#include "ARM.h"
#include "ARMAddressingModes.h"
#include "ARMConstantPoolValue.h"
+#include "ARMHazardRecognizer.h"
#include "ARMMachineFunctionInfo.h"
#include "ARMRegisterInfo.h"
#include "ARMGenInstrInfo.inc"
EnableARM3Addr("enable-arm-3-addr-conv", cl::Hidden,
cl::desc("Enable ARM 2-addr to 3-addr conv"));
+/// 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::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);
+ }
+}
+
+// Use a ScoreboardHazardRecognizer for prepass ARM scheduling. TargetInstrImpl
+// currently defaults to no prepass hazard recognizer.
+ScheduleHazardRecognizer *ARMBaseInstrInfo::
+CreateTargetHazardRecognizer(const TargetMachine *TM,
+ const ScheduleDAG *DAG) const {
+ if (usePreRAHazardRecognizer()) {
+ const InstrItineraryData *II = TM->getInstrItineraryData();
+ return new ScoreboardHazardRecognizer(II, DAG, "pre-RA-sched");
+ }
+ return TargetInstrInfoImpl::CreateTargetHazardRecognizer(TM, DAG);
+}
+
+ScheduleHazardRecognizer *ARMBaseInstrInfo::
+CreateTargetPostRAHazardRecognizer(const InstrItineraryData *II,
+ const ScheduleDAG *DAG) const {
+ if (Subtarget.isThumb2() || Subtarget.hasVFP2())
+ return (ScheduleHazardRecognizer *)
+ new ARMHazardRecognizer(II, *this, getRegisterInfo(), Subtarget, DAG);
+ return TargetInstrInfoImpl::CreateTargetPostRAHazardRecognizer(II, DAG);
}
MachineInstr *
if (LV) {
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
MachineOperand &MO = MI->getOperand(i);
- if (MO.isReg() && MO.getReg() &&
- TargetRegisterInfo::isVirtualRegister(MO.getReg())) {
+ if (MO.isReg() && TargetRegisterInfo::isVirtualRegister(MO.getReg())) {
unsigned Reg = MO.getReg();
LiveVariables::VarInfo &VI = LV->getVarInfo(Reg);
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 ARMII::Size2Bytes: return 2; // Thumb1 instruction.
case ARMII::SizeSpecial: {
switch (Opc) {
+ case ARM::MOVi16_ga_pcrel:
+ case ARM::MOVTi16_ga_pcrel:
+ case ARM::t2MOVi16_ga_pcrel:
+ case ARM::t2MOVTi16_ga_pcrel:
+ return 4;
case ARM::MOVi32imm:
case ARM::t2MOVi32imm:
return 8;
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.
ARMConstantPoolValue *ACPV =
static_cast<ARMConstantPoolValue*>(MCPE.Val.MachineCPVal);
- unsigned PCLabelId = AFI->createConstPoolEntryUId();
+ unsigned PCLabelId = AFI->createPICLabelUId();
ARMConstantPoolValue *NewCPV = 0;
// FIXME: The below assumes PIC relocation model and that the function
// is Thumb mode (t1 or t2). PCAdjustment would be 8 for ARM mode PIC, and
}
bool ARMBaseInstrInfo::produceSameValue(const MachineInstr *MI0,
- const MachineInstr *MI1) const {
+ const MachineInstr *MI1,
+ const MachineRegisterInfo *MRI) const {
int Opcode = MI0->getOpcode();
if (Opcode == ARM::t2LDRpci ||
Opcode == ARM::t2LDRpci_pic ||
Opcode == ARM::tLDRpci ||
- Opcode == ARM::tLDRpci_pic) {
+ Opcode == ARM::tLDRpci_pic ||
+ Opcode == ARM::MOV_ga_dyn ||
+ Opcode == ARM::MOV_ga_pcrel ||
+ Opcode == ARM::MOV_ga_pcrel_ldr ||
+ Opcode == ARM::t2MOV_ga_dyn ||
+ Opcode == ARM::t2MOV_ga_pcrel) {
if (MI1->getOpcode() != Opcode)
return false;
if (MI0->getNumOperands() != MI1->getNumOperands())
if (MO0.getOffset() != MO1.getOffset())
return false;
+ if (Opcode == ARM::MOV_ga_dyn ||
+ Opcode == ARM::MOV_ga_pcrel ||
+ Opcode == ARM::MOV_ga_pcrel_ldr ||
+ Opcode == ARM::t2MOV_ga_dyn ||
+ Opcode == ARM::t2MOV_ga_pcrel)
+ // Ignore the PC labels.
