#include "PPCTargetMachine.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/LiveIntervalAnalysis.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
+#include "llvm/CodeGen/ScheduleDAG.h"
+#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"
#include "llvm/Support/TargetRegistry.h"
#include "llvm/Support/raw_ostream.h"
+using namespace llvm;
+
+#define DEBUG_TYPE "ppc-instr-info"
+
#define GET_INSTRMAP_INFO
#define GET_INSTRINFO_CTOR_DTOR
#include "PPCGenInstrInfo.inc"
-using namespace llvm;
-
static cl::
opt<bool> DisableCTRLoopAnal("disable-ppc-ctrloop-analysis", cl::Hidden,
cl::desc("Disable analysis for CTR loops"));
static cl::opt<bool> DisableCmpOpt("disable-ppc-cmp-opt",
cl::desc("Disable compare instruction optimization"), cl::Hidden);
+static cl::opt<bool> VSXSelfCopyCrash("crash-on-ppc-vsx-self-copy",
+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(PPCTargetMachine &tm)
- : PPCGenInstrInfo(PPC::ADJCALLSTACKDOWN, PPC::ADJCALLSTACKUP),
- TM(tm), RI(*TM.getSubtargetImpl()) {}
+PPCInstrInfo::PPCInstrInfo(PPCSubtarget &STI)
+ : PPCGenInstrInfo(PPC::ADJCALLSTACKDOWN, PPC::ADJCALLSTACKUP),
+ Subtarget(STI), RI(STI.getTargetMachine()) {}
/// CreateTargetHazardRecognizer - Return the hazard recognizer to use for
/// this target when scheduling the DAG.
-ScheduleHazardRecognizer *PPCInstrInfo::CreateTargetHazardRecognizer(
- const TargetMachine *TM,
- const ScheduleDAG *DAG) const {
- unsigned Directive = TM->getSubtarget<PPCSubtarget>().getDarwinDirective();
+ScheduleHazardRecognizer *
+PPCInstrInfo::CreateTargetHazardRecognizer(const TargetSubtargetInfo *STI,
+ const ScheduleDAG *DAG) const {
+ unsigned Directive =
+ static_cast<const PPCSubtarget *>(STI)->getDarwinDirective();
if (Directive == PPC::DIR_440 || Directive == PPC::DIR_A2 ||
Directive == PPC::DIR_E500mc || Directive == PPC::DIR_E5500) {
- const InstrItineraryData *II = TM->getInstrItineraryData();
+ const InstrItineraryData *II =
+ static_cast<const PPCSubtarget *>(STI)->getInstrItineraryData();
return new ScoreboardHazardRecognizer(II, DAG);
}
- return TargetInstrInfo::CreateTargetHazardRecognizer(TM, DAG);
+ return TargetInstrInfo::CreateTargetHazardRecognizer(STI, DAG);
}
/// CreateTargetPostRAHazardRecognizer - Return the postRA hazard recognizer
/// to use for this target when scheduling the DAG.
-ScheduleHazardRecognizer *PPCInstrInfo::CreateTargetPostRAHazardRecognizer(
- const InstrItineraryData *II,
- const ScheduleDAG *DAG) const {
- unsigned Directive = TM.getSubtarget<PPCSubtarget>().getDarwinDirective();
+ScheduleHazardRecognizer *
+PPCInstrInfo::CreateTargetPostRAHazardRecognizer(const InstrItineraryData *II,
+ const ScheduleDAG *DAG) const {
+ unsigned Directive =
+ DAG->MF.getSubtarget<PPCSubtarget>().getDarwinDirective();
- if (Directive == PPC::DIR_PWR7)
+ if (Directive == PPC::DIR_PWR7 || Directive == PPC::DIR_PWR8)
return new PPCDispatchGroupSBHazardRecognizer(II, DAG);
// Most subtargets use a PPC970 recognizer.
if (Directive != PPC::DIR_440 && Directive != PPC::DIR_A2 &&
Directive != PPC::DIR_E500mc && Directive != PPC::DIR_E5500) {
- assert(TM.getInstrInfo() && "No InstrInfo?");
+ assert(DAG->TII && "No InstrInfo?");
- return new PPCHazardRecognizer970(TM);
+ return new PPCHazardRecognizer970(*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) ||
// On some cores, there is an additional delay between writing to a condition
// register, and using it from a branch.
- unsigned Directive = TM.getSubtarget<PPCSubtarget>().getDarwinDirective();
+ unsigned Directive = Subtarget.getDarwinDirective();
switch (Directive) {
default: break;
case PPC::DIR_7400:
case PPC::DIR_PWR6:
case PPC::DIR_PWR6X:
case PPC::DIR_PWR7:
+ case PPC::DIR_PWR8:
Latency += 2;
break;
}
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::LFS:
case PPC::LFD:
case PPC::RESTORE_CR:
+ 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::STFS:
case PPC::STFD:
case PPC::SPILL_CR:
+ 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
+ // to the high-bits of the mask (and, thus, the result).
