X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTransforms%2FUtils%2FLocal.cpp;h=d03f7a69c5a1df18fa8da929bad09c33930e4431;hb=84bd6b0c31f41cdd1d859dab54b6bc1177c4c6bb;hp=df831c78a42f8e1311d4699e2588fb3865689051;hpb=afe16e174264e3902bf376d7603b0cba2a0659bf;p=oota-llvm.git diff --git a/lib/Transforms/Utils/Local.cpp b/lib/Transforms/Utils/Local.cpp index df831c78a42..d03f7a69c5a 100644 --- a/lib/Transforms/Utils/Local.cpp +++ b/lib/Transforms/Utils/Local.cpp @@ -1,10 +1,10 @@ //===-- Local.cpp - Functions to perform local transformations ------------===// -// +// // The LLVM Compiler Infrastructure // -// This file was developed by the LLVM research group and is distributed under -// the University of Illinois Open Source License. See LICENSE.TXT for details. -// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// //===----------------------------------------------------------------------===// // // This family of functions perform various local transformations to the @@ -14,133 +14,156 @@ #include "llvm/Transforms/Utils/Local.h" #include "llvm/Constants.h" +#include "llvm/GlobalAlias.h" +#include "llvm/GlobalVariable.h" +#include "llvm/DerivedTypes.h" #include "llvm/Instructions.h" #include "llvm/Intrinsics.h" -#include -#include +#include "llvm/IntrinsicInst.h" +#include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/SmallPtrSet.h" +#include "llvm/Analysis/ConstantFolding.h" +#include "llvm/Analysis/InstructionSimplify.h" +#include "llvm/Analysis/ProfileInfo.h" +#include "llvm/Target/TargetData.h" +#include "llvm/Support/CFG.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/GetElementPtrTypeIterator.h" +#include "llvm/Support/MathExtras.h" +#include "llvm/Support/ValueHandle.h" +#include "llvm/Support/raw_ostream.h" using namespace llvm; -#if defined(__POWERPC__) && defined(__APPLE_CC__) -// FIXME: Currently it seems that isnan didn't make its way into the Apple -// C++ headers, although it IS in the C headers (which confuses autoconf -// in a big way). This is a quick fix to get things compiling, until one of -// us has time to write a more complicated autoconf test. -extern "C" int isnan (double d); -namespace std { int isnan (double d) { return ::isnan (d); } } -#endif - //===----------------------------------------------------------------------===// -// Local constant propagation... +// Local analysis. // -/// doConstantPropagation - If an instruction references constants, try to fold -/// them together... -/// -bool llvm::doConstantPropagation(BasicBlock::iterator &II) { - if (Constant *C = ConstantFoldInstruction(II)) { - // Replaces all of the uses of a variable with uses of the constant. - II->replaceAllUsesWith(C); - - // Remove the instruction from the basic block... - II = II->getParent()->getInstList().erase(II); - return true; +/// getUnderlyingObjectWithOffset - Strip off up to MaxLookup GEPs and +/// bitcasts to get back to the underlying object being addressed, keeping +/// track of the offset in bytes from the GEPs relative to the result. +/// This is closely related to Value::getUnderlyingObject but is located +/// here to avoid making VMCore depend on TargetData. +static Value *getUnderlyingObjectWithOffset(Value *V, const TargetData *TD, + uint64_t &ByteOffset, + unsigned MaxLookup = 6) { + if (!V->getType()->isPointerTy()) + return V; + for (unsigned Count = 0; MaxLookup == 0 || Count < MaxLookup; ++Count) { + if (GEPOperator *GEP = dyn_cast(V)) { + if (!GEP->hasAllConstantIndices()) + return V; + SmallVector Indices(GEP->op_begin() + 1, GEP->op_end()); + ByteOffset += TD->getIndexedOffset(GEP->getPointerOperandType(), + &Indices[0], Indices.size()); + V = GEP->getPointerOperand(); + } else if (Operator::getOpcode(V) == Instruction::BitCast) { + V = cast(V)->getOperand(0); + } else if (GlobalAlias *GA = dyn_cast(V)) { + if (GA->mayBeOverridden()) + return V; + V = GA->getAliasee(); + } else { + return V; + } + assert(V->getType()->isPointerTy() && "Unexpected operand type!"); } - - return false; + return V; } -/// ConstantFoldInstruction - Attempt to constant fold the specified -/// instruction. If successful, the constant result is returned, if not, null -/// is returned. Note that this function can only fail when attempting to fold -/// instructions like loads and stores, which have no constant expression form. -/// -Constant *llvm::ConstantFoldInstruction(Instruction *I) { - if (PHINode *PN = dyn_cast(I)) { - if (PN->getNumIncomingValues() == 0) - return Constant::getNullValue(PN->getType()); - - Constant *Result = dyn_cast(PN->getIncomingValue(0)); - if (Result == 0) return 0; - - // Handle PHI nodes specially here... - for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i) - if (PN->getIncomingValue(i) != Result && PN->getIncomingValue(i) != PN) - return 0; // Not all the same incoming constants... - - // If we reach here, all incoming values are the same constant. - return Result; - } else if (CallInst *CI = dyn_cast(I)) { - if (Function *F = CI->getCalledFunction()) - if (canConstantFoldCallTo(F)) { - std::vector Args; - for (unsigned i = 1, e = CI->getNumOperands(); i != e; ++i) - if (Constant *Op = dyn_cast(CI->getOperand(i))) - Args.push_back(Op); - else - return 0; - return ConstantFoldCall(F, Args); - } - return 0; +/// isSafeToLoadUnconditionally - Return true if we know that executing a load +/// from this value cannot trap. If it is not obviously safe to load from the +/// specified pointer, we do a quick local scan of the basic block containing +/// ScanFrom, to determine if the address is already accessed. +bool llvm::isSafeToLoadUnconditionally(Value *V, Instruction *ScanFrom, + unsigned Align, const TargetData *TD) { + uint64_t ByteOffset = 0; + Value *Base = V; + if (TD) + Base = getUnderlyingObjectWithOffset(V, TD, ByteOffset); + + const Type *BaseType = 0; + unsigned BaseAlign = 0; + if (const AllocaInst *AI = dyn_cast(Base)) { + // An alloca is safe to load from as load as it is suitably aligned. + BaseType = AI->getAllocatedType(); + BaseAlign = AI->getAlignment(); + } else if (const GlobalValue *GV = dyn_cast(Base)) { + // Global variables are safe to load from but their size cannot be + // guaranteed if they are overridden. + if (!isa(GV) && !GV->mayBeOverridden()) { + BaseType = GV->getType()->getElementType(); + BaseAlign = GV->getAlignment(); + } } - Constant *Op0 = 0, *Op1 = 0; - switch (I->getNumOperands()) { - default: - case 2: - Op1 = dyn_cast(I->getOperand(1)); - if (Op1 == 0) return 0; // Not a constant?, can't fold - case 1: - Op0 = dyn_cast(I->getOperand(0)); - if (Op0 == 0) return 0; // Not a constant?, can't fold - break; - case 0: return 0; + if (BaseType && BaseType->isSized()) { + if (TD && BaseAlign == 0) + BaseAlign = TD->getPrefTypeAlignment(BaseType); + + if (Align <= BaseAlign) { + if (!TD) + return true; // Loading directly from an alloca or global is OK. + + // Check if the load is within the bounds of the underlying object. + const PointerType *AddrTy = cast(V->getType()); + uint64_t LoadSize = TD->getTypeStoreSize(AddrTy->getElementType()); + if (ByteOffset + LoadSize <= TD->getTypeAllocSize(BaseType) && + (Align == 0 || (ByteOffset % Align) == 0)) + return true; + } } - if (isa(I) || isa(I)) - return ConstantExpr::get(I->getOpcode(), Op0, Op1); - - switch (I->getOpcode()) { - default: return 0; - case Instruction::Cast: - return ConstantExpr::getCast(Op0, I->getType()); - case Instruction::Select: - if (Constant *Op2 = dyn_cast(I->getOperand(2))) - return ConstantExpr::getSelect(Op0, Op1, Op2); - return 0; - case Instruction::GetElementPtr: - std::vector IdxList; - IdxList.reserve(I->getNumOperands()-1); - if (Op1) IdxList.push_back(Op1); - for (unsigned i = 2, e = I->getNumOperands(); i != e; ++i) - if (Constant *C = dyn_cast(I->getOperand(i))) - IdxList.push_back(C); - else - return 0; // Non-constant operand - return ConstantExpr::getGetElementPtr(Op0, IdxList); + // Otherwise, be a little bit aggressive by scanning the local block where we + // want to check to see if the pointer is already being loaded or stored + // from/to. If so, the previous load or store would have already trapped, + // so there is no harm doing an extra load (also, CSE will later eliminate + // the load entirely). + BasicBlock::iterator BBI = ScanFrom, E = ScanFrom->getParent()->begin(); + + while (BBI != E) { + --BBI; + + // If we see a free or a call which may write to memory (i.e. which might do + // a free) the pointer could be marked invalid. + if (isa(BBI) && BBI->mayWriteToMemory() && + !isa(BBI)) + return false; + + if (LoadInst *LI = dyn_cast(BBI)) { + if (LI->getOperand(0) == V) return true; + } else if (StoreInst *SI = dyn_cast(BBI)) { + if (SI->getOperand(1) == V) return true; + } } + return false; } + +//===----------------------------------------------------------------------===// +// Local constant propagation. +// + // ConstantFoldTerminator - If a terminator instruction is predicated on a // constant value, convert it into an unconditional branch to the constant // destination. // bool llvm::ConstantFoldTerminator(BasicBlock *BB) { TerminatorInst *T = BB->getTerminator(); - + // Branch - See if we are conditional jumping on constant if (BranchInst *BI = dyn_cast(T)) { if (BI->isUnconditional()) return false; // Can't optimize uncond branch - BasicBlock *Dest1 = cast(BI->getOperand(0)); - BasicBlock *Dest2 = cast(BI->getOperand(1)); + BasicBlock *Dest1 = BI->getSuccessor(0); + BasicBlock *Dest2 = BI->getSuccessor(1); - if (ConstantBool *Cond = dyn_cast(BI->getCondition())) { + if (ConstantInt *Cond = dyn_cast(BI->getCondition())) { // Are we branching on constant? // YES. Change to unconditional branch... - BasicBlock *Destination = Cond->getValue() ? Dest1 : Dest2; - BasicBlock *OldDest = Cond->getValue() ? Dest2 : Dest1; + BasicBlock *Destination = Cond->getZExtValue() ? Dest1 : Dest2; + BasicBlock *OldDest = Cond->getZExtValue() ? Dest2 : Dest1; - //cerr << "Function: " << T->getParent()->getParent() - // << "\nRemoving branch from " << T->getParent() + //cerr << "Function: " << T->getParent()->getParent() + // << "\nRemoving branch from " << T->getParent() // << "\n\nTo: " << OldDest << endl; // Let the basic block know that we are letting go of it. Based on this, @@ -152,8 +175,10 @@ bool llvm::ConstantFoldTerminator(BasicBlock *BB) { // unconditional branch. BI->setUnconditionalDest(Destination); return true; - } else if (Dest2 == Dest1) { // Conditional branch to same location? - // This branch matches something like this: + } + + if (Dest2 == Dest1) { // Conditional branch to same location? + // This branch matches something like this: // br bool %cond, label %Dest, label %Dest // and changes it into: br label %Dest @@ -165,7 +190,10 @@ bool llvm::ConstantFoldTerminator(BasicBlock *BB) { BI->setUnconditionalDest(Dest1); return true; } - } else if (SwitchInst *SI = dyn_cast(T)) { + return false; + } + + if (SwitchInst *SI = dyn_cast(T)) { // If we are switching on a constant, we can convert the switch into a // single branch instruction! ConstantInt *CI = dyn_cast(SI->getCondition()); @@ -174,7 +202,7 @@ bool llvm::ConstantFoldTerminator(BasicBlock *BB) { assert(TheOnlyDest == SI->getDefaultDest() && "Default destination is not successor #0?"); - // Figure out which case it goes to... + // Figure out which case it goes to. for (unsigned i = 1, e = SI->getNumSuccessors(); i != e; ++i) { // Found case matching a constant operand? if (SI->getSuccessorValue(i) == CI) { @@ -185,7 +213,7 @@ bool llvm::ConstantFoldTerminator(BasicBlock *BB) { // Check to see if this branch is going to the same place as the default // dest. If so, eliminate it as an explicit compare. if (SI->getSuccessor(i) == DefaultDest) { - // Remove this entry... + // Remove this entry. DefaultDest->removePredecessor(SI->getParent()); SI->removeCase(i); --i; --e; // Don't skip an entry... @@ -207,8 +235,8 @@ bool llvm::ConstantFoldTerminator(BasicBlock *BB) { // If we found a single destination that we can fold the switch into, do so // now. if (TheOnlyDest) { - // Insert the new branch.. - new BranchInst(TheOnlyDest, SI); + // Insert the new branch. + BranchInst::Create(TheOnlyDest, SI); BasicBlock *BB = SI->getParent(); // Remove entries from PHI nodes which we no longer branch to... @@ -221,157 +249,498 @@ bool llvm::ConstantFoldTerminator(BasicBlock *BB) { Succ->removePredecessor(BB); } - // Delete the old switch... + // Delete the old switch. BB->getInstList().erase(SI); return true; - } else if (SI->getNumSuccessors() == 2) { + } + + if (SI->getNumSuccessors() == 2) { // Otherwise, we can fold this switch into a conditional branch // instruction if it has only one non-default destination. - Value *Cond = new SetCondInst(Instruction::SetEQ, SI->getCondition(), - SI->getSuccessorValue(1), "cond", SI); - // Insert the new branch... - new BranchInst(SI->getSuccessor(1), SI->getSuccessor(0), Cond, SI); + Value *Cond = new ICmpInst(SI, ICmpInst::ICMP_EQ, SI->getCondition(), + SI->getSuccessorValue(1), "cond"); + // Insert the new