[InstSimplify][InstCombine] Modify the interface of decomposeBitTestICmp and use it in the InstSimplify

[craig.topper]

Summary

This addresses a fixme in InstSimplify about using decomposeBitTest.

I've modified the interface a little to return only the APInt version of the mask that InstSimplify needs. InstCombine now has a small wrapper routine to create a Constant out of it. I've also dropped the returning of 0 since InstSimplify doesn't need that. So InstCombine creates a zero constant itself.

I also had to make decomposeBitTest support vectors since InstSimplify needs that.

As InstSimplify can't use something from the Transforms library, I've moved the CmpInstAnalysis code to the Analysis library.

I'm happy to split this up into smaller pieces if that makes review easier.

[spatel]

LGTM.

[spatel]

Wait - this does have a testable change, right? You can check the ult/ugt cases from InstSimplify with this:
http://rise4fun.com/Alive/XAV

llvm/trunk/include/llvm/Transforms/Utils/CmpInstAnalysis.h

-//===-- CmpInstAnalysis.h - Utils to help fold compare insts ----*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file holds routines to help analyse compare instructions
-// and fold them into constants or other compare instructions
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef LLVM_TRANSFORMS_UTILS_CMPINSTANALYSIS_H
-#define LLVM_TRANSFORMS_UTILS_CMPINSTANALYSIS_H
-
-#include "llvm/IR/InstrTypes.h"
-
-namespace llvm {
-  class ICmpInst;
-  class Value;
-
-  /// Encode a icmp predicate into a three bit mask. These bits are carefully
-  /// arranged to allow folding of expressions such as:
-  ///
-  ///      (A < B) | (A > B) --> (A != B)
-  ///
-  /// Note that this is only valid if the first and second predicates have the
-  /// same sign. It is illegal to do: (A u< B) | (A s> B)
-  ///
-  /// Three bits are used to represent the condition, as follows:
-  ///   0  A > B
-  ///   1  A == B
-  ///   2  A < B
-  ///
-  /// <=>  Value  Definition
-  /// 000     0   Always false
-  /// 001     1   A >  B
-  /// 010     2   A == B
-  /// 011     3   A >= B
-  /// 100     4   A <  B
-  /// 101     5   A != B
-  /// 110     6   A <= B
-  /// 111     7   Always true
-  ///
-  unsigned getICmpCode(const ICmpInst *ICI, bool InvertPred = false);
-
-  /// This is the complement of getICmpCode, which turns an opcode and two
-  /// operands into either a constant true or false, or the predicate for a new
-  /// ICmp instruction. The sign is passed in to determine which kind of
-  /// predicate to use in the new icmp instruction.
-  /// Non-NULL return value will be a true or false constant.
-  /// NULL return means a new ICmp is needed. The predicate for which is output
-  /// in NewICmpPred.
-  Value *getICmpValue(bool Sign, unsigned Code, Value *LHS, Value *RHS,
-                      CmpInst::Predicate &NewICmpPred);
-
-  /// Return true if both predicates match sign or if at least one of them is an
-  /// equality comparison (which is signless).
-  bool PredicatesFoldable(CmpInst::Predicate p1, CmpInst::Predicate p2);
-
-  /// Decompose an icmp into the form ((X & Y) pred Z) if possible. The returned
-  /// predicate is either == or !=. Returns false if decomposition fails.
-  bool decomposeBitTestICmp(const ICmpInst *I, CmpInst::Predicate &Pred,
-                            Value *&X, Value *&Y, Value *&Z);
-
-} // end namespace llvm
-
-#endif

llvm/trunk/lib/Analysis/CMakeLists.txt

add_llvm_library(LLVMAnalysis
   CFGPrinter.cpp
   CFLAndersAliasAnalysis.cpp
   CFLSteensAliasAnalysis.cpp
   CGSCCPassManager.cpp
   CallGraph.cpp
   CallGraphSCCPass.cpp
   CallPrinter.cpp
   CaptureTracking.cpp
+  CmpInstAnalysis.cpp
   CostModel.cpp
   CodeMetrics.cpp
   ConstantFolding.cpp
   Delinearization.cpp
   DemandedBits.cpp
   DependenceAnalysis.cpp
   DivergenceAnalysis.cpp
   DomPrinter.cpp

