#include "llvm/CodeGen/SlotIndexes.h"
#include "llvm/CodeGen/StackMaps.h"
#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCInst.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
cl::desc("Causes the backend to crash instead of generating a nop VSX copy"),
cl::Hidden);
+static cl::opt<bool>
+UseOldLatencyCalc("ppc-old-latency-calc", cl::Hidden,
+ cl::desc("Use the old (incorrect) instruction latency calculation"));
+
// Pin the vtable to this file.
void PPCInstrInfo::anchor() {}
PPCInstrInfo::PPCInstrInfo(PPCSubtarget &STI)
: PPCGenInstrInfo(PPC::ADJCALLSTACKDOWN, PPC::ADJCALLSTACKUP),
- Subtarget(STI), RI(STI) {}
+ Subtarget(STI), RI(STI.getTargetMachine()) {}
/// CreateTargetHazardRecognizer - Return the hazard recognizer to use for
/// this target when scheduling the DAG.
return new ScoreboardHazardRecognizer(II, DAG);
}
+unsigned PPCInstrInfo::getInstrLatency(const InstrItineraryData *ItinData,
+ const MachineInstr *MI,
+ unsigned *PredCost) const {
+ if (!ItinData || UseOldLatencyCalc)
+ return PPCGenInstrInfo::getInstrLatency(ItinData, MI, PredCost);
+
+ // The default implementation of getInstrLatency calls getStageLatency, but
+ // getStageLatency does not do the right thing for us. While we have
+ // itinerary, most cores are fully pipelined, and so the itineraries only
+ // express the first part of the pipeline, not every stage. Instead, we need
+ // to use the listed output operand cycle number (using operand 0 here, which
+ // is an output).
+
+ unsigned Latency = 1;
+ unsigned DefClass = MI->getDesc().getSchedClass();
+ for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+ const MachineOperand &MO = MI->getOperand(i);
+ if (!MO.isReg() || !MO.isDef() || MO.isImplicit())
+ continue;
+
+ int Cycle = ItinData->getOperandCycle(DefClass, i);
+ if (Cycle < 0)
+ continue;
+
+ Latency = std::max(Latency, (unsigned) Cycle);
+ }
+
+ return Latency;
+}
int PPCInstrInfo::getOperandLatency(const InstrItineraryData *ItinData,
const MachineInstr *DefMI, unsigned DefIdx,
int Latency = PPCGenInstrInfo::getOperandLatency(ItinData, DefMI, DefIdx,
UseMI, UseIdx);
+ if (!DefMI->getParent())
+ return Latency;
+
const MachineOperand &DefMO = DefMI->getOperand(DefIdx);
unsigned Reg = DefMO.getReg();
- const TargetRegisterInfo *TRI = &getRegisterInfo();
bool IsRegCR;
- if (TRI->isVirtualRegister(Reg)) {
+ if (TargetRegisterInfo::isVirtualRegister(Reg)) {
const MachineRegisterInfo *MRI =
&DefMI->getParent()->getParent()->getRegInfo();
IsRegCR = MRI->getRegClass(Reg)->hasSuperClassEq(&PPC::CRRCRegClass) ||
return Latency;
}
+// This function does not list all associative and commutative operations, but
+// only those worth feeding through the machine combiner in an attempt to
+// reduce the critical path. Mostly, this means floating-point operations,
+// because they have high latencies (compared to other operations, such and
+// and/or, which are also associative and commutative, but have low latencies).
+bool PPCInstrInfo::isAssociativeAndCommutative(const MachineInstr &Inst) const {
+ switch (Inst.getOpcode()) {
+ // FP Add:
+ case PPC::FADD:
+ case PPC::FADDS:
+ // FP Multiply:
+ case PPC::FMUL:
+ case PPC::FMULS:
+ // Altivec Add:
+ case PPC::VADDFP:
+ // VSX Add:
+ case PPC::XSADDDP:
+ case PPC::XVADDDP:
+ case PPC::XVADDSP:
+ case PPC::XSADDSP:
+ // VSX Multiply:
+ case PPC::XSMULDP:
+ case PPC::XVMULDP:
+ case PPC::XVMULSP:
+ case PPC::XSMULSP:
+ // QPX Add:
+ case PPC::QVFADD:
+ case PPC::QVFADDS:
+ case PPC::QVFADDSs:
+ // QPX Multiply:
+ case PPC::QVFMUL:
+ case PPC::QVFMULS:
+ case PPC::QVFMULSs:
+ return true;
+ default:
+ return false;
+ }
+}
+
+bool PPCInstrInfo::getMachineCombinerPatterns(
+ MachineInstr &Root,
+ SmallVectorImpl<MachineCombinerPattern> &Patterns) const {
+ // Using the machine combiner in this way is potentially expensive, so
+ // restrict to when aggressive optimizations are desired.
