OSchip
/
llvm-project
mirror of https://github.com/THU-DSP-LAB/llvm-project.git
13
0
Fork 6
Code Issues Packages Projects Releases Wiki Activity
llvm-project/llvm/lib/Target/PowerPC/PPCInstrInfo.cpp

2023 lines
76 KiB
C++
Raw Blame History

//===-- PPCInstrInfo.cpp - PowerPC Instruction Information ----------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file contains the PowerPC implementation of the TargetInstrInfo class.
//
//===----------------------------------------------------------------------===//
#include "PPCInstrInfo.h"
#include "MCTargetDesc/PPCPredicates.h"
#include "PPC.h"
#include "PPCHazardRecognizers.h"
#include "PPCInstrBuilder.h"
#include "PPCMachineFunctionInfo.h"
#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"
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(PPCSubtarget &STI)
: PPCGenInstrInfo(PPC::ADJCALLSTACKDOWN, PPC::ADJCALLSTACKUP,
/* CatchRetOpcode */ -1,
STI.isPPC64() ? PPC::BLR8 : PPC::BLR),
Subtarget(STI), RI(STI.getTargetMachine()) {}
/// CreateTargetHazardRecognizer - Return the hazard recognizer to use for
/// this target when scheduling the DAG.
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 =
static_cast<const PPCSubtarget *>(STI)->getInstrItineraryData();
return new ScoreboardHazardRecognizer(II, 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 =
DAG->MF.getSubtarget<PPCSubtarget>().getDarwinDirective();
// FIXME: Leaving this as-is until we have POWER9 scheduling info
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(DAG->TII && "No InstrInfo?");
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,
const MachineInstr &UseMI,
unsigned UseIdx) const {
int Latency = PPCGenInstrInfo::getOperandLatency(ItinData, DefMI, DefIdx,
UseMI, UseIdx);
if (!DefMI.getParent())
return Latency;
const MachineOperand &DefMO = DefMI.getOperand(DefIdx);
unsigned Reg = DefMO.getReg();
bool IsRegCR;
if (TargetRegisterInfo::isVirtualRegister(Reg)) {
const MachineRegisterInfo *MRI =
&DefMI.getParent()->getParent()->getRegInfo();
IsRegCR = MRI->getRegClass(Reg)->hasSuperClassEq(&PPC::CRRCRegClass) ||
MRI->getRegClass(Reg)->hasSuperClassEq(&PPC::CRBITRCRegClass);
} else {
IsRegCR = PPC::CRRCRegClass.contains(Reg) ||
PPC::CRBITRCRegClass.contains(Reg);
}
if (UseMI.isBranch() && IsRegCR) {
if (Latency < 0)
Latency = getInstrLatency(ItinData, DefMI);
// On some cores, there is an additional delay between writing to a condition
// register, and using it from a branch.
unsigned Directive = Subtarget.getDarwinDirective();
switch (Directive) {
default: break;
case PPC::DIR_7400:
case PPC::DIR_750:
case PPC::DIR_970:
case PPC::DIR_E5500:
case PPC::DIR_PWR4:
case PPC::DIR_PWR5:
case PPC::DIR_PWR5X:
case PPC::DIR_PWR6:
case PPC::DIR_PWR6X:
case PPC::DIR_PWR7:
case PPC::DIR_PWR8:
// FIXME: Is this needed for POWER9?
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,
unsigned &SubIdx) const {
switch (MI.getOpcode()) {
default: return false;
case PPC::EXTSW:
case PPC::EXTSW_32_64:
SrcReg = MI.getOperand(1).getReg();
DstReg = MI.getOperand(0).getReg();
SubIdx = PPC::sub_32;
return true;
}
}
unsigned PPCInstrInfo::isLoadFromStackSlot(const MachineInstr &MI,
int &FrameIndex) const {
// Note: This list must be kept consistent with LoadRegFromStackSlot.
switch (MI.getOpcode()) {
default: break;
case PPC::LD:
case PPC::LWZ:
case PPC::LFS:
case PPC::LFD:
case PPC::RESTORE_CR:
case PPC::RESTORE_CRBIT:
case PPC::LVX:
case PPC::LXVD2X:
case PPC::LXVX:
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).
if (MI.getOperand(1).isImm() && !MI.getOperand(1).getImm() &&
MI.getOperand(2).isFI()) {
FrameIndex = MI.getOperand(2).getIndex();
return MI.getOperand(0).getReg();
}
break;
}
return 0;
}
// For opcodes with the ReMaterializable flag set, this function is called to
// verify the instruction is really rematable.
bool PPCInstrInfo::isReallyTriviallyReMaterializable(const MachineInstr &MI,
AliasAnalysis *AA) const {
switch (MI.getOpcode()) {
default:
// This function should only be called for opcodes with the ReMaterializable
// flag set.
llvm_unreachable("Unknown rematerializable operation!");
break;
case PPC::LI:
case PPC::LI8:
case PPC::LIS:
case PPC::LIS8:
case PPC::QVGPCI:
case PPC::ADDIStocHA:
case PPC::ADDItocL:
case PPC::LOAD_STACK_GUARD:
return true;
}
return false;
}
unsigned PPCInstrInfo::isStoreToStackSlot(const MachineInstr &MI,
int &FrameIndex) const {
// Note: This list must be kept consistent with StoreRegToStackSlot.
switch (MI.getOpcode()) {
default: break;
case PPC::STD:
case PPC::STW:
case PPC::STFS:
case PPC::STFD:
case PPC::SPILL_CR:
case PPC::SPILL_CRBIT:
case PPC::STVX:
case PPC::STXVD2X:
case PPC::STXVX:
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).
if (MI.getOperand(1).isImm() && !MI.getOperand(1).getImm() &&
MI.getOperand(2).isFI()) {
FrameIndex = MI.getOperand(2).getIndex();
return MI.getOperand(0).getReg();
}
break;
}
return 0;
}
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::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 nullptr;
// If we have a zero rotate count, we have:
// M = mask(MB,ME)
// Op0 = (Op1 & ~M) | (Op2 & M)
// Change this to:
// M = mask((ME+1)&31, (MB-1)&31)
// 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 SubReg1 = MI.getOperand(1).getSubReg();
unsigned SubReg2 = MI.getOperand(2).getSubReg();
bool Reg1IsKill = MI.getOperand(1).isKill();
bool Reg2IsKill = MI.getOperand(2).isKill();
bool ChangeReg0 = false;
// If machine instrs are no longer in two-address forms, update
// destination register as well.
if (Reg0 == Reg1) {
// Must be two address instruction!
assert(MI.getDesc().getOperandConstraint(0, MCOI::TIED_TO) &&
"Expecting a two-address instruction!");
assert(MI.getOperand(0).getSubReg() == SubReg1 && "Tied subreg mismatch");
Reg2IsKill = false;
ChangeReg0 = true;
}
// Masks.
