[llvm] [LAA] Avoid stray predicates from replaceSymStrides (PR #216350)

[Ramkumar Ramachandra via llvm-commits]

================
@@ -146,43 +146,35 @@ define void @versioned_sext_use_in_gep(i32 %scale, ptr %dst, i64 %scale.2) {
 ; CHECK-SAME: i32 [[SCALE:%.*]], ptr [[DST:%.*]], i64 [[SCALE_2:%.*]]) {
 ; CHECK-NEXT:  [[ENTRY:.*:]]
 ; CHECK-NEXT:    [[SCALE_EXT:%.*]] = sext i32 [[SCALE]] to i64
-; CHECK-NEXT:    br label %[[VECTOR_SCEVCHECK:.*]]
-; CHECK:       [[VECTOR_SCEVCHECK]]:
-; CHECK-NEXT:    [[IDENT_CHECK:%.*]] = icmp ne i32 [[SCALE]], 1
-; CHECK-NEXT:    br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK-NEXT:    br label %[[VECTOR_PH:.*]]
 ; CHECK:       [[VECTOR_PH]]:
-; CHECK-NEXT:    [[TMP83:%.*]] = getelementptr i8, ptr [[DST]], i64 [[SCALE_2]]
+; CHECK-NEXT:    [[TMP0:%.*]] = mul i64 [[SCALE_EXT]], [[SCALE_2]]
+; CHECK-NEXT:    [[TMP83:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP0]]
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[SCALE_EXT]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
 ; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
 ; CHECK:       [[VECTOR_BODY]]:
-; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[TMP12:%.*]] = add i64 [[INDEX]], 1
-; CHECK-NEXT:    [[TMP14:%.*]] = add i64 [[INDEX]], 2
-; CHECK-NEXT:    [[TMP16:%.*]] = add i64 [[INDEX]], 3
+; CHECK-NEXT:    [[INDEX1:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
----------------
artagnon wrote:

I'm unfortunately not able to save the test?

```diff
diff --git a/llvm/test/Transforms/LoopVectorize/version-stride-with-integer-casts.ll b/llvm/test/Transforms/LoopVectorize/version-stride-with-integer-casts.ll
index 3a57803775a5..4295fca7002e 100644
--- a/llvm/test/Transforms/LoopVectorize/version-stride-with-integer-casts.ll
+++ b/llvm/test/Transforms/LoopVectorize/version-stride-with-integer-casts.ll
@@ -144,37 +144,18 @@ exit:
 define void @versioned_sext_use_in_gep(i32 %scale, ptr %dst, i64 %scale.2) {
 ; CHECK-LABEL: define void @versioned_sext_use_in_gep(
 ; CHECK-SAME: i32 [[SCALE:%.*]], ptr [[DST:%.*]], i64 [[SCALE_2:%.*]]) {
-; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:  [[ENTRY:.*]]:
 ; CHECK-NEXT:    [[SCALE_EXT:%.*]] = sext i32 [[SCALE]] to i64
-; CHECK-NEXT:    br label %[[VECTOR_PH:.*]]
-; CHECK:       [[VECTOR_PH]]:
-; CHECK-NEXT:    [[TMP0:%.*]] = mul i64 [[SCALE_EXT]], [[SCALE_2]]
-; CHECK-NEXT:    [[TMP83:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP0]]
-; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[SCALE_EXT]], i64 0
-; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
 ; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
 ; CHECK:       [[VECTOR_BODY]]:
-; CHECK-NEXT:    [[INDEX1:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[TMP2:%.*]] = mul <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
-; CHECK-NEXT:    [[INDEX:%.*]] = extractelement <4 x i64> [[TMP2]], i64 0
-; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr i8, ptr [[DST]], i64 [[INDEX]]
-; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <4 x i64> [[TMP2]], i64 1
-; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP12]]
-; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <4 x i64> [[TMP2]], i64 2
+; CHECK-NEXT:    [[TMP14:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP14]]
-; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <4 x i64> [[TMP2]], i64 3
+; CHECK-NEXT:    [[INDEX_NEXT]] = add i64 [[TMP14]], [[SCALE_EXT]]
+; CHECK-NEXT:    [[TMP16:%.*]] = mul i64 [[SCALE_EXT]], [[SCALE_2]]
 ; CHECK-NEXT:    [[TMP17:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP16]]
-; CHECK-NEXT:    store ptr [[TMP83]], ptr [[TMP11]], align 8
-; CHECK-NEXT:    store ptr [[TMP83]], ptr [[TMP13]], align 8
-; CHECK-NEXT:    store ptr [[TMP83]], ptr [[TMP15]], align 8
-; CHECK-NEXT:    store ptr [[TMP83]], ptr [[TMP17]], align 8
-; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX1]], 4
-; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 4)
+; CHECK-NEXT:    store ptr [[TMP17]], ptr [[TMP15]], align 8
 ; CHECK-NEXT:    [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256
-; CHECK-NEXT:    br i1 [[TMP18]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
-; CHECK:       [[MIDDLE_BLOCK]]:
-; CHECK-NEXT:    br label %[[EXIT:.*]]
+; CHECK-NEXT:    br i1 [[TMP18]], label %[[EXIT:.*]], label %[[VECTOR_BODY]]
 ; CHECK:       [[EXIT]]:
 ; CHECK-NEXT:    ret void
 ;
@@ -184,9 +165,8 @@ entry:

 loop:
   %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
-  %iv.mul  = mul i64 %iv, %scale.ext
-  %gep.1 = getelementptr i8, ptr %dst, i64 %iv.mul
-  %iv.next = add i64 %iv, 1
+  %gep.1 = getelementptr i8, ptr %dst, i64 %iv
+  %iv.next = add i64 %iv, %scale.ext
   %scale.mul = mul i64 %scale.ext, %scale.2
   %gep.2 = getelementptr i8, ptr %dst, i64 %scale.mul
   store ptr %gep.2, ptr %gep.1, align 8
```

It's no longer profitable to vectorize -- any ideas on what else I can try?

https://github.com/llvm/llvm-project/pull/216350