https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122846
Jeffrey A. Law <law at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2025-12-02
Ever confirmed|0 |1
--- Comment #1 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Per the patchwork discussion, this is a known issue. Essentially we keep the
number of elements the same when doing widening/narrowing ops. In combination
with our LMUL=1 default it means those loads are going to be pretty narrow.
It's recognized this isn't great and fixing this starts to push us towards the
dynamic LMUL space where we want to be. It's unclear if there's a short term
milestone where we could profitably relax how we handle widening/narrowing ops.
But longer term this all just needs to be automatic without end-user
intervension with params and such.
The conclusion was we'd keep this on the list for a week or two as some
investigative work plays out.