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();
bool Reg0IsDead = MI.getOperand(0).isDead();
return BuildMI(MF, MI.getDebugLoc(), MI.getDesc())
.addReg(Reg0, RegState::Define | getDeadRegState(Reg0IsDead))
.addReg(Reg2, getKillRegState(Reg2IsKill))
.addReg(Reg1, getKillRegState(Reg1IsKill))
.addImm((ME + 1) & 31)
.addImm((MB - 1) & 31);
}
if (ChangeReg0) {
MI.getOperand(0).setReg(Reg2);
MI.getOperand(0).setSubReg(SubReg2);
}
MI.getOperand(2).setReg(Reg1);
MI.getOperand(1).setReg(Reg2);
MI.getOperand(2).setSubReg(SubReg1);
MI.getOperand(1).setSubReg(SubReg2);
MI.getOperand(2).setIsKill(Reg1IsKill);
MI.getOperand(1).setIsKill(Reg2IsKill);
// Swap the mask around.
MI.getOperand(4).setImm((ME + 1) & 31);
MI.getOperand(5).setImm((MB - 1) & 31);
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 = 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 */
// FIXME: Update when POWER9 scheduling model is ready.
case PPC::DIR_PWR9: Opcode = PPC::NOP_GT_PWR7; break;
}
DebugLoc DL;
BuildMI(MBB, MI, DL, get(Opcode));
}
/// Return the noop instruction to use for a noop.
void PPCInstrInfo::getNoop(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.
bool PPCInstrInfo::analyzeBranch(MachineBasicBlock &MBB,
MachineBasicBlock *&TBB,
MachineBasicBlock *&FBB,
SmallVectorImpl<MachineOperand> &Cond,
bool AllowModify) const {
bool isPPC64 = Subtarget.isPPC64();
// If the block has no terminators, it just falls into the block after it.
MachineBasicBlock::iterator I = MBB.getLastNonDebugInstr();
if (I == MBB.end())
return false;
if (!isUnpredicatedTerminator(*I))
return false;
// Get the last instruction in the block.
MachineInstr &LastInst = *I;
// If there is only one terminator instruction, process it.
if (I == MBB.begin() || !isUnpredicatedTerminator(*--I)) {
if (LastInst.getOpcode() == PPC::B) {
if (!LastInst.getOperand(0).isMBB())
return true;
TBB = LastInst.getOperand(0).getMBB();
return false;
} else if (LastInst.getOpcode() == PPC::BCC) {
if (!LastInst.getOperand(2).isMBB())
return true;
// Block ends with fall-through condbranch.
TBB = LastInst.getOperand(2).getMBB();
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())
return true;
if (DisableCTRLoopAnal)
return true;
TBB = LastInst.getOperand(0).getMBB();
Cond.push_back(MachineOperand::CreateImm(1));
Cond.push_back(MachineOperand::CreateReg(isPPC64 ? PPC::CTR8 : PPC::CTR,
true));
return false;
} else if (LastInst.getOpcode() == PPC::BDZ8 ||
LastInst.getOpcode() == PPC::BDZ) {
if (!LastInst.getOperand(0).isMBB())
return true;
if (DisableCTRLoopAnal)
return true;
TBB = LastInst.getOperand(0).getMBB();
Cond.push_back(MachineOperand::CreateImm(0));
Cond.push_back(MachineOperand::CreateReg(isPPC64 ? PPC::CTR8 : PPC::CTR,
true));
return false;
}
// Otherwise, don't know what this is.
return true;
}
// Get the instruction before it if it's a terminator.
MachineInstr &SecondLastInst = *I;
// If there are three terminators, we don't know what sort of block this is.
if (I != MBB.begin() && isUnpredicatedTerminator(*--I))
return true;
// If the block ends with PPC::B and PPC:BCC, handle it.
if (SecondLastInst.getOpcode() == PPC::BCC &&
LastInst.getOpcode() == PPC::B) {
if (!SecondLastInst.getOperand(2).isMBB() ||
!LastInst.getOperand(0).isMBB())
return true;
TBB = SecondLastInst.getOperand(2).getMBB();
Cond.push_back(SecondLastInst.getOperand(0));
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) {
if (!SecondLastInst.getOperand(0).isMBB() ||
!LastInst.getOperand(0).isMBB())
return true;
if (DisableCTRLoopAnal)
return true;
TBB = SecondLastInst.getOperand(0).getMBB();
Cond.push_back(MachineOperand::CreateImm(1));
Cond.push_back(MachineOperand::CreateReg(isPPC64 ? PPC::CTR8 : PPC::CTR,
true));
FBB = LastInst.getOperand(0).getMBB();
return false;
} else if ((SecondLastInst.getOpcode() == PPC::BDZ8 ||
SecondLastInst.getOpcode() == PPC::BDZ) &&
LastInst.getOpcode() == PPC::B) {
if (!SecondLastInst.getOperand(0).isMBB() ||
!LastInst.getOperand(0).isMBB())
return true;
if (DisableCTRLoopAnal)
return true;
TBB = SecondLastInst.getOperand(0).getMBB();
Cond.push_back(MachineOperand::CreateImm(0));
Cond.push_back(MachineOperand::CreateReg(isPPC64 ? PPC::CTR8 : PPC::CTR,
true));
FBB = LastInst.getOperand(0).getMBB();
return false;
}
// If the block ends with two PPC:Bs, handle it. The second one is not
// executed, so remove it.
if (SecondLastInst.getOpcode() == PPC::B && LastInst.getOpcode() == PPC::B) {
if (!SecondLastInst.getOperand(0).isMBB())
return true;
TBB = SecondLastInst.getOperand(0).getMBB();
I = LastInst;
if (AllowModify)
I->eraseFromParent();
return false;
}
// Otherwise, can't handle this.
return true;
}
unsigned PPCInstrInfo::removeBranch(MachineBasicBlock &MBB,
int *BytesRemoved) const {
assert(!BytesRemoved && "code size not handled");
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;
// Remove the branch.
I->eraseFromParent();
I = MBB.end();
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;
// Remove the branch.
I->eraseFromParent();
return 2;
}
unsigned PPCInstrInfo::insertBranch(MachineBasicBlock &MBB,
MachineBasicBlock *TBB,
MachineBasicBlock *FBB,
ArrayRef<MachineOperand> Cond,
const DebugLoc &DL,
int *BytesAdded) 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!");
assert(!BytesAdded && "code size not handled");
bool isPPC64 = Subtarget.isPPC64();
// One-way branch.
