RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-25 04:02:41 +01:00
Code Issues Projects Releases Wiki Activity

Reflects the chanegs made to PredicateOperand.

Browse Source 
llvm-svn: 37898
This commit is contained in:
Evan Cheng 2007-07-05 07:18:20 +00:00
parent 3d2cfd8bb1
commit be54fdf431
4 changed files with 93 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
lib/Target/ARM/ARMConstantIslandPass.cpp
View File

@ -1192,6 +1192,7 @@ ARMConstantIslands::FixUpConditionalBr(MachineFunction &Fn, ImmBranch &Br) {
// L2: // L2:
ARMCC::CondCodes CC = (ARMCC::CondCodes)MI->getOperand(1).getImmedValue(); ARMCC::CondCodes CC = (ARMCC::CondCodes)MI->getOperand(1).getImmedValue();
CC = ARMCC::getOppositeCondition(CC); CC = ARMCC::getOppositeCondition(CC);
unsigned CCReg = MI->getOperand(2).getReg();
// If the branch is at the end of its MBB and that has a fall-through block, // If the branch is at the end of its MBB and that has a fall-through block,
// direct the updated conditional branch to the fall-through block. Otherwise, // direct the updated conditional branch to the fall-through block. Otherwise,
@ -1241,7 +1242,8 @@ ARMConstantIslands::FixUpConditionalBr(MachineFunction &Fn, ImmBranch &Br) {
// Insert a new conditional branch and a new unconditional branch. // Insert a new conditional branch and a new unconditional branch.
// Also update the ImmBranch as well as adding a new entry for the new branch. // Also update the ImmBranch as well as adding a new entry for the new branch.
BuildMI(MBB, TII->get(MI->getOpcode())).addMBB(NextBB).addImm(CC); BuildMI(MBB, TII->get(MI->getOpcode())).addMBB(NextBB)
.addImm(CC).addReg(CCReg);
Br.MI = &MBB->back(); Br.MI = &MBB->back();
BBSizes[MBB->getNumber()] += ARM::GetInstSize(&MBB->back()); BBSizes[MBB->getNumber()] += ARM::GetInstSize(&MBB->back());
BuildMI(MBB, TII->get(Br.UncondBr)).addMBB(DestBB); BuildMI(MBB, TII->get(Br.UncondBr)).addMBB(DestBB);

25
lib/Target/ARM/ARMISelLowering.cpp
View File

@ -1127,8 +1127,9 @@ static SDOperand LowerSELECT_CC(SDOperand Op, SelectionDAG &DAG,
if (LHS.getValueType() == MVT::i32) { if (LHS.getValueType() == MVT::i32) {
SDOperand ARMCC; SDOperand ARMCC;
SDOperand CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
SDOperand Cmp = getARMCmp(LHS, RHS, CC, ARMCC, DAG, ST->isThumb()); SDOperand Cmp = getARMCmp(LHS, RHS, CC, ARMCC, DAG, ST->isThumb());
return DAG.getNode(ARMISD::CMOV, VT, FalseVal, TrueVal, ARMCC, Cmp); return DAG.getNode(ARMISD::CMOV, VT, FalseVal, TrueVal, ARMCC, CCR, Cmp);
} }
ARMCC::CondCodes CondCode, CondCode2; ARMCC::CondCodes CondCode, CondCode2;
@ -1136,14 +1137,15 @@ static SDOperand LowerSELECT_CC(SDOperand Op, SelectionDAG &DAG,
std::swap(TrueVal, FalseVal); std::swap(TrueVal, FalseVal);
SDOperand ARMCC = DAG.getConstant(CondCode, MVT::i32); SDOperand ARMCC = DAG.getConstant(CondCode, MVT::i32);
SDOperand CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
SDOperand Cmp = getVFPCmp(LHS, RHS, DAG); SDOperand Cmp = getVFPCmp(LHS, RHS, DAG);
SDOperand Result = DAG.getNode(ARMISD::CMOV, VT, FalseVal, TrueVal, SDOperand Result = DAG.getNode(ARMISD::CMOV, VT, FalseVal, TrueVal,
ARMCC, Cmp); ARMCC, CCR, Cmp);
if (CondCode2 != ARMCC::AL) { if (CondCode2 != ARMCC::AL) {
SDOperand ARMCC2 = DAG.getConstant(CondCode2, MVT::i32); SDOperand ARMCC2 = DAG.getConstant(CondCode2, MVT::i32);
// FIXME: Needs another CMP because flag can have but one use. // FIXME: Needs another CMP because flag can have but one use.
