#include "ARM.h"
#include "ARMAddressingModes.h"
#include "ARMConstantPoolValue.h"
+#include "ARMHazardRecognizer.h"
#include "ARMMachineFunctionInfo.h"
#include "ARMRegisterInfo.h"
#include "ARMGenInstrInfo.inc"
#include "llvm/Constants.h"
#include "llvm/Function.h"
#include "llvm/GlobalValue.h"
-#include "llvm/ADT/STLExtras.h"
#include "llvm/CodeGen/LiveVariables.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
+#include "llvm/ADT/STLExtras.h"
using namespace llvm;
static cl::opt<bool>
EnableARM3Addr("enable-arm-3-addr-conv", cl::Hidden,
cl::desc("Enable ARM 2-addr to 3-addr conv"));
-static cl::opt<bool>
-OldARMIfCvt("old-arm-ifcvt", cl::Hidden,
- cl::desc("Use old-style ARM if-conversion heuristics"));
+
+/// ARM_MLxEntry - Record information about MLA / MLS instructions.
+struct ARM_MLxEntry {
+ unsigned MLxOpc; // MLA / MLS opcode
+ unsigned MulOpc; // Expanded multiplication opcode
+ unsigned AddSubOpc; // Expanded add / sub opcode
+ bool NegAcc; // True if the acc is negated before the add / sub.
+ bool HasLane; // True if instruction has an extra "lane" operand.
+};
+
+static const ARM_MLxEntry ARM_MLxTable[] = {
+ // MLxOpc, MulOpc, AddSubOpc, NegAcc, HasLane
+ // fp scalar ops
+ { ARM::VMLAS, ARM::VMULS, ARM::VADDS, false, false },
+ { ARM::VMLSS, ARM::VMULS, ARM::VSUBS, false, false },
+ { ARM::VMLAD, ARM::VMULD, ARM::VADDD, false, false },
+ { ARM::VMLSD, ARM::VMULD, ARM::VSUBD, false, false },
+ { ARM::VMLAfd_sfp, ARM::VMULfd_sfp, ARM::VADDfd_sfp, false, false },
+ { ARM::VMLSfd_sfp, ARM::VMULfd_sfp, ARM::VSUBfd_sfp, false, false },
+ { ARM::VNMLAS, ARM::VNMULS, ARM::VSUBS, true, false },
+ { ARM::VNMLSS, ARM::VMULS, ARM::VSUBS, true, false },
+ { ARM::VNMLAD, ARM::VNMULD, ARM::VSUBD, true, false },
+ { ARM::VNMLSD, ARM::VMULD, ARM::VSUBD, true, false },
+
+ // fp SIMD ops
+ { ARM::VMLAfd, ARM::VMULfd, ARM::VADDfd, false, false },
+ { ARM::VMLSfd, ARM::VMULfd, ARM::VSUBfd, false, false },
+ { ARM::VMLAfq, ARM::VMULfq, ARM::VADDfq, false, false },
+ { ARM::VMLSfq, ARM::VMULfq, ARM::VSUBfq, false, false },
+ { ARM::VMLAslfd, ARM::VMULslfd, ARM::VADDfd, false, true },
+ { ARM::VMLSslfd, ARM::VMULslfd, ARM::VSUBfd, false, true },
+ { ARM::VMLAslfq, ARM::VMULslfq, ARM::VADDfq, false, true },
+ { ARM::VMLSslfq, ARM::VMULslfq, ARM::VSUBfq, false, true },
+};
ARMBaseInstrInfo::ARMBaseInstrInfo(const ARMSubtarget& STI)
: TargetInstrInfoImpl(ARMInsts, array_lengthof(ARMInsts)),
Subtarget(STI) {
+ for (unsigned i = 0, e = array_lengthof(ARM_MLxTable); i != e; ++i) {
+ if (!MLxEntryMap.insert(std::make_pair(ARM_MLxTable[i].MLxOpc, i)).second)
+ assert(false && "Duplicated entries?");
