[APInt] Assert correct values in APInt constructor

If the uint64_t constructor is used, assert that the value is
actuall a signed or unsigned N-bit integer depending on whether
the isSigned flag is set.

Currently, we allow values to be silently truncated, which is
a constant source of subtle bugs -- a particularly common mistake
is to create -1 values without setting the isSigned flag, which
will work fine for all common bit widths (<= 64-bit) and miscompile
for larger integers.

llvm/include/llvm/ADT/APFixedPoint.h

@@ -160,7 +160,9 @@ class APFixedPoint {
   }
 
   APFixedPoint(uint64_t Val, const FixedPointSemantics &Sema)
-      : APFixedPoint(APInt(Sema.getWidth(), Val, Sema.isSigned()), Sema) {}
+      : APFixedPoint(APInt(Sema.getWidth(), Val, Sema.isSigned(),
+                           /*implicitTrunc*/ true),
+                     Sema) {}
 
   // Zero initialization.
   APFixedPoint(const FixedPointSemantics &Sema) : APFixedPoint(0, Sema) {}

llvm/include/llvm/ADT/APInt.h

@@ -106,11 +106,26 @@ class [[nodiscard]] APInt {
   /// \param numBits the bit width of the constructed APInt
   /// \param val the initial value of the APInt
   /// \param isSigned how to treat signedness of val
-  APInt(unsigned numBits, uint64_t val, bool isSigned = false)
+  /// \param implicitTrunc allow implicit truncation of non-zero/sign bits of
+  ///                      val beyond the range of numBits
+  APInt(unsigned numBits, uint64_t val, bool isSigned = false,
+        bool implicitTrunc = false)
       : BitWidth(numBits) {
+    if (!implicitTrunc) {
+      if (BitWidth == 0) {
+        assert(val == 0 && "Value must be zero for 0-bit APInt");
+      } else if (isSigned) {
+        assert(llvm::isIntN(BitWidth, val) &&
+               "Value is not an N-bit signed value");
+      } else {
+        assert(llvm::isUIntN(BitWidth, val) &&
+               "Value is not an N-bit unsigned value");
+      }
+    }
     if (isSingleWord()) {
       U.VAL = val;
-      clearUnusedBits();
+      if (implicitTrunc || isSigned)
+        clearUnusedBits();
     } else {
       initSlowCase(val, isSigned);
     }

llvm/lib/Analysis/ConstantFolding.cpp

@@ -889,7 +889,8 @@ Constant *SymbolicallyEvaluateGEP(const GEPOperator *GEP,
   APInt Offset = APInt(
       BitWidth,
       DL.getIndexedOffsetInType(
-          SrcElemTy, ArrayRef((Value *const *)Ops.data() + 1, Ops.size() - 1)));
+          SrcElemTy, ArrayRef((Value *const *)Ops.data() + 1, Ops.size() - 1)),
+      /*isSigned*/ true, /*implicitTrunc*/ true);
 
   // If this is a GEP of a GEP, fold it all into a single GEP.
   while (auto *GEP = dyn_cast<GEPOperator>(Ptr)) {
@@ -3322,8 +3323,9 @@ ConstantFoldScalarFrexpCall(Constant *Op, Type *IntTy) {
 
   // The exponent is an "unspecified value" for inf/nan. We use zero to avoid
   // using undef.
-  Constant *Result1 = FrexpMant.isFinite() ? ConstantInt::get(IntTy, FrexpExp)
-                                           : ConstantInt::getNullValue(IntTy);
+  Constant *Result1 = FrexpMant.isFinite()
+                          ? ConstantInt::getSigned(IntTy, FrexpExp)
+                          : ConstantInt::getNullValue(IntTy);
   return {Result0, Result1};
 }
 

llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp

@@ -4542,7 +4542,7 @@ OpenMPIRBuilder::createTargetInit(const LocationDescription &Loc, bool IsSPMD,
       Builder.CreateCall(Fn, {KernelEnvironment, KernelLaunchEnvironment});
 
   Value *ExecUserCode = Builder.CreateICmpEQ(
-      ThreadKind, ConstantInt::get(ThreadKind->getType(), -1),
+      ThreadKind, Constant::getAllOnesValue(ThreadKind->getType()),
       "exec_user_code");
 
   // ThreadKind = __kmpc_target_init(...)

llvm/lib/FuzzMutate/OpDescriptor.cpp

@@ -17,7 +17,8 @@ void fuzzerop::makeConstantsWithType(Type *T, std::vector<Constant *> &Cs) {
     uint64_t W = IntTy->getBitWidth();
     Cs.push_back(ConstantInt::get(IntTy, 0));
     Cs.push_back(ConstantInt::get(IntTy, 1));
-    Cs.push_back(ConstantInt::get(IntTy, 42));
+    Cs.push_back(ConstantInt::get(
+        IntTy, APInt(W, 42, /*isSigned*/ false, /*implicitTrunc*/ true)));
     Cs.push_back(ConstantInt::get(IntTy, APInt::getMaxValue(W)));
     Cs.push_back(ConstantInt::get(IntTy, APInt::getMinValue(W)));
     Cs.push_back(ConstantInt::get(IntTy, APInt::getSignedMaxValue(W)));

