X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FCodeGen%2FCriticalAntiDepBreaker.cpp;h=c924ba30c8a11b54a4f1df39cc749f2493f1ec4b;hb=94d0a9154e9ec3df3398f11649c024ef8378b9c8;hp=cf8c71dab2ff95e8529a2028f787e0da13ed4df9;hpb=87f3dbc446181dc5b1c525bd28ca89760f63bc76;p=oota-llvm.git diff --git a/lib/CodeGen/CriticalAntiDepBreaker.cpp b/lib/CodeGen/CriticalAntiDepBreaker.cpp index cf8c71dab2f..c924ba30c8a 100644 --- a/lib/CodeGen/CriticalAntiDepBreaker.cpp +++ b/lib/CodeGen/CriticalAntiDepBreaker.cpp @@ -13,29 +13,27 @@ // //===----------------------------------------------------------------------===// -#define DEBUG_TYPE "post-RA-sched" #include "CriticalAntiDepBreaker.h" #include "llvm/CodeGen/MachineBasicBlock.h" #include "llvm/CodeGen/MachineFrameInfo.h" -#include "llvm/Target/TargetMachine.h" -#include "llvm/Target/TargetInstrInfo.h" -#include "llvm/Target/TargetRegisterInfo.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/raw_ostream.h" +#include "llvm/Target/TargetInstrInfo.h" +#include "llvm/Target/TargetRegisterInfo.h" +#include "llvm/Target/TargetSubtargetInfo.h" using namespace llvm; -CriticalAntiDepBreaker:: -CriticalAntiDepBreaker(MachineFunction& MFi, const RegisterClassInfo &RCI) : - AntiDepBreaker(), MF(MFi), - MRI(MF.getRegInfo()), - TII(MF.getTarget().getInstrInfo()), - TRI(MF.getTarget().getRegisterInfo()), - RegClassInfo(RCI), - Classes(TRI->getNumRegs(), static_cast(0)), - KillIndices(TRI->getNumRegs(), 0), - DefIndices(TRI->getNumRegs(), 0) {} +#define DEBUG_TYPE "post-RA-sched" + +CriticalAntiDepBreaker::CriticalAntiDepBreaker(MachineFunction &MFi, + const RegisterClassInfo &RCI) + : AntiDepBreaker(), MF(MFi), MRI(MF.getRegInfo()), + TII(MF.getSubtarget().getInstrInfo()), + TRI(MF.getSubtarget().getRegisterInfo()), RegClassInfo(RCI), + Classes(TRI->getNumRegs(), nullptr), KillIndices(TRI->getNumRegs(), 0), + DefIndices(TRI->getNumRegs(), 0), KeepRegs(TRI->getNumRegs(), false) {} CriticalAntiDepBreaker::~CriticalAntiDepBreaker() { } @@ -44,7 +42,7 @@ void CriticalAntiDepBreaker::StartBlock(MachineBasicBlock *BB) { const unsigned BBSize = BB->size(); for (unsigned i = 0, e = TRI->getNumRegs(); i != e; ++i) { // Clear out the register class data. - Classes[i] = static_cast(0); + Classes[i] = nullptr; // Initialize the indices to indicate that no registers are live. KillIndices[i] = ~0u; @@ -52,48 +50,19 @@ void CriticalAntiDepBreaker::StartBlock(MachineBasicBlock *BB) { } // Clear "do not change" set. - KeepRegs.clear(); - - bool IsReturnBlock = (BBSize != 0 && BB->back().isReturn()); - - // Determine the live-out physregs for this block. - if (IsReturnBlock) { - // In a return block, examine the function live-out regs. - for (MachineRegisterInfo::liveout_iterator I = MRI.liveout_begin(), - E = MRI.liveout_end(); I != E; ++I) { - unsigned Reg = *I; - Classes[Reg] = reinterpret_cast(-1); - KillIndices[Reg] = BBSize; - DefIndices[Reg] = ~0u; + KeepRegs.reset(); - // Repeat, for all aliases. - for (const uint16_t *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) { - unsigned AliasReg = *Alias; - Classes[AliasReg] = reinterpret_cast(-1); - KillIndices[AliasReg] = BBSize; - DefIndices[AliasReg] = ~0u; - } - } - } + bool IsReturnBlock = BB->isReturnBlock(); - // In a non-return block, examine the live-in regs of all successors. - // Note a return block can have successors if the return instruction is - // predicated. + // Examine the live-in regs of