llvm/trunk/lib/Analysis/InstructionSimplify.cpp

 //===----------------------------------------------------------------------===//
 
 #include "llvm/Analysis/InstructionSimplify.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/Analysis/AssumptionCache.h"
 #include "llvm/Analysis/CaptureTracking.h"
+#include "llvm/Analysis/CmpInstAnalysis.h"
 #include "llvm/Analysis/ConstantFolding.h"
 #include "llvm/Analysis/LoopAnalysisManager.h"
 #include "llvm/Analysis/MemoryBuiltins.h"
 #include "llvm/Analysis/OptimizationDiagnosticInfo.h"
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/Analysis/VectorUtils.h"
 #include "llvm/IR/ConstantRange.h"
 #include "llvm/IR/DataLayout.h"
if (TrueVal == X && match(FalseVal, m_Or(m_Specific(X), m_APInt(C))) &&
       return TrueWhenUnset ? TrueVal : FalseVal;
   }
 
   return nullptr;
 }
 
 /// An alternative way to test if a bit is set or not uses sgt/slt instead of
 /// eq/ne.
-static Value *simplifySelectWithFakeICmpEq(Value *CmpLHS, Value *TrueVal,
-                                           Value *FalseVal,
-                                           bool TrueWhenUnset) {
+static Value *simplifySelectWithFakeICmpEq(Value *CmpLHS, Value *CmpRHS,
+                                           ICmpInst::Predicate Pred,
+                                           Value *TrueVal, Value *FalseVal) {
+  Value *X;
+  APInt Mask;
+  if (!decomposeBitTestICmp(CmpLHS, CmpRHS, Pred, X, Mask))
+    return nullptr;
+
   unsigned BitWidth = TrueVal->getType()->getScalarSizeInBits();
   if (!BitWidth)
     return nullptr;
 
-  APInt MinSignedValue;
-  Value *X;
-  if (match(CmpLHS, m_Trunc(m_Value(X))) && (X == TrueVal || X == FalseVal)) {
+  Value *ExtX;
+  if (match(X, m_Trunc(m_Value(ExtX))) &&
+      (ExtX == TrueVal || ExtX == FalseVal)) {
     // icmp slt (trunc X), 0  <--> icmp ne (and X, C), 0
     // icmp sgt (trunc X), -1 <--> icmp eq (and X, C), 0
-    unsigned DestSize = CmpLHS->getType()->getScalarSizeInBits();
-    MinSignedValue = APInt::getSignedMinValue(DestSize).zext(BitWidth);
-  } else {
-    // icmp slt X, 0  <--> icmp ne (and X, C), 0
-    // icmp sgt X, -1 <--> icmp eq (and X, C), 0
-    X = CmpLHS;
-    MinSignedValue = APInt::getSignedMinValue(BitWidth);
+    X = ExtX;
+    Mask = Mask.zext(BitWidth);
   }
 
-  if (Value *V = simplifySelectBitTest(TrueVal, FalseVal, X, &MinSignedValue,
-                                       TrueWhenUnset))
-    return V;
-
-  return nullptr;
+  return simplifySelectBitTest(TrueVal, FalseVal, X, &Mask,
+                               Pred == ICmpInst::ICMP_EQ);
 }
 
 /// Try to simplify a select instruction when its condition operand is an
 /// integer comparison.
 static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal,
                                          Value *FalseVal, const SimplifyQuery &Q,
                                          unsigned MaxRecurse) {
   ICmpInst::Predicate Pred;
   Value *CmpLHS, *CmpRHS;
   if (!match(CondVal, m_ICmp(Pred, m_Value(CmpLHS), m_Value(CmpRHS))))
     return nullptr;
 
-  // FIXME: This code is nearly duplicated in InstCombine. Using/refactoring
-  // decomposeBitTestICmp() might help.
-  // FIXME this should support ICMP_SLE/SGE forms as well
   if (ICmpInst::isEquality(Pred) && match(CmpRHS, m_Zero())) {
     Value *X;
     const APInt *Y;
     if (match(CmpLHS, m_And(m_Value(X), m_APInt(Y))))
       if (Value *V = simplifySelectBitTest(TrueVal, FalseVal, X, Y,
                                            Pred == ICmpInst::ICMP_EQ))
         return V;
-  } else if (Pred == ICmpInst::ICMP_SLT && match(CmpRHS, m_Zero())) {
-    // Comparing signed-less-than 0 checks if the sign bit is set.
-    if (Value *V = simplifySelectWithFakeICmpEq(CmpLHS, TrueVal, FalseVal,
-                                                false))
-      return V;
-  } else if (Pred == ICmpInst::ICMP_SGT && match(CmpRHS, m_AllOnes())) {
-    // Comparing signed-greater-than -1 checks if the sign bit is not set.
-    if (Value *V = simplifySelectWithFakeICmpEq(CmpLHS, TrueVal, FalseVal,
-                                                true))
-      return V;
   }
 
