//===----------------------------------------------------------------------===//
//
// This pass reassociates commutative expressions in an order that is designed
-// to promote better constant propagation, GCSE, LICM, PRE...
+// to promote better constant propagation, GCSE, LICM, PRE, etc.
//
// For example: 4 + (x + 5) -> x + (4 + 5)
//
#define DEBUG_TYPE "reassociate"
#include "llvm/Transforms/Scalar.h"
+#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Function.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/DenseMap.h"
#include <algorithm>
-#include <map>
using namespace llvm;
STATISTIC(NumLinear , "Number of insts linearized");
#ifndef NDEBUG
/// PrintOps - Print out the expression identified in the Ops list.
///
-static void PrintOps(Instruction *I, const std::vector<ValueEntry> &Ops) {
+static void PrintOps(Instruction *I, const SmallVectorImpl<ValueEntry> &Ops) {
Module *M = I->getParent()->getParent()->getParent();
- errs() << Instruction::getOpcodeName(I->getOpcode()) << " "
+ dbgs() << Instruction::getOpcodeName(I->getOpcode()) << " "
<< *Ops[0].Op->getType() << '\t';
for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
- errs() << "[ ";
- WriteAsOperand(errs(), Ops[i].Op, false, M);
- errs() << ", #" << Ops[i].Rank << "] ";
+ dbgs() << "[ ";
+ WriteAsOperand(dbgs(), Ops[i].Op, false, M);
+ dbgs() << ", #" << Ops[i].Rank << "] ";
}
}
#endif
namespace {
class Reassociate : public FunctionPass {
- std::map<BasicBlock*, unsigned> RankMap;
- std::map<AssertingVH<>, unsigned> ValueRankMap;
+ DenseMap<BasicBlock*, unsigned> RankMap;
+ DenseMap<AssertingVH<>, unsigned> ValueRankMap;
+ SmallVector<WeakVH, 8> RedoInsts;
+ SmallVector<WeakVH, 8> DeadInsts;
bool MadeChange;
public:
static char ID; // Pass identification, replacement for typeid
- Reassociate() : FunctionPass(&ID) {}
+ Reassociate() : FunctionPass(ID) {
+ initializeReassociatePass(*PassRegistry::getPassRegistry());
+ }
bool runOnFunction(Function &F);
private:
void BuildRankMap(Function &F);
unsigned getRank(Value *V);
- void ReassociateExpression(BinaryOperator *I);
- void RewriteExprTree(BinaryOperator *I, std::vector<ValueEntry> &Ops,
+ Value *ReassociateExpression(BinaryOperator *I);
+ void RewriteExprTree(BinaryOperator *I, SmallVectorImpl<ValueEntry> &Ops,
unsigned Idx = 0);
- Value *OptimizeExpression(BinaryOperator *I, std::vector<ValueEntry> &Ops);
- Value *OptimizeAdd(Instruction *I, std::vector<ValueEntry> &Ops);
- void LinearizeExprTree(BinaryOperator *I, std::vector<ValueEntry> &Ops);
+ Value *OptimizeExpression(BinaryOperator *I,
+ SmallVectorImpl<ValueEntry> &Ops);
+ Value *OptimizeAdd(Instruction *I, SmallVectorImpl<ValueEntry> &Ops);
+ void LinearizeExprTree(BinaryOperator *I, SmallVectorImpl<ValueEntry> &Ops);
void LinearizeExpr(BinaryOperator *I);
Value *RemoveFactorFromExpression(Value *V, Value *Factor);
- void ReassociateBB(BasicBlock *BB);
+ void ReassociateInst(BasicBlock::iterator &BBI);
void RemoveDeadBinaryOp(Value *V);
};
}
char Reassociate::ID = 0;
-static RegisterPass<Reassociate> X("reassociate", "Reassociate expressions");
+INITIALIZE_PASS(Reassociate, "reassociate",
+ "Reassociate expressions", false, false)
// Public interface to the Reassociate pass
FunctionPass *llvm::createReassociatePass() { return new Reassociate(); }
void Reassociate::RemoveDeadBinaryOp(Value *V) {
Instruction *Op = dyn_cast<Instruction>(V);
- if (!Op || !isa<BinaryOperator>(Op) || !isa<CmpInst>(Op) || !Op->use_empty())
+ if (!Op || !isa<BinaryOperator>(Op))
return;
Value *LHS = Op->getOperand(0), *RHS = Op->getOperand(1);
+
+ ValueRankMap.erase(Op);
+ DeadInsts.push_back(Op);
RemoveDeadBinaryOp(LHS);
RemoveDeadBinaryOp(RHS);
}
}
unsigned Reassociate::getRank(Value *V) {
- if (isa<Argument>(V)) return ValueRankMap[V]; // Function argument...
-
Instruction *I = dyn_cast<Instruction>(V);
- if (I == 0) return 0; // Otherwise it's a global or constant, rank 0.
+ if (I == 0) {
+ if (isa<Argument>(V)) return ValueRankMap[V]; // Function argument.
+ return 0; // Otherwise it's a global or constant, rank 0.
+ }
- unsigned &CachedRank = ValueRankMap[I];
- if (CachedRank) return CachedRank; // Rank already known?
