-//===- PPCInstrInfo.cpp - PowerPC32 Instruction Information -----*- C++ -*-===//
+//===-- PPCInstrInfo.cpp - PowerPC Instruction Information ----------------===//
//
// The LLVM Compiler Infrastructure
//
//===----------------------------------------------------------------------===//
#include "PPCInstrInfo.h"
+#include "MCTargetDesc/PPCPredicates.h"
+#include "PPC.h"
+#include "PPCHazardRecognizers.h"
#include "PPCInstrBuilder.h"
#include "PPCMachineFunctionInfo.h"
-#include "PPCPredicates.h"
-#include "PPCGenInstrInfo.inc"
#include "PPCTargetMachine.h"
#include "llvm/ADT/STLExtras.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/PseudoSourceValue.h"
+#include "llvm/MC/MCAsmInfo.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/TargetRegistry.h"
#include "llvm/Support/raw_ostream.h"
-#include "llvm/MC/MCAsmInfo.h"
-namespace llvm {
-extern cl::opt<bool> EnablePPC32RS; // FIXME (64-bit): See PPCRegisterInfo.cpp.
-extern cl::opt<bool> EnablePPC64RS; // FIXME (64-bit): See PPCRegisterInfo.cpp.
-}
+#define GET_INSTRINFO_CTOR
+#include "PPCGenInstrInfo.inc"
using namespace llvm;
+static cl::
+opt<bool> DisableCTRLoopAnal("disable-ppc-ctrloop-analysis", cl::Hidden,
+ cl::desc("Disable analysis for CTR loops"));
+
PPCInstrInfo::PPCInstrInfo(PPCTargetMachine &tm)
- : TargetInstrInfoImpl(PPCInsts, array_lengthof(PPCInsts)), TM(tm),
- RI(*TM.getSubtargetImpl(), *this) {}
-
-bool PPCInstrInfo::isMoveInstr(const MachineInstr& MI,
- unsigned& sourceReg,
- unsigned& destReg,
- unsigned& sourceSubIdx,
- unsigned& destSubIdx) const {
- sourceSubIdx = destSubIdx = 0; // No sub-registers.
-
- unsigned oc = MI.getOpcode();
- if (oc == PPC::OR || oc == PPC::OR8 || oc == PPC::VOR ||
- oc == PPC::OR4To8 || oc == PPC::OR8To4) { // or r1, r2, r2
- assert(MI.getNumOperands() >= 3 &&
- MI.getOperand(0).isReg() &&
- MI.getOperand(1).isReg() &&
- MI.getOperand(2).isReg() &&
- "invalid PPC OR instruction!");
- if (MI.getOperand(1).getReg() == MI.getOperand(2).getReg()) {
- sourceReg = MI.getOperand(1).getReg();
- destReg = MI.getOperand(0).getReg();
- return true;
- }
- } else if (oc == PPC::ADDI) { // addi r1, r2, 0
- assert(MI.getNumOperands() >= 3 &&
- MI.getOperand(0).isReg() &&
- MI.getOperand(2).isImm() &&
- "invalid PPC ADDI instruction!");
- if (MI.getOperand(1).isReg() && MI.getOperand(2).getImm() == 0) {
- sourceReg = MI.getOperand(1).getReg();
- destReg = MI.getOperand(0).getReg();
- return true;
- }
- } else if (oc == PPC::ORI) { // ori r1, r2, 0
- assert(MI.getNumOperands() >= 3 &&
- MI.getOperand(0).isReg() &&
- MI.getOperand(1).isReg() &&
- MI.getOperand(2).isImm() &&
- "invalid PPC ORI instruction!");
- if (MI.getOperand(2).getImm() == 0) {
- sourceReg = MI.getOperand(1).getReg();
- destReg = MI.getOperand(0).getReg();
- return true;
- }
- } else if (oc == PPC::FMR || oc == PPC::FMRSD) { // fmr r1, r2
- assert(MI.getNumOperands() >= 2 &&
- MI.getOperand(0).isReg() &&
- MI.getOperand(1).isReg() &&
- "invalid PPC FMR instruction");
- sourceReg = MI.getOperand(1).getReg();
- destReg = MI.getOperand(0).getReg();
- return true;
- } else if (oc == PPC::MCRF) { // mcrf cr1, cr2
- assert(MI.getNumOperands() >= 2 &&
- MI.getOperand(0).isReg() &&
- MI.getOperand(1).isReg() &&
- "invalid PPC MCRF instruction");
- sourceReg = MI.getOperand(1).getReg();
- destReg = MI.getOperand(0).getReg();
+ : PPCGenInstrInfo(PPC::ADJCALLSTACKDOWN, PPC::ADJCALLSTACKUP),
+ TM(tm), RI(*TM.getSubtargetImpl(), *this) {}
+
+/// 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();
+ if (Directive == PPC::DIR_440 || Directive == PPC::DIR_A2 ||
+ Directive == PPC::DIR_E500mc || Directive == PPC::DIR_E5500) {
+ const InstrItineraryData *II = TM->getInstrItineraryData();
+ return new PPCScoreboardHazardRecognizer(II, DAG);
+ }
+
+ return TargetInstrInfo::CreateTargetHazardRecognizer(TM, 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();
+
+ // Most subtargets use a PPC970 recognizer.
