//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "machine-cse"
#include "llvm/CodeGen/Passes.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/ScopedHashTable.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/RecyclingAllocator.h"
+#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
using namespace llvm;
+#define DEBUG_TYPE "machine-cse"
+
STATISTIC(NumCoalesces, "Number of copies coalesced");
STATISTIC(NumCSEs, "Number of common subexpression eliminated");
STATISTIC(NumPhysCSEs,
MachineRegisterInfo *MRI;
public:
static char ID; // Pass identification
- MachineCSE() : MachineFunctionPass(ID), LookAheadLimit(5), CurrVN(0) {
+ MachineCSE() : MachineFunctionPass(ID), LookAheadLimit(0), CurrVN(0) {
initializeMachineCSEPass(*PassRegistry::getPassRegistry());
}
- virtual bool runOnMachineFunction(MachineFunction &MF);
+ bool runOnMachineFunction(MachineFunction &MF) override;
- virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.setPreservesCFG();
MachineFunctionPass::getAnalysisUsage(AU);
AU.addRequired<AliasAnalysis>();
AU.addPreserved<MachineDominatorTree>();
}
- virtual void releaseMemory() {
+ void releaseMemory() override {
ScopeMap.clear();
Exps.clear();
}
private:
- const unsigned LookAheadLimit;
+ unsigned LookAheadLimit;
typedef RecyclingAllocator<BumpPtrAllocator,
ScopedHashTableVal<MachineInstr*, unsigned> > AllocatorTy;
typedef ScopedHashTable<MachineInstr*, unsigned,
SmallVector<MachineInstr*, 64> Exps;
unsigned CurrVN;
- bool PerformTrivialCoalescing(MachineInstr *MI, MachineBasicBlock *MBB);
+ bool PerformTrivialCopyPropagation(MachineInstr *MI,
+ MachineBasicBlock *MBB);
bool isPhysDefTriviallyDead(unsigned Reg,
MachineBasicBlock::const_iterator I,
MachineBasicBlock::const_iterator E) const;
INITIALIZE_PASS_END(MachineCSE, "machine-cse",
"Machine Common Subexpression Elimination", false, false)
-bool MachineCSE::PerformTrivialCoalescing(MachineInstr *MI,
- MachineBasicBlock *MBB) {
+/// The source register of a COPY machine instruction can be propagated to all
+/// its users, and this propagation could increase the probability of finding
+/// common subexpressions. If the COPY has only one user, the COPY itself can
+/// be removed.
+bool MachineCSE::PerformTrivialCopyPropagation(MachineInstr *MI,
+ MachineBasicBlock *MBB) {
bool Changed = false;
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
MachineOperand &MO = MI->getOperand(i);
unsigned Reg = MO.getReg();
if (!TargetRegisterInfo::isVirtualRegister(Reg))
continue;
- if (!MRI->hasOneNonDBGUse(Reg))
- // Only coalesce single use copies. This ensure the copy will be
- // deleted.
- continue;
+ bool OnlyOneUse = MRI->hasOneNonDBGUse(Reg);
MachineInstr *DefMI = MRI->getVRegDef(Reg);
if (!DefMI->isCopy())
continue;
unsigned SrcReg = DefMI->getOperand(1).getReg();
if (!TargetRegisterInfo::isVirtualRegister(SrcReg))
continue;
- if (DefMI->getOperand(0).getSubReg() || DefMI->getOperand(1).getSubReg())
+ if (DefMI->getOperand(0).getSubReg())
continue;
- if (!MRI->constrainRegClass(SrcReg, MRI->getRegClass(Reg)))
+ // FIXME: We should trivially coalesce subregister copies to expose CSE
+ // opportunities on instructions with truncated operands (see
+ // cse-add-with-overflow.ll). This can be done here as follows:
+ // if (SrcSubReg)
+ // RC = TRI->getMatchingSuperRegClass(MRI->getRegClass(SrcReg), RC,
+ // SrcSubReg);
+ // MO.substVirtReg(SrcReg, SrcSubReg, *TRI);
+ //
+ // The 2-addr pass has been updated to handle coalesced subregs. However,
+ // some machine-specific code still can't handle it.
+ // To handle it properly we also need a way find a constrained subregister
+ // class given a super-reg class and subreg index.
+ if (DefMI->getOperand(1).getSubReg())
+ continue;
+ const TargetRegisterClass *RC = MRI->getRegClass(Reg);
+ if (!MRI->constrainRegClass(SrcReg, RC))
continue;
DEBUG(dbgs() << "Coalescing: " << *DefMI);
DEBUG(dbgs() << "*** to: " << *MI);
+ // Propagate SrcReg of copies to MI.
