[Bug tree-optimization/83359] [8 Regression] ICE in expand_LOOP_DIST_ALIAS, at internal-fn.c:2362

["rguenth at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83359

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #5)
> Created attachment 42830 [details]
> gcc8-pr83359.patch
> 
> In the mean time I wrote this patch.
> 
> That said, LOOP_DIST_ALIAS seems quite fragile to me, unlike ifcvt added
> LOOP_VECTORIZED calls which go away in the very next pass, there are many
> passes in between ldist and vectorizer and I'd be worried that the calls
> could be cloned with other loop copying, moved and all other kinds of
> things, especially in graphite.

IMHO given it's fragility we should see to tame it down for the release
with either folding the remaining calls during fold-all-builtins or
replacing expand_LOOP_DIST_ALIAS with code that does in fact expand
the function (to its boolean argument).

expand_LOOP_DIST_ALIAS should then emit a note in the .expand dump so
we can track the cases we leak it.