This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [patch] enabling vectorization by default at -O3
On 9/6/07, Zdenek Dvorak <rakdver@kam.mff.cuni.cz> wrote:
> Hello,
>
> > On 9/6/07, H.J. Lu <hjl@lucon.org> wrote:
> >
> > > I have been using this patch for several months on Linux/x86-64,
> > > Linux/ia64 and Linux/ia32 without any regressions. OK to install?
> >
> > I don't know. Zdenek, is this a fix you would like to propose or was
> > it just a quick hack around the issue? I don't see any indication in
> > the PR log.
>
> IIRC, I ment it seriously (possibly with a small change -- limiting it
> just to reassoc2).
You did, and i believe it is a valid fix :)
Reassociation moves things into loops sometimes. The patch takes into
account loops and avoids reassociation across loop boundaries.
The only other change than limiting to reassoc2 i think makes sense
would be to cause it to only not reassociate when the operation is
from a shallower loop than the existing ops (IE don't reassociate
depth 1 loops ops into depth 2 loops, but reassociating depth 2 loop
ops into depth 1 loops should be fine).