[llvm] [VPlan] Don't apply predication discount to non-originally-predicated blocks (PR #160449)

[Florian Hahn via llvm-commits]

================
@@ -760,28 +777,46 @@ define void @update_2_uses_in_same_recipe_in_merged_block(i32 %k) {
 ; CHECK-NEXT: vector.body:
 ; CHECK-NEXT:   EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
 ; CHECK-NEXT:   ir<%iv> = WIDEN-INDUCTION ir<0>, ir<1>, vp<[[VF]]>
+; CHECK-NEXT:   vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>, vp<[[VF]]>
----------------
fhahn wrote:

I think this is no not testing what it used to (replicate regions aren't merged any more), would be good to adjust the test so this merging still takes place

https://github.com/llvm/llvm-project/pull/160449