SDOperand Cmp2 = getVFPCmp(LHS, RHS, DAG); SDOperand Cmp2 = getVFPCmp(LHS, RHS, DAG);
Result = DAG.getNode(ARMISD::CMOV, VT, Result, TrueVal, ARMCC2, Cmp2); Result = DAG.getNode(ARMISD::CMOV, VT, Result, TrueVal, ARMCC2, CCR, Cmp2);
} }
return Result; return Result;
} }
@ -1158,8 +1160,9 @@ static SDOperand LowerBR_CC(SDOperand Op, SelectionDAG &DAG,
if (LHS.getValueType() == MVT::i32) { if (LHS.getValueType() == MVT::i32) {
SDOperand ARMCC; SDOperand ARMCC;
SDOperand CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
SDOperand Cmp = getARMCmp(LHS, RHS, CC, ARMCC, DAG, ST->isThumb()); SDOperand Cmp = getARMCmp(LHS, RHS, CC, ARMCC, DAG, ST->isThumb());
return DAG.getNode(ARMISD::BRCOND, MVT::Other, Chain, Dest, ARMCC, Cmp); return DAG.getNode(ARMISD::BRCOND, MVT::Other, Chain, Dest, ARMCC, CCR,Cmp);
} }
assert(LHS.getValueType() == MVT::f32 || LHS.getValueType() == MVT::f64); assert(LHS.getValueType() == MVT::f32 || LHS.getValueType() == MVT::f64);
@ -1170,13 +1173,14 @@ static SDOperand LowerBR_CC(SDOperand Op, SelectionDAG &DAG,
SDOperand Cmp = getVFPCmp(LHS, RHS, DAG); SDOperand Cmp = getVFPCmp(LHS, RHS, DAG);
SDOperand ARMCC = DAG.getConstant(CondCode, MVT::i32); SDOperand ARMCC = DAG.getConstant(CondCode, MVT::i32);
SDOperand CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
SDVTList VTList = DAG.getVTList(MVT::Other, MVT::Flag); SDVTList VTList = DAG.getVTList(MVT::Other, MVT::Flag);
SDOperand Ops[] = { Chain, Dest, ARMCC, Cmp }; SDOperand Ops[] = { Chain, Dest, ARMCC, CCR, Cmp };
SDOperand Res = DAG.getNode(ARMISD::BRCOND, VTList, Ops, 4); SDOperand Res = DAG.getNode(ARMISD::BRCOND, VTList, Ops, 5);
if (CondCode2 != ARMCC::AL) { if (CondCode2 != ARMCC::AL) {
ARMCC = DAG.getConstant(CondCode2, MVT::i32); ARMCC = DAG.getConstant(CondCode2, MVT::i32);
SDOperand Ops[] = { Res, Dest, ARMCC, Res.getValue(1) }; SDOperand Ops[] = { Res, Dest, ARMCC, CCR, Res.getValue(1) };
Res = DAG.getNode(ARMISD::BRCOND, VTList, Ops, 4); Res = DAG.getNode(ARMISD::BRCOND, VTList, Ops, 5);
} }
return Res; return Res;
} }
@ -1228,7 +1232,8 @@ static SDOperand LowerFCOPYSIGN(SDOperand Op, SelectionDAG &DAG) {
SDOperand AbsVal = DAG.getNode(ISD::FABS, VT, Tmp0); SDOperand AbsVal = DAG.getNode(ISD::FABS, VT, Tmp0);
SDOperand Cmp = getVFPCmp(Tmp1, DAG.getConstantFP(0.0, SrcVT), DAG); SDOperand Cmp = getVFPCmp(Tmp1, DAG.getConstantFP(0.0, SrcVT), DAG);
SDOperand ARMCC = DAG.getConstant(ARMCC::LT, MVT::i32); SDOperand ARMCC = DAG.getConstant(ARMCC::LT, MVT::i32);
return DAG.getNode(ARMISD::CNEG, VT, AbsVal, AbsVal, ARMCC, Cmp); SDOperand CCR = DAG.getRegister(ARM::CPSR, MVT::i32);
return DAG.getNode(ARMISD::CNEG, VT, AbsVal, AbsVal, ARMCC, CCR, Cmp);
} }
static SDOperand LowerBIT_CONVERT(SDOperand Op, SelectionDAG &DAG) { static SDOperand LowerBIT_CONVERT(SDOperand Op, SelectionDAG &DAG) {
@ -1472,7 +1477,7 @@ ARMTargetLowering::InsertAtEndOfBasicBlock(MachineInstr *MI,
MachineBasicBlock *copy0MBB = new MachineBasicBlock(LLVM_BB); MachineBasicBlock *copy0MBB = new MachineBasicBlock(LLVM_BB);
MachineBasicBlock *sinkMBB = new MachineBasicBlock(LLVM_BB); MachineBasicBlock *sinkMBB = new MachineBasicBlock(LLVM_BB);
BuildMI(BB, TII->get(ARM::tBcc)).addMBB(sinkMBB) BuildMI(BB, TII->get(ARM::tBcc)).addMBB(sinkMBB)
.addImm(MI->getOperand(3).getImm()); .addImm(MI->getOperand(3).getImm()).addReg(MI->getOperand(4).getReg());
MachineFunction *F = BB->getParent(); MachineFunction *F = BB->getParent();
F->getBasicBlockList().insert(It, copy0MBB); F->getBasicBlockList().insert(It, copy0MBB);
F->getBasicBlockList().insert(It, sinkMBB); F->getBasicBlockList().insert(It, sinkMBB);

14
lib/Target/ARM/ARMInstrInfo.cpp
View File

@ -322,6 +322,7 @@ bool ARMInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,MachineBasicBlock *&TBB,
// Block ends with fall-through condbranch. // Block ends with fall-through condbranch.
