http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56309
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |rguenth at gcc dot gnu.org
--- Comment #24 from Richard Biener <rguenth at gcc dot gnu.org> 2013-02-15
09:36:23 UTC ---
(In reply to comment #22)
> Yeah. We need to either find another way how to present vectorizer with the
> ifconverted statement (perhaps similar to how pattern matching works, I think
> if-conversion only handles inner-most loops, so maybe just as an alternative
> gimple_seq for the single bb of the loop body?), or we need some kind of
> if-unconversion pass.
> See the http://gcc.gnu.org/ml/gcc-patches/2012-11/msg01659.html patch which
> contains much more limited version of if-unconversion, unfortunately it is
> going to be harder to handle more complex conditions efficiently.
Well, ideally if-conversion would work as part of vectorization.
if-conversion is nothing else than pattern detection - but of course the
complication is that the vectorizer would need to be taught to vectorize
loops with more than a single basic-block (even though the vectorized
loop would again contain a single basic-block).