// Cannot commute if it has a non-zero rotate count.
if (MI->getOperand(3).getImm() != 0)
- return 0;
+ return nullptr;
// If we have a zero rotate count, we have:
// M = mask(MB,ME)
// 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();
return MI;
}
+bool PPCInstrInfo::findCommutedOpIndices(MachineInstr *MI, unsigned &SrcOpIdx1,
+ unsigned &SrcOpIdx2) const {
+ // For VSX A-Type FMA instructions, it is the first two operands that can be
+ // commuted, however, because the non-encoded tied input operand is listed
+ // first, the operands to swap are actually the second and third.
+
+ int AltOpc = PPC::getAltVSXFMAOpcode(MI->getOpcode());
+ if (AltOpc == -1)
+ return TargetInstrInfo::findCommutedOpIndices(MI, SrcOpIdx1, SrcOpIdx2);
+
+ // 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,
MachineBasicBlock::iterator MI) const {
// This function is used for scheduling, and the nop wanted here is the type
// that terminates dispatch groups on the POWER cores.
- unsigned Directive = TM.getSubtarget<PPCSubtarget>().getDarwinDirective();
+ unsigned Directive = Subtarget.getDarwinDirective();
unsigned Opcode;
switch (Directive) {
default: Opcode = PPC::NOP; break;
case PPC::DIR_PWR6: Opcode = PPC::NOP_GT_PWR6; break;
case PPC::DIR_PWR7: Opcode = PPC::NOP_GT_PWR7; break;
+ case PPC::DIR_PWR8: Opcode = PPC::NOP_GT_PWR7; break; /* FIXME: Update when P8 InstrScheduling model is ready */
}
DebugLoc DL;
BuildMI(MBB, MI, DL, get(Opcode));
}
+/// getNoopForMachoTarget - Return the noop instruction to use for a noop.
+void PPCInstrInfo::getNoopForMachoTarget(MCInst &NopInst) const {
+ NopInst.setOpcode(PPC::NOP);
+}
+
// Branch analysis.
// Note: If the condition register is set to CTR or CTR8 then this is a
// BDNZ (imm == 1) or BDZ (imm == 0) branch.
MachineBasicBlock *&FBB,
SmallVectorImpl<MachineOperand> &Cond,
bool AllowModify) const {
- bool isPPC64 = TM.getSubtargetImpl()->isPPC64();
+ 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;
Cond.push_back(LastInst->getOperand(0));
Cond.push_back(LastInst->getOperand(1));
return false;
+ } else if (LastInst->getOpcode() == PPC::BC) {
+ if (!LastInst->getOperand(1).isMBB())
+ return true;
+ // Block ends with fall-through condbranch.
+ TBB = LastInst->getOperand(1).getMBB();
+ Cond.push_back(MachineOperand::CreateImm(PPC::PRED_BIT_SET));
+ Cond.push_back(LastInst->getOperand(0));
+ return false;
+ } else if (LastInst->getOpcode() == PPC::BCn) {
+ if (!LastInst->getOperand(1).isMBB())
+ return true;
+ // Block ends with fall-through condbranch.
+ TBB = LastInst->getOperand(1).getMBB();
+ Cond.push_back(MachineOperand::CreateImm(PPC::PRED_BIT_UNSET));
+ Cond.push_back(LastInst->getOperand(0));
+ return false;
} else if (LastInst->getOpcode() == PPC::BDNZ8 ||
LastInst->getOpcode() == PPC::BDNZ) {
if (!LastInst->getOperand(0).isMBB())
Cond.push_back(SecondLastInst->getOperand(1));
FBB = LastInst->getOperand(0).getMBB();
return false;
+ } else if (SecondLastInst->getOpcode() == PPC::BC &&
+ LastInst->getOpcode() == PPC::B) {
+ if (!SecondLastInst->getOperand(1).isMBB() ||
+ !LastInst->getOperand(0).isMBB())
+ return true;
+ TBB = SecondLastInst->getOperand(1).getMBB();
+ Cond.push_back(MachineOperand::CreateImm(PPC::PRED_BIT_SET));
+ Cond.push_back(SecondLastInst->getOperand(0));
+ FBB = LastInst->getOperand(0).getMBB();
+ return false;
+ } else if (SecondLastInst->getOpcode() == PPC::BCn &&
+ LastInst->getOpcode() == PPC::B) {
+ if (!SecondLastInst->getOperand(1).isMBB() ||
+ !LastInst->getOperand(0).isMBB())
+ return true;
+ TBB = SecondLastInst->getOperand(1).getMBB();
+ Cond.push_back(MachineOperand::CreateImm(PPC::PRED_BIT_UNSET));
+ Cond.push_back(SecondLastInst->getOperand(0));
+ FBB = LastInst->getOperand(0).getMBB();
+ return false;
} else if ((SecondLastInst->getOpcode() == PPC::BDNZ8 ||
SecondLastInst->getOpcode() == PPC::BDNZ) &&
LastInst->getOpcode() == PPC::B) {
}
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 &&
I->getOpcode() != PPC::BDZ8 && I->getOpcode() != PPC::BDZ)
return 0;
if (I == MBB.begin()) return 1;
--I;
if (I->getOpcode() != PPC::BCC &&
+ I->getOpcode() != PPC::BC && I->getOpcode() != PPC::BCn &&
I->getOpcode() != PPC::BDNZ8 && I->getOpcode() != PPC::BDNZ &&
I->getOpcode() != PPC::BDZ8 && I->getOpcode() != PPC::BDZ)
return 1;
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");
assert((Cond.size() == 2 || Cond.size() == 0) &&
"PPC branch conditions have two components!");
- bool isPPC64 = TM.getSubtargetImpl()->isPPC64();
+ bool isPPC64 = Subtarget.isPPC64();
// One-way branch.
