X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTransforms%2FScalar%2FReassociate.cpp;h=71d787a839a73b380d514f7e33bc8de145839494;hb=ed5cb593efa6faaa8ed6dca36d5d200fb832496c;hp=f6e11bc646fef59c44e22d28725ca553054c5aed;hpb=bdff548e4dd577a72094d57b282de4e765643b96;p=oota-llvm.git diff --git a/lib/Transforms/Scalar/Reassociate.cpp b/lib/Transforms/Scalar/Reassociate.cpp index f6e11bc646f..71d787a839a 100644 --- a/lib/Transforms/Scalar/Reassociate.cpp +++ b/lib/Transforms/Scalar/Reassociate.cpp @@ -8,7 +8,7 @@ //===----------------------------------------------------------------------===// // // 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) // @@ -27,18 +27,16 @@ #include "llvm/Function.h" #include "llvm/Instructions.h" #include "llvm/IntrinsicInst.h" -#include "llvm/LLVMContext.h" #include "llvm/Pass.h" #include "llvm/Assembly/Writer.h" #include "llvm/Support/CFG.h" -#include "llvm/Support/Compiler.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ValueHandle.h" #include "llvm/Support/raw_ostream.h" #include "llvm/ADT/PostOrderIterator.h" #include "llvm/ADT/Statistic.h" +#include "llvm/ADT/DenseMap.h" #include -#include using namespace llvm; STATISTIC(NumLinear , "Number of insts linearized"); @@ -47,7 +45,7 @@ STATISTIC(NumAnnihil, "Number of expr tree annihilated"); STATISTIC(NumFactor , "Number of multiplies factored"); namespace { - struct VISIBILITY_HIDDEN ValueEntry { + struct ValueEntry { unsigned Rank; Value *Op; ValueEntry(unsigned R, Value *O) : Rank(R), Op(O) {} @@ -60,21 +58,22 @@ namespace { #ifndef NDEBUG /// PrintOps - Print out the expression identified in the Ops list. /// -static void PrintOps(Instruction *I, const std::vector &Ops) { +static void PrintOps(Instruction *I, const SmallVectorImpl &Ops) { Module *M = I->getParent()->getParent()->getParent(); - cerr << Instruction::getOpcodeName(I->getOpcode()) << " " - << *Ops[0].Op->getType(); + errs() << Instruction::getOpcodeName(I->getOpcode()) << " " + << *Ops[0].Op->getType() << '\t'; for (unsigned i = 0, e = Ops.size(); i != e; ++i) { - WriteAsOperand(*cerr.stream() << " ", Ops[i].Op, false, M); - cerr << "," << Ops[i].Rank; + errs() << "[ "; + WriteAsOperand(errs(), Ops[i].Op, false, M); + errs() << ", #" << Ops[i].Rank << "] "; } } #endif namespace { - class VISIBILITY_HIDDEN Reassociate : public FunctionPass { - std::map RankMap; - std::map, unsigned> ValueRankMap; + class Reassociate : public FunctionPass { + DenseMap RankMap; + DenseMap, unsigned> ValueRankMap; bool MadeChange; public: static char ID; // Pass identification, replacement for typeid @@ -88,11 +87,13 @@ namespace { private: void BuildRankMap(Function &F); unsigned getRank(Value *V); - void ReassociateExpression(BinaryOperator *I); - void RewriteExprTree(BinaryOperator *I, std::vector &Ops, + Value *ReassociateExpression(BinaryOperator *I); + void RewriteExprTree(BinaryOperator *I, SmallVectorImpl &Ops, unsigned Idx = 0); - Value *OptimizeExpression(BinaryOperator *I, std::vector &Ops); - void LinearizeExprTree(BinaryOperator *I, std::vector &Ops); + Value *OptimizeExpression(BinaryOperator *I, + SmallVectorImpl &Ops); + Value *OptimizeAdd(Instruction *I, SmallVectorImpl &Ops); + void LinearizeExprTree(BinaryOperator *I, SmallVectorImpl &Ops); void LinearizeExpr(BinaryOperator *I); Value *RemoveFactorFromExpression(Value *V, Value *Factor); void ReassociateBB(BasicBlock *BB); @@ -109,10 +110,13 @@ FunctionPass *llvm::createReassociatePass() { return new Reassociate(); } void Reassociate::RemoveDeadBinaryOp(Value *V) { Instruction *Op = dyn_cast(V); - if (!Op || !isa(Op) || !isa(Op) || !Op->use_empty()) + if (!Op || !isa(Op) || !Op->use_empty()) return; Value *LHS = Op->getOperand(0), *RHS = Op->getOperand(1); + + ValueRankMap.erase(Op); + Op->eraseFromParent(); RemoveDeadBinaryOp(LHS); RemoveDeadBinaryOp(RHS); } @@ -122,7 +126,6 @@ static bool isUnmovableInstruction(Instruction *I) { if (I->getOpcode() == Instruction::PHI || I->getOpcode() == Instruction::Alloca || I->getOpcode() == Instruction::Load || - I->getOpcode() == Instruction::Malloc || I->getOpcode() == Instruction::Invoke || (I->getOpcode() == Instruction::Call && !isa(I)) || @@ -159,13 +162,14 @@ void Reassociate::BuildRankMap(Function &F) { } unsigned Reassociate::getRank(Value *V) { - if (isa(V)) return ValueRankMap[V]; // Function argument... - Instruction *I = dyn_cast(V); - if (I == 0) return 0; // Otherwise it's a global or constant, rank 0. + if (I == 0) { + if (isa(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 @@ -185,7 +189,7 @@ unsigned Reassociate::getRank(Value *V) { //DEBUG(errs() << "Calculated Rank[" << V->getName() << "] = " // << Rank << "\n"); - return CachedRank = Rank; + return ValueRankMap[I] = Rank; } /// isReassociableOp - Return true if V is an instruction of the specified @@ -200,8 +204,7 @@ static BinaryOperator *isReassociableOp(Value *V, unsigned Opcode) { /// LowerNegateToMultiply - Replace 0-X with X*-1. /// static Instruction *LowerNegateToMultiply(Instruction *Neg, - std::map, unsigned> &ValueRankMap, - LLVMContext &Context) { + DenseMap, unsigned> &ValueRankMap) { Constant *Cst = Constant::getAllOnesValue(Neg->getType()); Instruction *Res = BinaryOperator::CreateMul(Neg->getOperand(1), Cst, "",Neg); @@ -254,10 +257,9 @@ void Reassociate::LinearizeExpr(BinaryOperator *I) { /// caller MUST use something like RewriteExprTree to put the values back in. /// void Reassociate::LinearizeExprTree(BinaryOperator *I, - std::vector &Ops) { + SmallVectorImpl &Ops) { Value *LHS = I->getOperand(0), *RHS = I->getOperand(1); unsigned Opcode = I->getOpcode(); - LLVMContext &Context = I->getContext(); // First step, linearize the expression if it is in ((A+B)+(C+D)) form. BinaryOperator *LHSBO = isReassociableOp(LHS, Opcode); @@ -267,13 +269,11 @@ void Reassociate::LinearizeExprTree(BinaryOperator *I, // transform them into multiplies by -1 so they can be reassociated. if (I->getOpcode() == Instruction::Mul) { if (!LHSBO && LHS->hasOneUse() && BinaryOperator::isNeg(LHS)) { - LHS = LowerNegateToMultiply(cast(LHS), - ValueRankMap, Context); + LHS = LowerNegateToMultiply(cast(LHS), ValueRankMap); LHSBO = isReassociableOp(LHS, Opcode); } if (!RHSBO && RHS->hasOneUse() && BinaryOperator::isNeg(RHS)) { - RHS = LowerNegateToMultiply(cast(RHS), - ValueRankMap, Context); + RHS = LowerNegateToMultiply(cast(RHS), ValueRankMap); RHSBO = isReassociableOp(RHS, Opcode); } } @@ -289,15 +289,15 @@ void Reassociate::LinearizeExprTree(BinaryOperator *I, I->setOperand(0, UndefValue::get(I->getType())); I->setOperand(1, UndefValue::get(I->getType())); return; - } else { - // Turn X+(Y+Z) -> (Y+Z)+X - std::swap(LHSBO, RHSBO); - std::swap(LHS, RHS); - bool Success = !I->swapOperands(); - assert(Success && "swapOperands failed"); - Success = false; - MadeChange = true; } + + // Turn X+(Y+Z) -> (Y+Z)+X + std::swap(LHSBO, RHSBO); + std::swap(LHS, RHS); + bool Success = !I->swapOperands(); + assert(Success && "swapOperands failed"); + Success = false; + MadeChange = true; } else if (RHSBO) { // Turn (A+B)+(C+D) -> (((A+B)+C)+D). This guarantees the the RHS is not // part of the expression tree. @@ -329,7 +329,7 @@ void Reassociate::LinearizeExprTree(BinaryOperator *I, // linearized and optimized, emit them in-order. This function is written to be // tail recursive. void Reassociate::RewriteExprTree(BinaryOperator *I, - std::vector &Ops, + SmallVectorImpl &Ops, unsigned i) { if (i+2 == Ops.size()) { if (I->getOperand(0) != Ops[i].Op || @@ -375,7 +375,10 @@ void Reassociate::RewriteExprTree(BinaryOperator *I, // version of the value is returned, and BI is left pointing at the instruction // that should be processed next by the reassociation pass. // -static Value *NegateValue(LLVMContext &Context, Value *V, Instruction *BI) { +static Value *NegateValue(Value *V, Instruction *BI) { + if (Constant *C = dyn_cast(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, @@ -383,13 +386,13 @@ static Value *NegateValue(LLVMContext &Context, Value *V, Instruction *BI) { // 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(V)) if (I->getOpcode() == Instruction::Add && I->hasOneUse()) { // Push the negates through the add. - I->setOperand(0, NegateValue(Context, I->getOperand(0), BI)); - I->setOperand(1, NegateValue(Context, I->getOperand(1), BI)); + I->setOperand(0, NegateValue(I->getOperand(0), BI)); + I->setOperand(1, NegateValue(I->getOperand(1), BI)); // We must move the add instruction here, because the neg instructions do // not dominate the old add instruction in general. By moving it, we are @@ -400,16 +403,46 @@ static Value *NegateValue(LLVMContext &Context, Value *V, Instruction *BI) { 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){ + if (!BinaryOperator::isNeg(*UI)) 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(*UI); + + // 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(V)) { + if (InvokeInst *II = dyn_cast(InstInput)) { + InsertPt = II->getNormalDest()->begin(); + } else { + InsertPt = InstInput; + ++InsertPt; + } + while (isa(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); } /// ShouldBreakUpSubtract - Return true if we should break up this subtract of /// X-Y into (X + -Y). -static bool ShouldBreakUpSubtract(LLVMContext &Context, Instruction *Sub) { +static bool ShouldBreakUpSubtract(Instruction *Sub) { // If this is a negation, we can't split it up! if (BinaryOperator::isNeg(Sub)) return false; @@ -433,15 +466,15 @@ static bool ShouldBreakUpSubtract(LLVMContext &Context, Instruction *Sub) { /// BreakUpSubtract - If we have (X-Y), and if either X is an add, or if this is /// only used by an add, transform this into (X+(0-Y)) to promote better /// reassociation. -static Instruction *BreakUpSubtract(LLVMContext &Context, Instruction *Sub, - std::map, unsigned> &ValueRankMap) { - // Convert a subtract into an add and a neg instruction... so that sub - // instructions can be commuted with other add instructions... +static Instruction *BreakUpSubtract(Instruction *Sub, + DenseMap, 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(Context, Sub->getOperand(1), Sub); + Value *NegVal = NegateValue(Sub->getOperand(1), Sub); Instruction *New = BinaryOperator::CreateAdd(Sub->getOperand(0), NegVal, "", Sub); New->takeName(Sub); @@ -459,8 +492,7 @@ static Instruction *BreakUpSubtract(LLVMContext &Context, Instruction *Sub, /// by one, change this into a multiply by a constant to assist with further /// reassociation. static Instruction *ConvertShiftToMul(Instruction *Shl, - std::map, unsigned> &ValueRankMap, - LLVMContext &Context) { + DenseMap, 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) || @@ -468,11 +500,10 @@ static Instruction *ConvertShiftToMul(Instruction *Shl, (isReassociableOp(Shl->use_back(), Instruction::Mul) || isReassociableOp(Shl->use_back(), Instruction::Add)))) { Constant *MulCst = ConstantInt::get(Shl->getType(), 1); - MulCst = - ConstantExpr::getShl(MulCst, cast(Shl->getOperand(1))); + MulCst = ConstantExpr::getShl(MulCst, cast(Shl->getOperand(1))); - Instruction *Mul = BinaryOperator::CreateMul(Shl->getOperand(0), MulCst, - "", Shl); + Instruction *Mul = + BinaryOperator::CreateMul(Shl->getOperand(0), MulCst, "", Shl); ValueRankMap.erase(Shl); Mul->takeName(Shl); Shl->replaceAllUsesWith(Mul); @@ -483,15 +514,16 @@ static Instruction *ConvertShiftToMul(Instruction *Shl, } // 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 &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 &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; @@ -500,7 +532,7 @@ static unsigned FindInOperandList(std::vector &Ops, unsigned i, /// EmitAddTreeOfValues - Emit a tree of add instructions, summing Ops together /// and returning the result. Insert the tree before I. -static Value *EmitAddTreeOfValues(Instruction *I, std::vector &Ops) { +static Value *EmitAddTreeOfValues(Instruction *I, SmallVectorImpl &Ops){ if (Ops.size() == 1) return Ops.back(); Value *V1 = Ops.back(); @@ -516,32 +548,57 @@ Value *Reassociate::RemoveFactorFromExpression(Value *V, Value *Factor) { BinaryOperator *BO = isReassociableOp(V, Instruction::Mul); if (!BO) return 0; - std::vector Factors; + SmallVector 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(Factor)) + if (ConstantInt *FC2 = dyn_cast(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); + BO->eraseFromParent(); + 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. static void FindSingleUseMultiplyFactors(Value *V, - std::vector &Factors) { + SmallVectorImpl &Factors) { BinaryOperator *BO; if ((!V->hasOneUse() && !V->use_empty()) || !(BO = dyn_cast(V)) || @@ -555,10 +612,228 @@ static void FindSingleUseMultiplyFactors(Value *V, FindSingleUseMultiplyFactors(BO->getOperand(0), Factors); } +/// 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, + SmallVectorImpl &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) { + // First, check for X and ~X in the operand list. + assert(i < Ops.size()); + if (BinaryOperator::isNot(Ops[i].Op)) { // Cannot occur for ^. + Value *X = BinaryOperator::getNotArgument(Ops[i].Op); + unsigned FoundX = FindInOperandList(Ops, i, X); + if (FoundX != i) { + if (Opcode == Instruction::And) // ...&X&~X = 0 + return Constant::getNullValue(X->getType()); + + if (Opcode == Instruction::Or) // ...|X|~X = -1 + return Constant::getAllOnesValue(X->getType()); + } + } + + // Next, check for duplicate pairs of values, which we assume are next to + // each other, due to our sorting criteria. + 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 for And and Or. + Ops.erase(Ops.begin()+i); + --i; --e; + ++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, + SmallVectorImpl &Ops) { + // Scan the operand lists looking for X and -X pairs. If we find any, we + // 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) { + 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(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 + Mul = ReassociateExpression(cast(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(TheOp)) + continue; + + Value *X = BinaryOperator::getNegArgument(TheOp); + unsigned FoundX = FindInOperandList(Ops, i, X); + if (FoundX == i) + continue; + + // Remove X and -X from the operand list. + if (Ops.size() == 2) + return Constant::getNullValue(X->getType()); + + Ops.erase(Ops.begin()+i); + if (i < FoundX) + --FoundX; + else + --i; // Need to back up an extra one. + Ops.erase(Ops.begin()+FoundX); + ++NumAnnihil; + --i; // Revisit element. + e -= 2; // Removed two elements. + } + + // Scan the operand list, checking to see if there are any common factors + // between operands. Consider something like A*A+A*B*C+D. We would like to + // reassociate this to A*(A+B*C)+D, which reduces the number of multiplies. + // To efficiently find this, we count the