//
// 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.
//
//===----------------------------------------------------------------------===//
//
// transformation from taking place, though currently the analysis cannot
// support moving any really useful instructions (only dead ones).
// 2. This pass transforms functions that are prevented from being tail
-// recursive by an associative expression to use an accumulator variable,
-// thus compiling the typical naive factorial or 'fib' implementation into
-// efficient code.
+// recursive by an associative and commutative expression to use an
+// accumulator variable, thus compiling the typical naive factorial or
+// 'fib' implementation into efficient code.
// 3. TRE is performed if the function returns void, if the return
// returns the result returned by the call, or if the function returns a
// run-time constant on all exits from the function. It is possible, though
// 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/Analysis/InlineCost.h"
+#include "llvm/Analysis/Loads.h"
+#include "llvm/Support/CallSite.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 {
- struct VISIBILITY_HIDDEN TailCallElim : public FunctionPass {
- static const int ID; // Pass identifcation, replacement for typeid
- TailCallElim() : FunctionPass((intptr_t)&ID) {}
+ 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);
};
- const int TailCallElim::ID = 0;
- RegisterPass<TailCallElim> X("tailcallelim", "Tail Call Elimination");
}
+char TailCallElim::ID = 0;
+INITIALIZE_PASS(TailCallElim, "tailcallelim",
+ "Tail Call Elimination", false, false);
+
// 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,
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)
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,
+ L->getAlignment()))
+ 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 we are passing this argument into call as the corresponding
// argument operand, then the argument is dynamically constant.
// Otherwise, we cannot transform this function safely.
- if (CI->getOperand(ArgNo+1) == Arg)
+ if (CI->getArgOperand(ArgNo) == 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;
}
// getCommonReturnValue - Check to see if the function containing the specified
-// return instruction and tail call consistently returns the same
-// runtime-constant value at all exit points. If so, return the returned value.
+// tail call consistently returns the same runtime-constant value at all exit
+// points except for IgnoreRI. If so, return the returned value.
//
-static Value *getCommonReturnValue(ReturnInst *TheRI, CallInst *CI) {
- Function *F = TheRI->getParent()->getParent();
+static Value *getCommonReturnValue(ReturnInst *IgnoreRI, CallInst *CI) {
+ Function *F = CI->getParent()->getParent();
Value *ReturnedValue = 0;
for (Function::iterator BBI = F->begin(), E = F->end(); BBI != E; ++BBI)
if (ReturnInst *RI = dyn_cast<ReturnInst>(BBI->getTerminator()))
- if (RI != TheRI) {
+ if (RI != IgnoreRI) {
Value *RetOp = RI->getOperand(0);
// We can only perform this transformation if the value returned is
// 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)
///
Value *TailCallElim::CanTransformAccumulatorRecursion(Instruction *I,
CallInst *CI) {
- if (!I->isAssociative()) return 0;
+ if (!I->isAssociative() || !I->isCommutative()) return 0;
assert(I->getNumOperands() == 2 &&
- "Associative operations should have 2 args!");
+ "Associative/commutative 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;
-
+
// Scan backwards from the return, checking to see if there is a tail call in
// this block. If so, set CI to it.
CallInst *CI;
if (CI->isTailCall() && CannotTailCallElimCallsMarkedTail)
return false;
- // If we are introducing accumulator recursion to eliminate associative
- // operations after the call instruction, this variable contains the initial
- // value for the accumulator. If this value is set, we actually perform
- // accumulator recursion elimination instead of simple tail recursion
- // elimination.
+ // As a special case, detect code like this:
+ // double fabs(double f) { return __builtin_fabs(f); } // a 'fabs' call
+ // and disable this xform in this case, because the code generator will
+ // lower the call to fabs into inline code.
+ if (BB == &F->getEntryBlock() &&
+ &BB->front() == CI && &*++BB->begin() == Ret &&
+ callIsSmall(F)) {
+ // A single-block function with just a call and a return. Check that
+ // the arguments match.
+ CallSite::arg_iterator I = CallSite(CI).arg_begin(),
+ E = CallSite(CI).arg_end();
+ Function::arg_iterator FI = F->arg_begin(),
+ FE = F->arg_end();
+ for (; I != E && FI != FE; ++I, ++FI)
+ if (*I != &*FI) break;
+ if (I == E && FI == FE)
+ return false;
+ }
+
+ // If we are introducing accumulator recursion to eliminate operations after
+ // the call instruction that are both associative and commutative, the initial
+ // value for the accumulator is placed in this variable. If this value is set
+ // then we actually perform accumulator recursion elimination instead of
+ // simple tail recursion elimination. If the operation is an LLVM instruction
+ // (eg: "add") then it is recorded in AccumulatorRecursionInstr. If not, then
+ // we are handling the case when the return instruction returns a constant C
+ // which is different to the constant returned by other return instructions
+ // (which is recorded in AccumulatorRecursionEliminationInitVal). This is a
+ // special case of accumulator recursion, the operation being "return C".