+ if (Subtarget.getTargetMachine().getOptLevel() != CodeGenOpt::Aggressive)
+ return false;
+
+ // FP reassociation is only legal when we don't need strict IEEE semantics.
+ if (!Root.getParent()->getParent()->getTarget().Options.UnsafeFPMath)
+ return false;
+
+ return TargetInstrInfo::getMachineCombinerPatterns(Root, Patterns);
+}
+
// Detect 32 -> 64-bit extensions where we may reuse the low sub-register.
bool PPCInstrInfo::isCoalescableExtInstr(const MachineInstr &MI,
unsigned &SrcReg, unsigned &DstReg,
case PPC::RESTORE_CRBIT:
case PPC::LVX:
case PPC::LXVD2X:
+ case PPC::QVLFDX:
+ case PPC::QVLFSXs:
+ case PPC::QVLFDXb:
case PPC::RESTORE_VRSAVE:
// Check for the operands added by addFrameReference (the immediate is the
// offset which defaults to 0).
case PPC::SPILL_CRBIT:
case PPC::STVX:
case PPC::STXVD2X:
+ case PPC::QVSTFDX:
+ case PPC::QVSTFSXs:
+ case PPC::QVSTFDXb:
case PPC::SPILL_VRSAVE:
// Check for the operands added by addFrameReference (the immediate is the
// offset which defaults to 0).
return 0;
}
-// commuteInstruction - We can commute rlwimi instructions, but only if the
-// rotate amt is zero. We also have to munge the immediates a bit.
-MachineInstr *
-PPCInstrInfo::commuteInstruction(MachineInstr *MI, bool NewMI) const {
+MachineInstr *PPCInstrInfo::commuteInstructionImpl(MachineInstr *MI,
+ bool NewMI,
+ unsigned OpIdx1,
+ unsigned OpIdx2) const {
MachineFunction &MF = *MI->getParent()->getParent();
// Normal instructions can be commuted the obvious way.
if (MI->getOpcode() != PPC::RLWIMI &&
MI->getOpcode() != PPC::RLWIMIo)
- return TargetInstrInfo::commuteInstruction(MI, NewMI);
+ return TargetInstrInfo::commuteInstructionImpl(MI, NewMI, OpIdx1, OpIdx2);
// Note that RLWIMI can be commuted as a 32-bit instruction, but not as a
// 64-bit instruction (so we don't handle PPC::RLWIMI8 here), because
// changing the relative order of the mask operands might change what happens
// Op0 = (Op2 & ~M) | (Op1 & M)
// Swap op1/op2
+ assert(((OpIdx1 == 1 && OpIdx2 == 2) || (OpIdx1 == 2 && OpIdx2 == 1)) &&
+ "Only the operands 1 and 2 can be swapped in RLSIMI/RLWIMIo.");
unsigned Reg0 = MI->getOperand(0).getReg();
unsigned Reg1 = MI->getOperand(1).getReg();
unsigned Reg2 = MI->getOperand(2).getReg();
unsigned MB = MI->getOperand(4).getImm();
unsigned ME = MI->getOperand(5).getImm();
+ // We can't commute a trivial mask (there is no way to represent an all-zero
+ // mask).
+ if (MB == 0 && ME == 31)
+ return nullptr;
+
if (NewMI) {
// Create a new instruction.
unsigned Reg0 = ChangeReg0 ? Reg2 : MI->getOperand(0).getReg();
if (AltOpc == -1)
return TargetInstrInfo::findCommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2);
- SrcOpIdx1 = 2;
- SrcOpIdx2 = 3;
- return true;
+ // The commutable operand indices are 2 and 3. Return them in SrcOpIdx1
+ // and SrcOpIdx2.
+ return fixCommutedOpIndices(SrcOpIdx1, SrcOpIdx2, 2, 3);
}
void PPCInstrInfo::insertNoop(MachineBasicBlock &MBB,
bool isPPC64 = Subtarget.isPPC64();
// If the block has no terminators, it just falls into the block after it.