- if (FBB == 0) {
+ if (!FBB) {
if (Cond.empty()) // Unconditional branch
BuildMI(&MBB, DL, get(PPC::B)).addMBB(TBB);
else if (Cond[1].getReg() == PPC::CTR || Cond[1].getReg() == PPC::CTR8)
BuildMI(&MBB, DL, get(Cond[0].getImm() ?
(isPPC64 ? PPC::BDNZ8 : PPC::BDNZ) :
(isPPC64 ? PPC::BDZ8 : PPC::BDZ))).addMBB(TBB);
+ else if (Cond[0].getImm() == PPC::PRED_BIT_SET)
+ BuildMI(&MBB, DL, get(PPC::BC)).addOperand(Cond[1]).addMBB(TBB);
+ else if (Cond[0].getImm() == PPC::PRED_BIT_UNSET)
+ BuildMI(&MBB, DL, get(PPC::BCn)).addOperand(Cond[1]).addMBB(TBB);
else // Conditional branch
BuildMI(&MBB, DL, get(PPC::BCC))
- .addImm(Cond[0].getImm()).addReg(Cond[1].getReg()).addMBB(TBB);
+ .addImm(Cond[0].getImm()).addOperand(Cond[1]).addMBB(TBB);
return 1;
}
BuildMI(&MBB, DL, get(Cond[0].getImm() ?
(isPPC64 ? PPC::BDNZ8 : PPC::BDNZ) :
(isPPC64 ? PPC::BDZ8 : PPC::BDZ))).addMBB(TBB);
+ else if (Cond[0].getImm() == PPC::PRED_BIT_SET)
+ BuildMI(&MBB, DL, get(PPC::BC)).addOperand(Cond[1]).addMBB(TBB);
+ else if (Cond[0].getImm() == PPC::PRED_BIT_UNSET)
+ BuildMI(&MBB, DL, get(PPC::BCn)).addOperand(Cond[1]).addMBB(TBB);
else
BuildMI(&MBB, DL, get(PPC::BCC))
- .addImm(Cond[0].getImm()).addReg(Cond[1].getReg()).addMBB(TBB);
+ .addImm(Cond[0].getImm()).addOperand(Cond[1]).addMBB(TBB);
BuildMI(&MBB, DL, get(PPC::B)).addMBB(FBB);
return 2;
}
// 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 (!TM.getSubtargetImpl()->hasISEL())
+ if (!Subtarget.hasISEL())
return false;
if (Cond.size() != 2)
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!");
- assert(TM.getSubtargetImpl()->hasISEL() &&
+ assert(Subtarget.hasISEL() &&
"Cannot insert select on target without ISEL support");
// Get the register classes.
case PPC::PRED_LE: SubIdx = PPC::sub_gt; SwapOps = true; break;
case PPC::PRED_UN: SubIdx = PPC::sub_un; SwapOps = false; break;
case PPC::PRED_NU: SubIdx = PPC::sub_un; SwapOps = true; break;
+ case PPC::PRED_BIT_SET: SubIdx = 0; SwapOps = false; break;
+ case PPC::PRED_BIT_UNSET: SubIdx = 0; SwapOps = true; break;
}
unsigned FirstReg = SwapOps ? FalseReg : TrueReg,
.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,
bool KillSrc) const {
+ // We can end up with self copies and similar things as a result of VSX copy
+ // legalization. Promote them here.