+  // Check for other compares that behave like bit test.
+  if (Value *V = simplifySelectWithFakeICmpEq(CmpLHS, CmpRHS, Pred,
+                                              TrueVal, FalseVal))
+    return V;
+
   if (CondVal->hasOneUse()) {
     const APInt *C;
     if (match(CmpRHS, m_APInt(C))) {
       // X < MIN ? T : F  -->  F
       if (Pred == ICmpInst::ICMP_SLT && C->isMinSignedValue())
         return FalseVal;
       // X < MIN ? T : F  -->  F
       if (Pred == ICmpInst::ICMP_ULT && C->isMinValue())

llvm/trunk/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp

 //===- InstCombineAndOrXor.cpp --------------------------------------------===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 //
 // This file implements the visitAnd, visitOr, and visitXor functions.
 //
 //===----------------------------------------------------------------------===//
 
 #include "InstCombineInternal.h"
+#include "llvm/Analysis/CmpInstAnalysis.h"
 #include "llvm/Analysis/InstructionSimplify.h"
 #include "llvm/IR/ConstantRange.h"
 #include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/PatternMatch.h"
-#include "llvm/Transforms/Utils/CmpInstAnalysis.h"
 #include "llvm/Transforms/Utils/Local.h"
 using namespace llvm;
 using namespace PatternMatch;
 
 #define DEBUG_TYPE "instcombine"
 
 /// Similar to getICmpCode but for FCmpInst. This encodes a fcmp predicate into
 /// a four bit mask.
static unsigned conjugateICmpMask(unsigned Mask) {
 
   NewMask |= (Mask & (AMask_NotAllOnes | BMask_NotAllOnes | Mask_NotAllZeros |
                       AMask_NotMixed | BMask_NotMixed))
              >> 1;
 
   return NewMask;
 }
 
+// Adapts the external decomposeBitTestICmp for local use.
+static bool decomposeBitTestICmp(Value *LHS, Value *RHS, CmpInst::Predicate &Pred,
+                                 Value *&X, Value *&Y, Value *&Z) {
+  APInt Mask;
+  if (!llvm::decomposeBitTestICmp(LHS, RHS, Pred, X, Mask))
+    return false;
+
+  Y = ConstantInt::get(X->getType(), Mask);
+  Z = ConstantInt::get(X->getType(), 0);
+  return true;
+}
+
 /// Handle (icmp(A & B) ==/!= C) &/| (icmp(A & D) ==/!= E).
 /// Return the set of pattern classes (from MaskedICmpType) that both LHS and
 /// RHS satisfy.
 static unsigned getMaskedTypeForICmpPair(Value *&A, Value *&B, Value *&C,
                                          Value *&D, Value *&E, ICmpInst *LHS,
                                          ICmpInst *RHS,
                                          ICmpInst::Predicate &PredL,
                                          ICmpInst::Predicate &PredR) {
   if (LHS->getOperand(0)->getType() != RHS->getOperand(0)->getType())
     return 0;
   // vectors are not (yet?) supported
   if (LHS->getOperand(0)->getType()->isVectorTy())
     return 0;
 
   // Here comes the tricky part:
   // LHS might be of the form L11 & L12 == X, X == L21 & L22,
   // and L11 & L12 == L21 & L22. The same goes for RHS.
   // Now we must find those components L** and R**, that are equal, so
   // that we can extract the parameters A, B, C, D, and E for the canonical
   // above.
   Value *L1 = LHS->getOperand(0);
   Value *L2 = LHS->getOperand(1);
   Value *L11, *L12, *L21, *L22;
   // Check whether the icmp can be decomposed into a bit test.
-  if (decomposeBitTestICmp(LHS, PredL, L11, L12, L2)) {
+  if (decomposeBitTestICmp(L1, L2, PredL, L11, L12, L2)) {
     L21 = L22 = L1 = nullptr;
   } else {
     // Look for ANDs in the LHS icmp.
     if (!L1->getType()->isIntegerTy()) {
       // You can icmp pointers, for example. They really aren't masks.
       L11 = L12 = nullptr;
     } else if (!match(L1, m_And(m_Value(L11), m_Value(L12)))) {
       // Any icmp can be viewed as being trivially masked; if it allows us to
static unsigned getMaskedTypeForICmpPair(Value *&A, Value *&B, Value *&C,
   // Bail if LHS was a icmp that can't be decomposed into an equality.
   if (!ICmpInst::isEquality(PredL))
     return 0;
 