+ MLxHazardOpcodes.insert(ARM_MLxTable[i].AddSubOpc);
+ MLxHazardOpcodes.insert(ARM_MLxTable[i].MulOpc);
+ }
+}
+
+ScheduleHazardRecognizer *ARMBaseInstrInfo::
+CreateTargetPostRAHazardRecognizer(const InstrItineraryData *II) const {
+ if (Subtarget.isThumb2() || Subtarget.hasVFP2())
+ return (ScheduleHazardRecognizer *)
+ new ARMHazardRecognizer(II, *this, getRegisterInfo(), Subtarget);
+ return TargetInstrInfoImpl::CreateTargetPostRAHazardRecognizer(II);
}
MachineInstr *
return NewMIs[0];
}
-bool
-ARMBaseInstrInfo::spillCalleeSavedRegisters(MachineBasicBlock &MBB,
- MachineBasicBlock::iterator MI,
- const std::vector<CalleeSavedInfo> &CSI,
- const TargetRegisterInfo *TRI) const {
- if (CSI.empty())
- return false;
-
- DebugLoc DL;
- if (MI != MBB.end()) DL = MI->getDebugLoc();
-
- for (unsigned i = 0, e = CSI.size(); i != e; ++i) {
- unsigned Reg = CSI[i].getReg();
- bool isKill = true;
-
- // Add the callee-saved register as live-in unless it's LR and
- // @llvm.returnaddress is called. If LR is returned for @llvm.returnaddress
- // then it's already added to the function and entry block live-in sets.
- if (Reg == ARM::LR) {
- MachineFunction &MF = *MBB.getParent();
- if (MF.getFrameInfo()->isReturnAddressTaken() &&
- MF.getRegInfo().isLiveIn(Reg))
- isKill = false;
- }
-
- if (isKill)
- MBB.addLiveIn(Reg);
-
- // Insert the spill to the stack frame. The register is killed at the spill
- //
- const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
- storeRegToStackSlot(MBB, MI, Reg, isKill,
- CSI[i].getFrameIdx(), RC, TRI);
- }
- return true;
-}
-
// Branch analysis.
bool
ARMBaseInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,MachineBasicBlock *&TBB,
case ARM::BR_JTadd:
case ARM::tBR_JTr:
case ARM::t2BR_JT:
- case ARM::t2TBB:
- case ARM::t2TBH: {
+ case ARM::t2TBB_JT:
+ case ARM::t2TBH_JT: {
// These are jumptable branches, i.e. a branch followed by an inlined
// jumptable. The size is 4 + 4 * number of entries. For TBB, each
// entry is one byte; TBH two byte each.
- unsigned EntrySize = (Opc == ARM::t2TBB)
- ? 1 : ((Opc == ARM::t2TBH) ? 2 : 4);
+ unsigned EntrySize = (Opc == ARM::t2TBB_JT)
+ ? 1 : ((Opc == ARM::t2TBH_JT) ? 2 : 4);
unsigned NumOps = TID.getNumOperands();
MachineOperand JTOP =
MI->getOperand(NumOps - (TID.isPredicable() ? 3 : 2));
// alignment issue.
unsigned InstSize = (Opc == ARM::tBR_JTr || Opc == ARM::t2BR_JT) ? 2 : 4;
unsigned NumEntries = getNumJTEntries(JT, JTI);
- if (Opc == ARM::t2TBB && (NumEntries & 1))
+ if (Opc == ARM::t2TBB_JT && (NumEntries & 1))
// Make sure the instruction that follows TBB is 2-byte aligned.
// FIXME: Constant island pass should insert an "ALIGN" instruction
// instead.