+ const TargetRegisterInfo *TRI = &getRegisterInfo();
+ if (PPC::F8RCRegClass.contains(DestReg) &&
+ PPC::VSRCRegClass.contains(SrcReg)) {
+ unsigned SuperReg =
+ TRI->getMatchingSuperReg(DestReg, PPC::sub_64, &PPC::VSRCRegClass);
+
+ if (VSXSelfCopyCrash && SrcReg == SuperReg)
+ llvm_unreachable("nop VSX copy");
+
+ DestReg = SuperReg;
+ } else if (PPC::VRRCRegClass.contains(DestReg) &&
+ PPC::VSRCRegClass.contains(SrcReg)) {
+ unsigned SuperReg =
+ TRI->getMatchingSuperReg(DestReg, PPC::sub_128, &PPC::VSRCRegClass);
+
+ if (VSXSelfCopyCrash && SrcReg == SuperReg)
+ llvm_unreachable("nop VSX copy");
+
+ DestReg = SuperReg;
+ } else if (PPC::F8RCRegClass.contains(SrcReg) &&
+ PPC::VSRCRegClass.contains(DestReg)) {
+ unsigned SuperReg =
+ TRI->getMatchingSuperReg(SrcReg, PPC::sub_64, &PPC::VSRCRegClass);
+
+ if (VSXSelfCopyCrash && DestReg == SuperReg)
+ llvm_unreachable("nop VSX copy");
+
+ SrcReg = SuperReg;
+ } else if (PPC::VRRCRegClass.contains(SrcReg) &&
+ PPC::VSRCRegClass.contains(DestReg)) {
+ unsigned SuperReg =
+ TRI->getMatchingSuperReg(SrcReg, PPC::sub_128, &PPC::VSRCRegClass);
+
+ if (VSXSelfCopyCrash && DestReg == SuperReg)
+ llvm_unreachable("nop VSX copy");
+
+ 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;
Opc = PPC::MCRF;
else if (PPC::VRRCRegClass.contains(DestReg, SrcReg))
Opc = PPC::VOR;
+ else if (PPC::VSRCRegClass.contains(DestReg, SrcReg))
+ // There are two different ways this can be done:
+ // 1. xxlor : This has lower latency (on the P7), 2 cycles, but can only
+ // issue in VSU pipeline 0.
+ // 2. xmovdp/xmovsp: This has higher latency (on the P7), 6 cycles, but
+ // can go to either pipeline.
+ // We'll always use xxlor here, because in practically all cases where
+ // 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) ||
+ 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
FrameIdx));
return true;
} else if (PPC::CRBITRCRegClass.hasSubClassEq(RC)) {
- // FIXME: We use CRi here because there is no mtcrf on a bit. Since the
- // backend currently only uses CR1EQ as an individual bit, this should
- // not cause any bug. If we need other uses of CR bits, the following
- // code may be invalid.
- unsigned Reg = 0;
- if (SrcReg == PPC::CR0LT || SrcReg == PPC::CR0GT ||
- SrcReg == PPC::CR0EQ || SrcReg == PPC::CR0UN)
- Reg = PPC::CR0;
- else if (SrcReg == PPC::CR1LT || SrcReg == PPC::CR1GT ||
- SrcReg == PPC::CR1EQ || SrcReg == PPC::CR1UN)
- Reg = PPC::CR1;
- else if (SrcReg == PPC::CR2LT || SrcReg == PPC::CR2GT ||
- SrcReg == PPC::CR2EQ || SrcReg == PPC::CR2UN)
- Reg = PPC::CR2;
- else if (SrcReg == PPC::CR3LT || SrcReg == PPC::CR3GT ||
- SrcReg == PPC::CR3EQ || SrcReg == PPC::CR3UN)
- Reg = PPC::CR3;
- else if (SrcReg == PPC::CR4LT || SrcReg == PPC::CR4GT ||
- SrcReg == PPC::CR4EQ || SrcReg == PPC::CR4UN)
- Reg = PPC::CR4;
- else if (SrcReg == PPC::CR5LT || SrcReg == PPC::CR5GT ||
- SrcReg == PPC::CR5EQ || SrcReg == PPC::CR5UN)
- Reg = PPC::CR5;
- else if (SrcReg == PPC::CR6LT || SrcReg == PPC::CR6GT ||
- SrcReg == PPC::CR6EQ || SrcReg == PPC::CR6UN)
- Reg = PPC::CR6;
- else if (SrcReg == PPC::CR7LT || SrcReg == PPC::CR7GT ||
- SrcReg == PPC::CR7EQ || SrcReg == PPC::CR7UN)
- Reg = PPC::CR7;
-
- return StoreRegToStackSlot(MF, Reg, isKill, FrameIdx,
- &PPC::CRRCRegClass, NewMIs, NonRI, SpillsVRS);
-
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::SPILL_CRBIT))
+ .addReg(SrcReg,
+ getKillRegState(isKill)),
+ FrameIdx));
+ return true;
} else if (PPC::VRRCRegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STVX))
.addReg(SrcReg,
getKillRegState(isKill)),
FrameIdx));
NonRI = true;
+ } else if (PPC::VSRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STXVD2X))
+ .addReg(SrcReg,
+ getKillRegState(isKill)),
+ FrameIdx));
