On Tuesday 21 June 2005 18:08, Scott Robert Ladd wrote:
> I've noticed that -floop-optimize2 tends to be a pessimism on many
> algorithms.
>
> I'm hesitant to file this as a "bug", given that -floop-optimize2 is a
> "new" replacement for the older loop optimizer.
It is not that new anymore ;-) Bug reports about it are useful
only if you can show some specific (preferably small) piece of
code that regresses. But there are some open issues that make a
comparison with the old loop optimizer not quite fair at this
point.
> Is -floop-optimize2 still in development, and not ready yet -- or are
> the problems I'm seeing something that should be analyzed and reported
> as a bug?
The problems with it, at least those that I know of, are:
1) There are some unreviewed patches to make it work better, in
particular for loop invariant code motion.
2) Someone should look into GIV splitting. My feeling is that
just running the webizer on the subcfg for the loop should be
enough to get this right.
3) loop2 is placed in a totally unsuitable place in the schedule,
it should probably be moved to where the current ("old") loop
optimizer is. Or otherwise jump bypassing should be moved to
after loop2. There is a problem report about this in bugzilla.
Gr.
Steven