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)).add(Cond[1]).addMBB(TBB);
else if (Cond[0].getImm() == PPC::PRED_BIT_UNSET)
BuildMI(&MBB, DL, get(PPC::BCn)).add(Cond[1]).addMBB(TBB);
else // Conditional branch
BuildMI(&MBB, DL, get(PPC::BCC))
.addImm(Cond[0].getImm())
.add(Cond[1])
.addMBB(TBB);
return 1;
}
// Two-way Conditional Branch.
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)).add(Cond[1]).addMBB(TBB);
else if (Cond[0].getImm() == PPC::PRED_BIT_UNSET)
BuildMI(&MBB, DL, get(PPC::BCn)).add(Cond[1]).addMBB(TBB);
else
BuildMI(&MBB, DL, get(PPC::BCC))
.addImm(Cond[0].getImm())
.add(Cond[1])
.addMBB(TBB);
BuildMI(&MBB, DL, get(PPC::B)).addMBB(FBB);
return 2;
}
// Select analysis.
bool PPCInstrInfo::canInsertSelect(const MachineBasicBlock &MBB,
ArrayRef<MachineOperand> Cond,
unsigned TrueReg, unsigned FalseReg,
int &CondCycles, int &TrueCycles, int &FalseCycles) const {
if (Cond.size() != 2)
return false;
// If this is really a bdnz-like condition, then it cannot be turned into a
// select.
if (Cond[1].getReg() == PPC::CTR || Cond[1].getReg() == PPC::CTR8)
return false;
// Check register classes.
const MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
const TargetRegisterClass *RC =
RI.getCommonSubClass(MRI.getRegClass(TrueReg), MRI.getRegClass(FalseReg));
if (!RC)
return false;
// isel is for regular integer GPRs only.
if (!PPC::GPRCRegClass.hasSubClassEq(RC) &&
!PPC::GPRC_NOR0RegClass.hasSubClassEq(RC) &&
!PPC::G8RCRegClass.hasSubClassEq(RC) &&
!PPC::G8RC_NOX0RegClass.hasSubClassEq(RC))
return false;
// FIXME: These numbers are for the A2, how well they work for other cores is
// an open question. On the A2, the isel instruction has a 2-cycle latency
// but single-cycle throughput. These numbers are used in combination with
// the MispredictPenalty setting from the active SchedMachineModel.
CondCycles = 1;
TrueCycles = 1;
FalseCycles = 1;
return true;
}
void PPCInstrInfo::insertSelect(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
const DebugLoc &dl, unsigned DestReg,
ArrayRef<MachineOperand> Cond, unsigned TrueReg,
unsigned FalseReg) const {
assert(Cond.size() == 2 &&
"PPC branch conditions have two components!");
// Get the register classes.
MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
const TargetRegisterClass *RC =
RI.getCommonSubClass(MRI.getRegClass(TrueReg), MRI.getRegClass(FalseReg));
assert(RC && "TrueReg and FalseReg must have overlapping register classes");
bool Is64Bit = PPC::G8RCRegClass.hasSubClassEq(RC) ||
PPC::G8RC_NOX0RegClass.hasSubClassEq(RC);
assert((Is64Bit ||
PPC::GPRCRegClass.hasSubClassEq(RC) ||
PPC::GPRC_NOR0RegClass.hasSubClassEq(RC)) &&
"isel is for regular integer GPRs only");
unsigned OpCode = Is64Bit ? PPC::ISEL8 : PPC::ISEL;
auto SelectPred = static_cast<PPC::Predicate>(Cond[0].getImm());
unsigned SubIdx = 0;
bool SwapOps = false;
switch (SelectPred) {
case PPC::PRED_EQ:
case PPC::PRED_EQ_MINUS:
case PPC::PRED_EQ_PLUS:
SubIdx = PPC::sub_eq; SwapOps = false; break;
case PPC::PRED_NE:
case PPC::PRED_NE_MINUS:
case PPC::PRED_NE_PLUS:
SubIdx = PPC::sub_eq; SwapOps = true; break;
case PPC::PRED_LT:
case PPC::PRED_LT_MINUS:
case PPC::PRED_LT_PLUS:
SubIdx = PPC::sub_lt; SwapOps = false; break;
case PPC::PRED_GE:
case PPC::PRED_GE_MINUS:
case PPC::PRED_GE_PLUS:
SubIdx = PPC::sub_lt; SwapOps = true; break;
case PPC::PRED_GT:
case PPC::PRED_GT_MINUS:
case PPC::PRED_GT_PLUS:
SubIdx = PPC::sub_gt; SwapOps = false; break;
case PPC::PRED_LE:
case PPC::PRED_LE_MINUS:
case PPC::PRED_LE_PLUS:
SubIdx = PPC::sub_gt; SwapOps = true; break;
case PPC::PRED_UN:
case PPC::PRED_UN_MINUS:
case PPC::PRED_UN_PLUS:
SubIdx = PPC::sub_un; SwapOps = false; break;
case PPC::PRED_NU:
case PPC::PRED_NU_MINUS:
case PPC::PRED_NU_PLUS:
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,
SecondReg = SwapOps ? TrueReg : FalseReg;
// The first input register of isel cannot be r0. If it is a member
// of a register class that can be r0, then copy it first (the
// register allocator should eliminate the copy).
if (MRI.getRegClass(FirstReg)->contains(PPC::R0) ||
MRI.getRegClass(FirstReg)->contains(PPC::X0)) {
const TargetRegisterClass *FirstRC =
MRI.getRegClass(FirstReg)->contains(PPC::X0) ?
&PPC::G8RC_NOX0RegClass : &PPC::GPRC_NOR0RegClass;
unsigned OldFirstReg = FirstReg;
FirstReg = MRI.createVirtualRegister(FirstRC);
BuildMI(MBB, MI, dl, get(TargetOpcode::COPY), FirstReg)
.addReg(OldFirstReg);
}
BuildMI(MBB, MI, dl, get(OpCode), DestReg)
.addReg(FirstReg).addReg(SecondReg)
.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,
const 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::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;
}
// 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;
else if (PPC::G8RCRegClass.contains(DestReg, SrcReg))
Opc = PPC::OR8;
else if (PPC::F4RCRegClass.contains(DestReg, SrcReg))
Opc = PPC::FMR;
else if (PPC::CRRCRegClass.contains(DestReg, SrcReg))
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
llvm_unreachable("Impossible reg-to-reg copy");
const MCInstrDesc &MCID = get(Opc);
if (MCID.getNumOperands() == 3)
BuildMI(MBB, I, DL, MCID, DestReg)
.addReg(SrcReg).addReg(SrcReg, getKillRegState(KillSrc));
else
BuildMI(MBB, I, DL, MCID, DestReg).addReg(SrcReg, getKillRegState(KillSrc));
}
// This function returns true if a CR spill is necessary and false otherwise.
bool
PPCInstrInfo::StoreRegToStackSlot(MachineFunction &MF,
unsigned SrcReg, bool isKill,
int FrameIdx,
const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr*> &NewMIs,
bool &NonRI, bool &SpillsVRS) const{
// Note: If additional store instructions are added here,
// update isStoreToStackSlot.