+ NonRI = true;
+ } else if (PPC::VSFRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STXSDX))
+ .addReg(SrcReg,
+ 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(TM.getSubtargetImpl()->isDarwin() &&
+ assert(Subtarget.isDarwin() &&
"VRSAVE only needs spill/restore on Darwin");
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::SPILL_VRSAVE))
.addReg(SrcReg,
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);
}
FrameIdx));
return true;
} else if (PPC::CRBITRCRegClass.hasSubClassEq(RC)) {
-
- unsigned Reg = 0;
- if (DestReg == PPC::CR0LT || DestReg == PPC::CR0GT ||
- DestReg == PPC::CR0EQ || DestReg == PPC::CR0UN)
- Reg = PPC::CR0;
- else if (DestReg == PPC::CR1LT || DestReg == PPC::CR1GT ||
- DestReg == PPC::CR1EQ || DestReg == PPC::CR1UN)
- Reg = PPC::CR1;
- else if (DestReg == PPC::CR2LT || DestReg == PPC::CR2GT ||
- DestReg == PPC::CR2EQ || DestReg == PPC::CR2UN)
- Reg = PPC::CR2;
- else if (DestReg == PPC::CR3LT || DestReg == PPC::CR3GT ||
- DestReg == PPC::CR3EQ || DestReg == PPC::CR3UN)
- Reg = PPC::CR3;
- else if (DestReg == PPC::CR4LT || DestReg == PPC::CR4GT ||
- DestReg == PPC::CR4EQ || DestReg == PPC::CR4UN)
- Reg = PPC::CR4;
- else if (DestReg == PPC::CR5LT || DestReg == PPC::CR5GT ||
- DestReg == PPC::CR5EQ || DestReg == PPC::CR5UN)
- Reg = PPC::CR5;
- else if (DestReg == PPC::CR6LT || DestReg == PPC::CR6GT ||
- DestReg == PPC::CR6EQ || DestReg == PPC::CR6UN)
- Reg = PPC::CR6;
- else if (DestReg == PPC::CR7LT || DestReg == PPC::CR7GT ||
- DestReg == PPC::CR7EQ || DestReg == PPC::CR7UN)
- Reg = PPC::CR7;
-
- return LoadRegFromStackSlot(MF, DL, Reg, FrameIdx,
- &PPC::CRRCRegClass, NewMIs, NonRI, SpillsVRS);
-
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL,
+ get(PPC::RESTORE_CRBIT), DestReg),
+ FrameIdx));
+ return true;
} else if (PPC::VRRCRegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LVX), DestReg),
FrameIdx));
NonRI = true;
+ } else if (PPC::VSRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LXVD2X), DestReg),
+ FrameIdx));
+ NonRI = true;
+ } else if (PPC::VSFRCRegClass.hasSubClassEq(RC)) {
+ 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(TM.getSubtargetImpl()->isDarwin() &&
+ assert(Subtarget.isDarwin() &&
"VRSAVE only needs spill/restore on Darwin");
NewMIs.push_back(addFrameReference(BuildMI(MF, DL,
get(PPC::RESTORE_VRSAVE),
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 ZeroReg;
if (UseInfo->isLookupPtrRegClass()) {
- bool isPPC64 = TM.getSubtargetImpl()->isPPC64();
+ bool isPPC64 = Subtarget.isPPC64();
ZeroReg = isPPC64 ? PPC::ZERO8 : PPC::ZERO;
} else {
ZeroReg = UseInfo->RegClass == PPC::G8RC_NOX0RegClassID ?
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) {
+ if (OpC == PPC::BLR || OpC == PPC::BLR8) {
if (Pred[1].getReg() == PPC::CTR8 || Pred[1].getReg() == PPC::CTR) {
- bool isPPC64 = TM.getSubtargetImpl()->isPPC64();
+ bool isPPC64 = Subtarget.isPPC64();
MI->setDesc(get(Pred[0].getImm() ?
(isPPC64 ? PPC::BDNZLR8 : PPC::BDNZLR) :
(isPPC64 ? PPC::BDZLR8 : PPC::BDZLR)));
- } else {
+ } else if (Pred[0].getImm() == PPC::PRED_BIT_SET) {
MI->setDesc(get(PPC::BCLR));
+ MachineInstrBuilder(*MI->getParent()->getParent(), MI)
+ .addReg(Pred[1].getReg());
+ } else if (Pred[0].getImm() == PPC::PRED_BIT_UNSET) {
+ MI->setDesc(get(PPC::BCLRn));
+ MachineInstrBuilder(*MI->getParent()->getParent(), MI)
+ .addReg(Pred[1].getReg());
+ } else {
+ MI->setDesc(get(PPC::BCCLR));
MachineInstrBuilder(*MI->getParent()->getParent(), MI)
.addImm(Pred[0].getImm())
.addReg(Pred[1].getReg());
return true;
} else if (OpC == PPC::B) {
if (Pred[1].getReg() == PPC::CTR8 || Pred[1].getReg() == PPC::CTR) {
- bool isPPC64 = TM.getSubtargetImpl()->isPPC64();
+ bool isPPC64 = Subtarget.isPPC64();
MI->setDesc(get(Pred[0].getImm() ?