.addReg(SrcReg, getKillRegState(isKill))
.addMemOperand(MMO));
} else {
- AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMQ))
+ AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMQIA))
.addReg(SrcReg, getKillRegState(isKill))
.addFrameIndex(FI)
- .addImm(ARM_AM::getAM4ModeImm(ARM_AM::ia))
.addMemOperand(MMO));
}
break;
.addMemOperand(MMO));
} else {
MachineInstrBuilder MIB =
- AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMD))
- .addFrameIndex(FI)
- .addImm(ARM_AM::getAM4ModeImm(ARM_AM::ia)))
+ AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMDIA))
+ .addFrameIndex(FI))
.addMemOperand(MMO);
MIB = AddDReg(MIB, SrcReg, ARM::dsub_0, getKillRegState(isKill), TRI);
MIB = AddDReg(MIB, SrcReg, ARM::dsub_1, 0, TRI);
break;
case ARM::QQQQPRRegClassID: {
MachineInstrBuilder MIB =
- AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMD))
- .addFrameIndex(FI)
- .addImm(ARM_AM::getAM4ModeImm(ARM_AM::ia)))
+ AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VSTMDIA))
+ .addFrameIndex(FI))
.addMemOperand(MMO);
MIB = AddDReg(MIB, SrcReg, ARM::dsub_0, getKillRegState(isKill), TRI);
MIB = AddDReg(MIB, SrcReg, ARM::dsub_1, 0, TRI);
return MI->getOperand(2).getReg();
}
break;
- case ARM::VSTMQ:
+ case ARM::VSTMQIA:
if (MI->getOperand(1).isFI() &&
- MI->getOperand(2).isImm() &&
- MI->getOperand(2).getImm() == ARM_AM::getAM4ModeImm(ARM_AM::ia) &&
MI->getOperand(0).getSubReg() == 0) {
FrameIndex = MI->getOperand(1).getIndex();
return MI->getOperand(0).getReg();
.addFrameIndex(FI).addImm(16)
.addMemOperand(MMO));
} else {
- AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDMQ), DestReg)
+ AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDMQIA), DestReg)
.addFrameIndex(FI)
- .addImm(ARM_AM::getAM4ModeImm(ARM_AM::ia))
.addMemOperand(MMO));
}
break;
.addMemOperand(MMO));
} else {
MachineInstrBuilder MIB =
- AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDMD))
- .addFrameIndex(FI)
- .addImm(ARM_AM::getAM4ModeImm(ARM_AM::ia)))
+ AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDMDIA))
+ .addFrameIndex(FI))
.addMemOperand(MMO);
MIB = AddDReg(MIB, DestReg, ARM::dsub_0, RegState::Define, TRI);
MIB = AddDReg(MIB, DestReg, ARM::dsub_1, RegState::Define, TRI);
break;
case ARM::QQQQPRRegClassID: {
MachineInstrBuilder MIB =
- AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDMD))
- .addFrameIndex(FI)
- .addImm(ARM_AM::getAM4ModeImm(ARM_AM::ia)))
+ AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::VLDMDIA))
+ .addFrameIndex(FI))
.addMemOperand(MMO);
MIB = AddDReg(MIB, DestReg, ARM::dsub_0, RegState::Define, TRI);
MIB = AddDReg(MIB, DestReg, ARM::dsub_1, RegState::Define, TRI);
return MI->getOperand(0).getReg();
}
break;
- case ARM::VLDMQ:
+ case ARM::VLDMQIA:
if (MI->getOperand(1).isFI() &&
- MI->getOperand(2).isImm() &&
- MI->getOperand(2).getImm() == ARM_AM::getAM4ModeImm(ARM_AM::ia) &&
MI->getOperand(0).getSubReg() == 0) {
FrameIndex = MI->getOperand(1).getIndex();
return MI->getOperand(0).getReg();
}
bool ARMBaseInstrInfo::isProfitableToIfCvt(MachineBasicBlock &MBB,
- unsigned NumInstrs,
+ unsigned NumCyles,
+ unsigned ExtraPredCycles,
float Probability,
float Confidence) const {
- if (!NumInstrs)
+ if (!NumCyles)
return false;
- // Use old-style heuristics
- if (OldARMIfCvt) {
- if (Subtarget.getCPUString() == "generic")
- // Generic (and overly aggressive) if-conversion limits for testing.
- return NumInstrs <= 10;
- if (Subtarget.hasV7Ops())
- return NumInstrs <= 3;
- return NumInstrs <= 2;
- }
-
// Attempt to estimate the relative costs of predication versus branching.
