//
// 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.
//
//===----------------------------------------------------------------------===//
//
//
//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "tailcallelim"
#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/Pass.h"
#include "llvm/Support/CFG.h"
#include "llvm/ADT/Statistic.h"
+#include "llvm/Support/Compiler.h"
using namespace llvm;
+STATISTIC(NumEliminated, "Number of tail calls removed");
+STATISTIC(NumAccumAdded, "Number of accumulators introduced");
+
namespace {
- Statistic<> NumEliminated("tailcallelim", "Number of tail calls removed");
- Statistic<> NumAccumAdded("tailcallelim","Number of accumulators introduced");
+ struct VISIBILITY_HIDDEN TailCallElim : public FunctionPass {
+ static char ID; // Pass identification, replacement for typeid
+ TailCallElim() : FunctionPass(&ID) {}
- struct TailCallElim : public FunctionPass {
virtual bool runOnFunction(Function &F);
private:
bool CanMoveAboveCall(Instruction *I, CallInst *CI);
Value *CanTransformAccumulatorRecursion(Instruction *I, CallInst *CI);
};
- RegisterOpt<TailCallElim> X("tailcallelim", "Tail Call Elimination");
}
+char TailCallElim::ID = 0;
+static RegisterPass<TailCallElim> X("tailcallelim", "Tail Call Elimination");
+
// Public interface to the TailCallElimination pass
FunctionPass *llvm::createTailCallEliminationPass() {
return new TailCallElim();
// If this alloca is in the body of the function, or if it is a variable
// sized allocation, we cannot tail call eliminate calls marked 'tail'
// with this mechanism.
- if (BB != &BB->getParent()->front() ||
+ if (BB != &BB->getParent()->getEntryBlock() ||
!isa<ConstantInt>(AI->getArraySize()))
CannotTCETailMarkedCall = true;
}
FunctionContainsEscapingAllocas |=
CheckForEscapingAllocas(BB, CannotTCETailMarkedCall);
}
+
+ /// FIXME: The code generator produces really bad code when an 'escaping
+ /// alloca' is changed from being a static alloca to being a dynamic alloca.
+ /// Until this is resolved, disable this transformation if that would ever
+ /// happen. This bug is PR962.
+ if (FunctionContainsEscapingAllocas)
+ return false;
+
// Second pass, change any tail calls to loops.
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
// occurs when a function passes an argument straight through to its tail
// call.
if (!ArgumentPHIs.empty()) {
- unsigned NumIncoming = ArgumentPHIs[0]->getNumIncomingValues();
for (unsigned i = 0, e = ArgumentPHIs.size(); i != e; ++i) {
PHINode *PN = ArgumentPHIs[i];
if (!FunctionContainsEscapingAllocas)
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
- if (CallInst *CI = dyn_cast<CallInst>(I))
+ if (CallInst *CI = dyn_cast<CallInst>(I)) {
CI->setTailCall();
+ MadeChange = true;
+ }
return MadeChange;
}
bool TailCallElim::CanMoveAboveCall(Instruction *I, CallInst *CI) {
// FIXME: We can move load/store/call/free instructions above the call if the
// call does not mod/ref the memory location being processed.
- if (I->mayWriteToMemory() || isa<LoadInst>(I))
+ if (I->mayHaveSideEffects()) // This also handles volatile loads.
return false;
+
+ if (LoadInst* L = dyn_cast<LoadInst>(I)) {
+ // Loads may always be moved above calls without side effects.
+ if (CI->mayHaveSideEffects()) {
+ // Non-volatile loads may be moved above a call with side effects if it
+ // does not write to memory and the load provably won't trap.
+ // FIXME: Writes to memory only matter if they may alias the pointer
+ // being loaded from.