+ if (unsigned Rank = ValueRankMap[I])
+ return Rank; // Rank already known?
// If this is an expression, return the 1+MAX(rank(LHS), rank(RHS)) so that
// we can reassociate expressions for code motion! Since we do not recurse
// If this is a not or neg instruction, do not count it for rank. This
// assures us that X and ~X will have the same rank.
- if (!I->getType()->isInteger() ||
+ if (!I->getType()->isIntegerTy() ||
(!BinaryOperator::isNot(I) && !BinaryOperator::isNeg(I)))
++Rank;
- //DEBUG(errs() << "Calculated Rank[" << V->getName() << "] = "
+ //DEBUG(dbgs() << "Calculated Rank[" << V->getName() << "] = "
// << Rank << "\n");
- return CachedRank = Rank;
+ return ValueRankMap[I] = Rank;
}
/// isReassociableOp - Return true if V is an instruction of the specified
/// LowerNegateToMultiply - Replace 0-X with X*-1.
///
static Instruction *LowerNegateToMultiply(Instruction *Neg,
- std::map<AssertingVH<>, unsigned> &ValueRankMap) {
+ DenseMap<AssertingVH<>, unsigned> &ValueRankMap) {
Constant *Cst = Constant::getAllOnesValue(Neg->getType());
Instruction *Res = BinaryOperator::CreateMul(Neg->getOperand(1), Cst, "",Neg);
ValueRankMap.erase(Neg);
Res->takeName(Neg);
Neg->replaceAllUsesWith(Res);
+ Res->setDebugLoc(Neg->getDebugLoc());
Neg->eraseFromParent();
return Res;
}
isReassociableOp(RHS, I->getOpcode()) &&
"Not an expression that needs linearization?");
- DEBUG(errs() << "Linear" << *LHS << '\n' << *RHS << '\n' << *I << '\n');
+ DEBUG(dbgs() << "Linear" << *LHS << '\n' << *RHS << '\n' << *I << '\n');
// Move the RHS instruction to live immediately before I, avoiding breaking
// dominator properties.
RHS->setOperand(0, LHS);
I->setOperand(0, RHS);
+ // Conservatively clear all the optional flags, which may not hold
+ // after the reassociation.
+ I->clearSubclassOptionalData();
+ LHS->clearSubclassOptionalData();
+ RHS->clearSubclassOptionalData();
+
++NumLinear;
MadeChange = true;
- DEBUG(errs() << "Linearized: " << *I << '\n');
+ DEBUG(dbgs() << "Linearized: " << *I << '\n');
// If D is part of this expression tree, tail recurse.
if (isReassociableOp(I->getOperand(1), I->getOpcode()))
/// LinearizeExprTree - Given an associative binary expression tree, traverse
/// all of the uses putting it into canonical form. This forces a left-linear
-/// form of the the expression (((a+b)+c)+d), and collects information about the
+/// form of the expression (((a+b)+c)+d), and collects information about the
/// rank of the non-tree operands.
///
/// NOTE: These intentionally destroys the expression tree operands (turning
/// caller MUST use something like RewriteExprTree to put the values back in.
///
void Reassociate::LinearizeExprTree(BinaryOperator *I,
- std::vector<ValueEntry> &Ops) {
+ SmallVectorImpl<ValueEntry> &Ops) {
Value *LHS = I->getOperand(0), *RHS = I->getOperand(1);
unsigned Opcode = I->getOpcode();
std::swap(LHS, RHS);
bool Success = !I->swapOperands();
assert(Success && "swapOperands failed");
- Success = false;
+ (void)Success;
MadeChange = true;
} else if (RHSBO) {
- // Turn (A+B)+(C+D) -> (((A+B)+C)+D). This guarantees the the RHS is not
+ // Turn (A+B)+(C+D) -> (((A+B)+C)+D). This guarantees the RHS is not
// part of the expression tree.
LinearizeExpr(I);
LHS = LHSBO = cast<BinaryOperator>(I->getOperand(0));
// linearized and optimized, emit them in-order. This function is written to be
// tail recursive.
void Reassociate::RewriteExprTree(BinaryOperator *I,
- std::vector<ValueEntry> &Ops,
+ SmallVectorImpl<ValueEntry> &Ops,
unsigned i) {
if (i+2 == Ops.size()) {
if (I->getOperand(0) != Ops[i].Op ||
I->getOperand(1) != Ops[i+1].Op) {
Value *OldLHS = I->getOperand(0);
- DEBUG(errs() << "RA: " << *I << '\n');
+ DEBUG(dbgs() << "RA: " << *I << '\n');
I->setOperand(0, Ops[i].Op);
I->setOperand(1, Ops[i+1].Op);
- DEBUG(errs() << "TO: " << *I << '\n');
+
+ // Clear all the optional flags, which may not hold after the
+ // reassociation if the expression involved more than just this operation.