- float UnpredCost = Probability * NumInstrs;
+ float UnpredCost = Probability * NumCyles;
UnpredCost += 1.0; // The branch itself
UnpredCost += (1.0 - Confidence) * Subtarget.getMispredictionPenalty();
- float PredCost = NumInstrs;
-
- return PredCost < UnpredCost;
-
+ return (float)(NumCyles + ExtraPredCycles) < UnpredCost;
}
bool ARMBaseInstrInfo::
-isProfitableToIfCvt(MachineBasicBlock &TMBB, unsigned NumT,
- MachineBasicBlock &FMBB, unsigned NumF,
+isProfitableToIfCvt(MachineBasicBlock &TMBB,
+ unsigned TCycles, unsigned TExtra,
+ MachineBasicBlock &FMBB,
+ unsigned FCycles, unsigned FExtra,
float Probability, float Confidence) const {
- // Use old-style if-conversion heuristics
- if (OldARMIfCvt) {
- return NumT && NumF && NumT <= 2 && NumF <= 2;
- }
-
- if (!NumT || !NumF)
+ if (!TCycles || !FCycles)
return false;
// Attempt to estimate the relative costs of predication versus branching.
- float UnpredCost = Probability * NumT + (1.0 - Probability) * NumF;
+ float UnpredCost = Probability * TCycles + (1.0 - Probability) * FCycles;
UnpredCost += 1.0; // The branch itself
UnpredCost += (1.0 - Confidence) * Subtarget.getMispredictionPenalty();
- float PredCost = NumT + NumF;
-
- return PredCost < UnpredCost;
+ return (float)(TCycles + FCycles + TExtra + FExtra) < UnpredCost;
}
/// getInstrPredicate - If instruction is predicated, returns its predicate
switch (MI->getOpcode()) {
default: break;
case ARM::CMPri:
- case ARM::CMPzri:
case ARM::t2CMPri:
- case ARM::t2CMPzri:
SrcReg = MI->getOperand(0).getReg();
CmpMask = ~0;
CmpValue = MI->getOperand(1).getImm();
}
/// OptimizeCompareInstr - Convert the instruction supplying the argument to the
-/// comparison into one that sets the zero bit in the flags register. Update the
-/// iterator *only* if a transformation took place.
+/// comparison into one that sets the zero bit in the flags register.
bool ARMBaseInstrInfo::
OptimizeCompareInstr(MachineInstr *CmpInstr, unsigned SrcReg, int CmpMask,
- int CmpValue, const MachineRegisterInfo *MRI,
- MachineBasicBlock::iterator &MII) const {
+ int CmpValue, const MachineRegisterInfo *MRI) const {
if (CmpValue != 0)
return false;
for (unsigned IO = 0, EO = Instr.getNumOperands(); IO != EO; ++IO) {
const MachineOperand &MO = Instr.getOperand(IO);
- if (!MO.isReg() || !MO.isDef()) continue;
+ if (!MO.isReg()) continue;
- // This instruction modifies CPSR before the one we want to change. We
- // can't do this transformation.
+ // This instruction modifies or uses CPSR after the one we want to
+ // change. We can't do this transformation.
if (MO.getReg() == ARM::CPSR)
return false;
}
case ARM::SUBri:
case ARM::t2ADDri:
case ARM::t2SUBri:
- MI->RemoveOperand(5);
- MachineInstrBuilder(MI)
- .addReg(ARM::CPSR, RegState::Define | RegState::Implicit);
- MII = llvm::next(MachineBasicBlock::iterator(CmpInstr));
+ // Toggle the optional operand to CPSR.
+ MI->getOperand(5).setReg(ARM::CPSR);
+ MI->getOperand(5).setIsDef(true);
CmpInstr->eraseFromParent();
return true;
}
return false;
}
+bool ARMBaseInstrInfo::FoldImmediate(MachineInstr *UseMI,
+ MachineInstr *DefMI, unsigned Reg,
+ MachineRegisterInfo *MRI) const {
+ // Fold large immediates into add, sub, or, xor.
