[llvm] [IA] Fix crash when dealing with deinterleave(interleave) (PR #150713)

[Craig Topper via llvm-commits]

topperc wrote:

> > Should we just bail out if the load intrinsics are not vp.load or masked.load early on in InterleavedAccessImpl::lowerDeinterleaveIntrinsic (and similar thing for stores)?
> 
> If I understood your right, you suggest to add a check `isa<Intrinsic::vp_load, masked_load>(II)`, right ? It seems to be extra boilerplate as `getMaskOperand` does this already. Not strong argument, but prior to #150241 [these checks](https://github.com/llvm/llvm-project/pull/150241/files#diff-5d08820e48ae7cfb270db4934014822ec75324d9342b980e10ee7c9e94cecab1L595-L607) were in the same place I added

>From an interface design, it feels a little weird to me that the `getMaskOperand` function is telling the caller whether the case is interesting or not.

https://github.com/llvm/llvm-project/pull/150713