X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FTarget%2FPowerPC%2FPPCRegisterInfo.cpp;h=6f364bc05771adceac900a8127e80b3f738f16e8;hp=c90889aeeeba57be8d59daa1cd134ff087a9fa9a;hb=f606a6ed992d5e4e2877419b51e2a9b540b5e3f0;hpb=2e95afa04cd1c89de7e4abaeead66cd8e51ec929 diff --git a/lib/Target/PowerPC/PPCRegisterInfo.cpp b/lib/Target/PowerPC/PPCRegisterInfo.cpp index c90889aeeeb..6f364bc0577 100644 --- a/lib/Target/PowerPC/PPCRegisterInfo.cpp +++ b/lib/Target/PowerPC/PPCRegisterInfo.cpp @@ -1,4 +1,4 @@ -//===- PPCRegisterInfo.cpp - PowerPC Register Information -------*- C++ -*-===// +//===-- PPCRegisterInfo.cpp - PowerPC Register Information ----------------===// // // The LLVM Compiler Infrastructure // @@ -12,66 +12,56 @@ // //===----------------------------------------------------------------------===// -#define DEBUG_TYPE "reginfo" +#include "PPCRegisterInfo.h" #include "PPC.h" +#include "PPCFrameLowering.h" #include "PPCInstrBuilder.h" #include "PPCMachineFunctionInfo.h" -#include "PPCRegisterInfo.h" -#include "PPCFrameLowering.h" #include "PPCSubtarget.h" -#include "llvm/CallingConv.h" -#include "llvm/Constants.h" -#include "llvm/Function.h" -#include "llvm/Type.h" -#include "llvm/CodeGen/ValueTypes.h" +#include "PPCTargetMachine.h" +#include "llvm/ADT/BitVector.h" +#include "llvm/ADT/STLExtras.h" +#include "llvm/CodeGen/MachineFrameInfo.h" +#include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineInstrBuilder.h" #include "llvm/CodeGen/MachineModuleInfo.h" -#include "llvm/CodeGen/MachineFunction.h" -#include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineRegisterInfo.h" #include "llvm/CodeGen/RegisterScavenging.h" -#include "llvm/Target/TargetFrameLowering.h" -#include "llvm/Target/TargetInstrInfo.h" -#include "llvm/Target/TargetMachine.h" -#include "llvm/Target/TargetOptions.h" +#include "llvm/IR/CallingConv.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/Function.h" +#include "llvm/IR/Type.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/MathExtras.h" #include "llvm/Support/raw_ostream.h" -#include "llvm/ADT/BitVector.h" -#include "llvm/ADT/STLExtras.h" +#include "llvm/Target/TargetFrameLowering.h" +#include "llvm/Target/TargetInstrInfo.h" +#include "llvm/Target/TargetMachine.h" +#include "llvm/Target/TargetOptions.h" #include +using namespace llvm; + +#define DEBUG_TYPE "reginfo" + #define GET_REGINFO_TARGET_DESC #include "PPCGenRegisterInfo.inc" -namespace llvm { -cl::opt DisablePPC32RS("disable-ppc32-regscavenger", - cl::init(false), - cl::desc("Disable PPC32 register scavenger"), - cl::Hidden); -cl::opt DisablePPC64RS("disable-ppc64-regscavenger", - cl::init(false), - cl::desc("Disable PPC64 register scavenger"), - cl::Hidden); -} - -using namespace llvm; +static cl::opt +EnableBasePointer("ppc-use-base-pointer", cl::Hidden, cl::init(true), + cl::desc("Enable use of a base pointer for complex stack frames")); -// FIXME (64-bit): Should be inlined. -bool -PPCRegisterInfo::requiresRegisterScavenging(const