+ unsigned DefOpc = DefMI->getOpcode();
+ if (DefOpc != ARM::t2MOVi32imm && DefOpc != ARM::MOVi32imm)
+ return false;
+ if (!DefMI->getOperand(1).isImm())
+ // Could be t2MOVi32imm <ga:xx>
+ return false;
+
+ if (!MRI->hasOneNonDBGUse(Reg))
+ return false;
+
+ unsigned UseOpc = UseMI->getOpcode();
+ unsigned NewUseOpc = 0;
+ uint32_t ImmVal = (uint32_t)DefMI->getOperand(1).getImm();
+ uint32_t SOImmValV1 = 0, SOImmValV2 = 0;
+ bool Commute = false;
+ switch (UseOpc) {
+ default: return false;
+ case ARM::SUBrr:
+ case ARM::ADDrr:
+ case ARM::ORRrr:
+ case ARM::EORrr:
+ case ARM::t2SUBrr:
+ case ARM::t2ADDrr:
+ case ARM::t2ORRrr:
+ case ARM::t2EORrr: {
+ Commute = UseMI->getOperand(2).getReg() != Reg;
+ switch (UseOpc) {
+ default: break;
+ case ARM::SUBrr: {
+ if (Commute)
+ return false;
+ ImmVal = -ImmVal;
+ NewUseOpc = ARM::SUBri;
+ // Fallthrough
+ }
+ case ARM::ADDrr:
+ case ARM::ORRrr:
+ case ARM::EORrr: {
+ if (!ARM_AM::isSOImmTwoPartVal(ImmVal))
+ return false;
+ SOImmValV1 = (uint32_t)ARM_AM::getSOImmTwoPartFirst(ImmVal);
+ SOImmValV2 = (uint32_t)ARM_AM::getSOImmTwoPartSecond(ImmVal);
+ switch (UseOpc) {
+ default: break;
+ case ARM::ADDrr: NewUseOpc = ARM::ADDri; break;
+ case ARM::ORRrr: NewUseOpc = ARM::ORRri; break;
+ case ARM::EORrr: NewUseOpc = ARM::EORri; break;
+ }
+ break;
+ }
+ case ARM::t2SUBrr: {
+ if (Commute)
+ return false;
+ ImmVal = -ImmVal;
+ NewUseOpc = ARM::t2SUBri;
+ // Fallthrough
+ }
+ case ARM::t2ADDrr:
+ case ARM::t2ORRrr:
+ case ARM::t2EORrr: {
+ if (!ARM_AM::isT2SOImmTwoPartVal(ImmVal))
+ return false;
+ SOImmValV1 = (uint32_t)ARM_AM::getT2SOImmTwoPartFirst(ImmVal);
+ SOImmValV2 = (uint32_t)ARM_AM::getT2SOImmTwoPartSecond(ImmVal);
+ switch (UseOpc) {
+ default: break;
+ case ARM::t2ADDrr: NewUseOpc = ARM::t2ADDri; break;
+ case ARM::t2ORRrr: NewUseOpc = ARM::t2ORRri; break;
+ case ARM::t2EORrr: NewUseOpc = ARM::t2EORri; break;
+ }
+ break;
+ }
+ }
+ }
+ }
+
+ unsigned OpIdx = Commute ? 2 : 1;
+ unsigned Reg1 = UseMI->getOperand(OpIdx).getReg();
+ bool isKill = UseMI->getOperand(OpIdx).isKill();
+ unsigned NewReg = MRI->createVirtualRegister(MRI->getRegClass(Reg));
+ AddDefaultCC(AddDefaultPred(BuildMI(*UseMI->getParent(),
+ *UseMI, UseMI->getDebugLoc(),
+ get(NewUseOpc), NewReg)
+ .addReg(Reg1, getKillRegState(isKill))
+ .addImm(SOImmValV1)));
+ UseMI->setDesc(get(NewUseOpc));
+ UseMI->getOperand(1).setReg(NewReg);
+ UseMI->getOperand(1).setIsKill();
+ UseMI->getOperand(2).ChangeToImmediate(SOImmValV2);
+ DefMI->eraseFromParent();
+ return true;
+}
+
unsigned
-ARMBaseInstrInfo::getNumMicroOps(const MachineInstr *MI,
- const InstrItineraryData *ItinData) const {
+ARMBaseInstrInfo::getNumMicroOps(const InstrItineraryData *ItinData,
+ const MachineInstr *MI) const {
if (!ItinData || ItinData->isEmpty())
return 1;
default:
llvm_unreachable("Unexpected multi-uops instruction!");
break;
- case ARM::VLDMQ:
- case ARM::VSTMQ:
+ case ARM::VLDMQIA:
+ case ARM::VLDMQDB:
+ case ARM::VSTMQIA:
+ case ARM::VSTMQDB:
return 2;
// The number of uOps for load / store multiple are determined by the number
// registers.