DebugLoc DL;
if (PPC::GPRCRegClass.hasSubClassEq(RC) ||
PPC::GPRC_NOR0RegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STW))
.addReg(SrcReg,
getKillRegState(isKill)),
FrameIdx));
} else if (PPC::G8RCRegClass.hasSubClassEq(RC) ||
PPC::G8RC_NOX0RegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STD))
.addReg(SrcReg,
getKillRegState(isKill)),
FrameIdx));
} else if (PPC::F8RCRegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STFD))
.addReg(SrcReg,
getKillRegState(isKill)),
FrameIdx));
} else if (PPC::F4RCRegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STFS))
.addReg(SrcReg,
getKillRegState(isKill)),
FrameIdx));
} else if (PPC::CRRCRegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::SPILL_CR))
.addReg(SrcReg,
getKillRegState(isKill)),
FrameIdx));
return true;
} else if (PPC::CRBITRCRegClass.hasSubClassEq(RC)) {
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)) {
unsigned Op = Subtarget.hasP9Vector() ? PPC::STXVX : PPC::STXVD2X;
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(Op))
.addReg(SrcReg,
getKillRegState(isKill)),
FrameIdx));
NonRI = true;
} else if (PPC::VSFRCRegClass.hasSubClassEq(RC)) {
unsigned Opc = Subtarget.hasP9Vector() ? PPC::DFSTOREf64 : PPC::STXSDX;
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(Opc))
.addReg(SrcReg,
getKillRegState(isKill)),
FrameIdx));
NonRI = true;
} else if (PPC::VSSRCRegClass.hasSubClassEq(RC)) {
unsigned Opc = Subtarget.hasP9Vector() ? PPC::DFSTOREf32 : PPC::STXSSPX;
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(Opc))
.addReg(SrcReg,
getKillRegState(isKill)),
FrameIdx));
NonRI = true;
} else if (PPC::VRSAVERCRegClass.hasSubClassEq(RC)) {
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!");
}
return false;
}
void
PPCInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
unsigned SrcReg, bool isKill, int FrameIdx,
const TargetRegisterClass *RC,
const TargetRegisterInfo *TRI) const {
MachineFunction &MF = *MBB.getParent();
SmallVector<MachineInstr*, 4> NewMIs;
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
FuncInfo->setHasSpills();
// We need to avoid a situation in which the value from a VRRC register is
// spilled using an Altivec instruction and reloaded into a VSRC register
// using a VSX instruction. The issue with this is that the VSX
// load/store instructions swap the doublewords in the vector and the Altivec
// ones don't. The register classes on the spill/reload may be different if
// the register is defined using an Altivec instruction and is then used by a
// VSX instruction.
RC = updatedRC(RC);
bool NonRI = false, SpillsVRS = false;
if (StoreRegToStackSlot(MF, SrcReg, isKill, FrameIdx, RC, NewMIs,
NonRI, SpillsVRS))
FuncInfo->setSpillsCR();
if (SpillsVRS)
FuncInfo->setSpillsVRSAVE();
if (NonRI)
FuncInfo->setHasNonRISpills();
for (unsigned i = 0, e = NewMIs.size(); i != e; ++i)
MBB.insert(MI, NewMIs[i]);
const MachineFrameInfo &MFI = MF.getFrameInfo();
MachineMemOperand *MMO = MF.getMachineMemOperand(
MachinePointerInfo::getFixedStack(MF, FrameIdx),
MachineMemOperand::MOStore, MFI.getObjectSize(FrameIdx),
MFI.getObjectAlignment(FrameIdx));
NewMIs.back()->addMemOperand(MF, MMO);
}
bool PPCInstrInfo::LoadRegFromStackSlot(MachineFunction &MF, const DebugLoc &DL,
unsigned DestReg, int FrameIdx,
const TargetRegisterClass *RC,
SmallVectorImpl<MachineInstr *> &NewMIs,
bool &NonRI, bool &SpillsVRS) const {
// Note: If additional load instructions are added here,
// update isLoadFromStackSlot.
if (PPC::GPRCRegClass.hasSubClassEq(RC) ||
PPC::GPRC_NOR0RegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LWZ),
DestReg), FrameIdx));
} else if (PPC::G8RCRegClass.hasSubClassEq(RC) ||
PPC::G8RC_NOX0RegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LD), DestReg),
FrameIdx));
} else if (PPC::F8RCRegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LFD), DestReg),
FrameIdx));
} else if (PPC::F4RCRegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LFS), DestReg),
FrameIdx));
} else if (PPC::CRRCRegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL,
get(PPC::RESTORE_CR), DestReg),
FrameIdx));
return true;
} else if (PPC::CRBITRCRegClass.hasSubClassEq(RC)) {
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)) {
unsigned Op = Subtarget.hasP9Vector() ? PPC::LXVX : PPC::LXVD2X;
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(Op), DestReg),
FrameIdx));
NonRI = true;
} else if (PPC::VSFRCRegClass.hasSubClassEq(RC)) {
unsigned Opc = Subtarget.hasP9Vector() ? PPC::DFLOADf64 : PPC::LXSDX;
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(Opc),
DestReg), FrameIdx));
NonRI = true;
} else if (PPC::VSSRCRegClass.hasSubClassEq(RC)) {
unsigned Opc = Subtarget.hasP9Vector() ? PPC::DFLOADf32 : PPC::LXSSPX;
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(Opc),
DestReg), FrameIdx));
NonRI = true;
} else if (PPC::VRSAVERCRegClass.hasSubClassEq(RC)) {
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!");
}
return false;
}
void
PPCInstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
unsigned DestReg, int FrameIdx,
const TargetRegisterClass *RC,
const TargetRegisterInfo *TRI) const {
MachineFunction &MF = *MBB.getParent();
SmallVector<MachineInstr*, 4> NewMIs;
DebugLoc DL;
if (MI != MBB.end()) DL = MI->getDebugLoc();
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
FuncInfo->setHasSpills();
// We need to avoid a situation in which the value from a VRRC register is
// spilled using an Altivec instruction and reloaded into a VSRC register
// using a VSX instruction. The issue with this is that the VSX
// load/store instructions swap the doublewords in the vector and the Altivec
// ones don't. The register classes on the spill/reload may be different if
// the register is defined using an Altivec instruction and is then used by a
// VSX instruction.