number of times a factor occurs + // for any ADD operands that are MULs. + DenseMap FactorOccurrences; + + // Keep track of each multiply we see, to avoid triggering on (X*4)+(X*4) + // where they are actually the same multiply. + unsigned MaxOcc = 0; + Value *MaxOccVal = 0; + for (unsigned i = 0, e = Ops.size(); i != e; ++i) { + BinaryOperator *BOp = dyn_cast(Ops[i].Op); + if (BOp == 0 || BOp->getOpcode() != Instruction::Mul || !BOp->use_empty()) + continue; + + // Compute all of the factors of this added value. + SmallVector Factors; + FindSingleUseMultiplyFactors(BOp, Factors); + assert(Factors.size() > 1 && "Bad linearize!"); + + // Add one to FactorOccurrences for each unique factor in this op. + SmallPtrSet 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(Factor)) + if (CI->getValue().isNegative() && !CI->getValue().isMinSignedValue()) { + 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'); + ++NumFactor; + + // Create a new instruction that uses the MaxOccVal twice. If we don't do + // this, we could otherwise run into situations where removing a factor + // from an expression will drop a use of maxocc, and this can cause + // RemoveFactorFromExpression on successive values to behave differently. + Instruction *DummyInst = BinaryOperator::CreateAdd(MaxOccVal, MaxOccVal); + SmallVector NewMulOps; + for (unsigned i = 0, e = Ops.size(); i != e; ++i) { + if (Value *V = RemoveFactorFromExpression(Ops[i].Op, MaxOccVal)) { + NewMulOps.push_back(V); + Ops.erase(Ops.begin()+i); + --i; --e; + } + } + + // No need for extra uses anymore. + delete DummyInst; + + unsigned NumAddedValues = NewMulOps.size(); + Value *V = EmitAddTreeOfValues(I, NewMulOps); + + // 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)) + assert(NumAddedValues > 1 && "Each occurrence should contribute a value"); + V = ReassociateExpression(cast(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(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 factor, such as + // "A*B + A*C + D" -> "A*(B+C) + D". Add the new multiply to the list of + // things being added by this operation. + Ops.insert(Ops.begin(), ValueEntry(getRank(V2), V2)); + } + + return 0; +} Value *Reassociate::OptimizeExpression(BinaryOperator *I, - std::vector &Ops) { + SmallVectorImpl &Ops) { // Now that we have the linearized expression tree, try to optimize it. // Start by folding any constants that we found. bool IterateOptimization = false; @@ -578,198 +853,53 @@ Value *Reassociate::OptimizeExpression(BinaryOperator *I, switch (Opcode) { default: break; case Instruction::And: - if (CstVal->isZero()) { // ... & 0 -> 0 - ++NumAnnihil; + if (CstVal->isZero()) // X & 0 -> 0 return CstVal; - } else 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; - } else if (cast(CstVal)->isOne()) { - Ops.pop_back(); // ... * 1 -> ... } + + if (cast(CstVal)->isOne()) + Ops.pop_back(); // X * 1 -> X break; case Instruction::Or: - if (CstVal->isAllOnesValue()) { // ... | -1 -> -1 - ++NumAnnihil; + 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; } if (Ops.size() == 1) return Ops[0].Op; - // Handle destructive annihilation do to identities between elements in the + // Handle destructive annihilation due to identities between elements in the // argument list here. switch (Opcode) { default: break; case Instruction::And: case Instruction::Or: - case Instruction::Xor: - // 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) { - // First, check for X and ~X in the operand list. - assert(i < Ops.size()); - if (BinaryOperator::isNot(Ops[i].Op)) { // Cannot occur for ^. - Value *X = BinaryOperator::getNotArgument(Ops[i].Op); - unsigned FoundX = FindInOperandList(Ops, i, X); - if (FoundX != i) { - if (Opcode == Instruction::And) { // ...