+ return MO0.getGlobal() == MO1.getGlobal();
+
const MachineFunction *MF = MI0->getParent()->getParent();
const MachineConstantPool *MCP = MF->getConstantPool();
int CPI0 = MO0.getIndex();
int CPI1 = MO1.getIndex();
const MachineConstantPoolEntry &MCPE0 = MCP->getConstants()[CPI0];
const MachineConstantPoolEntry &MCPE1 = MCP->getConstants()[CPI1];
- ARMConstantPoolValue *ACPV0 =
- static_cast<ARMConstantPoolValue*>(MCPE0.Val.MachineCPVal);
- ARMConstantPoolValue *ACPV1 =
- static_cast<ARMConstantPoolValue*>(MCPE1.Val.MachineCPVal);
- return ACPV0->hasSameValue(ACPV1);
+ bool isARMCP0 = MCPE0.isMachineConstantPoolEntry();
+ bool isARMCP1 = MCPE1.isMachineConstantPoolEntry();
+ if (isARMCP0 && isARMCP1) {
+ ARMConstantPoolValue *ACPV0 =
+ static_cast<ARMConstantPoolValue*>(MCPE0.Val.MachineCPVal);
+ ARMConstantPoolValue *ACPV1 =
+ static_cast<ARMConstantPoolValue*>(MCPE1.Val.MachineCPVal);
+ return ACPV0->hasSameValue(ACPV1);
+ } else if (!isARMCP0 && !isARMCP1) {
+ return MCPE0.Val.ConstVal == MCPE1.Val.ConstVal;
+ }
+ return false;
+ } else if (Opcode == ARM::PICLDR) {
+ if (MI1->getOpcode() != Opcode)
+ return false;
+ if (MI0->getNumOperands() != MI1->getNumOperands())
+ return false;
+
+ unsigned Addr0 = MI0->getOperand(1).getReg();
+ unsigned Addr1 = MI1->getOperand(1).getReg();
+ if (Addr0 != Addr1) {
+ if (!MRI ||
+ !TargetRegisterInfo::isVirtualRegister(Addr0) ||
+ !TargetRegisterInfo::isVirtualRegister(Addr1))
+ return false;
+
+ // This assumes SSA form.
+ MachineInstr *Def0 = MRI->getVRegDef(Addr0);
+ MachineInstr *Def1 = MRI->getVRegDef(Addr1);
+ // Check if the loaded value, e.g. a constantpool of a global address, are
+ // the same.
+ if (!produceSameValue(Def0, Def1, MRI))
+ return false;
+ }
+
+ for (unsigned i = 3, e = MI0->getNumOperands(); i != e; ++i) {
+ // %vreg12<def> = PICLDR %vreg11, 0, pred:14, pred:%noreg
+ const MachineOperand &MO0 = MI0->getOperand(i);
+ const MachineOperand &MO1 = MI1->getOperand(i);
+ if (!MO0.isIdenticalTo(MO1))
+ return false;
+ }
+ return true;
}
return MI0->isIdenticalTo(MI1, MachineInstr::IgnoreVRegDefs);
}
/// shouldScheduleLoadsNear - This is a used by the pre-regalloc scheduler to
-/// determine (in conjuction with areLoadsFromSameBasePtr) if two loads should
+/// determine (in conjunction with areLoadsFromSameBasePtr) if two loads should
/// be scheduled togther. On some targets if two loads are loading from
/// addresses in the same cache line, it's better if they are scheduled
/// together. This function takes two integers that represent the load offsets
}
bool ARMBaseInstrInfo::isProfitableToIfCvt(MachineBasicBlock &MBB,
- unsigned NumCyles,
+ unsigned NumCycles,
unsigned ExtraPredCycles,
float Probability,
float Confidence) const {
- if (!NumCyles)
+ if (!NumCycles)
return false;
// Attempt to estimate the relative costs of predication versus branching.