all successors. for (MachineBasicBlock::succ_iterator SI = BB->succ_begin(), SE = BB->succ_end(); SI != SE; ++SI) - for (MachineBasicBlock::livein_iterator I = (*SI)->livein_begin(), - E = (*SI)->livein_end(); I != E; ++I) { - unsigned Reg = *I; - Classes[Reg] = reinterpret_cast(-1); - KillIndices[Reg] = BBSize; - DefIndices[Reg] = ~0u; - - // Repeat, for all aliases. - for (const uint16_t *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) { - unsigned AliasReg = *Alias; - Classes[AliasReg] = reinterpret_cast(-1); - KillIndices[AliasReg] = BBSize; - DefIndices[AliasReg] = ~0u; + for (const auto &LI : (*SI)->liveins()) { + for (MCRegAliasIterator AI(LI.PhysReg, TRI, true); AI.isValid(); ++AI) { + unsigned Reg = *AI; + Classes[Reg] = reinterpret_cast(-1); + KillIndices[Reg] = BBSize; + DefIndices[Reg] = ~0u; } } @@ -101,32 +70,33 @@ void CriticalAntiDepBreaker::StartBlock(MachineBasicBlock *BB) { // all callee-saved registers. In non-return this is any // callee-saved register that is not saved in the prolog. const MachineFrameInfo *MFI = MF.getFrameInfo(); - BitVector Pristine = MFI->getPristineRegs(BB); - for (const uint16_t *I = TRI->getCalleeSavedRegs(&MF); *I; ++I) { - unsigned Reg = *I; - if (!IsReturnBlock && !Pristine.test(Reg)) continue; - Classes[Reg] = reinterpret_cast(-1); - KillIndices[Reg] = BBSize; - DefIndices[Reg] = ~0u; - - // Repeat, for all aliases. - for (const uint16_t *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) { - unsigned AliasReg = *Alias; - Classes[AliasReg] = reinterpret_cast(-1); - KillIndices[AliasReg] = BBSize; - DefIndices[AliasReg] = ~0u; + BitVector Pristine = MFI->getPristineRegs(MF); + for (const MCPhysReg *I = TRI->getCalleeSavedRegs(&MF); *I; ++I) { + if (!IsReturnBlock && !Pristine.test(*I)) continue; + for (MCRegAliasIterator AI(*I, TRI, true); AI.isValid(); ++AI) { + unsigned Reg = *AI; + Classes[Reg] = reinterpret_cast(-1); + KillIndices[Reg] = BBSize; + DefIndices[Reg] = ~0u; } } } void CriticalAntiDepBreaker::FinishBlock() { RegRefs.clear(); - KeepRegs.clear(); + KeepRegs.reset(); } void CriticalAntiDepBreaker::Observe(MachineInstr *MI, unsigned Count, unsigned InsertPosIndex) { - if (MI->isDebugValue()) + // Kill instructions can define registers but are really nops, and there might + // be a real definition earlier that needs to be paired with uses dominated by + // this kill. + + // FIXME: It may be possible to remove the isKill() restriction once PR18663 + // has been properly fixed. There can be value in processing kills as seen in + // the AggressiveAntiDepBreaker class. + if (MI->isDebugValue() || MI->isKill()) return; assert(Count < InsertPosIndex && "Instruction index out of expected range!"); @@ -157,7 +127,7 @@ void CriticalAntiDepBreaker::Observe(MachineInstr *MI, unsigned Count, /// CriticalPathStep - Return the next SUnit after SU on the bottom-up /// critical path. static const SDep *CriticalPathStep(const SUnit *SU) { - const SDep *Next = 0; + const SDep *Next = nullptr; unsigned NextDepth = 0; // Find the predecessor edge with the greatest depth. for (SUnit::const_pred_iterator P = SU->Preds.begin(), PE = SU->Preds.end(); @@ -178,8 +148,8 @@ static const SDep *CriticalPathStep(const SUnit *SU) { void CriticalAntiDepBreaker::PrescanInstruction(MachineInstr *MI) { // It's not safe to change register allocation for source operands of - // that have special allocation requirements. Also assume all registers - // used in a call must not be changed (ABI). + // instructions that have special allocation requirements. Also assume all + // registers used in a call must not be changed (ABI). // FIXME: The issue with predicated instruction is more complex. We are being // conservative here because the kill markers cannot be trusted after // if-conversion: @@ -204,10 +174,10 @@ void CriticalAntiDepBreaker::PrescanInstruction(MachineInstr *MI) { if (!MO.isReg()) continue; unsigned Reg = MO.getReg(); if (Reg == 0) continue; - const TargetRegisterClass *NewRC = 0; + const TargetRegisterClass *NewRC = nullptr; if (i < MI->getDesc().getNumOperands()) - NewRC = TII->getRegClass(MI->getDesc(), i, TRI); + NewRC = TII->getRegClass(MI->getDesc(), i, TRI, MF); // For now, only allow the register to be changed if its register // class is consistent across all uses. @@ -217,11 +187,11 @@ void CriticalAntiDepBreaker::PrescanInstruction(MachineInstr *MI) { Classes[Reg] = reinterpret_cast(-1); // Now check for aliases. - for (const uint16_t *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) { + for (MCRegAliasIterator AI(Reg, TRI, false); AI.isValid(); ++AI) { // If an alias of the reg is used during the live range, give up. // Note that this allows us to skip checking if AntiDepReg // overlaps with any of the aliases, among other things. - unsigned AliasReg = *Alias; + unsigned AliasReg = *AI; if (Classes[AliasReg]) { Classes[AliasReg] = reinterpret_cast(-1); Classes[Reg] = reinterpret_cast(-1); @@ -232,11 +202,33 @@ void CriticalAntiDepBreaker::PrescanInstruction(MachineInstr *MI) { if (Classes[Reg] != reinterpret_cast(-1)) RegRefs.insert(std::make_pair(Reg, &MO)); + // If this reg is tied and live (Classes[Reg] is set to -1), we can't change + // it or any of its sub or super regs. We need to use KeepRegs to mark the + // reg because not all uses of the same reg within an instruction are + // necessarily tagged as tied. + // Example: an x86 "xor %eax, %eax" will have one source operand tied to the + // def register but not the second (see PR20020 for details). + // FIXME: can this check be relaxed to account for undef uses + // of a register? In the above 'xor' example, the uses of %eax are undef, so + // earlier instructions could still replace %eax even though the 'xor' + // itself can't be changed. + if (MI->isRegTiedToUseOperand(i) && + Classes[Reg] == reinterpret_cast(-1)) { + for (MCSubRegIterator SubRegs(Reg, TRI, /*IncludeSelf=*/true); + SubRegs.isValid(); ++SubRegs) { + KeepRegs.set(*SubRegs); + } + for (MCSuperRegIterator SuperRegs(Reg, TRI); + SuperRegs.isValid(); ++SuperRegs) { + KeepRegs.set(*SuperRegs); + } + } + if (MO.isUse() && Special) { - if (KeepRegs.insert(Reg)) { - for (const uint16_t *Subreg = TRI->getSubRegisters(Reg); - *Subreg; ++Subreg) - KeepRegs.insert(*Subreg); + if (!KeepRegs.test(Reg)) { + for (MCSubRegIterator SubRegs(Reg, TRI, /*IncludeSelf=*/true); + SubRegs.isValid(); ++SubRegs) + KeepRegs.set(*SubRegs); } } } @@ -245,8 +237,9 @@ void CriticalAntiDepBreaker::PrescanInstruction(MachineInstr *MI) { void CriticalAntiDepBreaker::ScanInstruction(MachineInstr *MI, unsigned Count) { // Update liveness. - // Proceding upwards, registers that are defed but not used in this + // Proceeding upwards, registers that are defed but not used in this // instruction are now dead. + assert(!MI->isKill() && "Attempting to scan a kill instruction"); if (!TII->isPredicated(MI)) { // Predicated