#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
+#include "llvm/ADT/DenseSet.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SetVector.h"
#include "llvm/ADT/Statistic.h"
SmallSetVector<MachineBasicBlock*, 8> &Succs);
bool TailDuplicateBlocks(MachineFunction &MF);
bool shouldTailDuplicate(const MachineFunction &MF,
- MachineBasicBlock &TailBB);
- bool TailDuplicate(MachineBasicBlock *TailBB, MachineFunction &MF,
+ bool IsSimple, MachineBasicBlock &TailBB);
+ bool isSimpleBB(MachineBasicBlock *TailBB);
+ bool canCompletelyDuplicateBB(MachineBasicBlock &BB);
+ bool duplicateSimpleBB(MachineBasicBlock *TailBB,
+ SmallVector<MachineBasicBlock*, 8> &TDBBs,
+ const DenseSet<unsigned> &RegsUsedByPhi,
+ SmallVector<MachineInstr*, 16> &Copies);
+ bool TailDuplicate(MachineBasicBlock *TailBB,
+ bool IsSimple,
+ MachineFunction &MF,
SmallVector<MachineBasicBlock*, 8> &TDBBs,
SmallVector<MachineInstr*, 16> &Copies);
+ bool TailDuplicateAndUpdate(MachineBasicBlock *MBB,
+ bool IsSimple,
+ MachineFunction &MF);
+
void RemoveDeadBlock(MachineBasicBlock *MBB);
};
}
}
+/// TailDuplicateAndUpdate - Tail duplicate the block and cleanup.
+bool
+TailDuplicatePass::TailDuplicateAndUpdate(MachineBasicBlock *MBB,
+ bool IsSimple,
+ MachineFunction &MF) {
+ // Save the successors list.
+ SmallSetVector<MachineBasicBlock*, 8> Succs(MBB->succ_begin(),
+ MBB->succ_end());
+
+ SmallVector<MachineBasicBlock*, 8> TDBBs;
+ SmallVector<MachineInstr*, 16> Copies;
+ if (!TailDuplicate(MBB, IsSimple, MF, TDBBs, Copies))
+ return false;
+
+ ++NumTails;
+
+ SmallVector<MachineInstr*, 8> NewPHIs;
+ MachineSSAUpdater SSAUpdate(MF, &NewPHIs);
+
+ // TailBB's immediate successors are now successors of those predecessors
+ // which duplicated TailBB. Add the predecessors as sources to the PHI
+ // instructions.
+ bool isDead = MBB->pred_empty() && !MBB->hasAddressTaken();
+ if (PreRegAlloc)
+ UpdateSuccessorsPHIs(MBB, isDead, TDBBs, Succs);
+
+ // If it is dead, remove it.
+ if (isDead) {
+ NumInstrDups -= MBB->size();
+ RemoveDeadBlock(MBB);
+ ++NumDeadBlocks;
+ }
+
+ // Update SSA form.
+ if (!SSAUpdateVRs.empty()) {
+ for (unsigned i = 0, e = SSAUpdateVRs.size(); i != e; ++i) {
+ unsigned VReg = SSAUpdateVRs[i];
+ SSAUpdate.Initialize(VReg);
+
+ // If the original definition is still around, add it as an available
+ // value.
+ MachineInstr *DefMI = MRI->getVRegDef(VReg);
+ MachineBasicBlock *DefBB = 0;
+ if (DefMI) {
+ DefBB = DefMI->getParent();
+ SSAUpdate.AddAvailableValue(DefBB, VReg);
+ }
+
+ // Add the new vregs as available values.
+ DenseMap<unsigned, AvailableValsTy>::iterator LI =
+ SSAUpdateVals.find(VReg);
+ for (unsigned j = 0, ee = LI->second.size(); j != ee; ++j) {
+ MachineBasicBlock *SrcBB = LI->second[j].first;
+ unsigned SrcReg = LI->second[j].second;
+ SSAUpdate.AddAvailableValue(SrcBB, SrcReg);
+ }
+
+ // Rewrite uses that are outside of the original def's block.
+ MachineRegisterInfo::use_iterator UI = MRI->use_begin(VReg);
+ while (UI != MRI->use_end()) {
+ MachineOperand &UseMO = UI.getOperand();
+ MachineInstr *UseMI = &*UI;
+ ++UI;
+ if (UseMI->isDebugValue()) {
+ // SSAUpdate can replace the use with an undef. That creates
+ // a debug instruction that is a kill.
+ // FIXME: Should it SSAUpdate job to delete debug instructions
+ // instead of replacing the use with undef?