branch. + BranchInst::Create(SI->getSuccessor(1), SI->getSuccessor(0), Cond, SI); - // Delete the old switch... - SI->getParent()->getInstList().erase(SI); + // Delete the old switch. + SI->eraseFromParent(); return true; } + return false; } + + if (IndirectBrInst *IBI = dyn_cast(T)) { + // indirectbr blockaddress(@F, @BB) -> br label @BB + if (BlockAddress *BA = + dyn_cast(IBI->getAddress()->stripPointerCasts())) { + BasicBlock *TheOnlyDest = BA->getBasicBlock(); + // Insert the new branch. + BranchInst::Create(TheOnlyDest, IBI); + + for (unsigned i = 0, e = IBI->getNumDestinations(); i != e; ++i) { + if (IBI->getDestination(i) == TheOnlyDest) + TheOnlyDest = 0; + else + IBI->getDestination(i)->removePredecessor(IBI->getParent()); + } + IBI->eraseFromParent(); + + // If we didn't find our destination in the IBI successor list, then we + // have undefined behavior. Replace the unconditional branch with an + // 'unreachable' instruction. + if (TheOnlyDest) { + BB->getTerminator()->eraseFromParent(); + new UnreachableInst(BB->getContext(), BB); + } + + return true; + } + } + return false; } -/// canConstantFoldCallTo - Return true if its even possible to fold a call to -/// the specified function. -bool llvm::canConstantFoldCallTo(Function *F) { - const std::string &Name = F->getName(); - switch (F->getIntrinsicID()) { - case Intrinsic::isunordered: return true; - default: break; - } +//===----------------------------------------------------------------------===// +// Local dead code elimination. +// - return Name == "sin" || Name == "cos" || Name == "tan" || Name == "sqrt" || - Name == "log" || Name == "log10" || Name == "exp" || Name == "pow" || - Name == "acos" || Name == "asin" || Name == "atan" || Name == "fmod"; -} +/// isInstructionTriviallyDead - Return true if the result produced by the +/// instruction is not used, and the instruction has no side effects. +/// +bool llvm::isInstructionTriviallyDead(Instruction *I) { + if (!I->use_empty() || isa(I)) return false; + + // We don't want debug info removed by anything this general. + if (isa(I)) return false; + + // Likewise for memory use markers. + if (isa(I)) return false; + + if (!I->mayHaveSideEffects()) return true; -static Constant *ConstantFoldFP(double (*NativeFP)(double), double V, - const Type *Ty) { - errno = 0; - V = NativeFP(V); - if (errno == 0) - return ConstantFP::get(Ty, V); - return 0; + // Special case intrinsics that "may have side effects" but can be deleted + // when dead. + if (IntrinsicInst *II = dyn_cast(I)) + // Safe to delete llvm.stacksave if dead. + if (II->getIntrinsicID() == Intrinsic::stacksave) + return true; + return false; } -/// ConstantFoldCall - Attempt to constant fold a call to the specified function -/// with the specified arguments, returning null if unsuccessful. -Constant *llvm::ConstantFoldCall(Function *F, - const std::vector &Operands) { - const std::string &Name = F->getName(); - const Type *Ty = F->getReturnType(); - - if (Operands.size() == 1) { - if (ConstantFP *Op = dyn_cast(Operands[0])) { - double V = Op->getValue(); - if (Name == "sin") - return ConstantFP::get(Ty, sin(V)); - else if (Name == "cos") - return ConstantFP::get(Ty, cos(V)); - else if (Name == "tan") - return ConstantFP::get(Ty, tan(V)); - else if (Name == "sqrt" && V >= 0) - return ConstantFP::get(Ty, sqrt(V)); - else if (Name == "exp") - return ConstantFP::get(Ty, exp(V)); - else if (Name == "log" && V > 0) - return ConstantFP::get(Ty, log(V)); - else if (Name == "log10") - return ConstantFoldFP(log10, V, Ty); - else if (Name == "acos") - return ConstantFoldFP(acos, V, Ty); - else if (Name == "asin") - return ConstantFoldFP(asin, V, Ty); - else if (Name == "atan") - return ConstantFP::get(Ty, atan(V)); +/// RecursivelyDeleteTriviallyDeadInstructions - If the specified value is a +/// trivially dead instruction, delete it. If that makes any of its operands +/// trivially dead, delete them too, recursively. Return true if any +/// instructions were deleted. +bool llvm::RecursivelyDeleteTriviallyDeadInstructions(Value *V) { + Instruction *I = dyn_cast(V); + if (!I || !I->use_empty() || !isInstructionTriviallyDead(I)) + return false; + + SmallVector DeadInsts; + DeadInsts.push_back(I); + + do { + I = DeadInsts.pop_back_val(); + + // Null out all of the instruction's operands to see if any operand