MachineFunction &) const { - return ((!DisablePPC32RS && !Subtarget.isPPC64()) || - (!DisablePPC64RS && Subtarget.isPPC64())); -} +static cl::opt +AlwaysBasePointer("ppc-always-use-base-pointer", cl::Hidden, cl::init(false), + cl::desc("Force the use of a base pointer in every function")); -PPCRegisterInfo::PPCRegisterInfo(const PPCSubtarget &ST, - const TargetInstrInfo &tii) - : PPCGenRegisterInfo(ST.isPPC64() ? PPC::LR8 : PPC::LR, - ST.isPPC64() ? 0 : 1, - ST.isPPC64() ? 0 : 1), - Subtarget(ST), TII(tii) { +PPCRegisterInfo::PPCRegisterInfo(const PPCTargetMachine &TM) + : PPCGenRegisterInfo(TM.isPPC64() ? PPC::LR8 : PPC::LR, + TM.isPPC64() ? 0 : 1, + TM.isPPC64() ? 0 : 1), + TM(TM) { ImmToIdxMap[PPC::LD] = PPC::LDX; ImmToIdxMap[PPC::STD] = PPC::STDX; ImmToIdxMap[PPC::LBZ] = PPC::LBZX; ImmToIdxMap[PPC::STB] = PPC::STBX; ImmToIdxMap[PPC::LHZ] = PPC::LHZX; ImmToIdxMap[PPC::LHA] = PPC::LHAX; @@ -80,188 +70,199 @@ PPCRegisterInfo::PPCRegisterInfo(const PPCSubtarget &ST, ImmToIdxMap[PPC::STH] = PPC::STHX; ImmToIdxMap[PPC::STW] = PPC::STWX; ImmToIdxMap[PPC::STFS] = PPC::STFSX; ImmToIdxMap[PPC::STFD] = PPC::STFDX; ImmToIdxMap[PPC::ADDI] = PPC::ADD4; + ImmToIdxMap[PPC::LWA_32] = PPC::LWAX_32; // 64-bit ImmToIdxMap[PPC::LHA8] = PPC::LHAX8; ImmToIdxMap[PPC::LBZ8] = PPC::LBZX8; ImmToIdxMap[PPC::LHZ8] = PPC::LHZX8; ImmToIdxMap[PPC::LWZ8] = PPC::LWZX8; ImmToIdxMap[PPC::STB8] = PPC::STBX8; ImmToIdxMap[PPC::STH8] = PPC::STHX8; ImmToIdxMap[PPC::STW8] = PPC::STWX8; ImmToIdxMap[PPC::STDU] = PPC::STDUX; - ImmToIdxMap[PPC::ADDI8] = PPC::ADD8; ImmToIdxMap[PPC::STD_32] = PPC::STDX_32; + ImmToIdxMap[PPC::ADDI8] = PPC::ADD8; } /// getPointerRegClass - Return the register class to use to hold pointers. /// This is used for addressing modes. const TargetRegisterClass * -PPCRegisterInfo::getPointerRegClass(unsigned Kind) const { - if (Subtarget.isPPC64()) +PPCRegisterInfo::getPointerRegClass(const MachineFunction &MF, unsigned Kind) + const { + // Note that PPCInstrInfo::FoldImmediate also directly uses this Kind value + // when it checks for ZERO folding. + if (Kind == 1) { + if (TM.isPPC64()) + return &PPC::G8RC_NOX0RegClass; + return &PPC::GPRC_NOR0RegClass; + } + + if (TM.isPPC64()) return &PPC::G8RCRegClass; return &PPC::GPRCRegClass; } -const unsigned* +const MCPhysReg* PPCRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const { - // 32-bit Darwin calling convention. - static const unsigned Darwin32_CalleeSavedRegs[] = { - PPC::R13, PPC::R14, PPC::R15, - PPC::R16, PPC::R17, PPC::R18, PPC::R19, - PPC::R20, PPC::R21, PPC::R22, PPC::R23, - PPC::R24, PPC::R25, PPC::R26, PPC::R27, - PPC::R28, PPC::R29, PPC::R30, PPC::R31, - - PPC::F14, PPC::F15, PPC::F16, PPC::F17, - PPC::F18, PPC::F19, PPC::F20, PPC::F21, - PPC::F22, PPC::F23, PPC::F24, PPC::F25, - PPC::F26, PPC::F27, PPC::F28, PPC::F29, - PPC::F30, PPC::F31, - - PPC::CR2, PPC::CR3, PPC::CR4, - PPC::V20, PPC::V21, PPC::V22, PPC::V23, - PPC::V24, PPC::V25, PPC::V26, PPC::V27, - PPC::V28, PPC::V29, PPC::V30, PPC::V31, - - PPC::LR, 