[llvm] [VP][RISCV] Add a vp.load.ff intrinsic for fault only first load. (PR #128593)

[Luke Lau via llvm-commits]

lukel97 wrote:

> This seems to suggest that there is a generic lowering, but I don't see how it would work.

Oh I forgot about that lowering actually. It's definitely not optimal but it should be correct, because if the first lane traps then the intrinsic is also allowed to trap, but only for the first lane:

> If ``evl`` is non-zero and the first lane is masked-on, then the
first lane of the vector needs to be inbounds of an allocation.

We should probably implement this

https://github.com/llvm/llvm-project/pull/128593