+ //
// On Cortex-A8, each pair of register loads / stores can be scheduled on the
// same cycle. The scheduling for the first load / store must be done
// separately by assuming the the address is not 64-bit aligned.
+ //
// On Cortex-A9, the formula is simply (#reg / 2) + (#reg % 2). If the address
- // is not 64-bit aligned, then AGU would take an extra cycle.
- // For VFP / NEON load / store multiple, the formula is
- // (#reg / 2) + (#reg % 2) + 1.
- case ARM::VLDMD:
- case ARM::VLDMS:
- case ARM::VLDMD_UPD:
- case ARM::VLDMS_UPD:
- case ARM::VSTMD:
- case ARM::VSTMS:
- case ARM::VSTMD_UPD:
- case ARM::VSTMS_UPD: {
+ // is not 64-bit aligned, then AGU would take an extra cycle. For VFP / NEON
+ // load / store multiple, the formula is (#reg / 2) + (#reg % 2) + 1.
+ case ARM::VLDMDIA:
+ case ARM::VLDMDDB:
+ case ARM::VLDMDIA_UPD:
+ case ARM::VLDMDDB_UPD:
+ case ARM::VLDMSIA:
+ case ARM::VLDMSDB:
+ case ARM::VLDMSIA_UPD:
+ case ARM::VLDMSDB_UPD:
+ case ARM::VSTMDIA:
+ case ARM::VSTMDDB:
+ case ARM::VSTMDIA_UPD:
+ case ARM::VSTMDDB_UPD:
+ case ARM::VSTMSIA:
+ case ARM::VSTMSDB:
+ case ARM::VSTMSIA_UPD:
+ case ARM::VSTMSDB_UPD: {
unsigned NumRegs = MI->getNumOperands() - Desc.getNumOperands();
return (NumRegs / 2) + (NumRegs % 2) + 1;
}
- case ARM::LDM_RET:
- case ARM::LDM:
- case ARM::LDM_UPD:
- case ARM::STM:
- case ARM::STM_UPD:
- case ARM::tLDM:
- case ARM::tLDM_UPD:
- case ARM::tSTM_UPD:
+
+ case ARM::LDMIA_RET:
+ case ARM::LDMIA:
+ case ARM::LDMDA:
+ case ARM::LDMDB:
+ case ARM::LDMIB:
+ case ARM::LDMIA_UPD:
+ case ARM::LDMDA_UPD:
+ case ARM::LDMDB_UPD:
+ case ARM::LDMIB_UPD:
+ case ARM::STMIA:
+ case ARM::STMDA:
+ case ARM::STMDB:
+ case ARM::STMIB:
+ case ARM::STMIA_UPD:
+ case ARM::STMDA_UPD:
+ case ARM::STMDB_UPD:
+ case ARM::STMIB_UPD:
+ case ARM::tLDMIA:
+ case ARM::tLDMIA_UPD:
+ case ARM::tSTMIA:
+ case ARM::tSTMIA_UPD:
case ARM::tPOP_RET:
case ARM::tPOP:
case ARM::tPUSH:
- case ARM::t2LDM_RET:
- case ARM::t2LDM:
- case ARM::t2LDM_UPD:
- case ARM::t2STM:
- case ARM::t2STM_UPD: {
+ case ARM::t2LDMIA_RET:
+ case ARM::t2LDMIA:
+ case ARM::t2LDMDB:
+ case ARM::t2LDMIA_UPD:
+ case ARM::t2LDMDB_UPD:
+ case ARM::t2STMIA:
+ case ARM::t2STMDB:
+ case ARM::t2STMIA_UPD:
+ case ARM::t2STMDB_UPD: {
unsigned NumRegs = MI->getNumOperands() - Desc.getNumOperands() + 1;
if (Subtarget.isCortexA8()) {
- // 4 registers would be issued: 1, 2, 1.
- // 5 registers would be issued: 1, 2, 2.
