[ValueTracking] Introduce a KnownBits struct to wrap the two APInts for computeKnownBits
authorCraig Topper <craig.topper@gmail.com>
Wed, 26 Apr 2017 16:39:58 +0000 (16:39 +0000)
committerCraig Topper <craig.topper@gmail.com>
Wed, 26 Apr 2017 16:39:58 +0000 (16:39 +0000)
This patch introduces a new KnownBits struct that wraps the two APInt used by computeKnownBits. This allows us to treat them as more of a unit.

Initially I've just altered the signatures of computeKnownBits and InstCombine's simplifyDemandedBits to pass a KnownBits reference instead of two separate APInt references. I'll do similar to the SelectionDAG version of computeKnownBits/simplifyDemandedBits as a separate patch.

I've added a constructor that allows initializing both APInts to the same bit width with a starting value of 0. This reduces the repeated pattern of initializing both APInts. Once place default constructed the APInts so I added a default constructor for those cases.

Going forward I would like to add more methods that will work on the pairs. For example trunc, zext, and sext occur on both APInts together in several places. We should probably add a clear method that can be used to clear both pieces. Maybe a method to check for conflicting information. A method to return (Zero|One) so we don't write it out everywhere. Maybe a method for (Zero|One).isAllOnesValue() to determine if all bits are known. I'm sure there are many other methods we can come up with.

Differential Revision: https://reviews.llvm.org/D32376

llvm-svn: 301432

25 files changed:
llvm/include/llvm/Analysis/DemandedBits.h
llvm/include/llvm/Analysis/ValueTracking.h
llvm/include/llvm/Support/KnownBits.h [new file with mode: 0644]
llvm/lib/Analysis/ConstantFolding.cpp
llvm/lib/Analysis/DemandedBits.cpp
llvm/lib/Analysis/InstructionSimplify.cpp
llvm/lib/Analysis/Lint.cpp
llvm/lib/Analysis/ScalarEvolution.cpp
llvm/lib/Analysis/ValueTracking.cpp
llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
llvm/lib/Target/Hexagon/HexagonLoopIdiomRecognition.cpp
llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
llvm/lib/Transforms/InstCombine/InstCombineInternal.h
llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
llvm/lib/Transforms/Scalar/GuardWidening.cpp
llvm/lib/Transforms/Utils/BypassSlowDivision.cpp
llvm/lib/Transforms/Utils/Local.cpp
llvm/lib/Transforms/Utils/SimplifyCFG.cpp
llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp
llvm/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp

index c603274..e5fd8a0 100644 (file)
@@ -35,6 +35,7 @@ class Function;
 class Instruction;
 class DominatorTree;
 class AssumptionCache;
+struct KnownBits;
 
 class DemandedBits {
 public:
@@ -58,8 +59,7 @@ private:
   void determineLiveOperandBits(const Instruction *UserI,
     const Instruction *I, unsigned OperandNo,
     const APInt &AOut, APInt &AB,
-    APInt &KnownZero, APInt &KnownOne,
-    APInt &KnownZero2, APInt &KnownOne2);
+    KnownBits &Known, KnownBits &Known2);
 
   bool Analyzed;
 
index e3c2f3b..764308d 100644 (file)
@@ -29,6 +29,7 @@ template <typename T> class ArrayRef;
   class DominatorTree;
   class GEPOperator;
   class Instruction;
+  struct KnownBits;
   class Loop;
   class LoopInfo;
   class OptimizationRemarkEmitter;
@@ -49,7 +50,7 @@ template <typename T> class ArrayRef;
   /// where V is a vector, the known zero and known one values are the
   /// same width as the vector element, and the bit is set only if it is true
   /// for all of the elements in the vector.
-  void computeKnownBits(const Value *V, APInt &KnownZero, APInt &KnownOne,
+  void computeKnownBits(const Value *V, KnownBits &Known,
                         const DataLayout &DL, unsigned Depth = 0,
                         AssumptionCache *AC = nullptr,
                         const Instruction *CxtI = nullptr,
diff --git a/llvm/include/llvm/Support/KnownBits.h b/llvm/include/llvm/Support/KnownBits.h
new file mode 100644 (file)
index 0000000..08d4ded
--- /dev/null
@@ -0,0 +1,43 @@
+//===- llvm/Support/KnownBits.h - Stores known zeros/ones -------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains a class for representing known zeros and ones used by
+// computeKnownBits.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_SUPPORT_KNOWNBITS_H
+#define LLVM_SUPPORT_KNOWNBITS_H
+
+#include "llvm/ADT/APInt.h"
+
+namespace llvm {
+
+// For now this is a simple wrapper around two APInts.
+struct KnownBits {
+  APInt Zero;
+  APInt One;
+
+  // Default construct Zero and One.
+  KnownBits() {}
+
+  /// Create a known bits object of BitWidth bits initialized to unknown.
+  KnownBits(unsigned BitWidth) : Zero(BitWidth, 0), One(BitWidth, 0) {}
+
+  /// Get the bit width of this value.
+  unsigned getBitWidth() const {
+    assert(Zero.getBitWidth() == One.getBitWidth() &&
+           "Zero and One should have the same width!");
+    return Zero.getBitWidth();
+  }
+};
+
+} // end namespace llvm
+
+#endif
index 14176da..863fbdb 100644 (file)
@@ -42,6 +42,7 @@
 #include "llvm/IR/Value.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Support/MathExtras.h"
 #include <cassert>
 #include <cerrno>
@@ -687,21 +688,21 @@ Constant *SymbolicallyEvaluateBinop(unsigned Opc, Constant *Op0, Constant *Op1,
 
   if (Opc == Instruction::And) {
     unsigned BitWidth = DL.getTypeSizeInBits(Op0->getType()->getScalarType());
-    APInt KnownZero0(BitWidth, 0), KnownOne0(BitWidth, 0);
-    APInt KnownZero1(BitWidth, 0), KnownOne1(BitWidth, 0);
-    computeKnownBits(Op0, KnownZero0, KnownOne0, DL);
-    computeKnownBits(Op1, KnownZero1, KnownOne1, DL);
-    if ((KnownOne1 | KnownZero0).isAllOnesValue()) {
+    KnownBits Known0(BitWidth);
+    KnownBits Known1(BitWidth);
+    computeKnownBits(Op0, Known0, DL);
+    computeKnownBits(Op1, Known1, DL);
+    if ((Known1.One | Known0.Zero).isAllOnesValue()) {
       // All the bits of Op0 that the 'and' could be masking are already zero.
       return Op0;
     }
-    if ((KnownOne0 | KnownZero1).isAllOnesValue()) {
+    if ((Known0.One | Known1.Zero).isAllOnesValue()) {
       // All the bits of Op1 that the 'and' could be masking are already zero.
       return Op1;
     }
 
-    APInt KnownZero = KnownZero0 | KnownZero1;
-    APInt KnownOne = KnownOne0 & KnownOne1;
+    APInt KnownZero = Known0.Zero | Known1.Zero;
+    APInt KnownOne = Known0.One & Known1.One;
     if ((KnownZero | KnownOne).isAllOnesValue()) {
       return ConstantInt::get(Op0->getType(), KnownOne);
     }
index 151c0b0..285339d 100644 (file)
@@ -37,6 +37,7 @@
 #include "llvm/IR/Operator.h"
 #include "llvm/Pass.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Support/raw_ostream.h"
 using namespace llvm;
 
@@ -72,8 +73,7 @@ static bool isAlwaysLive(Instruction *I) {
 
 void DemandedBits::determineLiveOperandBits(
     const Instruction *UserI, const Instruction *I, unsigned OperandNo,
-    const APInt &AOut, APInt &AB, APInt &KnownZero, APInt &KnownOne,
-    APInt &KnownZero2, APInt &KnownOne2) {
+    const APInt &AOut, APInt &AB, KnownBits &Known, KnownBits &Known2) {
   unsigned BitWidth = AB.getBitWidth();
 
   // We're called once per operand, but for some instructions, we need to
@@ -85,15 +85,13 @@ void DemandedBits::determineLiveOperandBits(
   auto ComputeKnownBits =
       [&](unsigned BitWidth, const Value *V1, const Value *V2) {
         const DataLayout &DL = I->getModule()->getDataLayout();
-        KnownZero = APInt(BitWidth, 0);
-        KnownOne = APInt(BitWidth, 0);
-        computeKnownBits(const_cast<Value *>(V1), KnownZero, KnownOne, DL, 0,
+        Known = KnownBits(BitWidth);
+        computeKnownBits(const_cast<Value *>(V1), Known, DL, 0,
                          &AC, UserI, &DT);
 
         if (V2) {
-          KnownZero2 = APInt(BitWidth, 0);
-          KnownOne2 = APInt(BitWidth, 0);
-          computeKnownBits(const_cast<Value *>(V2), KnownZero2, KnownOne2, DL,
+          Known2 = KnownBits(BitWidth);
+          computeKnownBits(const_cast<Value *>(V2), Known2, DL,
                            0, &AC, UserI, &DT);
         }
       };
@@ -120,7 +118,7 @@ void DemandedBits::determineLiveOperandBits(
           // known to be one.
           ComputeKnownBits(BitWidth, I, nullptr);
           AB = APInt::getHighBitsSet(BitWidth,
-                 std::min(BitWidth, KnownOne.countLeadingZeros()+1));
+                 std::min(BitWidth, Known.One.countLeadingZeros()+1));
         }
         break;
       case Intrinsic::cttz:
@@ -130,7 +128,7 @@ void DemandedBits::determineLiveOperandBits(
           // known to be one.
           ComputeKnownBits(BitWidth, I, nullptr);
           AB = APInt::getLowBitsSet(BitWidth,
-                 std::min(BitWidth, KnownOne.countTrailingZeros()+1));
+                 std::min(BitWidth, Known.One.countTrailingZeros()+1));
         }
         break;
       }
@@ -200,11 +198,11 @@ void DemandedBits::determineLiveOperandBits(
     // dead).
     if (OperandNo == 0) {
       ComputeKnownBits(BitWidth, I, UserI->getOperand(1));
-      AB &= ~KnownZero2;
+      AB &= ~Known2.Zero;
     } else {
       if (!isa<Instruction>(UserI->getOperand(0)))
         ComputeKnownBits(BitWidth, UserI->getOperand(0), I);
-      AB &= ~(KnownZero & ~KnownZero2);
+      AB &= ~(Known.Zero & ~Known2.Zero);
     }
     break;
   case Instruction::Or:
@@ -216,11 +214,11 @@ void DemandedBits::determineLiveOperandBits(
     // dead).
     if (OperandNo == 0) {
       ComputeKnownBits(BitWidth, I, UserI->getOperand(1));
-      AB &= ~KnownOne2;
+      AB &= ~Known2.One;
     } else {
       if (!isa<Instruction>(UserI->getOperand(0)))
         ComputeKnownBits(BitWidth, UserI->getOperand(0), I);
-      AB &= ~(KnownOne & ~KnownOne2);
+      AB &= ~(Known.One & ~Known2.One);
     }
     break;
   case Instruction::Xor:
@@ -318,7 +316,7 @@ void DemandedBits::performAnalysis() {
     if (!UserI->getType()->isIntegerTy())
       Visited.insert(UserI);
 
-    APInt KnownZero, KnownOne, KnownZero2, KnownOne2;
+    KnownBits Known, Known2;
     // Compute the set of alive bits for each operand. These are anded into the
     // existing set, if any, and if that changes the set of alive bits, the
     // operand is added to the work-list.
@@ -335,8 +333,7 @@ void DemandedBits::performAnalysis() {
             // Bits of each operand that are used to compute alive bits of the
             // output are alive, all others are dead.
             determineLiveOperandBits(UserI, I, OI.getOperandNo(), AOut, AB,
-                                     KnownZero, KnownOne,
-                                     KnownZero2, KnownOne2);
+                                     Known, Known2);
           }
 
           // If we've added to the set of alive bits (or the operand has not
index 4c707b6..e720e3e 100644 (file)
@@ -35,6 +35,7 @@
 #include "llvm/IR/Operator.h"
 #include "llvm/IR/PatternMatch.h"
 #include "llvm/IR/ValueHandle.h"
+#include "llvm/Support/KnownBits.h"
 #include <algorithm>
 using namespace llvm;
 using namespace llvm::PatternMatch;
@@ -693,10 +694,9 @@ static Value *SimplifySubInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW,
       return Op0;
 
     unsigned BitWidth = Op1->getType()->getScalarSizeInBits();
-    APInt KnownZero(BitWidth, 0);
-    APInt KnownOne(BitWidth, 0);
-    computeKnownBits(Op1, KnownZero, KnownOne, Q.DL, 0, Q.AC, Q.CxtI, Q.DT);
-    if (KnownZero.isMaxSignedValue()) {
+    KnownBits Known(BitWidth);
+    computeKnownBits(Op1, Known, Q.DL, 0, Q.AC, Q.CxtI, Q.DT);
+    if (Known.Zero.isMaxSignedValue()) {
       // Op1 is either 0 or the minimum signed value. If the sub is NSW, then
       // Op1 must be 0 because negating the minimum signed value is undefined.
       if (isNSW)
@@ -1402,16 +1402,15 @@ static Value *SimplifyShift(Instruction::BinaryOps Opcode, Value *Op0,
   // If any bits in the shift amount make that value greater than or equal to
   // the number of bits in the type, the shift is undefined.
   unsigned BitWidth = Op1->getType()->getScalarSizeInBits();
-  APInt KnownZero(BitWidth, 0);
-  APInt KnownOne(BitWidth, 0);
-  computeKnownBits(Op1, KnownZero, KnownOne, Q.DL, 0, Q.AC, Q.CxtI, Q.DT);
-  if (KnownOne.getLimitedValue() >= BitWidth)
+  KnownBits Known(BitWidth);
+  computeKnownBits(Op1, Known, Q.DL, 0, Q.AC, Q.CxtI, Q.DT);
+  if (Known.One.getLimitedValue() >= BitWidth)
     return UndefValue::get(Op0->getType());
 
   // If all valid bits in the shift amount are known zero, the first operand is
   // unchanged.
   unsigned NumValidShiftBits = Log2_32_Ceil(BitWidth);
-  if (KnownZero.countTrailingOnes() >= NumValidShiftBits)
+  if (Known.Zero.countTrailingOnes() >= NumValidShiftBits)
     return Op0;
 
   return nullptr;
@@ -1437,11 +1436,9 @@ static Value *SimplifyRightShift(Instruction::BinaryOps Opcode, Value *Op0,
   // The low bit cannot be shifted out of an exact shift if it is set.
   if (isExact) {
     unsigned BitWidth = Op0->getType()->getScalarSizeInBits();
-    APInt Op0KnownZero(BitWidth, 0);
-    APInt Op0KnownOne(BitWidth, 0);
-    computeKnownBits(Op0, Op0KnownZero, Op0KnownOne, Q.DL, /*Depth=*/0, Q.AC,
-                     Q.CxtI, Q.DT);
-    if (Op0KnownOne[0])
+    KnownBits Op0Known(BitWidth);
+    computeKnownBits(Op0, Op0Known, Q.DL, /*Depth=*/0, Q.AC, Q.CxtI, Q.DT);
+    if (Op0Known.One[0])
       return Op0;
   }
 
@@ -3427,11 +3424,10 @@ static Value *SimplifyICmpInst(unsigned Predicate, Value *LHS, Value *RHS,
     const APInt *RHSVal;
     if (match(RHS, m_APInt(RHSVal))) {
       unsigned BitWidth = RHSVal->getBitWidth();
-      APInt LHSKnownZero(BitWidth, 0);
-      APInt LHSKnownOne(BitWidth, 0);
-      computeKnownBits(LHS, LHSKnownZero, LHSKnownOne, Q.DL, /*Depth=*/0, Q.AC,
-                       Q.CxtI, Q.DT);
-      if (LHSKnownZero.intersects(*RHSVal) || !LHSKnownOne.isSubsetOf(*RHSVal))
+      KnownBits LHSKnown(BitWidth);
+      computeKnownBits(LHS, LHSKnown, Q.DL, /*Depth=*/0, Q.AC, Q.CxtI, Q.DT);
+      if (LHSKnown.Zero.intersects(*RHSVal) ||
+          !LHSKnown.One.isSubsetOf(*RHSVal))
         return Pred == ICmpInst::ICMP_EQ ? ConstantInt::getFalse(ITy)
                                          : ConstantInt::getTrue(ITy);
     }
@@ -4854,12 +4850,10 @@ Value *llvm::SimplifyInstruction(Instruction *I, const SimplifyQuery &Q,
   // value even when the operands are not all constants.
   if (!Result && I->getType()->isIntOrIntVectorTy()) {
     unsigned BitWidth = I->getType()->getScalarSizeInBits();
-    APInt KnownZero(BitWidth, 0);
-    APInt KnownOne(BitWidth, 0);
-    computeKnownBits(I, KnownZero, KnownOne, Q.DL, /*Depth*/ 0, Q.AC, I, Q.DT,
-                     ORE);
-    if ((KnownZero | KnownOne).isAllOnesValue())
-      Result = ConstantInt::get(I->getType(), KnownOne);
+    KnownBits Known(BitWidth);
+    computeKnownBits(I, Known, Q.DL, /*Depth*/ 0, Q.AC, I, Q.DT, ORE);
+    if ((Known.Zero | Known.One).isAllOnesValue())
+      Result = ConstantInt::get(I->getType(), Known.One);
   }
 
   /// If called on unreachable code, the above logic may report that the
index 2ca46b1..0f04af5 100644 (file)
@@ -70,6 +70,7 @@
 #include "llvm/Pass.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
 #include <cassert>
@@ -534,10 +535,9 @@ static bool isZero(Value *V, const DataLayout &DL, DominatorTree *DT,
   VectorType *VecTy = dyn_cast<VectorType>(V->getType());
   if (!VecTy) {
     unsigned BitWidth = V->getType()->getIntegerBitWidth();
-    APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
-    computeKnownBits(V, KnownZero, KnownOne, DL, 0, AC,
-                     dyn_cast<Instruction>(V), DT);
-    return KnownZero.isAllOnesValue();
+    KnownBits Known(BitWidth);
+    computeKnownBits(V, Known, DL, 0, AC, dyn_cast<Instruction>(V), DT);
+    return Known.Zero.isAllOnesValue();
   }
 
   // Per-component check doesn't work with zeroinitializer
@@ -556,9 +556,9 @@ static bool isZero(Value *V, const DataLayout &DL, DominatorTree *DT,
     if (isa<UndefValue>(Elem))
       return true;
 
-    APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
-    computeKnownBits(Elem, KnownZero, KnownOne, DL);
-    if (KnownZero.isAllOnesValue())
+    KnownBits Known(BitWidth);
+    computeKnownBits(Elem, Known, DL);
+    if (Known.Zero.isAllOnesValue())
       return true;
   }
 
index 1cc0554..3ac4bf1 100644 (file)
@@ -89,6 +89,7 @@
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Support/SaveAndRestore.h"
@@ -4575,10 +4576,10 @@ uint32_t ScalarEvolution::GetMinTrailingZerosImpl(const SCEV *S) {
   if (const SCEVUnknown *U = dyn_cast<SCEVUnknown>(S)) {
     // For a SCEVUnknown, ask ValueTracking.
     unsigned BitWidth = getTypeSizeInBits(U->getType());
-    APInt Zeros(BitWidth, 0), Ones(BitWidth, 0);
-    computeKnownBits(U->getValue(), Zeros, Ones, getDataLayout(), 0, &AC,
+    KnownBits Known(BitWidth);
+    computeKnownBits(U->getValue(), Known, getDataLayout(), 0, &AC,
                      nullptr, &DT);
-    return Zeros.countTrailingOnes();
+    return Known.Zero.countTrailingOnes();
   }
 
   // SCEVUDivExpr
@@ -4757,11 +4758,12 @@ ScalarEvolution::getRange(const SCEV *S,
     const DataLayout &DL = getDataLayout();
     if (SignHint == ScalarEvolution::HINT_RANGE_UNSIGNED) {
       // For a SCEVUnknown, ask ValueTracking.
-      APInt Zeros(BitWidth, 0), Ones(BitWidth, 0);
-      computeKnownBits(U->getValue(), Zeros, Ones, DL, 0, &AC, nullptr, &DT);
-      if (Ones != ~Zeros + 1)
+      KnownBits Known(BitWidth);
+      computeKnownBits(U->getValue(), Known, DL, 0, &AC, nullptr, &DT);
+      if (Known.One != ~Known.Zero + 1)
         ConservativeResult =
-            ConservativeResult.intersectWith(ConstantRange(Ones, ~Zeros + 1));
+            ConservativeResult.intersectWith(ConstantRange(Known.One,
+                                                           ~Known.Zero + 1));
     } else {
       assert(SignHint == ScalarEvolution::HINT_RANGE_SIGNED &&
              "generalize as needed!");
@@ -5292,13 +5294,13 @@ const SCEV *ScalarEvolution::createSCEV(Value *V) {
         unsigned LZ = A.countLeadingZeros();
         unsigned TZ = A.countTrailingZeros();
         unsigned BitWidth = A.getBitWidth();
-        APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
-        computeKnownBits(BO->LHS, KnownZero, KnownOne, getDataLayout(),
+        KnownBits Known(BitWidth);
+        computeKnownBits(BO->LHS, Known, getDataLayout(),
                          0, &AC, nullptr, &DT);
 
         APInt EffectiveMask =
             APInt::getLowBitsSet(BitWidth, BitWidth - LZ - TZ).shl(TZ);
-        if ((LZ != 0 || TZ != 0) && !((~A & ~KnownZero) & EffectiveMask)) {
+        if ((LZ != 0 || TZ != 0) && !((~A & ~Known.Zero) & EffectiveMask)) {
           const SCEV *MulCount = getConstant(APInt::getOneBitSet(BitWidth, TZ));
           const SCEV *LHS = getSCEV(BO->LHS);
           const SCEV *ShiftedLHS = nullptr;
index b39783d..af964b6 100644 (file)
@@ -38,6 +38,7 @@
 #include "llvm/IR/PatternMatch.h"
 #include "llvm/IR/Statepoint.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Support/MathExtras.h"
 #include <algorithm>
 #include <array>
@@ -130,15 +131,15 @@ static const Instruction *safeCxtI(const Value *V, const Instruction *CxtI) {
   return nullptr;
 }
 
-static void computeKnownBits(const Value *V, APInt &KnownZero, APInt &KnownOne,
+static void computeKnownBits(const Value *V, KnownBits &Known,
                              unsigned Depth, const Query &Q);
 
-void llvm::computeKnownBits(const Value *V, APInt &KnownZero, APInt &KnownOne,
+void llvm::computeKnownBits(const Value *V, KnownBits &Known,
                             const DataLayout &DL, unsigned Depth,
                             AssumptionCache *AC, const Instruction *CxtI,
                             const DominatorTree *DT,
                             OptimizationRemarkEmitter *ORE) {
-  ::computeKnownBits(V, KnownZero, KnownOne, Depth,
+  ::computeKnownBits(V, Known, Depth,
                      Query(DL, AC, safeCxtI(V, CxtI), DT, ORE));
 }
 