+ if (Directive != PPC::DIR_440 && Directive != PPC::DIR_A2 &&
+ Directive != PPC::DIR_E500mc && Directive != PPC::DIR_E5500) {
+ const TargetInstrInfo *TII = TM.getInstrInfo();
+ assert(TII && "No InstrInfo?");
+
+ return new PPCHazardRecognizer970(*TII);
+ }
+
+ return new PPCScoreboardHazardRecognizer(II, DAG);
+}
+
+// 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;
}
- return false;
}
-unsigned PPCInstrInfo::isLoadFromStackSlot(const MachineInstr *MI,
+unsigned PPCInstrInfo::isLoadFromStackSlot(const MachineInstr *MI,
int &FrameIndex) const {
switch (MI->getOpcode()) {
default: break;
return 0;
}
-unsigned PPCInstrInfo::isStoreToStackSlot(const MachineInstr *MI,
+unsigned PPCInstrInfo::isStoreToStackSlot(const MachineInstr *MI,
int &FrameIndex) const {
switch (MI->getOpcode()) {
default: break;
// Normal instructions can be commuted the obvious way.
if (MI->getOpcode() != PPC::RLWIMI)
- return TargetInstrInfoImpl::commuteInstruction(MI, NewMI);
-
+ return TargetInstrInfo::commuteInstruction(MI, NewMI);
+
// Cannot commute if it has a non-zero rotate count.
if (MI->getOperand(3).getImm() != 0)
return 0;
-
+
// If we have a zero rotate count, we have:
// M = mask(MB,ME)
// Op0 = (Op1 & ~M) | (Op2 & M)
// destination register as well.
if (Reg0 == Reg1) {
// Must be two address instruction!
- assert(MI->getDesc().getOperandConstraint(0, TOI::TIED_TO) &&
+ assert(MI->getDesc().getOperandConstraint(0, MCOI::TIED_TO) &&
"Expecting a two-address instruction!");
Reg2IsKill = false;
ChangeReg0 = true;
MI->getOperand(1).setReg(Reg2);
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;
}
-void PPCInstrInfo::insertNoop(MachineBasicBlock &MBB,
+void PPCInstrInfo::insertNoop(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI) const {
DebugLoc DL;
BuildMI(MBB, MI, DL, get(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 = TM.getSubtargetImpl()->isPPC64();
+
// If the block has no terminators, it just falls into the block after it.
MachineBasicBlock::iterator I = MBB.end();
if (I == MBB.begin())
// 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) {
Cond.push_back(LastInst->getOperand(0));
Cond.push_back(LastInst->getOperand(1));
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 (SecondLastInst && I != MBB.begin() &&
isUnpredicatedTerminator(--I))
return true;
-
+
// If the block ends with PPC::B and PPC:BCC, handle it.
- if (SecondLastInst->getOpcode() == PPC::BCC &&
+ if (SecondLastInst->getOpcode() == PPC::BCC &&
LastInst->getOpcode() == PPC::B) {
if (!SecondLastInst->getOperand(2).isMBB() ||
!LastInst->getOperand(0).isMBB())
Cond.push_back(SecondLastInst->getOperand(1));
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 &&
+ if (SecondLastInst->getOpcode() == PPC::B &&
LastInst->getOpcode() == PPC::B) {
if (!SecondLastInst->getOperand(0).isMBB())
return true;
return 0;
--I;
}
- if (I->getOpcode() != PPC::B && I->getOpcode() != PPC::BCC)
+ if (I->getOpcode() != PPC::B && I->getOpcode() != PPC::BCC &&
+ 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)
+ if (I->getOpcode() != PPC::BCC &&
+ 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;
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) &&
+ assert((Cond.size() == 2 || Cond.size() == 0) &&
"PPC branch conditions have two components!");
-
+
+ bool isPPC64 = TM.getSubtargetImpl()->isPPC64();
+
// One-way branch.
if (FBB == 0) {
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 // Conditional branch
BuildMI(&MBB, DL, get(PPC::BCC))
.addImm(Cond[0].getImm()).addReg(Cond[1].getReg()).addMBB(TBB);
return 1;
}
-
+
// Two-way Conditional Branch.