TBB = LastInst->getOperand(0).getMachineBasicBlock(); TBB = LastInst->getOperand(0).getMachineBasicBlock();
Cond.push_back(LastInst->getOperand(1)); Cond.push_back(LastInst->getOperand(1));
Cond.push_back(LastInst->getOperand(2));
return false; return false;
} }
return true; // Can't handle indirect branch. return true; // Can't handle indirect branch.
@ -341,6 +342,7 @@ bool ARMInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,MachineBasicBlock *&TBB,
(SecondLastOpc == ARM::tBcc && LastOpc == ARM::tB)) { (SecondLastOpc == ARM::tBcc && LastOpc == ARM::tB)) {
TBB = SecondLastInst->getOperand(0).getMachineBasicBlock(); TBB = SecondLastInst->getOperand(0).getMachineBasicBlock();
Cond.push_back(SecondLastInst->getOperand(1)); Cond.push_back(SecondLastInst->getOperand(1));
Cond.push_back(SecondLastInst->getOperand(2));
FBB = LastInst->getOperand(0).getMachineBasicBlock(); FBB = LastInst->getOperand(0).getMachineBasicBlock();
return false; return false;
} }
@ -397,19 +399,21 @@ unsigned ARMInstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *T
// Shouldn't be a fall through. // Shouldn't be a fall through.
assert(TBB && "InsertBranch must not be told to insert a fallthrough"); assert(TBB && "InsertBranch must not be told to insert a fallthrough");
assert((Cond.size() == 1 || Cond.size() == 0) && assert((Cond.size() == 2 || Cond.size() == 0) &&
"ARM branch conditions have two components!"); "ARM branch conditions have two components!");
if (FBB == 0) { if (FBB == 0) {
if (Cond.empty()) // Unconditional branch? if (Cond.empty()) // Unconditional branch?
BuildMI(&MBB, get(BOpc)).addMBB(TBB); BuildMI(&MBB, get(BOpc)).addMBB(TBB);
else else
BuildMI(&MBB, get(BccOpc)).addMBB(TBB).addImm(Cond[0].getImm()); BuildMI(&MBB, get(BccOpc)).addMBB(TBB)
.addImm(Cond[0].getImm()).addReg(Cond[1].getReg());
return 1; return 1;
} }
// Two-way conditional branch. // Two-way conditional branch.
BuildMI(&MBB, get(BccOpc)).addMBB(TBB).addImm(Cond[0].getImm()); BuildMI(&MBB, get(BccOpc)).addMBB(TBB)
.addImm(Cond[0].getImm()).addReg(Cond[1].getReg());
BuildMI(&MBB, get(BOpc)).addMBB(FBB); BuildMI(&MBB, get(BOpc)).addMBB(FBB);
return 2; return 2;
} }
@ -452,6 +456,7 @@ bool ARMInstrInfo::PredicateInstruction(MachineInstr *MI,
if (Opc == ARM::B || Opc == ARM::tB) { if (Opc == ARM::B || Opc == ARM::tB) {
MI->setInstrDescriptor(get(Opc == ARM::B ? ARM::Bcc : ARM::tBcc)); MI->setInstrDescriptor(get(Opc == ARM::B ? ARM::Bcc : ARM::tBcc));
MI->addImmOperand(Pred[0].getImmedValue()); MI->addImmOperand(Pred[0].getImmedValue());
MI->addRegOperand(Pred[1].getReg(), false);
return true; return true;
} }
@ -459,6 +464,7 @@ bool ARMInstrInfo::PredicateInstruction(MachineInstr *MI,
if (PIdx != -1) { if (PIdx != -1) {
MachineOperand &PMO = MI->getOperand(PIdx); MachineOperand &PMO = MI->getOperand(PIdx);
PMO.setImm(Pred[0].getImmedValue()); PMO.setImm(Pred[0].getImmedValue());
MI->getOperand(PIdx+1).setReg(Pred[1].getReg());
return true; return true;
} }
return false; return false;
@ -467,7 +473,7 @@ bool ARMInstrInfo::PredicateInstruction(MachineInstr *MI,
bool bool
ARMInstrInfo::SubsumesPredicate(const std::vector<MachineOperand> &Pred1, ARMInstrInfo::SubsumesPredicate(const std::vector<MachineOperand> &Pred1,
const std::vector<MachineOperand> &Pred2) const{ const std::vector<MachineOperand> &Pred2) const{
if (Pred1.size() > 1 || Pred2.size() > 1) if (Pred1.size() > 2 || Pred2.size() > 2)
return false; return false;
ARMCC::CondCodes CC1 = (ARMCC::CondCodes)Pred1[0].getImmedValue(); ARMCC::CondCodes CC1 = (ARMCC::CondCodes)Pred1[0].getImmedValue();

106
lib/Target/ARM/ARMLoadStoreOptimizer.cpp
View File

@ -66,7 +66,8 @@ namespace {
SmallVector<MachineBasicBlock::iterator, 4> SmallVector<MachineBasicBlock::iterator, 4>
MergeLDR_STR(MachineBasicBlock &MBB, unsigned SIndex, unsigned Base, MergeLDR_STR(MachineBasicBlock &MBB, unsigned SIndex, unsigned Base,
int Opcode, unsigned Size, ARMCC::CondCodes Pred, int Opcode, unsigned Size,
ARMCC::CondCodes Pred, unsigned PredReg,
unsigned Scratch, MemOpQueue &MemOps); unsigned Scratch, MemOpQueue &MemOps);
void AdvanceRS(MachineBasicBlock &MBB, MemOpQueue &MemOps); void AdvanceRS(MachineBasicBlock &MBB, MemOpQueue &MemOps);
@ -112,7 +113,7 @@ static int getLoadStoreMultipleOpcode(int Opcode) {
/// It returns true if the transformation is done. /// It returns true if the transformation is done.