@@ -151,11 +152,11 @@ bool llvm::haveNoCommonBitsSet(const Value *LHS, const Value *RHS,
   assert(LHS->getType()->isIntOrIntVectorTy() &&
          "LHS and RHS should be integers");
   IntegerType *IT = cast<IntegerType>(LHS->getType()->getScalarType());
-  APInt LHSKnownZero(IT->getBitWidth(), 0), LHSKnownOne(IT->getBitWidth(), 0);
-  APInt RHSKnownZero(IT->getBitWidth(), 0), RHSKnownOne(IT->getBitWidth(), 0);
-  computeKnownBits(LHS, LHSKnownZero, LHSKnownOne, DL, 0, AC, CxtI, DT);
-  computeKnownBits(RHS, RHSKnownZero, RHSKnownOne, DL, 0, AC, CxtI, DT);
-  return (LHSKnownZero | RHSKnownZero).isAllOnesValue();
+  KnownBits LHSKnown(IT->getBitWidth());
+  KnownBits RHSKnown(IT->getBitWidth());
+  computeKnownBits(LHS, LHSKnown, DL, 0, AC, CxtI, DT);
+  computeKnownBits(RHS, RHSKnown, DL, 0, AC, CxtI, DT);
+  return (LHSKnown.Zero | RHSKnown.Zero).isAllOnesValue();
 }
 
 static void ComputeSignBit(const Value *V, bool &KnownZero, bool &KnownOne,
@@ -252,67 +253,65 @@ unsigned llvm::ComputeNumSignBits(const Value *V, const DataLayout &DL,
 
 static void computeKnownBitsAddSub(bool Add, const Value *Op0, const Value *Op1,
                                    bool NSW,
-                                   APInt &KnownZero, APInt &KnownOne,
-                                   APInt &KnownZero2, APInt &KnownOne2,
+                                   KnownBits &KnownOut, KnownBits &Known2,
                                    unsigned Depth, const Query &Q) {
-  unsigned BitWidth = KnownZero.getBitWidth();
+  unsigned BitWidth = KnownOut.getBitWidth();
 
   // If an initial sequence of bits in the result is not needed, the
   // corresponding bits in the operands are not needed.
-  APInt LHSKnownZero(BitWidth, 0), LHSKnownOne(BitWidth, 0);
-  computeKnownBits(Op0, LHSKnownZero, LHSKnownOne, Depth + 1, Q);
-  computeKnownBits(Op1, KnownZero2, KnownOne2, Depth + 1, Q);
+  KnownBits LHSKnown(BitWidth);
+  computeKnownBits(Op0, LHSKnown, Depth + 1, Q);
+  computeKnownBits(Op1, Known2, Depth + 1, Q);
 
   // Carry in a 1 for a subtract, rather than a 0.
   uint64_t CarryIn = 0;
   if (!Add) {
     // Sum = LHS + ~RHS + 1
-    std::swap(KnownZero2, KnownOne2);
+    std::swap(Known2.Zero, Known2.One);
     CarryIn = 1;
   }
 
-  APInt PossibleSumZero = ~LHSKnownZero + ~KnownZero2 + CarryIn;
-  APInt PossibleSumOne = LHSKnownOne + KnownOne2 + CarryIn;
+  APInt PossibleSumZero = ~LHSKnown.Zero + ~Known2.Zero + CarryIn;
+  APInt PossibleSumOne = LHSKnown.One + Known2.One + CarryIn;
 
   // Compute known bits of the carry.
-  APInt CarryKnownZero = ~(PossibleSumZero ^ LHSKnownZero ^ KnownZero2);
-  APInt CarryKnownOne = PossibleSumOne ^ LHSKnownOne ^ KnownOne2;
+  APInt CarryKnownZero = ~(PossibleSumZero ^ LHSKnown.Zero ^ Known2.Zero);
+  APInt CarryKnownOne = PossibleSumOne ^ LHSKnown.One ^ Known2.One;
 
   // Compute set of known bits (where all three relevant bits are known).
-  APInt LHSKnown = LHSKnownZero | LHSKnownOne;
-  APInt RHSKnown = KnownZero2 | KnownOne2;
-  APInt CarryKnown = CarryKnownZero | CarryKnownOne;
-  APInt Known = LHSKnown & RHSKnown & CarryKnown;
+  APInt LHSKnownUnion = LHSKnown.Zero | LHSKnown.One;
+  APInt RHSKnownUnion = Known2.Zero | Known2.One;
+  APInt CarryKnownUnion = CarryKnownZero | CarryKnownOne;
+  APInt Known = LHSKnownUnion & RHSKnownUnion & CarryKnownUnion;
 
   assert((PossibleSumZero & Known) == (PossibleSumOne & Known) &&
          "known bits of sum differ");
 
   // Compute known bits of the result.
-  KnownZero = ~PossibleSumOne & Known;
-  KnownOne = PossibleSumOne & Known;
+  KnownOut.Zero = ~PossibleSumOne & Known;
+  KnownOut.One = PossibleSumOne & Known;
 
   // Are we still trying to solve for the sign bit?
   if (!Known.isSignBitSet()) {
     if (NSW) {
       // Adding two non-negative numbers, or subtracting a negative number from
       // a non-negative one, can't wrap into negative.
-      if (LHSKnownZero.isSignBitSet() && KnownZero2.isSignBitSet())
-        KnownZero.setSignBit();
+      if (LHSKnown.Zero.isSignBitSet() && Known2.Zero.isSignBitSet())
+        KnownOut.Zero.setSignBit();
       // Adding two negative numbers, or subtracting a non-negative number from
       // a negative one, can't wrap into non-negative.
-      else if (LHSKnownOne.isSignBitSet() && KnownOne2.isSignBitSet())
-        KnownOne.setSignBit();
+      else if (LHSKnown.One.isSignBitSet() && Known2.One.isSignBitSet())
+        KnownOut.One.setSignBit();
     }
   }
 }
 
 static void computeKnownBitsMul(const Value *Op0, const Value *Op1, bool NSW,
-                                APInt &KnownZero, APInt &KnownOne,
-                                APInt &KnownZero2, APInt &KnownOne2,
+                                KnownBits &Known, KnownBits &Known2,
                                 unsigned Depth, const Query &Q) {
-  unsigned BitWidth = KnownZero.getBitWidth();
-  computeKnownBits(Op1, KnownZero, KnownOne, Depth + 1, Q);
-  computeKnownBits(Op0, KnownZero2, KnownOne2, Depth + 1, Q);
+  unsigned BitWidth = Known.getBitWidth();
+  computeKnownBits(Op1, Known, Depth + 1, Q);
+  computeKnownBits(Op0, Known2, Depth + 1, Q);
 
   bool isKnownNegative = false;
   bool isKnownNonNegative = false;
@@ -322,10 +321,10 @@ static void computeKnownBitsMul(const Value *Op0, const Value *Op1, bool NSW,
       // The product of a number with itself is non-negative.
       isKnownNonNegative = true;
     } else {
-      bool isKnownNonNegativeOp1 = KnownZero.isSignBitSet();
-      bool isKnownNonNegativeOp0 = KnownZero2.isSignBitSet();
-      bool isKnownNegativeOp1 = KnownOne.isSignBitSet();
-      bool isKnownNegativeOp0 = KnownOne2.isSignBitSet();
+      bool isKnownNonNegativeOp1 = Known.Zero.isSignBitSet();
+      bool isKnownNonNegativeOp0 = Known2.Zero.isSignBitSet();
+      bool isKnownNegativeOp1 = Known.One.isSignBitSet();
+      bool isKnownNegativeOp0 = Known2.One.isSignBitSet();
       // The product of two numbers with the same sign is non-negative.
       isKnownNonNegative = (isKnownNegativeOp1 && isKnownNegativeOp0) ||
         (isKnownNonNegativeOp1 && isKnownNonNegativeOp0);
@@ -343,28 +342,28 @@ static void computeKnownBitsMul(const Value *Op0, const Value *Op1, bool NSW,
   // Also compute a conservative estimate for high known-0 bits.
   // More trickiness is possible, but this is sufficient for the
   // interesting case of alignment computation.
-  KnownOne.clearAllBits();
-  unsigned TrailZ = KnownZero.countTrailingOnes() +
-                    KnownZero2.countTrailingOnes();
-  unsigned LeadZ =  std::max(KnownZero.countLeadingOnes() +
-                             KnownZero2.countLeadingOnes(),
+  Known.One.clearAllBits();
+  unsigned TrailZ = Known.Zero.countTrailingOnes() +
+                    Known2.Zero.countTrailingOnes();
+  unsigned LeadZ =  std::max(Known.Zero.countLeadingOnes() +
+                             Known2.Zero.countLeadingOnes(),
                              BitWidth) - BitWidth;
 
   TrailZ = std::min(TrailZ, BitWidth);
   LeadZ = std::min(LeadZ, BitWidth);
-  KnownZero.clearAllBits();
-  KnownZero.setLowBits(TrailZ);
-  KnownZero.setHighBits(LeadZ);
+  Known.Zero.clearAllBits();
+  Known.Zero.setLowBits(TrailZ);
+  Known.Zero.setHighBits(LeadZ);
 
   // Only make use of no-wrap flags if we failed to compute the sign bit
   // directly.  This matters if the multiplication always overflows, in
   // which case we prefer to follow the result of the direct computation,
   // though as the program is invoking undefined behaviour we can choose
   // whatever we like here.
-  if (isKnownNonNegative && !KnownOne.isSignBitSet())
-    KnownZero.setSignBit();
-  else if (isKnownNegative && !KnownZero.isSignBitSet())
-    KnownOne.setSignBit();
+  if (isKnownNonNegative && !Known.One.isSignBitSet())
+    Known.Zero.setSignBit();
+  else if (isKnownNegative && !Known.Zero.isSignBitSet())
+    Known.One.setSignBit();
 }
 
 void llvm::computeKnownBitsFromRangeMetadata(const MDNode &Ranges,
@@ -499,15 +498,14 @@ bool llvm::isValidAssumeForContext(const Instruction *Inv,
   return !isEphemeralValueOf(Inv, CxtI);
 }
 
-static void computeKnownBitsFromAssume(const Value *V, APInt &KnownZero,
-                                       APInt &KnownOne, unsigned Depth,
-                                       const Query &Q) {
+static void computeKnownBitsFromAssume(const Value *V, KnownBits &Known,
+                                       unsigned Depth, const Query &Q) {
   // Use of assumptions is context-sensitive. If we don't have a context, we
   // cannot use them!
   if (!Q.AC || !Q.CxtI)
     return;
 
-  unsigned BitWidth = KnownZero.getBitWidth();
+  unsigned BitWidth = Known.getBitWidth();
 
   // Note that the patterns below need to be kept in sync with the code
   // in AssumptionCache::updateAffectedValues.
@@ -532,15 +530,15 @@ static void computeKnownBitsFromAssume(const Value *V, APInt &KnownZero,
 
     if (Arg == V && isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
       assert(BitWidth == 1 && "assume operand is not i1?");
-      KnownZero.clearAllBits();
-      KnownOne.setAllBits();
+      Known.Zero.clearAllBits();
+      Known.One.setAllBits();
       return;
     }
     if (match(Arg, m_Not(m_Specific(V))) &&
         isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
       assert(BitWidth == 1 && "assume operand is not i1?");
-      KnownZero.setAllBits();
-      KnownOne.clearAllBits();
+      Known.Zero.setAllBits();
+      Known.One.clearAllBits();
       return;
     }
 
@@ -558,126 +556,126 @@ static void computeKnownBitsFromAssume(const Value *V, APInt &KnownZero,
     // assume(v = a)
     if (match(Arg, m_c_ICmp(Pred, m_V, m_Value(A))) &&
         Pred == ICmpInst::ICMP_EQ && isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
-      KnownZero |= RHSKnownZero;
-      KnownOne  |= RHSKnownOne;
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
+      Known.Zero |= RHSKnown.Zero;
+      Known.One  |= RHSKnown.One;
     // assume(v & b = a)
     } else if (match(Arg,
                      m_c_ICmp(Pred, m_c_And(m_V, m_Value(B)), m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
-      APInt MaskKnownZero(BitWidth, 0), MaskKnownOne(BitWidth, 0);
-      computeKnownBits(B, MaskKnownZero, MaskKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
+      KnownBits MaskKnown(BitWidth);
+      computeKnownBits(B, MaskKnown, Depth+1, Query(Q, I));
 
       // For those bits in the mask that are known to be one, we can propagate
       // known bits from the RHS to V.
-      KnownZero |= RHSKnownZero & MaskKnownOne;
-      KnownOne  |= RHSKnownOne  & MaskKnownOne;
+      Known.Zero |= RHSKnown.Zero & MaskKnown.One;
+      Known.One  |= RHSKnown.One  & MaskKnown.One;
     // assume(~(v & b) = a)
     } else if (match(Arg, m_c_ICmp(Pred, m_Not(m_c_And(m_V, m_Value(B))),
                                    m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
-      APInt MaskKnownZero(BitWidth, 0), MaskKnownOne(BitWidth, 0);
-      computeKnownBits(B, MaskKnownZero, MaskKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
+      KnownBits MaskKnown(BitWidth);
+      computeKnownBits(B, MaskKnown, Depth+1, Query(Q, I));
 
       // For those bits in the mask that are known to be one, we can propagate
       // inverted known bits from the RHS to V.
-      KnownZero |= RHSKnownOne  & MaskKnownOne;
-      KnownOne  |= RHSKnownZero & MaskKnownOne;
+      Known.Zero |= RHSKnown.One  & MaskKnown.One;
+      Known.One  |= RHSKnown.Zero & MaskKnown.One;
     // assume(v | b = a)
     } else if (match(Arg,
                      m_c_ICmp(Pred, m_c_Or(m_V, m_Value(B)), m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
-      APInt BKnownZero(BitWidth, 0), BKnownOne(BitWidth, 0);
-      computeKnownBits(B, BKnownZero, BKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
+      KnownBits BKnown(BitWidth);
+      computeKnownBits(B, BKnown, Depth+1, Query(Q, I));
 
       // For those bits in B that are known to be zero, we can propagate known
       // bits from the RHS to V.
-      KnownZero |= RHSKnownZero & BKnownZero;
-      KnownOne  |= RHSKnownOne  & BKnownZero;
+      Known.Zero |= RHSKnown.Zero & BKnown.Zero;
+      Known.One  |= RHSKnown.One  & BKnown.Zero;
     // assume(~(v | b) = a)
     } else if (match(Arg, m_c_ICmp(Pred, m_Not(m_c_Or(m_V, m_Value(B))),
                                    m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
-      APInt BKnownZero(BitWidth, 0), BKnownOne(BitWidth, 0);
-      computeKnownBits(B, BKnownZero, BKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
+      KnownBits BKnown(BitWidth);
+      computeKnownBits(B, BKnown, Depth+1, Query(Q, I));
 
       // For those bits in B that are known to be zero, we can propagate
       // inverted known bits from the RHS to V.
-      KnownZero |= RHSKnownOne  & BKnownZero;
-      KnownOne  |= RHSKnownZero & BKnownZero;
+      Known.Zero |= RHSKnown.One  & BKnown.Zero;
+      Known.One  |= RHSKnown.Zero & BKnown.Zero;
     // assume(v ^ b = a)
     } else if (match(Arg,
                      m_c_ICmp(Pred, m_c_Xor(m_V, m_Value(B)), m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
-      APInt BKnownZero(BitWidth, 0), BKnownOne(BitWidth, 0);
-      computeKnownBits(B, BKnownZero, BKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
+      KnownBits BKnown(BitWidth);
+      computeKnownBits(B, BKnown, Depth+1, Query(Q, I));
 
       // For those bits in B that are known to be zero, we can propagate known
       // bits from the RHS to V. For those bits in B that are known to be one,
       // we can propagate inverted known bits from the RHS to V.
-      KnownZero |= RHSKnownZero & BKnownZero;
-      KnownOne  |= RHSKnownOne  & BKnownZero;
-      KnownZero |= RHSKnownOne  & BKnownOne;
-      KnownOne  |= RHSKnownZero & BKnownOne;
+      Known.Zero |= RHSKnown.Zero & BKnown.Zero;
+      Known.One  |= RHSKnown.One  & BKnown.Zero;
+      Known.Zero |= RHSKnown.One  & BKnown.One;
+      Known.One  |= RHSKnown.Zero & BKnown.One;
     // assume(~(v ^ b) = a)
     } else if (match(Arg, m_c_ICmp(Pred, m_Not(m_c_Xor(m_V, m_Value(B))),
                                    m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
-      APInt BKnownZero(BitWidth, 0), BKnownOne(BitWidth, 0);
-      computeKnownBits(B, BKnownZero, BKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
+      KnownBits BKnown(BitWidth);
+      computeKnownBits(B, BKnown, Depth+1, Query(Q, I));
 
       // For those bits in B that are known to be zero, we can propagate
       // inverted known bits from the RHS to V. For those bits in B that are
       // known to be one, we can propagate known bits from the RHS to V.
-      KnownZero |= RHSKnownOne  & BKnownZero;
-      KnownOne  |= RHSKnownZero & BKnownZero;
-      KnownZero |= RHSKnownZero & BKnownOne;
-      KnownOne  |= RHSKnownOne  & BKnownOne;
+      Known.Zero |= RHSKnown.One  & BKnown.Zero;
+      Known.One  |= RHSKnown.Zero & BKnown.Zero;
+      Known.Zero |= RHSKnown.Zero & BKnown.One;
+      Known.One  |= RHSKnown.One  & BKnown.One;
     // assume(v << c = a)
     } else if (match(Arg, m_c_ICmp(Pred, m_Shl(m_V, m_ConstantInt(C)),
                                    m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
       // For those bits in RHS that are known, we can propagate them to known
       // bits in V shifted to the right by C.
-      RHSKnownZero.lshrInPlace(C->getZExtValue());
-      KnownZero |= RHSKnownZero;
-      RHSKnownOne.lshrInPlace(C->getZExtValue());
-      KnownOne  |= RHSKnownOne;
+      RHSKnown.Zero.lshrInPlace(C->getZExtValue());
+      Known.Zero |= RHSKnown.Zero;
+      RHSKnown.One.lshrInPlace(C->getZExtValue());
+      Known.One  |= RHSKnown.One;
     // assume(~(v << c) = a)
     } else if (match(Arg, m_c_ICmp(Pred, m_Not(m_Shl(m_V, m_ConstantInt(C))),
                                    m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
       // For those bits in RHS that are known, we can propagate them inverted
       // to known bits in V shifted to the right by C.
-      RHSKnownOne.lshrInPlace(C->getZExtValue());
-      KnownZero |= RHSKnownOne;
-      RHSKnownZero.lshrInPlace(C->getZExtValue());
-      KnownOne  |= RHSKnownZero;
+      RHSKnown.One.lshrInPlace(C->getZExtValue());
+      Known.Zero |= RHSKnown.One;
+      RHSKnown.Zero.lshrInPlace(C->getZExtValue());
+      Known.One  |= RHSKnown.Zero;
     // assume(v >> c = a)
     } else if (match(Arg,
                      m_c_ICmp(Pred, m_CombineOr(m_LShr(m_V, m_ConstantInt(C)),
@@ -685,12 +683,12 @@ static void computeKnownBitsFromAssume(const Value *V, APInt &KnownZero,
                               m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
       // For those bits in RHS that are known, we can propagate them to known
       // bits in V shifted to the right by C.
-      KnownZero |= RHSKnownZero << C->getZExtValue();
-      KnownOne  |= RHSKnownOne  << C->getZExtValue();
+      Known.Zero |= RHSKnown.Zero << C->getZExtValue();
+      Known.One  |= RHSKnown.One  << C->getZExtValue();
     // assume(~(v >> c) = a)
     } else if (match(Arg, m_c_ICmp(Pred, m_Not(m_CombineOr(
                                              m_LShr(m_V, m_ConstantInt(C)),
@@ -698,78 +696,78 @@ static void computeKnownBitsFromAssume(const Value *V, APInt &KnownZero,
                                    m_Value(A))) &&
                Pred == ICmpInst::ICMP_EQ &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
       // For those bits in RHS that are known, we can propagate them inverted
       // to known bits in V shifted to the right by C.
-      KnownZero |= RHSKnownOne  << C->getZExtValue();
-      KnownOne  |= RHSKnownZero << C->getZExtValue();
+      Known.Zero |= RHSKnown.One  << C->getZExtValue();
+      Known.One  |= RHSKnown.Zero << C->getZExtValue();
     // assume(v >=_s c) where c is non-negative
     } else if (match(Arg, m_ICmp(Pred, m_V, m_Value(A))) &&
                Pred == ICmpInst::ICMP_SGE &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
 
-      if (RHSKnownZero.isSignBitSet()) {
+      if (RHSKnown.Zero.isSignBitSet()) {
         // We know that the sign bit is zero.
-        KnownZero.setSignBit();
+        Known.Zero.setSignBit();
       }
     // assume(v >_s c) where c is at least -1.
     } else if (match(Arg, m_ICmp(Pred, m_V, m_Value(A))) &&
                Pred == ICmpInst::ICMP_SGT &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
 
-      if (RHSKnownOne.isAllOnesValue() || RHSKnownZero.isSignBitSet()) {
+      if (RHSKnown.One.isAllOnesValue() || RHSKnown.Zero.isSignBitSet()) {
         // We know that the sign bit is zero.
-        KnownZero.setSignBit();
+        Known.Zero.setSignBit();
       }
     // assume(v <=_s c) where c is negative
     } else if (match(Arg, m_ICmp(Pred, m_V, m_Value(A))) &&
                Pred == ICmpInst::ICMP_SLE &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
 
-      if (RHSKnownOne.isSignBitSet()) {
+      if (RHSKnown.One.isSignBitSet()) {
         // We know that the sign bit is one.
-        KnownOne.setSignBit();
+        Known.One.setSignBit();
       }
     // assume(v <_s c) where c is non-positive
     } else if (match(Arg, m_ICmp(Pred, m_V, m_Value(A))) &&
                Pred == ICmpInst::ICMP_SLT &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
 
-      if (RHSKnownZero.isAllOnesValue() || RHSKnownOne.isSignBitSet()) {
+      if (RHSKnown.Zero.isAllOnesValue() || RHSKnown.One.isSignBitSet()) {
         // We know that the sign bit is one.
-        KnownOne.setSignBit();
+        Known.One.setSignBit();
       }
     // assume(v <=_u c)
     } else if (match(Arg, m_ICmp(Pred, m_V, m_Value(A))) &&
                Pred == ICmpInst::ICMP_ULE &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
 
       // Whatever high bits in c are zero are known to be zero.
-      KnownZero.setHighBits(RHSKnownZero.countLeadingOnes());
+      Known.Zero.setHighBits(RHSKnown.Zero.countLeadingOnes());
     // assume(v <_u c)
     } else if (match(Arg, m_ICmp(Pred, m_V, m_Value(A))) &&
                Pred == ICmpInst::ICMP_ULT &&
                isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
-      APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
-      computeKnownBits(A, RHSKnownZero, RHSKnownOne, Depth+1, Query(Q, I));
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(A, RHSKnown, Depth+1, Query(Q, I));
 
