[Bug target/122751] missed fusing of sign-extend and MLA

[tnfchris at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122751

--- Comment #2 from Tamar Christina <tnfchris at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #1)
> My view was always that we want bswap-pass like analysis of 'bytes' also in
> vectors and then pattern-match useful alterations.  Of course the bswap
> analysis does not represent 'sign-extend' at the moment.  There's also the
> symbolic execution stuff.

Hmm how would this work, since the predicate also changes. i.e. you need to
change both the data and mask lanes. So doing it after vectorization would be
pretty hard.

Or did you mean a bswap-pass during vectorization?