if (Subtarget.hasVSX() && RC == &PPC::VRRCRegClass)
RC = &PPC::VSRCRegClass;
bool NonRI = false, SpillsVRS = false;
if (LoadRegFromStackSlot(MF, DL, DestReg, FrameIdx, RC, NewMIs,
NonRI, SpillsVRS))
FuncInfo->setSpillsCR();
if (SpillsVRS)
FuncInfo->setSpillsVRSAVE();
if (NonRI)
FuncInfo->setHasNonRISpills();
for (unsigned i = 0, e = NewMIs.size(); i != e; ++i)
MBB.insert(MI, NewMIs[i]);
const MachineFrameInfo &MFI = MF.getFrameInfo();
MachineMemOperand *MMO = MF.getMachineMemOperand(
MachinePointerInfo::getFixedStack(MF, FrameIdx),
MachineMemOperand::MOLoad, MFI.getObjectSize(FrameIdx),
MFI.getObjectAlignment(FrameIdx));
NewMIs.back()->addMemOperand(MF, MMO);
}
bool PPCInstrInfo::
reverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const {
assert(Cond.size() == 2 && "Invalid PPC branch opcode!");
if (Cond[1].getReg() == PPC::CTR8 || Cond[1].getReg() == PPC::CTR)
Cond[0].setImm(Cond[0].getImm() == 0 ? 1 : 0);
else
// Leave the CR# the same, but invert the condition.
Cond[0].setImm(PPC::InvertPredicate((PPC::Predicate)Cond[0].getImm()));
return false;
}
bool PPCInstrInfo::FoldImmediate(MachineInstr &UseMI, MachineInstr &DefMI,
unsigned Reg, MachineRegisterInfo *MRI) const {
// For some instructions, it is legal to fold ZERO into the RA register field.
// A zero immediate should always be loaded with a single li.
unsigned DefOpc = DefMI.getOpcode();
if (DefOpc != PPC::LI && DefOpc != PPC::LI8)
return false;
if (!DefMI.getOperand(1).isImm())
return false;
if (DefMI.getOperand(1).getImm() != 0)
return false;
// Note that we cannot here invert the arguments of an isel in order to fold
// a ZERO into what is presented as the second argument. All we have here
// is the condition bit, and that might come from a CR-logical bit operation.
const MCInstrDesc &UseMCID = UseMI.getDesc();
// Only fold into real machine instructions.
if (UseMCID.isPseudo())
return false;
unsigned UseIdx;
for (UseIdx = 0; UseIdx < UseMI.getNumOperands(); ++UseIdx)
if (UseMI.getOperand(UseIdx).isReg() &&
UseMI.getOperand(UseIdx).getReg() == Reg)
break;
assert(UseIdx < UseMI.getNumOperands() && "Cannot find Reg in UseMI");
assert(UseIdx < UseMCID.getNumOperands() && "No operand description for Reg");
const MCOperandInfo *UseInfo = &UseMCID.OpInfo[UseIdx];
// We can fold the zero if this register requires a GPRC_NOR0/G8RC_NOX0
// register (which might also be specified as a pointer class kind).
if (UseInfo->isLookupPtrRegClass()) {
if (UseInfo->RegClass /* Kind */ != 1)
return false;
} else {
if (UseInfo->RegClass != PPC::GPRC_NOR0RegClassID &&
UseInfo->RegClass != PPC::G8RC_NOX0RegClassID)
return false;
}
// Make sure this is not tied to an output register (or otherwise
// constrained). This is true for ST?UX registers, for example, which
// are tied to their output registers.
if (UseInfo->Constraints != 0)
return false;
unsigned ZeroReg;
if (UseInfo->isLookupPtrRegClass()) {
bool isPPC64 = Subtarget.isPPC64();
ZeroReg = isPPC64 ? PPC::ZERO8 : PPC::ZERO;
} else {
ZeroReg = UseInfo->RegClass == PPC::G8RC_NOX0RegClassID ?
PPC::ZERO8 : PPC::ZERO;
}
bool DeleteDef = MRI->hasOneNonDBGUse(Reg);
UseMI.getOperand(UseIdx).setReg(ZeroReg);
if (DeleteDef)
DefMI.eraseFromParent();
return true;
}
static bool MBBDefinesCTR(MachineBasicBlock &MBB) {
for (MachineBasicBlock::iterator I = MBB.begin(), IE = MBB.end();
I != IE; ++I)
if (I->definesRegister(PPC::CTR) || I->definesRegister(PPC::CTR8))
return true;
return false;
}
// We should make sure that, if we're going to predicate both sides of a
// condition (a diamond), that both sides don't define the counter register. We
// can predicate counter-decrement-based branches, but while that predicates
// the branching, it does not predicate the counter decrement. If we tried to
// merge the triangle into one predicated block, we'd decrement the counter
// twice.
bool PPCInstrInfo::isProfitableToIfCvt(MachineBasicBlock &TMBB,
unsigned NumT, unsigned ExtraT,
MachineBasicBlock &FMBB,
unsigned NumF, unsigned ExtraF,
BranchProbability Probability) const {
return !(MBBDefinesCTR(TMBB) && MBBDefinesCTR(FMBB));
}
bool PPCInstrInfo::isPredicated(const MachineInstr &MI) const {
// The predicated branches are identified by their type, not really by the
// explicit presence of a predicate. Furthermore, some of them can be
// predicated more than once. Because if conversion won't try to predicate
// any instruction which already claims to be predicated (by returning true
// here), always return false. In doing so, we let isPredicable() be the
// final word on whether not the instruction can be (further) predicated.
return false;
}
bool PPCInstrInfo::isUnpredicatedTerminator(const MachineInstr &MI) const {
if (!MI.isTerminator())
return false;
// Conditional branch is a special case.
if (MI.isBranch() && !MI.isBarrier())
return true;
return !isPredicated(MI);
}
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) {
bool isPPC64 = Subtarget.isPPC64();
MI.setDesc(get(Pred[0].getImm() ? (isPPC64 ? PPC::BDNZLR8 : PPC::BDNZLR)
: (isPPC64 ? PPC::BDZLR8 : PPC::BDZLR)));
} 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 = 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);
MI.setDesc(get(PPC::BCC));
MachineInstrBuilder(*MI.getParent()->getParent(), MI)
.addImm(Pred[0].getImm())
.addReg(Pred[1].getReg())
.addMBB(MBB);
}
return true;
} else if (OpC == PPC::BCTR || OpC == PPC::BCTR8 ||
OpC == PPC::BCTRL || OpC == PPC::BCTRL8) {
if (Pred[1].getReg() == PPC::CTR8 || Pred[1].getReg() == PPC::CTR)
llvm_unreachable("Cannot predicate bctr[l] on the ctr register");
bool setLR = OpC == PPC::BCTRL || OpC == PPC::BCTRL8;
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 true;
}
return false;
}
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");
if (Pred1[1].getReg() == PPC::CTR8 || Pred1[1].getReg() == PPC::CTR)
return false;
if (Pred2[1].getReg() == PPC::CTR8 || Pred2[1].getReg() == PPC::CTR)
return false;
// P1 can only subsume P2 if they test the same condition register.