       // Whatever high bits in c are zero are known to be zero (if c is a power
       // of 2, then one more).
       if (isKnownToBeAPowerOfTwo(A, false, Depth + 1, Query(Q, I)))
-        KnownZero.setHighBits(RHSKnownZero.countLeadingOnes()+1);
+        Known.Zero.setHighBits(RHSKnown.Zero.countLeadingOnes()+1);
       else
-        KnownZero.setHighBits(RHSKnownZero.countLeadingOnes());
+        Known.Zero.setHighBits(RHSKnown.Zero.countLeadingOnes());
     }
   }
 
@@ -778,9 +776,9 @@ static void computeKnownBitsFromAssume(const Value *V, APInt &KnownZero,
   // so this isn't a real bug. On the other hand, the program may have undefined
   // behavior, or we might have a bug in the compiler. We can't assert/crash, so
   // clear out the known bits, try to warn the user, and hope for the best.
-  if (KnownZero.intersects(KnownOne)) {
-    KnownZero.clearAllBits();
-    KnownOne.clearAllBits();
+  if (Known.Zero.intersects(Known.One)) {
+    Known.Zero.clearAllBits();
+    Known.One.clearAllBits();
 
     if (Q.ORE) {
       auto *CxtI = const_cast<Instruction *>(Q.CxtI);
@@ -793,57 +791,57 @@ static void computeKnownBitsFromAssume(const Value *V, APInt &KnownZero,
 }
 
 // Compute known bits from a shift operator, including those with a
-// non-constant shift amount. KnownZero and KnownOne are the outputs of this
-// function. KnownZero2 and KnownOne2 are pre-allocated temporaries with the
-// same bit width as KnownZero and KnownOne. KZF and KOF are operator-specific
-// functors that, given the known-zero or known-one bits respectively, and a
-// shift amount, compute the implied known-zero or known-one bits of the shift
-// operator's result respectively for that shift amount. The results from calling
-// KZF and KOF are conservatively combined for all permitted shift amounts.
+// non-constant shift amount. Known is the outputs of this function. Known2 is a
+// pre-allocated temporary with the/ same bit width as Known. KZF and KOF are
+// operator-specific functors that, given the known-zero or known-one bits
+// respectively, and a shift amount, compute the implied known-zero or known-one
+// bits of the shift operator's result respectively for that shift amount. The
+// results from calling KZF and KOF are conservatively combined for all
+// permitted shift amounts.
 static void computeKnownBitsFromShiftOperator(
-    const Operator *I, APInt &KnownZero, APInt &KnownOne, APInt &KnownZero2,
-    APInt &KnownOne2, unsigned Depth, const Query &Q,
+    const Operator *I, KnownBits &Known, KnownBits &Known2,
+    unsigned Depth, const Query &Q,
     function_ref<APInt(const APInt &, unsigned)> KZF,
     function_ref<APInt(const APInt &, unsigned)> KOF) {
-  unsigned BitWidth = KnownZero.getBitWidth();
+  unsigned BitWidth = Known.getBitWidth();
 
   if (auto *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
     unsigned ShiftAmt = SA->getLimitedValue(BitWidth-1);
 
-    computeKnownBits(I->getOperand(0), KnownZero, KnownOne, Depth + 1, Q);
-    KnownZero = KZF(KnownZero, ShiftAmt);
-    KnownOne  = KOF(KnownOne, ShiftAmt);
-    // If there is conflict between KnownZero and KnownOne, this must be an
-    // overflowing left shift, so the shift result is undefined. Clear KnownZero
-    // and KnownOne bits so that other code could propagate this undef.
-    if ((KnownZero & KnownOne) != 0) {
-      KnownZero.clearAllBits();
-      KnownOne.clearAllBits();
+    computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
+    Known.Zero = KZF(Known.Zero, ShiftAmt);
+    Known.One  = KOF(Known.One, ShiftAmt);
+    // If there is conflict between Known.Zero and Known.One, this must be an
+    // overflowing left shift, so the shift result is undefined. Clear Known
+    // bits so that other code could propagate this undef.
+    if ((Known.Zero & Known.One) != 0) {
+      Known.Zero.clearAllBits();
+      Known.One.clearAllBits();
     }
 
     return;
   }
 
-  computeKnownBits(I->getOperand(1), KnownZero, KnownOne, Depth + 1, Q);
+  computeKnownBits(I->getOperand(1), Known, Depth + 1, Q);
 
   // If the shift amount could be greater than or equal to the bit-width of the LHS, the
   // value could be undef, so we don't know anything about it.
-  if ((~KnownZero).uge(BitWidth)) {
-    KnownZero.clearAllBits();
-    KnownOne.clearAllBits();
+  if ((~Known.Zero).uge(BitWidth)) {
+    Known.Zero.clearAllBits();
+    Known.One.clearAllBits();
     return;
   }
 
-  // Note: We cannot use KnownZero.getLimitedValue() here, because if
+  // Note: We cannot use Known.Zero.getLimitedValue() here, because if
   // BitWidth > 64 and any upper bits are known, we'll end up returning the
   // limit value (which implies all bits are known).
-  uint64_t ShiftAmtKZ = KnownZero.zextOrTrunc(64).getZExtValue();
-  uint64_t ShiftAmtKO = KnownOne.zextOrTrunc(64).getZExtValue();
+  uint64_t ShiftAmtKZ = Known.Zero.zextOrTrunc(64).getZExtValue();
+  uint64_t ShiftAmtKO = Known.One.zextOrTrunc(64).getZExtValue();
 
   // It would be more-clearly correct to use the two temporaries for this
   // calculation. Reusing the APInts here to prevent unnecessary allocations.
-  KnownZero.clearAllBits();
-  KnownOne.clearAllBits();
+  Known.Zero.clearAllBits();
+  Known.One.clearAllBits();
 
   // If we know the shifter operand is nonzero, we can sometimes infer more
   // known bits. However this is expensive to compute, so be lazy about it and
@@ -858,10 +856,10 @@ static void computeKnownBitsFromShiftOperator(
       return;
   }
 
-  computeKnownBits(I->getOperand(0), KnownZero2, KnownOne2, Depth + 1, Q);
+  computeKnownBits(I->getOperand(0), Known2, Depth + 1, Q);
 
-  KnownZero.setAllBits();
-  KnownOne.setAllBits();
+  Known.Zero.setAllBits();
+  Known.One.setAllBits();
   for (unsigned ShiftAmt = 0; ShiftAmt < BitWidth; ++ShiftAmt) {
     // Combine the shifted known input bits only for those shift amounts
     // compatible with its known constraints.
@@ -880,8 +878,8 @@ static void computeKnownBitsFromShiftOperator(
         continue;
     }
 
-    KnownZero &= KZF(KnownZero2, ShiftAmt);
-    KnownOne  &= KOF(KnownOne2, ShiftAmt);
+    Known.Zero &= KZF(Known2.Zero, ShiftAmt);
+    Known.One  &= KOF(Known2.One, ShiftAmt);
   }
 
   // If there are no compatible shift amounts, then we've proven that the shift
@@ -889,33 +887,32 @@ static void computeKnownBitsFromShiftOperator(
   // return anything we'd like, but we need to make sure the sets of known bits
   // stay disjoint (it should be better for some other code to actually
   // propagate the undef than to pick a value here using known bits).
-  if (KnownZero.intersects(KnownOne)) {
-    KnownZero.clearAllBits();
-    KnownOne.clearAllBits();
+  if (Known.Zero.intersects(Known.One)) {
+    Known.Zero.clearAllBits();
+    Known.One.clearAllBits();
   }
 }
 
-static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
-                                         APInt &KnownOne, unsigned Depth,
-                                         const Query &Q) {
-  unsigned BitWidth = KnownZero.getBitWidth();
+static void computeKnownBitsFromOperator(const Operator *I, KnownBits &Known,
+                                         unsigned Depth, const Query &Q) {
+  unsigned BitWidth = Known.getBitWidth();
 
-  APInt KnownZero2(KnownZero), KnownOne2(KnownOne);
+  KnownBits Known2(Known);
   switch (I->getOpcode()) {
   default: break;
   case Instruction::Load:
     if (MDNode *MD = cast<LoadInst>(I)->getMetadata(LLVMContext::MD_range))
-      computeKnownBitsFromRangeMetadata(*MD, KnownZero, KnownOne);
+      computeKnownBitsFromRangeMetadata(*MD, Known.Zero, Known.One);
     break;
   case Instruction::And: {
     // If either the LHS or the RHS are Zero, the result is zero.
-    computeKnownBits(I->getOperand(1), KnownZero, KnownOne, Depth + 1, Q);
-    computeKnownBits(I->getOperand(0), KnownZero2, KnownOne2, Depth + 1, Q);
+    computeKnownBits(I->getOperand(1), Known, Depth + 1, Q);
+    computeKnownBits(I->getOperand(0), Known2, Depth + 1, Q);
 
     // Output known-1 bits are only known if set in both the LHS & RHS.
-    KnownOne &= KnownOne2;
+    Known.One &= Known2.One;
     // Output known-0 are known to be clear if zero in either the LHS | RHS.
-    KnownZero |= KnownZero2;
+    Known.Zero |= Known2.Zero;
 
     // and(x, add (x, -1)) is a common idiom that always clears the low bit;
     // here we handle the more general case of adding any odd number by
@@ -923,115 +920,115 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
     // TODO: This could be generalized to clearing any bit set in y where the
     // following bit is known to be unset in y.
     Value *Y = nullptr;
-    if (!KnownZero[0] && !KnownOne[0] &&
+    if (!Known.Zero[0] && !Known.One[0] &&
         (match(I->getOperand(0), m_Add(m_Specific(I->getOperand(1)),
                                        m_Value(Y))) ||
          match(I->getOperand(1), m_Add(m_Specific(I->getOperand(0)),
                                        m_Value(Y))))) {
-      KnownZero2.clearAllBits(); KnownOne2.clearAllBits();
-      computeKnownBits(Y, KnownZero2, KnownOne2, Depth + 1, Q);
-      if (KnownOne2.countTrailingOnes() > 0)
-        KnownZero.setBit(0);
+      Known2.Zero.clearAllBits(); Known2.One.clearAllBits();
+      computeKnownBits(Y, Known2, Depth + 1, Q);
+      if (Known2.One.countTrailingOnes() > 0)
+        Known.Zero.setBit(0);
     }
     break;
   }
   case Instruction::Or: {
-    computeKnownBits(I->getOperand(1), KnownZero, KnownOne, Depth + 1, Q);
-    computeKnownBits(I->getOperand(0), KnownZero2, KnownOne2, Depth + 1, Q);
+    computeKnownBits(I->getOperand(1), Known, Depth + 1, Q);
+    computeKnownBits(I->getOperand(0), Known2, Depth + 1, Q);
 
     // Output known-0 bits are only known if clear in both the LHS & RHS.
-    KnownZero &= KnownZero2;
+    Known.Zero &= Known2.Zero;
     // Output known-1 are known to be set if set in either the LHS | RHS.
-    KnownOne |= KnownOne2;
+    Known.One |= Known2.One;
     break;
   }
   case Instruction::Xor: {
-    computeKnownBits(I->getOperand(1), KnownZero, KnownOne, Depth + 1, Q);
-    computeKnownBits(I->getOperand(0), KnownZero2, KnownOne2, Depth + 1, Q);
+    computeKnownBits(I->getOperand(1), Known, Depth + 1, Q);
+    computeKnownBits(I->getOperand(0), Known2, Depth + 1, Q);
 
     // Output known-0 bits are known if clear or set in both the LHS & RHS.
-    APInt KnownZeroOut = (KnownZero & KnownZero2) | (KnownOne & KnownOne2);
+    APInt KnownZeroOut = (Known.Zero & Known2.Zero) | (Known.One & Known2.One);
     // Output known-1 are known to be set if set in only one of the LHS, RHS.
-    KnownOne = (KnownZero & KnownOne2) | (KnownOne & KnownZero2);
-    KnownZero = std::move(KnownZeroOut);
+    Known.One = (Known.Zero & Known2.One) | (Known.One & Known2.Zero);
+    Known.Zero = std::move(KnownZeroOut);
     break;
   }
   case Instruction::Mul: {
     bool NSW = cast<OverflowingBinaryOperator>(I)->hasNoSignedWrap();
-    computeKnownBitsMul(I->getOperand(0), I->getOperand(1), NSW, KnownZero,
-                        KnownOne, KnownZero2, KnownOne2, Depth, Q);
+    computeKnownBitsMul(I->getOperand(0), I->getOperand(1), NSW, Known,
+                        Known2, Depth, Q);
     break;
   }
   case Instruction::UDiv: {
     // For the purposes of computing leading zeros we can conservatively
     // treat a udiv as a logical right shift by the power of 2 known to
     // be less than the denominator.
-    computeKnownBits(I->getOperand(0), KnownZero2, KnownOne2, Depth + 1, Q);
-    unsigned LeadZ = KnownZero2.countLeadingOnes();
+    computeKnownBits(I->getOperand(0), Known2, Depth + 1, Q);
+    unsigned LeadZ = Known2.Zero.countLeadingOnes();
 
-    KnownOne2.clearAllBits();
-    KnownZero2.clearAllBits();
-    computeKnownBits(I->getOperand(1), KnownZero2, KnownOne2, Depth + 1, Q);
-    unsigned RHSUnknownLeadingOnes = KnownOne2.countLeadingZeros();
+    Known2.One.clearAllBits();
+    Known2.Zero.clearAllBits();
+    computeKnownBits(I->getOperand(1), Known2, Depth + 1, Q);
+    unsigned RHSUnknownLeadingOnes = Known2.One.countLeadingZeros();
     if (RHSUnknownLeadingOnes != BitWidth)
       LeadZ = std::min(BitWidth,
                        LeadZ + BitWidth - RHSUnknownLeadingOnes - 1);
 
-    KnownZero.setHighBits(LeadZ);
+    Known.Zero.setHighBits(LeadZ);
     break;
   }
   case Instruction::Select: {
     const Value *LHS, *RHS;
     SelectPatternFlavor SPF = matchSelectPattern(I, LHS, RHS).Flavor;
     if (SelectPatternResult::isMinOrMax(SPF)) {
-      computeKnownBits(RHS, KnownZero, KnownOne, Depth + 1, Q);
-      computeKnownBits(LHS, KnownZero2, KnownOne2, Depth + 1, Q);
+      computeKnownBits(RHS, Known, Depth + 1, Q);
+      computeKnownBits(LHS, Known2, Depth + 1, Q);
     } else {
-      computeKnownBits(I->getOperand(2), KnownZero, KnownOne, Depth + 1, Q);
-      computeKnownBits(I->getOperand(1), KnownZero2, KnownOne2, Depth + 1, Q);
+      computeKnownBits(I->getOperand(2), Known, Depth + 1, Q);
+      computeKnownBits(I->getOperand(1), Known2, Depth + 1, Q);
     }
 
     unsigned MaxHighOnes = 0;
     unsigned MaxHighZeros = 0;
     if (SPF == SPF_SMAX) {
       // If both sides are negative, the result is negative.
-      if (KnownOne.isSignBitSet() && KnownOne2.isSignBitSet())
+      if (Known.One.isSignBitSet() && Known2.One.isSignBitSet())
         // We can derive a lower bound on the result by taking the max of the
         // leading one bits.
-        MaxHighOnes =
-            std::max(KnownOne.countLeadingOnes(), KnownOne2.countLeadingOnes());
+        MaxHighOnes = std::max(Known.One.countLeadingOnes(),
+                               Known2.One.countLeadingOnes());
       // If either side is non-negative, the result is non-negative.
-      else if (KnownZero.isSignBitSet() || KnownZero2.isSignBitSet())
+      else if (Known.Zero.isSignBitSet() || Known2.Zero.isSignBitSet())
         MaxHighZeros = 1;
     } else if (SPF == SPF_SMIN) {
       // If both sides are non-negative, the result is non-negative.
-      if (KnownZero.isSignBitSet() && KnownZero2.isSignBitSet())
+      if (Known.Zero.isSignBitSet() && Known2.Zero.isSignBitSet())
         // We can derive an upper bound on the result by taking the max of the
         // leading zero bits.
-        MaxHighZeros = std::max(KnownZero.countLeadingOnes(),
-                                KnownZero2.countLeadingOnes());
+        MaxHighZeros = std::max(Known.Zero.countLeadingOnes(),
+                                Known2.Zero.countLeadingOnes());
       // If either side is negative, the result is negative.
-      else if (KnownOne.isSignBitSet() || KnownOne2.isSignBitSet())
+      else if (Known.One.isSignBitSet() || Known2.One.isSignBitSet())
         MaxHighOnes = 1;
     } else if (SPF == SPF_UMAX) {
       // We can derive a lower bound on the result by taking the max of the
       // leading one bits.
       MaxHighOnes =
-          std::max(KnownOne.countLeadingOnes(), KnownOne2.countLeadingOnes());
+          std::max(Known.One.countLeadingOnes(), Known2.One.countLeadingOnes());
     } else if (SPF == SPF_UMIN) {
       // We can derive an upper bound on the result by taking the max of the
       // leading zero bits.
       MaxHighZeros =
-          std::max(KnownZero.countLeadingOnes(), KnownZero2.countLeadingOnes());
+          std::max(Known.Zero.countLeadingOnes(), Known2.Zero.countLeadingOnes());
     }
 
     // Only known if known in both the LHS and RHS.
-    KnownOne &= KnownOne2;
-    KnownZero &= KnownZero2;
+    Known.One &= Known2.One;
+    Known.Zero &= Known2.Zero;
     if (MaxHighOnes > 0)
-      KnownOne.setHighBits(MaxHighOnes);
+      Known.One.setHighBits(MaxHighOnes);
     if (MaxHighZeros > 0)
-      KnownZero.setHighBits(MaxHighZeros);
+      Known.Zero.setHighBits(MaxHighZeros);
     break;
   }
   case Instruction::FPTrunc:
@@ -1055,14 +1052,14 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
     SrcBitWidth = Q.DL.getTypeSizeInBits(SrcTy->getScalarType());
 
     assert(SrcBitWidth && "SrcBitWidth can't be zero");
-    KnownZero = KnownZero.zextOrTrunc(SrcBitWidth);
-    KnownOne = KnownOne.zextOrTrunc(SrcBitWidth);
-    computeKnownBits(I->getOperand(0), KnownZero, KnownOne, Depth + 1, Q);
-    KnownZero = KnownZero.zextOrTrunc(BitWidth);
-    KnownOne = KnownOne.zextOrTrunc(BitWidth);
+    Known.Zero = Known.Zero.zextOrTrunc(SrcBitWidth);
+    Known.One = Known.One.zextOrTrunc(SrcBitWidth);
+    computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
+    Known.Zero = Known.Zero.zextOrTrunc(BitWidth);
+    Known.One = Known.One.zextOrTrunc(BitWidth);
     // Any top bits are known to be zero.
     if (BitWidth > SrcBitWidth)
-      KnownZero.setBitsFrom(SrcBitWidth);
+      Known.Zero.setBitsFrom(SrcBitWidth);
     break;
   }
   case Instruction::BitCast: {
@@ -1071,7 +1068,7 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
         // TODO: For now, not handling conversions like:
         // (bitcast i64 %x to <2 x i32>)
         !I->getType()->isVectorTy()) {
-      computeKnownBits(I->getOperand(0), KnownZero, KnownOne, Depth + 1, Q);
+      computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
       break;
     }
     break;
@@ -1080,13 +1077,13 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
     // Compute the bits in the result that are not present in the input.
     unsigned SrcBitWidth = I->getOperand(0)->getType()->getScalarSizeInBits();
 
-    KnownZero = KnownZero.trunc(SrcBitWidth);
-    KnownOne = KnownOne.trunc(SrcBitWidth);
-    computeKnownBits(I->getOperand(0), KnownZero, KnownOne, Depth + 1, Q);
+    Known.Zero = Known.Zero.trunc(SrcBitWidth);
+    Known.One = Known.One.trunc(SrcBitWidth);
+    computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
     // If the sign bit of the input is known set or clear, then we know the
     // top bits of the result.
-    KnownZero = KnownZero.sext(BitWidth);
-    KnownOne = KnownOne.sext(BitWidth);
+    Known.Zero = Known.Zero.sext(BitWidth);
+    Known.One = Known.One.sext(BitWidth);
     break;
   }
   case Instruction::Shl: {
@@ -1109,9 +1106,7 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
       return KOResult;
     };
 
-    computeKnownBitsFromShiftOperator(I, KnownZero, KnownOne,
-                                      KnownZero2, KnownOne2, Depth, Q, KZF,
-                                      KOF);
+    computeKnownBitsFromShiftOperator(I, Known, Known2, Depth, Q, KZF, KOF);
     break;
   }
   case Instruction::LShr: {
@@ -1127,9 +1122,7 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
       return KnownOne.lshr(ShiftAmt);
     };
 
-    computeKnownBitsFromShiftOperator(I, KnownZero, KnownOne,
-                                      KnownZero2, KnownOne2, Depth, Q, KZF,
-                                      KOF);
+    computeKnownBitsFromShiftOperator(I, Known, Known2, Depth, Q, KZF, KOF);
     break;
   }
   case Instruction::AShr: {
@@ -1142,23 +1135,19 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
       return KnownOne.ashr(ShiftAmt);
     };
 
-    computeKnownBitsFromShiftOperator(I, KnownZero, KnownOne,
-                                      KnownZero2, KnownOne2, Depth, Q, KZF,
-                                      KOF);
+    computeKnownBitsFromShiftOperator(I, Known, Known2, Depth, Q, KZF, KOF);
     break;
   }
   case Instruction::Sub: {
     bool NSW = cast<OverflowingBinaryOperator>(I)->hasNoSignedWrap();
     computeKnownBitsAddSub(false, I->getOperand(0), I->getOperand(1), NSW,
-                           KnownZero, KnownOne, KnownZero2, KnownOne2, Depth,
-                           Q);
+                           Known, Known2, Depth, Q);
     break;
   }
   case Instruction::Add: {
     bool NSW = cast<OverflowingBinaryOperator>(I)->hasNoSignedWrap();
     computeKnownBitsAddSub(true, I->getOperand(0), I->getOperand(1), NSW,
-                           KnownZero, KnownOne, KnownZero2, KnownOne2, Depth,
-                           Q);
+                           Known, Known2, Depth, Q);
     break;
   }
   case Instruction::SRem:
@@ -1166,34 +1155,33 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
       APInt RA = Rem->getValue().abs();
       if (RA.isPowerOf2()) {
         APInt LowBits = RA - 1;
-        computeKnownBits(I->getOperand(0), KnownZero2, KnownOne2, Depth + 1,
-                         Q);
+        computeKnownBits(I->getOperand(0), Known2, Depth + 1, Q);
 
         // The low bits of the first operand are unchanged by the srem.
-        KnownZero = KnownZero2 & LowBits;
-        KnownOne = KnownOne2 & LowBits;
+        Known.Zero = Known2.Zero & LowBits;
+        Known.One = Known2.One & LowBits;
 