- BuildMI(&MBB, DL, get(PPC::BCC))
- .addImm(Cond[0].getImm()).addReg(Cond[1].getReg()).addMBB(TBB);
+ 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
+ BuildMI(&MBB, DL, get(PPC::BCC))
+ .addImm(Cond[0].getImm()).addReg(Cond[1].getReg()).addMBB(TBB);
BuildMI(&MBB, DL, get(PPC::B)).addMBB(FBB);
return 2;
}
Opc = PPC::VOR;
else if (PPC::CRBITRCRegClass.contains(DestReg, SrcReg))
Opc = PPC::CROR;
+
+ // Asymmetric copies:
+
+ else if (PPC::GPRCRegClass.contains(DestReg) &&
+ PPC::G8RCRegClass.contains(SrcReg))
+ Opc = PPC::OR_64;
+ else if (PPC::G8RCRegClass.contains(DestReg) &&
+ PPC::GPRCRegClass.contains(SrcReg))
+ Opc = PPC::OR8_32;
else
llvm_unreachable("Impossible reg-to-reg copy");
- const TargetInstrDesc &TID = get(Opc);
- if (TID.getNumOperands() == 3)
- BuildMI(MBB, I, DL, TID, DestReg)
+ 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, TID, DestReg).addReg(SrcReg, getKillRegState(KillSrc));
+ 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) const{
+ SmallVectorImpl<MachineInstr*> &NewMIs,
+ bool &NonRI) const{
DebugLoc DL;
- if (RC == PPC::GPRCRegisterClass) {
+ if (PPC::GPRCRegClass.hasSubClassEq(RC)) {
if (SrcReg != PPC::LR) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STW))
.addReg(SrcReg,
getKillRegState(isKill)),
FrameIdx));
}
- } else if (RC == PPC::G8RCRegisterClass) {
+ } else if (PPC::G8RCRegClass.hasSubClassEq(RC)) {
if (SrcReg != PPC::LR8) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STD))
.addReg(SrcReg,
FrameIdx));
} else {
// FIXME: this spills LR immediately to memory in one step. To do this,
- // we use R11, which we know cannot be used in the prolog/epilog. This is
+ // we use X11, which we know cannot be used in the prolog/epilog. This is
// a hack.
NewMIs.push_back(BuildMI(MF, DL, get(PPC::MFLR8), PPC::X11));
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STD))
getKillRegState(isKill)),
FrameIdx));
}
- } else if (RC == PPC::F8RCRegisterClass) {
+ } else if (PPC::F8RCRegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STFD))
.addReg(SrcReg,
getKillRegState(isKill)),
FrameIdx));
- } else if (RC == PPC::F4RCRegisterClass) {
+ } else if (PPC::F4RCRegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STFS))
.addReg(SrcReg,
getKillRegState(isKill)),
FrameIdx));
- } else if (RC == PPC::CRRCRegisterClass) {
- if ((EnablePPC32RS && !TM.getSubtargetImpl()->isPPC64()) ||
- (EnablePPC64RS && TM.getSubtargetImpl()->isPPC64())) {
- // FIXME (64-bit): Enable
- NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::SPILL_CR))
- .addReg(SrcReg,
- getKillRegState(isKill)),
- FrameIdx));
- return true;
- } else {
- // FIXME: We need a scatch reg here. The trouble with using R0 is that
- // it's possible for the stack frame to be so big the save location is
- // out of range of immediate offsets, necessitating another register.
- // We hack this on Darwin by reserving R2. It's probably broken on Linux
- // at the moment.
-
- // We need to store the CR in the low 4-bits of the saved value. First,
- // issue a MFCR to save all of the CRBits.
- unsigned ScratchReg = TM.getSubtargetImpl()->isDarwinABI() ?
- PPC::R2 : PPC::R0;
- NewMIs.push_back(BuildMI(MF, DL, get(PPC::MFCRpseud), ScratchReg)
- .addReg(SrcReg, getKillRegState(isKill)));
-
- // If the saved register wasn't CR0, shift the bits left so that they are
- // in CR0's slot.