- MachineBasicBlock::iterator I = MBB.end();
- if (I == MBB.begin())
+ MachineBasicBlock::iterator I = MBB.getLastNonDebugInstr();
+ if (I == MBB.end())
return false;
- --I;
- while (I->isDebugValue()) {
- if (I == MBB.begin())
- return false;
- --I;
- }
+
if (!isUnpredicatedTerminator(I))
return false;
}
unsigned PPCInstrInfo::RemoveBranch(MachineBasicBlock &MBB) const {
- MachineBasicBlock::iterator I = MBB.end();
- if (I == MBB.begin()) return 0;
- --I;
- while (I->isDebugValue()) {
- if (I == MBB.begin())
- return 0;
- --I;
- }
+ MachineBasicBlock::iterator I = MBB.getLastNonDebugInstr();
+ if (I == MBB.end())
+ return 0;
+
if (I->getOpcode() != PPC::B && I->getOpcode() != PPC::BCC &&
I->getOpcode() != PPC::BC && I->getOpcode() != PPC::BCn &&
I->getOpcode() != PPC::BDNZ8 && I->getOpcode() != PPC::BDNZ &&
unsigned
PPCInstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
MachineBasicBlock *FBB,
- const SmallVectorImpl<MachineOperand> &Cond,
+ ArrayRef<MachineOperand> Cond,
DebugLoc DL) const {
// Shouldn't be a fall through.
assert(TBB && "InsertBranch must not be told to insert a fallthrough");
// Select analysis.
bool PPCInstrInfo::canInsertSelect(const MachineBasicBlock &MBB,
- const SmallVectorImpl<MachineOperand> &Cond,
+ ArrayRef<MachineOperand> Cond,
unsigned TrueReg, unsigned FalseReg,
int &CondCycles, int &TrueCycles, int &FalseCycles) const {
if (!Subtarget.hasISEL())
void PPCInstrInfo::insertSelect(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI, DebugLoc dl,
- unsigned DestReg,
- const SmallVectorImpl<MachineOperand> &Cond,
+ unsigned DestReg, ArrayRef<MachineOperand> Cond,
unsigned TrueReg, unsigned FalseReg) const {
assert(Cond.size() == 2 &&
"PPC branch conditions have two components!");
.addReg(Cond[1].getReg(), 0, SubIdx);
}
+static unsigned getCRBitValue(unsigned CRBit) {
+ unsigned Ret = 4;
+ if (CRBit == PPC::CR0LT || CRBit == PPC::CR1LT ||
+ CRBit == PPC::CR2LT || CRBit == PPC::CR3LT ||
+ CRBit == PPC::CR4LT || CRBit == PPC::CR5LT ||
+ CRBit == PPC::CR6LT || CRBit == PPC::CR7LT)
+ Ret = 3;
+ if (CRBit == PPC::CR0GT || CRBit == PPC::CR1GT ||
+ CRBit == PPC::CR2GT || CRBit == PPC::CR3GT ||
+ CRBit == PPC::CR4GT || CRBit == PPC::CR5GT ||
+ CRBit == PPC::CR6GT || CRBit == PPC::CR7GT)
+ Ret = 2;
+ if (CRBit == PPC::CR0EQ || CRBit == PPC::CR1EQ ||
+ CRBit == PPC::CR2EQ || CRBit == PPC::CR3EQ ||
+ CRBit == PPC::CR4EQ || CRBit == PPC::CR5EQ ||
+ CRBit == PPC::CR6EQ || CRBit == PPC::CR7EQ)
+ Ret = 1;
+ if (CRBit == PPC::CR0UN || CRBit == PPC::CR1UN ||
+ CRBit == PPC::CR2UN || CRBit == PPC::CR3UN ||
+ CRBit == PPC::CR4UN || CRBit == PPC::CR5UN ||
+ CRBit == PPC::CR6UN || CRBit == PPC::CR7UN)
+ Ret = 0;
+
+ assert(Ret != 4 && "Invalid CR bit register");
+ return Ret;
+}
+
void PPCInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
MachineBasicBlock::iterator I, DebugLoc DL,
unsigned DestReg, unsigned SrcReg,
SrcReg = SuperReg;
}
+ // Different class register copy
+ if (PPC::CRBITRCRegClass.contains(SrcReg) &&
+ PPC::GPRCRegClass.contains(DestReg)) {
+ unsigned CRReg = getCRFromCRBit(SrcReg);
+ BuildMI(MBB, I, DL, get(PPC::MFOCRF), DestReg)
+ .addReg(CRReg), getKillRegState(KillSrc);
+ // Rotate the CR bit in the CR fields to be the least significant bit and
+ // then mask with 0x1 (MB = ME = 31).
