[llvm] [VPlan] Skip uses-scalars restriction if one of ops needs broadcast. (PR #168246)

[Florian Hahn via llvm-commits]

================
@@ -161,18 +161,16 @@ define void @narrow_widen_store_user(i32 %x, ptr noalias %A, ptr noalias %B) {
 ; VF4IC1-NEXT:    br label %[[VECTOR_PH:.*]]
 ; VF4IC1:       [[VECTOR_PH]]:
 ; VF4IC1-NEXT:    [[TMP0:%.*]] = add i32 [[X]], 1
-; VF4IC1-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[TMP0]], i64 0
-; VF4IC1-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
-; VF4IC1-NEXT:    [[TMP5:%.*]] = mul i32 [[TMP0]], 3
-; VF4IC1-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i32> poison, i32 [[TMP5]], i64 0
+; VF4IC1-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i32> poison, i32 [[TMP0]], i64 0
 ; VF4IC1-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT1]], <4 x i32> poison, <4 x i32> zeroinitializer
+; VF4IC1-NEXT:    [[TMP5:%.*]] = mul <4 x i32> [[TMP1]], splat (i32 3)
----------------
fhahn wrote:

not converting to single-scalar, as it introduces an additional broadcast 

https://github.com/llvm/llvm-project/pull/168246