+ if (Ops.size() != 2)
+ I->clearSubclassOptionalData();
+
+ DEBUG(dbgs() << "TO: " << *I << '\n');
MadeChange = true;
++NumChanged;
assert(i+2 < Ops.size() && "Ops index out of range!");
if (I->getOperand(1) != Ops[i].Op) {
- DEBUG(errs() << "RA: " << *I << '\n');
+ DEBUG(dbgs() << "RA: " << *I << '\n');
I->setOperand(1, Ops[i].Op);
- DEBUG(errs() << "TO: " << *I << '\n');
+
+ // Conservatively clear all the optional flags, which may not hold
+ // after the reassociation.
+ I->clearSubclassOptionalData();
+
+ DEBUG(dbgs() << "TO: " << *I << '\n');
MadeChange = true;
++NumChanged;
}
// that should be processed next by the reassociation pass.
//
static Value *NegateValue(Value *V, Instruction *BI) {
+ if (Constant *C = dyn_cast<Constant>(V))
+ return ConstantExpr::getNeg(C);
+
// We are trying to expose opportunity for reassociation. One of the things
// that we want to do to achieve this is to push a negation as deep into an
// expression chain as possible, to expose the add instructions. In practice,
// X = -(A+12+C+D) into X = -A + -12 + -C + -D = -12 + -A + -C + -D
// so that later, a: Y = 12+X could get reassociated with the -12 to eliminate
// the constants. We assume that instcombine will clean up the mess later if
- // we introduce tons of unnecessary negation instructions...
+ // we introduce tons of unnecessary negation instructions.
//
if (Instruction *I = dyn_cast<Instruction>(V))
if (I->getOpcode() == Instruction::Add && I->hasOneUse()) {
I->setName(I->getName()+".neg");
return I;
}
+
+ // Okay, we need to materialize a negated version of V with an instruction.
+ // Scan the use lists of V to see if we have one already.
+ for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;++UI){
+ User *U = *UI;
+ if (!BinaryOperator::isNeg(U)) continue;
+
+ // We found one! Now we have to make sure that the definition dominates
+ // this use. We do this by moving it to the entry block (if it is a
+ // non-instruction value) or right after the definition. These negates will
+ // be zapped by reassociate later, so we don't need much finesse here.
+ BinaryOperator *TheNeg = cast<BinaryOperator>(U);
+
+ // Verify that the negate is in this function, V might be a constant expr.
+ if (TheNeg->getParent()->getParent() != BI->getParent()->getParent())
+ continue;
+
+ BasicBlock::iterator InsertPt;
+ if (Instruction *InstInput = dyn_cast<Instruction>(V)) {
+ if (InvokeInst *II = dyn_cast<InvokeInst>(InstInput)) {
+ InsertPt = II->getNormalDest()->begin();
+ } else {
+ InsertPt = InstInput;
+ ++InsertPt;
+ }
+ while (isa<PHINode>(InsertPt)) ++InsertPt;
+ } else {
+ InsertPt = TheNeg->getParent()->getParent()->getEntryBlock().begin();
+ }
+ TheNeg->moveBefore(InsertPt);
+ return TheNeg;
+ }
// Insert a 'neg' instruction that subtracts the value from zero to get the
// negation.
- //
return BinaryOperator::CreateNeg(V, V->getName() + ".neg", BI);
}
/// only used by an add, transform this into (X+(0-Y)) to promote better
/// reassociation.
static Instruction *BreakUpSubtract(Instruction *Sub,
- std::map<AssertingVH<>, unsigned> &ValueRankMap) {
- // Convert a subtract into an add and a neg instruction... so that sub
- // instructions can be commuted with other add instructions...
+ DenseMap<AssertingVH<>, unsigned> &ValueRankMap) {
+ // Convert a subtract into an add and a neg instruction. This allows sub
+ // instructions to be commuted with other add instructions.
//
- // Calculate the negative value of Operand 1 of the sub instruction...
- // and set it as the RHS of the add instruction we just made...
+ // Calculate the negative value of Operand 1 of the sub instruction,
+ // and set it as the RHS of the add instruction we just made.
//
Value *NegVal = NegateValue(Sub->getOperand(1), Sub);
Instruction *New =
// Everyone now refers to the add instruction.
ValueRankMap.erase(Sub);
Sub->replaceAllUsesWith(New);
+ New->setDebugLoc(Sub->getDebugLoc());
Sub->eraseFromParent();
- DEBUG(errs() << "Negated: " << *New << '\n');
+ DEBUG(dbgs() << "Negated: " << *New << '\n');
return New;
}
/// by one, change this into a multiply by a constant to assist with further
/// reassociation.
static Instruction *ConvertShiftToMul(Instruction *Shl,
- std::map<AssertingVH<>, unsigned> &ValueRankMap) {
+ DenseMap<AssertingVH<>, unsigned> &ValueRankMap) {
// If an operand of this shift is a reassociable multiply, or if the shift
// is used by a reassociable multiply or add, turn into a multiply.
if (isReassociableOp(Shl->getOperand(0), Instruction::Mul) ||
ValueRankMap.erase(Shl);
Mul->takeName(Shl);
Shl->replaceAllUsesWith(Mul);
+ Mul->setDebugLoc(Shl->getDebugLoc());
Shl->eraseFromParent();
return Mul;
}
}
// Scan backwards and forwards among values with the same rank as element i to
-// see if X exists. If X does not exist, return i.