   Value *R1 = RHS->getOperand(0);
   Value *R2 = RHS->getOperand(1);
   Value *R11, *R12;
   bool Ok = false;
-  if (decomposeBitTestICmp(RHS, PredR, R11, R12, R2)) {
+  if (decomposeBitTestICmp(R1, R2, PredR, R11, R12, R2)) {
     if (R11 == L11 || R11 == L12 || R11 == L21 || R11 == L22) {
       A = R11;
       D = R12;
     } else if (R12 == L11 || R12 == L12 || R12 == L21 || R12 == L22) {
       A = R12;
       D = R11;
     } else {
       return 0;

llvm/trunk/lib/Transforms/Utils/CMakeLists.txt

 add_llvm_library(LLVMTransformUtils
   ASanStackFrameLayout.cpp
   AddDiscriminators.cpp
   BasicBlockUtils.cpp
   BreakCriticalEdges.cpp
   BuildLibCalls.cpp
   BypassSlowDivision.cpp
   CloneFunction.cpp
   CloneModule.cpp
-  CmpInstAnalysis.cpp
   CodeExtractor.cpp
   CtorUtils.cpp
   DemoteRegToStack.cpp
   EscapeEnumerator.cpp
   Evaluator.cpp
   FlattenCFG.cpp
   FunctionComparator.cpp
   FunctionImportUtils.cpp

llvm/trunk/lib/Transforms/Utils/CmpInstAnalysis.cpp

-//===- CmpInstAnalysis.cpp - Utils to help fold compares ---------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file holds routines to help analyse compare instructions
-// and fold them into constants or other compare instructions
-//
-//===----------------------------------------------------------------------===//
-
-#include "llvm/Transforms/Utils/CmpInstAnalysis.h"
-#include "llvm/IR/Constants.h"
-#include "llvm/IR/Instructions.h"
-
-using namespace llvm;
-
-unsigned llvm::getICmpCode(const ICmpInst *ICI, bool InvertPred) {
-  ICmpInst::Predicate Pred = InvertPred ? ICI->getInversePredicate()
-                                        : ICI->getPredicate();
-  switch (Pred) {
-      // False -> 0
-    case ICmpInst::ICMP_UGT: return 1;  // 001
-    case ICmpInst::ICMP_SGT: return 1;  // 001
-    case ICmpInst::ICMP_EQ:  return 2;  // 010
-    case ICmpInst::ICMP_UGE: return 3;  // 011
-    case ICmpInst::ICMP_SGE: return 3;  // 011
-    case ICmpInst::ICMP_ULT: return 4;  // 100
-    case ICmpInst::ICMP_SLT: return 4;  // 100
-    case ICmpInst::ICMP_NE:  return 5;  // 101
-    case ICmpInst::ICMP_ULE: return 6;  // 110
-    case ICmpInst::ICMP_SLE: return 6;  // 110
-      // True -> 7
-    default:
-      llvm_unreachable("Invalid ICmp predicate!");
-  }
-}
-
-Value *llvm::getICmpValue(bool Sign, unsigned Code, Value *LHS, Value *RHS,
-                          CmpInst::Predicate &NewICmpPred) {
-  switch (Code) {
-    default: llvm_unreachable("Illegal ICmp code!");
-    case 0: // False.
-      return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);
-    case 1: NewICmpPred = Sign ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT; break;
-    case 2: NewICmpPred = ICmpInst::ICMP_EQ; break;
-    case 3: NewICmpPred = Sign ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE; break;
-    case 4: NewICmpPred = Sign ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT; break;
-    case 5: NewICmpPred = ICmpInst::ICMP_NE; break;
-    case 6: NewICmpPred = Sign ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE; break;
-    case 7: // True.
-      return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 1);
-  }
-  return nullptr;
-}
-
-bool llvm::PredicatesFoldable(ICmpInst::Predicate p1, ICmpInst::Predicate p2) {
-  return (CmpInst::isSigned(p1) == CmpInst::isSigned(p2)) ||
-         (CmpInst::isSigned(p1) && ICmpInst::isEquality(p2)) ||
-         (CmpInst::isSigned(p2) && ICmpInst::isEquality(p1));
-}
-
-bool llvm::decomposeBitTestICmp(const ICmpInst *I, CmpInst::Predicate &Pred,
-                                Value *&X, Value *&Y, Value *&Z) {
-  ConstantInt *C = dyn_cast<ConstantInt>(I->getOperand(1));
-  if (!C)
-    return false;
-
-  switch (I->getPredicate()) {
-  default:
-    return false;
-  case ICmpInst::ICMP_SLT:
-    // X < 0 is equivalent to (X & SignMask) != 0.
-    if (!C->isZero())
-      return false;
-    Y = ConstantInt::get(I->getContext(), APInt::getSignMask(C->getBitWidth()));
-    Pred = ICmpInst::ICMP_NE;
-    break;
-  case ICmpInst::ICMP_SGT:
-    // X > -1 is equivalent to (X & SignMask) == 0.
-    if (!C->isMinusOne())
-      return false;
-    Y = ConstantInt::get(I->getContext(), APInt::getSignMask(C->getBitWidth()));
-    Pred = ICmpInst::ICMP_EQ;
-    break;
-  case ICmpInst::ICMP_ULT:
-    // X <u 2^n is equivalent to (X & ~(2^n-1)) == 0.
-    if (!C->getValue().isPowerOf2())
-      return false;
-    Y = ConstantInt::get(I->getContext(), -C->getValue());
-    Pred = ICmpInst::ICMP_EQ;
-    break;
-  case ICmpInst::ICMP_UGT:
-    // X >u 2^n-1 is equivalent to (X & ~(2^n-1)) != 0.
-    if (!(C->getValue() + 1).isPowerOf2())
-      return false;
-    Y = ConstantInt::get(I->getContext(), ~C->getValue());
-    Pred = ICmpInst::ICMP_NE;
-    break;
-  }
-
-  X = I->getOperand(0);
-  Z = ConstantInt::getNullValue(C->getType());
-  return true;
-}

