[VectorCombine] Limit scalarization known non-poison indices.

[fhahn]

Summary

We can only trust the range of the index if it is guaranteed
non-poison.

Fixes PR50949.

[lebedev.ri]

I think this could use some comments (both before the check and in testcases) that freeze doesn't help.

[fhahn]

In D107364#2922705, @lebedev.ri wrote:

I think this could use some comments (both before the check and in testcases) that freeze doesn't help.

freeze would help, if we push it to the and and urem operands respectively I think. But I don't think there's an easy way to do this when using computeConstantRange. Not sure if there's potential for a convenient API that transforms a constant range which may include poison to one without? Otherwise I could match the and and urem patterns manually, which would allow inserting the freeze at the right place.

[efriedma]

In D107364#2923374, @fhahn wrote:

Not sure if there's potential for a convenient API that transforms a constant range which may include poison to one without?

Not sure how useful that would be in general, but maybe. I mean, "match the and and urem patterns manually" is basically such an API.

llvm/test/Transforms/VectorCombine/load-insert-store.ll

419

Please make sure we have a corresponding testcase with the appropriate freeze.

[fhahn]

rebased after adding additional tests in 8d2a8ced00dc45605cb85.

[fhahn]

llvm/test/Transforms/VectorCombine/load-insert-store.ll

419

I added variants that either freeze the %idx operand or %idx.clamped in 8d2a8ced00dc

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

 
 /// Check if it is legal to scalarize a memory access to \p VecTy at index \p
 /// Idx. \p Idx must access a valid vector element.
 static bool canScalarizeAccess(FixedVectorType *VecTy, Value *Idx,
                                Instruction *CtxI, AssumptionCache &AC) {
   if (auto *C = dyn_cast<ConstantInt>(Idx))
     return C->getValue().ult(VecTy->getNumElements());
 
+  if (!isGuaranteedNotToBePoison(Idx, &AC))
+    return false;
+
   APInt Zero(Idx->getType()->getScalarSizeInBits(), 0);
   APInt MaxElts(Idx->getType()->getScalarSizeInBits(), VecTy->getNumElements());
   ConstantRange ValidIndices(Zero, MaxElts);
   ConstantRange IdxRange = computeConstantRange(Idx, true, &AC, CtxI, 0);
   return ValidIndices.contains(IdxRange);
 }
 
 /// The memory operation on a vector of \p ScalarType had alignment of

llvm/test/Transforms/VectorCombine/load-insert-store.ll

entry:
   %vecins = insertelement <16 x i8> %0, i8 %s, i32 %idx.clamped.frozen
   store <16 x i8> %vecins, <16 x i8>* %q
   ret void
 }
 
 define void @insert_store_nonconst_index_known_valid_by_and_but_may_be_poison(<16 x i8>* %q, i8 zeroext %s, i32 %idx) {
 ; CHECK-LABEL: @insert_store_nonconst_index_known_valid_by_and_but_may_be_poison(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = load <16 x i8>, <16 x i8>* [[Q:%.*]], align 16
 ; CHECK-NEXT:    [[IDX_CLAMPED:%.*]] = and i32 [[IDX:%.*]], 7
-; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds <16 x i8>, <16 x i8>* [[Q:%.*]], i32 0, i32 [[IDX_CLAMPED]]
-; CHECK-NEXT:    store i8 [[S:%.*]], i8* [[TMP0]], align 1
+; CHECK-NEXT:    [[VECINS:%.*]] = insertelement <16 x i8> [[TMP0]], i8 [[S:%.*]], i32 [[IDX_CLAMPED]]
+; CHECK-NEXT:    store <16 x i8> [[VECINS]], <16 x i8>* [[Q]], align 16
 ; CHECK-NEXT:    ret void
 ;
 entry:
   %0 = load <16 x i8>, <16 x i8>* %q
   %idx.clamped = and i32 %idx, 7
   %vecins = insertelement <16 x i8> %0, i8 %s, i32 %idx.clamped
   store <16 x i8> %vecins, <16 x i8>* %q
   ret void
entry:
   %vecins = insertelement <16 x i8> %0, i8 %s, i32 %idx.clamped.frozen
   store <16 x i8> %vecins, <16 x i8>* %q
   ret void
 }
 
 define void @insert_store_nonconst_index_known_valid_by_urem_but_may_be_poison(<16 x i8>* %q, i8 zeroext %s, i32 %idx) {
 ; CHECK-LABEL: @insert_store_nonconst_index_known_valid_by_urem_but_may_be_poison(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = load <16 x i8>, <16 x i8>* [[Q:%.*]], align 16
 ; CHECK-NEXT:    [[IDX_CLAMPED:%.*]] = urem i32 [[IDX:%.*]], 16
-; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds <16 x i8>, <16 x i8>* [[Q:%.*]], i32 0, i32 [[IDX_CLAMPED]]
-; CHECK-NEXT:    store i8 [[S:%.*]], i8* [[TMP0]], align 1
+; CHECK-NEXT:    [[VECINS:%.*]] = insertelement <16 x i8> [[TMP0]], i8 [[S:%.*]], i32 [[IDX_CLAMPED]]
+; CHECK-NEXT:    store <16 x i8> [[VECINS]], <16 x i8>* [[Q]], align 16
 ; CHECK-NEXT:    ret void
 ;
 entry:
   %0 = load <16 x i8>, <16 x i8>* %q
   %idx.clamped = urem i32 %idx, 16
   %vecins = insertelement <16 x i8> %0, i8 %s, i32 %idx.clamped
   store <16 x i8> %vecins, <16 x i8>* %q
   ret void