         // If the first operand is non-negative or has all low bits zero, then
         // the upper bits are all zero.
-        if (KnownZero2.isSignBitSet() || ((KnownZero2 & LowBits) == LowBits))
-          KnownZero |= ~LowBits;
+        if (Known2.Zero.isSignBitSet() || ((Known2.Zero & LowBits) == LowBits))
+          Known.Zero |= ~LowBits;
 
         // If the first operand is negative and not all low bits are zero, then
         // the upper bits are all one.
-        if (KnownOne2.isSignBitSet() && ((KnownOne2 & LowBits) != 0))
-          KnownOne |= ~LowBits;
+        if (Known2.One.isSignBitSet() && ((Known2.One & LowBits) != 0))
+          Known.One |= ~LowBits;
 
-        assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+        assert((Known.Zero & Known.One) == 0 && "Bits known to be one AND zero?");
         break;
       }
     }
 
     // The sign bit is the LHS's sign bit, except when the result of the
     // remainder is zero.
-    computeKnownBits(I->getOperand(0), KnownZero2, KnownOne2, Depth + 1, Q);
+    computeKnownBits(I->getOperand(0), Known2, Depth + 1, Q);
     // If it's known zero, our sign bit is also zero.
-    if (KnownZero2.isSignBitSet())
-      KnownZero.setSignBit();
+    if (Known2.Zero.isSignBitSet())
+      Known.Zero.setSignBit();
 
     break;
   case Instruction::URem: {
@@ -1201,23 +1189,23 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
       const APInt &RA = Rem->getValue();
       if (RA.isPowerOf2()) {
         APInt LowBits = (RA - 1);
-        computeKnownBits(I->getOperand(0), KnownZero, KnownOne, Depth + 1, Q);
-        KnownZero |= ~LowBits;
-        KnownOne &= LowBits;
+        computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
+        Known.Zero |= ~LowBits;
+        Known.One &= LowBits;
         break;
       }
     }
 
     // Since the result is less than or equal to either operand, any leading
     // zero bits in either operand must also exist in the result.
-    computeKnownBits(I->getOperand(0), KnownZero, KnownOne, Depth + 1, Q);
-    computeKnownBits(I->getOperand(1), KnownZero2, KnownOne2, Depth + 1, Q);
-
-    unsigned Leaders = std::max(KnownZero.countLeadingOnes(),
-                                KnownZero2.countLeadingOnes());
-    KnownOne.clearAllBits();
-    KnownZero.clearAllBits();
-    KnownZero.setHighBits(Leaders);
+    computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
+    computeKnownBits(I->getOperand(1), Known2, Depth + 1, Q);
+
+    unsigned Leaders = std::max(Known.Zero.countLeadingOnes(),
+                                Known2.Zero.countLeadingOnes());
+    Known.One.clearAllBits();
+    Known.Zero.clearAllBits();
+    Known.Zero.setHighBits(Leaders);
     break;
   }
 
@@ -1228,16 +1216,15 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
       Align = Q.DL.getABITypeAlignment(AI->getAllocatedType());
 
     if (Align > 0)
-      KnownZero.setLowBits(countTrailingZeros(Align));
+      Known.Zero.setLowBits(countTrailingZeros(Align));
     break;
   }
   case Instruction::GetElementPtr: {
     // Analyze all of the subscripts of this getelementptr instruction
     // to determine if we can prove known low zero bits.
-    APInt LocalKnownZero(BitWidth, 0), LocalKnownOne(BitWidth, 0);
-    computeKnownBits(I->getOperand(0), LocalKnownZero, LocalKnownOne, Depth + 1,
-                     Q);
-    unsigned TrailZ = LocalKnownZero.countTrailingOnes();
+    KnownBits LocalKnown(BitWidth);
+    computeKnownBits(I->getOperand(0), LocalKnown, Depth + 1, Q);
+    unsigned TrailZ = LocalKnown.Zero.countTrailingOnes();
 
     gep_type_iterator GTI = gep_type_begin(I);
     for (unsigned i = 1, e = I->getNumOperands(); i != e; ++i, ++GTI) {
@@ -1267,15 +1254,15 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
         }
         unsigned GEPOpiBits = Index->getType()->getScalarSizeInBits();
         uint64_t TypeSize = Q.DL.getTypeAllocSize(IndexedTy);
-        LocalKnownZero = LocalKnownOne = APInt(GEPOpiBits, 0);
-        computeKnownBits(Index, LocalKnownZero, LocalKnownOne, Depth + 1, Q);
+        LocalKnown.Zero = LocalKnown.One = APInt(GEPOpiBits, 0);
+        computeKnownBits(Index, LocalKnown, Depth + 1, Q);
         TrailZ = std::min(TrailZ,
                           unsigned(countTrailingZeros(TypeSize) +
-                                   LocalKnownZero.countTrailingOnes()));
+                                   LocalKnown.Zero.countTrailingOnes()));
       }
     }
 
-    KnownZero.setLowBits(TrailZ);
+    Known.Zero.setLowBits(TrailZ);
     break;
   }
   case Instruction::PHI: {
@@ -1310,14 +1297,14 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
             break;
           // Ok, we have a PHI of the form L op= R. Check for low
           // zero bits.
-          computeKnownBits(R, KnownZero2, KnownOne2, Depth + 1, Q);
+          computeKnownBits(R, Known2, Depth + 1, Q);
 
           // We need to take the minimum number of known bits
-          APInt KnownZero3(KnownZero), KnownOne3(KnownOne);
-          computeKnownBits(L, KnownZero3, KnownOne3, Depth + 1, Q);
+          KnownBits Known3(Known);
+          computeKnownBits(L, Known3, Depth + 1, Q);
 
-          KnownZero.setLowBits(std::min(KnownZero2.countTrailingOnes(),
-                                        KnownZero3.countTrailingOnes()));
+          Known.Zero.setLowBits(std::min(Known2.Zero.countTrailingOnes(),
+                                         Known3.Zero.countTrailingOnes()));
 
           if (DontImproveNonNegativePhiBits)
             break;
@@ -1334,25 +1321,25 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
             // (add non-negative, non-negative) --> non-negative
             // (add negative, negative) --> negative
             if (Opcode == Instruction::Add) {
-              if (KnownZero2.isSignBitSet() && KnownZero3.isSignBitSet())
-                KnownZero.setSignBit();
-              else if (KnownOne2.isSignBitSet() && KnownOne3.isSignBitSet())
-                KnownOne.setSignBit();
+              if (Known2.Zero.isSignBitSet() && Known3.Zero.isSignBitSet())
+                Known.Zero.setSignBit();
+              else if (Known2.One.isSignBitSet() && Known3.One.isSignBitSet())
+                Known.One.setSignBit();
             }
 
             // (sub nsw non-negative, negative) --> non-negative
             // (sub nsw negative, non-negative) --> negative
             else if (Opcode == Instruction::Sub && LL == I) {
-              if (KnownZero2.isSignBitSet() && KnownOne3.isSignBitSet())
-                KnownZero.setSignBit();
-              else if (KnownOne2.isSignBitSet() && KnownZero3.isSignBitSet())
-                KnownOne.setSignBit();
+              if (Known2.Zero.isSignBitSet() && Known3.One.isSignBitSet())
+                Known.Zero.setSignBit();
+              else if (Known2.One.isSignBitSet() && Known3.Zero.isSignBitSet())
+                Known.One.setSignBit();
             }
 
             // (mul nsw non-negative, non-negative) --> non-negative
-            else if (Opcode == Instruction::Mul && KnownZero2.isSignBitSet() &&
-                     KnownZero3.isSignBitSet())
-              KnownZero.setSignBit();
+            else if (Opcode == Instruction::Mul && Known2.Zero.isSignBitSet() &&
+                     Known3.Zero.isSignBitSet())
+              Known.Zero.setSignBit();
           }
 
           break;
@@ -1366,27 +1353,26 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
 
     // Otherwise take the unions of the known bit sets of the operands,
     // taking conservative care to avoid excessive recursion.
-    if (Depth < MaxDepth - 1 && !KnownZero && !KnownOne) {
+    if (Depth < MaxDepth - 1 && !Known.Zero && !Known.One) {
       // Skip if every incoming value references to ourself.
       if (dyn_cast_or_null<UndefValue>(P->hasConstantValue()))
         break;
 
-      KnownZero.setAllBits();
-      KnownOne.setAllBits();
+      Known.Zero.setAllBits();
+      Known.One.setAllBits();
       for (Value *IncValue : P->incoming_values()) {
         // Skip direct self references.
         if (IncValue == P) continue;
 
-        KnownZero2 = APInt(BitWidth, 0);
-        KnownOne2 = APInt(BitWidth, 0);
+        Known2 = KnownBits(BitWidth);
         // Recurse, but cap the recursion to one level, because we don't
         // want to waste time spinning around in loops.
-        computeKnownBits(IncValue, KnownZero2, KnownOne2, MaxDepth - 1, Q);
-        KnownZero &= KnownZero2;
-        KnownOne &= KnownOne2;
+        computeKnownBits(IncValue, Known2, MaxDepth - 1, Q);
+        Known.Zero &= Known2.Zero;
+        Known.One &= Known2.One;
         // If all bits have been ruled out, there's no need to check
         // more operands.
-        if (!KnownZero && !KnownOne)
+        if (!Known.Zero && !Known.One)
           break;
       }
     }
@@ -1398,24 +1384,24 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
     // and then intersect with known bits based on other properties of the
     // function.
     if (MDNode *MD = cast<Instruction>(I)->getMetadata(LLVMContext::MD_range))
-      computeKnownBitsFromRangeMetadata(*MD, KnownZero, KnownOne);
+      computeKnownBitsFromRangeMetadata(*MD, Known.Zero, Known.One);
     if (const Value *RV = ImmutableCallSite(I).getReturnedArgOperand()) {
-      computeKnownBits(RV, KnownZero2, KnownOne2, Depth + 1, Q);
-      KnownZero |= KnownZero2;
-      KnownOne |= KnownOne2;
+      computeKnownBits(RV, Known2, Depth + 1, Q);
+      Known.Zero |= Known2.Zero;
+      Known.One |= Known2.One;
     }
     if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
       switch (II->getIntrinsicID()) {
       default: break;
       case Intrinsic::bitreverse:
-        computeKnownBits(I->getOperand(0), KnownZero2, KnownOne2, Depth + 1, Q);
-        KnownZero |= KnownZero2.reverseBits();
-        KnownOne |= KnownOne2.reverseBits();
+        computeKnownBits(I->getOperand(0), Known2, Depth + 1, Q);
+        Known.Zero |= Known2.Zero.reverseBits();
+        Known.One |= Known2.One.reverseBits();
         break;
       case Intrinsic::bswap:
-        computeKnownBits(I->getOperand(0), KnownZero2, KnownOne2, Depth + 1, Q);
-        KnownZero |= KnownZero2.byteSwap();
-        KnownOne |= KnownOne2.byteSwap();
+        computeKnownBits(I->getOperand(0), Known2, Depth + 1, Q);
+        Known.Zero |= Known2.Zero.byteSwap();
+        Known.One |= Known2.One.byteSwap();
         break;
       case Intrinsic::ctlz:
       case Intrinsic::cttz: {
@@ -1423,22 +1409,22 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
         // If this call is undefined for 0, the result will be less than 2^n.
         if (II->getArgOperand(1) == ConstantInt::getTrue(II->getContext()))
           LowBits -= 1;
-        KnownZero.setBitsFrom(LowBits);
+        Known.Zero.setBitsFrom(LowBits);
         break;
       }
       case Intrinsic::ctpop: {
-        computeKnownBits(I->getOperand(0), KnownZero2, KnownOne2, Depth + 1, Q);
+        computeKnownBits(I->getOperand(0), Known2, Depth + 1, Q);
         // We can bound the space the count needs.  Also, bits known to be zero
         // can't contribute to the population.
-        unsigned BitsPossiblySet = BitWidth - KnownZero2.countPopulation();
+        unsigned BitsPossiblySet = BitWidth - Known2.Zero.countPopulation();
         unsigned LowBits = Log2_32(BitsPossiblySet)+1;
-        KnownZero.setBitsFrom(LowBits);
+        Known.Zero.setBitsFrom(LowBits);
         // TODO: we could bound KnownOne using the lower bound on the number
         // of bits which might be set provided by popcnt KnownOne2.
         break;
       }
       case Intrinsic::x86_sse42_crc32_64_64:
-        KnownZero.setBitsFrom(32);
+        Known.Zero.setBitsFrom(32);
         break;
       }
     }
@@ -1448,7 +1434,7 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
     // tracking the specific element. But at least we might find information
     // valid for all elements of the vector (for example if vector is sign
     // extended, shifted, etc).
-    computeKnownBits(I->getOperand(0), KnownZero, KnownOne, Depth + 1, Q);
+    computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
     break;
   case Instruction::ExtractValue:
     if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I->getOperand(0))) {
@@ -1460,20 +1446,19 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
         case Intrinsic::uadd_with_overflow:
         case Intrinsic::sadd_with_overflow:
           computeKnownBitsAddSub(true, II->getArgOperand(0),
-                                 II->getArgOperand(1), false, KnownZero,
-                                 KnownOne, KnownZero2, KnownOne2, Depth, Q);
+                                 II->getArgOperand(1), false, Known, Known2,
+                                 Depth, Q);
           break;
         case Intrinsic::usub_with_overflow:
         case Intrinsic::ssub_with_overflow:
           computeKnownBitsAddSub(false, II->getArgOperand(0),
-                                 II->getArgOperand(1), false, KnownZero,
-                                 KnownOne, KnownZero2, KnownOne2, Depth, Q);
+                                 II->getArgOperand(1), false, Known, Known2,
+                                 Depth, Q);
           break;
         case Intrinsic::umul_with_overflow:
         case Intrinsic::smul_with_overflow:
           computeKnownBitsMul(II->getArgOperand(0), II->getArgOperand(1), false,
-                              KnownZero, KnownOne, KnownZero2, KnownOne2, Depth,
-                              Q);
+                              Known, Known2, Depth, Q);
           break;
         }
       }
@@ -1482,7 +1467,7 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
 }
 
 /// Determine which bits of V are known to be either zero or one and return
-/// them in the KnownZero/KnownOne bit sets.
+/// them in the Known bit set.
 ///
 /// NOTE: we cannot consider 'undef' to be "IsZero" here.  The problem is that
 /// we cannot optimize based on the assumption that it is zero without changing
@@ -1496,11 +1481,11 @@ static void computeKnownBitsFromOperator(const Operator *I, APInt &KnownZero,
 /// where V is a vector, known zero, and known one values are the
 /// same width as the vector element, and the bit is set only if it is true
 /// for all of the elements in the vector.
-void computeKnownBits(const Value *V, APInt &KnownZero, APInt &KnownOne,
-                      unsigned Depth, const Query &Q) {
+void computeKnownBits(const Value *V, KnownBits &Known, unsigned Depth,
+                      const Query &Q) {
   assert(V && "No Value?");
   assert(Depth <= MaxDepth && "Limit Search Depth");
-  unsigned BitWidth = KnownZero.getBitWidth();
+  unsigned BitWidth = Known.getBitWidth();
 
   assert((V->getType()->isIntOrIntVectorTy() ||
           V->getType()->getScalarType()->isPointerTy()) &&
@@ -1508,22 +1493,20 @@ void computeKnownBits(const Value *V, APInt &KnownZero, APInt &KnownOne,
   assert((Q.DL.getTypeSizeInBits(V->getType()->getScalarType()) == BitWidth) &&
          (!V->getType()->isIntOrIntVectorTy() ||
           V->getType()->getScalarSizeInBits() == BitWidth) &&
-         KnownZero.getBitWidth() == BitWidth &&
-         KnownOne.getBitWidth() == BitWidth &&
-         "V, KnownOne and KnownZero should have same BitWidth");
+         "V and Known should have same BitWidth");
   (void)BitWidth;
 
   const APInt *C;
   if (match(V, m_APInt(C))) {
     // We know all of the bits for a scalar constant or a splat vector constant!
-    KnownOne = *C;
-    KnownZero = ~KnownOne;
+    Known.One = *C;
+    Known.Zero = ~Known.One;
     return;
   }
   // Null and aggregate-zero are all-zeros.
   if (isa<ConstantPointerNull>(V) || isa<ConstantAggregateZero>(V)) {
-    KnownOne.clearAllBits();
-    KnownZero.setAllBits();
+    Known.One.clearAllBits();
+    Known.Zero.setAllBits();
     return;
   }
   // Handle a constant vector by taking the intersection of the known bits of
@@ -1531,12 +1514,12 @@ void computeKnownBits(const Value *V, APInt &KnownZero, APInt &KnownOne,
   if (const ConstantDataSequential *CDS = dyn_cast<ConstantDataSequential>(V)) {
     // We know that CDS must be a vector of integers. Take the intersection of
     // each element.
-    KnownZero.setAllBits(); KnownOne.setAllBits();
+    Known.Zero.setAllBits(); Known.One.setAllBits();
     APInt Elt(BitWidth, 0);
     for (unsigned i = 0, e = CDS->getNumElements(); i != e; ++i) {
       Elt = CDS->getElementAsInteger(i);
-      KnownZero &= ~Elt;
-      KnownOne &= Elt;
+      Known.Zero &= ~Elt;
+      Known.One &= Elt;
     }
     return;
   }
@@ -1544,25 +1527,25 @@ void computeKnownBits(const Value *V, APInt &KnownZero, APInt &KnownOne,
   if (const auto *CV = dyn_cast<ConstantVector>(V)) {
     // We know that CV must be a vector of integers. Take the intersection of
     // each element.
-    KnownZero.setAllBits(); KnownOne.setAllBits();
-    APInt Elt(KnownZero.getBitWidth(), 0);
+    Known.Zero.setAllBits(); Known.One.setAllBits();
+    APInt Elt(BitWidth, 0);
     for (unsigned i = 0, e = CV->getNumOperands(); i != e; ++i) {
       Constant *Element = CV->getAggregateElement(i);
       auto *ElementCI = dyn_cast_or_null<ConstantInt>(Element);
       if (!ElementCI) {
-        KnownZero.clearAllBits();
-        KnownOne.clearAllBits();
+        Known.Zero.clearAllBits();
+        Known.One.clearAllBits();
         return;
       }
       Elt = ElementCI->getValue();
-      KnownZero &= ~Elt;
-      KnownOne &= Elt;
+      Known.Zero &= ~Elt;
+      Known.One &= Elt;
     }
     return;
   }
 
   // Start out not knowing anything.
-  KnownZero.clearAllBits(); KnownOne.clearAllBits();
+  Known.Zero.clearAllBits(); Known.One.clearAllBits();
 
   // We can't imply anything about undefs.
   if (isa<UndefValue>(V))
@@ -1581,27 +1564,27 @@ void computeKnownBits(const Value *V, APInt &KnownZero, APInt &KnownOne,
   // the bits of its aliasee.
   if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) {
     if (!GA->isInterposable())
-      computeKnownBits(GA->getAliasee(), KnownZero, KnownOne, Depth + 1, Q);
+      computeKnownBits(GA->getAliasee(), Known, Depth + 1, Q);
     return;
   }
 
   if (const Operator *I = dyn_cast<Operator>(V))
-    computeKnownBitsFromOperator(I, KnownZero, KnownOne, Depth, Q);
+    computeKnownBitsFromOperator(I, Known, Depth, Q);
 
-  // Aligned pointers have trailing zeros - refine KnownZero set
+  // Aligned pointers have trailing zeros - refine Known.Zero set
   if (V->getType()->isPointerTy()) {
     unsigned Align = V->getPointerAlignment(Q.DL);
     if (Align)
-      KnownZero.setLowBits(countTrailingZeros(Align));
+      Known.Zero.setLowBits(countTrailingZeros(Align));
   }
 
-  // computeKnownBitsFromAssume strictly refines KnownZero and
-  // KnownOne. Therefore, we run them after computeKnownBitsFromOperator.
+  // computeKnownBitsFromAssume strictly refines Known.
+  // Therefore, we run them after computeKnownBitsFromOperator.
 