-static unsigned FindInOperandList(std::vector<ValueEntry> &Ops, unsigned i,
+// see if X exists. If X does not exist, return i. This is useful when
+// scanning for 'x' when we see '-x' because they both get the same rank.
+static unsigned FindInOperandList(SmallVectorImpl<ValueEntry> &Ops, unsigned i,
Value *X) {
unsigned XRank = Ops[i].Rank;
unsigned e = Ops.size();
for (unsigned j = i+1; j != e && Ops[j].Rank == XRank; ++j)
if (Ops[j].Op == X)
return j;
- // Scan backwards
+ // Scan backwards.
for (unsigned j = i-1; j != ~0U && Ops[j].Rank == XRank; --j)
if (Ops[j].Op == X)
return j;
BinaryOperator *BO = isReassociableOp(V, Instruction::Mul);
if (!BO) return 0;
- std::vector<ValueEntry> Factors;
+ SmallVector<ValueEntry, 8> Factors;
LinearizeExprTree(BO, Factors);
bool FoundFactor = false;
- for (unsigned i = 0, e = Factors.size(); i != e; ++i)
+ bool NeedsNegate = false;
+ for (unsigned i = 0, e = Factors.size(); i != e; ++i) {
if (Factors[i].Op == Factor) {
FoundFactor = true;
Factors.erase(Factors.begin()+i);
break;
}
+
+ // If this is a negative version of this factor, remove it.
+ if (ConstantInt *FC1 = dyn_cast<ConstantInt>(Factor))
+ if (ConstantInt *FC2 = dyn_cast<ConstantInt>(Factors[i].Op))
+ if (FC1->getValue() == -FC2->getValue()) {
+ FoundFactor = NeedsNegate = true;
+ Factors.erase(Factors.begin()+i);
+ break;
+ }
+ }
+
if (!FoundFactor) {
// Make sure to restore the operands to the expression tree.
RewriteExprTree(BO, Factors);
return 0;
}
- if (Factors.size() == 1) return Factors[0].Op;
+ BasicBlock::iterator InsertPt = BO; ++InsertPt;
- RewriteExprTree(BO, Factors);
- return BO;
+ // If this was just a single multiply, remove the multiply and return the only
+ // remaining operand.
+ if (Factors.size() == 1) {
+ ValueRankMap.erase(BO);
+ DeadInsts.push_back(BO);
+ V = Factors[0].Op;
+ } else {
+ RewriteExprTree(BO, Factors);
+ V = BO;
+ }
+
+ if (NeedsNegate)
+ V = BinaryOperator::CreateNeg(V, "neg", InsertPt);
+
+ return V;
}
/// FindSingleUseMultiplyFactors - If V is a single-use multiply, recursively
/// add its operands as factors, otherwise add V to the list of factors.
+///
+/// Ops is the top-level list of add operands we're trying to factor.
static void FindSingleUseMultiplyFactors(Value *V,
- SmallVectorImpl<Value*> &Factors) {
+ SmallVectorImpl<Value*> &Factors,
+ const SmallVectorImpl<ValueEntry> &Ops,
+ bool IsRoot) {
BinaryOperator *BO;
- if ((!V->hasOneUse() && !V->use_empty()) ||
+ if (!(V->hasOneUse() || V->use_empty()) || // More than one use.
!(BO = dyn_cast<BinaryOperator>(V)) ||
BO->getOpcode() != Instruction::Mul) {
Factors.push_back(V);
return;
}
+ // If this value has a single use because it is another input to the add
+ // tree we're reassociating and we dropped its use, it actually has two
+ // uses and we can't factor it.
+ if (!IsRoot) {
+ for (unsigned i = 0, e = Ops.size(); i != e; ++i)
+ if (Ops[i].Op == V) {
+ Factors.push_back(V);
+ return;
+ }
+ }
+
+
// Otherwise, add the LHS and RHS to the list of factors.
- FindSingleUseMultiplyFactors(BO->getOperand(1), Factors);
- FindSingleUseMultiplyFactors(BO->getOperand(0), Factors);
+ FindSingleUseMultiplyFactors(BO->getOperand(1), Factors, Ops, false);
+ FindSingleUseMultiplyFactors(BO->getOperand(0), Factors, Ops, false);
}
/// OptimizeAndOrXor - Optimize a series of operands to an 'and', 'or', or 'xor'
/// instruction. This optimizes based on identities. If it can be reduced to
/// a single Value, it is returned, otherwise the Ops list is mutated as
/// necessary.
-static Value *OptimizeAndOrXor(unsigned Opcode, std::vector<ValueEntry> &Ops) {
+static Value *OptimizeAndOrXor(unsigned Opcode,
+ SmallVectorImpl<ValueEntry> &Ops) {
// Scan the operand lists looking for X and ~X pairs, along with X,X pairs.
// If we find any, we can simplify the expression. X&~X == 0, X|~X == -1.
for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
assert(i < Ops.size());
if (i+1 != Ops.size() && Ops[i+1].Op == Ops[i].Op) {
if (Opcode == Instruction::And || Opcode == Instruction::Or) {
- // Drop duplicate values.
