https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125875
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Component|rtl-optimization |middle-end
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
And then vector lowering allows for vec_shr_optab/vec_shl_optab, and in the x86
case we can do vec_shl, so we don't lower. can_vec_perm_const_p says false.
IMHO this paths in vector lowering have to vanish. When a target can do
vec_shl the target should better update vec_perm_const handling to cover it.
I'm fixing the middle-end part, but not the backend part of this (missed
vec_perm_const handling for V4HI vector shift).