static bool mergeOps(MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, static bool mergeOps(MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI,
int Offset, unsigned Base, bool BaseKill, int Opcode, int Offset, unsigned Base, bool BaseKill, int Opcode,
ARMCC::CondCodes Pred, unsigned Scratch, ARMCC::CondCodes Pred, unsigned PredReg, unsigned Scratch,
SmallVector<std::pair<unsigned, bool>, 8> &Regs, SmallVector<std::pair<unsigned, bool>, 8> &Regs,
const TargetInstrInfo *TII) { const TargetInstrInfo *TII) {
// Only a single register to load / store. Don't bother. // Only a single register to load / store. Don't bother.
@ -156,7 +157,8 @@ static bool mergeOps(MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI,
return false; // Probably not worth it then. return false; // Probably not worth it then.
BuildMI(MBB, MBBI, TII->get(BaseOpc), NewBase) BuildMI(MBB, MBBI, TII->get(BaseOpc), NewBase)
.addReg(Base, false, false, BaseKill).addImm(ImmedOffset).addImm(Pred); .addReg(Base, false, false, BaseKill).addImm(ImmedOffset)
.addImm(Pred).addReg(PredReg);
Base = NewBase; Base = NewBase;
BaseKill = true; // New base is always killed right its use. BaseKill = true; // New base is always killed right its use.
} }
@ -166,10 +168,10 @@ static bool mergeOps(MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI,
Opcode = getLoadStoreMultipleOpcode(Opcode); Opcode = getLoadStoreMultipleOpcode(Opcode);
MachineInstrBuilder MIB = (isAM4) MachineInstrBuilder MIB = (isAM4)
? BuildMI(MBB, MBBI, TII->get(Opcode)).addReg(Base, false, false, BaseKill) ? BuildMI(MBB, MBBI, TII->get(Opcode)).addReg(Base, false, false, BaseKill)
.addImm(ARM_AM::getAM4ModeImm(Mode)).addImm(Pred) .addImm(ARM_AM::getAM4ModeImm(Mode)).addImm(Pred).addReg(PredReg)
: BuildMI(MBB, MBBI, TII->get(Opcode)).addReg(Base, false, false, BaseKill) : BuildMI(MBB, MBBI, TII->get(Opcode)).addReg(Base, false, false, BaseKill)
.addImm(ARM_AM::getAM5Opc(Mode, false, isDPR ? NumRegs<<1 : NumRegs)) .addImm(ARM_AM::getAM5Opc(Mode, false, isDPR ? NumRegs<<1 : NumRegs))
.addImm(Pred); .addImm(Pred).addReg(PredReg);
for (unsigned i = 0; i != NumRegs; ++i) for (unsigned i = 0; i != NumRegs; ++i)
MIB = MIB.addReg(Regs[i].first, isDef, false, Regs[i].second); MIB = MIB.addReg(Regs[i].first, isDef, false, Regs[i].second);
@ -181,8 +183,8 @@ static bool mergeOps(MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI,
SmallVector<MachineBasicBlock::iterator, 4> SmallVector<MachineBasicBlock::iterator, 4>
ARMLoadStoreOpt::MergeLDR_STR(MachineBasicBlock &MBB, unsigned SIndex, ARMLoadStoreOpt::MergeLDR_STR(MachineBasicBlock &MBB, unsigned SIndex,
unsigned Base, int Opcode, unsigned Size, unsigned Base, int Opcode, unsigned Size,
ARMCC::CondCodes Pred, unsigned Scratch, ARMCC::CondCodes Pred, unsigned PredReg,
MemOpQueue &MemOps) { unsigned Scratch, MemOpQueue &MemOps) {
SmallVector<MachineBasicBlock::iterator, 4> Merges; SmallVector<MachineBasicBlock::iterator, 4> Merges;
bool isAM4 = Opcode == ARM::LDR || Opcode == ARM::STR; bool isAM4 = Opcode == ARM::LDR || Opcode == ARM::STR;
int Offset = MemOps[SIndex].Offset; int Offset = MemOps[SIndex].Offset;
@ -209,8 +211,8 @@ ARMLoadStoreOpt::MergeLDR_STR(MachineBasicBlock &MBB, unsigned SIndex,
PRegNum = RegNum; PRegNum = RegNum;
} else { } else {
// Can't merge this in. Try merge the earlier ones first. // Can't merge this in. Try merge the earlier ones first.