Value *AccumulatorRecursionEliminationInitVal = 0;
Instruction *AccumulatorRecursionInstr = 0;
for (BBI = CI, ++BBI; &*BBI != Ret; ++BBI)
if (!CanMoveAboveCall(BBI, CI)) {
// If we can't move the instruction above the call, it might be because it
- // is an associative operation that could be tranformed using accumulator
- // recursion elimination. Check to see if this is the case, and if so,
- // remember the initial accumulator value for later.
+ // is an associative and commutative operation that could be tranformed
+ // using accumulator recursion elimination. Check to see if this is the
+ // case, and if so, remember the initial accumulator value for later.
if ((AccumulatorRecursionEliminationInitVal =
CanTransformAccumulatorRecursion(BBI, CI))) {
// Yes, this is accumulator recursion. Remember which instruction
// 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;
+ !getCommonReturnValue(0, CI)) {
+ // One case remains that we are able to handle: the current return
+ // instruction returns a constant, and all other return instructions
+ // return a different constant.
+ if (!isDynamicConstant(Ret->getReturnValue(), CI, Ret))
+ return false; // Current return instruction does not return a constant.
+ // Check that all other return instructions return a common constant. If
+ // so, record it in AccumulatorRecursionEliminationInitVal.
+ AccumulatorRecursionEliminationInitVal = getCommonReturnValue(Ret, CI);
+ if (!AccumulatorRecursionEliminationInitVal)
+ return false;
+ }
// OK! We can transform this tail call. If this is the first one found,
// create the new entry block, allowing us to branch back to the old entry.
if (OldEntry == 0) {
OldEntry = &F->getEntryBlock();
- BasicBlock *NewEntry = new BasicBlock("", F, OldEntry);
+ BasicBlock *NewEntry = BasicBlock::Create(F->getContext(), "", F, OldEntry);
NewEntry->takeName(OldEntry);
OldEntry->setName("tailrecurse");
- new BranchInst(OldEntry, NewEntry);
+ 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);
// Ok, now that we know we have a pseudo-entry block WITH all of the
// required PHI nodes, add entries into the PHI node for the actual
// parameters passed into the tail-recursive call.
- for (unsigned i = 0, e = CI->getNumOperands()-1; i != e; ++i)
- ArgumentPHIs[i]->addIncoming(CI->getOperand(i+1), BB);
+ for (unsigned i = 0, e = CI->getNumArgOperands(); i != e; ++i)
+ ArgumentPHIs[i]->addIncoming(CI->getArgOperand(i), BB);
// If we are introducing an accumulator variable to eliminate the recursion,
// do so now. Note that we _know_ that no subsequent tail recursion
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(AccumulatorRecursionEliminationInitVal->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,
// it will not show up as a predecessor.
for (pred_iterator PI = pred_begin(OldEntry), PE = pred_end(OldEntry);
PI != PE; ++PI) {
- if (*PI == &F->getEntryBlock())
- AccPN->addIncoming(AccumulatorRecursionEliminationInitVal, *PI);
+ BasicBlock *P = *PI;
+ if (P == &F->getEntryBlock())
+ AccPN->addIncoming(AccumulatorRecursionEliminationInitVal, P);
else
- AccPN->addIncoming(AccPN, *PI);
+ AccPN->addIncoming(AccPN, P);
}
- // Add an incoming argument for the current block, which is computed by our
- // associative accumulator instruction.
- AccPN->addIncoming(AccRecInstr, BB);
-
- // Next, rewrite the accumulator recursion instruction so that it does not
- // use the result of the call anymore, instead, use the PHI node we just
- // inserted.
- AccRecInstr->setOperand(AccRecInstr->getOperand(0) != CI, AccPN);
+ if (AccRecInstr) {
+ // Add an incoming argument for the current block, which is computed by
+ // our associative and commutative accumulator instruction.
+ AccPN->addIncoming(AccRecInstr, BB);
+
+ // Next, rewrite the accumulator recursion instruction so that it does not
+ // use the result of the call anymore, instead, use the PHI node we just
+ // inserted.
+ AccRecInstr->setOperand(AccRecInstr->getOperand(0) != CI, AccPN);
+ } else {
+ // Add an incoming argument for the current block, which is just the
+ // constant returned by the current return instruction.
+ AccPN->addIncoming(Ret->getReturnValue(), BB);
+ }
// Finally, rewrite any return instructions in the program to return the PHI
// node instead of the "initval" that they do currently. This loop will
// 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