+ UseMI->eraseFromParent();
+ continue;
+ }
+ if (UseMI->getParent() == DefBB && !UseMI->isPHI())
+ continue;
+ SSAUpdate.RewriteUse(UseMO);
+ }
+ }
+
+ SSAUpdateVRs.clear();
+ SSAUpdateVals.clear();
+ }
+
+ // Eliminate some of the copies inserted by tail duplication to maintain
+ // SSA form.
+ for (unsigned i = 0, e = Copies.size(); i != e; ++i) {
+ MachineInstr *Copy = Copies[i];
+ if (!Copy->isCopy())
+ continue;
+ unsigned Dst = Copy->getOperand(0).getReg();
+ unsigned Src = Copy->getOperand(1).getReg();
+ MachineRegisterInfo::use_iterator UI = MRI->use_begin(Src);
+ if (++UI == MRI->use_end()) {
+ // Copy is the only use. Do trivial copy propagation here.
+ MRI->replaceRegWith(Dst, Src);
+ Copy->eraseFromParent();
+ }
+ }
+
+ if (NewPHIs.size())
+ NumAddedPHIs += NewPHIs.size();
+
+ return true;
+}
+
/// TailDuplicateBlocks - Look for small blocks that are unconditionally
/// branched to and do not fall through. Tail-duplicate their instructions
/// into their predecessors to eliminate (dynamic) branches.
VerifyPHIs(MF, true);
}
- SmallVector<MachineInstr*, 8> NewPHIs;
- MachineSSAUpdater SSAUpdate(MF, &NewPHIs);
-
for (MachineFunction::iterator I = ++MF.begin(), E = MF.end(); I != E; ) {
MachineBasicBlock *MBB = I++;
if (NumTails == TailDupLimit)
break;
- // Save the successors list.
- SmallSetVector<MachineBasicBlock*, 8> Succs(MBB->succ_begin(),
- MBB->succ_end());
-
- SmallVector<MachineBasicBlock*, 8> TDBBs;
- SmallVector<MachineInstr*, 16> Copies;
- if (TailDuplicate(MBB, MF, TDBBs, Copies)) {
- ++NumTails;
-
- // TailBB's immediate successors are now successors of those predecessors
- // which duplicated TailBB. Add the predecessors as sources to the PHI
- // instructions.
- bool isDead = MBB->pred_empty();
- if (PreRegAlloc)
- UpdateSuccessorsPHIs(MBB, isDead, TDBBs, Succs);
-
- // If it is dead, remove it.
- if (isDead) {
- NumInstrDups -= MBB->size();
- RemoveDeadBlock(MBB);
- ++NumDeadBlocks;
- }
-
- // Update SSA form.
- if (!SSAUpdateVRs.empty()) {
- for (unsigned i = 0, e = SSAUpdateVRs.size(); i != e; ++i) {
- unsigned VReg = SSAUpdateVRs[i];
- SSAUpdate.Initialize(VReg);
-
- // If the original definition is still around, add it as an available
- // value.
- MachineInstr *DefMI = MRI->getVRegDef(VReg);
- MachineBasicBlock *DefBB = 0;
- if (DefMI) {
- DefBB = DefMI->getParent();
- SSAUpdate.AddAvailableValue(DefBB, VReg);
- }
-
- // Add the new vregs as available values.
- DenseMap<unsigned, AvailableValsTy>::iterator LI =
- SSAUpdateVals.find(VReg);
- for (unsigned j = 0, ee = LI->second.size(); j != ee; ++j) {
- MachineBasicBlock *SrcBB = LI->second[j].first;
- unsigned SrcReg = LI->second[j].second;
- SSAUpdate.AddAvailableValue(SrcBB, SrcReg);
- }
-
- // Rewrite uses that are outside of the original def's block.
- MachineRegisterInfo::use_iterator UI = MRI->use_begin(VReg);
- while (UI != MRI->use_end()) {
- MachineOperand &UseMO = UI.getOperand();
- MachineInstr *UseMI = &*UI;
- ++UI;
- if (UseMI->isDebugValue()) {
- // SSAUpdate can replace the use with an undef. That creates
- // a debug instruction that is a kill.
- // FIXME: Should it SSAUpdate job to delete debug instructions
- // instead of replacing the use with undef?
- UseMI->eraseFromParent();
- continue;
- }
- if (UseMI->getParent() == DefBB && !UseMI->isPHI())
- continue;
- SSAUpdate.RewriteUse(UseMO);
- }
- }
-
- SSAUpdateVRs.clear();
- SSAUpdateVals.clear();
- }
+ bool IsSimple = isSimpleBB(MBB);
- // Eliminate some of the copies inserted by tail duplication to maintain
- // SSA form.