if (mergeOps(MBB, ++Loc, SOffset, Base, false, Opcode, Pred, Scratch, if (mergeOps(MBB, ++Loc, SOffset, Base, false, Opcode, Pred, PredReg,
Regs, TII)) { Scratch, Regs, TII)) {
Merges.push_back(prior(Loc)); Merges.push_back(prior(Loc));
for (unsigned j = SIndex; j < i; ++j) { for (unsigned j = SIndex; j < i; ++j) {
MBB.erase(MemOps[j].MBBI); MBB.erase(MemOps[j].MBBI);
@ -218,7 +220,7 @@ ARMLoadStoreOpt::MergeLDR_STR(MachineBasicBlock &MBB, unsigned SIndex,
} }
} }
SmallVector<MachineBasicBlock::iterator, 4> Merges2 = SmallVector<MachineBasicBlock::iterator, 4> Merges2 =
MergeLDR_STR(MBB, i, Base, Opcode, Size, Pred, Scratch, MemOps); MergeLDR_STR(MBB, i, Base, Opcode, Size, Pred, PredReg, Scratch,MemOps);
Merges.append(Merges2.begin(), Merges2.end()); Merges.append(Merges2.begin(), Merges2.end());
return Merges; return Merges;
} }
@ -230,8 +232,8 @@ ARMLoadStoreOpt::MergeLDR_STR(MachineBasicBlock &MBB, unsigned SIndex,
} }
bool BaseKill = Loc->findRegisterUseOperandIdx(Base, true) != -1; bool BaseKill = Loc->findRegisterUseOperandIdx(Base, true) != -1;
if (mergeOps(MBB, ++Loc, SOffset, Base, BaseKill, Opcode, Pred, Scratch, if (mergeOps(MBB, ++Loc, SOffset, Base, BaseKill, Opcode, Pred, PredReg,
Regs, TII)) { Scratch, Regs, TII)) {
Merges.push_back(prior(Loc)); Merges.push_back(prior(Loc));
for (unsigned i = SIndex, e = MemOps.size(); i != e; ++i) { for (unsigned i = SIndex, e = MemOps.size(); i != e; ++i) {
MBB.erase(MemOps[i].MBBI); MBB.erase(MemOps[i].MBBI);
@ -243,29 +245,41 @@ ARMLoadStoreOpt::MergeLDR_STR(MachineBasicBlock &MBB, unsigned SIndex,
} }
/// getInstrPredicate - If instruction is predicated, returns its predicate /// getInstrPredicate - If instruction is predicated, returns its predicate
/// condition, otherwise returns AL. /// condition, otherwise returns AL. It also returns the condition code
static ARMCC::CondCodes getInstrPredicate(MachineInstr *MI) { /// register by reference.
static ARMCC::CondCodes getInstrPredicate(MachineInstr *MI, unsigned &PredReg) {
int PIdx = MI->findFirstPredOperandIdx(); int PIdx = MI->findFirstPredOperandIdx();
return PIdx == -1 ? ARMCC::AL if (PIdx == -1) {
: (ARMCC::CondCodes)MI->getOperand(PIdx).getImmedValue(); PredReg = 0;
return ARMCC::AL;
}
PredReg = MI->getOperand(PIdx+1).getReg();
return (ARMCC::CondCodes)MI->getOperand(PIdx).getImmedValue();
} }
static inline bool isMatchingDecrement(MachineInstr *MI, unsigned Base, static inline bool isMatchingDecrement(MachineInstr *MI, unsigned Base,
unsigned Bytes, ARMCC::CondCodes Pred) { unsigned Bytes, ARMCC::CondCodes Pred,
unsigned PredReg) {
unsigned MyPredReg = 0;
return (MI && MI->getOpcode() == ARM::SUBri && return (MI && MI->getOpcode() == ARM::SUBri &&
MI->getOperand(0).getReg() == Base && MI->getOperand(0).getReg() == Base &&
MI->getOperand(1).getReg() == Base && MI->getOperand(1).getReg() == Base &&
ARM_AM::getAM2Offset(MI->getOperand(2).getImm()) == Bytes && ARM_AM::getAM2Offset(MI->getOperand(2).getImm()) == Bytes &&
getInstrPredicate(MI) == Pred); getInstrPredicate(MI, MyPredReg) == Pred &&
MyPredReg == PredReg);
} }
static inline bool isMatchingIncrement(MachineInstr *MI, unsigned Base, static inline bool isMatchingIncrement(MachineInstr *MI, unsigned Base,
unsigned Bytes, ARMCC::CondCodes Pred) { unsigned Bytes, ARMCC::CondCodes Pred,
unsigned PredReg) {
unsigned MyPredReg = 0;
return (MI && MI->getOpcode() == ARM::ADDri && return (MI && MI->getOpcode() == ARM::ADDri &&
MI->getOperand(0).getReg() == Base && MI->getOperand(0).getReg() == Base &&
MI->getOperand(1).getReg() == Base && MI->getOperand(1).getReg() == Base &&
ARM_AM::getAM2Offset(MI->getOperand(2).getImm()) == Bytes && ARM_AM::getAM2Offset(MI->getOperand(2).getImm()) == Bytes &&