+ BuildMI(MBB, I, DL, get(PPC::RLWINM), DestReg)
+ .addReg(DestReg, RegState::Kill)
+ .addImm(TRI->getEncodingValue(CRReg) * 4 + (4 - getCRBitValue(SrcReg)))
+ .addImm(31)
+ .addImm(31);
+ return;
+ } else if (PPC::CRRCRegClass.contains(SrcReg) &&
+ PPC::G8RCRegClass.contains(DestReg)) {
+ BuildMI(MBB, I, DL, get(PPC::MFOCRF8), DestReg)
+ .addReg(SrcReg), getKillRegState(KillSrc);
+ return;
+ } else if (PPC::CRRCRegClass.contains(SrcReg) &&
+ PPC::GPRCRegClass.contains(DestReg)) {
+ BuildMI(MBB, I, DL, get(PPC::MFOCRF), DestReg)
+ .addReg(SrcReg), getKillRegState(KillSrc);
+ return;
+ }
+
unsigned Opc;
if (PPC::GPRCRegClass.contains(DestReg, SrcReg))
Opc = PPC::OR;
// copies are generated, they are close enough to some use that the
// lower-latency form is preferable.
Opc = PPC::XXLOR;
- else if (PPC::VSFRCRegClass.contains(DestReg, SrcReg))
+ else if (PPC::VSFRCRegClass.contains(DestReg, SrcReg) ||
+ PPC::VSSRCRegClass.contains(DestReg, SrcReg))
Opc = PPC::XXLORf;
+ else if (PPC::QFRCRegClass.contains(DestReg, SrcReg))
+ Opc = PPC::QVFMR;
+ else if (PPC::QSRCRegClass.contains(DestReg, SrcReg))
+ Opc = PPC::QVFMRs;
+ else if (PPC::QBRCRegClass.contains(DestReg, SrcReg))
+ Opc = PPC::QVFMRb;
else if (PPC::CRBITRCRegClass.contains(DestReg, SrcReg))
Opc = PPC::CROR;
else
getKillRegState(isKill)),
FrameIdx));
NonRI = true;
+ } else if (PPC::VSSRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STXSSPX))
+ .addReg(SrcReg,
+ getKillRegState(isKill)),
+ FrameIdx));
+ NonRI = true;
} else if (PPC::VRSAVERCRegClass.hasSubClassEq(RC)) {
assert(Subtarget.isDarwin() &&
"VRSAVE only needs spill/restore on Darwin");
getKillRegState(isKill)),
FrameIdx));
SpillsVRS = true;
+ } else if (PPC::QFRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::QVSTFDX))
+ .addReg(SrcReg,
+ getKillRegState(isKill)),
+ FrameIdx));
+ NonRI = true;
+ } else if (PPC::QSRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::QVSTFSXs))
+ .addReg(SrcReg,
+ getKillRegState(isKill)),
+ FrameIdx));
+ NonRI = true;
+ } else if (PPC::QBRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::QVSTFDXb))
+ .addReg(SrcReg,
+ getKillRegState(isKill)),
+ FrameIdx));
+ NonRI = true;
} else {
llvm_unreachable("Unknown regclass!");
}
MBB.insert(MI, NewMIs[i]);
const MachineFrameInfo &MFI = *MF.getFrameInfo();
- MachineMemOperand *MMO =
- MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(FrameIdx),
- MachineMemOperand::MOStore,
- MFI.getObjectSize(FrameIdx),
- MFI.getObjectAlignment(FrameIdx));
+ MachineMemOperand *MMO = MF.getMachineMemOperand(
+ MachinePointerInfo::getFixedStack(MF, FrameIdx),
+ MachineMemOperand::MOStore, MFI.getObjectSize(FrameIdx),
+ MFI.getObjectAlignment(FrameIdx));
NewMIs.back()->addMemOperand(MF, MMO);
}
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LXSDX), DestReg),
FrameIdx));
NonRI = true;
+ } else if (PPC::VSSRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LXSSPX), DestReg),
+ FrameIdx));
+ NonRI = true;
} else if (PPC::VRSAVERCRegClass.hasSubClassEq(RC)) {
assert(Subtarget.isDarwin() &&
"VRSAVE only needs spill/restore on Darwin");
DestReg),
FrameIdx));
SpillsVRS = true;
+ } else if (PPC::QFRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::QVLFDX), DestReg),
+ FrameIdx));
+ NonRI = true;
+ } else if (PPC::QSRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::QVLFSXs), DestReg),
+ FrameIdx));
