[llvm] [VPlan] Directly unroll VectorPointerRecipe (PR #168886)

[Florian Hahn via llvm-commits]

================
@@ -528,7 +528,8 @@ define void @multiple_exit_conditions(ptr %src, ptr noalias %dst) #1 {
 ; DEFAULT-NEXT:    br i1 [[MIN_ITERS_CHECK1]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
 ; DEFAULT:       [[VECTOR_PH]]:
 ; DEFAULT-NEXT:    [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
-; DEFAULT-NEXT:    [[TMP5:%.*]] = mul nuw i64 [[TMP4]], 16
+; DEFAULT-NEXT:    [[TMP11:%.*]] = mul nuw i64 [[TMP4]], 4
+; DEFAULT-NEXT:    [[TMP5:%.*]] = mul i64 [[TMP11]], 4
----------------
fhahn wrote:

This is somewhat unrelated directly to the patch, but could we do something to keep the  NUW when computing VF and VFxUF separately?

https://github.com/llvm/llvm-project/pull/168886