[llvm] [NFCI][VPlan] Split initial mem-widening into a separate transformation (PR #182592)

[Florian Hahn via llvm-commits]

================
@@ -0,0 +1,114 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 6
+; RUN: opt -S -p loop-vectorize -mtriple=arm64 < %s | FileCheck %s
+
+; Crashed during refactoring if reduction store is not sunk out of the loop.
+define void @ordered_reduction(ptr %dst, float %a, float %b) {
+; CHECK-LABEL: define void @ordered_reduction(
+; CHECK-SAME: ptr [[DST:%.*]], float [[A:%.*]], float [[B:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    br label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[B]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x float> poison, float [[A]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT1]], <2 x float> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP0:%.*]] = fmul <2 x float> [[BROADCAST_SPLAT2]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi float [ 0.000000e+00, %[[VECTOR_PH]] ], [ [[TMP2:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = call float @llvm.vector.reduce.fadd.v2f32(float [[VEC_PHI]], <2 x float> [[TMP0]])
+; CHECK-NEXT:    [[TMP2]] = call float @llvm.vector.reduce.fadd.v2f32(float [[TMP1]], <2 x float> [[TMP0]])
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
+; CHECK-NEXT:    br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    store float [[TMP2]], ptr [[DST]], align 4
+; CHECK-NEXT:    br label %[[SCALAR_PH:.*]]
+; CHECK:       [[SCALAR_PH]]:
+; CHECK-NEXT:    br label %[[LOOP:.*]]
+; CHECK:       [[LOOP]]:
+; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ 100, %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[SUM:%.*]] = phi float [ [[TMP2]], %[[SCALAR_PH]] ], [ [[MULADD:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[MULADD]] = tail call float @llvm.fmuladd.f32(float [[A]], float [[B]], float [[SUM]])
+; CHECK-NEXT:    store float [[MULADD]], ptr [[DST]], align 4
+; CHECK-NEXT:    [[IV_NEXT]] = add i64 [[IV]], 1
+; CHECK-NEXT:    [[DONE:%.*]] = icmp eq i64 [[IV]], 100
+; CHECK-NEXT:    br i1 [[DONE]], label %[[EXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK:       [[EXIT]]:
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %loop
+
+loop:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+  %sum = phi float [ 0.000000e+00, %entry ], [ %muladd, %loop ]
+  %muladd = tail call float @llvm.fmuladd.f32(float %a, float %b, float %sum)
+  store float %muladd, ptr %dst, align 4
+  %iv.next = add i64 %iv, 1
+  %done = icmp eq i64 %iv, 100
+  br i1 %done, label %exit, label %loop
+
+exit:
+  ret void
+}
+
+; Crashed due to not updating "return -> continue" after cut-paste during refactoring.
+define void @ordered_reduction2(ptr %dst, ptr %src) {
----------------
fhahn wrote:

```suggestion
define void @ordered_reduction2(ptr no alias %dst, ptr no alias %src) {
```

could be no alias I think

https://github.com/llvm/llvm-project/pull/182592