(isPPC64 ? PPC::BDNZ8 : PPC::BDNZ) :
(isPPC64 ? PPC::BDZ8 : PPC::BDZ)));
+ } else if (Pred[0].getImm() == PPC::PRED_BIT_SET) {
+ MachineBasicBlock *MBB = MI->getOperand(0).getMBB();
+ MI->RemoveOperand(0);
+
+ MI->setDesc(get(PPC::BC));
+ MachineInstrBuilder(*MI->getParent()->getParent(), MI)
+ .addReg(Pred[1].getReg())
+ .addMBB(MBB);
+ } else if (Pred[0].getImm() == PPC::PRED_BIT_UNSET) {
+ MachineBasicBlock *MBB = MI->getOperand(0).getMBB();
+ MI->RemoveOperand(0);
+
+ MI->setDesc(get(PPC::BCn));
+ MachineInstrBuilder(*MI->getParent()->getParent(), MI)
+ .addReg(Pred[1].getReg())
+ .addMBB(MBB);
} else {
MachineBasicBlock *MBB = MI->getOperand(0).getMBB();
MI->RemoveOperand(0);
llvm_unreachable("Cannot predicate bctr[l] on the ctr register");
bool setLR = OpC == PPC::BCTRL || OpC == PPC::BCTRL8;
- bool isPPC64 = TM.getSubtargetImpl()->isPPC64();
- MI->setDesc(get(isPPC64 ? (setLR ? PPC::BCCTRL8 : PPC::BCCTR8) :
- (setLR ? PPC::BCCTRL : PPC::BCCTR)));
+ bool isPPC64 = Subtarget.isPPC64();
+
+ if (Pred[0].getImm() == PPC::PRED_BIT_SET) {
+ MI->setDesc(get(isPPC64 ? (setLR ? PPC::BCCTRL8 : PPC::BCCTR8) :
+ (setLR ? PPC::BCCTRL : PPC::BCCTR)));
+ MachineInstrBuilder(*MI->getParent()->getParent(), MI)
+ .addReg(Pred[1].getReg());
+ return true;
+ } else if (Pred[0].getImm() == PPC::PRED_BIT_UNSET) {
+ MI->setDesc(get(isPPC64 ? (setLR ? PPC::BCCTRL8n : PPC::BCCTR8n) :
+ (setLR ? PPC::BCCTRLn : PPC::BCCTRn)));
+ MachineInstrBuilder(*MI->getParent()->getParent(), MI)
+ .addReg(Pred[1].getReg());
+ return true;
+ }
+
+ MI->setDesc(get(isPPC64 ? (setLR ? PPC::BCCCTRL8 : PPC::BCCCTR8) :
+ (setLR ? PPC::BCCCTRL : PPC::BCCCTR)));
MachineInstrBuilder(*MI->getParent()->getParent(), MI)
.addImm(Pred[0].getImm())
.addReg(Pred[1].getReg());
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");
return false;
case PPC::B:
case PPC::BLR:
+ case PPC::BLR8:
case PPC::BCTR:
case PPC::BCTR8:
case PPC::BCTRL:
// for equality checks (as those don't depend on the sign). On PPC64,
// we are restricted to equality for unsigned 64-bit comparisons and for
// signed 32-bit comparisons the applicability is more restricted.
- bool isPPC64 = TM.getSubtargetImpl()->isPPC64();
+ bool isPPC64 = Subtarget.isPPC64();
bool is32BitSignedCompare = OpC == PPC::CMPWI || OpC == PPC::CMPW;
bool is32BitUnsignedCompare = OpC == PPC::CMPLWI || OpC == PPC::CMPLW;
bool is64BitUnsignedCompare = OpC == PPC::CMPLDI || OpC == PPC::CMPLD;
if (equalityOnly) {
// We need to check the uses of the condition register in order to reject
// non-equality comparisons.
