[ConstraintElim] Use nusw flag for GEP decomposition

Check for nusw instead of inbounds when decomposing GEPs.

In this particular case, we can also look through multiple nusw
flags, because we will ultimately be working in the unsigned
constraint system.

Author: nikic

llvm/lib/Transforms/Scalar/ConstraintElimination.cpp

@@ -387,12 +387,12 @@ struct OffsetResult {
   Value *BasePtr;
   APInt ConstantOffset;
   SmallMapVector<Value *, APInt, 4> VariableOffsets;
-  bool AllInbounds;
+  GEPNoWrapFlags NW;
 
   OffsetResult() : BasePtr(nullptr), ConstantOffset(0, uint64_t(0)) {}
 
   OffsetResult(GEPOperator &GEP, const DataLayout &DL)
-      : BasePtr(GEP.getPointerOperand()), AllInbounds(GEP.isInBounds()) {
+      : BasePtr(GEP.getPointerOperand()), NW(GEP.getNoWrapFlags()) {
     ConstantOffset = APInt(DL.getIndexTypeSizeInBits(BasePtr->getType()), 0);
   }
 };
@@ -426,7 +426,7 @@ static OffsetResult collectOffsets(GEPOperator &GEP, const DataLayout &DL) {
     Result.ConstantOffset += ConstantOffset2;
     if (Result.VariableOffsets.size() == 0 && VariableOffsets2.size() == 1)
       Result.VariableOffsets = VariableOffsets2;
-    Result.AllInbounds &= InnerGEP->isInBounds();
+    Result.NW &= InnerGEP->getNoWrapFlags();
   }
   return Result;
 }
@@ -450,9 +450,9 @@ static Decomposition decomposeGEP(GEPOperator &GEP,
 
   assert(!IsSigned && "The logic below only supports decomposition for "
                       "unsigned predicates at the moment.");
-  const auto &[BasePtr, ConstantOffset, VariableOffsets, AllInbounds] =
+  const auto &[BasePtr, ConstantOffset, VariableOffsets, NW] =
       collectOffsets(GEP, DL);
-  if (!BasePtr || !AllInbounds)
+  if (!BasePtr || !NW.hasNoUnsignedSignedWrap())
     return &GEP;
 
   Decomposition Result(ConstantOffset.getSExtValue(), DecompEntry(1, BasePtr));

llvm/test/Transforms/ConstraintElimination/gep-arithmetic.ll

@@ -630,3 +630,68 @@ ptr.check:
 exit:
   ret i4 3
 }
+
+define i1 @test_nusw(ptr %p, i32 %x, i32 %y) {
+; CHECK-LABEL: @test_nusw(
+; CHECK-NEXT:    [[X_EXT:%.*]] = zext i32 [[X:%.*]] to i64
+; CHECK-NEXT:    [[Y_EXT:%.*]] = zext i32 [[Y:%.*]] to i64
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp ugt i64 [[X_EXT]], [[Y_EXT]]
+; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP1]])
+; CHECK-NEXT:    [[GEP_X:%.*]] = getelementptr nusw i8, ptr [[P:%.*]], i64 [[X_EXT]]
+; CHECK-NEXT:    [[GEP_Y:%.*]] = getelementptr nusw i8, ptr [[P]], i64 [[Y_EXT]]
+; CHECK-NEXT:    ret i1 true
+;
+  %x.ext = zext i32 %x to i64
+  %y.ext = zext i32 %y to i64
+  %cmp1 = icmp ugt i64 %x.ext, %y.ext
+  call void @llvm.assume(i1 %cmp1)
+  %gep.x = getelementptr nusw i8, ptr %p, i64 %x.ext
+  %gep.y = getelementptr nusw i8, ptr %p, i64 %y.ext
+  %cmp2 = icmp ugt ptr %gep.x, %gep.y
+  ret i1 %cmp2
+}
+
+define i1 @test_nusw_nested(ptr %p, i32 %x, i32 %y) {
+; CHECK-LABEL: @test_nusw_nested(
+; CHECK-NEXT:    [[X_EXT:%.*]] = zext i32 [[X:%.*]] to i64
+; CHECK-NEXT:    [[Y_EXT:%.*]] = zext i32 [[Y:%.*]] to i64
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp ugt i64 [[X_EXT]], [[Y_EXT]]
+; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP1]])
+; CHECK-NEXT:    [[GEP_X:%.*]] = getelementptr nusw i8, ptr [[P:%.*]], i64 [[X_EXT]]
+; CHECK-NEXT:    [[GEP_X1:%.*]] = getelementptr nusw i8, ptr [[GEP_X]], i64 1
+; CHECK-NEXT:    [[GEP_Y:%.*]] = getelementptr nusw i8, ptr [[P]], i64 [[Y_EXT]]
+; CHECK-NEXT:    ret i1 true
+;
+  %x.ext = zext i32 %x to i64
+  %y.ext = zext i32 %y to i64
+  %cmp1 = icmp ugt i64 %x.ext, %y.ext
+  call void @llvm.assume(i1 %cmp1)
+  %gep.x = getelementptr nusw i8, ptr %p, i64 %x.ext
+  %gep.x1 = getelementptr nusw i8, ptr %gep.x, i64 1
+  %gep.y = getelementptr nusw i8, ptr %p, i64 %y.ext
+  %cmp2 = icmp ugt ptr %gep.x1, %gep.y
+  ret i1 %cmp2
+}
+
+define i1 @test_missing_nusw(ptr %p, i32 %x, i32 %y) {
+; CHECK-LABEL: @test_missing_nusw(
+; CHECK-NEXT:    [[X_EXT:%.*]] = zext i32 [[X:%.*]] to i64
+; CHECK-NEXT:    [[Y_EXT:%.*]] = zext i32 [[Y:%.*]] to i64
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp ugt i64 [[X_EXT]], [[Y_EXT]]
+; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP1]])
+; CHECK-NEXT:    [[GEP_X:%.*]] = getelementptr nusw i8, ptr [[P:%.*]], i64 [[X_EXT]]
+; CHECK-NEXT:    [[GEP_X1:%.*]] = getelementptr i8, ptr [[GEP_X]], i64 1
+; CHECK-NEXT:    [[GEP_Y:%.*]] = getelementptr nusw i8, ptr [[P]], i64 [[Y_EXT]]
+; CHECK-NEXT:    [[CMP2:%.*]] = icmp ugt ptr [[GEP_X1]], [[GEP_Y]]
+; CHECK-NEXT:    ret i1 [[CMP2]]
+;
+  %x.ext = zext i32 %x to i64
+  %y.ext = zext i32 %y to i64
+  %cmp1 = icmp ugt i64 %x.ext, %y.ext
+  call void @llvm.assume(i1 %cmp1)
+  %gep.x = getelementptr nusw i8, ptr %p, i64 %x.ext
+  %gep.x1 = getelementptr i8, ptr %gep.x, i64 1
+  %gep.y = getelementptr nusw i8, ptr %p, i64 %y.ext
+  %cmp2 = icmp ugt ptr %gep.x1, %gep.y
+  ret i1 %cmp2
+}