getInstrPredicate(MI) == Pred); getInstrPredicate(MI, MyPredReg) == Pred &&
MyPredReg == PredReg);
} }
static inline unsigned getLSMultipleTransferSize(MachineInstr *MI) { static inline unsigned getLSMultipleTransferSize(MachineInstr *MI) {
@ -281,7 +295,7 @@ static inline unsigned getLSMultipleTransferSize(MachineInstr *MI) {
return 8; return 8;
case ARM::LDM: case ARM::LDM:
case ARM::STM: case ARM::STM:
return (MI->getNumOperands() - 3) * 4; return (MI->getNumOperands() - 4) * 4;
case ARM::FLDMS: case ARM::FLDMS:
case ARM::FSTMS: case ARM::FSTMS:
case ARM::FLDMD: case ARM::FLDMD:
@ -307,7 +321,8 @@ static bool mergeBaseUpdateLSMultiple(MachineBasicBlock &MBB,
MachineInstr *MI = MBBI; MachineInstr *MI = MBBI;
unsigned Base = MI->getOperand(0).getReg(); unsigned Base = MI->getOperand(0).getReg();
unsigned Bytes = getLSMultipleTransferSize(MI); unsigned Bytes = getLSMultipleTransferSize(MI);
ARMCC::CondCodes Pred = getInstrPredicate(MI); unsigned PredReg = 0;
ARMCC::CondCodes Pred = getInstrPredicate(MI, PredReg);
int Opcode = MI->getOpcode(); int Opcode = MI->getOpcode();
bool isAM4 = Opcode == ARM::LDM || Opcode == ARM::STM; bool isAM4 = Opcode == ARM::LDM || Opcode == ARM::STM;
@ -326,12 +341,12 @@ static bool mergeBaseUpdateLSMultiple(MachineBasicBlock &MBB,
if (MBBI != MBB.begin()) { if (MBBI != MBB.begin()) {
MachineBasicBlock::iterator PrevMBBI = prior(MBBI); MachineBasicBlock::iterator PrevMBBI = prior(MBBI);
if (Mode == ARM_AM::ia && if (Mode == ARM_AM::ia &&
isMatchingDecrement(PrevMBBI, Base, Bytes, Pred)) { isMatchingDecrement(PrevMBBI, Base, Bytes, Pred, PredReg)) {
MI->getOperand(1).setImm(ARM_AM::getAM4ModeImm(ARM_AM::db, true)); MI->getOperand(1).setImm(ARM_AM::getAM4ModeImm(ARM_AM::db, true));
MBB.erase(PrevMBBI); MBB.erase(PrevMBBI);
return true; return true;
} else if (Mode == ARM_AM::ib && } else if (Mode == ARM_AM::ib &&
isMatchingDecrement(PrevMBBI, Base, Bytes, Pred)) { isMatchingDecrement(PrevMBBI, Base, Bytes, Pred, PredReg)) {
MI->getOperand(1).setImm(ARM_AM::getAM4ModeImm(ARM_AM::da, true)); MI->getOperand(1).setImm(ARM_AM::getAM4ModeImm(ARM_AM::da, true));
MBB.erase(PrevMBBI); MBB.erase(PrevMBBI);
return true; return true;
@ -341,12 +356,12 @@ static bool mergeBaseUpdateLSMultiple(MachineBasicBlock &MBB,
if (MBBI != MBB.end()) { if (MBBI != MBB.end()) {
MachineBasicBlock::iterator NextMBBI = next(MBBI); MachineBasicBlock::iterator NextMBBI = next(MBBI);
if ((Mode == ARM_AM::ia || Mode == ARM_AM::ib) && if ((Mode == ARM_AM::ia || Mode == ARM_AM::ib) &&
isMatchingIncrement(NextMBBI, Base, Bytes, Pred)) { isMatchingIncrement(NextMBBI, Base, Bytes, Pred, PredReg)) {
MI->getOperand(1).setImm(ARM_AM::getAM4ModeImm(Mode, true)); MI->getOperand(1).setImm(ARM_AM::getAM4ModeImm(Mode, true));
MBB.erase(NextMBBI); MBB.erase(NextMBBI);
return true; return true;
} else if ((Mode == ARM_AM::da || Mode == ARM_AM::db) && } else if ((Mode == ARM_AM::da || Mode == ARM_AM::db) &&
isMatchingDecrement(NextMBBI, Base, Bytes, Pred)) { isMatchingDecrement(NextMBBI, Base, Bytes, Pred, PredReg)) {
MI->getOperand(1).setImm(ARM_AM::getAM4ModeImm(Mode, true)); MI->getOperand(1).setImm(ARM_AM::getAM4ModeImm(Mode, true));
MBB.erase(NextMBBI); MBB.erase(NextMBBI);
return true; return true;
@ -362,7 +377,7 @@ static bool mergeBaseUpdateLSMultiple(MachineBasicBlock &MBB,
if (MBBI != MBB.begin()) { if (MBBI != MBB.begin()) {
MachineBasicBlock::iterator PrevMBBI = prior(MBBI); MachineBasicBlock::iterator PrevMBBI = prior(MBBI);
if (Mode == ARM_AM::ia && if (Mode == ARM_AM::ia &&
isMatchingDecrement(PrevMBBI, Base, Bytes, Pred)) { isMatchingDecrement(PrevMBBI, Base, Bytes, Pred, PredReg)) {