   // Check whether a nearby assume intrinsic can determine some known bits.
-  computeKnownBitsFromAssume(V, KnownZero, KnownOne, Depth, Q);
+  computeKnownBitsFromAssume(V, Known, Depth, Q);
 
-  assert((KnownZero & KnownOne) == 0 && "Bits known to be one AND zero?");
+  assert((Known.Zero & Known.One) == 0 && "Bits known to be one AND zero?");
 }
 
 /// Determine whether the sign bit is known to be zero or one.
@@ -1614,11 +1597,10 @@ void ComputeSignBit(const Value *V, bool &KnownZero, bool &KnownOne,
     KnownOne = false;
     return;
   }
-  APInt ZeroBits(BitWidth, 0);
-  APInt OneBits(BitWidth, 0);
-  computeKnownBits(V, ZeroBits, OneBits, Depth, Q);
-  KnownOne = OneBits.isSignBitSet();
-  KnownZero = ZeroBits.isSignBitSet();
+  KnownBits Bits(BitWidth);
+  computeKnownBits(V, Bits, Depth, Q);
+  KnownOne = Bits.One.isSignBitSet();
+  KnownZero = Bits.Zero.isSignBitSet();
 }
 
 /// Return true if the given value is known to have exactly one
@@ -1690,18 +1672,18 @@ bool isKnownToBeAPowerOfTwo(const Value *V, bool OrZero, unsigned Depth,
           return true;
 
       unsigned BitWidth = V->getType()->getScalarSizeInBits();
-      APInt LHSZeroBits(BitWidth, 0), LHSOneBits(BitWidth, 0);
-      computeKnownBits(X, LHSZeroBits, LHSOneBits, Depth, Q);
+      KnownBits LHSBits(BitWidth);
+      computeKnownBits(X, LHSBits, Depth, Q);
 
-      APInt RHSZeroBits(BitWidth, 0), RHSOneBits(BitWidth, 0);
-      computeKnownBits(Y, RHSZeroBits, RHSOneBits, Depth, Q);
+      KnownBits RHSBits(BitWidth);
+      computeKnownBits(Y, RHSBits, Depth, Q);
       // If i8 V is a power of two or zero:
       //  ZeroBits: 1 1 1 0 1 1 1 1
       // ~ZeroBits: 0 0 0 1 0 0 0 0
-      if ((~(LHSZeroBits & RHSZeroBits)).isPowerOf2())
+      if ((~(LHSBits.Zero & RHSBits.Zero)).isPowerOf2())
         // If OrZero isn't set, we cannot give back a zero result.
         // Make sure either the LHS or RHS has a bit set.
-        if (OrZero || RHSOneBits.getBoolValue() || LHSOneBits.getBoolValue())
+        if (OrZero || RHSBits.One.getBoolValue() || LHSBits.One.getBoolValue())
           return true;
     }
   }
@@ -1872,10 +1854,9 @@ bool isKnownNonZero(const Value *V, unsigned Depth, const Query &Q) {
     if (BO->hasNoUnsignedWrap())
       return isKnownNonZero(X, Depth, Q);
 
-    APInt KnownZero(BitWidth, 0);
-    APInt KnownOne(BitWidth, 0);
-    computeKnownBits(X, KnownZero, KnownOne, Depth, Q);
-    if (KnownOne[0])
+    KnownBits Known(BitWidth);
+    computeKnownBits(X, Known, Depth, Q);
+    if (Known.One[0])
       return true;
   }
   // shr X, Y != 0 if X is negative.  Note that the value of the shift is not
@@ -1895,16 +1876,15 @@ bool isKnownNonZero(const Value *V, unsigned Depth, const Query &Q) {
     // out are known to be zero, and X is known non-zero then at least one
     // non-zero bit must remain.
     if (ConstantInt *Shift = dyn_cast<ConstantInt>(Y)) {
-      APInt KnownZero(BitWidth, 0);
-      APInt KnownOne(BitWidth, 0);
-      computeKnownBits(X, KnownZero, KnownOne, Depth, Q);
+      KnownBits Known(BitWidth);
+      computeKnownBits(X, Known, Depth, Q);
 
       auto ShiftVal = Shift->getLimitedValue(BitWidth - 1);
       // Is there a known one in the portion not shifted out?
-      if (KnownOne.countLeadingZeros() < BitWidth - ShiftVal)
+      if (Known.One.countLeadingZeros() < BitWidth - ShiftVal)
         return true;
       // Are all the bits to be shifted out known zero?
-      if (KnownZero.countTrailingOnes() >= ShiftVal)
+      if (Known.Zero.countTrailingOnes() >= ShiftVal)
         return isKnownNonZero(X, Depth, Q);
     }
   }
@@ -1928,18 +1908,17 @@ bool isKnownNonZero(const Value *V, unsigned Depth, const Query &Q) {
     // If X and Y are both negative (as signed values) then their sum is not
     // zero unless both X and Y equal INT_MIN.
     if (BitWidth && XKnownNegative && YKnownNegative) {
-      APInt KnownZero(BitWidth, 0);
-      APInt KnownOne(BitWidth, 0);
+      KnownBits Known(BitWidth);
       APInt Mask = APInt::getSignedMaxValue(BitWidth);
       // The sign bit of X is set.  If some other bit is set then X is not equal
       // to INT_MIN.
-      computeKnownBits(X, KnownZero, KnownOne, Depth, Q);
-      if (KnownOne.intersects(Mask))
+      computeKnownBits(X, Known, Depth, Q);
+      if (Known.One.intersects(Mask))
         return true;
       // The sign bit of Y is set.  If some other bit is set then Y is not equal
       // to INT_MIN.
-      computeKnownBits(Y, KnownZero, KnownOne, Depth, Q);
-      if (KnownOne.intersects(Mask))
+      computeKnownBits(Y, Known, Depth, Q);
+      if (Known.One.intersects(Mask))
         return true;
     }
 
@@ -1994,10 +1973,9 @@ bool isKnownNonZero(const Value *V, unsigned Depth, const Query &Q) {
   }
 
   if (!BitWidth) return false;
-  APInt KnownZero(BitWidth, 0);
-  APInt KnownOne(BitWidth, 0);
-  computeKnownBits(V, KnownZero, KnownOne, Depth, Q);
-  return KnownOne != 0;
+  KnownBits Known(BitWidth);
+  computeKnownBits(V, Known, Depth, Q);
+  return Known.One != 0;
 }
 
 /// Return true if V2 == V1 + X, where X is known non-zero.
@@ -2029,14 +2007,13 @@ static bool isKnownNonEqual(const Value *V1, const Value *V2, const Query &Q) {
     // Are any known bits in V1 contradictory to known bits in V2? If V1
     // has a known zero where V2 has a known one, they must not be equal.
     auto BitWidth = Ty->getBitWidth();
-    APInt KnownZero1(BitWidth, 0);
-    APInt KnownOne1(BitWidth, 0);
-    computeKnownBits(V1, KnownZero1, KnownOne1, 0, Q);
-    APInt KnownZero2(BitWidth, 0);
-    APInt KnownOne2(BitWidth, 0);
-    computeKnownBits(V2, KnownZero2, KnownOne2, 0, Q);
-
-    APInt OppositeBits = (KnownZero1 & KnownOne2) | (KnownZero2 & KnownOne1);
+    KnownBits Known1(BitWidth);
+    computeKnownBits(V1, Known1, 0, Q);
+    KnownBits Known2(BitWidth);
+    computeKnownBits(V2, Known2, 0, Q);
+
+    APInt OppositeBits = (Known1.Zero & Known2.One) |
+                         (Known2.Zero & Known1.One);
     if (OppositeBits.getBoolValue())
       return true;
   }
@@ -2054,9 +2031,9 @@ static bool isKnownNonEqual(const Value *V1, const Value *V2, const Query &Q) {
 /// for all of the elements in the vector.
 bool MaskedValueIsZero(const Value *V, const APInt &Mask, unsigned Depth,
                        const Query &Q) {
-  APInt KnownZero(Mask.getBitWidth(), 0), KnownOne(Mask.getBitWidth(), 0);
-  computeKnownBits(V, KnownZero, KnownOne, Depth, Q);
-  return Mask.isSubsetOf(KnownZero);
+  KnownBits Known(Mask.getBitWidth());
+  computeKnownBits(V, Known, Depth, Q);
+  return Mask.isSubsetOf(Known.Zero);
 }
 
 /// For vector constants, loop over the elements and find the constant with the
@@ -2234,17 +2211,17 @@ static unsigned ComputeNumSignBitsImpl(const Value *V, unsigned Depth,
     // Special case decrementing a value (ADD X, -1):
     if (const auto *CRHS = dyn_cast<Constant>(U->getOperand(1)))
       if (CRHS->isAllOnesValue()) {
-        APInt KnownZero(TyBits, 0), KnownOne(TyBits, 0);
-        computeKnownBits(U->getOperand(0), KnownZero, KnownOne, Depth + 1, Q);
+        KnownBits Known(TyBits);
+        computeKnownBits(U->getOperand(0), Known, Depth + 1, Q);
 
         // If the input is known to be 0 or 1, the output is 0/-1, which is all
         // sign bits set.
-        if ((KnownZero | 1).isAllOnesValue())
+        if ((Known.Zero | 1).isAllOnesValue())
           return TyBits;
 
         // If we are subtracting one from a positive number, there is no carry
         // out of the result.
-        if (KnownZero.isSignBitSet())
+        if (Known.Zero.isSignBitSet())
           return Tmp;
       }
 
@@ -2259,16 +2236,16 @@ static unsigned ComputeNumSignBitsImpl(const Value *V, unsigned Depth,
     // Handle NEG.
     if (const auto *CLHS = dyn_cast<Constant>(U->getOperand(0)))
       if (CLHS->isNullValue()) {
-        APInt KnownZero(TyBits, 0), KnownOne(TyBits, 0);
-        computeKnownBits(U->getOperand(1), KnownZero, KnownOne, Depth + 1, Q);
+        KnownBits Known(TyBits);
+        computeKnownBits(U->getOperand(1), Known, Depth + 1, Q);
         // If the input is known to be 0 or 1, the output is 0/-1, which is all
         // sign bits set.
-        if ((KnownZero | 1).isAllOnesValue())
+        if ((Known.Zero | 1).isAllOnesValue())
           return TyBits;
 
         // If the input is known to be positive (the sign bit is known clear),
         // the output of the NEG has the same number of sign bits as the input.
-        if (KnownZero.isSignBitSet())
+        if (Known.Zero.isSignBitSet())
           return Tmp2;
 
         // Otherwise, we treat this like a SUB.
@@ -2320,16 +2297,16 @@ static unsigned ComputeNumSignBitsImpl(const Value *V, unsigned Depth,
   if (unsigned VecSignBits = computeNumSignBitsVectorConstant(V, TyBits))
     return VecSignBits;
 
-  APInt KnownZero(TyBits, 0), KnownOne(TyBits, 0);
-  computeKnownBits(V, KnownZero, KnownOne, Depth, Q);
+  KnownBits Known(TyBits);
+  computeKnownBits(V, Known, Depth, Q);
 
   // If we know that the sign bit is either zero or one, determine the number of
   // identical bits in the top of the input value.
-  if (KnownZero.isSignBitSet())
-    return std::max(FirstAnswer, KnownZero.countLeadingOnes());
+  if (Known.Zero.isSignBitSet())
+    return std::max(FirstAnswer, Known.Zero.countLeadingOnes());
 
-  if (KnownOne.isSignBitSet())
-    return std::max(FirstAnswer, KnownOne.countLeadingOnes());
+  if (Known.One.isSignBitSet())
+    return std::max(FirstAnswer, Known.One.countLeadingOnes());
 
   // computeKnownBits gave us no extra information about the top bits.
   return FirstAnswer;
@@ -3535,26 +3512,22 @@ OverflowResult llvm::computeOverflowForUnsignedMul(const Value *LHS,
   // we can guarantee that the result does not overflow.
   // Ref: "Hacker's Delight" by Henry Warren
   unsigned BitWidth = LHS->getType()->getScalarSizeInBits();
-  APInt LHSKnownZero(BitWidth, 0);
-  APInt LHSKnownOne(BitWidth, 0);
-  APInt RHSKnownZero(BitWidth, 0);
-  APInt RHSKnownOne(BitWidth, 0);
-  computeKnownBits(LHS, LHSKnownZero, LHSKnownOne, DL, /*Depth=*/0, AC, CxtI,
-                   DT);
-  computeKnownBits(RHS, RHSKnownZero, RHSKnownOne, DL, /*Depth=*/0, AC, CxtI,
-                   DT);
+  KnownBits LHSKnown(BitWidth);
+  KnownBits RHSKnown(BitWidth);
+  computeKnownBits(LHS, LHSKnown, DL, /*Depth=*/0, AC, CxtI, DT);
+  computeKnownBits(RHS, RHSKnown, DL, /*Depth=*/0, AC, CxtI, DT);
   // Note that underestimating the number of zero bits gives a more
   // conservative answer.
-  unsigned ZeroBits = LHSKnownZero.countLeadingOnes() +
-                      RHSKnownZero.countLeadingOnes();
+  unsigned ZeroBits = LHSKnown.Zero.countLeadingOnes() +
+                      RHSKnown.Zero.countLeadingOnes();
   // First handle the easy case: if we have enough zero bits there's
   // definitely no overflow.
   if (ZeroBits >= BitWidth)
     return OverflowResult::NeverOverflows;
 
   // Get the largest possible values for each operand.
-  APInt LHSMax = ~LHSKnownZero;
-  APInt RHSMax = ~RHSKnownZero;
+  APInt LHSMax = ~LHSKnown.Zero;
+  APInt RHSMax = ~RHSKnown.Zero;
 
   // We know the multiply operation doesn't overflow if the maximum values for
   // each operand will not overflow after we multiply them together.
@@ -3566,7 +3539,7 @@ OverflowResult llvm::computeOverflowForUnsignedMul(const Value *LHS,
   // We know it always overflows if multiplying the smallest possible values for
   // the operands also results in overflow.
   bool MinOverflow;
-  (void)LHSKnownOne.umul_ov(RHSKnownOne, MinOverflow);
+  (void)LHSKnown.One.umul_ov(RHSKnown.One, MinOverflow);
   if (MinOverflow)
     return OverflowResult::AlwaysOverflows;
 
@@ -4285,11 +4258,10 @@ static bool isTruePredicate(CmpInst::Predicate Pred,
       // If X & C == 0 then (X | C) == X +_{nuw} C
       if (match(A, m_Or(m_Value(X), m_APInt(CA))) &&
           match(B, m_Or(m_Specific(X), m_APInt(CB)))) {
-        unsigned BitWidth = CA->getBitWidth();
-        APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
-        computeKnownBits(X, KnownZero, KnownOne, DL, Depth + 1, AC, CxtI, DT);
+        KnownBits Known(CA->getBitWidth());
+        computeKnownBits(X, Known, DL, Depth + 1, AC, CxtI, DT);
 
-        if (CA->isSubsetOf(KnownZero) && CB->isSubsetOf(KnownZero))
+        if (CA->isSubsetOf(Known.Zero) && CB->isSubsetOf(Known.Zero))
           return true;
       }
 
index 63db98a..439f67f 100644 (file)
@@ -36,6 +36,7 @@
 #include "llvm/IR/Intrinsics.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Support/ManagedStatic.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/Mutex.h"
@@ -7539,10 +7540,10 @@ unsigned SelectionDAG::InferPtrAlignment(SDValue Ptr) const {
   int64_t GVOffset = 0;
   if (TLI->isGAPlusOffset(Ptr.getNode(), GV, GVOffset)) {
     unsigned PtrWidth = getDataLayout().getPointerTypeSizeInBits(GV->getType());
-    APInt KnownZero(PtrWidth, 0), KnownOne(PtrWidth, 0);
-    llvm::computeKnownBits(const_cast<GlobalValue *>(GV), KnownZero, KnownOne,
+    KnownBits Known(PtrWidth);
+    llvm::computeKnownBits(const_cast<GlobalValue *>(GV), Known,
                            getDataLayout());
-    unsigned AlignBits = KnownZero.countTrailingOnes();
+    unsigned AlignBits = Known.Zero.countTrailingOnes();
     unsigned Align = AlignBits ? 1 << std::min(31U, AlignBits) : 0;
     if (Align)
       return MinAlign(Align, GVOffset);
index b594847..1829c5d 100644 (file)
@@ -26,6 +26,7 @@
 #include "llvm/Transforms/Scalar.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Support/raw_ostream.h"
 
 #include <algorithm>
@@ -1206,10 +1207,9 @@ bool PolynomialMultiplyRecognize::highBitsAreZero(Value *V,
   if (!T)
     return false;
 
-  unsigned BW = T->getBitWidth();
-  APInt K0(BW, 0), K1(BW, 0);
-  computeKnownBits(V, K0, K1, DL);
-  return K0.countLeadingOnes() >= IterCount;
+  KnownBits Known(T->getBitWidth());
+  computeKnownBits(V, Known, DL);
+  return Known.Zero.countLeadingOnes() >= IterCount;
 }
 
 
index e946acf..0304610 100644 (file)
@@ -17,6 +17,7 @@
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/GetElementPtrTypeIterator.h"
 #include "llvm/IR/PatternMatch.h"
+#include "llvm/Support/KnownBits.h"
 
 using namespace llvm;
 using namespace PatternMatch;
@@ -899,24 +900,22 @@ bool InstCombiner::WillNotOverflowSignedAdd(Value *LHS, Value *RHS,
     return true;
 
   unsigned BitWidth = LHS->getType()->getScalarSizeInBits();
-  APInt LHSKnownZero(BitWidth, 0);
-  APInt LHSKnownOne(BitWidth, 0);
-  computeKnownBits(LHS, LHSKnownZero, LHSKnownOne, 0, &CxtI);
+  KnownBits LHSKnown(BitWidth);
+  computeKnownBits(LHS, LHSKnown, 0, &CxtI);
 
-  APInt RHSKnownZero(BitWidth, 0);
-  APInt RHSKnownOne(BitWidth, 0);
-  computeKnownBits(RHS, RHSKnownZero, RHSKnownOne, 0, &CxtI);
+  KnownBits RHSKnown(BitWidth);
+  computeKnownBits(RHS, RHSKnown, 0, &CxtI);
 
   // Addition of two 2's complement numbers having opposite signs will never
   // overflow.
-  if ((LHSKnownOne[BitWidth - 1] && RHSKnownZero[BitWidth - 1]) ||
-      (LHSKnownZero[BitWidth - 1] && RHSKnownOne[BitWidth - 1]))
+  if ((LHSKnown.One[BitWidth - 1] && RHSKnown.Zero[BitWidth - 1]) ||
+      (LHSKnown.Zero[BitWidth - 1] && RHSKnown.One[BitWidth - 1]))
     return true;
 
   // Check if carry bit of addition will not cause overflow.
-  if (checkRippleForAdd(LHSKnownZero, RHSKnownZero))
+  if (checkRippleForAdd(LHSKnown.Zero, RHSKnown.Zero))
     return true;
-  if (checkRippleForAdd(RHSKnownZero, LHSKnownZero))
+  if (checkRippleForAdd(RHSKnown.Zero, LHSKnown.Zero))
     return true;
 
   return false;
@@ -936,18 +935,16 @@ bool InstCombiner::WillNotOverflowSignedSub(Value *LHS, Value *RHS,
     return true;
 
   unsigned BitWidth = LHS->getType()->getScalarSizeInBits();
-  APInt LHSKnownZero(BitWidth, 0);
-  APInt LHSKnownOne(BitWidth, 0);
-  computeKnownBits(LHS, LHSKnownZero, LHSKnownOne, 0, &CxtI);
+  KnownBits LHSKnown(BitWidth);
+  computeKnownBits(LHS, LHSKnown, 0, &CxtI);
 
-  APInt RHSKnownZero(BitWidth, 0);
-  APInt RHSKnownOne(BitWidth, 0);
-  computeKnownBits(RHS, RHSKnownZero, RHSKnownOne, 0, &CxtI);
+  KnownBits RHSKnown(BitWidth);
+  computeKnownBits(RHS, RHSKnown, 0, &CxtI);
 
   // Subtraction of two 2's complement numbers having identical signs will
   // never overflow.
-  if ((LHSKnownOne[BitWidth - 1] && RHSKnownOne[BitWidth - 1]) ||
-      (LHSKnownZero[BitWidth - 1] && RHSKnownZero[BitWidth - 1]))
+  if ((LHSKnown.One[BitWidth - 1] && RHSKnown.One[BitWidth - 1]) ||
+      (LHSKnown.Zero[BitWidth - 1] && RHSKnown.Zero[BitWidth - 1]))
     return true;
 
   // TODO: implement logic similar to checkRippleForAdd
@@ -1118,10 +1115,9 @@ Instruction *InstCombiner::visitAdd(BinaryOperator &I) {
       // a sub and fuse this add with it.
       if (LHS->hasOneUse() && (XorRHS->getValue()+1).isPowerOf2()) {
         IntegerType *IT = cast<IntegerType>(I.getType());
-        APInt LHSKnownOne(IT->getBitWidth(), 0);
-        APInt LHSKnownZero(IT->getBitWidth(), 0);
-        computeKnownBits(XorLHS, LHSKnownZero, LHSKnownOne, 0, &I);
-        if ((XorRHS->getValue() | LHSKnownZero).isAllOnesValue())
+        KnownBits LHSKnown(IT->getBitWidth());
+        computeKnownBits(XorLHS, LHSKnown, 0, &I);
+        if ((XorRHS->getValue() | LHSKnown.Zero).isAllOnesValue())
           return BinaryOperator::CreateSub(ConstantExpr::getAdd(XorRHS, CI),
                                            XorLHS);
       }
@@ -1641,10 +1637,9 @@ Instruction *InstCombiner::visitSub(BinaryOperator &I) {
     // Turn this into a xor if LHS is 2^n-1 and the remaining bits are known
     // zero.
     if (Op0C->isMask()) {
-      APInt RHSKnownZero(BitWidth, 0);
-      APInt RHSKnownOne(BitWidth, 0);
-      computeKnownBits(Op1, RHSKnownZero, RHSKnownOne, 0, &I);
-      if ((*Op0C | RHSKnownZero).isAllOnesValue())
+      KnownBits RHSKnown(BitWidth);
+      computeKnownBits(Op1, RHSKnown, 0, &I);
+      if ((*Op0C | RHSKnown.Zero).isAllOnesValue())
         return BinaryOperator::CreateXor(Op1, Op0);
     }
   }
index 6890b06..313ab13 100644 (file)
@@ -44,6 +44,7 @@
 #include "llvm/IR/ValueHandle.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Transforms/Utils/SimplifyLibCalls.h"
@@ -1378,14 +1379,13 @@ static Instruction *foldCttzCtlz(IntrinsicInst &II, InstCombiner &IC) {
     return nullptr;
 
   unsigned BitWidth = IT->getBitWidth();
-  APInt KnownZero(BitWidth, 0);
-  APInt KnownOne(BitWidth, 0);
-  IC.computeKnownBits(Op0, KnownZero, KnownOne, 0, &II);
+  KnownBits Known(BitWidth);
+  IC.computeKnownBits(Op0, Known, 0, &II);
 
   // Create a mask for bits above (ctlz) or below (cttz) the first known one.
   bool IsTZ = II.getIntrinsicID() == Intrinsic::cttz;
-  unsigned NumMaskBits = IsTZ ? KnownOne.countTrailingZeros()
-                              : KnownOne.countLeadingZeros();
+  unsigned NumMaskBits = IsTZ ? Known.One.countTrailingZeros()
+                              : Known.One.countLeadingZeros();
   APInt Mask = IsTZ ? APInt::getLowBitsSet(BitWidth, NumMaskBits)
                     : APInt::getHighBitsSet(BitWidth, NumMaskBits);
 
@@ -1393,7 +1393,7 @@ static Instruction *foldCttzCtlz(IntrinsicInst &II, InstCombiner &IC) {
   // zero, this value is constant.
   // FIXME: This should be in InstSimplify because we're replacing an
   // instruction with a constant.
-  if (Mask.isSubsetOf(KnownZero)) {
+  if (Mask.isSubsetOf(Known.Zero)) {
     auto *C = ConstantInt::get(IT, APInt(BitWidth, NumMaskBits));
     return IC.replaceInstUsesWith(II, C);
   }
@@ -1401,7 +1401,7 @@ static Instruction *foldCttzCtlz(IntrinsicInst &II, InstCombiner &IC) {
   // If the input to cttz/ctlz is known to be non-zero,
   // then change the 'ZeroIsUndef' parameter to 'true'
   // because we know the zero behavior can't affect the result.
-  if (KnownOne != 0 || isKnownNonZero(Op0, IC.getDataLayout())) {
+  if (Known.One != 0 || isKnownNonZero(Op0, IC.getDataLayout())) {
     if (!match(II.getArgOperand(1), m_One())) {
       II.setOperand(1, IC.Builder->getTrue());
       return &II;
@@ -3617,9 +3617,9 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
 
     // If there is a dominating assume with the same condition as this one,
     // then this one is redundant, and should be removed.
-    APInt KnownZero(1, 0), KnownOne(1, 0);
-    computeKnownBits(IIOperand, KnownZero, KnownOne, 0, II);
-    if (KnownOne.isAllOnesValue())
+    KnownBits Known(1);
+    computeKnownBits(IIOperand, Known, 0, II);
+    if (Known.One.isAllOnesValue())
       return eraseInstFromFunction(*II);
 
     // Update the cache of affected values for this assumption (we might be
index 9127ddc..312d9ba 100644 (file)
 #include "InstCombineInternal.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/Analysis/ConstantFolding.h"
+#include "llvm/Analysis/TargetLibraryInfo.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/PatternMatch.h"
-#include "llvm/Analysis/TargetLibraryInfo.h"
+#include "llvm/Support/KnownBits.h"
 using namespace llvm;
 using namespace PatternMatch;
 