becomes + // dead as we go. + for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) { + Value *OpV = I->getOperand(i); + I->setOperand(i, 0); + + if (!OpV->use_empty()) continue; + + // If the operand is an instruction that became dead as we nulled out the + // operand, and if it is 'trivially' dead, delete it in a future loop + // iteration. + if (Instruction *OpI = dyn_cast(OpV)) + if (isInstructionTriviallyDead(OpI)) + DeadInsts.push_back(OpI); } - } else if (Operands.size() == 2) { - if (ConstantFP *Op1 = dyn_cast(Operands[0])) - if (ConstantFP *Op2 = dyn_cast(Operands[1])) { - double Op1V = Op1->getValue(), Op2V = Op2->getValue(); - - if (Name == "llvm.isunordered") - return ConstantBool::get(std::isnan(Op1V) | std::isnan(Op2V)); - else if (Name == "pow") { - errno = 0; - double V = pow(Op1V, Op2V); - if (errno == 0) - return ConstantFP::get(Ty, V); - } else if (Name == "fmod") { - errno = 0; - double V = fmod(Op1V, Op2V); - if (errno == 0) - return ConstantFP::get(Ty, V); - } + + I->eraseFromParent(); + } while (!DeadInsts.empty()); + + return true; +} + +/// RecursivelyDeleteDeadPHINode - If the specified value is an effectively +/// dead PHI node, due to being a def-use chain of single-use nodes that +/// either forms a cycle or is terminated by a trivially dead instruction, +/// delete it. If that makes any of its operands trivially dead, delete them +/// too, recursively. Return true if the PHI node is actually deleted. +bool +llvm::RecursivelyDeleteDeadPHINode(PHINode *PN) { + // We can remove a PHI if it is on a cycle in the def-use graph + // where each node in the cycle has degree one, i.e. only one use, + // and is an instruction with no side effects. + if (!PN->hasOneUse()) + return false; + + bool Changed = false; + SmallPtrSet PHIs; + PHIs.insert(PN); + for (Instruction *J = cast(*PN->use_begin()); + J->hasOneUse() && !J->mayHaveSideEffects(); + J = cast(*J->use_begin())) + // If we find a PHI more than once, we're on a cycle that + // won't prove fruitful. + if (PHINode *JP = dyn_cast(J)) + if (!PHIs.insert(cast(JP))) { + // Break the cycle and delete the PHI and its operands. + JP->replaceAllUsesWith(UndefValue::get(JP->getType())); + (void)RecursivelyDeleteTriviallyDeadInstructions(JP); + Changed = true; + break; } + return Changed; +} + +/// SimplifyInstructionsInBlock - Scan the specified basic block and try to +/// simplify any instructions in it and recursively delete dead instructions. +/// +/// This returns true if it changed the code, note that it can delete +/// instructions in other blocks as well in this block. +bool llvm::SimplifyInstructionsInBlock(BasicBlock *BB, const TargetData *TD) { + bool MadeChange = false; + for (BasicBlock::iterator BI = BB->begin(), E = BB->end(); BI != E; ) { + Instruction *Inst = BI++; + + if (Value *V = SimplifyInstruction(Inst, TD)) { + WeakVH BIHandle(BI); + ReplaceAndSimplifyAllUses(Inst, V, TD); + MadeChange = true; + if (BIHandle == 0) + BI = BB->begin(); + continue; + } + + MadeChange |= RecursivelyDeleteTriviallyDeadInstructions(Inst); } - return 0; + return MadeChange; } +//===----------------------------------------------------------------------===// +// Control Flow Graph Restructuring. +// +/// RemovePredecessorAndSimplify - Like BasicBlock::removePredecessor, this +/// method is called when we're about to delete Pred as a predecessor of BB. If +/// BB contains any PHI nodes, this drops the entries in the PHI nodes for Pred. +/// +/// Unlike the removePredecessor method, this attempts to simplify uses of PHI +/// nodes that collapse into identity values. For example, if we have: +/// x = phi(1, 0, 0, 0) +/// y = and x, z +/// +/// .. and delete the predecessor corresponding to the '1', this will attempt to +/// recursively fold the and to 0. +void llvm::RemovePredecessorAndSimplify(BasicBlock *BB, BasicBlock *Pred, + TargetData *TD) { + // This only adjusts blocks with PHI nodes. + if (!isa(BB->begin())) + return; + + // Remove the entries for Pred from the PHI nodes in BB, but do not simplify + // them down. This will leave us with single entry phi nodes and other phis + // that can be removed. + BB->removePredecessor(Pred, true); + + WeakVH PhiIt = &BB->front(); + while (PHINode *PN = dyn_cast(PhiIt)) { + PhiIt = &*++BasicBlock::iterator(cast(PhiIt)); + + Value *PNV = PN->hasConstantValue(); + if (PNV == 0) continue; + + // If we're able to simplify the phi to a single value, substitute the new + // value into all of its uses. + assert(PNV != PN && "hasConstantValue broken"); + + ReplaceAndSimplifyAllUses(PN, PNV, TD); + + // If recursive simplification ended up deleting the next PHI node we would + // iterate to, then our iterator is invalid, restart scanning from the top + // of the block. + if (PhiIt == 0) PhiIt = &BB->front(); + } +} -//===----------------------------------------------------------------------===// -// Local dead code elimination... -// -bool llvm::isInstructionTriviallyDead(Instruction *I) { - return I->use_empty() && !I->mayWriteToMemory() && !isa(I); +/// MergeBasicBlockIntoOnlyPred - DestBB is a block with one predecessor and its +/// predecessor is known to have one successor (DestBB!). Eliminate the edge +/// between them, moving the instructions in the predecessor into DestBB and +/// deleting the predecessor block. +/// +void llvm::MergeBasicBlockIntoOnlyPred(BasicBlock *DestBB, Pass *P) { + // If BB has single-entry PHI nodes, fold them. + while (PHINode *PN = dyn_cast(DestBB->begin())) { + Value *NewVal = PN->getIncomingValue(0); + // Replace self referencing PHI with undef, it must be dead. + if (NewVal == PN) NewVal = UndefValue::get(PN->getType()); + PN->replaceAllUsesWith(NewVal); + PN->eraseFromParent(); + } + + BasicBlock *PredBB = DestBB->getSinglePredecessor(); + assert(PredBB && "Block doesn't have a single predecessor!"); + + // Splice all the instructions from PredBB to DestBB. + PredBB->getTerminator()->eraseFromParent(); + DestBB->getInstList().splice(DestBB->begin(), PredBB->getInstList()); + + // Zap anything that took the address of DestBB. Not doing this will give the + // address an invalid value. + if (DestBB->hasAddressTaken()) { + BlockAddress *BA = BlockAddress::get(DestBB); + Constant *Replacement = + ConstantInt::get(llvm::Type::getInt32Ty(BA->getContext()), 1); + BA->replaceAllUsesWith(ConstantExpr::getIntToPtr(Replacement, + BA->getType())); + BA->destroyConstant(); + } + + // Anything that branched to PredBB now branches to DestBB. + PredBB->replaceAllUsesWith(DestBB); + + if (P) { + ProfileInfo *PI = P->getAnalysisIfAvailable(); + if (PI) { + PI->replaceAllUses(PredBB, DestBB); + PI->removeEdge(ProfileInfo::getEdge(PredBB, DestBB)); + } + } + // Nuke BB. + PredBB->eraseFromParent(); } -// dceInstruction - Inspect the instruction at *BBI and figure out if it's -// [trivially] dead. If so, remove the instruction and update the iterator -// to point to the instruction that immediately succeeded the original -// instruction. -// -bool llvm::dceInstruction(BasicBlock::iterator &BBI) { - // Look for un"used" definitions... - if (isInstructionTriviallyDead(BBI)) { - BBI = BBI->getParent()->getInstList().erase(BBI); // Bye bye +/// CanPropagatePredecessorsForPHIs - Return true if we can fold BB, an +/// almost-empty BB ending in an unconditional branch to Succ, into succ. +/// +/// Assumption: Succ is the single successor for BB. +/// +static bool CanPropagatePredecessorsForPHIs(BasicBlock *BB, BasicBlock *Succ) { + assert(*succ_begin(BB) == Succ && "Succ is not successor of BB!"); + + DEBUG(dbgs() << "Looking to fold " << BB->getName() << " into " + << Succ->getName() << "\n"); + // Shortcut, if there is only a single predecessor it must be BB and merging + // is always safe + if (Succ->getSinglePredecessor()) return true; + + // Make a list of the predecessors of BB + typedef SmallPtrSet BlockSet; + BlockSet BBPreds(pred_begin(BB), pred_end(BB)); + + // Use that list to make another list of common predecessors of BB and Succ + BlockSet CommonPreds; + for (pred_iterator PI = pred_begin(Succ), PE = pred_end(Succ); + PI != PE; ++PI) + if (BBPreds.count(*PI)) + CommonPreds.insert(*PI); + + // Shortcut, if there are no common predecessors, merging is always safe + if (CommonPreds.empty()) return true; + + // Look at all the phi nodes in Succ, to see if they present a conflict when + // merging these blocks + for (BasicBlock::iterator I = Succ->begin(); isa(I); ++I) { + PHINode *PN = cast(I); + + // If the incoming value from BB is again a PHINode in + // BB which has the same incoming value for *PI as PN does, we can + // merge the phi nodes and then the blocks can still be merged + PHINode *BBPN = dyn_cast(PN->getIncomingValueForBlock(BB)); + if (BBPN && BBPN->getParent() == BB) { + for (BlockSet::iterator PI = CommonPreds.begin(), PE = CommonPreds.end(); + PI != PE; PI++) { + if (BBPN->getIncomingValueForBlock(*PI) + != PN->getIncomingValueForBlock(*PI)) { + DEBUG(dbgs() << "Can't fold, phi node " << PN->getName() << " in " + << Succ->getName() << " is conflicting with " + << BBPN->getName() << " with regard to common predecessor " + << (*PI)->getName() << "\n"); + return false; + } + } + } else { + Value* Val = PN->getIncomingValueForBlock(BB); + for (BlockSet::iterator PI = CommonPreds.begin(), PE = CommonPreds.end(); + PI != PE; PI++) { + // See if the incoming value for the common predecessor is equal to the + // one for BB, in which case this phi node will not prevent the merging + // of the block. + if (Val != PN->getIncomingValueForBlock(*PI)) { + DEBUG(dbgs() << "Can't fold, phi node " << PN->getName() << " in " + << Succ->getName() << " is conflicting with regard to common " + << "predecessor " << (*PI)->getName() << "\n"); + return false; + } + } + } } - return false; + + return true; } -//===----------------------------------------------------------------------===// -// PHI Instruction Simplification -// +/// TryToSimplifyUncondBranchFromEmptyBlock - BB is known to contain an +/// unconditional branch, and contains no instructions other than PHI nodes, +/// potential debug intrinsics and the branch. If possible, eliminate BB by +/// rewriting all the predecessors to branch to the successor block and return +/// true. If we can't transform, return false. +bool llvm::TryToSimplifyUncondBranchFromEmptyBlock(BasicBlock *BB) { + // We can't eliminate infinite loops. + BasicBlock *Succ = cast(BB->getTerminator())->getSuccessor(0); + if (BB == Succ) return false; + + // Check to see if merging these blocks would cause conflicts for any of the + // phi nodes in BB or Succ. If not, we can safely merge. + if (!CanPropagatePredecessorsForPHIs(BB, Succ)) return false; + + // Check for cases where Succ has multiple predecessors and a PHI node in BB + // has uses which will not disappear when the PHI nodes are merged. It is + // possible to handle such cases, but difficult: it requires checking whether + // BB dominates Succ, which is non-trivial to calculate in the case where + // Succ has multiple predecessors. Also, it requires checking whether + // constructing the necessary self-referential PHI node doesn't intoduce any + // conflicts; this isn't too difficult, but the previous code for doing this + // was incorrect. + // + // Note that if this check finds a live use, BB dominates Succ, so BB is + // something like a loop pre-header (or rarely, a part of an irreducible CFG); + // folding the branch isn't profitable in that case anyway. + if (!Succ->getSinglePredecessor()) { + BasicBlock::iterator BBI = BB->begin(); + while (isa(*BBI)) { + for (Value::use_iterator UI = BBI->use_begin(), E = BBI->use_end(); + UI != E; ++UI) { + if (PHINode* PN = dyn_cast(*UI)) { + if (PN->getIncomingBlock(UI) != BB) + return false; + } else { + return false; + } + } + ++BBI; + } + } -/// hasConstantValue - If the specified PHI node always merges together the same -/// value, return the value, otherwise return null. -/// -Value *llvm::hasConstantValue(PHINode *PN) { - // If the PHI node only has one incoming value, eliminate the PHI node... - if (PN->getNumIncomingValues() == 1) - return PN->getIncomingValue(0); + DEBUG(dbgs() << "Killing Trivial BB: \n" << *BB); + + if (isa(Succ->begin())) { + // If there is more than one pred of succ, and there are PHI nodes in + // the successor, then we need to add incoming edges for the PHI nodes + // + const SmallVector BBPreds(pred_begin(BB), pred_end(BB)); + + // Loop over all of the PHI nodes in the successor of BB. + for (BasicBlock::iterator I = Succ->begin(); isa(I); ++I) { + PHINode *PN = cast(I); + Value *OldVal = PN->removeIncomingValue(BB, false); + assert(OldVal && "No entry in PHI for Pred BB!"); + + // If this incoming value is one of the PHI nodes in BB, the new entries + // in the PHI node are the entries from the old PHI. + if (isa(OldVal) && cast(OldVal)->getParent() == BB) { + PHINode *OldValPN = cast(OldVal); + for (unsigned i = 0, e = OldValPN->getNumIncomingValues(); i != e; ++i) + // Note that, since we are merging phi