- float UnpredCost = Probability * NumCyles;
+ float UnpredCost = Probability * NumCycles;
UnpredCost += 1.0; // The branch itself
UnpredCost += (1.0 - Confidence) * Subtarget.getMispredictionPenalty();
- return (float)(NumCyles + ExtraPredCycles) < UnpredCost;
+ return (float)(NumCycles + ExtraPredCycles) < UnpredCost;
}
bool ARMBaseInstrInfo::
MachineBasicBlock::iterator &MBBI, DebugLoc dl,
unsigned DestReg, unsigned BaseReg, int NumBytes,
ARMCC::CondCodes Pred, unsigned PredReg,
- const ARMBaseInstrInfo &TII) {
+ const ARMBaseInstrInfo &TII, unsigned MIFlags) {
bool isSub = NumBytes < 0;
if (isSub) NumBytes = -NumBytes;
unsigned Opc = isSub ? ARM::SUBri : ARM::ADDri;
BuildMI(MBB, MBBI, dl, TII.get(Opc), DestReg)
.addReg(BaseReg, RegState::Kill).addImm(ThisVal)
- .addImm((unsigned)Pred).addReg(PredReg).addReg(0);
+ .addImm((unsigned)Pred).addReg(PredReg).addReg(0)
+ .setMIFlags(MIFlags);
BaseReg = DestReg;
}
}
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();
// Set the "zero" bit in CPSR.
switch (MI->getOpcode()) {
default: break;
+ case ARM::RSBrr:
+ case ARM::RSBri:
+ case ARM::RSCrr:
+ case ARM::RSCri:
+ case ARM::ADDrr:
case ARM::ADDri:
- case ARM::ANDri:
- case ARM::t2ANDri:
+ case ARM::ADCrr:
+ case ARM::ADCri:
+ case ARM::SUBrr:
case ARM::SUBri:
+ case ARM::SBCrr:
+ case ARM::SBCri:
+ case ARM::t2RSBri:
+ case ARM::t2ADDrr:
case ARM::t2ADDri:
+ case ARM::t2ADCrr:
+ case ARM::t2ADCri:
+ case ARM::t2SUBrr:
case ARM::t2SUBri:
+ case ARM::t2SBCrr:
+ case ARM::t2SBCri:
+ case ARM::ANDrr:
+ case ARM::ANDri:
+ case ARM::t2ANDrr:
+ case ARM::t2ANDri:
+ case ARM::ORRrr:
+ case ARM::ORRri:
+ case ARM::t2ORRrr:
+ case ARM::t2ORRri:
+ case ARM::EORrr:
+ case ARM::EORri:
+ case ARM::t2EORrr:
+ case ARM::t2EORri: {
+ // Scan forward for the use of CPSR, if it's a conditional code requires
+ // checking of V bit, then this is not safe to do. If we can't find the
+ // CPSR use (i.e. used in another block), then it's not safe to perform
+ // the optimization.
+ bool isSafe = false;
+ I = CmpInstr;
+ E = MI->getParent()->end();
+ while (!isSafe && ++I != E) {
+ const MachineInstr &Instr = *I;
+ for (unsigned IO = 0, EO = Instr.getNumOperands();
+ !isSafe && IO != EO; ++IO) {
+ const MachineOperand &MO = Instr.getOperand(IO);
+ if (!MO.isReg() || MO.getReg() != ARM::CPSR)
+ continue;
+ if (MO.isDef()) {
+ isSafe = true;
+ break;
+ }
+ // Condition code is after the operand before CPSR.
+ ARMCC::CondCodes CC = (ARMCC::CondCodes)Instr.getOperand(IO-1).getImm();
+ switch (CC) {
+ default:
+ isSafe = true;
+ break;
+ case ARMCC::VS:
+ case ARMCC::VC:
+ case ARMCC::GE:
+ case ARMCC::LT:
+ case ARMCC::GT:
+ case ARMCC::LE:
+ return false;
+ }
+ }
+ }
+
+ if (!isSafe)
+ return false;
+
// Toggle the optional operand to CPSR.