+ if (CI->mayWriteToMemory() ||
+ !isSafeToLoadUnconditionally(L->getPointerOperand(), L))
+ return false;
+ }
+ }
// Otherwise, if this is a side-effect free instruction, check to make sure
// that it does not use the return value of the call. If it doesn't use the
Function *F = TheRI->getParent()->getParent();
Value *ReturnedValue = 0;
+ // TODO: Handle multiple value ret instructions;
+ if (isa<StructType>(F->getReturnType()))
+ return 0;
+
for (Function::iterator BBI = F->begin(), E = F->end(); BBI != E; ++BBI)
if (ReturnInst *RI = dyn_cast<ReturnInst>(BBI->getTerminator()))
if (RI != TheRI) {
"Associative operations should have 2 args!");
// Exactly one operand should be the result of the call instruction...
- if (I->getOperand(0) == CI && I->getOperand(1) == CI ||
- I->getOperand(0) != CI && I->getOperand(1) != CI)
+ if ((I->getOperand(0) == CI && I->getOperand(1) == CI) ||
+ (I->getOperand(0) != CI && I->getOperand(1) != CI))
return 0;
// The only user of this instruction we allow is a single return instruction.
if (&BB->front() == Ret) // Make sure there is something before the ret...
return false;
+
+ // If the return is in the entry block, then making this transformation would
+ // turn infinite recursion into an infinite loop. This transformation is ok
+ // in theory, but breaks some code like:
+ // double fabs(double f) { return __builtin_fabs(f); } // a 'fabs' call
+ // disable this xform in this case, because the code generator will lower the
+ // call to fabs into inline code.
+ if (BB == &F->getEntryBlock())
+ return false;
// Scan backwards from the return, checking to see if there is a tail call in
// this block. If so, set CI to it.
// of the call and return void, ignore the value of the call and return a
// constant, return the value returned by the tail call, or that are being
// accumulator recursion variable eliminated.
- if (Ret->getNumOperands() != 0 && Ret->getReturnValue() != CI &&
+ if (Ret->getNumOperands() == 1 && Ret->getReturnValue() != CI &&
+ !isa<UndefValue>(Ret->getReturnValue()) &&
AccumulatorRecursionEliminationInitVal == 0 &&
!getCommonReturnValue(Ret, CI))
return false;
// create the new entry block, allowing us to branch back to the old entry.
if (OldEntry == 0) {
OldEntry = &F->getEntryBlock();
- std::string OldName = OldEntry->getName(); OldEntry->setName("tailrecurse");
- BasicBlock *NewEntry = new BasicBlock(OldName, F, OldEntry);
- new BranchInst(OldEntry, NewEntry);
+ BasicBlock *NewEntry = BasicBlock::Create("", F, OldEntry);
+ NewEntry->takeName(OldEntry);
+ OldEntry->setName("tailrecurse");
+ BranchInst::Create(OldEntry, NewEntry);
// If this tail call is marked 'tail' and if there are any allocas in the
// entry block, move them up to the new entry block.
Instruction *InsertPos = OldEntry->begin();
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I) {
- PHINode *PN = new PHINode(I->getType(), I->getName()+".tr", InsertPos);
+ PHINode *PN = PHINode::Create(I->getType(),
+ I->getName() + ".tr", InsertPos);
I->replaceAllUsesWith(PN); // Everyone use the PHI node now!
PN->addIncoming(I, NewEntry);
ArgumentPHIs.push_back(PN);
if (AccumulatorRecursionEliminationInitVal) {
Instruction *AccRecInstr = AccumulatorRecursionInstr;
// Start by inserting a new PHI node for the accumulator.
- PHINode *AccPN = new PHINode(AccRecInstr->getType(), "accumulator.tr",
- OldEntry->begin());
+ PHINode *AccPN = PHINode::Create(AccRecInstr->getType(), "accumulator.tr",
+ OldEntry->begin());
// Loop over all of the predecessors of the tail recursion block. For the
// real entry into the function we seed the PHI with the initial value,
// Now that all of the PHI nodes are in place, remove the call and
// ret instructions, replacing them with an unconditional branch.
- new BranchInst(OldEntry, Ret);
+ BranchInst::Create(OldEntry, Ret);
BB->getInstList().erase(Ret); // Remove return.
BB->getInstList().erase(CI); // Remove call.
++NumEliminated;