[VectorCombine] Limit scalarization to non-poison indices for now.

As Eli mentioned post-commit in D103378, the result of the freeze may
still be out-of-range according to Alive2. So for now, just limit the
transform to indices that are non-poison.

diff --git a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
index 43a2172..5fae270 100644 (file)
--- a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
+++ b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
@@ -888,6 +888,9 @@ bool VectorCombine::scalarizeLoadExtract(Instruction &I) {
     if (!UI || UI->getParent() != LI->getParent())
       return false;
 
+    if (!isGuaranteedNotToBePoison(UI->getOperand(1), &AC, LI, &DT))
+      return false;
+
     // Check if any instruction between the load and the extract may modify
     // memory.
     if (LastCheckedInst->comesBefore(UI)) {
@@ -928,9 +931,6 @@ bool VectorCombine::scalarizeLoadExtract(Instruction &I) {
     Builder.SetInsertPoint(EI);
 
     Value *Idx = EI->getOperand(1);
-    if (!isGuaranteedNotToBePoison(Idx, &AC, LI, &DT))
-      Idx = Builder.CreateFreeze(Idx);
-
     Value *GEP =
         Builder.CreateInBoundsGEP(FixedVT, Ptr, {Builder.getInt32(0), Idx});
     auto *NewLoad = cast<LoadInst>(Builder.CreateLoad(

diff --git a/llvm/test/Transforms/VectorCombine/AArch64/load-extractelement-scalarization.ll b/llvm/test/Transforms/VectorCombine/AArch64/load-extractelement-scalarization.ll
index 35777ee..ecd1762 100644 (file)
--- a/llvm/test/Transforms/VectorCombine/AArch64/load-extractelement-scalarization.ll
+++ b/llvm/test/Transforms/VectorCombine/AArch64/load-extractelement-scalarization.ll
@@ -147,9 +147,8 @@ define i32 @load_extract_idx_var_i64_known_valid_by_and(<4 x i32>* %x, i64 %idx)
 ; CHECK-LABEL: @load_extract_idx_var_i64_known_valid_by_and(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[IDX_CLAMPED:%.*]] = and i64 [[IDX:%.*]], 3
-; CHECK-NEXT:    [[TMP0:%.*]] = freeze i64 [[IDX_CLAMPED]]
-; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds <4 x i32>, <4 x i32>* [[X:%.*]], i32 0, i64 [[TMP0]]
-; CHECK-NEXT:    [[R:%.*]] = load i32, i32* [[TMP1]], align 4
+; CHECK-NEXT:    [[LV:%.*]] = load <4 x i32>, <4 x i32>* [[X:%.*]], align 16
+; CHECK-NEXT:    [[R:%.*]] = extractelement <4 x i32> [[LV]], i64 [[IDX_CLAMPED]]
 ; CHECK-NEXT:    ret i32 [[R]]
 ;
 entry:
@@ -193,9 +192,8 @@ define i32 @load_extract_idx_var_i64_known_valid_by_urem(<4 x i32>* %x, i64 %idx
 ; CHECK-LABEL: @load_extract_idx_var_i64_known_valid_by_urem(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[IDX_CLAMPED:%.*]] = urem i64 [[IDX:%.*]], 4
-; CHECK-NEXT:    [[TMP0:%.*]] = freeze i64 [[IDX_CLAMPED]]
-; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds <4 x i32>, <4 x i32>* [[X:%.*]], i32 0, i64 [[TMP0]]
-; CHECK-NEXT:    [[R:%.*]] = load i32, i32* [[TMP1]], align 4
+; CHECK-NEXT:    [[LV:%.*]] = load <4 x i32>, <4 x i32>* [[X:%.*]], align 16
+; CHECK-NEXT:    [[R:%.*]] = extractelement <4 x i32> [[LV]], i64 [[IDX_CLAMPED]]
 ; CHECK-NEXT:    ret i32 [[R]]
 ;
 entry:
@@ -571,12 +569,9 @@ define i32 @load_multiple_extracts_with_variable_indices_large_vector_all_valid_
 ; CHECK-LABEL: @load_multiple_extracts_with_variable_indices_large_vector_all_valid_by_and(
 ; CHECK-NEXT:    [[IDX_0_CLAMPED:%.*]] = and i64 [[IDX_0:%.*]], 15
 ; CHECK-NEXT:    [[IDX_1_CLAMPED:%.*]] = and i64 [[IDX_1:%.*]], 15
-; CHECK-NEXT:    [[TMP1:%.*]] = freeze i64 [[IDX_0_CLAMPED]]
-; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds <16 x i32>, <16 x i32>* [[X:%.*]], i32 0, i64 [[TMP1]]
-; CHECK-NEXT:    [[E_0:%.*]] = load i32, i32* [[TMP2]], align 4
-; CHECK-NEXT:    [[TMP3:%.*]] = freeze i64 [[IDX_1_CLAMPED]]
-; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds <16 x i32>, <16 x i32>* [[X]], i32 0, i64 [[TMP3]]
-; CHECK-NEXT:    [[E_1:%.*]] = load i32, i32* [[TMP4]], align 4
+; CHECK-NEXT:    [[LV:%.*]] = load <16 x i32>, <16 x i32>* [[X:%.*]], align 64
+; CHECK-NEXT:    [[E_0:%.*]] = extractelement <16 x i32> [[LV]], i64 [[IDX_0_CLAMPED]]
+; CHECK-NEXT:    [[E_1:%.*]] = extractelement <16 x i32> [[LV]], i64 [[IDX_1_CLAMPED]]
 ; CHECK-NEXT:    [[RES:%.*]] = add i32 [[E_0]], [[E_1]]
 ; CHECK-NEXT:    ret i32 [[RES]]
 ;
@@ -594,11 +589,9 @@ define i32 @load_multiple_extracts_with_variable_indices_large_vector_all_valid_
 ; CHECK-LABEL: @load_multiple_extracts_with_variable_indices_large_vector_all_valid_by_and_some_noundef(
 ; CHECK-NEXT:    [[IDX_0_CLAMPED:%.*]] = and i64 [[IDX_0:%.*]], 15
 ; CHECK-NEXT:    [[IDX_1_CLAMPED:%.*]] = and i64 [[IDX_1:%.*]], 15
-; CHECK-NEXT:    [[TMP1:%.*]] = freeze i64 [[IDX_0_CLAMPED]]
-; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds <16 x i32>, <16 x i32>* [[X:%.*]], i32 0, i64 [[TMP1]]
-; CHECK-NEXT:    [[E_0:%.*]] = load i32, i32* [[TMP2]], align 4
-; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds <16 x i32>, <16 x i32>* [[X]], i32 0, i64 [[IDX_1_CLAMPED]]
-; CHECK-NEXT:    [[E_1:%.*]] = load i32, i32* [[TMP3]], align 4
+; CHECK-NEXT:    [[LV:%.*]] = load <16 x i32>, <16 x i32>* [[X:%.*]], align 64
+; CHECK-NEXT:    [[E_0:%.*]] = extractelement <16 x i32> [[LV]], i64 [[IDX_0_CLAMPED]]
+; CHECK-NEXT:    [[E_1:%.*]] = extractelement <16 x i32> [[LV]], i64 [[IDX_1_CLAMPED]]
 ; CHECK-NEXT:    [[RES:%.*]] = add i32 [[E_0]], [[E_1]]
 ; CHECK-NEXT:    ret i32 [[RES]]
 ;