- for (unsigned i = 0, e = Copies.size(); i != e; ++i) {
- MachineInstr *Copy = Copies[i];
- if (!Copy->isCopy())
- continue;
- unsigned Dst = Copy->getOperand(0).getReg();
- unsigned Src = Copy->getOperand(1).getReg();
- MachineRegisterInfo::use_iterator UI = MRI->use_begin(Src);
- if (++UI == MRI->use_end()) {
- // Copy is the only use. Do trivial copy propagation here.
- MRI->replaceRegWith(Dst, Src);
- Copy->eraseFromParent();
- }
- }
+ if (!shouldTailDuplicate(MF, IsSimple, *MBB))
+ continue;
- if (PreRegAlloc && TailDupVerify)
- VerifyPHIs(MF, false);
- MadeChange = true;
- }
+ MadeChange |= TailDuplicateAndUpdate(MBB, IsSimple, MF);
}
- NumAddedPHIs += NewPHIs.size();
+
+ if (PreRegAlloc && TailDupVerify)
+ VerifyPHIs(MF, false);
return MadeChange;
}
/// shouldTailDuplicate - Determine if it is profitable to duplicate this block.
bool
TailDuplicatePass::shouldTailDuplicate(const MachineFunction &MF,
+ bool IsSimple,
MachineBasicBlock &TailBB) {
// Only duplicate blocks that end with unconditional branches.
if (TailBB.canFallThrough())
// to allow undoing the effects of tail merging and other optimizations
// that rearrange the predecessors of the indirect branch.
- if (PreRegAlloc && !TailBB.empty()) {
- const TargetInstrDesc &TID = TailBB.back().getDesc();
- if (TID.isIndirectBranch())
- MaxDuplicateCount = 20;
- }
+ bool HasIndirectbr = false;
+ if (!TailBB.empty())
+ HasIndirectbr = TailBB.back().getDesc().isIndirectBranch();
+
+ if (HasIndirectbr && PreRegAlloc)
+ MaxDuplicateCount = 20;
// Check the instructions in the block to determine whether tail-duplication
// is invalid or unlikely to be profitable.
return false;
}
+ if (HasIndirectbr && PreRegAlloc)
+ return true;
+
+ if (IsSimple)
+ return true;
+
+ if (!PreRegAlloc)
+ return true;
+
+ return canCompletelyDuplicateBB(TailBB);
+}
+
+/// isSimpleBB - True if this BB has only one unconditional jump.
+bool
+TailDuplicatePass::isSimpleBB(MachineBasicBlock *TailBB) {
+ if (TailBB->succ_size() != 1)
+ return false;
+ if (TailBB->pred_empty())
+ return false;
+ MachineBasicBlock::iterator I = TailBB->begin();
+ MachineBasicBlock::iterator E = TailBB->end();
+ while (I != E && I->isDebugValue())
+ ++I;
+ if (I == E)
+ return true;
+ return I->getDesc().isUnconditionalBranch();
+}
+
+static bool
+bothUsedInPHI(const MachineBasicBlock &A,
+ SmallPtrSet<MachineBasicBlock*, 8> SuccsB) {
+ for (MachineBasicBlock::const_succ_iterator SI = A.succ_begin(),
+ SE = A.succ_end(); SI != SE; ++SI) {
+ MachineBasicBlock *BB = *SI;
+ if (SuccsB.count(BB) && !BB->empty() && BB->begin()->isPHI())
+ return true;
+ }
+
+ return false;
+}
+
+bool
+TailDuplicatePass::canCompletelyDuplicateBB(MachineBasicBlock &BB) {
+ SmallPtrSet<MachineBasicBlock*, 8> Succs(BB.succ_begin(), BB.succ_end());
+
+ for (MachineBasicBlock::pred_iterator PI = BB.pred_begin(),
+ PE = BB.pred_end(); PI != PE; ++PI) {
+ MachineBasicBlock *PredBB = *PI;
+
+ if (PredBB->succ_size() > 1)
+ return false;
+
+ MachineBasicBlock *PredTBB = NULL, *PredFBB = NULL;
+ SmallVector<MachineOperand, 4> PredCond;
+ if (TII->AnalyzeBranch(*PredBB, PredTBB, PredFBB, PredCond, true))
+ return false;
+
+ if (!PredCond.empty())
+ return false;
+ }
return true;
}
+bool