- for (MachineRegisterInfo::use_iterator I = MRI->use_begin(CRReg),
- IE = MRI->use_end(); I != IE; ++I) {
+ for (MachineRegisterInfo::use_instr_iterator I =MRI->use_instr_begin(CRReg),
+ IE = MRI->use_instr_end(); I != IE; ++I) {
MachineInstr *UseMI = &*I;
if (UseMI->getOpcode() == PPC::BCC) {
unsigned Pred = UseMI->getOperand(0).getImm();
for (MachineBasicBlock::iterator EL = CmpInstr->getParent()->end();
I != EL; ++I) {
bool FoundUse = false;
- for (MachineRegisterInfo::use_iterator J = MRI->use_begin(CRReg),
- JE = MRI->use_end(); J != JE; ++J)
+ for (MachineRegisterInfo::use_instr_iterator J =MRI->use_instr_begin(CRReg),
+ JE = MRI->use_instr_end(); J != JE; ++J)
if (&*J == &*I) {
FoundUse = true;
break;
// There are two possible candidates which can be changed to set CR[01].
// One is MI, the other is a SUB instruction.
// For CMPrr(r1,r2), we are looking for SUB(r1,r2) or SUB(r2,r1).
- MachineInstr *Sub = NULL;
+ MachineInstr *Sub = nullptr;
if (SrcReg2 != 0)
// MI is not a candidate for CMPrr.
- MI = NULL;
+ MI = nullptr;
// FIXME: Conservatively refuse to convert an instruction which isn't in the
// same BB as the comparison. This is to allow the check below to avoid calls
// (and other explicit clobbers); instead we should really check for these
}
if (ShouldSwap)
- for (MachineRegisterInfo::use_iterator I = MRI->use_begin(CRReg),
- IE = MRI->use_end(); I != IE; ++I) {
+ for (MachineRegisterInfo::use_instr_iterator
+ I = MRI->use_instr_begin(CRReg), IE = MRI->use_instr_end();
+ I != IE; ++I) {
MachineInstr *UseMI = &*I;
if (UseMI->getOpcode() == PPC::BCC) {
PPC::Predicate Pred = (PPC::Predicate) UseMI->getOperand(0).getImm();
assert((!equalityOnly ||
Pred == PPC::PRED_EQ || Pred == PPC::PRED_NE) &&
"Invalid predicate for equality-only optimization");
- PredsToUpdate.push_back(std::make_pair(&((*I).getOperand(0)),
+ PredsToUpdate.push_back(std::make_pair(&(UseMI->getOperand(0)),
PPC::getSwappedPredicate(Pred)));
} else if (UseMI->getOpcode() == PPC::ISEL ||
UseMI->getOpcode() == PPC::ISEL8) {
else if (NewSubReg == PPC::sub_gt)
NewSubReg = PPC::sub_lt;
- SubRegsToUpdate.push_back(std::make_pair(&((*I).getOperand(3)),
+ SubRegsToUpdate.push_back(std::make_pair(&(UseMI->getOperand(3)),
NewSubReg));
} else // We need to abort on a user we don't understand.
return false;
CmpInstr->eraseFromParent();
MachineBasicBlock::iterator MII = MI;
- BuildMI(*MI->getParent(), llvm::next(MII), MI->getDebugLoc(),
+ BuildMI(*MI->getParent(), std::next(MII), MI->getDebugLoc(),
get(TargetOpcode::COPY), CRReg)
.addReg(PPC::CR0, MIOpC != NewOpC ? RegState::Kill : 0);
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(),
const MachineFunction *MF = MI->getParent()->getParent();
const char *AsmStr = MI->getOperand(0).getSymbolName();
return getInlineAsmLength(AsmStr, *MF->getTarget().getMCAsmInfo());
+ } else if (Opcode == TargetOpcode::STACKMAP) {
+ return MI->getOperand(1).getImm();
+ } else if (Opcode == TargetOpcode::PATCHPOINT) {
+ PatchPointOpers Opers(MI);
+ return Opers.getMetaOper(PatchPointOpers::NBytesPos).getImm();
} else {
const MCInstrDesc &Desc = get(Opcode);
return Desc.getSize();
}
}
-#undef DEBUG_TYPE
-#define DEBUG_TYPE "ppc-early-ret"
-STATISTIC(NumBCLR, "Number of early conditional returns");
-STATISTIC(NumBLR, "Number of early returns");
-
-namespace llvm {
- void initializePPCEarlyReturnPass(PassRegistry&);
+std::pair<unsigned, unsigned>
+PPCInstrInfo::decomposeMachineOperandsTargetFlags(unsigned TF) const {
+ const unsigned Mask = PPCII::MO_ACCESS_MASK;
+ return std::make_pair(TF & Mask, TF & ~Mask);
}
-namespace {
- // PPCEarlyReturn pass - For simple functions without epilogue code, move
- // returns up, and create conditional returns, to avoid unnecessary
- // branch-to-blr sequences.