+ // Drop duplicate values for And and Or.
Ops.erase(Ops.begin()+i);
--i; --e;
++NumAnnihil;
- } else {
- assert(Opcode == Instruction::Xor);
- if (e == 2)
- return Constant::getNullValue(Ops[0].Op->getType());
-
- // ... X^X -> ...
- Ops.erase(Ops.begin()+i, Ops.begin()+i+2);
- i -= 1; e -= 2;
- ++NumAnnihil;
+ continue;
}
+
+ // Drop pairs of values for Xor.
+ assert(Opcode == Instruction::Xor);
+ if (e == 2)
+ return Constant::getNullValue(Ops[0].Op->getType());
+
+ // Y ^ X^X -> Y
+ Ops.erase(Ops.begin()+i, Ops.begin()+i+2);
+ i -= 1; e -= 2;
+ ++NumAnnihil;
}
}
return 0;
/// OptimizeAdd - Optimize a series of operands to an 'add' instruction. This
/// optimizes based on identities. If it can be reduced to a single Value, it
/// is returned, otherwise the Ops list is mutated as necessary.
-Value *Reassociate::OptimizeAdd(Instruction *I, std::vector<ValueEntry> &Ops) {
+Value *Reassociate::OptimizeAdd(Instruction *I,
+ SmallVectorImpl<ValueEntry> &Ops) {
// Scan the operand lists looking for X and -X pairs. If we find any, we
- // can simplify the expression. X+-X == 0.
+ // can simplify the expression. X+-X == 0. While we're at it, scan for any
+ // duplicates. We want to canonicalize Y+Y+Y+Z -> 3*Y+Z.
+ //
+ // TODO: We could handle "X + ~X" -> "-1" if we wanted, since "-X = ~X+1".
+ //
for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
- assert(i < Ops.size());
+ Value *TheOp = Ops[i].Op;
+ // Check to see if we've seen this operand before. If so, we factor all
+ // instances of the operand together. Due to our sorting criteria, we know
+ // that these need to be next to each other in the vector.
+ if (i+1 != Ops.size() && Ops[i+1].Op == TheOp) {
+ // Rescan the list, remove all instances of this operand from the expr.
+ unsigned NumFound = 0;
+ do {
+ Ops.erase(Ops.begin()+i);
+ ++NumFound;
+ } while (i != Ops.size() && Ops[i].Op == TheOp);
+
+ DEBUG(errs() << "\nFACTORING [" << NumFound << "]: " << *TheOp << '\n');
+ ++NumFactor;
+
+ // Insert a new multiply.
+ Value *Mul = ConstantInt::get(cast<IntegerType>(I->getType()), NumFound);
+ Mul = BinaryOperator::CreateMul(TheOp, Mul, "factor", I);
+
+ // Now that we have inserted a multiply, optimize it. This allows us to
+ // handle cases that require multiple factoring steps, such as this:
+ // (X*2) + (X*2) + (X*2) -> (X*2)*3 -> X*6
+ RedoInsts.push_back(Mul);
+
+ // If every add operand was a duplicate, return the multiply.
+ if (Ops.empty())
+ return Mul;
+
+ // Otherwise, we had some input that didn't have the dupe, such as
+ // "A + A + B" -> "A*2 + B". Add the new multiply to the list of
+ // things being added by this operation.
+ Ops.insert(Ops.begin(), ValueEntry(getRank(Mul), Mul));
+
+ --i;
+ e = Ops.size();
+ continue;
+ }
+
// Check for X and -X in the operand list.
- if (!BinaryOperator::isNeg(Ops[i].Op))
+ if (!BinaryOperator::isNeg(TheOp))
continue;
- Value *X = BinaryOperator::getNegArgument(Ops[i].Op);
+ Value *X = BinaryOperator::getNegArgument(TheOp);
unsigned FoundX = FindInOperandList(Ops, i, X);
if (FoundX == i)
continue;
// Keep track of each multiply we see, to avoid triggering on (X*4)+(X*4)
// where they are actually the same multiply.
- SmallPtrSet<BinaryOperator*, 4> Multiplies;
unsigned MaxOcc = 0;
Value *MaxOccVal = 0;
for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
if (BOp == 0 || BOp->getOpcode() != Instruction::Mul || !BOp->use_empty())
continue;
- // If we've already seen this multiply, don't revisit it.
- if (!Multiplies.insert(BOp)) continue;
-
// Compute all of the factors of this added value.
SmallVector<Value*, 8> Factors;
- FindSingleUseMultiplyFactors(BOp, Factors);
+ FindSingleUseMultiplyFactors(BOp, Factors, Ops, true);
assert(Factors.size() > 1 && "Bad linearize!");
// Add one to FactorOccurrences for each unique factor in this op.
- if (Factors.size() == 2) {
- unsigned Occ = ++FactorOccurrences[Factors[0]];
- if (Occ > MaxOcc) { MaxOcc = Occ; MaxOccVal = Factors[0]; }
- if (Factors[0] != Factors[1]) { // Don't double count A*A.