@@ -676,11 +677,10 @@ Instruction *InstCombiner::transformZExtICmp(ICmpInst *ICI, ZExtInst &CI,
         // This only works for EQ and NE
         ICI->isEquality()) {
       // If Op1C some other power of two, convert:
-      uint32_t BitWidth = Op1C->getType()->getBitWidth();
-      APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
-      computeKnownBits(ICI->getOperand(0), KnownZero, KnownOne, 0, &CI);
+      KnownBits Known(Op1C->getType()->getBitWidth());
+      computeKnownBits(ICI->getOperand(0), Known, 0, &CI);
 
-      APInt KnownZeroMask(~KnownZero);
+      APInt KnownZeroMask(~Known.Zero);
       if (KnownZeroMask.isPowerOf2()) { // Exactly 1 possible 1?
         if (!DoTransform) return ICI;
 
@@ -726,13 +726,13 @@ Instruction *InstCombiner::transformZExtICmp(ICmpInst *ICI, ZExtInst &CI,
       Value *LHS = ICI->getOperand(0);
       Value *RHS = ICI->getOperand(1);
 
-      APInt KnownZeroLHS(BitWidth, 0), KnownOneLHS(BitWidth, 0);
-      APInt KnownZeroRHS(BitWidth, 0), KnownOneRHS(BitWidth, 0);
-      computeKnownBits(LHS, KnownZeroLHS, KnownOneLHS, 0, &CI);
-      computeKnownBits(RHS, KnownZeroRHS, KnownOneRHS, 0, &CI);
+      KnownBits KnownLHS(BitWidth);
+      KnownBits KnownRHS(BitWidth);
+      computeKnownBits(LHS, KnownLHS, 0, &CI);
+      computeKnownBits(RHS, KnownRHS, 0, &CI);
 
-      if (KnownZeroLHS == KnownZeroRHS && KnownOneLHS == KnownOneRHS) {
-        APInt KnownBits = KnownZeroLHS | KnownOneLHS;
+      if (KnownLHS.Zero == KnownRHS.Zero && KnownLHS.One == KnownRHS.One) {
+        APInt KnownBits = KnownLHS.Zero | KnownLHS.One;
         APInt UnknownBit = ~KnownBits;
         if (UnknownBit.countPopulation() == 1) {
           if (!DoTransform) return ICI;
@@ -740,7 +740,7 @@ Instruction *InstCombiner::transformZExtICmp(ICmpInst *ICI, ZExtInst &CI,
           Value *Result = Builder->CreateXor(LHS, RHS);
 
           // Mask off any bits that are set and won't be shifted away.
-          if (KnownOneLHS.uge(UnknownBit))
+          if (KnownLHS.One.uge(UnknownBit))
             Result = Builder->CreateAnd(Result,
                                         ConstantInt::get(ITy, UnknownBit));
 
@@ -1049,10 +1049,10 @@ Instruction *InstCombiner::transformSExtICmp(ICmpInst *ICI, Instruction &CI) {
     if (ICI->hasOneUse() &&
         ICI->isEquality() && (Op1C->isZero() || Op1C->getValue().isPowerOf2())){
       unsigned BitWidth = Op1C->getType()->getBitWidth();
-      APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
-      computeKnownBits(Op0, KnownZero, KnownOne, 0, &CI);
+      KnownBits Known(BitWidth);
+      computeKnownBits(Op0, Known, 0, &CI);
 
-      APInt KnownZeroMask(~KnownZero);
+      APInt KnownZeroMask(~Known.Zero);
       if (KnownZeroMask.isPowerOf2()) {
         Value *In = ICI->getOperand(0);
 
index 003029a..d846a63 100644 (file)
@@ -26,6 +26,7 @@
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/PatternMatch.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/KnownBits.h"
 
 using namespace llvm;
 using namespace PatternMatch;
@@ -175,19 +176,18 @@ static bool isSignTest(ICmpInst::Predicate &Pred, const APInt &C) {
 /// Given a signed integer type and a set of known zero and one bits, compute
 /// the maximum and minimum values that could have the specified known zero and
 /// known one bits, returning them in Min/Max.
-static void computeSignedMinMaxValuesFromKnownBits(const APInt &KnownZero,
-                                                   const APInt &KnownOne,
+/// TODO: Move to method on KnownBits struct?
+static void computeSignedMinMaxValuesFromKnownBits(const KnownBits &Known,
                                                    APInt &Min, APInt &Max) {
-  assert(KnownZero.getBitWidth() == KnownOne.getBitWidth() &&
-         KnownZero.getBitWidth() == Min.getBitWidth() &&
-         KnownZero.getBitWidth() == Max.getBitWidth() &&
+  assert(Known.getBitWidth() == Min.getBitWidth() &&
+         Known.getBitWidth() == Max.getBitWidth() &&
          "KnownZero, KnownOne and Min, Max must have equal bitwidth.");
-  APInt UnknownBits = ~(KnownZero|KnownOne);
+  APInt UnknownBits = ~(Known.Zero|Known.One);
 
   // The minimum value is when all unknown bits are zeros, EXCEPT for the sign
   // bit if it is unknown.
-  Min = KnownOne;
-  Max = KnownOne|UnknownBits;
+  Min = Known.One;
+  Max = Known.One|UnknownBits;
 
   if (UnknownBits.isNegative()) { // Sign bit is unknown
     Min.setBit(Min.getBitWidth()-1);
@@ -198,19 +198,18 @@ static void computeSignedMinMaxValuesFromKnownBits(const APInt &KnownZero,
 /// Given an unsigned integer type and a set of known zero and one bits, compute
 /// the maximum and minimum values that could have the specified known zero and
 /// known one bits, returning them in Min/Max.
-static void computeUnsignedMinMaxValuesFromKnownBits(const APInt &KnownZero,
-                                                     const APInt &KnownOne,
+/// TODO: Move to method on KnownBits struct?
+static void computeUnsignedMinMaxValuesFromKnownBits(const KnownBits &Known,
                                                      APInt &Min, APInt &Max) {
-  assert(KnownZero.getBitWidth() == KnownOne.getBitWidth() &&
-         KnownZero.getBitWidth() == Min.getBitWidth() &&
-         KnownZero.getBitWidth() == Max.getBitWidth() &&
+  assert(Known.getBitWidth() == Min.getBitWidth() &&
+         Known.getBitWidth() == Max.getBitWidth() &&
          "Ty, KnownZero, KnownOne and Min, Max must have equal bitwidth.");
-  APInt UnknownBits = ~(KnownZero|KnownOne);
+  APInt UnknownBits = ~(Known.Zero|Known.One);
 
   // The minimum value is when the unknown bits are all zeros.
-  Min = KnownOne;
+  Min = Known.One;
   // The maximum value is when the unknown bits are all ones.
-  Max = KnownOne|UnknownBits;
+  Max = Known.One|UnknownBits;
 }
 
 /// This is called when we see this pattern:
@@ -1479,14 +1478,14 @@ Instruction *InstCombiner::foldICmpTruncConstant(ICmpInst &Cmp,
     // of the high bits truncated out of x are known.
     unsigned DstBits = Trunc->getType()->getScalarSizeInBits(),
              SrcBits = X->getType()->getScalarSizeInBits();
-    APInt KnownZero(SrcBits, 0), KnownOne(SrcBits, 0);
-    computeKnownBits(X, KnownZero, KnownOne, 0, &Cmp);
+    KnownBits Known(SrcBits);
+    computeKnownBits(X, Known, 0, &Cmp);
 
     // If all the high bits are known, we can do this xform.
-    if ((KnownZero | KnownOne).countLeadingOnes() >= SrcBits - DstBits) {
+    if ((Known.Zero | Known.One).countLeadingOnes() >= SrcBits - DstBits) {
       // Pull in the high bits from known-ones set.
       APInt NewRHS = C->zext(SrcBits);
-      NewRHS |= KnownOne & APInt::getHighBitsSet(SrcBits, SrcBits - DstBits);
+      NewRHS |= Known.One & APInt::getHighBitsSet(SrcBits, SrcBits - DstBits);
       return new ICmpInst(Pred, X, ConstantInt::get(X->getType(), NewRHS));
     }
   }
@@ -4001,16 +4000,16 @@ Instruction *InstCombiner::foldICmpUsingKnownBits(ICmpInst &I) {
     IsSignBit = isSignBitCheck(Pred, *CmpC, UnusedBit);
   }
 
-  APInt Op0KnownZero(BitWidth, 0), Op0KnownOne(BitWidth, 0);
-  APInt Op1KnownZero(BitWidth, 0), Op1KnownOne(BitWidth, 0);
+  KnownBits Op0Known(BitWidth);
+  KnownBits Op1Known(BitWidth);
 
   if (SimplifyDemandedBits(&I, 0,
                            getDemandedBitsLHSMask(I, BitWidth, IsSignBit),
-                           Op0KnownZero, Op0KnownOne, 0))
+                           Op0Known, 0))
     return &I;
 
   if (SimplifyDemandedBits(&I, 1, APInt::getAllOnesValue(BitWidth),
-                           Op1KnownZero, Op1KnownOne, 0))
+                           Op1Known, 0))
     return &I;
 
   // Given the known and unknown bits, compute a range that the LHS could be
@@ -4019,15 +4018,11 @@ Instruction *InstCombiner::foldICmpUsingKnownBits(ICmpInst &I) {
   APInt Op0Min(BitWidth, 0), Op0Max(BitWidth, 0);
   APInt Op1Min(BitWidth, 0), Op1Max(BitWidth, 0);
   if (I.isSigned()) {
-    computeSignedMinMaxValuesFromKnownBits(Op0KnownZero, Op0KnownOne, Op0Min,
-                                           Op0Max);
-    computeSignedMinMaxValuesFromKnownBits(Op1KnownZero, Op1KnownOne, Op1Min,
-                                           Op1Max);
+    computeSignedMinMaxValuesFromKnownBits(Op0Known, Op0Min, Op0Max);
+    computeSignedMinMaxValuesFromKnownBits(Op1Known, Op1Min, Op1Max);
   } else {
-    computeUnsignedMinMaxValuesFromKnownBits(Op0KnownZero, Op0KnownOne, Op0Min,
-                                             Op0Max);
-    computeUnsignedMinMaxValuesFromKnownBits(Op1KnownZero, Op1KnownOne, Op1Min,
-                                             Op1Max);
+    computeUnsignedMinMaxValuesFromKnownBits(Op0Known, Op0Min, Op0Max);
+    computeUnsignedMinMaxValuesFromKnownBits(Op1Known, Op1Min, Op1Max);
   }
 
   // If Min and Max are known to be the same, then SimplifyDemandedBits
@@ -4054,8 +4049,8 @@ Instruction *InstCombiner::foldICmpUsingKnownBits(ICmpInst &I) {
     // If all bits are known zero except for one, then we know at most one bit
     // is set. If the comparison is against zero, then this is a check to see if
     // *that* bit is set.
-    APInt Op0KnownZeroInverted = ~Op0KnownZero;
-    if (~Op1KnownZero == 0) {
+    APInt Op0KnownZeroInverted = ~Op0Known.Zero;
+    if (~Op1Known.Zero == 0) {
       // If the LHS is an AND with the same constant, look through it.
       Value *LHS = nullptr;
       const APInt *LHSC;
@@ -4193,8 +4188,8 @@ Instruction *InstCombiner::foldICmpUsingKnownBits(ICmpInst &I) {
   // Turn a signed comparison into an unsigned one if both operands are known to
   // have the same sign.
   if (I.isSigned() &&
-      ((Op0KnownZero.isNegative() && Op1KnownZero.isNegative()) ||
-       (Op0KnownOne.isNegative() && Op1KnownOne.isNegative())))
+      ((Op0Known.Zero.isNegative() && Op1Known.Zero.isNegative()) ||
+       (Op0Known.One.isNegative() && Op1Known.One.isNegative())))
     return new ICmpInst(I.getUnsignedPredicate(), Op0, Op1);
 
   return nullptr;
index c1eeef6..776686d 100644 (file)
@@ -489,10 +489,9 @@ public:
     return nullptr; // Don't do anything with FI
   }
 
-  void computeKnownBits(Value *V, APInt &KnownZero, APInt &KnownOne,
+  void computeKnownBits(Value *V, KnownBits &Known,
                         unsigned Depth, Instruction *CxtI) const {
-    return llvm::computeKnownBits(V, KnownZero, KnownOne, DL, Depth, &AC, CxtI,
-                                  &DT);
+    return llvm::computeKnownBits(V, Known, DL, Depth, &AC, CxtI, &DT);
   }
 
   bool MaskedValueIsZero(Value *V, const APInt &Mask, unsigned Depth = 0,
@@ -536,25 +535,23 @@ private:
 
   /// \brief Attempts to replace V with a simpler value based on the demanded
   /// bits.
-  Value *SimplifyDemandedUseBits(Value *V, APInt DemandedMask, APInt &KnownZero,
-                                 APInt &KnownOne, unsigned Depth,
-                                 Instruction *CxtI);
+  Value *SimplifyDemandedUseBits(Value *V, APInt DemandedMask, KnownBits &Known,
+                                 unsigned Depth, Instruction *CxtI);
   bool SimplifyDemandedBits(Instruction *I, unsigned Op,
-                            const APInt &DemandedMask, APInt &KnownZero,
-                            APInt &KnownOne, unsigned Depth = 0);
+                            const APInt &DemandedMask, KnownBits &Known,
+                            unsigned Depth = 0);
   /// Helper routine of SimplifyDemandedUseBits. It computes KnownZero/KnownOne
   /// bits. It also tries to handle simplifications that can be done based on
   /// DemandedMask, but without modifying the Instruction.
   Value *SimplifyMultipleUseDemandedBits(Instruction *I,
                                          const APInt &DemandedMask,
-                                         APInt &KnownZero, APInt &KnownOne,
+                                         KnownBits &Known,
                                          unsigned Depth, Instruction *CxtI);
   /// Helper routine of SimplifyDemandedUseBits. It tries to simplify demanded
   /// bit for "r1 = shr x, c1; r2 = shl r1, c2" instruction sequence.
   Value *simplifyShrShlDemandedBits(
       Instruction *Shr, const APInt &ShrOp1, Instruction *Shl,
-      const APInt &ShlOp1, const APInt &DemandedMask, APInt &KnownZero,
-      APInt &KnownOne);
+      const APInt &ShlOp1, const APInt &DemandedMask, KnownBits &Known);
 
   /// \brief Tries to simplify operands to an integer instruction based on its
   /// demanded bits.
index 5d6d899..76829c5 100644 (file)
@@ -17,6 +17,7 @@
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/IR/MDBuilder.h"
 #include "llvm/IR/PatternMatch.h"
+#include "llvm/Support/KnownBits.h"
 using namespace llvm;
 using namespace PatternMatch;
 
@@ -1476,11 +1477,11 @@ Instruction *InstCombiner::visitSelectInst(SelectInst &SI) {
   // The motivation for this call into value tracking is to take advantage of
   // the assumption cache, so make sure that is populated.
   if (!CondVal->getType()->isVectorTy() && !AC.assumptions().empty()) {
-    APInt KnownOne(1, 0), KnownZero(1, 0);
-    computeKnownBits(CondVal, KnownZero, KnownOne, 0, &SI);
-    if (KnownOne == 1)
+    KnownBits Known(1);
+    computeKnownBits(CondVal, Known, 0, &SI);
+    if (Known.One == 1)
       return replaceInstUsesWith(SI, TrueVal);
-    if (KnownZero == 1)
+    if (Known.Zero == 1)
       return replaceInstUsesWith(SI, FalseVal);
   }
 
index 4dabbda..8d0ed85 100644 (file)
@@ -16,6 +16,7 @@
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/PatternMatch.h"
+#include "llvm/Support/KnownBits.h"
 
 using namespace llvm;
 using namespace llvm::PatternMatch;
@@ -52,10 +53,10 @@ static bool ShrinkDemandedConstant(Instruction *I, unsigned OpNo,
 /// the instruction has any properties that allow us to simplify its operands.
 bool InstCombiner::SimplifyDemandedInstructionBits(Instruction &Inst) {
   unsigned BitWidth = Inst.getType()->getScalarSizeInBits();
-  APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
+  KnownBits Known(BitWidth);
   APInt DemandedMask(APInt::getAllOnesValue(BitWidth));
 
-  Value *V = SimplifyDemandedUseBits(&Inst, DemandedMask, KnownZero, KnownOne,
+  Value *V = SimplifyDemandedUseBits(&Inst, DemandedMask, Known,
                                      0, &Inst);
   if (!V) return false;
   if (V == &Inst) return true;
@@ -68,11 +69,11 @@ bool InstCombiner::SimplifyDemandedInstructionBits(Instruction &Inst) {
 /// change and false otherwise.
 bool InstCombiner::SimplifyDemandedBits(Instruction *I, unsigned OpNo,
                                         const APInt &DemandedMask,
-                                        APInt &KnownZero, APInt &KnownOne,
+                                        KnownBits &Known,
                                         unsigned Depth) {
   Use &U = I->getOperandUse(OpNo);
-  Value *NewVal = SimplifyDemandedUseBits(U.get(), DemandedMask, KnownZero,
-                                          KnownOne, Depth, I);
+  Value *NewVal = SimplifyDemandedUseBits(U.get(), DemandedMask, Known,
+                                          Depth, I);
   if (!NewVal) return false;
   U = NewVal;
   return true;
@@ -86,15 +87,16 @@ bool InstCombiner::SimplifyDemandedBits(Instruction *I, unsigned OpNo,
 /// with a constant or one of its operands. In such cases, this function does
 /// the replacement and returns true. In all other cases, it returns false after
 /// analyzing the expression and setting KnownOne and known to be one in the
-/// expression. KnownZero contains all the bits that are known to be zero in the
-/// expression. These are provided to potentially allow the caller (which might
-/// recursively be SimplifyDemandedBits itself) to simplify the expression.
-/// KnownOne and KnownZero always follow the invariant that:
-///   KnownOne & KnownZero == 0.
-/// That is, a bit can't be both 1 and 0. Note that the bits in KnownOne and
-/// KnownZero may only be accurate for those bits set in DemandedMask. Note also
-/// that the bitwidth of V, DemandedMask, KnownZero and KnownOne must all be the
-/// same.
+/// expression. Known.Zero contains all the bits that are known to be zero in
+/// the expression. These are provided to potentially allow the caller (which
+/// might recursively be SimplifyDemandedBits itself) to simplify the
+/// expression.
+/// Known.One and Known.Zero always follow the invariant that:
+///   Known.One & Known.Zero == 0.
+/// That is, a bit can't be both 1 and 0. Note that the bits in Known.One and
+/// Known.Zero may only be accurate for those bits set in DemandedMask. Note
+/// also that the bitwidth of V, DemandedMask, Known.Zero and Known.One must all
+/// be the same.
 ///
 /// This returns null if it did not change anything and it permits no
 /// simplification.  This returns V itself if it did some simplification of V's
@@ -102,8 +104,7 @@ bool InstCombiner::SimplifyDemandedBits(Instruction *I, unsigned OpNo,
 /// some other non-null value if it found out that V is equal to another value
 /// in the context where the specified bits are demanded, but not for all users.
 Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
-                                             APInt &KnownZero, APInt &KnownOne,
-                                             unsigned Depth,
+                                             KnownBits &Known, unsigned Depth,
                                              Instruction *CxtI) {
   assert(V != nullptr && "Null pointer of Value???");
   assert(Depth <= 6 && "Limit Search Depth");
@@ -111,18 +112,16 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
   Type *VTy = V->getType();
   assert(
       (!VTy->isIntOrIntVectorTy() || VTy->getScalarSizeInBits() == BitWidth) &&
-      KnownZero.getBitWidth() == BitWidth &&
-      KnownOne.getBitWidth() == BitWidth &&
-      "Value *V, DemandedMask, KnownZero and KnownOne "
-      "must have same BitWidth");
+      Known.getBitWidth() == BitWidth &&
+      "Value *V, DemandedMask and Known must have same BitWidth");
 
   if (isa<Constant>(V)) {
-    computeKnownBits(V, KnownZero, KnownOne, Depth, CxtI);
+    computeKnownBits(V, Known, Depth, CxtI);
     return nullptr;
   }
 
-  KnownZero.clearAllBits();
-  KnownOne.clearAllBits();
+  Known.Zero.clearAllBits();
+  Known.One.clearAllBits();
   if (DemandedMask == 0)     // Not demanding any bits from V.
     return UndefValue::get(VTy);
 
@@ -131,7 +130,7 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
 
   Instruction *I = dyn_cast<Instruction>(V);
   if (!I) {
-    computeKnownBits(V, KnownZero, KnownOne, Depth, CxtI);
+    computeKnownBits(V, Known, Depth, CxtI);
     return nullptr;        // Only analyze instructions.
   }
 
@@ -139,11 +138,9 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
   // we can't do any simplifications of the operands, because DemandedMask
   // only reflects the bits demanded by *one* of the users.
   if (Depth != 0 && !I->hasOneUse())
-    return SimplifyMultipleUseDemandedBits(I, DemandedMask, KnownZero, KnownOne,
-                                           Depth, CxtI);
+    return SimplifyMultipleUseDemandedBits(I, DemandedMask, Known, Depth, CxtI);
 
-  APInt LHSKnownZero(BitWidth, 0), LHSKnownOne(BitWidth, 0);
-  APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
+  KnownBits LHSKnown(BitWidth), RHSKnown(BitWidth);
 
   // If this is the root being simplified, allow it to have multiple uses,
   // just set the DemandedMask to all bits so that we can try to simplify the
@@ -154,22 +151,21 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
 
   switch (I->getOpcode()) {
   default:
-    computeKnownBits(I, KnownZero, KnownOne, Depth, CxtI);
+    computeKnownBits(I, Known, Depth, CxtI);
     break;
   case Instruction::And: {
     // If either the LHS or the RHS are Zero, the result is zero.
-    if (SimplifyDemandedBits(I, 1, DemandedMask, RHSKnownZero, RHSKnownOne,
-                             Depth + 1) ||
-        SimplifyDemandedBits(I, 0, DemandedMask & ~RHSKnownZero, LHSKnownZero,
-                             LHSKnownOne, Depth + 1))
+    if (SimplifyDemandedBits(I, 1, DemandedMask, RHSKnown, Depth + 1) ||
+        SimplifyDemandedBits(I, 0, DemandedMask & ~RHSKnown.Zero, LHSKnown,
+                             Depth + 1))
       return I;
-    assert(!(RHSKnownZero & RHSKnownOne) && "Bits known to be one AND zero?");
-    assert(!(LHSKnownZero & LHSKnownOne) && "Bits known to be one AND zero?");
+    assert(!(RHSKnown.Zero & RHSKnown.One) && "Bits known to be one AND zero?");
+    assert(!(LHSKnown.Zero & LHSKnown.One) && "Bits known to be one AND zero?");
 
     // Output known-0 are known to be clear if zero in either the LHS | RHS.
-    APInt IKnownZero = RHSKnownZero | LHSKnownZero;
+    APInt IKnownZero = RHSKnown.Zero | LHSKnown.Zero;
     // Output known-1 bits are only known if set in both the LHS & RHS.
-    APInt IKnownOne = RHSKnownOne & LHSKnownOne;
+    APInt IKnownOne = RHSKnown.One & LHSKnown.One;
 