- if (SrcReg != PPC::CR0) {
- unsigned ShiftBits = PPCRegisterInfo::getRegisterNumbering(SrcReg)*4;
- // rlwinm scratch, scratch, ShiftBits, 0, 31.
- NewMIs.push_back(BuildMI(MF, DL, get(PPC::RLWINM), ScratchReg)
- .addReg(ScratchReg).addImm(ShiftBits)
- .addImm(0).addImm(31));
- }
-
- NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::STW))
- .addReg(ScratchReg,
- getKillRegState(isKill)),
- FrameIdx));
- }
- } else if (RC == PPC::CRBITRCRegisterClass) {
+ } 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)) {
// 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
SrcReg == PPC::CR7EQ || SrcReg == PPC::CR7UN)
Reg = PPC::CR7;
- return StoreRegToStackSlot(MF, Reg, isKill, FrameIdx,
- PPC::CRRCRegisterClass, NewMIs);
-
- } else if (RC == PPC::VRRCRegisterClass) {
- // We don't have indexed addressing for vector loads. Emit:
- // R0 = ADDI FI#
- // STVX VAL, 0, R0
- //
- // FIXME: We use R0 here, because it isn't available for RA.
- NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::ADDI), PPC::R0),
- FrameIdx, 0, 0));
- NewMIs.push_back(BuildMI(MF, DL, get(PPC::STVX))
- .addReg(SrcReg, getKillRegState(isKill))
- .addReg(PPC::R0)
- .addReg(PPC::R0));
+ return StoreRegToStackSlot(MF, Reg, isKill, FrameIdx,
+ &PPC::CRRCRegClass, NewMIs, NonRI);
+
+ } 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::VRSAVERCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::SPILL_VRSAVE))
+ .addReg(SrcReg,
+ getKillRegState(isKill)),
+ FrameIdx));
+ NonRI = true;
} else {
llvm_unreachable("Unknown regclass!");
}
MachineFunction &MF = *MBB.getParent();
SmallVector<MachineInstr*, 4> NewMIs;
- if (StoreRegToStackSlot(MF, SrcReg, isKill, FrameIdx, RC, NewMIs)) {
- PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
+ PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
+ FuncInfo->setHasSpills();
+
+ bool NonRI = false;
+ if (StoreRegToStackSlot(MF, SrcReg, isKill, FrameIdx, RC, NewMIs, NonRI))
FuncInfo->setSpillsCR();
- }
+
+ 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(FrameIdx),
+ MachineMemOperand::MOStore,
+ MFI.getObjectSize(FrameIdx),
+ MFI.getObjectAlignment(FrameIdx));
+ NewMIs.back()->addMemOperand(MF, MMO);
}
-void
+bool
PPCInstrInfo::LoadRegFromStackSlot(MachineFunction &MF, DebugLoc DL,
unsigned DestReg, int FrameIdx,
const TargetRegisterClass *RC,
- SmallVectorImpl<MachineInstr*> &NewMIs)const{
- if (RC == PPC::GPRCRegisterClass) {
+ SmallVectorImpl<MachineInstr*> &NewMIs,
+ bool &NonRI) const{
+ if (PPC::GPRCRegClass.hasSubClassEq(RC)) {
if (DestReg != PPC::LR) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LWZ),
DestReg), FrameIdx));
PPC::R11), FrameIdx));
NewMIs.push_back(BuildMI(MF, DL, get(PPC::MTLR)).addReg(PPC::R11));
}
- } else if (RC == PPC::G8RCRegisterClass) {
+ } else if (PPC::G8RCRegClass.hasSubClassEq(RC)) {
if (DestReg != PPC::LR8) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LD), DestReg),
FrameIdx));
} else {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LD),
- PPC::R11), FrameIdx));
- NewMIs.push_back(BuildMI(MF, DL, get(PPC::MTLR8)).addReg(PPC::R11));
+ PPC::X11), FrameIdx));
+ NewMIs.push_back(BuildMI(MF, DL, get(PPC::MTLR8)).addReg(PPC::X11));
}
- } else if (RC == PPC::F8RCRegisterClass) {
+ } else if (PPC::F8RCRegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LFD), DestReg),
FrameIdx));
- } else if (RC == PPC::F4RCRegisterClass) {
+ } else if (PPC::F4RCRegClass.hasSubClassEq(RC)) {
NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LFS), DestReg),
FrameIdx));
- } else if (RC == PPC::CRRCRegisterClass) {
- // FIXME: We need a scatch reg here. The trouble with using R0 is that
- // it's possible for the stack frame to be so big the save location is
- // out of range of immediate offsets, necessitating another register.