- return 1 + (NumRegs / 2);
+ if (NumRegs < 4)
+ return 2;
+ // 4 registers would be issued: 2, 2.
+ // 5 registers would be issued: 2, 2, 1.
+ UOps = (NumRegs / 2);
+ if (NumRegs % 2)
+ ++UOps;
+ return UOps;
} else if (Subtarget.isCortexA9()) {
UOps = (NumRegs / 2);
// If there are odd number of registers or if it's not 64-bit aligned,
} else if (Subtarget.isCortexA9()) {
DefCycle = RegNo;
bool isSLoad = false;
+
switch (DefTID.getOpcode()) {
default: break;
- case ARM::VLDMS:
- case ARM::VLDMS_UPD:
+ case ARM::VLDMSIA:
+ case ARM::VLDMSDB:
+ case ARM::VLDMSIA_UPD:
+ case ARM::VLDMSDB_UPD:
isSLoad = true;
break;
}
+
// If there are odd number of 'S' registers or if it's not 64-bit aligned,
// then it takes an extra cycle.
if ((isSLoad && (RegNo % 2)) || DefAlign < 8)
} else if (Subtarget.isCortexA9()) {
UseCycle = RegNo;
bool isSStore = false;
+
switch (UseTID.getOpcode()) {
default: break;
- case ARM::VSTMS:
- case ARM::VSTMS_UPD:
+ case ARM::VSTMSIA:
+ case ARM::VSTMSDB:
+ case ARM::VSTMSIA_UPD:
+ case ARM::VSTMSDB_UPD:
isSStore = true;
break;
}
+
// If there are odd number of 'S' registers or if it's not 64-bit aligned,
// then it takes an extra cycle.
if ((isSStore && (RegNo % 2)) || UseAlign < 8)
default:
DefCycle = ItinData->getOperandCycle(DefClass, DefIdx);
break;
- case ARM::VLDMD:
- case ARM::VLDMS:
- case ARM::VLDMD_UPD:
- case ARM::VLDMS_UPD: {
+
+ case ARM::VLDMDIA:
+ case ARM::VLDMDDB:
+ case ARM::VLDMDIA_UPD:
+ case ARM::VLDMDDB_UPD:
+ case ARM::VLDMSIA:
+ case ARM::VLDMSDB:
+ case ARM::VLDMSIA_UPD:
+ case ARM::VLDMSDB_UPD:
DefCycle = getVLDMDefCycle(ItinData, DefTID, DefClass, DefIdx, DefAlign);
break;
- }
- case ARM::LDM_RET:
- case ARM::LDM:
- case ARM::LDM_UPD:
- case ARM::tLDM:
- case ARM::tLDM_UPD:
+
+ case ARM::LDMIA_RET:
+ case ARM::LDMIA:
+ case ARM::LDMDA:
+ case ARM::LDMDB:
+ case ARM::LDMIB:
+ case ARM::LDMIA_UPD:
+ case ARM::LDMDA_UPD:
+ case ARM::LDMDB_UPD:
+ case ARM::LDMIB_UPD:
+ case ARM::tLDMIA:
+ case ARM::tLDMIA_UPD:
case ARM::tPUSH:
- case ARM::t2LDM_RET:
- case ARM::t2LDM:
- case ARM::t2LDM_UPD: {
+ case ARM::t2LDMIA_RET:
+ case ARM::t2LDMIA:
+ case ARM::t2LDMDB:
+ case ARM::t2LDMIA_UPD:
+ case ARM::t2LDMDB_UPD:
LdmBypass = 1;
DefCycle = getLDMDefCycle(ItinData, DefTID, DefClass, DefIdx, DefAlign);
break;
}
- }
if (DefCycle == -1)
// We can't seem to determine the result latency of the def, assume it's 2.