     // If the client is only demanding bits that we know, return the known
     // constant.
@@ -178,33 +174,32 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
 
     // If all of the demanded bits are known 1 on one side, return the other.
     // These bits cannot contribute to the result of the 'and'.
-    if (DemandedMask.isSubsetOf(LHSKnownZero | RHSKnownOne))
+    if (DemandedMask.isSubsetOf(LHSKnown.Zero | RHSKnown.One))
       return I->getOperand(0);
-    if (DemandedMask.isSubsetOf(RHSKnownZero | LHSKnownOne))
+    if (DemandedMask.isSubsetOf(RHSKnown.Zero | LHSKnown.One))
       return I->getOperand(1);
 
     // If the RHS is a constant, see if we can simplify it.
-    if (ShrinkDemandedConstant(I, 1, DemandedMask & ~LHSKnownZero))
+    if (ShrinkDemandedConstant(I, 1, DemandedMask & ~LHSKnown.Zero))
       return I;
 
-    KnownZero = std::move(IKnownZero);
-    KnownOne  = std::move(IKnownOne);
+    Known.Zero = std::move(IKnownZero);
+    Known.One  = std::move(IKnownOne);
     break;
   }
   case Instruction::Or: {
     // If either the LHS or the RHS are One, the result is One.
-    if (SimplifyDemandedBits(I, 1, DemandedMask, RHSKnownZero, RHSKnownOne,
-                             Depth + 1) ||
-        SimplifyDemandedBits(I, 0, DemandedMask & ~RHSKnownOne, LHSKnownZero,
-                             LHSKnownOne, Depth + 1))
+    if (SimplifyDemandedBits(I, 1, DemandedMask, RHSKnown, Depth + 1) ||
+        SimplifyDemandedBits(I, 0, DemandedMask & ~RHSKnown.One, LHSKnown,
+                             Depth + 1))
       return I;
-    assert(!(RHSKnownZero & RHSKnownOne) && "Bits known to be one AND zero?");
-    assert(!(LHSKnownZero & LHSKnownOne) && "Bits known to be one AND zero?");
+    assert(!(RHSKnown.Zero & RHSKnown.One) && "Bits known to be one AND zero?");
+    assert(!(LHSKnown.Zero & LHSKnown.One) && "Bits known to be one AND zero?");
 
     // Output known-0 bits are only known if clear in both the LHS & RHS.
-    APInt IKnownZero = RHSKnownZero & LHSKnownZero;
-    // Output known-1 are known to be set if set in either the LHS | RHS.
-    APInt IKnownOne = RHSKnownOne | LHSKnownOne;
+    APInt IKnownZero = RHSKnown.Zero & LHSKnown.Zero;
+    // Output known-1 are known. to be set if s.et in either the LHS | RHS.
+    APInt IKnownOne = RHSKnown.One | LHSKnown.One;
 
     // If the client is only demanding bits that we know, return the known
     // constant.
@@ -213,34 +208,32 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
 
     // If all of the demanded bits are known zero on one side, return the other.
     // These bits cannot contribute to the result of the 'or'.
-    if (DemandedMask.isSubsetOf(LHSKnownOne | RHSKnownZero))
+    if (DemandedMask.isSubsetOf(LHSKnown.One | RHSKnown.Zero))
       return I->getOperand(0);
-    if (DemandedMask.isSubsetOf(RHSKnownOne | LHSKnownZero))
+    if (DemandedMask.isSubsetOf(RHSKnown.One | LHSKnown.Zero))
       return I->getOperand(1);
 
     // If the RHS is a constant, see if we can simplify it.
     if (ShrinkDemandedConstant(I, 1, DemandedMask))
       return I;
 
-    KnownZero = std::move(IKnownZero);
-    KnownOne  = std::move(IKnownOne);
+    Known.Zero = std::move(IKnownZero);
+    Known.One  = std::move(IKnownOne);
     break;
   }
   case Instruction::Xor: {
-    if (SimplifyDemandedBits(I, 1, DemandedMask, RHSKnownZero, RHSKnownOne,
-                             Depth + 1) ||
-        SimplifyDemandedBits(I, 0, DemandedMask, LHSKnownZero, LHSKnownOne,
-                             Depth + 1))
+    if (SimplifyDemandedBits(I, 1, DemandedMask, RHSKnown, Depth + 1) ||
+        SimplifyDemandedBits(I, 0, DemandedMask, LHSKnown, Depth + 1))
       return I;
-    assert(!(RHSKnownZero & RHSKnownOne) && "Bits known to be one AND zero?");
-    assert(!(LHSKnownZero & LHSKnownOne) && "Bits known to be one AND zero?");
+    assert(!(RHSKnown.Zero & RHSKnown.One) && "Bits known to be one AND zero?");
+    assert(!(LHSKnown.Zero & LHSKnown.One) && "Bits known to be one AND zero?");
 
     // Output known-0 bits are known if clear or set in both the LHS & RHS.
-    APInt IKnownZero = (RHSKnownZero & LHSKnownZero) |
-                       (RHSKnownOne & LHSKnownOne);
+    APInt IKnownZero = (RHSKnown.Zero & LHSKnown.Zero) |
+                       (RHSKnown.One & LHSKnown.One);
     // Output known-1 are known to be set if set in only one of the LHS, RHS.
-    APInt IKnownOne =  (RHSKnownZero & LHSKnownOne) |
-                       (RHSKnownOne & LHSKnownZero);
+    APInt IKnownOne =  (RHSKnown.Zero & LHSKnown.One) |
+                       (RHSKnown.One & LHSKnown.Zero);
 
     // If the client is only demanding bits that we know, return the known
     // constant.
@@ -249,15 +242,15 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
 
     // If all of the demanded bits are known zero on one side, return the other.
     // These bits cannot contribute to the result of the 'xor'.
-    if (DemandedMask.isSubsetOf(RHSKnownZero))
+    if (DemandedMask.isSubsetOf(RHSKnown.Zero))
       return I->getOperand(0);
-    if (DemandedMask.isSubsetOf(LHSKnownZero))
+    if (DemandedMask.isSubsetOf(LHSKnown.Zero))
       return I->getOperand(1);
 
     // If all of the demanded bits are known to be zero on one side or the
     // other, turn this into an *inclusive* or.
     //    e.g. (A & C1)^(B & C2) -> (A & C1)|(B & C2) iff C1&C2 == 0
-    if (DemandedMask.isSubsetOf(RHSKnownZero | LHSKnownZero)) {
+    if (DemandedMask.isSubsetOf(RHSKnown.Zero | LHSKnown.Zero)) {
       Instruction *Or =
         BinaryOperator::CreateOr(I->getOperand(0), I->getOperand(1),
                                  I->getName());
@@ -268,10 +261,10 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
     // bits on that side are also known to be set on the other side, turn this
     // into an AND, as we know the bits will be cleared.
     //    e.g. (X | C1) ^ C2 --> (X | C1) & ~C2 iff (C1&C2) == C2
-    if (DemandedMask.isSubsetOf(RHSKnownZero|RHSKnownOne) &&
-        RHSKnownOne.isSubsetOf(LHSKnownOne)) {
+    if (DemandedMask.isSubsetOf(RHSKnown.Zero|RHSKnown.One) &&
+        RHSKnown.One.isSubsetOf(LHSKnown.One)) {
       Constant *AndC = Constant::getIntegerValue(VTy,
-                                                 ~RHSKnownOne & DemandedMask);
+                                                 ~RHSKnown.One & DemandedMask);
       Instruction *And = BinaryOperator::CreateAnd(I->getOperand(0), AndC);
       return InsertNewInstWith(And, *I);
     }
@@ -289,10 +282,10 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
       if (LHSInst->getOpcode() == Instruction::And && LHSInst->hasOneUse() &&
           isa<ConstantInt>(I->getOperand(1)) &&
           isa<ConstantInt>(LHSInst->getOperand(1)) &&
-          (LHSKnownOne & RHSKnownOne & DemandedMask) != 0) {
+          (LHSKnown.One & RHSKnown.One & DemandedMask) != 0) {
         ConstantInt *AndRHS = cast<ConstantInt>(LHSInst->getOperand(1));
         ConstantInt *XorRHS = cast<ConstantInt>(I->getOperand(1));
-        APInt NewMask = ~(LHSKnownOne & RHSKnownOne & DemandedMask);
+        APInt NewMask = ~(LHSKnown.One & RHSKnown.One & DemandedMask);
 
         Constant *AndC =
           ConstantInt::get(I->getType(), NewMask & AndRHS->getValue());
@@ -306,9 +299,9 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
       }
 
     // Output known-0 bits are known if clear or set in both the LHS & RHS.
-    KnownZero = std::move(IKnownZero);
+    Known.Zero = std::move(IKnownZero);
     // Output known-1 are known to be set if set in only one of the LHS, RHS.
-    KnownOne  = std::move(IKnownOne);
+    Known.One  = std::move(IKnownOne);
     break;
   }
   case Instruction::Select:
@@ -318,13 +311,11 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
     if (matchSelectPattern(I, LHS, RHS).Flavor != SPF_UNKNOWN)
       return nullptr;
 
-    if (SimplifyDemandedBits(I, 2, DemandedMask, RHSKnownZero, RHSKnownOne,
-                             Depth + 1) ||
-        SimplifyDemandedBits(I, 1, DemandedMask, LHSKnownZero, LHSKnownOne,
-                             Depth + 1))
+    if (SimplifyDemandedBits(I, 2, DemandedMask, RHSKnown, Depth + 1) ||
+        SimplifyDemandedBits(I, 1, DemandedMask, LHSKnown, Depth + 1))
       return I;
-    assert(!(RHSKnownZero & RHSKnownOne) && "Bits known to be one AND zero?");
-    assert(!(LHSKnownZero & LHSKnownOne) && "Bits known to be one AND zero?");
+    assert(!(RHSKnown.Zero & RHSKnown.One) && "Bits known to be one AND zero?");
+    assert(!(LHSKnown.Zero & LHSKnown.One) && "Bits known to be one AND zero?");
 
     // If the operands are constants, see if we can simplify them.
     if (ShrinkDemandedConstant(I, 1, DemandedMask) ||
@@ -332,21 +323,20 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
       return I;
 
     // Only known if known in both the LHS and RHS.
-    KnownOne = RHSKnownOne & LHSKnownOne;
-    KnownZero = RHSKnownZero & LHSKnownZero;
+    Known.One = RHSKnown.One & LHSKnown.One;
+    Known.Zero = RHSKnown.Zero & LHSKnown.Zero;
     break;
   case Instruction::Trunc: {
     unsigned truncBf = I->getOperand(0)->getType()->getScalarSizeInBits();
     DemandedMask = DemandedMask.zext(truncBf);
-    KnownZero = KnownZero.zext(truncBf);
-    KnownOne = KnownOne.zext(truncBf);
-    if (SimplifyDemandedBits(I, 0, DemandedMask, KnownZero, KnownOne,
-                             Depth + 1))
+    Known.Zero = Known.Zero.zext(truncBf);
+    Known.One = Known.One.zext(truncBf);
+    if (SimplifyDemandedBits(I, 0, DemandedMask, Known, Depth + 1))
       return I;
     DemandedMask = DemandedMask.trunc(BitWidth);
-    KnownZero = KnownZero.trunc(BitWidth);
-    KnownOne = KnownOne.trunc(BitWidth);
-    assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
+    Known.Zero = Known.Zero.trunc(BitWidth);
+    Known.One = Known.One.trunc(BitWidth);
+    assert(!(Known.Zero & Known.One) && "Bits known to be one AND zero?");
     break;
   }
   case Instruction::BitCast:
@@ -366,27 +356,25 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
       // Don't touch a vector-to-scalar bitcast.
       return nullptr;
 
-    if (SimplifyDemandedBits(I, 0, DemandedMask, KnownZero, KnownOne,
-                             Depth + 1))
+    if (SimplifyDemandedBits(I, 0, DemandedMask, Known, Depth + 1))
       return I;
-    assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
+    assert(!(Known.Zero & Known.One) && "Bits known to be one AND zero?");
     break;
   case Instruction::ZExt: {
     // Compute the bits in the result that are not present in the input.
     unsigned SrcBitWidth =I->getOperand(0)->getType()->getScalarSizeInBits();
 
     DemandedMask = DemandedMask.trunc(SrcBitWidth);
-    KnownZero = KnownZero.trunc(SrcBitWidth);
-    KnownOne = KnownOne.trunc(SrcBitWidth);
-    if (SimplifyDemandedBits(I, 0, DemandedMask, KnownZero, KnownOne,
-                             Depth + 1))
+    Known.Zero = Known.Zero.trunc(SrcBitWidth);
+    Known.One = Known.One.trunc(SrcBitWidth);
+    if (SimplifyDemandedBits(I, 0, DemandedMask, Known, Depth + 1))
       return I;
     DemandedMask = DemandedMask.zext(BitWidth);
-    KnownZero = KnownZero.zext(BitWidth);
-    KnownOne = KnownOne.zext(BitWidth);
-    assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
+    Known.Zero = Known.Zero.zext(BitWidth);
+    Known.One = Known.One.zext(BitWidth);
+    assert(!(Known.Zero & Known.One) && "Bits known to be one AND zero?");
     // The top bits are known to be zero.
-    KnownZero.setBitsFrom(SrcBitWidth);
+    Known.Zero.setBitsFrom(SrcBitWidth);
     break;
   }
   case Instruction::SExt: {
@@ -403,27 +391,26 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
       InputDemandedBits.setBit(SrcBitWidth-1);
 
     InputDemandedBits = InputDemandedBits.trunc(SrcBitWidth);
-    KnownZero = KnownZero.trunc(SrcBitWidth);
-    KnownOne = KnownOne.trunc(SrcBitWidth);
-    if (SimplifyDemandedBits(I, 0, InputDemandedBits, KnownZero, KnownOne,
-                             Depth + 1))
+    Known.Zero = Known.Zero.trunc(SrcBitWidth);
+    Known.One = Known.One.trunc(SrcBitWidth);
+    if (SimplifyDemandedBits(I, 0, InputDemandedBits, Known, Depth + 1))
       return I;
     InputDemandedBits = InputDemandedBits.zext(BitWidth);
-    KnownZero = KnownZero.zext(BitWidth);
-    KnownOne = KnownOne.zext(BitWidth);
-    assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
+    Known.Zero = Known.Zero.zext(BitWidth);
+    Known.One = Known.One.zext(BitWidth);
+    assert(!(Known.Zero & Known.One) && "Bits known to be one AND zero?");
 
     // If the sign bit of the input is known set or clear, then we know the
     // top bits of the result.
 
     // If the input sign bit is known zero, or if the NewBits are not demanded
     // convert this into a zero extension.
-    if (KnownZero[SrcBitWidth-1] || (NewBits & ~DemandedMask) == NewBits) {
+    if (Known.Zero[SrcBitWidth-1] || (NewBits & ~DemandedMask) == NewBits) {
       // Convert to ZExt cast
       CastInst *NewCast = new ZExtInst(I->getOperand(0), VTy, I->getName());
       return InsertNewInstWith(NewCast, *I);
-    } else if (KnownOne[SrcBitWidth-1]) {    // Input sign bit known set
-      KnownOne |= NewBits;
+    } else if (Known.One[SrcBitWidth-1]) {    // Input sign bit known set
+      Known.One |= NewBits;
     }
     break;
   }
@@ -437,11 +424,9 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
       // significant bit and all those below it.
       APInt DemandedFromOps(APInt::getLowBitsSet(BitWidth, BitWidth-NLZ));
       if (ShrinkDemandedConstant(I, 0, DemandedFromOps) ||
-          SimplifyDemandedBits(I, 0, DemandedFromOps, LHSKnownZero, LHSKnownOne,
-                               Depth + 1) ||
+          SimplifyDemandedBits(I, 0, DemandedFromOps, LHSKnown, Depth + 1) ||
           ShrinkDemandedConstant(I, 1, DemandedFromOps) ||
-          SimplifyDemandedBits(I, 1, DemandedFromOps, RHSKnownZero, RHSKnownOne,
-                               Depth + 1)) {
+          SimplifyDemandedBits(I, 1, DemandedFromOps, RHSKnown, Depth + 1)) {
         // Disable the nsw and nuw flags here: We can no longer guarantee that
         // we won't wrap after simplification. Removing the nsw/nuw flags is
         // legal here because the top bit is not demanded.
@@ -453,17 +438,17 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
 
       // If we are known to be adding/subtracting zeros to every bit below
       // the highest demanded bit, we just return the other side.
-      if (DemandedFromOps.isSubsetOf(RHSKnownZero))
+      if (DemandedFromOps.isSubsetOf(RHSKnown.Zero))
         return I->getOperand(0);
       // We can't do this with the LHS for subtraction.
       if (I->getOpcode() == Instruction::Add &&
-          DemandedFromOps.isSubsetOf(LHSKnownZero))
+          DemandedFromOps.isSubsetOf(LHSKnown.Zero))
         return I->getOperand(1);
     }
 
     // Otherwise just hand the add/sub off to computeKnownBits to fill in
     // the known zeros and ones.
-    computeKnownBits(V, KnownZero, KnownOne, Depth, CxtI);
+    computeKnownBits(V, Known, Depth, CxtI);
     break;
   }
   case Instruction::Shl: {
@@ -473,7 +458,7 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
       if (match(I->getOperand(0), m_Shr(m_Value(), m_APInt(ShrAmt)))) {
         Instruction *Shr = cast<Instruction>(I->getOperand(0));
         if (Value *R = simplifyShrShlDemandedBits(
-                Shr, *ShrAmt, I, *SA, DemandedMask, KnownZero, KnownOne))
+                Shr, *ShrAmt, I, *SA, DemandedMask, Known))
           return R;
       }
 
@@ -487,15 +472,14 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
       else if (IOp->hasNoUnsignedWrap())
         DemandedMaskIn.setHighBits(ShiftAmt);
 
-      if (SimplifyDemandedBits(I, 0, DemandedMaskIn, KnownZero, KnownOne,
-                               Depth + 1))
+      if (SimplifyDemandedBits(I, 0, DemandedMaskIn, Known, Depth + 1))
         return I;
-      assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
-      KnownZero <<= ShiftAmt;
-      KnownOne  <<= ShiftAmt;
+      assert(!(Known.Zero & Known.One) && "Bits known to be one AND zero?");
+      Known.Zero <<= ShiftAmt;
+      Known.One  <<= ShiftAmt;
       // low bits known zero.
       if (ShiftAmt)
-        KnownZero.setLowBits(ShiftAmt);
+        Known.Zero.setLowBits(ShiftAmt);
     }
     break;
   }
@@ -512,14 +496,13 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
       if (cast<LShrOperator>(I)->isExact())
         DemandedMaskIn.setLowBits(ShiftAmt);
 
-      if (SimplifyDemandedBits(I, 0, DemandedMaskIn, KnownZero, KnownOne,
-                               Depth + 1))
+      if (SimplifyDemandedBits(I, 0, DemandedMaskIn, Known, Depth + 1))
         return I;
-      assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
-      KnownZero.lshrInPlace(ShiftAmt);
-      KnownOne.lshrInPlace(ShiftAmt);
+      assert(!(Known.Zero & Known.One) && "Bits known to be one AND zero?");
+      Known.Zero.lshrInPlace(ShiftAmt);
+      Known.One.lshrInPlace(ShiftAmt);
       if (ShiftAmt)
-        KnownZero.setHighBits(ShiftAmt);  // high bits known zero.
+        Known.Zero.setHighBits(ShiftAmt);  // high bits known zero.
     }
     break;
   }
@@ -556,15 +539,14 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
       if (cast<AShrOperator>(I)->isExact())
         DemandedMaskIn.setLowBits(ShiftAmt);
 
-      if (SimplifyDemandedBits(I, 0, DemandedMaskIn, KnownZero, KnownOne,
-                               Depth + 1))
+      if (SimplifyDemandedBits(I, 0, DemandedMaskIn, Known, Depth + 1))
         return I;
 
-      assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
+      assert(!(Known.Zero & Known.One) && "Bits known to be one AND zero?");
       // Compute the new bits that are at the top now.
       APInt HighBits(APInt::getHighBitsSet(BitWidth, ShiftAmt));
-      KnownZero.lshrInPlace(ShiftAmt);
-      KnownOne.lshrInPlace(ShiftAmt);
+      Known.Zero.lshrInPlace(ShiftAmt);
+      Known.One.lshrInPlace(ShiftAmt);
 
       // Handle the sign bits.
       APInt SignMask(APInt::getSignMask(BitWidth));
@@ -573,14 +555,14 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
 
       // If the input sign bit is known to be zero, or if none of the top bits
       // are demanded, turn this into an unsigned shift right.
-      if (BitWidth <= ShiftAmt || KnownZero[BitWidth-ShiftAmt-1] ||
+      if (BitWidth <= ShiftAmt || Known.Zero[BitWidth-ShiftAmt-1] ||
           !DemandedMask.intersects(HighBits)) {
         BinaryOperator *LShr = BinaryOperator::CreateLShr(I->getOperand(0),
                                                           I->getOperand(1));
         LShr->setIsExact(cast<BinaryOperator>(I)->isExact());
         return InsertNewInstWith(LShr, *I);
-      } else if (KnownOne.intersects(SignMask)) { // New bits are known one.
-        KnownOne |= HighBits;
+      } else if (Known.One.intersects(SignMask)) { // New bits are known one.
+        Known.One |= HighBits;
       }
     }
     break;
@@ -598,25 +580,24 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
 
         APInt LowBits = RA - 1;
         APInt Mask2 = LowBits | APInt::getSignMask(BitWidth);
-        if (SimplifyDemandedBits(I, 0, Mask2, LHSKnownZero, LHSKnownOne,
-                                 Depth + 1))
+        if (SimplifyDemandedBits(I, 0, Mask2, LHSKnown, Depth + 1))
           return I;
 
         // The low bits of LHS are unchanged by the srem.
-        KnownZero = LHSKnownZero & LowBits;
-        KnownOne = LHSKnownOne & LowBits;
+        Known.Zero = LHSKnown.Zero & LowBits;
+        Known.One = LHSKnown.One & LowBits;
 
         // If LHS is non-negative or has all low bits zero, then the upper bits
         // are all zero.
-        if (LHSKnownZero.isSignBitSet() || LowBits.isSubsetOf(LHSKnownZero))
-          KnownZero |= ~LowBits;
+        if (LHSKnown.Zero.isSignBitSet() || LowBits.isSubsetOf(LHSKnown.Zero))
+          Known.Zero |= ~LowBits;
 
         // If LHS is negative and not all low bits are zero, then the upper bits
         // are all one.
-        if (LHSKnownOne.isSignBitSet() && LowBits.intersects(LHSKnownOne))
-          KnownOne |= ~LowBits;
+        if (LHSKnown.One.isSignBitSet() && LowBits.intersects(LHSKnown.One))
+          Known.One |= ~LowBits;
 