&X&~X = 0 - ++NumAnnihil; - return Constant::getNullValue(X->getType()); - } else if (Opcode == Instruction::Or) { // ...|X|~X = -1 - ++NumAnnihil; - return Constant::getAllOnesValue(X->getType()); - } - } - } - - // Next, check for duplicate pairs of values, which we assume are next to - // each other, due to our sorting criteria. - 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. - Ops.erase(Ops.begin()+i); - --i; --e; - IterateOptimization = true; - ++NumAnnihil; - } else { - assert(Opcode == Instruction::Xor); - if (e == 2) { - ++NumAnnihil; - return Constant::getNullValue(Ops[0].Op->getType()); - } - // ... X^X -> ... - Ops.erase(Ops.begin()+i, Ops.begin()+i+2); - i -= 1; e -= 2; - IterateOptimization = true; - ++NumAnnihil; - } - } - } + case Instruction::Xor: { + unsigned NumOps = Ops.size(); + if (Value *Result = OptimizeAndOrXor(Opcode, Ops)) + return Result; + IterateOptimization |= Ops.size() != NumOps; break; + } - case Instruction::Add: - // Scan the operand lists looking for X and -X pairs. If we find any, we - // can simplify the expression. X+-X == 0. - for (unsigned i = 0, e = Ops.size(); i != e; ++i) { - assert(i < Ops.size()); - // Check for X and -X in the operand list. - if (BinaryOperator::isNeg(Ops[i].Op)) { - Value *X = BinaryOperator::getNegArgument(Ops[i].Op); - unsigned FoundX = FindInOperandList(Ops, i, X); - if (FoundX != i) { - // Remove X and -X from the operand list. - if (Ops.size() == 2) { - ++NumAnnihil; - return Constant::getNullValue(X->getType()); - } else { - Ops.erase(Ops.begin()+i); - if (i < FoundX) - --FoundX; - else - --i; // Need to back up an extra one. - Ops.erase(Ops.begin()+FoundX); - IterateOptimization = true; - ++NumAnnihil; - --i; // Revisit element. - e -= 2; // Removed two elements. - } - } - } - } - - - // Scan the operand list, checking to see if there are any common factors - // between operands. Consider something like A*A+A*B*C+D. We would like to - // reassociate this to A*(A+B*C)+D, which reduces the number of multiplies. - // To efficiently find this, we count the number of times a factor occurs - // for any ADD operands that are MULs. - std::map FactorOccurrences; - unsigned MaxOcc = 0; - Value *MaxOccVal = 0; - for (unsigned i = 0, e = Ops.size(); i != e; ++i) { - if (BinaryOperator *BOp = dyn_cast(Ops[i].Op)) { - if (BOp->getOpcode() == Instruction::Mul && BOp->use_empty()) { - // Compute all of the factors of this added value. - std::vector Factors; - FindSingleUseMultiplyFactors(BOp, Factors); - 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 { - std::set Duplicates; - for (unsigned i = 0, e = Factors.size(); i != e; ++i) { - if (Duplicates.insert(Factors[i]).second) { - unsigned Occ = ++FactorOccurrences[Factors[i]]; - if (Occ > MaxOcc) { MaxOcc = Occ; MaxOccVal = Factors[i]; } - } - } - } - } - } - } - - // If any factor occurred more than one time, we can pull it out. - if (MaxOcc > 1) { - DEBUG(errs() << "\nFACTORING [" << MaxOcc << "]: " << *MaxOccVal << "\n"); - - // Create a new instruction that uses the MaxOccVal twice. If we don't do - // this, we could otherwise run into situations where removing a factor - // from an expression will drop a use of maxocc, and this can cause - // RemoveFactorFromExpression on successive values to behave differently. - Instruction *DummyInst = BinaryOperator::CreateAdd(MaxOccVal, MaxOccVal); - std::vector NewMulOps; - for (unsigned i = 0, e = Ops.size(); i != e; ++i) { - if (Value *V = RemoveFactorFromExpression(Ops[i].Op, MaxOccVal)) { - NewMulOps.push_back(V); - Ops.erase(Ops.begin()+i); - --i; --e; - } - } - - // 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: - // A*A*B + A*A*C --> A*(A*B+A*C) --> A*(A*(B+C)) - if (NumAddedValues > 1) - ReassociateExpression(cast(V)); - - ++NumFactor; - - if (Ops.empty()) - return V2; + case Instruction::Add: { + unsigned NumOps = Ops.size(); + if (Value *Result = OptimizeAdd(I, Ops)) + return Result; + IterateOptimization |= Ops.size() != NumOps; + } - // Add the new value to the list of things being added. - Ops.insert(Ops.begin(), ValueEntry(getRank(V2), V2)); - - // Rewrite the tree so that there is now a use of V. - RewriteExprTree(I, Ops); - return OptimizeExpression(I, Ops); - } break; //case Instruction::Mul: } @@ -783,13 +913,11 @@ Value *Reassociate::OptimizeExpression(BinaryOperator *I, /// ReassociateBB - Inspect all of the instructions in this basic block, /// reassociating them as we go. void Reassociate::ReassociateBB(BasicBlock *BB) { - LLVMContext &Context = BB->getContext(); - for (BasicBlock::iterator BBI = BB->begin(); BBI != BB->end(); ) { Instruction *BI = BBI++; if (BI->getOpcode() == Instruction::Shl && isa(BI->getOperand(1))) - if (Instruction *NI = ConvertShiftToMul(BI, ValueRankMap, Context)) { + if (Instruction *NI = ConvertShiftToMul(BI, ValueRankMap)) { MadeChange = true; BI = NI; } @@ -802,8 +930,8 @@ void Reassociate::ReassociateBB(BasicBlock *BB) { // 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(Context, BI)) { - BI = BreakUpSubtract(Context, BI, ValueRankMap); + if (ShouldBreakUpSubtract(BI)) { + BI = BreakUpSubtract(BI, ValueRankMap); MadeChange = true; } else if (BinaryOperator::isNeg(BI)) { // Otherwise, this is a negation. See if the operand is a multiply tree @@ -811,7 +939,7 @@ void Reassociate::ReassociateBB(BasicBlock *BB) { if (isReassociableOp(BI->getOperand(1), Instruction::Mul) && (!BI->hasOneUse() || !isReassociableOp(BI->use_back(), Instruction::Mul))) { - BI = LowerNegateToMultiply(BI, ValueRankMap, Context); + BI = LowerNegateToMultiply(BI, ValueRankMap); MadeChange = true; } } @@ -836,13 +964,14 @@ void Reassociate::ReassociateBB(BasicBlock *BB) { } } -void Reassociate::ReassociateExpression(BinaryOperator *I) { +Value *Reassociate::ReassociateExpression(BinaryOperator *I) { - // First, walk the expression tree, linearizing the tree, collecting - std::vector Ops; + // First, walk the expression tree, linearizing the tree, collecting the + // operand information. + SmallVector Ops; LinearizeExprTree(I, Ops); - DEBUG(errs() << "RAIn:\t"; PrintOps(I, Ops); errs() << "\n"); + DEBUG(errs() << "RAIn:\t"; PrintOps(I, Ops); errs() << '\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 @@ -857,10 +986,11 @@ void Reassociate::ReassociateExpression(BinaryOperator *I) { 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(errs() << "Reassoc to scalar: " << *V << '\n'); I->replaceAllUsesWith(V); RemoveDeadBinaryOp(I); - return; + ++NumAnnihil; + return V; } // We want to sink immediates as deeply as possible except in the case where @@ -871,22 +1001,24 @@ void Reassociate::ReassociateExpression(BinaryOperator *I) { cast(I->use_back())->getOpcode() == Instruction::Add && isa(Ops.back().Op) && cast(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(errs() << "RAOut:\t"; PrintOps(I, Ops); errs() << '\n'); if (Ops.size() == 1) { // This expression tree simplified to something that isn't a tree, // eliminate it. I->replaceAllUsesWith(Ops[0].Op); 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; } @@ -898,7 +1030,7 @@ bool Reassociate::runOnFunction(Function &F) { for (Function::iterator FI = F.begin(), FE = F.end(); FI != FE; ++FI) ReassociateBB(FI); - // We are done with the rank map... + // We are done with the rank map. RankMap.clear(); ValueRankMap.clear(); return MadeChange;