if (Pred1[1].getReg() != Pred2[1].getReg())
return false;
PPC::Predicate P1 = (PPC::Predicate) Pred1[0].getImm();
PPC::Predicate P2 = (PPC::Predicate) Pred2[0].getImm();
if (P1 == P2)
return true;
// Does P1 subsume P2, e.g. GE subsumes GT.
if (P1 == PPC::PRED_LE &&
(P2 == PPC::PRED_LT || P2 == PPC::PRED_EQ))
return true;
if (P1 == PPC::PRED_GE &&
(P2 == PPC::PRED_GT || P2 == PPC::PRED_EQ))
return true;
return false;
}
bool PPCInstrInfo::DefinesPredicate(MachineInstr &MI,
std::vector<MachineOperand> &Pred) const {
// Note: At the present time, the contents of Pred from this function is
// unused by IfConversion. This implementation follows ARM by pushing the
// CR-defining operand. Because the 'DZ' and 'DNZ' count as types of
// predicate, instructions defining CTR or CTR8 are also included as
// predicate-defining instructions.
const TargetRegisterClass *RCs[] =
{ &PPC::CRRCRegClass, &PPC::CRBITRCRegClass,
&PPC::CTRRCRegClass, &PPC::CTRRC8RegClass };
bool Found = false;
for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
const MachineOperand &MO = MI.getOperand(i);
for (unsigned c = 0; c < array_lengthof(RCs) && !Found; ++c) {
const TargetRegisterClass *RC = RCs[c];
if (MO.isReg()) {
if (MO.isDef() && RC->contains(MO.getReg())) {
Pred.push_back(MO);
Found = true;
}
} else if (MO.isRegMask()) {
for (TargetRegisterClass::iterator I = RC->begin(),
IE = RC->end(); I != IE; ++I)
if (MO.clobbersPhysReg(*I)) {
Pred.push_back(MO);
Found = true;
}
}
}
}
return Found;
}
bool PPCInstrInfo::isPredicable(const MachineInstr &MI) const {
unsigned OpC = MI.getOpcode();
switch (OpC) {
default:
return false;
case PPC::B:
case PPC::BLR:
case PPC::BLR8:
case PPC::BCTR:
case PPC::BCTR8:
case PPC::BCTRL:
case PPC::BCTRL8:
return true;
}
}
bool PPCInstrInfo::analyzeCompare(const MachineInstr &MI, unsigned &SrcReg,
unsigned &SrcReg2, int &Mask,
int &Value) const {
unsigned Opc = MI.getOpcode();
switch (Opc) {
default: return false;
case PPC::CMPWI:
case PPC::CMPLWI:
case PPC::CMPDI:
case PPC::CMPLDI:
SrcReg = MI.getOperand(1).getReg();
SrcReg2 = 0;
Value = MI.getOperand(2).getImm();
Mask = 0xFFFF;
return true;
case PPC::CMPW:
case PPC::CMPLW:
case PPC::CMPD:
case PPC::CMPLD:
case PPC::FCMPUS:
case PPC::FCMPUD:
SrcReg = MI.getOperand(1).getReg();
SrcReg2 = MI.getOperand(2).getReg();
Value = 0;
Mask = 0;
return true;
}
}
bool PPCInstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, unsigned SrcReg,
unsigned SrcReg2, int Mask, int Value,
const MachineRegisterInfo *MRI) const {
if (DisableCmpOpt)
return false;
int OpC = CmpInstr.getOpcode();
unsigned CRReg = CmpInstr.getOperand(0).getReg();
// FP record forms set CR1 based on the execption status bits, not a
// comparison with zero.
if (OpC == PPC::FCMPUS || OpC == PPC::FCMPUD)
return false;
// The record forms set the condition register based on a signed comparison
// with zero (so says the ISA manual). This is not as straightforward as it
// seems, however, because this is always a 64-bit comparison on PPC64, even
// for instructions that are 32-bit in nature (like slw for example).
// So, on PPC32, for unsigned comparisons, we can use the record forms only
// 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 = 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;
// Get the unique definition of SrcReg.
MachineInstr *MI = MRI->getUniqueVRegDef(SrcReg);
if (!MI) return false;
int MIOpC = MI->getOpcode();
bool equalityOnly = false;
bool noSub = false;
if (isPPC64) {
if (is32BitSignedCompare) {
// We can perform this optimization only if MI is sign-extending.
if (MIOpC == PPC::SRAW || MIOpC == PPC::SRAWo ||
MIOpC == PPC::SRAWI || MIOpC == PPC::SRAWIo ||
MIOpC == PPC::EXTSB || MIOpC == PPC::EXTSBo ||
MIOpC == PPC::EXTSH || MIOpC == PPC::EXTSHo ||
MIOpC == PPC::EXTSW || MIOpC == PPC::EXTSWo) {
noSub = true;
} else
return false;
} else if (is32BitUnsignedCompare) {
// 32-bit rotate and mask instructions are zero extending only if MB <= ME
bool isZeroExtendingRotate =
(MIOpC == PPC::RLWINM || MIOpC == PPC::RLWINMo ||
MIOpC == PPC::RLWNM || MIOpC == PPC::RLWNMo)
&& MI->getOperand(3).getImm() <= MI->getOperand(4).getImm();
// We can perform this optimization, equality only, if MI is
// zero-extending.
// FIXME: Other possible target instructions include ANDISo and
// RLWINM aliases, such as ROTRWI, EXTLWI, SLWI and SRWI.
if (MIOpC == PPC::CNTLZW || MIOpC == PPC::CNTLZWo ||
MIOpC == PPC::SLW || MIOpC == PPC::SLWo ||
MIOpC == PPC::SRW || MIOpC == PPC::SRWo ||
MIOpC == PPC::ANDIo ||
isZeroExtendingRotate) {
noSub = true;
equalityOnly = true;
} else
return false;
} else
equalityOnly = is64BitUnsignedCompare;
} else
equalityOnly = is32BitUnsignedCompare;
if (equalityOnly) {
// We need to check the uses of the condition register in order to reject
// non-equality comparisons.