-        assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
+        assert(!(Known.Zero & Known.One) && "Bits known to be one AND zero?");
         break;
       }
     }
@@ -624,22 +605,21 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
     // The sign bit is the LHS's sign bit, except when the result of the
     // remainder is zero.
     if (DemandedMask.isSignBitSet()) {
-      computeKnownBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, Depth + 1,
-                       CxtI);
+      computeKnownBits(I->getOperand(0), LHSKnown, Depth + 1, CxtI);
       // If it's known zero, our sign bit is also zero.
-      if (LHSKnownZero.isSignBitSet())
-        KnownZero.setSignBit();
+      if (LHSKnown.Zero.isSignBitSet())
+        Known.Zero.setSignBit();
     }
     break;
   case Instruction::URem: {
-    APInt KnownZero2(BitWidth, 0), KnownOne2(BitWidth, 0);
+    KnownBits Known2(BitWidth);
     APInt AllOnes = APInt::getAllOnesValue(BitWidth);
-    if (SimplifyDemandedBits(I, 0, AllOnes, KnownZero2, KnownOne2, Depth + 1) ||
-        SimplifyDemandedBits(I, 1, AllOnes, KnownZero2, KnownOne2, Depth + 1))
+    if (SimplifyDemandedBits(I, 0, AllOnes, Known2, Depth + 1) ||
+        SimplifyDemandedBits(I, 1, AllOnes, Known2, Depth + 1))
       return I;
 
-    unsigned Leaders = KnownZero2.countLeadingOnes();
-    KnownZero = APInt::getHighBitsSet(BitWidth, Leaders) & DemandedMask;
+    unsigned Leaders = Known2.Zero.countLeadingOnes();
+    Known.Zero = APInt::getHighBitsSet(BitWidth, Leaders) & DemandedMask;
     break;
   }
   case Instruction::Call:
@@ -703,56 +683,54 @@ Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
           return ConstantInt::getNullValue(VTy);
 
         // We know that the upper bits are set to zero.
-        KnownZero.setBitsFrom(ArgWidth);
+        Known.Zero.setBitsFrom(ArgWidth);
         return nullptr;
       }
       case Intrinsic::x86_sse42_crc32_64_64:
-        KnownZero.setBitsFrom(32);
+        Known.Zero.setBitsFrom(32);
         return nullptr;
       }
     }
-    computeKnownBits(V, KnownZero, KnownOne, Depth, CxtI);
+    computeKnownBits(V, Known, Depth, CxtI);
     break;
   }
 
   // If the client is only demanding bits that we know, return the known
   // constant.
-  if (DemandedMask.isSubsetOf(KnownZero|KnownOne))
-    return Constant::getIntegerValue(VTy, KnownOne);
+  if (DemandedMask.isSubsetOf(Known.Zero|Known.One))
+    return Constant::getIntegerValue(VTy, Known.One);
   return nullptr;
 }
 
-/// Helper routine of SimplifyDemandedUseBits. It computes KnownZero/KnownOne
+/// Helper routine of SimplifyDemandedUseBits. It computes Known
 /// bits. It also tries to handle simplifications that can be done based on
 /// DemandedMask, but without modifying the Instruction.
 Value *InstCombiner::SimplifyMultipleUseDemandedBits(Instruction *I,
                                                      const APInt &DemandedMask,
-                                                     APInt &KnownZero,
-                                                     APInt &KnownOne,
+                                                     KnownBits &Known,
                                                      unsigned Depth,
                                                      Instruction *CxtI) {
   unsigned BitWidth = DemandedMask.getBitWidth();
   Type *ITy = I->getType();
 
-  APInt LHSKnownZero(BitWidth, 0), LHSKnownOne(BitWidth, 0);
-  APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
+  KnownBits LHSKnown(BitWidth);
+  KnownBits RHSKnown(BitWidth);
 
   // Despite the fact that we can't simplify this instruction in all User's
-  // context, we can at least compute the knownzero/knownone bits, and we can
+  // context, we can at least compute the known bits, and we can
   // do simplifications that apply to *just* the one user if we know that
   // this instruction has a simpler value in that context.
   switch (I->getOpcode()) {
   case Instruction::And: {
     // If either the LHS or the RHS are Zero, the result is zero.
-    computeKnownBits(I->getOperand(1), RHSKnownZero, RHSKnownOne, Depth + 1,
-                     CxtI);
-    computeKnownBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, Depth + 1,
+    computeKnownBits(I->getOperand(1), RHSKnown, Depth + 1, CxtI);
+    computeKnownBits(I->getOperand(0), LHSKnown, Depth + 1,
                      CxtI);
 
     // Output known-0 are known to be clear if zero in either the LHS | RHS.
-    APInt IKnownZero = RHSKnownZero | LHSKnownZero;
+    APInt IKnownZero = RHSKnown.Zero | LHSKnown.Zero;
     // Output known-1 bits are only known if set in both the LHS & RHS.
-    APInt IKnownOne = RHSKnownOne & LHSKnownOne;
+    APInt IKnownOne = RHSKnown.One & LHSKnown.One;
 
     // If the client is only demanding bits that we know, return the known
     // constant.
@@ -762,13 +740,13 @@ Value *InstCombiner::SimplifyMultipleUseDemandedBits(Instruction *I,
     // If all of the demanded bits are known 1 on one side, return the other.
     // These bits cannot contribute to the result of the 'and' in this
     // context.
-    if (DemandedMask.isSubsetOf(LHSKnownZero | RHSKnownOne))
+    if (DemandedMask.isSubsetOf(LHSKnown.Zero | RHSKnown.One))
       return I->getOperand(0);
-    if (DemandedMask.isSubsetOf(RHSKnownZero | LHSKnownOne))
+    if (DemandedMask.isSubsetOf(RHSKnown.Zero | LHSKnown.One))
       return I->getOperand(1);
 
-    KnownZero = std::move(IKnownZero);
-    KnownOne  = std::move(IKnownOne);
+    Known.Zero = std::move(IKnownZero);
+    Known.One  = std::move(IKnownOne);
     break;
   }
   case Instruction::Or: {
@@ -776,15 +754,14 @@ Value *InstCombiner::SimplifyMultipleUseDemandedBits(Instruction *I,
     // only bits from X or Y are demanded.
 
     // If either the LHS or the RHS are One, the result is One.
-    computeKnownBits(I->getOperand(1), RHSKnownZero, RHSKnownOne, Depth + 1,
-                     CxtI);
-    computeKnownBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, Depth + 1,
+    computeKnownBits(I->getOperand(1), RHSKnown, Depth + 1, CxtI);
+    computeKnownBits(I->getOperand(0), LHSKnown, Depth + 1,
                      CxtI);
 
     // Output known-0 bits are only known if clear in both the LHS & RHS.
-    APInt IKnownZero = RHSKnownZero & LHSKnownZero;
+    APInt IKnownZero = RHSKnown.Zero & LHSKnown.Zero;
     // Output known-1 are known to be set if set in either the LHS | RHS.
-    APInt IKnownOne = RHSKnownOne | LHSKnownOne;
+    APInt IKnownOne = RHSKnown.One | LHSKnown.One;
 
     // If the client is only demanding bits that we know, return the known
     // constant.
@@ -794,30 +771,29 @@ Value *InstCombiner::SimplifyMultipleUseDemandedBits(Instruction *I,
     // If all of the demanded bits are known zero on one side, return the
     // other.  These bits cannot contribute to the result of the 'or' in this
     // context.
-    if (DemandedMask.isSubsetOf(LHSKnownOne | RHSKnownZero))
+    if (DemandedMask.isSubsetOf(LHSKnown.One | RHSKnown.Zero))
       return I->getOperand(0);
-    if (DemandedMask.isSubsetOf(RHSKnownOne | LHSKnownZero))
+    if (DemandedMask.isSubsetOf(RHSKnown.One | LHSKnown.Zero))
       return I->getOperand(1);
 
-    KnownZero = std::move(IKnownZero);
-    KnownOne  = std::move(IKnownOne);
+    Known.Zero = std::move(IKnownZero);
+    Known.One  = std::move(IKnownOne);
     break;
   }
   case Instruction::Xor: {
     // We can simplify (X^Y) -> X or Y in the user's context if we know that
     // only bits from X or Y are demanded.
 
-    computeKnownBits(I->getOperand(1), RHSKnownZero, RHSKnownOne, Depth + 1,
-                     CxtI);
-    computeKnownBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, Depth + 1,
+    computeKnownBits(I->getOperand(1), RHSKnown, Depth + 1, CxtI);
+    computeKnownBits(I->getOperand(0), LHSKnown, Depth + 1,
                      CxtI);
 
     // Output known-0 bits are known if clear or set in both the LHS & RHS.
-    APInt IKnownZero = (RHSKnownZero & LHSKnownZero) |
-                       (RHSKnownOne & LHSKnownOne);
+    APInt IKnownZero = (RHSKnown.Zero & LHSKnown.Zero) |
+                       (RHSKnown.One & LHSKnown.One);
     // Output known-1 are known to be set if set in only one of the LHS, RHS.
-    APInt IKnownOne =  (RHSKnownZero & LHSKnownOne) |
-                       (RHSKnownOne & LHSKnownZero);
+    APInt IKnownOne =  (RHSKnown.Zero & LHSKnown.One) |
+                       (RHSKnown.One & LHSKnown.Zero);
 
     // If the client is only demanding bits that we know, return the known
     // constant.
@@ -826,25 +802,25 @@ Value *InstCombiner::SimplifyMultipleUseDemandedBits(Instruction *I,
 
     // If all of the demanded bits are known zero on one side, return the
     // other.
-    if (DemandedMask.isSubsetOf(RHSKnownZero))
+    if (DemandedMask.isSubsetOf(RHSKnown.Zero))
       return I->getOperand(0);
-    if (DemandedMask.isSubsetOf(LHSKnownZero))
+    if (DemandedMask.isSubsetOf(LHSKnown.Zero))
       return I->getOperand(1);
 
     // Output known-0 bits are known if clear or set in both the LHS & RHS.
-    KnownZero = std::move(IKnownZero);
+    Known.Zero = std::move(IKnownZero);
     // Output known-1 are known to be set if set in only one of the LHS, RHS.
-    KnownOne  = std::move(IKnownOne);
+    Known.One  = std::move(IKnownOne);
     break;
   }
   default:
-    // Compute the KnownZero/KnownOne bits to simplify things downstream.
-    computeKnownBits(I, KnownZero, KnownOne, Depth, CxtI);
+    // Compute the Known bits to simplify things downstream.
+    computeKnownBits(I, Known, Depth, CxtI);
 
     // If this user is only demanding bits that we know, return the known
     // constant.
-    if (DemandedMask.isSubsetOf(KnownZero|KnownOne))
-      return Constant::getIntegerValue(ITy, KnownOne);
+    if (DemandedMask.isSubsetOf(Known.Zero|Known.One))
+      return Constant::getIntegerValue(ITy, Known.One);
 
     break;
   }
@@ -874,7 +850,7 @@ Value *
 InstCombiner::simplifyShrShlDemandedBits(Instruction *Shr, const APInt &ShrOp1,
                                          Instruction *Shl, const APInt &ShlOp1,
                                          const APInt &DemandedMask,
-                                         APInt &KnownZero, APInt &KnownOne) {
+                                         KnownBits &Known) {
   if (!ShlOp1 || !ShrOp1)
     return nullptr; // No-op.
 
@@ -887,9 +863,9 @@ InstCombiner::simplifyShrShlDemandedBits(Instruction *Shr, const APInt &ShrOp1,
   unsigned ShlAmt = ShlOp1.getZExtValue();
   unsigned ShrAmt = ShrOp1.getZExtValue();
 
-  KnownOne.clearAllBits();
-  KnownZero.setLowBits(ShlAmt - 1);
-  KnownZero &= DemandedMask;
+  Known.One.clearAllBits();
+  Known.Zero.setLowBits(ShlAmt - 1);
+  Known.Zero &= DemandedMask;
 
   APInt BitMask1(APInt::getAllOnesValue(BitWidth));
   APInt BitMask2(APInt::getAllOnesValue(BitWidth));
index c23b45c..4729c79 100644 (file)
@@ -60,6 +60,7 @@
 #include "llvm/IR/ValueHandle.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Transforms/Scalar.h"
 #include "llvm/Transforms/Utils/Local.h"
@@ -2180,11 +2181,10 @@ Instruction *InstCombiner::visitReturnInst(ReturnInst &RI) {
 
   // There might be assume intrinsics dominating this return that completely
   // determine the value. If so, constant fold it.
-  unsigned BitWidth = VTy->getPrimitiveSizeInBits();
-  APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
-  computeKnownBits(ResultOp, KnownZero, KnownOne, 0, &RI);
-  if ((KnownZero|KnownOne).isAllOnesValue())
-    RI.setOperand(0, Constant::getIntegerValue(VTy, KnownOne));
+  KnownBits Known(VTy->getPrimitiveSizeInBits());
+  computeKnownBits(ResultOp, Known, 0, &RI);
+  if ((Known.Zero|Known.One).isAllOnesValue())
+    RI.setOperand(0, Constant::getIntegerValue(VTy, Known.One));
 
   return nullptr;
 }
@@ -2263,10 +2263,10 @@ Instruction *InstCombiner::visitSwitchInst(SwitchInst &SI) {
   }
 
   unsigned BitWidth = cast<IntegerType>(Cond->getType())->getBitWidth();
-  APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
-  computeKnownBits(Cond, KnownZero, KnownOne, 0, &SI);
-  unsigned LeadingKnownZeros = KnownZero.countLeadingOnes();
-  unsigned LeadingKnownOnes = KnownOne.countLeadingOnes();
+  KnownBits Known(BitWidth);
+  computeKnownBits(Cond, Known, 0, &SI);
+  unsigned LeadingKnownZeros = Known.Zero.countLeadingOnes();
+  unsigned LeadingKnownOnes = Known.One.countLeadingOnes();
 
   // Compute the number of leading bits we can ignore.
   // TODO: A better way to determine this would use ComputeNumSignBits().
@@ -2863,11 +2863,10 @@ bool InstCombiner::run() {
     Type *Ty = I->getType();
     if (ExpensiveCombines && !I->use_empty() && Ty->isIntOrIntVectorTy()) {
       unsigned BitWidth = Ty->getScalarSizeInBits();
-      APInt KnownZero(BitWidth, 0);
-      APInt KnownOne(BitWidth, 0);
-      computeKnownBits(I, KnownZero, KnownOne, /*Depth*/0, I);
-      if ((KnownZero | KnownOne).isAllOnesValue()) {
-        Constant *C = ConstantInt::get(Ty, KnownOne);
+      KnownBits Known(BitWidth);
+      computeKnownBits(I, Known, /*Depth*/0, I);
+      if ((Known.Zero | Known.One).isAllOnesValue()) {
+        Constant *C = ConstantInt::get(Ty, Known.One);
         DEBUG(dbgs() << "IC: ConstFold (all bits known) to: " << *C <<
                         " from: " << *I << '\n');
 
index 7019287..48eda09 100644 (file)
@@ -51,6 +51,7 @@
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/PatternMatch.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Transforms/Scalar.h"
 
 using namespace llvm;
@@ -537,9 +538,9 @@ bool GuardWideningImpl::parseRangeChecks(
     } else if (match(Check.getBase(),
                      m_Or(m_Value(OpLHS), m_ConstantInt(OpRHS)))) {
       unsigned BitWidth = OpLHS->getType()->getScalarSizeInBits();
-      APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
-      computeKnownBits(OpLHS, KnownZero, KnownOne, DL);
-      if ((OpRHS->getValue() & KnownZero) == OpRHS->getValue()) {
+      KnownBits Known(BitWidth);
+      computeKnownBits(OpLHS, Known, DL);
+      if ((OpRHS->getValue() & Known.Zero) == OpRHS->getValue()) {
         Check.setBase(OpLHS);
         APInt NewOffset = Check.getOffsetValue() + OpRHS->getValue();
         Check.setOffset(ConstantInt::get(Ctx, NewOffset));
index 1cfe3bd..7ffdad5 100644 (file)
@@ -22,6 +22,7 @@
 #include "llvm/IR/Function.h"
 #include "llvm/IR/IRBuilder.h"
 #include "llvm/IR/Instructions.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Transforms/Utils/Local.h"
 
 using namespace llvm;
@@ -256,14 +257,14 @@ ValueRange FastDivInsertionTask::getValueRange(Value *V,
   unsigned HiBits = LongLen - ShortLen;
 
   const DataLayout &DL = SlowDivOrRem->getModule()->getDataLayout();
-  APInt Zeros(LongLen, 0), Ones(LongLen, 0);
+  KnownBits Known(LongLen);
 
-  computeKnownBits(V, Zeros, Ones, DL);
+  computeKnownBits(V, Known, DL);
 
-  if (Zeros.countLeadingOnes() >= HiBits)
+  if (Known.Zero.countLeadingOnes() >= HiBits)
     return VALRNG_KNOWN_SHORT;
 
-  if (Ones.countLeadingZeros() < HiBits)
+  if (Known.One.countLeadingZeros() < HiBits)
     return VALRNG_LIKELY_LONG;
 
   // Long integer divisions are often used in hashtable implementations. It's
index 357c303..d3002c5 100644 (file)
@@ -45,6 +45,7 @@
 #include "llvm/IR/PatternMatch.h"
 #include "llvm/IR/ValueHandle.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
 using namespace llvm;
@@ -1038,9 +1039,9 @@ unsigned llvm::getOrEnforceKnownAlignment(Value *V, unsigned PrefAlign,
          "getOrEnforceKnownAlignment expects a pointer!");
   unsigned BitWidth = DL.getPointerTypeSizeInBits(V->getType());
 
-  APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
-  computeKnownBits(V, KnownZero, KnownOne, DL, 0, AC, CxtI, DT);
-  unsigned TrailZ = KnownZero.countTrailingOnes();
+  KnownBits Known(BitWidth);
+  computeKnownBits(V, Known, DL, 0, AC, CxtI, DT);
+  unsigned TrailZ = Known.Zero.countTrailingOnes();
 
   // Avoid trouble with ridiculously large TrailZ values, such as
   // those computed from a null pointer.
index 1c2c763..f86e97b 100644 (file)
@@ -60,6 +60,7 @@
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
@@ -4367,8 +4368,8 @@ static bool EliminateDeadSwitchCases(SwitchInst *SI, AssumptionCache *AC,
                                      const DataLayout &DL) {
   Value *Cond = SI->getCondition();
   unsigned Bits = Cond->getType()->getIntegerBitWidth();
-  APInt KnownZero(Bits, 0), KnownOne(Bits, 0);
-  computeKnownBits(Cond, KnownZero, KnownOne, DL, 0, AC, SI);
+  KnownBits Known(Bits);
+  computeKnownBits(Cond, Known, DL, 0, AC, SI);
 
   // We can also eliminate cases by determining that their values are outside of
   // the limited range of the condition based on how many significant (non-sign)
@@ -4380,7 +4381,7 @@ static bool EliminateDeadSwitchCases(SwitchInst *SI, AssumptionCache *AC,
   SmallVector<ConstantInt *, 8> DeadCases;
   for (auto &Case : SI->cases()) {
     APInt CaseVal = Case.getCaseValue()->getValue();
-    if (KnownZero.intersects(CaseVal) || !KnownOne.isSubsetOf(CaseVal) ||
+    if (Known.Zero.intersects(CaseVal) || !Known.One.isSubsetOf(CaseVal) ||
         (CaseVal.getMinSignedBits() > MaxSignificantBitsInCond)) {
       DeadCases.push_back(Case.getCaseValue());
       DEBUG(dbgs() << "SimplifyCFG: switch case " << CaseVal << " is dead.\n");
@@ -4394,7 +4395,7 @@ static bool EliminateDeadSwitchCases(SwitchInst *SI, AssumptionCache *AC,
   bool HasDefault =
       !isa<UnreachableInst>(SI->getDefaultDest()->getFirstNonPHIOrDbg());
   const unsigned NumUnknownBits =
-      Bits - (KnownZero | KnownOne).countPopulation();
+      Bits - (Known.Zero | Known.One).countPopulation();
   assert(NumUnknownBits <= Bits);
   if (HasDefault && DeadCases.empty() &&
       NumUnknownBits < 64 /* avoid overflow */ &&
index 2640c1f..2c1c304 100644 (file)
@@ -30,6 +30,7 @@
 #include "llvm/IR/Module.h"
 #include "llvm/IR/PatternMatch.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Transforms/Utils/BuildLibCalls.h"
 #include "llvm/Transforms/Utils/Local.h"
 
@@ -455,11 +456,9 @@ Value *LibCallSimplifier::optimizeStrLen(CallInst *CI, IRBuilder<> &B) {
      
       Value *Offset = GEP->getOperand(2);
       unsigned BitWidth = Offset->getType()->getIntegerBitWidth();
-      APInt KnownZero(BitWidth, 0);
-      APInt KnownOne(BitWidth, 0);
-      computeKnownBits(Offset, KnownZero, KnownOne, DL, 0, nullptr, CI, 
-                       nullptr);
-      KnownZero.flipAllBits();
+      KnownBits Known(BitWidth);
+      computeKnownBits(Offset, Known, DL, 0, nullptr, CI, nullptr);
+      Known.Zero.flipAllBits();
       size_t ArrSize = 
              cast<ArrayType>(GEP->getSourceElementType())->getNumElements();
 
@@ -473,7 +472,7 @@ Value *LibCallSimplifier::optimizeStrLen(CallInst *CI, IRBuilder<> &B) {
       // optimize if we can prove that the program has undefined behavior when 
       // Offset is outside that range. That is the case when GEP->getOperand(0) 
       // is a pointer to an object whose memory extent is NullTermIdx+1.
-      if ((KnownZero.isNonNegative() && KnownZero.ule(NullTermIdx)) || 
+      if ((Known.Zero.isNonNegative() && Known.Zero.ule(NullTermIdx)) || 
           (GEP->isInBounds() && isa<GlobalVariable>(GEP->getOperand(0)) &&
            NullTermIdx == ArrSize - 1))
         return B.CreateSub(ConstantInt::get(CI->getType(), NullTermIdx), 
index b0324d0..97dcb40 100644 (file)
@@ -30,6 +30,7 @@
 #include "llvm/IR/Value.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/KnownBits.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Transforms/Vectorize.h"
@@ -343,10 +344,9 @@ bool Vectorizer::isConsecutiveAccess(Value *A, Value *B) {
   // If any bits are known to be zero other than the sign bit in OpA, we can
   // add 1 to it while guaranteeing no overflow of any sort.
   if (!Safe) {
-    APInt KnownZero(BitWidth, 0);
-    APInt KnownOne(BitWidth, 0);
-    computeKnownBits(OpA, KnownZero, KnownOne, DL, 0, nullptr, OpA, &DT);
-    if (KnownZero.countTrailingZeros() < (BitWidth - 1))
+    KnownBits Known(BitWidth);
+    computeKnownBits(OpA, Known, DL, 0, nullptr, OpA, &DT);
+    if (Known.Zero.countTrailingZeros() < (BitWidth - 1))
       Safe = true;
   }