MI->getOperand(1).setImm(ARM_AM::getAM5Opc(ARM_AM::db, true, Offset)); MI->getOperand(1).setImm(ARM_AM::getAM5Opc(ARM_AM::db, true, Offset));
MBB.erase(PrevMBBI); MBB.erase(PrevMBBI);
return true; return true;
@ -372,7 +387,7 @@ static bool mergeBaseUpdateLSMultiple(MachineBasicBlock &MBB,
if (MBBI != MBB.end()) { if (MBBI != MBB.end()) {
MachineBasicBlock::iterator NextMBBI = next(MBBI); MachineBasicBlock::iterator NextMBBI = next(MBBI);
if (Mode == ARM_AM::ia && if (Mode == ARM_AM::ia &&
isMatchingIncrement(NextMBBI, Base, Bytes, Pred)) { isMatchingIncrement(NextMBBI, Base, Bytes, Pred, PredReg)) {
MI->getOperand(1).setImm(ARM_AM::getAM5Opc(ARM_AM::ia, true, Offset)); MI->getOperand(1).setImm(ARM_AM::getAM5Opc(ARM_AM::ia, true, Offset));
MBB.erase(NextMBBI); MBB.erase(NextMBBI);
} }
@ -430,17 +445,19 @@ static bool mergeBaseUpdateLoadStore(MachineBasicBlock &MBB,
if (isLd && MI->getOperand(0).getReg() == Base) if (isLd && MI->getOperand(0).getReg() == Base)
return false; return false;
ARMCC::CondCodes Pred = getInstrPredicate(MI); unsigned PredReg = 0;
ARMCC::CondCodes Pred = getInstrPredicate(MI, PredReg);
bool DoMerge = false; bool DoMerge = false;
ARM_AM::AddrOpc AddSub = ARM_AM::add; ARM_AM::AddrOpc AddSub = ARM_AM::add;
unsigned NewOpc = 0; unsigned NewOpc = 0;
if (MBBI != MBB.begin()) { if (MBBI != MBB.begin()) {
MachineBasicBlock::iterator PrevMBBI = prior(MBBI); MachineBasicBlock::iterator PrevMBBI = prior(MBBI);
if (isMatchingDecrement(PrevMBBI, Base, Bytes, Pred)) { if (isMatchingDecrement(PrevMBBI, Base, Bytes, Pred, PredReg)) {
DoMerge = true; DoMerge = true;
AddSub = ARM_AM::sub; AddSub = ARM_AM::sub;
NewOpc = getPreIndexedLoadStoreOpcode(Opcode); NewOpc = getPreIndexedLoadStoreOpcode(Opcode);
} else if (isAM2 && isMatchingIncrement(PrevMBBI, Base, Bytes, Pred)) { } else if (isAM2 && isMatchingIncrement(PrevMBBI, Base, Bytes,
Pred, PredReg)) {
DoMerge = true; DoMerge = true;
NewOpc = getPreIndexedLoadStoreOpcode(Opcode); NewOpc = getPreIndexedLoadStoreOpcode(Opcode);
} }
@ -450,11 +467,11 @@ static bool mergeBaseUpdateLoadStore(MachineBasicBlock &MBB,
if (!DoMerge && MBBI != MBB.end()) { if (!DoMerge && MBBI != MBB.end()) {
MachineBasicBlock::iterator NextMBBI = next(MBBI); MachineBasicBlock::iterator NextMBBI = next(MBBI);
if (isAM2 && isMatchingDecrement(NextMBBI, Base, Bytes, Pred)) { if (isAM2 && isMatchingDecrement(NextMBBI, Base, Bytes, Pred, PredReg)) {
DoMerge = true; DoMerge = true;
AddSub = ARM_AM::sub; AddSub = ARM_AM::sub;
NewOpc = getPostIndexedLoadStoreOpcode(Opcode); NewOpc = getPostIndexedLoadStoreOpcode(Opcode);
} else if (isMatchingIncrement(NextMBBI, Base, Bytes, Pred)) { } else if (isMatchingIncrement(NextMBBI, Base, Bytes, Pred, PredReg)) {
DoMerge = true; DoMerge = true;
NewOpc = getPostIndexedLoadStoreOpcode(Opcode); NewOpc = getPostIndexedLoadStoreOpcode(Opcode);
} }
@ -474,22 +491,24 @@ static bool mergeBaseUpdateLoadStore(MachineBasicBlock &MBB,
// LDR_PRE, LDR_POST; // LDR_PRE, LDR_POST;
BuildMI(MBB, MBBI, TII->get(NewOpc), MI->getOperand(0).getReg()) BuildMI(MBB, MBBI, TII->get(NewOpc), MI->getOperand(0).getReg())
.addReg(Base, true) .addReg(Base, true)
.addReg(Base).addReg(0).addImm(Offset).addImm(Pred); .addReg(Base).addReg(0).addImm(Offset).addImm(Pred).addReg(PredReg);
else else
// FLDMS, FLDMD // FLDMS, FLDMD
BuildMI(MBB, MBBI, TII->get(NewOpc)).addReg(Base, false, false, BaseKill) BuildMI(MBB, MBBI, TII->get(NewOpc)).addReg(Base, false, false, BaseKill)
.addImm(Offset).addImm(Pred).addReg(MI->getOperand(0).getReg(), true); .addImm(Offset).addImm(Pred).addReg(PredReg)
.addReg(MI->getOperand(0).getReg(), true);
} else { } else {
MachineOperand &MO = MI->getOperand(0); MachineOperand &MO = MI->getOperand(0);
if (isAM2) if (isAM2)
// STR_PRE, STR_POST; // STR_PRE, STR_POST;
