[lno] performance comparison to mainline

[Richard Guenther]

Hi!

The first complete run of the automatic POOMA/tramp3d tester completed
tonight and the results are encouraging!

Bootstrap times for lno are slightly larger than for mainline (54min vs.
50min) and also tramp3d compile-times regress from 3.8min to 4.7min
(with leafify) and from 2.6min to 2.9min (without leafify).  But runtime
gets a big boost for leafified mode from 4.2s/it to 2.7s/it!  That's
a 50% improvement!  (runtime for non-leafified mode didn't change - I
guess loop optimizers don't deal well with calls inside their loops)

Bootstrapping is with checking disabled and the runtime tests were
performed with -O2 -funroll-loops -ffast-math.  All on a ia64 host.

Any suggestions on what would be even better options for lno to test?

Btw., the leafify patch is still unreviewed (though I pinged two times
already). http://gcc.gnu.org/ml/gcc-patches/2004-05/msg00187.html

Thanks,
Richard.

--
Richard Guenther <richard dot guenther at uni-tuebingen dot de>
WWW: http://www.tat.physik.uni-tuebingen.de/~rguenth/