nodes and BB and Succ might + // have common predecessors, we could end up with a phi node with + // identical incoming branches. This will be cleaned up later (and + // will trigger asserts if we try to clean it up now, without also + // simplifying the corresponding conditional branch). + PN->addIncoming(OldValPN->getIncomingValue(i), + OldValPN->getIncomingBlock(i)); + } else { + // Add an incoming value for each of the new incoming values. + for (unsigned i = 0, e = BBPreds.size(); i != e; ++i) + PN->addIncoming(OldVal, BBPreds[i]); + } + } + } + + while (PHINode *PN = dyn_cast(&BB->front())) { + if (Succ->getSinglePredecessor()) { + // BB is the only predecessor of Succ, so Succ will end up with exactly + // the same predecessors BB had. + Succ->getInstList().splice(Succ->begin(), + BB->getInstList(), BB->begin()); + } else { + // We explicitly check for such uses in CanPropagatePredecessorsForPHIs. + assert(PN->use_empty() && "There shouldn't be any uses here!"); + PN->eraseFromParent(); + } + } + + // Everything that jumped to BB now goes to Succ. + BB->replaceAllUsesWith(Succ); + if (!Succ->hasName()) Succ->takeName(BB); + BB->eraseFromParent(); // Delete the old basic block. + return true; +} - // Otherwise if all of the incoming values are the same for the PHI, replace - // the PHI node with the incoming value. - // - Value *InVal = 0; - for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) - if (PN->getIncomingValue(i) != PN) // Not the PHI node itself... - if (InVal && PN->getIncomingValue(i) != InVal) - return 0; // Not the same, bail out. - else - InVal = PN->getIncomingValue(i); - - // The only case that could cause InVal to be null is if we have a PHI node - // that only has entries for itself. In this case, there is no entry into the - // loop, so kill the PHI. - // - if (InVal == 0) InVal = Constant::getNullValue(PN->getType()); +/// EliminateDuplicatePHINodes - Check for and eliminate duplicate PHI +/// nodes in this block. This doesn't try to be clever about PHI nodes +/// which differ only in the order of the incoming values, but instcombine +/// orders them so it usually won't matter. +/// +bool llvm::EliminateDuplicatePHINodes(BasicBlock *BB) { + bool Changed = false; + + // This implementation doesn't currently consider undef operands + // specially. Theroetically, two phis which are identical except for + // one having an undef where the other doesn't could be collapsed. + + // Map from PHI hash values to PHI nodes. If multiple PHIs have + // the same hash value, the element is the first PHI in the + // linked list in CollisionMap. + DenseMap HashMap; + + // Maintain linked lists of PHI nodes with common hash values. + DenseMap CollisionMap; + + // Examine each PHI. + for (BasicBlock::iterator I = BB->begin(); + PHINode *PN = dyn_cast(I++); ) { + // Compute a hash value on the operands. Instcombine will likely have sorted + // them, which helps expose duplicates, but we have to check all the + // operands to be safe in case instcombine hasn't run. + uintptr_t Hash = 0; + for (User::op_iterator I = PN->op_begin(), E = PN->op_end(); I != E; ++I) { + // This hash algorithm is quite weak as hash functions go, but it seems + // to do a good enough job for this particular purpose, and is very quick. + Hash ^= reinterpret_cast(static_cast(*I)); + Hash = (Hash << 7) | (Hash >> (sizeof(uintptr_t) * CHAR_BIT - 7)); + } + // If we've never seen this hash value before, it's a unique PHI. + std::pair::iterator, bool> Pair = + HashMap.insert(std::make_pair(Hash, PN)); + if (Pair.second) continue; + // Otherwise it's either a duplicate or a hash collision. + for (PHINode *OtherPN = Pair.first->second; ; ) { + if (OtherPN->isIdenticalTo(PN)) { + // A duplicate. Replace this PHI with its duplicate. + PN->replaceAllUsesWith(OtherPN); + PN->eraseFromParent(); + Changed = true; + break; + } + // A non-duplicate hash collision. + DenseMap::iterator I = CollisionMap.find(OtherPN); + if (I == CollisionMap.end()) { + // Set this PHI to be the head of the linked list of colliding PHIs. + PHINode *Old = Pair.first->second; + Pair.first->second = PN; + CollisionMap[PN] = Old; + break; + } + // Procede to the next PHI in the list. + OtherPN = I->second; + } + } - // All of the incoming values are the same, return the value now. - return InVal; + return Changed; }