- Occ = ++FactorOccurrences[Factors[1]];
- if (Occ > MaxOcc) { MaxOcc = Occ; MaxOccVal = Factors[1]; }
- }
- } else {
- SmallPtrSet<Value*, 4> Duplicates;
- for (unsigned i = 0, e = Factors.size(); i != e; ++i) {
- if (!Duplicates.insert(Factors[i])) continue;
-
- unsigned Occ = ++FactorOccurrences[Factors[i]];
- if (Occ > MaxOcc) { MaxOcc = Occ; MaxOccVal = Factors[i]; }
- }
+ SmallPtrSet<Value*, 8> Duplicates;
+ for (unsigned i = 0, e = Factors.size(); i != e; ++i) {
+ Value *Factor = Factors[i];
+ if (!Duplicates.insert(Factor)) continue;
+
+ unsigned Occ = ++FactorOccurrences[Factor];
+ if (Occ > MaxOcc) { MaxOcc = Occ; MaxOccVal = Factor; }
+
+ // If Factor is a negative constant, add the negated value as a factor
+ // because we can percolate the negate out. Watch for minint, which
+ // cannot be positivified.
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(Factor))
+ if (CI->isNegative() && !CI->isMinValue(true)) {
+ Factor = ConstantInt::get(CI->getContext(), -CI->getValue());
+ assert(!Duplicates.count(Factor) &&
+ "Shouldn't have two constant factors, missed a canonicalize");
+
+ unsigned Occ = ++FactorOccurrences[Factor];
+ if (Occ > MaxOcc) { MaxOcc = Occ; MaxOccVal = Factor; }
+ }
}
}
// If any factor occurred more than one time, we can pull it out.
if (MaxOcc > 1) {
- DEBUG(errs() << "\nFACTORING [" << MaxOcc << "]: " << *MaxOccVal << "\n");
+ DEBUG(errs() << "\nFACTORING [" << MaxOcc << "]: " << *MaxOccVal << '\n');
++NumFactor;
// Create a new instruction that uses the MaxOccVal twice. If we don't do
// RemoveFactorFromExpression on successive values to behave differently.
Instruction *DummyInst = BinaryOperator::CreateAdd(MaxOccVal, MaxOccVal);
SmallVector<Value*, 4> NewMulOps;
- for (unsigned i = 0, e = Ops.size(); i != e; ++i) {
+ for (unsigned i = 0; i != Ops.size(); ++i) {
+ // Only try to remove factors from expressions we're allowed to.
+ BinaryOperator *BOp = dyn_cast<BinaryOperator>(Ops[i].Op);
+ if (BOp == 0 || BOp->getOpcode() != Instruction::Mul || !BOp->use_empty())
+ continue;
+
if (Value *V = RemoveFactorFromExpression(Ops[i].Op, MaxOccVal)) {
- NewMulOps.push_back(V);
- Ops.erase(Ops.begin()+i);
- --i; --e;
+ // The factorized operand may occur several times. Convert them all in
+ // one fell swoop.
+ for (unsigned j = Ops.size(); j != i;) {
+ --j;
+ if (Ops[j].Op == Ops[i].Op) {
+ NewMulOps.push_back(V);
+ Ops.erase(Ops.begin()+j);
+ }
+ }
+ --i;
}
}
// No need for extra uses anymore.
delete DummyInst;
-
+
unsigned NumAddedValues = NewMulOps.size();
Value *V = EmitAddTreeOfValues(I, NewMulOps);
- Value *V2 = BinaryOperator::CreateMul(V, MaxOccVal, "tmp", I);
-
- // Now that we have inserted V and its sole use, optimize it. This allows
- // us to handle cases that require multiple factoring steps, such as this:
+
+ // Now that we have inserted the add tree, optimize it. This allows us to
+ // handle cases that require multiple factoring steps, such as this:
// A*A*B + A*A*C --> A*(A*B+A*C) --> A*(A*(B+C))
- if (NumAddedValues > 1)
- ReassociateExpression(cast<BinaryOperator>(V));
+ assert(NumAddedValues > 1 && "Each occurrence should contribute a value");
+ (void)NumAddedValues;
+ V = ReassociateExpression(cast<BinaryOperator>(V));
+
+ // Create the multiply.
+ Value *V2 = BinaryOperator::CreateMul(V, MaxOccVal, "tmp", I);
+
+ // Rerun associate on the multiply in case the inner expression turned into
+ // a multiply. We want to make sure that we keep things in canonical form.
+ V2 = ReassociateExpression(cast<BinaryOperator>(V2));
// If every add operand included the factor (e.g. "A*B + A*C"), then the
// entire result expression is just the multiply "A*(B+C)".
if (Ops.empty())
return V2;
- // Otherwise, we had some input that didn't have the fact, such as
+ // Otherwise, we had some input that didn't have the factor, such as
// "A*B + A*C + D" -> "A*(B+C) + D". Add the new multiply to the list of
- // things being added.
+ // things being added by this operation.
Ops.insert(Ops.begin(), ValueEntry(getRank(V2), V2));
}
}
Value *Reassociate::OptimizeExpression(BinaryOperator *I,
- std::vector<ValueEntry> &Ops) {
+ SmallVectorImpl<ValueEntry> &Ops) {
// Now that we have the linearized expression tree, try to optimize it.