defs are modeled as read + write, i.e. similar to two @@ -259,8 +252,8 @@ void CriticalAntiDepBreaker::ScanInstruction(MachineInstr *MI, if (MO.clobbersPhysReg(i)) { DefIndices[i] = Count; KillIndices[i] = ~0u; - KeepRegs.erase(i); - Classes[i] = 0; + KeepRegs.reset(i); + Classes[i] = nullptr; RegRefs.erase(i); } @@ -268,33 +261,26 @@ void CriticalAntiDepBreaker::ScanInstruction(MachineInstr *MI, unsigned Reg = MO.getReg(); if (Reg == 0) continue; if (!MO.isDef()) continue; + + // If we've already marked this reg as unchangeable, carry on. + if (KeepRegs.test(Reg)) continue; + // Ignore two-addr defs. if (MI->isRegTiedToUseOperand(i)) continue; - DefIndices[Reg] = Count; - KillIndices[Reg] = ~0u; - assert(((KillIndices[Reg] == ~0u) != - (DefIndices[Reg] == ~0u)) && - "Kill and Def maps aren't consistent for Reg!"); - KeepRegs.erase(Reg); - Classes[Reg] = 0; - RegRefs.erase(Reg); - // Repeat, for all subregs. - for (const uint16_t *Subreg = TRI->getSubRegisters(Reg); - *Subreg; ++Subreg) { - unsigned SubregReg = *Subreg; + // For the reg itself and all subregs: update the def to current; + // reset the kill state, any restrictions, and references. + for (MCSubRegIterator SRI(Reg, TRI, true); SRI.isValid(); ++SRI) { + unsigned SubregReg = *SRI; DefIndices[SubregReg] = Count; KillIndices[SubregReg] = ~0u; - KeepRegs.erase(SubregReg); - Classes[SubregReg] = 0; + KeepRegs.reset(SubregReg); + Classes[SubregReg] = nullptr; RegRefs.erase(SubregReg); } // Conservatively mark super-registers as unusable. - for (const uint16_t *Super = TRI->getSuperRegisters(Reg); - *Super; ++Super) { - unsigned SuperReg = *Super; - Classes[SuperReg] = reinterpret_cast(-1); - } + for (MCSuperRegIterator SR(Reg, TRI); SR.isValid(); ++SR) + Classes[*SR] = reinterpret_cast(-1); } } for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { @@ -304,9 +290,9 @@ void CriticalAntiDepBreaker::ScanInstruction(MachineInstr *MI, if (Reg == 0) continue; if (!MO.isUse()) continue; - const TargetRegisterClass *NewRC = 0; + const TargetRegisterClass *NewRC = nullptr; if (i < MI->getDesc().getNumOperands()) - NewRC = TII->getRegClass(MI->getDesc(), i, TRI); + NewRC = TII->getRegClass(MI->getDesc(), i, TRI, MF); // For now, only allow the register to be changed if its register // class is consistent across all uses. @@ -318,16 +304,9 @@ void CriticalAntiDepBreaker::ScanInstruction(MachineInstr *MI, RegRefs.insert(std::make_pair(Reg, &MO)); // It wasn't previously live but now it is, this is a kill. - if (KillIndices[Reg] == ~0u) { - KillIndices[Reg] = Count; - DefIndices[Reg] = ~0u; - assert(((KillIndices[Reg] == ~0u) != - (DefIndices[Reg] == ~0u)) && - "Kill and Def maps aren't consistent for Reg!"); - } - // Repeat, for all aliases. - for (const uint16_t *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) { - unsigned AliasReg = *Alias; + // Repeat for all aliases. + for (MCRegAliasIterator AI(Reg, TRI, true); AI.isValid(); ++AI) { + unsigned AliasReg = *AI; if (KillIndices[AliasReg] == ~0u) { KillIndices[AliasReg] = Count; DefIndices[AliasReg] = ~0u; @@ -345,7 +324,7 @@ void CriticalAntiDepBreaker::ScanInstruction(MachineInstr *MI, // the two-address instruction also defines NewReg, as may happen with // pre/postincrement loads. In this case, both the use and def operands are in // RegRefs because the def is inserted by PrescanInstruction and not erased -// during ScanInstruction. So checking