+TailDuplicatePass::duplicateSimpleBB(MachineBasicBlock *TailBB,
+ SmallVector<MachineBasicBlock*, 8> &TDBBs,
+ const DenseSet<unsigned> &UsedByPhi,
+ SmallVector<MachineInstr*, 16> &Copies) {
+ SmallPtrSet<MachineBasicBlock*, 8> Succs(TailBB->succ_begin(),
+ TailBB->succ_end());
+ SmallVector<MachineBasicBlock*, 8> Preds(TailBB->pred_begin(),
+ TailBB->pred_end());
+ bool Changed = false;
+ for (SmallSetVector<MachineBasicBlock *, 8>::iterator PI = Preds.begin(),
+ PE = Preds.end(); PI != PE; ++PI) {
+ MachineBasicBlock *PredBB = *PI;
+
+ if (PredBB->getLandingPadSuccessor())
+ continue;
+
+ if (bothUsedInPHI(*PredBB, Succs))
+ continue;
+
+ MachineBasicBlock *PredTBB = NULL, *PredFBB = NULL;
+ SmallVector<MachineOperand, 4> PredCond;
+ if (TII->AnalyzeBranch(*PredBB, PredTBB, PredFBB, PredCond, true))
+ continue;
+
+ Changed = true;
+ DEBUG(dbgs() << "\nTail-duplicating into PredBB: " << *PredBB
+ << "From simple Succ: " << *TailBB);
+
+ MachineBasicBlock *NewTarget = *TailBB->succ_begin();
+ MachineBasicBlock *NextBB = llvm::next(MachineFunction::iterator(PredBB));
+
+ // Make PredFBB explicit.
+ if (PredCond.empty())
+ PredFBB = PredTBB;
+
+ // Make fall through explicit.
+ if (!PredTBB)
+ PredTBB = NextBB;
+ if (!PredFBB)
+ PredFBB = NextBB;
+
+ // Redirect
+ if (PredFBB == TailBB)
+ PredFBB = NewTarget;
+ if (PredTBB == TailBB)
+ PredTBB = NewTarget;
+
+ // Make the branch unconditional if possible
+ if (PredTBB == PredFBB) {
+ PredCond.clear();
+ PredFBB = NULL;
+ }
+
+ // Avoid adding fall through branches.
+ if (PredFBB == NextBB)
+ PredFBB = NULL;
+ if (PredTBB == NextBB && PredFBB == NULL)
+ PredTBB = NULL;
+
+ TII->RemoveBranch(*PredBB);
+
+ if (PredTBB)
+ TII->InsertBranch(*PredBB, PredTBB, PredFBB, PredCond, DebugLoc());
+
+ PredBB->removeSuccessor(TailBB);
+ unsigned NumSuccessors = PredBB->succ_size();
+ assert(NumSuccessors <= 1);
+ if (NumSuccessors == 0 || *PredBB->succ_begin() != NewTarget)
+ PredBB->addSuccessor(NewTarget);
+
+ TDBBs.push_back(PredBB);
+ }
+ return Changed;
+}
+
/// TailDuplicate - If it is profitable, duplicate TailBB's contents in each
/// of its predecessors.
bool
-TailDuplicatePass::TailDuplicate(MachineBasicBlock *TailBB, MachineFunction &MF,
+TailDuplicatePass::TailDuplicate(MachineBasicBlock *TailBB,
+ bool IsSimple,
+ MachineFunction &MF,
SmallVector<MachineBasicBlock*, 8> &TDBBs,
SmallVector<MachineInstr*, 16> &Copies) {
- if (!shouldTailDuplicate(MF, *TailBB))
- return false;
-
DEBUG(dbgs() << "\n*** Tail-duplicating BB#" << TailBB->getNumber() << '\n');
+ DenseSet<unsigned> UsedByPhi;
+ getRegsUsedByPHIs(*TailBB, &UsedByPhi);
+
+ if (IsSimple)
+ return duplicateSimpleBB(TailBB, TDBBs, UsedByPhi, Copies);
+
// Iterate through all the unique predecessors and tail-duplicate this
// block into them, if possible. Copying the list ahead of time also
// avoids trouble with the predecessor list reallocating.
bool Changed = false;
SmallSetVector<MachineBasicBlock*, 8> Preds(TailBB->pred_begin(),
TailBB->pred_end());
- DenseSet<unsigned> UsedByPhi;
- getRegsUsedByPHIs(*TailBB, &UsedByPhi);
for (SmallSetVector<MachineBasicBlock *, 8>::iterator PI = Preds.begin(),
PE = Preds.end(); PI != PE; ++PI) {
MachineBasicBlock *PredBB = *PI;
projects / oota-llvm.git / blobdiff