// Start by folding any constants that we found.
bool IterateOptimization = false;
switch (Opcode) {
default: break;
case Instruction::And:
- if (CstVal->isZero()) // ... & 0 -> 0
+ if (CstVal->isZero()) // X & 0 -> 0
return CstVal;
- if (CstVal->isAllOnesValue()) // ... & -1 -> ...
+ if (CstVal->isAllOnesValue()) // X & -1 -> X
Ops.pop_back();
break;
case Instruction::Mul:
- if (CstVal->isZero()) { // ... * 0 -> 0
+ if (CstVal->isZero()) { // X * 0 -> 0
++NumAnnihil;
return CstVal;
}
if (cast<ConstantInt>(CstVal)->isOne())
- Ops.pop_back(); // ... * 1 -> ...
+ Ops.pop_back(); // X * 1 -> X
break;
case Instruction::Or:
- if (CstVal->isAllOnesValue()) // ... | -1 -> -1
+ if (CstVal->isAllOnesValue()) // X | -1 -> -1
return CstVal;
// FALLTHROUGH!
case Instruction::Add:
case Instruction::Xor:
- if (CstVal->isZero()) // ... [|^+] 0 -> ...
+ if (CstVal->isZero()) // X [|^+] 0 -> X
Ops.pop_back();
break;
}
}
-/// ReassociateBB - Inspect all of the instructions in this basic block,
-/// reassociating them as we go.
-void Reassociate::ReassociateBB(BasicBlock *BB) {
- for (BasicBlock::iterator BBI = BB->begin(); BBI != BB->end(); ) {
- Instruction *BI = BBI++;
- if (BI->getOpcode() == Instruction::Shl &&
- isa<ConstantInt>(BI->getOperand(1)))
- if (Instruction *NI = ConvertShiftToMul(BI, ValueRankMap)) {
- MadeChange = true;
- BI = NI;
- }
+/// ReassociateInst - Inspect and reassociate the instruction at the
+/// given position, post-incrementing the position.
+void Reassociate::ReassociateInst(BasicBlock::iterator &BBI) {
+ Instruction *BI = BBI++;
+ if (BI->getOpcode() == Instruction::Shl &&
+ isa<ConstantInt>(BI->getOperand(1)))
+ if (Instruction *NI = ConvertShiftToMul(BI, ValueRankMap)) {
+ MadeChange = true;
+ BI = NI;
+ }
- // Reject cases where it is pointless to do this.
- if (!isa<BinaryOperator>(BI) || BI->getType()->isFloatingPoint() ||
- isa<VectorType>(BI->getType()))
- continue; // Floating point ops are not associative.
+ // Reject cases where it is pointless to do this.
+ if (!isa<BinaryOperator>(BI) || BI->getType()->isFloatingPointTy() ||
+ BI->getType()->isVectorTy())
+ return; // Floating point ops are not associative.
+
+ // Do not reassociate boolean (i1) expressions. We want to preserve the
+ // original order of evaluation for short-circuited comparisons that
+ // SimplifyCFG has folded to AND/OR expressions. If the expression
+ // is not further optimized, it is likely to be transformed back to a
+ // short-circuited form for code gen, and the source order may have been
+ // optimized for the most likely conditions.
+ if (BI->getType()->isIntegerTy(1))
+ return;
- // If this is a subtract instruction which is not already in negate form,
- // see if we can convert it to X+-Y.
- if (BI->getOpcode() == Instruction::Sub) {
- if (ShouldBreakUpSubtract(BI)) {
- BI = BreakUpSubtract(BI, ValueRankMap);
+ // If this is a subtract instruction which is not already in negate form,
+ // see if we can convert it to X+-Y.
+ if (BI->getOpcode() == Instruction::Sub) {
+ if (ShouldBreakUpSubtract(BI)) {
+ BI = BreakUpSubtract(BI, ValueRankMap);
+ // Reset the BBI iterator in case BreakUpSubtract changed the
+ // instruction it points to.
+ BBI = BI;
+ ++BBI;
+ MadeChange = true;
+ } else if (BinaryOperator::isNeg(BI)) {
+ // Otherwise, this is a negation. See if the operand is a multiply tree
+ // and if this is not an inner node of a multiply tree.
+ if (isReassociableOp(BI->getOperand(1), Instruction::Mul) &&
+ (!BI->hasOneUse() ||
+ !isReassociableOp(BI->use_back(), Instruction::Mul))) {
+ BI = LowerNegateToMultiply(BI, ValueRankMap);
MadeChange = true;
- } else if (BinaryOperator::isNeg(BI)) {
- // Otherwise, this is a negation. See if the operand is a multiply tree
- // and if this is not an inner node of a multiply tree.
- if (isReassociableOp(BI->getOperand(1), Instruction::Mul) &&
- (!BI->hasOneUse() ||
- !isReassociableOp(BI->use_back(), Instruction::Mul))) {
- BI = LowerNegateToMultiply(BI, ValueRankMap);
- MadeChange = true;
- }
}
}
+ }
- // If this instruction is a commutative binary operator, process it.
