------- Additional Comments From law at redhat dot com 2004-12-23 17:22 -------
Subject: Re: [4.0 Regression] jump threading
on trees is slow with switch statements with large # of cases
On Thu, 2004-12-23 at 13:31 +0000, steven at gcc dot gnu dot org wrote:
> ------- Additional Comments From steven at gcc dot gnu dot org 2004-12-23
> 13:31 -------
> Jeff, as I said, you're attacking problems not related to the original
> problem reported in this PR. Maybe you don't use bugzilla often enough
> to see how annoying it is when the PR summary and the problem described
> in the bug report don't match. But oh well, whatever, if you can win
> 10% here, why am I complaining ;-)
The bug report really should have been called "compilation speed
regression" rather than calling out "jump threading" -- a goodly number
of the changes necessary to get 15524 to a point of almost acceptable
were not jump threading related.
How about this, open a new report with the same testcase and a generic
description about compile-time regression, then close 15524. Seems like
a make-work project, but hey, if it makes you happy....
> I'm very curious about your tree alias rewrite. Got a plan/overview of
> it somewhere?
In simplest terms, we lose the braindamaged type memory tag nonsense and
only do type based alias analysis on objects that have not been
disambiguated by other means. This as the effect of greatly reducing
number of type based aliasing checks -- which accounts for the vast
majority of the compilation time for 15855.
> Is it GCC 4.0 material?
Unsure, mostly because I will not have much time to work on it for the
next 3 weeks.
Jeff
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15524
