https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114032
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2024-02-22
Ever confirmed|0 |1
CC| |rguenth at gcc dot gnu.org
Status|UNCONFIRMED |NEW
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
With other cases the fear is we'd completely unroll and decompose to scalars
again and then VRP eventually deriving wrong ranges.
So yeah, ifcvt would have to use a .COND_CALL here or some other tricks to
avoid the situation - like a .CLZ with the out-of-range behavior specified
according to what the target does. OTOH that target behavior might differ
between scalar and vector modes (maybe even different vector modes ...).
Do we actually treat CLZ (0) as invoking undefined behavior? IIRC we have
match patterns that rely on that but conditional on that target hook?
IL semantics based on hooks is ... bad.
Anyway - did we create a meta-bug to link all these issues together?