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();
unsigned PredCond = PPC::getPredicateCondition(Pred);
// We ignore hint bits when checking for non-equality comparisons.
if (PredCond != PPC::PRED_EQ && PredCond != PPC::PRED_NE)
return false;
} else if (UseMI->getOpcode() == PPC::ISEL ||
UseMI->getOpcode() == PPC::ISEL8) {
unsigned SubIdx = UseMI->getOperand(3).getSubReg();
if (SubIdx != PPC::sub_eq)
return false;
} else
return false;
}
}
MachineBasicBlock::iterator I = CmpInstr;
// Scan forward to find the first use of the compare.
for (MachineBasicBlock::iterator EL = CmpInstr.getParent()->end(); I != EL;
++I) {
bool FoundUse = false;
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;
}
if (FoundUse)
break;
}
SmallVector<std::pair<MachineOperand*, PPC::Predicate>, 4> PredsToUpdate;
SmallVector<std::pair<MachineOperand*, unsigned>, 4> SubRegsToUpdate;
// 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 = nullptr;
if (SrcReg2 != 0)
// MI is not a candidate for CMPrr.
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
// more explicitly (in at least a few predecessors).
else if (MI->getParent() != CmpInstr.getParent())
return false;
else if (Value != 0) {
// The record-form instructions set CR bit based on signed comparison against 0.
// We try to convert a compare against 1 or -1 into a compare against 0.
bool Success = false;
if (!equalityOnly && MRI->hasOneUse(CRReg)) {
MachineInstr *UseMI = &*MRI->use_instr_begin(CRReg);
if (UseMI->getOpcode() == PPC::BCC) {
PPC::Predicate Pred = (PPC::Predicate)UseMI->getOperand(0).getImm();
unsigned PredCond = PPC::getPredicateCondition(Pred);
unsigned PredHint = PPC::getPredicateHint(Pred);
int16_t Immed = (int16_t)Value;
// When modyfing the condition in the predicate, we propagate hint bits
// from the original predicate to the new one.
if (Immed == -1 && PredCond == PPC::PRED_GT) {
// We convert "greater than -1" into "greater than or equal to 0",
// since we are assuming signed comparison by !equalityOnly
PredsToUpdate.push_back(std::make_pair(&(UseMI->getOperand(0)),
PPC::getPredicate(PPC::PRED_GE, PredHint)));
Success = true;
}
else if (Immed == 1 && PredCond == PPC::PRED_LT) {
// We convert "less than 1" into "less than or equal to 0".
PredsToUpdate.push_back(std::make_pair(&(UseMI->getOperand(0)),
PPC::getPredicate(PPC::PRED_LE, PredHint)));
Success = true;
}
}
}
// PPC does not have a record-form SUBri.
if (!Success)
return false;
}
// Search for Sub.
const TargetRegisterInfo *TRI = &getRegisterInfo();
--I;
// Get ready to iterate backward from CmpInstr.
MachineBasicBlock::iterator E = MI, B = CmpInstr.getParent()->begin();
for (; I != E && !noSub; --I) {
const MachineInstr &Instr = *I;
unsigned IOpC = Instr.getOpcode();
if (&*I != &CmpInstr && (Instr.modifiesRegister(PPC::CR0, TRI) ||
Instr.readsRegister(PPC::CR0, TRI)))
// This instruction modifies or uses the record condition register after
// the one we want to change. While we could do this transformation, it
// would likely not be profitable. This transformation removes one
// instruction, and so even forcing RA to generate one move probably
// makes it unprofitable.
return false;
// Check whether CmpInstr can be made redundant by the current instruction.
if ((OpC == PPC::CMPW || OpC == PPC::CMPLW ||
OpC == PPC::CMPD || OpC == PPC::CMPLD) &&
(IOpC == PPC::SUBF || IOpC == PPC::SUBF8) &&
((Instr.getOperand(1).getReg() == SrcReg &&
Instr.getOperand(2).getReg() == SrcReg2) ||
(Instr.getOperand(1).getReg() == SrcReg2 &&
Instr.getOperand(2).getReg() == SrcReg))) {
Sub = &*I;
break;
}
if (I == B)
// The 'and' is below the comparison instruction.
return false;
}
// Return false if no candidates exist.
if (!MI && !Sub)
return false;
// The single candidate is called MI.
if (!MI) MI = Sub;
int NewOpC = -1;
MIOpC = MI->getOpcode();
if (MIOpC == PPC::ANDIo || MIOpC == PPC::ANDIo8)
NewOpC = MIOpC;
else {
NewOpC = PPC::getRecordFormOpcode(MIOpC);
if (NewOpC == -1 && PPC::getNonRecordFormOpcode(MIOpC) != -1)
NewOpC = MIOpC;
}
// FIXME: On the non-embedded POWER architectures, only some of the record
// forms are fast, and we should use only the fast ones.
// The defining instruction has a record form (or is already a record
// form). It is possible, however, that we'll need to reverse the condition
// code of the users.
if (NewOpC == -1)
return false;
// If we have SUB(r1, r2) and CMP(r2, r1), the condition code based on CMP
// needs to be updated to be based on SUB. Push the condition code
// operands to OperandsToUpdate. If it is safe to remove CmpInstr, the
// condition code of these operands will be modified.
// Here, Value == 0 means we haven't converted comparison against 1 or -1 to
// comparison against 0, which may modify predicate.
bool ShouldSwap = false;
if (Sub && Value == 0) {
ShouldSwap = SrcReg2 != 0 && Sub->getOperand(1).getReg() == SrcReg2 &&
Sub->getOperand(2).getReg() == SrcReg;
// The operands to subf are the opposite of sub, so only in the fixed-point
// case, invert the order.
ShouldSwap = !ShouldSwap;
}
if (ShouldSwap)
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();
unsigned PredCond = PPC::getPredicateCondition(Pred);
assert((!equalityOnly ||
PredCond == PPC::PRED_EQ || PredCond == PPC::PRED_NE) &&
"Invalid predicate for equality-only optimization");
(void)PredCond; // To suppress warning in release build.
PredsToUpdate.push_back(std::make_pair(&(UseMI->getOperand(0)),
PPC::getSwappedPredicate(Pred)));
} else if (UseMI->getOpcode() == PPC::ISEL ||
UseMI->getOpcode() == PPC::ISEL8) {
unsigned NewSubReg = UseMI->getOperand(3).getSubReg();
assert((!equalityOnly || NewSubReg == PPC::sub_eq) &&
"Invalid CR bit for equality-only optimization");
if (NewSubReg == PPC::sub_lt)
NewSubReg = PPC::sub_gt;
else if (NewSubReg == PPC::sub_gt)
NewSubReg = PPC::sub_lt;
SubRegsToUpdate.push_back(std::make_pair(&(UseMI->getOperand(3)),
NewSubReg));
} else // We need to abort on a user we don't understand.
return false;
}
assert(!(Value != 0 && ShouldSwap) &&
"Non-zero immediate support and ShouldSwap"
"may conflict in updating predicate");
// Create a new virtual register to hold the value of the CR set by the
// record-form instruction. If the instruction was not previously in
// record form, then set the kill flag on the CR.