MI->getOperand(5).setReg(ARM::CPSR);
MI->getOperand(5).setIsDef(true);
CmpInstr->eraseFromParent();
return true;
}
+ }
return false;
}
llvm_unreachable("Unexpected multi-uops instruction!");
break;
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.
// 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();
switch (DefTID.getOpcode()) {
default: break;
case ARM::VLDMSIA:
- case ARM::VLDMSDB:
case ARM::VLDMSIA_UPD:
case ARM::VLDMSDB_UPD:
isSLoad = true;
switch (UseTID.getOpcode()) {
default: break;
case ARM::VSTMSIA:
- case ARM::VSTMSDB:
case ARM::VSTMSIA_UPD:
case ARM::VSTMSDB_UPD:
isSStore = true;
break;
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::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);
if (!ItinData || ItinData->isEmpty())
return DefTID.mayLoad() ? 3 : 1;
-
const TargetInstrDesc &UseTID = UseMI->getDesc();
const MachineOperand &DefMO = DefMI->getOperand(DefIdx);
if (DefMO.getReg() == ARM::CPSR) {
}
}
+ if (DefAlign < 8 && Subtarget.isCortexA9())
+ switch (DefTID.getOpcode()) {
+ default: break;
+ case ARM::VLD1q8:
+ case ARM::VLD1q16:
+ case ARM::VLD1q32:
+ case ARM::VLD1q64:
+ case ARM::VLD1q8_UPD:
+ case ARM::VLD1q16_UPD:
+ case ARM::VLD1q32_UPD:
+ case ARM::VLD1q64_UPD:
+ case ARM::VLD2d8:
+ case ARM::VLD2d16:
+ case ARM::VLD2d32:
+ case ARM::VLD2q8:
+ case ARM::VLD2q16:
+ case ARM::VLD2q32:
+ case ARM::VLD2d8_UPD:
+ case ARM::VLD2d16_UPD:
+ case ARM::VLD2d32_UPD:
+ case ARM::VLD2q8_UPD:
+ case ARM::VLD2q16_UPD:
+ case ARM::VLD2q32_UPD:
+ case ARM::VLD3d8:
+ case ARM::VLD3d16:
+ case ARM::VLD3d32:
+ case ARM::VLD1d64T:
+ case ARM::VLD3d8_UPD:
+ case ARM::VLD3d16_UPD:
+ case ARM::VLD3d32_UPD:
+ case ARM::VLD1d64T_UPD:
+ case ARM::VLD3q8_UPD:
+ case ARM::VLD3q16_UPD:
+ case ARM::VLD3q32_UPD:
+ case ARM::VLD4d8:
+ case ARM::VLD4d16:
+ case ARM::VLD4d32:
+ case ARM::VLD1d64Q:
+ case ARM::VLD4d8_UPD:
+ case ARM::VLD4d16_UPD:
+ case ARM::VLD4d32_UPD:
+ case ARM::VLD1d64Q_UPD:
+ case ARM::VLD4q8_UPD:
+ case ARM::VLD4q16_UPD:
+ case ARM::VLD4q32_UPD:
+ case ARM::VLD1DUPq8:
+ case ARM::VLD1DUPq16:
+ case ARM::VLD1DUPq32:
+ case ARM::VLD1DUPq8_UPD:
+ case ARM::VLD1DUPq16_UPD:
+ case ARM::VLD1DUPq32_UPD:
+ case ARM::VLD2DUPd8:
+ case ARM::VLD2DUPd16:
+ case ARM::VLD2DUPd32:
+ case ARM::VLD2DUPd8_UPD:
+ case ARM::VLD2DUPd16_UPD:
+ case ARM::VLD2DUPd32_UPD:
+ case ARM::VLD4DUPd8:
+ case ARM::VLD4DUPd16:
+ case ARM::VLD4DUPd32:
+ case ARM::VLD4DUPd8_UPD:
+ case ARM::VLD4DUPd16_UPD:
+ case ARM::VLD4DUPd32_UPD:
+ case ARM::VLD1LNd8:
+ case ARM::VLD1LNd16:
+ case ARM::VLD1LNd32:
+ case ARM::VLD1LNd8_UPD:
+ case ARM::VLD1LNd16_UPD:
+ case ARM::VLD1LNd32_UPD:
+ case ARM::VLD2LNd8:
+ case ARM::VLD2LNd16:
+ case ARM::VLD2LNd32:
+ case ARM::VLD2LNq16:
+ case ARM::VLD2LNq32:
+ case ARM::VLD2LNd8_UPD:
+ case ARM::VLD2LNd16_UPD:
+ case ARM::VLD2LNd32_UPD:
+ case ARM::VLD2LNq16_UPD:
+ case ARM::VLD2LNq32_UPD:
+ case ARM::VLD4LNd8:
+ case ARM::VLD4LNd16:
+ case ARM::VLD4LNd32:
+ case ARM::VLD4LNq16:
+ case ARM::VLD4LNq32:
+ case ARM::VLD4LNd8_UPD:
+ case ARM::VLD4LNd16_UPD:
+ case ARM::VLD4LNd32_UPD:
+ case ARM::VLD4LNq16_UPD:
+ case ARM::VLD4LNq32_UPD:
+ // If the address is not 64-bit aligned, the latencies of these
+ // instructions increases by one.
+ ++Latency;
+ break;
+ }
+
return Latency;
}
return 1;
const TargetInstrDesc &DefTID = get(DefNode->getMachineOpcode());
+
+ if (isZeroCost(DefTID.Opcode))
+ return 0;
+
if (!ItinData || ItinData->isEmpty())
return DefTID.mayLoad() ? 3 : 1;
}
}
+ if (DefAlign < 8 && Subtarget.isCortexA9())
+ switch (DefTID.getOpcode()) {
+ default: break;
+ case ARM::VLD1q8Pseudo:
+ case ARM::VLD1q16Pseudo:
+ case ARM::VLD1q32Pseudo:
+ case ARM::VLD1q64Pseudo:
+ case ARM::VLD1q8Pseudo_UPD:
+ case ARM::VLD1q16Pseudo_UPD:
+ case ARM::VLD1q32Pseudo_UPD:
+ case ARM::VLD1q64Pseudo_UPD:
+ case ARM::VLD2d8Pseudo:
+ case ARM::VLD2d16Pseudo:
+ case ARM::VLD2d32Pseudo:
+ case ARM::VLD2q8Pseudo:
+ case ARM::VLD2q16Pseudo:
+ case ARM::VLD2q32Pseudo:
+ case ARM::VLD2d8Pseudo_UPD:
+ case ARM::VLD2d16Pseudo_UPD:
+ case ARM::VLD2d32Pseudo_UPD:
+ case ARM::VLD2q8Pseudo_UPD:
+ case ARM::VLD2q16Pseudo_UPD:
+ case ARM::VLD2q32Pseudo_UPD:
+ case ARM::VLD3d8Pseudo:
+ case ARM::VLD3d16Pseudo:
+ case ARM::VLD3d32Pseudo:
+ case ARM::VLD1d64TPseudo:
+ case ARM::VLD3d8Pseudo_UPD:
+ case ARM::VLD3d16Pseudo_UPD:
+ case ARM::VLD3d32Pseudo_UPD:
+ case ARM::VLD1d64TPseudo_UPD:
+ case ARM::VLD3q8Pseudo_UPD:
+ case ARM::VLD3q16Pseudo_UPD:
+ case ARM::VLD3q32Pseudo_UPD:
+ case ARM::VLD3q8oddPseudo:
+ case ARM::VLD3q16oddPseudo:
+ case ARM::VLD3q32oddPseudo:
+ case ARM::VLD3q8oddPseudo_UPD:
+ case ARM::VLD3q16oddPseudo_UPD:
