>
> # of expected passes              974
> # of unexpected failures          314
> make[4]: *** [check-DEJAGNU] Error 1
> make[4]: Leaving directory
`/sources/gcc-build/i686-pc-linux-gnu/libmudflap/testsuite`
> make[3]: *** [check-am] Error 2
> make[3]: Target 'check' not remade because of errors.
> WARNING: program timed out.
> FAIL: libmudflap.cth/pass39-frag.c (-O2) execution
test
> FAIL: libmudflap.cth/pass39-frag.c (-O2) output pattern
test
> WARNING: program timed out.

Yep, they're timing out. Wow, that's a spectacular amount of
failing.
Do this sed before you start the build to give the test
programs some
more time before they quit.

sed -i.bak '/dg-timeout/s/3|10/20/'
libmudflap/testsuite/libmudflap.cth/*.c

That's from DIY Linux. We should probably do the same here.
That
should help get the number of failures down. And yes, 6 is
the normal
number of failures.

--
Dan
-- 
http://linuxfromscratch.org/mailman/listinfo/lfs-support

FAQ: http://w
ww.linuxfromscratch.org/lfs/faq.html
Unsubscribe: See the above information page