- struct PPCEarlyReturn : public MachineFunctionPass {
- static char ID;
- PPCEarlyReturn() : MachineFunctionPass(ID) {
- initializePPCEarlyReturnPass(*PassRegistry::getPassRegistry());
- }
-
- const PPCTargetMachine *TM;
- const PPCInstrInfo *TII;
-
-protected:
- bool processBlock(MachineBasicBlock &ReturnMBB) {
- bool Changed = false;
-
- MachineBasicBlock::iterator I = ReturnMBB.begin();
- I = ReturnMBB.SkipPHIsAndLabels(I);
-
- // The block must be essentially empty except for the blr.
- if (I == ReturnMBB.end() || I->getOpcode() != PPC::BLR ||
- I != ReturnMBB.getLastNonDebugInstr())
- return Changed;
-
- SmallVector<MachineBasicBlock*, 8> PredToRemove;
- for (MachineBasicBlock::pred_iterator PI = ReturnMBB.pred_begin(),
- PIE = ReturnMBB.pred_end(); PI != PIE; ++PI) {
- bool OtherReference = false, BlockChanged = false;
- for (MachineBasicBlock::iterator J = (*PI)->getLastNonDebugInstr();;) {
- if (J->getOpcode() == PPC::B) {
- if (J->getOperand(0).getMBB() == &ReturnMBB) {
- // This is an unconditional branch to the return. Replace the
- // branch with a blr.
- BuildMI(**PI, J, J->getDebugLoc(), TII->get(PPC::BLR));
- MachineBasicBlock::iterator K = J--;
- K->eraseFromParent();
- BlockChanged = true;
- ++NumBLR;
- continue;
- }
- } else if (J->getOpcode() == PPC::BCC) {
- if (J->getOperand(2).getMBB() == &ReturnMBB) {
- // This is a conditional branch to the return. Replace the branch
- // with a bclr.
- BuildMI(**PI, J, J->getDebugLoc(), TII->get(PPC::BCLR))
- .addImm(J->getOperand(0).getImm())
- .addReg(J->getOperand(1).getReg());
- MachineBasicBlock::iterator K = J--;
- K->eraseFromParent();
- BlockChanged = true;
- ++NumBCLR;
- continue;
- }
- } else if (J->isBranch()) {
- if (J->isIndirectBranch()) {
- if (ReturnMBB.hasAddressTaken())
- OtherReference = true;
- } else
- for (unsigned i = 0; i < J->getNumOperands(); ++i)
- if (J->getOperand(i).isMBB() &&
- J->getOperand(i).getMBB() == &ReturnMBB)
- OtherReference = true;
- } else if (!J->isTerminator() && !J->isDebugValue())
- break;
-
- if (J == (*PI)->begin())
- break;
-
- --J;
- }
-
- if ((*PI)->canFallThrough() && (*PI)->isLayoutSuccessor(&ReturnMBB))
- OtherReference = true;
-
- // Predecessors are stored in a vector and can't be removed here.
- if (!OtherReference && BlockChanged) {
- PredToRemove.push_back(*PI);
- }
-
- if (BlockChanged)
- Changed = true;
- }
-
- for (unsigned i = 0, ie = PredToRemove.size(); i != ie; ++i)
- PredToRemove[i]->removeSuccessor(&ReturnMBB);
-
- if (Changed && !ReturnMBB.hasAddressTaken()) {
- // We now might be able to merge this blr-only block into its
- // by-layout predecessor.
- if (ReturnMBB.pred_size() == 1 &&
- (*ReturnMBB.pred_begin())->isLayoutSuccessor(&ReturnMBB)) {
- // Move the blr into the preceding block.
- MachineBasicBlock &PrevMBB = **ReturnMBB.pred_begin();
- PrevMBB.splice(PrevMBB.end(), &ReturnMBB, I);
- PrevMBB.removeSuccessor(&ReturnMBB);
- }
-
- if (ReturnMBB.pred_empty())
- ReturnMBB.eraseFromParent();
- }
-
- return Changed;
- }
-
-public:
- virtual bool runOnMachineFunction(MachineFunction &MF) {
- TM = static_cast<const PPCTargetMachine *>(&MF.getTarget());
- TII = TM->getInstrInfo();
-
- bool Changed = false;
-
- // If the function does not have at least two blocks, then there is
- // nothing to do.
- if (MF.size() < 2)
- return Changed;
-
- for (MachineFunction::iterator I = MF.begin(); I != MF.end();) {
- MachineBasicBlock &B = *I++;
- if (processBlock(B))
- Changed = true;
- }
-
- return Changed;
- }
-
- virtual void getAnalysisUsage(AnalysisUsage &AU) const {
- MachineFunctionPass::getAnalysisUsage(AU);
- }
- };
+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);
}
-INITIALIZE_PASS(PPCEarlyReturn, DEBUG_TYPE,
- "PowerPC Early-Return Creation", false, false)
+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);
+}
-char PPCEarlyReturn::ID = 0;
-FunctionPass*
-llvm::createPPCEarlyReturnPass() { return new PPCEarlyReturn(); }