default:
UseCycle = ItinData->getOperandCycle(UseClass, UseIdx);
break;
- case ARM::VSTMD:
- case ARM::VSTMS:
- case ARM::VSTMD_UPD:
- case ARM::VSTMS_UPD: {
+
+ case ARM::VSTMDIA:
+ case ARM::VSTMDDB:
+ case ARM::VSTMDIA_UPD:
+ case ARM::VSTMDDB_UPD:
+ case ARM::VSTMSIA:
+ case ARM::VSTMSDB:
+ case ARM::VSTMSIA_UPD:
+ case ARM::VSTMSDB_UPD:
UseCycle = getVSTMUseCycle(ItinData, UseTID, UseClass, UseIdx, UseAlign);
break;
- }
- case ARM::STM:
- case ARM::STM_UPD:
- case ARM::tSTM_UPD:
+
+ case ARM::STMIA:
+ case ARM::STMDA:
+ case ARM::STMDB:
+ case ARM::STMIB:
+ case ARM::STMIA_UPD:
+ case ARM::STMDA_UPD:
+ case ARM::STMDB_UPD:
+ case ARM::STMIB_UPD:
+ case ARM::tSTMIA:
+ case ARM::tSTMIA_UPD:
case ARM::tPOP_RET:
case ARM::tPOP:
- case ARM::t2STM:
- case ARM::t2STM_UPD: {
+ case ARM::t2STMIA:
+ case ARM::t2STMDB:
+ case ARM::t2STMIA_UPD:
+ case ARM::t2STMDB_UPD:
UseCycle = getSTMUseCycle(ItinData, UseTID, UseClass, UseIdx, UseAlign);
break;
}
- }
if (UseCycle == -1)
// Assume it's read in the first stage.
UseClass, UseIdx))
--UseCycle;
} else if (ItinData->hasPipelineForwarding(DefClass, DefIdx,
- UseClass, UseIdx))
+ UseClass, UseIdx)) {
--UseCycle;
+ }
}
return UseCycle;
return Latency;
}
+int ARMBaseInstrInfo::getInstrLatency(const InstrItineraryData *ItinData,
+ const MachineInstr *MI,
+ unsigned *PredCost) const {
+ if (MI->isCopyLike() || MI->isInsertSubreg() ||
+ MI->isRegSequence() || MI->isImplicitDef())
+ return 1;
+
+ if (!ItinData || ItinData->isEmpty())
+ return 1;
+
+ const TargetInstrDesc &TID = MI->getDesc();
+ unsigned Class = TID.getSchedClass();
+ unsigned UOps = ItinData->Itineraries[Class].NumMicroOps;
+ if (PredCost && TID.hasImplicitDefOfPhysReg(ARM::CPSR))
+ // When predicated, CPSR is an additional source operand for CPSR updating
+ // instructions, this apparently increases their latencies.
+ *PredCost = 1;
+ if (UOps)
+ return ItinData->getStageLatency(Class);
+ return getNumMicroOps(ItinData, MI);
+}
+
+int ARMBaseInstrInfo::getInstrLatency(const InstrItineraryData *ItinData,
+ SDNode *Node) const {
+ if (!Node->isMachineOpcode())
+ return 1;
+
+ if (!ItinData || ItinData->isEmpty())
+ return 1;
+
+ unsigned Opcode = Node->getMachineOpcode();
+ switch (Opcode) {
+ default:
+ return ItinData->getStageLatency(get(Opcode).getSchedClass());
+ case ARM::VLDMQIA:
+ case ARM::VLDMQDB:
+ case ARM::VSTMQIA:
+ case ARM::VSTMQDB:
+ return 2;
+ }
+}
+
bool ARMBaseInstrInfo::
hasHighOperandLatency(const InstrItineraryData *ItinData,
const MachineRegisterInfo *MRI,
}
return false;
}
+
+bool
+ARMBaseInstrInfo::isFpMLxInstruction(unsigned Opcode, unsigned &MulOpc,
+ unsigned &AddSubOpc,
+ bool &NegAcc, bool &HasLane) const {
+ DenseMap<unsigned, unsigned>::const_iterator I = MLxEntryMap.find(Opcode);
+ if (I == MLxEntryMap.end())
+ return false;
+
+ const ARM_MLxEntry &Entry = ARM_MLxTable[I->second];
+ MulOpc = Entry.MulOpc;
+ AddSubOpc = Entry.AddSubOpc;
+ NegAcc = Entry.NegAcc;
+ HasLane = Entry.HasLane;
+ return true;
+}
projects / oota-llvm.git / blobdiff