- // We hack this on Darwin by reserving R2. It's probably broken on Linux
- // at the moment.
- unsigned ScratchReg = TM.getSubtargetImpl()->isDarwinABI() ?
- PPC::R2 : PPC::R0;
- NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LWZ),
- ScratchReg), FrameIdx));
-
- // If the reloaded register isn't CR0, shift the bits right so that they are
- // in the right CR's slot.
- if (DestReg != PPC::CR0) {
- unsigned ShiftBits = PPCRegisterInfo::getRegisterNumbering(DestReg)*4;
- // rlwinm r11, r11, 32-ShiftBits, 0, 31.
- NewMIs.push_back(BuildMI(MF, DL, get(PPC::RLWINM), ScratchReg)
- .addReg(ScratchReg).addImm(32-ShiftBits).addImm(0)
- .addImm(31));
- }
-
- NewMIs.push_back(BuildMI(MF, DL, get(PPC::MTCRF), DestReg)
- .addReg(ScratchReg));
- } else if (RC == PPC::CRBITRCRegisterClass) {
-
+ } 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)) {
+
unsigned Reg = 0;
if (DestReg == PPC::CR0LT || DestReg == PPC::CR0GT ||
DestReg == PPC::CR0EQ || DestReg == PPC::CR0UN)
DestReg == PPC::CR7EQ || DestReg == PPC::CR7UN)
Reg = PPC::CR7;
- return LoadRegFromStackSlot(MF, DL, Reg, FrameIdx,
- PPC::CRRCRegisterClass, NewMIs);
-
- } else if (RC == PPC::VRRCRegisterClass) {
- // We don't have indexed addressing for vector loads. Emit:
- // R0 = ADDI FI#
- // Dest = LVX 0, R0
- //
- // FIXME: We use R0 here, because it isn't available for RA.
- NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::ADDI), PPC::R0),
- FrameIdx, 0, 0));
- NewMIs.push_back(BuildMI(MF, DL, get(PPC::LVX),DestReg).addReg(PPC::R0)
- .addReg(PPC::R0));
+ return LoadRegFromStackSlot(MF, DL, Reg, FrameIdx,
+ &PPC::CRRCRegClass, NewMIs, NonRI);
+
+ } else if (PPC::VRRCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::LVX), DestReg),
+ FrameIdx));
+ NonRI = true;
+ } else if (PPC::VRSAVERCRegClass.hasSubClassEq(RC)) {
+ NewMIs.push_back(addFrameReference(BuildMI(MF, DL,
+ get(PPC::RESTORE_VRSAVE),
+ DestReg),
+ FrameIdx));
+ NonRI = true;
} else {
llvm_unreachable("Unknown regclass!");
}
+
+ return false;
}
void
SmallVector<MachineInstr*, 4> NewMIs;
DebugLoc DL;
if (MI != MBB.end()) DL = MI->getDebugLoc();
- LoadRegFromStackSlot(MF, DL, DestReg, FrameIdx, RC, NewMIs);
+
+ PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
+ FuncInfo->setHasSpills();
+
+ bool NonRI = false;
+ if (LoadRegFromStackSlot(MF, DL, DestReg, FrameIdx, RC, NewMIs, NonRI))
+ FuncInfo->setSpillsCR();
+
+ 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(FrameIdx),
+ MachineMemOperand::MOLoad,
+ MFI.getObjectSize(FrameIdx),
+ MFI.getObjectAlignment(FrameIdx));
+ NewMIs.back()->addMemOperand(MF, MMO);
}
MachineInstr*
bool PPCInstrInfo::
ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const {
assert(Cond.size() == 2 && "Invalid PPC branch opcode!");
- // Leave the CR# the same, but invert the condition.
- Cond[0].setImm(PPC::InvertPredicate((PPC::Predicate)Cond[0].getImm()));
+ 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;
}
const char *AsmStr = MI->getOperand(0).getSymbolName();
return getInlineAsmLength(AsmStr, *MF->getTarget().getMCAsmInfo());
}
- case PPC::DBG_LABEL:
+ case PPC::PROLOG_LABEL:
case PPC::EH_LABEL:
case PPC::GC_LABEL:
case PPC::DBG_VALUE:
return 0;
+ case PPC::BL8_NOP_ELF:
+ case PPC::BLA8_NOP_ELF:
+ return 8;
default:
return 4; // PowerPC instructions are all 4 bytes
}
projects / oota-llvm.git / blobdiff