Re: 3.0-pre vs 3.0 on Haney speed

[Paolo Carlini <pcarlini at unitus dot it>]

Hi Joern,

> Could it be related to this patch?
>
> 2001-06-09  Mark Mitchell  <mark@codesourcery.com>
>
>         * toplev.c (rest_of_compilation): Remove dead code before
>         purge_addressof.

Thanks for your feedback! I would really like to understand more of this
strange story...
Indeed, on 2001-06-09 Mitchell applied that nice patch, which for some
days, until its revertion about a week later due to a latent bug, just
before the release that is, improved greatly the Stepanov test and
*perhaps* also that complex matrix test in the Haney speed suite.
On the other hand, as I should have said, I had consistentently obtained
in the last 3-4 months results similar to those corresponding to the
20010611 pre, that is much better than those corresponding to the
released 3.0 (or current 3.1 snapshots), well before that patch going in
and out... However, I must admit, I did'nt run the test frequently in
the last month before the release, and may be possible that the patch
temporarily masked a pessimization already present since a few weeks...
I'm planning to revert from 20010611 pre only that patch and see what
happen.

What else you suggest may be worth trying ???

Thanks,
Paolo.