CmpInstr.eraseFromParent();
MachineBasicBlock::iterator MII = MI;
BuildMI(*MI->getParent(), std::next(MII), MI->getDebugLoc(),
get(TargetOpcode::COPY), CRReg)
.addReg(PPC::CR0, MIOpC != NewOpC ? RegState::Kill : 0);
// Even if CR0 register were dead before, it is alive now since the
// instruction we just built uses it.
MI->clearRegisterDeads(PPC::CR0);
if (MIOpC != NewOpC) {
// We need to be careful here: we're replacing one instruction with
// another, and we need to make sure that we get all of the right
// implicit uses and defs. On the other hand, the caller may be holding
// an iterator to this instruction, and so we can't delete it (this is
// specifically the case if this is the instruction directly after the
// compare).
const MCInstrDesc &NewDesc = get(NewOpC);
MI->setDesc(NewDesc);
if (NewDesc.ImplicitDefs)
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 MCPhysReg *ImpUses = NewDesc.getImplicitUses();
*ImpUses; ++ImpUses)
if (!MI->readsRegister(*ImpUses))
MI->addOperand(*MI->getParent()->getParent(),
MachineOperand::CreateReg(*ImpUses, false, true));
}
assert(MI->definesRegister(PPC::CR0) &&
"Record-form instruction does not define cr0?");
// Modify the condition code of operands in OperandsToUpdate.
// Since we have SUB(r1, r2) and CMP(r2, r1), the condition code needs to
// be changed from r2 > r1 to r1 < r2, from r2 < r1 to r1 > r2, etc.
for (unsigned i = 0, e = PredsToUpdate.size(); i < e; i++)
PredsToUpdate[i].first->setImm(PredsToUpdate[i].second);
for (unsigned i = 0, e = SubRegsToUpdate.size(); i < e; i++)
SubRegsToUpdate[i].first->setSubReg(SubRegsToUpdate[i].second);
return true;
}
/// GetInstSize - Return the number of bytes of code the specified
/// instruction may be. This returns the maximum number of bytes.
///
unsigned PPCInstrInfo::getInstSizeInBytes(const MachineInstr &MI) const {
unsigned Opcode = MI.getOpcode();
if (Opcode == PPC::INLINEASM) {
const MachineFunction *MF = MI.getParent()->getParent();
const char *AsmStr = MI.getOperand(0).getSymbolName();
return getInlineAsmLength(AsmStr, *MF->getTarget().getMCAsmInfo());
} else if (Opcode == TargetOpcode::STACKMAP) {
StackMapOpers Opers(&MI);
return Opers.getNumPatchBytes();
} else if (Opcode == TargetOpcode::PATCHPOINT) {
PatchPointOpers Opers(&MI);
return Opers.getNumPatchBytes();
} else {
return get(Opcode).getSize();
}
}
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, "ppc-plt"},
{MO_PIC_FLAG, "ppc-pic"},
{MO_NLP_FLAG, "ppc-nlp"},
{MO_NLP_HIDDEN_FLAG, "ppc-nlp-hidden"}};
return makeArrayRef(TargetFlags);
}
bool PPCInstrInfo::expandPostRAPseudo(MachineInstr &MI) const {
auto &MBB = *MI.getParent();
auto DL = MI.getDebugLoc();
switch (MI.getOpcode()) {
case TargetOpcode::LOAD_STACK_GUARD: {
assert(Subtarget.isTargetLinux() &&
"Only Linux target is expected to contain LOAD_STACK_GUARD");
const int64_t Offset = Subtarget.isPPC64() ? -0x7010 : -0x7008;
const unsigned Reg = Subtarget.isPPC64() ? PPC::X13 : PPC::R2;
MI.setDesc(get(Subtarget.isPPC64() ? PPC::LD : PPC::LWZ));
MachineInstrBuilder(*MI.getParent()->getParent(), MI)
.addImm(Offset)
.addReg(Reg);
return true;
}
case PPC::DFLOADf32:
case PPC::DFLOADf64:
case PPC::DFSTOREf32:
case PPC::DFSTOREf64: {
assert(Subtarget.hasP9Vector() &&
"Invalid D-Form Pseudo-ops on non-P9 target.");
assert(MI.getOperand(2).isReg() && MI.getOperand(1).isImm() &&
"D-form op must have register and immediate operands");
unsigned UpperOpcode, LowerOpcode;
switch (MI.getOpcode()) {
case PPC::DFLOADf32:
UpperOpcode = PPC::LXSSP;
LowerOpcode = PPC::LFS;
break;
case PPC::DFLOADf64:
UpperOpcode = PPC::LXSD;
LowerOpcode = PPC::LFD;
break;
case PPC::DFSTOREf32:
UpperOpcode = PPC::STXSSP;
LowerOpcode = PPC::STFS;
break;
case PPC::DFSTOREf64:
UpperOpcode = PPC::STXSD;
LowerOpcode = PPC::STFD;
break;
}
unsigned TargetReg = MI.getOperand(0).getReg();
unsigned Opcode;
if ((TargetReg >= PPC::F0 && TargetReg <= PPC::F31) ||
(TargetReg >= PPC::VSL0 && TargetReg <= PPC::VSL31))
Opcode = LowerOpcode;
else
Opcode = UpperOpcode;
MI.setDesc(get(Opcode));
return true;
}
case PPC::CFENCE8: {
auto Val = MI.getOperand(0).getReg();
BuildMI(MBB, MI, DL, get(PPC::CMPD), PPC::CR7).addReg(Val).addReg(Val);
BuildMI(MBB, MI, DL, get(PPC::CTRL_DEP))
.addImm(PPC::PRED_NE_MINUS)
.addReg(PPC::CR7)
.addImm(1);
MI.setDesc(get(PPC::ISYNC));
MI.RemoveOperand(0);
return true;
}
}
return false;
}
const TargetRegisterClass *
PPCInstrInfo::updatedRC(const TargetRegisterClass *RC) const {
if (Subtarget.hasVSX() && RC == &PPC::VRRCRegClass)
return &PPC::VSRCRegClass;
return RC;
}
int PPCInstrInfo::getRecordFormOpcode(unsigned Opcode) {
return PPC::getRecordFormOpcode(Opcode);
}
Reference in New Issue View Git Blame Copy Permalink