MO.setReg(SrcReg);
MRI->clearKillFlags(SrcReg);
- DefMI->eraseFromParent();
- ++NumCoalesces;
+ // Coalesce single use copies.
+ if (OnlyOneUse) {
+ DefMI->eraseFromParent();
+ ++NumCoalesces;
+ }
Changed = true;
}
// Next, collect all defs into PhysDefs. If any is already in PhysRefs
// (which currently contains only uses), set the PhysUseDef flag.
PhysUseDef = false;
- MachineBasicBlock::const_iterator I = MI; I = llvm::next(I);
+ MachineBasicBlock::const_iterator I = MI; I = std::next(I);
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
const MachineOperand &MO = MI->getOperand(i);
if (!MO.isReg() || !MO.isDef())
}
CrossMBB = true;
}
- MachineBasicBlock::const_iterator I = CSMI; I = llvm::next(I);
+ MachineBasicBlock::const_iterator I = CSMI; I = std::next(I);
MachineBasicBlock::const_iterator E = MI;
MachineBasicBlock::const_iterator EE = CSMBB->end();
unsigned LookAheadLeft = LookAheadLimit;
}
bool MachineCSE::isCSECandidate(MachineInstr *MI) {
- if (MI->isLabel() || MI->isPHI() || MI->isImplicitDef() ||
- MI->isKill() || MI->isInlineAsm() || MI->isDebugValue())
+ if (MI->isPosition() || MI->isPHI() || MI->isImplicitDef() || MI->isKill() ||
+ MI->isInlineAsm() || MI->isDebugValue())
return false;
// Ignore copies.
TargetRegisterInfo::isVirtualRegister(Reg)) {
MayIncreasePressure = false;
SmallPtrSet<MachineInstr*, 8> CSUses;
- for (MachineRegisterInfo::use_nodbg_iterator I =MRI->use_nodbg_begin(CSReg),
- E = MRI->use_nodbg_end(); I != E; ++I) {
- MachineInstr *Use = &*I;
- CSUses.insert(Use);
+ for (MachineInstr &MI : MRI->use_nodbg_instructions(CSReg)) {
+ CSUses.insert(&MI);
}
- for (MachineRegisterInfo::use_nodbg_iterator I = MRI->use_nodbg_begin(Reg),
- E = MRI->use_nodbg_end(); I != E; ++I) {
- MachineInstr *Use = &*I;
- if (!CSUses.count(Use)) {
+ for (MachineInstr &MI : MRI->use_nodbg_instructions(Reg)) {
+ if (!CSUses.count(&MI)) {
MayIncreasePressure = true;
break;
}
// Heuristics #1: Don't CSE "cheap" computation if the def is not local or in
// an immediate predecessor. We don't want to increase register pressure and
// end up causing other computation to be spilled.
- if (MI->isAsCheapAsAMove()) {
+ if (TII->isAsCheapAsAMove(MI)) {
MachineBasicBlock *CSBB = CSMI->getParent();
MachineBasicBlock *BB = MI->getParent();
if (CSBB != BB && !CSBB->isSuccessor(BB))
}
if (!HasVRegUse) {
bool HasNonCopyUse = false;
- for (MachineRegisterInfo::use_nodbg_iterator I = MRI->use_nodbg_begin(Reg),
- E = MRI->use_nodbg_end(); I != E; ++I) {
- MachineInstr *Use = &*I;
+ for (MachineInstr &MI : MRI->use_nodbg_instructions(Reg)) {
// Ignore copies.
- if (!Use->isCopyLike()) {
+ if (!MI.isCopyLike()) {
HasNonCopyUse = true;
break;
}
// it unless the defined value is already used in the BB of the new use.
bool HasPHI = false;
SmallPtrSet<MachineBasicBlock*, 4> CSBBs;
- for (MachineRegisterInfo::use_nodbg_iterator I = MRI->use_nodbg_begin(CSReg),
- E = MRI->use_nodbg_end(); I != E; ++I) {
- MachineInstr *Use = &*I;
- HasPHI |= Use->isPHI();
- CSBBs.insert(Use->getParent());
+ for (MachineInstr &MI : MRI->use_nodbg_instructions(CSReg)) {
+ HasPHI |= MI.isPHI();
+ CSBBs.insert(MI.getParent());
}
if (!HasPHI)
SmallVector<std::pair<unsigned, unsigned>, 8> CSEPairs;
SmallVector<unsigned, 2> ImplicitDefsToUpdate;
+ SmallVector<unsigned, 2> ImplicitDefs;
for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end(); I != E; ) {
MachineInstr *MI = &*I;
++I;
bool FoundCSE = VNT.count(MI);
if (!FoundCSE) {
- // Look for trivial copy coalescing opportunities.
