[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.

Author: nikic

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

@@ -107,11 +107,26 @@ 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);
 
   std::optional<ConstantRange> InRange = GEP->getInRange();
   if (InRange)
@@ -3401,8 +3402,9 @@ 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/Analysis/MemoryBuiltins.cpp

@@ -675,7 +675,7 @@ Value *llvm::lowerObjectSizeCall(
   if (!MustSucceed)
     return nullptr;
 
-  return ConstantInt::get(ResultType, MaxVal ? -1ULL : 0);
+  return ConstantInt::get(ResultType, MaxVal ? -1ULL : 0, true);
 }
 
 STATISTIC(ObjectVisitorArgument,

llvm/lib/Analysis/ScalarEvolution.cpp

@@ -1460,7 +1460,7 @@ bool ScalarEvolution::proveNoWrapByVaryingStart(const SCEV *Start,
 
   APInt StartAI = StartC->getAPInt();
 
-  for (unsigned Delta : {-2, -1, 1, 2}) {
+  for (int Delta : {-2, -1, 1, 2}) {
     const SCEV *PreStart = getConstant(StartAI - Delta);
 
     FoldingSetNodeID ID;
@@ -1475,7 +1475,7 @@ bool ScalarEvolution::proveNoWrapByVaryingStart(const SCEV *Start,
     // Give up if we don't already have the add recurrence we need because
     // actually constructing an add recurrence is relatively expensive.
     if (PreAR && PreAR->getNoWrapFlags(WrapType)) {  // proves (2)
-      const SCEV *DeltaS = getConstant(StartC->getType(), Delta);
+      const SCEV *DeltaS = getConstant(StartC->getType(), Delta, true);
       ICmpInst::Predicate Pred = ICmpInst::BAD_ICMP_PREDICATE;
       const SCEV *Limit = ExtendOpTraits<ExtendOpTy>::getOverflowLimitForStep(
           DeltaS, &Pred, this);

llvm/lib/Analysis/ValueTracking.cpp

@@ -9303,7 +9303,7 @@ static ConstantRange getRangeForIntrinsic(const IntrinsicInst &II) {
   case Intrinsic::cttz:
     // Maximum of set/clear bits is the bit width.
     return ConstantRange::getNonEmpty(APInt::getZero(Width),
-                                      APInt(Width, Width + 1));
+                                      APInt(Width, Width) + 1);
   case Intrinsic::uadd_sat:
     // uadd.sat(x, C) produces [C, UINT_MAX].
     if (match(II.getOperand(0), m_APInt(C)) ||
@@ -9454,7 +9454,7 @@ static void setLimitForFPToI(const Instruction *I, APInt &Lower, APInt &Upper) {
   if (!I->getOperand(0)->getType()->getScalarType()->isHalfTy())
     return;
   if (isa<FPToSIInst>(I) && BitWidth >= 17) {
-    Lower = APInt(BitWidth, -65504);
+    Lower = APInt(BitWidth, -65504, true);
     Upper = APInt(BitWidth, 65505);
   }
 

llvm/lib/Bitcode/Reader/BitcodeReader.cpp

@@ -3147,7 +3147,7 @@ Error BitcodeReader::parseConstants() {
     case bitc::CST_CODE_INTEGER:   // INTEGER: [intval]
       if (!CurTy->isIntOrIntVectorTy() || Record.empty())
         return error("Invalid integer const record");
-      V = ConstantInt::get(CurTy, decodeSignRotatedValue(Record[0]));
+      V = ConstantInt::getSigned(CurTy, decodeSignRotatedValue(Record[0]));
       break;
     case bitc::CST_CODE_WIDE_INTEGER: {// WIDE_INTEGER: [n x intval]
       if (!CurTy->isIntOrIntVectorTy() || Record.empty())

llvm/lib/CodeGen/CodeGenPrepare.cpp

@@ -1643,7 +1643,7 @@ static bool matchUAddWithOverflowConstantEdgeCases(CmpInst *Cmp,
   if (Pred == ICmpInst::ICMP_EQ && match(B, m_AllOnes()))
     B = ConstantInt::get(B->getType(), 1);
   else if (Pred == ICmpInst::ICMP_NE && match(B, m_ZeroInt()))
-    B = ConstantInt::get(B->getType(), -1);
+    B = Constant::getAllOnesValue(B->getType());
   else
     return false;
 

llvm/lib/CodeGen/ExpandMemCmp.cpp

@@ -590,7 +590,7 @@ void MemCmpExpansion::emitMemCmpResultBlock() {
                                   ResBlock.PhiSrc2);
 
   Value *Res =
-      Builder.CreateSelect(Cmp, ConstantInt::get(Builder.getInt32Ty(), -1),
+      Builder.CreateSelect(Cmp, Constant::getAllOnesValue(Builder.getInt32Ty()),
                            ConstantInt::get(Builder.getInt32Ty(), 1));
 
   PhiRes->addIncoming(Res, ResBlock.BB);

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

@@ -1625,7 +1625,10 @@ SDValue SelectionDAG::getConstant(uint64_t Val, const SDLoc &DL, EVT VT,
   assert((EltVT.getSizeInBits() >= 64 ||
           (uint64_t)((int64_t)Val >> EltVT.getSizeInBits()) + 1 < 2) &&
          "getConstant with a uint64_t value that doesn't fit in the type!");
-  return getConstant(APInt(EltVT.getSizeInBits(), Val), DL, VT, isT, isO);
+  // TODO: Avoid implicit trunc?
+  return getConstant(
+      APInt(EltVT.getSizeInBits(), Val, false, /*implicitTrunc*/ true), DL, VT,
+      isT, isO);
 }
 
 SDValue SelectionDAG::getConstant(const APInt &Val, const SDLoc &DL, EVT VT,