+ NonRI = true;
+ } else if (PPC::QBRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::QVLFDXb), DestReg),
+ FrameIdx));
+ NonRI = true;
} else {
llvm_unreachable("Unknown regclass!");
}
MBB.insert(MI, NewMIs[i]);
const MachineFrameInfo &MFI = *MF.getFrameInfo();
- MachineMemOperand *MMO =
- MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(FrameIdx),
- MachineMemOperand::MOLoad,
- MFI.getObjectSize(FrameIdx),
- MFI.getObjectAlignment(FrameIdx));
+ MachineMemOperand *MMO = MF.getMachineMemOperand(
+ MachinePointerInfo::getFixedStack(MF, FrameIdx),
+ MachineMemOperand::MOLoad, MFI.getObjectSize(FrameIdx),
+ MFI.getObjectAlignment(FrameIdx));
NewMIs.back()->addMemOperand(MF, MMO);
}
unsigned NumT, unsigned ExtraT,
MachineBasicBlock &FMBB,
unsigned NumF, unsigned ExtraF,
- const BranchProbability &Probability) const {
+ BranchProbability Probability) const {
return !(MBBDefinesCTR(TMBB) && MBBDefinesCTR(FMBB));
}
return !isPredicated(MI);
}
-bool PPCInstrInfo::PredicateInstruction(
- MachineInstr *MI,
- const SmallVectorImpl<MachineOperand> &Pred) const {
+bool PPCInstrInfo::PredicateInstruction(MachineInstr *MI,
+ ArrayRef<MachineOperand> Pred) const {
unsigned OpC = MI->getOpcode();
if (OpC == PPC::BLR || OpC == PPC::BLR8) {
if (Pred[1].getReg() == PPC::CTR8 || Pred[1].getReg() == PPC::CTR) {
return false;
}
-bool PPCInstrInfo::SubsumesPredicate(
- const SmallVectorImpl<MachineOperand> &Pred1,
- const SmallVectorImpl<MachineOperand> &Pred2) const {
+bool PPCInstrInfo::SubsumesPredicate(ArrayRef<MachineOperand> Pred1,
+ ArrayRef<MachineOperand> Pred2) const {
assert(Pred1.size() == 2 && "Invalid PPC first predicate");
assert(Pred2.size() == 2 && "Invalid PPC second predicate");
MI->setDesc(NewDesc);
if (NewDesc.ImplicitDefs)
- for (const uint16_t *ImpDefs = NewDesc.getImplicitDefs();
+ for (const MCPhysReg *ImpDefs = NewDesc.getImplicitDefs();
*ImpDefs; ++ImpDefs)
if (!MI->definesRegister(*ImpDefs))
MI->addOperand(*MI->getParent()->getParent(),
MachineOperand::CreateReg(*ImpDefs, true, true));
if (NewDesc.ImplicitUses)
- for (const uint16_t *ImpUses = NewDesc.getImplicitUses();
+ for (const MCPhysReg *ImpUses = NewDesc.getImplicitUses();
*ImpUses; ++ImpUses)
if (!MI->readsRegister(*ImpUses))
MI->addOperand(*MI->getParent()->getParent(),
}
}
+std::pair<unsigned, unsigned>
+PPCInstrInfo::decomposeMachineOperandsTargetFlags(unsigned TF) const {
+ const unsigned Mask = PPCII::MO_ACCESS_MASK;
+ return std::make_pair(TF & Mask, TF & ~Mask);
+}
+
+ArrayRef<std::pair<unsigned, const char *>>
+PPCInstrInfo::getSerializableDirectMachineOperandTargetFlags() const {
+ using namespace PPCII;
+ static const std::pair<unsigned, const char *> TargetFlags[] = {
+ {MO_LO, "ppc-lo"},
+ {MO_HA, "ppc-ha"},
+ {MO_TPREL_LO, "ppc-tprel-lo"},
+ {MO_TPREL_HA, "ppc-tprel-ha"},
+ {MO_DTPREL_LO, "ppc-dtprel-lo"},
+ {MO_TLSLD_LO, "ppc-tlsld-lo"},
+ {MO_TOC_LO, "ppc-toc-lo"},
+ {MO_TLS, "ppc-tls"}};
+ return makeArrayRef(TargetFlags);
+}
+
+ArrayRef<std::pair<unsigned, const char *>>
+PPCInstrInfo::getSerializableBitmaskMachineOperandTargetFlags() const {
+ using namespace PPCII;
+ static const std::pair<unsigned, const char *> TargetFlags[] = {
+ {MO_PLT_OR_STUB, "ppc-plt-or-stub"},
+ {MO_PIC_FLAG, "ppc-pic"},
+ {MO_NLP_FLAG, "ppc-nlp"},
+ {MO_NLP_HIDDEN_FLAG, "ppc-nlp-hidden"}};
+ return makeArrayRef(TargetFlags);
+}
+