https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115886
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
Last reconfirmed| |2024-07-12
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
wrt concat4 there's possibly also missed canonicalization on GIMPLE of
a CTOR vs. permute. But it really depends on whether vec_init with vector
components has better coverage than vec_perm with constant permute. Since
vec_init can handle more than two sources it might feel that it's more
canonical than vec_perm. OTOH we might choose to turn all two-component
CTORs to permutes if supported. But on GIMPLE the "if supported" conditional
comes in the way with CTORs having RTL expand fallback (bad one, via memory
and guaranteed store-to-load forward issue) but permute not.
So I'd lean towards CTORs for concat permutes.