for an instructions with definitions of +// during ScanInstruction. So checking for an instruction with definitions of // both NewReg and AntiDepReg covers it. bool CriticalAntiDepBreaker::isNewRegClobberedByRefs(RegRefIter RegRefBegin, @@ -361,7 +340,7 @@ CriticalAntiDepBreaker::isNewRegClobberedByRefs(RegRefIter RegRefBegin, if (RefOper->isDef() && RefOper->isEarlyClobber()) return true; - // Handle cases in which this instructions defines NewReg. + // Handle cases in which this instruction defines NewReg. MachineInstr *MI = RefOper->getParent(); for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { const MachineOperand &CheckOper = MI->getOperand(i); @@ -379,11 +358,11 @@ CriticalAntiDepBreaker::isNewRegClobberedByRefs(RegRefIter RegRefBegin, return true; // Don't allow an instruction using AntiDepReg to be earlyclobbered by - // NewReg + // NewReg. if (CheckOper.isEarlyClobber()) return true; - // Don't allow inline asm to define NewReg at all. Who know what it's + // Don't allow inline asm to define NewReg at all. Who knows what it's // doing with it. if (MI->isInlineAsm()) return true; @@ -392,14 +371,15 @@ CriticalAntiDepBreaker::isNewRegClobberedByRefs(RegRefIter RegRefBegin, return false; } -unsigned -CriticalAntiDepBreaker::findSuitableFreeRegister(RegRefIter RegRefBegin, - RegRefIter RegRefEnd, - unsigned AntiDepReg, - unsigned LastNewReg, - const TargetRegisterClass *RC) +unsigned CriticalAntiDepBreaker:: +findSuitableFreeRegister(RegRefIter RegRefBegin, + RegRefIter RegRefEnd, + unsigned AntiDepReg, + unsigned LastNewReg, + const TargetRegisterClass *RC, + SmallVectorImpl &Forbid) { - ArrayRef Order = RegClassInfo.getOrder(RC); + ArrayRef Order = RegClassInfo.getOrder(RC); for (unsigned i = 0; i != Order.size(); ++i) { unsigned NewReg = Order[i]; // Don't replace a register with itself. @@ -422,6 +402,15 @@ CriticalAntiDepBreaker::findSuitableFreeRegister(RegRefIter RegRefBegin, Classes[NewReg] == reinterpret_cast(-1) || KillIndices[AntiDepReg] > DefIndices[NewReg]) continue; + // If NewReg overlaps any of the forbidden registers, we can't use it. + bool Forbidden = false; + for (SmallVectorImpl::iterator it = Forbid.begin(), + ite = Forbid.end(); it != ite; ++it) + if (TRI->regsOverlap(NewReg, *it)) { + Forbidden = true; + break; + } + if (Forbidden) continue; return NewReg; } @@ -446,7 +435,7 @@ BreakAntiDependencies(const std::vector& SUnits, DenseMap MISUnitMap; // Find the node at the bottom of the critical path. - const SUnit *Max = 0; + const SUnit *Max = nullptr; for (unsigned i = 0, e = SUnits.size(); i != e; ++i) { const SUnit *SU = &SUnits[i]; MISUnitMap[SU->getInstr()] = SU; @@ -520,10 +509,16 @@ BreakAntiDependencies(const std::vector& SUnits, // as we go to help determine which registers are available. unsigned Broken = 0; unsigned Count = InsertPosIndex - 1; - for (MachineBasicBlock::iterator I = End, E = Begin; - I != E; --Count) { + for (MachineBasicBlock::iterator I = End, E = Begin; I != E; --Count) { MachineInstr *MI = --I; - if (MI->isDebugValue()) + // Kill instructions can define registers but are really nops, and there + // might be a real definition earlier that needs to be paired with uses + // dominated by this kill. + + // FIXME: It may be possible to remove the isKill() restriction once PR18663 + // has been properly fixed. There can be value in processing kills as seen + // in the AggressiveAntiDepBreaker class. + if (MI->isDebugValue() || MI->isKill()) continue; // Check if this instruction has a dependence on the critical path that @@ -548,11 +543,11 @@ BreakAntiDependencies(const std::vector& SUnits, if (Edge->getKind() == SDep::Anti) { AntiDepReg = Edge->getReg(); assert(AntiDepReg != 0 && "Anti-dependence on reg0?"); - if (!RegClassInfo.isAllocatable(AntiDepReg)) + if (!MRI.isAllocatable(AntiDepReg)) // Don't break anti-dependencies on non-allocatable registers. AntiDepReg = 0; - else if (KeepRegs.count(AntiDepReg)) - // Don't break anti-dependencies if an use down below requires + else if (KeepRegs.test(AntiDepReg)) + // Don't break anti-dependencies if a use down below requires // this exact register. AntiDepReg = 0; else { @@ -578,24 +573,27 @@ BreakAntiDependencies(const std::vector& SUnits, CriticalPathMI = CriticalPathSU->getInstr(); } else { // We've reached the end of the critical path. - CriticalPathSU = 0; - CriticalPathMI = 0; + CriticalPathSU = nullptr; + CriticalPathMI = nullptr; } } PrescanInstruction(MI); + SmallVector ForbidRegs; + // If MI's defs have a special allocation requirement, don't allow // any def registers to be changed. Also assume all registers // defined in a call must not be changed (ABI). - if (MI->isCall() || MI->hasExtraDefRegAllocReq() || - TII->isPredicated(MI)) + if (MI->isCall() || MI->hasExtraDefRegAllocReq() || TII->isPredicated(MI)) // If this instruction's defs have special allocation requirement, don't // break this anti-dependency. AntiDepReg = 0; else if (AntiDepReg) { // If this instruction has a use of AntiDepReg, breaking it - // is invalid. + // is invalid. If the instruction defines other registers, + // save a list of them so that we don't pick a new register + // that overlaps any of them. for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { MachineOperand &MO = MI->getOperand(i); if (!MO.isReg()) continue; @@ -605,18 +603,21 @@ BreakAntiDependencies(const std::vector& SUnits, AntiDepReg = 0; break; } + if (MO.isDef() && Reg != AntiDepReg) + ForbidRegs.push_back(Reg); } } // Determine AntiDepReg's register class, if it is live and is // consistently used within a single class. - const TargetRegisterClass *RC = AntiDepReg != 0 ? Classes[AntiDepReg] : 0; - assert((AntiDepReg == 0 || RC != NULL) && + const TargetRegisterClass *RC = AntiDepReg != 0 ? Classes[AntiDepReg] + : nullptr; + assert((AntiDepReg == 0 || RC != nullptr) && "Register should be live if it's causing an anti-dependence!"); if (RC == reinterpret_cast(-1)) AntiDepReg = 0; - // Look for a suitable register to use to break the anti-depenence. + // Look for a suitable register to use to break the anti-dependence. // // TODO: Instead of picking the first free register, consider which might // be the best. @@ -627,7 +628,7 @@ BreakAntiDependencies(const std::vector& SUnits, if (unsigned NewReg = findSuitableFreeRegister(Range.first, Range.second, AntiDepReg, LastNewReg[AntiDepReg], - RC)) { + RC, ForbidRegs)) { DEBUG(dbgs() << "Breaking anti-dependence edge on " << TRI->getName(AntiDepReg) << " with " << RegRefs.count(AntiDepReg) << " references" @@ -659,7 +660,7 @@ BreakAntiDependencies(const std::vector& SUnits, (DefIndices[NewReg] == ~0u)) && "Kill and Def maps aren't consistent for NewReg!"); - Classes[AntiDepReg] = 0; + Classes[AntiDepReg] = nullptr; DefIndices[AntiDepReg] = KillIndices[AntiDepReg]; KillIndices[AntiDepReg] = ~0u; assert(((KillIndices[AntiDepReg] == ~0u) !=