Addressed comments

Author: npanchen

llvm/lib/Analysis/ConstantFolding.cpp

@@ -3763,15 +3763,21 @@ static Constant *ConstantFoldFixedVectorCall(
     unsigned NumElements = FVTy->getNumElements();
     unsigned VecNumElements =
         cast<FixedVectorType>(Vec->getType())->getNumElements();
+    unsigned StartingIndex = Idx->getZExtValue();
     // Extracting entire vector is nop
-    if (NumElements == VecNumElements)
+    if (NumElements == VecNumElements && StartingIndex == 0)
       return Vec;
 
-    unsigned StartingIndex = Idx->getZExtValue();
-    assert(StartingIndex + NumElements <= VecNumElements &&
-           "Cannot extract more elements than exist in the vector");
-    for (unsigned I = 0; I != NumElements; ++I)
-      Result[I] = Vec->getAggregateElement(StartingIndex + I);
+    const unsigned NonPoisonNumElements =
+        std::min(StartingIndex + NumElements, VecNumElements);
+    for (unsigned I = StartingIndex; I < NonPoisonNumElements; ++I)
+      Result[I - StartingIndex] = Vec->getAggregateElement(I);
+
+    // Remaining elements are poison since they are out of bounds.
+    for (unsigned I = NonPoisonNumElements, E = StartingIndex + NumElements;
+         I < E; ++I)
+      Result[I - StartingIndex] = PoisonValue::get(FVTy->getElementType());
+
     return ConstantVector::get(Result);
   }
   case Intrinsic::vector_insert: {
@@ -3787,9 +3793,15 @@ static Constant *ConstantFoldFixedVectorCall(
         cast<FixedVectorType>(Vec->getType())->getNumElements();
     unsigned IdxN = Idx->getZExtValue();
     // Replacing entire vector with a subvec is nop
-    if (SubVecNumElements == VecNumElements)
+    if (SubVecNumElements == VecNumElements && IdxN == 0)
       return SubVec;
 
+    // Make sure indices are in the range [0, VecNumElements), otherwise the
+    // result is a poison value.
+    if (IdxN >= VecNumElements || IdxN + SubVecNumElements > VecNumElements ||
+        (IdxN && (SubVecNumElements % IdxN) != 0))
+      return PoisonValue::get(FVTy);
+
     unsigned I = 0;
     for (; I < IdxN; ++I)
       Result[I] = Vec->getAggregateElement(I);
@@ -3980,7 +3992,8 @@ ConstantFoldStructCall(StringRef Name, Intrinsic::ID IntrinsicID,
       return nullptr;
 
     unsigned NumElements =
-        cast<FixedVectorType>(Vec->getType())->getNumElements() / 2;
+        cast<VectorType>(Vec->getType())->getElementCount().getKnownMinValue() /
+        2;
     SmallVector<Constant *, 4> Res0(NumElements), Res1(NumElements);
     for (unsigned I = 0; I < NumElements; ++I) {
       Res0[I] = Vec->getAggregateElement(2 * I);

llvm/test/Transforms/InstSimplify/ConstProp/vector-calls.ll

@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
-; RUN: opt < %s -passes=instsimplify,verify -S | FileCheck %s
+; RUN: opt < %s -passes=instsimplify,verify -disable-verify -S | FileCheck %s
 
 define <3 x i32> @fold_vector_extract() {
 ; CHECK-LABEL: define <3 x i32> @fold_vector_extract() {
@@ -9,6 +9,22 @@ define <3 x i32> @fold_vector_extract() {
   ret <3 x i32> %1
 }
 
+define <3 x i32> @fold_vector_extract_last_poison() {
+; CHECK-LABEL: define <3 x i32> @fold_vector_extract_last_poison() {
+; CHECK-NEXT:    ret <3 x i32> <i32 6, i32 7, i32 poison>
+;
+  %1 = call <3 x i32> @llvm.vector.extract.v3i32.v8i32(<8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>, i64 6)
+  ret <3 x i32> %1
+}
+
+define <3 x i32> @fold_vector_extract_poison() {
+; CHECK-LABEL: define <3 x i32> @fold_vector_extract_poison() {
+; CHECK-NEXT:    ret <3 x i32> poison
+;
+  %1 = call <3 x i32> @llvm.vector.extract.v3i32.v8i32(<8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>, i64 8)
+  ret <3 x i32> %1
+}
+
 define <8 x i32> @fold_vector_extract_nop() {
 ; CHECK-LABEL: define <8 x i32> @fold_vector_extract_nop() {
 ; CHECK-NEXT:    ret <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
@@ -33,6 +49,22 @@ define <8 x i32> @fold_vector_insert_nop() {
   ret <8 x i32> %1
 }
 
+define <8 x i32> @fold_vector_insert_poison_idx_range() {
+; CHECK-LABEL: define <8 x i32> @fold_vector_insert_poison_idx_range() {
+; CHECK-NEXT:    ret <8 x i32> poison
+;
+  %1 = call <8 x i32> @llvm.vector.insert.v8i32(<8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>, <6 x i32> <i32 9, i32 10, i32 11, i32 12, i32 13, i32 14>, i64 6)
+  ret <8 x i32> %1
+}
+
+define <8 x i32> @fold_vector_insert_poison_large_idx() {
+; CHECK-LABEL: define <8 x i32> @fold_vector_insert_poison_large_idx() {
+; CHECK-NEXT:    ret <8 x i32> poison
+;
+  %1 = call <8 x i32> @llvm.vector.insert.v8i32(<8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>, <6 x i32> <i32 9, i32 10, i32 11, i32 12, i32 13, i32 14>, i64 -2)
+  ret <8 x i32> %1
+}
+
 define <8 x i32> @fold_vector_interleave2() {
 ; CHECK-LABEL: define <8 x i32> @fold_vector_interleave2() {
 ; CHECK-NEXT:    ret <8 x i32> <i32 1, i32 5, i32 2, i32 6, i32 3, i32 7, i32 4, i32 8>
@@ -48,3 +80,11 @@ define {<4 x i32>, <4 x i32>} @fold_vector_deinterleav2() {
   %1 = call {<4 x i32>, <4 x i32>} @llvm.vector.deinterleave2.v4i32.v8i32(<8 x i32> <i32 1, i32 5, i32 2, i32 6, i32 3, i32 7, i32 4, i32 8>)
   ret {<4 x i32>, <4 x i32>} %1
 }
+
+define {<vscale x 4 x i32>, <vscale x 4 x i32>} @fold_scalable_vector_deinterleav2() {
+; CHECK-LABEL: define { <vscale x 4 x i32>, <vscale x 4 x i32> } @fold_scalable_vector_deinterleav2() {
+; CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } zeroinitializer
+;
+  %1 = call {<vscale x 4 x i32>, <vscale x 4 x i32>} @llvm.vector.deinterleave2.v4i32.v8i32(<vscale x 8 x i32> zeroinitializer)
+  ret {<vscale x 4 x i32>, <vscale x 4 x i32>} %1
+}