[llvm] [VPlan] Scalarize to first-lane-only directly on VPlan (PR #184267)

[Andrei Elovikov via llvm-commits]

================
@@ -8,14 +8,14 @@ target triple = "arm64-apple-macosx"
 define void @udiv_rhs_opt_cost(ptr %dst) #0 {
 ; CHECK-LABEL: 'udiv_rhs_opt_cost'
 ; CHECK:  LV: Found an estimated cost of 5 for VF 1 For instruction: %div = udiv i8 %iv.trunc, 3
-; CHECK:  Cost of 5 for VF 2: CLONE ir<%div> = udiv vp<[[VP7:%[0-9]+]]>, ir<3>
+; CHECK:  Cost of 5 for VF 2: CLONE ir<%div> = udiv ir<%iv.trunc>, ir<3>
----------------
eas wrote:

We avoid `tryToOptimizeInductionTruncate` processing by scalarizing before widening. Generated LLVM IR is the same and I think this is fine.

https://github.com/llvm/llvm-project/pull/184267