------- Comment #2 from rakdver at gcc dot gnu dot org 2007-07-01 13:28 -------
(In reply to comment #1)
> On the mainline, we have ivtmp.33 going from 1 to 101 so it does not equal the
> same as what i is.
... and ivopts do this transformation because of my favourite hack -- md of
i386 claims that more complex addressing modes are cheaper, so ivopts prefer to
introduce unnecessary offsets. I guess it is time to check again whether
removing this nonsense still causes performance regressions.
--
rakdver at gcc dot gnu dot org changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Ever Confirmed|0 |1
Last reconfirmed|0000-00-00 00:00:00 |2007-07-01 13:28:49
date| |
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32445