llvm/lib/IR/ConstantRange.cpp

@@ -1743,7 +1743,7 @@ ConstantRange ConstantRange::ctlz(bool ZeroIsPoison) const {
   // Zero is either safe or not in the range. The output range is composed by
   // the result of countLeadingZero of the two extremes.
   return getNonEmpty(APInt(getBitWidth(), getUnsignedMax().countl_zero()),
-                     APInt(getBitWidth(), getUnsignedMin().countl_zero() + 1));
+                     APInt(getBitWidth(), getUnsignedMin().countl_zero()) + 1);
 }
 
 static ConstantRange getUnsignedCountTrailingZerosRange(const APInt &Lower,
@@ -1802,7 +1802,7 @@ ConstantRange ConstantRange::cttz(bool ZeroIsPoison) const {
   }
 
   if (isFullSet())
-    return getNonEmpty(Zero, APInt(BitWidth, BitWidth + 1));
+    return getNonEmpty(Zero, APInt(BitWidth, BitWidth) + 1);
   if (!isWrappedSet())
     return getUnsignedCountTrailingZerosRange(Lower, Upper);
   // The range is wrapped. We decompose it into two ranges, [0, Upper) and
@@ -1847,7 +1847,7 @@ ConstantRange ConstantRange::ctpop() const {
   unsigned BitWidth = getBitWidth();
   APInt Zero = APInt::getZero(BitWidth);
   if (isFullSet())
-    return getNonEmpty(Zero, APInt(BitWidth, BitWidth + 1));
+    return getNonEmpty(Zero, APInt(BitWidth, BitWidth) + 1);
   if (!isWrappedSet())
     return getUnsignedPopCountRange(Lower, Upper);
   // The range is wrapped. We decompose it into two ranges, [0, Upper) and

llvm/lib/Support/APInt.cpp

@@ -234,7 +234,8 @@ APInt& APInt::operator-=(uint64_t RHS) {
 APInt APInt::operator*(const APInt& RHS) const {
   assert(BitWidth == RHS.BitWidth && "Bit widths must be the same");
   if (isSingleWord())
-    return APInt(BitWidth, U.VAL * RHS.U.VAL);
+    return APInt(BitWidth, U.VAL * RHS.U.VAL, /*isSigned*/ false,
+                 /*implicitTrunc*/ true);
 
   APInt Result(getMemory(getNumWords()), getBitWidth());
   tcMultiply(Result.U.pVal, U.pVal, RHS.U.pVal, getNumWords());
@@ -455,15 +456,17 @@ APInt APInt::extractBits(unsigned numBits, unsigned bitPosition) const {
          "Illegal bit extraction");
 
   if (isSingleWord())
-    return APInt(numBits, U.VAL >> bitPosition);
+    return APInt(numBits, U.VAL >> bitPosition, /*isSigned*/ false,
+                 /*implicitTrunc*/ true);
 
   unsigned loBit = whichBit(bitPosition);
   unsigned loWord = whichWord(bitPosition);
   unsigned hiWord = whichWord(bitPosition + numBits - 1);
 
   // Single word result extracting bits from a single word source.
   if (loWord == hiWord)
-    return APInt(numBits, U.pVal[loWord] >> loBit);
+    return APInt(numBits, U.pVal[loWord] >> loBit, /*isSigned*/ false,
+                 /*implicitTrunc*/ true);
 
   // Extracting bits that start on a source word boundary can be done
   // as a fast memory copy.
@@ -907,7 +910,8 @@ APInt APInt::trunc(unsigned width) const {
   assert(width <= BitWidth && "Invalid APInt Truncate request");
 
   if (width <= APINT_BITS_PER_WORD)
-    return APInt(width, getRawData()[0]);
+    return APInt(width, getRawData()[0], /*isSigned*/ false,
+                 /*implicitTrunc*/ true);
 
   if (width == BitWidth)
     return *this;
@@ -955,7 +959,7 @@ APInt APInt::sext(unsigned Width) const {
   assert(Width >= BitWidth && "Invalid APInt SignExtend request");
 
   if (Width <= APINT_BITS_PER_WORD)
-    return APInt(Width, SignExtend64(U.VAL, BitWidth));
+    return APInt(Width, SignExtend64(U.VAL, BitWidth), /*isSigned*/ true);
 
   if (Width == BitWidth)
     return *this;

llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp

@@ -308,7 +308,7 @@ Instruction *InstCombinerImpl::foldCmpLoadFromIndexedGlobal(
       DL.getTypeAllocSize(Init->getType()->getArrayElementType());
   auto MaskIdx = [&](Value *Idx) {
     if (!GEP->isInBounds() && llvm::countr_zero(ElementSize) != 0) {
-      Value *Mask = ConstantInt::get(Idx->getType(), -1);
+      Value *Mask = Constant::getAllOnesValue(Idx->getType());
       Mask = Builder.CreateLShr(Mask, llvm::countr_zero(ElementSize));
       Idx = Builder.CreateAnd(Idx, Mask);
     }

llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp

@@ -665,11 +665,11 @@ static Value *foldSelectICmpLshrAshr(const ICmpInst *IC, Value *TrueVal,
   Value *X, *Y;
   unsigned Bitwidth = CmpRHS->getType()->getScalarSizeInBits();
   if ((Pred != ICmpInst::ICMP_SGT ||
-       !match(CmpRHS,
-              m_SpecificInt_ICMP(ICmpInst::ICMP_SGE, APInt(Bitwidth, -1)))) &&
+       !match(CmpRHS, m_SpecificInt_ICMP(ICmpInst::ICMP_SGE,
+                                         APInt::getAllOnes(Bitwidth)))) &&
       (Pred != ICmpInst::ICMP_SLT ||
-       !match(CmpRHS,
-              m_SpecificInt_ICMP(ICmpInst::ICMP_SGE, APInt(Bitwidth, 0)))))
+       !match(CmpRHS, m_SpecificInt_ICMP(ICmpInst::ICMP_SGE,
+                                         APInt::getZero(Bitwidth)))))
     return nullptr;
 
   // Canonicalize so that ashr is in FalseVal.

llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp

@@ -224,7 +224,7 @@ static Value *convertStrToInt(CallInst *CI, StringRef &Str, Value *EndPtr,
     // Unsigned negation doesn't overflow.
     Result = -Result;
 
-  return ConstantInt::get(RetTy, Result);
+  return ConstantInt::get(RetTy, Result, AsSigned);
 }
 
 static bool isOnlyUsedInComparisonWithZero(Value *V) {
@@ -553,7 +553,8 @@ Value *LibCallSimplifier::optimizeStrCmp(CallInst *CI, IRBuilderBase &B) {
   // strcmp(x, y)  -> cnst  (if both x and y are constant strings)
   if (HasStr1 && HasStr2)
     return ConstantInt::get(CI->getType(),
-                            std::clamp(Str1.compare(Str2), -1, 1));
+                            std::clamp(Str1.compare(Str2), -1, 1),
+                            /*isSigned*/ true);
 
   if (HasStr1 && Str1.empty()) // strcmp("", x) -> -*x
     return B.CreateNeg(B.CreateZExt(
@@ -638,7 +639,8 @@ Value *LibCallSimplifier::optimizeStrNCmp(CallInst *CI, IRBuilderBase &B) {
     StringRef SubStr1 = substr(Str1, Length);
     StringRef SubStr2 = substr(Str2, Length);
     return ConstantInt::get(CI->getType(),
-                            std::clamp(SubStr1.compare(SubStr2), -1, 1));
+                            std::clamp(SubStr1.compare(SubStr2), -1, 1),
+                            /*isSigned*/ true);
   }
 
   if (HasStr1 && Str1.empty()) // strncmp("", x, n) -> -*x
@@ -1518,7 +1520,7 @@ static Value *optimizeMemCmpVarSize(CallInst *CI, Value *LHS, Value *RHS,
   int IRes = UChar(LStr[Pos]) < UChar(RStr[Pos]) ? -1 : 1;
   Value *MaxSize = ConstantInt::get(Size->getType(), Pos);
   Value *Cmp = B.CreateICmp(ICmpInst::ICMP_ULE, Size, MaxSize);
-  Value *Res = ConstantInt::get(CI->getType(), IRes);
+  Value *Res = ConstantInt::get(CI->getType(), IRes, /*isSigned*/ true);
   return B.CreateSelect(Cmp, Zero, Res);
 }
 

llvm/unittests/ADT/APFixedPointTest.cpp

@@ -240,27 +240,28 @@ void CheckIntPart(const FixedPointSemantics &Sema, int64_t IntPart) {
   APFixedPoint ValWithFract(
       APInt(Sema.getWidth(),
             relativeShr(IntPart, Sema.getLsbWeight()) + FullFactPart,
-            Sema.isSigned()),
+            Sema.isSigned(), /*implicitTrunc*/ true),
       Sema);
   ASSERT_EQ(ValWithFract.getIntPart(), IntPart);
 
   // Just fraction
-  APFixedPoint JustFract(APInt(Sema.getWidth(), FullFactPart, Sema.isSigned()),
+  APFixedPoint JustFract(APInt(Sema.getWidth(), FullFactPart, Sema.isSigned(),
+                               /*implicitTrunc*/ true),
                          Sema);
   ASSERT_EQ(JustFract.getIntPart(), 0);
 
   // Whole number
   APFixedPoint WholeNum(APInt(Sema.getWidth(),
                               relativeShr(IntPart, Sema.getLsbWeight()),
-                              Sema.isSigned()),
+                              Sema.isSigned(), /*implicitTrunc*/ true),
                         Sema);
   ASSERT_EQ(WholeNum.getIntPart(), IntPart);
 
   // Negative
   if (Sema.isSigned()) {
     APFixedPoint Negative(APInt(Sema.getWidth(),
                                 relativeShr(IntPart, Sema.getLsbWeight()),
-                                Sema.isSigned()),
+                                Sema.isSigned(), /*implicitTrunc*/ true),
                           Sema);
     ASSERT_EQ(Negative.getIntPart(), IntPart);
   }