[llvm] [VPlan] Use predicated intrinsics for trapping divisors (PR #154076)

[Mel Chen via llvm-commits]

Mel-Chen wrote:

> Spent some time revisiting this, I think it's probably worth the effort if we introduce a dedicated @llvm.masked.udiv intrinsic and friends. It also helps avoid the mask on AArch64 SVE and means we could do the regular masked op -> VP intrinsic transform for EVL tail folding.
> 
> But I would need to check to see if there's any impact from lost combines/folds using the intrinsic vs. the binary op.

+1

https://github.com/llvm/llvm-project/pull/154076