https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123954
Robin Dapp <rdapp at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2026-02-05
Ever confirmed|0 |1
--- Comment #2 from Robin Dapp <rdapp at gcc dot gnu.org> ---
(In reply to Ilya Kurdyukov from comment #1)
> It might be worth asking for clarification in the RVV spec as to which of
> vs1/vs2 is more important in this case.
Hmm when reading the spec literally (the way it should be read)
I think your interpretation is correct. The sentence refers only to individual
registers or register groups and not the whole LHS or RHS of an insn.
Our implementation, however, relies on an earlyclobber on the target which is
too coarse to distinguish between individual source operands.
We had patches in the same space but IIRC we only handled the LMUL > 1 case
and not this case here.
