[llvm] [ExpandVectorPredication] Expand vp.load.ff. (PR #154440)

[Philip Reames via llvm-commits]

preames wrote:

> The EVL is folded into the mask in `foldEVLIntoMask` so if the EVL is zero then the mask in the expansion will also be zero

Is there guaranteed?  If so, then yeah, that's sound modulo your point about returning 1 in this case.  Hm, don't we have a forward progress problem here though?  If EVL is non-zero, but the 0-th lane of the mask is, don't we need to return at least one for forward progress in an idiomatic loop?

https://github.com/llvm/llvm-project/pull/154440