+ case ARM::VLD3q32oddPseudo_UPD:
+ case ARM::VLD4d8Pseudo:
+ case ARM::VLD4d16Pseudo:
+ case ARM::VLD4d32Pseudo:
+ case ARM::VLD1d64QPseudo:
+ case ARM::VLD4d8Pseudo_UPD:
+ case ARM::VLD4d16Pseudo_UPD:
+ case ARM::VLD4d32Pseudo_UPD:
+ case ARM::VLD1d64QPseudo_UPD:
+ case ARM::VLD4q8Pseudo_UPD:
+ case ARM::VLD4q16Pseudo_UPD:
+ case ARM::VLD4q32Pseudo_UPD:
+ case ARM::VLD4q8oddPseudo:
+ case ARM::VLD4q16oddPseudo:
+ case ARM::VLD4q32oddPseudo:
+ case ARM::VLD4q8oddPseudo_UPD:
+ case ARM::VLD4q16oddPseudo_UPD:
+ case ARM::VLD4q32oddPseudo_UPD:
+ case ARM::VLD1DUPq8Pseudo:
+ case ARM::VLD1DUPq16Pseudo:
+ case ARM::VLD1DUPq32Pseudo:
+ case ARM::VLD1DUPq8Pseudo_UPD:
+ case ARM::VLD1DUPq16Pseudo_UPD:
+ case ARM::VLD1DUPq32Pseudo_UPD:
+ case ARM::VLD2DUPd8Pseudo:
+ case ARM::VLD2DUPd16Pseudo:
+ case ARM::VLD2DUPd32Pseudo:
+ case ARM::VLD2DUPd8Pseudo_UPD:
+ case ARM::VLD2DUPd16Pseudo_UPD:
+ case ARM::VLD2DUPd32Pseudo_UPD:
+ case ARM::VLD4DUPd8Pseudo:
+ case ARM::VLD4DUPd16Pseudo:
+ case ARM::VLD4DUPd32Pseudo:
+ case ARM::VLD4DUPd8Pseudo_UPD:
+ case ARM::VLD4DUPd16Pseudo_UPD:
+ case ARM::VLD4DUPd32Pseudo_UPD:
+ case ARM::VLD1LNq8Pseudo:
+ case ARM::VLD1LNq16Pseudo:
+ case ARM::VLD1LNq32Pseudo:
+ case ARM::VLD1LNq8Pseudo_UPD:
+ case ARM::VLD1LNq16Pseudo_UPD:
+ case ARM::VLD1LNq32Pseudo_UPD:
+ case ARM::VLD2LNd8Pseudo:
+ case ARM::VLD2LNd16Pseudo:
+ case ARM::VLD2LNd32Pseudo:
+ case ARM::VLD2LNq16Pseudo:
+ case ARM::VLD2LNq32Pseudo:
+ case ARM::VLD2LNd8Pseudo_UPD:
+ case ARM::VLD2LNd16Pseudo_UPD:
+ case ARM::VLD2LNd32Pseudo_UPD:
+ case ARM::VLD2LNq16Pseudo_UPD:
+ case ARM::VLD2LNq32Pseudo_UPD:
+ case ARM::VLD4LNd8Pseudo:
+ case ARM::VLD4LNd16Pseudo:
+ case ARM::VLD4LNd32Pseudo:
+ case ARM::VLD4LNq16Pseudo:
+ case ARM::VLD4LNq32Pseudo:
+ case ARM::VLD4LNd8Pseudo_UPD:
+ case ARM::VLD4LNd16Pseudo_UPD:
+ case ARM::VLD4LNd32Pseudo_UPD:
+ case ARM::VLD4LNq16Pseudo_UPD:
+ case ARM::VLD4LNq32Pseudo_UPD:
+ // If the address is not 64-bit aligned, the latencies of these
+ // instructions increases by one.
+ ++Latency;
+ break;
+ }
+
return Latency;
}
default:
return ItinData->getStageLatency(get(Opcode).getSchedClass());
case ARM::VLDMQIA:
- case ARM::VLDMQDB:
case ARM::VSTMQIA:
- case ARM::VSTMQDB:
return 2;
- }
+ }
}
bool ARMBaseInstrInfo::
}
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;
+}