https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123997

--- Comment #3 from vekumar at gcc dot gnu.org ---
(In reply to Hongtao Liu from comment #2)
> > I suspect quite some explosions in patterns if we want to handle this (and
> > other operations that can do fault suppression) memory forwarding via
> > combine.  Unsure if there's another, better, way to achieve such forwarding
> > with some md-reorg?
> 
> Yes, because vec_merge itself does not have the meaning of suppressing
> memory faults, we specifically added UNSPEC_MASKLOAD to prevent errors in
> PR110309. This causes all subsequent <mask_name> patterns to be unable to
> generated by combine with .MASKLOAD. + operations.

Hongtao Liu and Richard - We will attempt to fix this by adding missing
patterns. Let us know if there is a better way to avoid explosion of adding
patterns.

Reply via email to