llvm/trunk/test/Transforms/InstSimplify/select.ll

 ; CHECK-NEXT:    ret <2 x i8> [[AND]]
 ;
   %cmp = icmp sgt <2 x i8> %X, <i8 -1, i8 -1>
   %and = and <2 x i8> %X, <i8 127, i8 127>
   %sel = select <2 x i1> %cmp, <2 x i8> %X, <2 x i8> %and
   ret <2 x i8> %sel
 }
 
-; TODO: we should be able to simplify this
 define i32 @test12(i32 %X) {
 ; CHECK-LABEL: @test12(
-; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[X:%.*]], 4
-; CHECK-NEXT:    [[AND:%.*]] = and i32 [[X]], 3
-; CHECK-NEXT:    [[COND:%.*]] = select i1 [[CMP]], i32 [[X]], i32 [[AND]]
-; CHECK-NEXT:    ret i32 [[COND]]
+; CHECK-NEXT:    [[AND:%.*]] = and i32 [[X:%.*]], 3
+; CHECK-NEXT:    ret i32 [[AND]]
 ;
   %cmp = icmp ult i32 %X, 4
   %and = and i32 %X, 3
   %cond = select i1 %cmp, i32 %X, i32 %and
   ret i32 %cond
 }
 
 ; Same as above, but the compare isn't canonical
 ; TODO: we should be able to simplify this
 define i32 @test12noncanon(i32 %X) {
 ; CHECK-LABEL: @test12noncanon(
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp ule i32 [[X:%.*]], 3
 ; CHECK-NEXT:    [[AND:%.*]] = and i32 [[X]], 3
 ; CHECK-NEXT:    [[COND:%.*]] = select i1 [[CMP]], i32 [[X]], i32 [[AND]]
 ; CHECK-NEXT:    ret i32 [[COND]]
 ;
   %cmp = icmp ule i32 %X, 3
   %and = and i32 %X, 3
   %cond = select i1 %cmp, i32 %X, i32 %and
   ret i32 %cond
 }
 
-; TODO: we should be able to simplify this
 define i32 @test13(i32 %X) {
 ; CHECK-LABEL: @test13(
-; CHECK-NEXT:    [[CMP:%.*]] = icmp ugt i32 [[X:%.*]], 3
-; CHECK-NEXT:    [[AND:%.*]] = and i32 [[X]], 3
-; CHECK-NEXT:    [[COND:%.*]] = select i1 [[CMP]], i32 [[AND]], i32 [[X]]
-; CHECK-NEXT:    ret i32 [[COND]]
+; CHECK-NEXT:    [[AND:%.*]] = and i32 [[X:%.*]], 3
+; CHECK-NEXT:    ret i32 [[AND]]
 ;
   %cmp = icmp ugt i32 %X, 3
   %and = and i32 %X, 3
   %cond = select i1 %cmp, i32 %and, i32 %X
   ret i32 %cond
 }
 
 ; Same as above, but the compare isn't canonical