- if (!BI->isAssociative()) continue;
- BinaryOperator *I = cast<BinaryOperator>(BI);
+ // If this instruction is a commutative binary operator, process it.
+ if (!BI->isAssociative()) return;
+ BinaryOperator *I = cast<BinaryOperator>(BI);
- // If this is an interior node of a reassociable tree, ignore it until we
- // get to the root of the tree, to avoid N^2 analysis.
- if (I->hasOneUse() && isReassociableOp(I->use_back(), I->getOpcode()))
- continue;
+ // If this is an interior node of a reassociable tree, ignore it until we
+ // get to the root of the tree, to avoid N^2 analysis.
+ if (I->hasOneUse() && isReassociableOp(I->use_back(), I->getOpcode()))
+ return;
- // If this is an add tree that is used by a sub instruction, ignore it
- // until we process the subtract.
- if (I->hasOneUse() && I->getOpcode() == Instruction::Add &&
- cast<Instruction>(I->use_back())->getOpcode() == Instruction::Sub)
- continue;
+ // If this is an add tree that is used by a sub instruction, ignore it
+ // until we process the subtract.
+ if (I->hasOneUse() && I->getOpcode() == Instruction::Add &&
+ cast<Instruction>(I->use_back())->getOpcode() == Instruction::Sub)
+ return;
- ReassociateExpression(I);
- }
+ ReassociateExpression(I);
}
-void Reassociate::ReassociateExpression(BinaryOperator *I) {
+Value *Reassociate::ReassociateExpression(BinaryOperator *I) {
- // First, walk the expression tree, linearizing the tree, collecting
- std::vector<ValueEntry> Ops;
+ // First, walk the expression tree, linearizing the tree, collecting the
+ // operand information.
+ SmallVector<ValueEntry, 8> Ops;
LinearizeExprTree(I, Ops);
- DEBUG(errs() << "RAIn:\t"; PrintOps(I, Ops); errs() << '\n');
+ DEBUG(dbgs() << "RAIn:\t"; PrintOps(I, Ops); dbgs() << '\n');
// Now that we have linearized the tree to a list and have gathered all of
// the operands and their ranks, sort the operands by their rank. Use a
if (Value *V = OptimizeExpression(I, Ops)) {
// This expression tree simplified to something that isn't a tree,
// eliminate it.
- DEBUG(errs() << "Reassoc to scalar: " << *V << '\n');
+ DEBUG(dbgs() << "Reassoc to scalar: " << *V << '\n');
I->replaceAllUsesWith(V);
+ if (Instruction *VI = dyn_cast<Instruction>(V))
+ VI->setDebugLoc(I->getDebugLoc());
RemoveDeadBinaryOp(I);
++NumAnnihil;
- return;
+ return V;
}
// We want to sink immediates as deeply as possible except in the case where
cast<Instruction>(I->use_back())->getOpcode() == Instruction::Add &&
isa<ConstantInt>(Ops.back().Op) &&
cast<ConstantInt>(Ops.back().Op)->isAllOnesValue()) {
- Ops.insert(Ops.begin(), Ops.back());
- Ops.pop_back();
+ ValueEntry Tmp = Ops.pop_back_val();
+ Ops.insert(Ops.begin(), Tmp);
}
- DEBUG(errs() << "RAOut:\t"; PrintOps(I, Ops); errs() << '\n');
+ DEBUG(dbgs() << "RAOut:\t"; PrintOps(I, Ops); dbgs() << '\n');
if (Ops.size() == 1) {
// This expression tree simplified to something that isn't a tree,
// eliminate it.
I->replaceAllUsesWith(Ops[0].Op);
+ if (Instruction *OI = dyn_cast<Instruction>(Ops[0].Op))
+ OI->setDebugLoc(I->getDebugLoc());
RemoveDeadBinaryOp(I);
- } else {
- // Now that we ordered and optimized the expressions, splat them back into
- // the expression tree, removing any unneeded nodes.
- RewriteExprTree(I, Ops);
+ return Ops[0].Op;
}
+
+ // Now that we ordered and optimized the expressions, splat them back into
+ // the expression tree, removing any unneeded nodes.
+ RewriteExprTree(I, Ops);
+ return I;
}
MadeChange = false;
for (Function::iterator FI = F.begin(), FE = F.end(); FI != FE; ++FI)
- ReassociateBB(FI);
+ for (BasicBlock::iterator BBI = FI->begin(); BBI != FI->end(); )
+ ReassociateInst(BBI);
+
+ // Now that we're done, revisit any instructions which are likely to
+ // have secondary reassociation opportunities.
+ while (!RedoInsts.empty())
+ if (Value *V = RedoInsts.pop_back_val()) {
+ BasicBlock::iterator BBI = cast<Instruction>(V);
+ ReassociateInst(BBI);
+ }
+
+ // Now that we're done, delete any instructions which are no longer used.
+ while (!DeadInsts.empty())
+ if (Value *V = DeadInsts.pop_back_val())
+ RecursivelyDeleteTriviallyDeadInstructions(V);
- // We are done with the rank map...
+ // We are done with the rank map.
RankMap.clear();
ValueRankMap.clear();
return MadeChange;