//
// 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.
//
//===----------------------------------------------------------------------===//
//
// unlikely, that the return returns something else (like constant 0), and
// can still be TRE'd. It can be TRE'd if ALL OTHER return instructions in
// the function return the exact same value.
-// 4. If it can prove that callees do not access theier caller stack frame,
+// 4. If it can prove that callees do not access their caller stack frame,
// they are marked as eligible for tail call elimination (by the code
// generator).
//
//
//===----------------------------------------------------------------------===//
+#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/Instructions.h"
#include "llvm/Pass.h"
+#include "llvm/Analysis/CaptureTracking.h"
#include "llvm/Support/CFG.h"
#include "llvm/ADT/Statistic.h"
using namespace llvm;
-namespace {
- Statistic<> NumEliminated("tailcallelim", "Number of tail calls removed");
- Statistic<> NumAccumAdded("tailcallelim","Number of accumulators introduced");
+STATISTIC(NumEliminated, "Number of tail calls removed");
+STATISTIC(NumAccumAdded, "Number of accumulators introduced");
+namespace {
struct TailCallElim : public FunctionPass {
+ static char ID; // Pass identification, replacement for typeid
+ TailCallElim() : FunctionPass(&ID) {}
+
virtual bool runOnFunction(Function &F);
private:
bool ProcessReturningBlock(ReturnInst *RI, BasicBlock *&OldEntry,
bool &TailCallsAreMarkedTail,
- std::vector<PHINode*> &ArgumentPHIs,
+ SmallVector<PHINode*, 8> &ArgumentPHIs,
bool CannotTailCallElimCallsMarkedTail);
bool CanMoveAboveCall(Instruction *I, CallInst *CI);
Value *CanTransformAccumulatorRecursion(Instruction *I, CallInst *CI);
};
- RegisterPass<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();
}
-
/// AllocaMightEscapeToCalls - Return true if this alloca may be accessed by
/// callees of this function. We only do very simple analysis right now, this
/// could be expanded in the future to use mod/ref information for particular
return true;
}
-/// FunctionContainsAllocas - Scan the specified basic block for alloca
+/// CheckForEscapingAllocas - Scan the specified basic block for alloca
/// instructions. If it contains any that might be accessed by calls, return
/// true.
static bool CheckForEscapingAllocas(BasicBlock *BB,
// 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;
}
BasicBlock *OldEntry = 0;
bool TailCallsAreMarkedTail = false;
- std::vector<PHINode*> ArgumentPHIs;
+ SmallVector<PHINode*, 8> ArgumentPHIs;
bool MadeChange = false;
bool FunctionContainsEscapingAllocas = false;
/// 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
// We currently handle static constants and arguments that are not modified as
// part of the recursion.
//
-static bool isDynamicConstant(Value *V, CallInst *CI) {
+static bool isDynamicConstant(Value *V, CallInst *CI, ReturnInst *RI) {
if (isa<Constant>(V)) return true; // Static constants are always dyn consts
// Check to see if this is an immutable argument, if so, the value
if (CI->getOperand(ArgNo+1) == Arg)
return true;
}
+
+ // Switch cases are always constant integers. If the value is being switched
+ // on and the return is only reachable from one of its cases, it's
+ // effectively constant.
+ if (BasicBlock *UniquePred = RI->getParent()->getUniquePredecessor())
+ if (SwitchInst *SI = dyn_cast<SwitchInst>(UniquePred->getTerminator()))
+ if (SI->getCondition() == V)
+ return SI->getDefaultDest() != RI->getParent();
+
// Not a constant or immutable argument, we can't safely transform.
return false;
}
// evaluatable at the start of the initial invocation of the function,
// instead of at the end of the evaluation.
//
- if (!isDynamicConstant(RetOp, CI))
+ if (!isDynamicConstant(RetOp, CI, RI))
return 0;
if (ReturnedValue && RetOp != ReturnedValue)
"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.
bool TailCallElim::ProcessReturningBlock(ReturnInst *Ret, BasicBlock *&OldEntry,
bool &TailCallsAreMarkedTail,
- std::vector<PHINode*> &ArgumentPHIs,
+ SmallVector<PHINode*, 8> &ArgumentPHIs,
bool CannotTailCallElimCallsMarkedTail) {
BasicBlock *BB = Ret->getParent();
Function *F = BB->getParent();
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))
// 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->getContext(), "", 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;
projects / oota-llvm.git / blobdiff