BuildMI(MBB, MBBI, TII->get(NewOpc), Base) BuildMI(MBB, MBBI, TII->get(NewOpc), Base)
.addReg(MO.getReg(), false, false, MO.isKill()) .addReg(MO.getReg(), false, false, MO.isKill())
.addReg(Base).addReg(0).addImm(Offset).addImm(Pred); .addReg(Base).addReg(0).addImm(Offset).addImm(Pred).addReg(PredReg);
else else
// FSTMS, FSTMD // FSTMS, FSTMD
BuildMI(MBB, MBBI, TII->get(NewOpc)).addReg(Base).addImm(Offset) BuildMI(MBB, MBBI, TII->get(NewOpc)).addReg(Base).addImm(Offset)
.addImm(Pred).addReg(MO.getReg(), false, false, MO.isKill()); .addImm(Pred).addReg(PredReg)
.addReg(MO.getReg(), false, false, MO.isKill());
} }
MBB.erase(MBBI); MBB.erase(MBBI);
@ -541,6 +560,7 @@ bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) {
int CurrOpc = -1; int CurrOpc = -1;
unsigned CurrSize = 0; unsigned CurrSize = 0;
ARMCC::CondCodes CurrPred = ARMCC::AL; ARMCC::CondCodes CurrPred = ARMCC::AL;
unsigned CurrPredReg = 0;
unsigned Position = 0; unsigned Position = 0;
RS->enterBasicBlock(&MBB); RS->enterBasicBlock(&MBB);
@ -556,9 +576,10 @@ bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) {
bool isAM2 = Opcode == ARM::LDR || Opcode == ARM::STR; bool isAM2 = Opcode == ARM::LDR || Opcode == ARM::STR;
unsigned Size = getLSMultipleTransferSize(MBBI); unsigned Size = getLSMultipleTransferSize(MBBI);
unsigned Base = MBBI->getOperand(1).getReg(); unsigned Base = MBBI->getOperand(1).getReg();
ARMCC::CondCodes Pred = getInstrPredicate(MBBI); unsigned PredReg = 0;
ARMCC::CondCodes Pred = getInstrPredicate(MBBI, PredReg);
const TargetInstrDescriptor *TID = MBBI->getInstrDescriptor(); const TargetInstrDescriptor *TID = MBBI->getInstrDescriptor();
unsigned OffField = MBBI->getOperand(TID->numOperands-2).getImm(); unsigned OffField = MBBI->getOperand(TID->numOperands-3).getImm();
int Offset = isAM2 int Offset = isAM2
? ARM_AM::getAM2Offset(OffField) : ARM_AM::getAM5Offset(OffField) * 4; ? ARM_AM::getAM2Offset(OffField) : ARM_AM::getAM5Offset(OffField) * 4;
if (isAM2) { if (isAM2) {
@ -584,6 +605,7 @@ bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) {
CurrOpc = Opcode; CurrOpc = Opcode;
CurrSize = Size; CurrSize = Size;
CurrPred = Pred; CurrPred = Pred;
CurrPredReg = PredReg;
MemOps.push_back(MemOpQueueEntry(Offset, Position, MBBI)); MemOps.push_back(MemOpQueueEntry(Offset, Position, MBBI));
NumMemOps++; NumMemOps++;
Advance = true; Advance = true;
@ -594,6 +616,7 @@ bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) {
} }
if (CurrOpc == Opcode && CurrBase == Base && CurrPred == Pred) { if (CurrOpc == Opcode && CurrBase == Base && CurrPred == Pred) {
// No need to match PredReg.
// Continue adding to the queue. // Continue adding to the queue.
if (Offset > MemOps.back().Offset) { if (Offset > MemOps.back().Offset) {
MemOps.push_back(MemOpQueueEntry(Offset, Position, MBBI)); MemOps.push_back(MemOpQueueEntry(Offset, Position, MBBI));
@ -639,8 +662,8 @@ bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) {
// Merge ops. // Merge ops.
SmallVector<MachineBasicBlock::iterator,4> MBBII = SmallVector<MachineBasicBlock::iterator,4> MBBII =
MergeLDR_STR(MBB, 0, CurrBase, CurrOpc, CurrSize, CurrPred, MergeLDR_STR(MBB, 0, CurrBase, CurrOpc, CurrSize,
Scratch, MemOps); CurrPred, CurrPredReg, Scratch, MemOps);
// Try folding preceeding/trailing base inc/dec into the generated // Try folding preceeding/trailing base inc/dec into the generated
// LDM/STM ops. // LDM/STM ops.
@ -664,6 +687,7 @@ bool ARMLoadStoreOpt::LoadStoreMultipleOpti(MachineBasicBlock &MBB) {
CurrOpc = -1; CurrOpc = -1;
CurrSize = 0; CurrSize = 0;
CurrPred = ARMCC::AL; CurrPred = ARMCC::AL;
CurrPredReg = 0;
if (NumMemOps) { if (NumMemOps) {
MemOps.clear(); MemOps.clear();
NumMemOps = 0; NumMemOps = 0;