- if (PerformTrivialCoalescing(MI, MBB)) {
+ // Using trivial copy propagation to find more CSE opportunities.
+ if (PerformTrivialCopyPropagation(MI, MBB)) {
Changed = true;
// After coalescing MI itself may become a copy.
if (MI->isCopyLike())
continue;
+
+ // Try again to see if CSE is possible.
FoundCSE = VNT.count(MI);
}
}
bool DoCSE = true;
unsigned NumDefs = MI->getDesc().getNumDefs() +
MI->getDesc().getNumImplicitDefs();
-
+
for (unsigned i = 0, e = MI->getNumOperands(); NumDefs && i != e; ++i) {
MachineOperand &MO = MI->getOperand(i);
if (!MO.isReg() || !MO.isDef())
// we should make sure it is not dead at CSMI.
if (MO.isImplicit() && !MO.isDead() && CSMI->getOperand(i).isDead())
ImplicitDefsToUpdate.push_back(i);
+
+ // Keep track of implicit defs of CSMI and MI, to clear possibly
+ // made-redundant kill flags.
+ if (MO.isImplicit() && !MO.isDead() && OldReg == NewReg)
+ ImplicitDefs.push_back(OldReg);
+
if (OldReg == NewReg) {
--NumDefs;
continue;
// Actually perform the elimination.
if (DoCSE) {
for (unsigned i = 0, e = CSEPairs.size(); i != e; ++i) {
- MRI->replaceRegWith(CSEPairs[i].first, CSEPairs[i].second);
- MRI->clearKillFlags(CSEPairs[i].second);
+ unsigned OldReg = CSEPairs[i].first;
+ unsigned NewReg = CSEPairs[i].second;
+ // OldReg may have been unused but is used now, clear the Dead flag
+ MachineInstr *Def = MRI->getUniqueVRegDef(NewReg);
+ assert(Def != nullptr && "CSEd register has no unique definition?");
+ Def->clearRegisterDeads(NewReg);
+ // Replace with NewReg and clear kill flags which may be wrong now.
+ MRI->replaceRegWith(OldReg, NewReg);
+ MRI->clearKillFlags(NewReg);
}
// Go through implicit defs of CSMI and MI, if a def is not dead at MI,
for (unsigned i = 0, e = ImplicitDefsToUpdate.size(); i != e; ++i)
CSMI->getOperand(ImplicitDefsToUpdate[i]).setIsDead(false);
+ // Go through implicit defs of CSMI and MI, and clear the kill flags on
+ // their uses in all the instructions between CSMI and MI.
+ // We might have made some of the kill flags redundant, consider:
+ // subs ... %NZCV<imp-def> <- CSMI
+ // csinc ... %NZCV<imp-use,kill> <- this kill flag isn't valid anymore
+ // subs ... %NZCV<imp-def> <- MI, to be eliminated
+ // csinc ... %NZCV<imp-use,kill>
+ // Since we eliminated MI, and reused a register imp-def'd by CSMI
+ // (here %NZCV), that register, if it was killed before MI, should have
+ // that kill flag removed, because it's lifetime was extended.
+ if (CSMI->getParent() == MI->getParent()) {
+ for (MachineBasicBlock::iterator II = CSMI, IE = MI; II != IE; ++II)
+ for (auto ImplicitDef : ImplicitDefs)
+ if (MachineOperand *MO = II->findRegisterUseOperand(
+ ImplicitDef, /*isKill=*/true, TRI))
+ MO->setIsKill(false);
+ } else {
+ // If the instructions aren't in the same BB, bail out and clear the
+ // kill flag on all uses of the imp-def'd register.
+ for (auto ImplicitDef : ImplicitDefs)
+ MRI->clearKillFlags(ImplicitDef);
+ }
+
if (CrossMBBPhysDef) {
// Add physical register defs now coming in from a predecessor to MBB
// livein list.
}
CSEPairs.clear();
ImplicitDefsToUpdate.clear();
+ ImplicitDefs.clear();
}
return Changed;
}
bool MachineCSE::runOnMachineFunction(MachineFunction &MF) {
- TII = MF.getTarget().getInstrInfo();
- TRI = MF.getTarget().getRegisterInfo();
+ if (skipOptnoneFunction(*MF.getFunction()))
+ return false;
+
+ TII = MF.getSubtarget().getInstrInfo();
+ TRI = MF.getSubtarget().getRegisterInfo();
MRI = &MF.getRegInfo();
AA = &getAnalysis<AliasAnalysis>();
DT = &getAnalysis<MachineDominatorTree>();
+ LookAheadLimit = TII->getMachineCSELookAheadLimit();
return PerformCSE(DT->getRootNode());
}