0 - }; - - // 32-bit SVR4 calling convention. - static const unsigned SVR4_CalleeSavedRegs[] = { - PPC::R14, PPC::R15, - PPC::R16, PPC::R17, PPC::R18, PPC::R19, - PPC::R20, PPC::R21, PPC::R22, PPC::R23, - PPC::R24, PPC::R25, PPC::R26, PPC::R27, - PPC::R28, PPC::R29, PPC::R30, PPC::R31, - - PPC::F14, PPC::F15, PPC::F16, PPC::F17, - PPC::F18, PPC::F19, PPC::F20, PPC::F21, - PPC::F22, PPC::F23, PPC::F24, PPC::F25, - PPC::F26, PPC::F27, PPC::F28, PPC::F29, - PPC::F30, PPC::F31, - - PPC::CR2, PPC::CR3, PPC::CR4, - - PPC::VRSAVE, - - PPC::V20, PPC::V21, PPC::V22, PPC::V23, - PPC::V24, PPC::V25, PPC::V26, PPC::V27, - PPC::V28, PPC::V29, PPC::V30, PPC::V31, - - 0 - }; - // 64-bit Darwin calling convention. - static const unsigned Darwin64_CalleeSavedRegs[] = { - PPC::X14, PPC::X15, - PPC::X16, PPC::X17, PPC::X18, PPC::X19, - PPC::X20, PPC::X21, PPC::X22, PPC::X23, - PPC::X24, PPC::X25, PPC::X26, PPC::X27, - PPC::X28, PPC::X29, PPC::X30, PPC::X31, - - PPC::F14, PPC::F15, PPC::F16, PPC::F17, - PPC::F18, PPC::F19, PPC::F20, PPC::F21, - PPC::F22, PPC::F23, PPC::F24, PPC::F25, - PPC::F26, PPC::F27, PPC::F28, PPC::F29, - PPC::F30, PPC::F31, - - PPC::CR2, PPC::CR3, PPC::CR4, - PPC::V20, PPC::V21, PPC::V22, PPC::V23, - PPC::V24, PPC::V25, PPC::V26, PPC::V27, - PPC::V28, PPC::V29, PPC::V30, PPC::V31, - - PPC::LR8, 0 - }; - - // 64-bit SVR4 calling convention. - static const unsigned SVR4_64_CalleeSavedRegs[] = { - PPC::X14, PPC::X15, - PPC::X16, PPC::X17, PPC::X18, PPC::X19, - PPC::X20, PPC::X21, PPC::X22, PPC::X23, - PPC::X24, PPC::X25, PPC::X26, PPC::X27, - PPC::X28, PPC::X29, PPC::X30, PPC::X31, - - PPC::F14, PPC::F15, PPC::F16, PPC::F17, - PPC::F18, PPC::F19, PPC::F20, PPC::F21, - PPC::F22, PPC::F23, PPC::F24, PPC::F25, - PPC::F26, PPC::F27, PPC::F28, PPC::F29, - PPC::F30, PPC::F31, - - PPC::CR2, PPC::CR3, PPC::CR4, + const PPCSubtarget &Subtarget = MF->getSubtarget(); + if (MF->getFunction()->getCallingConv() == CallingConv::AnyReg) { + if (Subtarget.hasVSX()) + return CSR_64_AllRegs_VSX_SaveList; + if (Subtarget.hasAltivec()) + return CSR_64_AllRegs_Altivec_SaveList; + return CSR_64_AllRegs_SaveList; + } - PPC::VRSAVE, + if (Subtarget.isDarwinABI()) + return TM.isPPC64() + ? (Subtarget.hasAltivec() ? CSR_Darwin64_Altivec_SaveList + : CSR_Darwin64_SaveList) + : (Subtarget.hasAltivec() ? CSR_Darwin32_Altivec_SaveList + : CSR_Darwin32_SaveList); + + // On PPC64, we might need to save r2 (but only if it is not reserved). + bool SaveR2 = MF->getRegInfo().isAllocatable(PPC::X2); + + return TM.isPPC64() + ? (Subtarget.hasAltivec() + ? (SaveR2 ? CSR_SVR464_R2_Altivec_SaveList + : CSR_SVR464_Altivec_SaveList) + : (SaveR2 ? CSR_SVR464_R2_SaveList : CSR_SVR464_SaveList)) + : (Subtarget.hasAltivec() ? CSR_SVR432_Altivec_SaveList + : CSR_SVR432_SaveList); +} - PPC::V20, PPC::V21, PPC::V22, PPC::V23, - PPC::V24, PPC::V25, PPC::V26, PPC::V27, - PPC::V28, PPC::V29, PPC::V30, PPC::V31, +const uint32_t * +PPCRegisterInfo::getCallPreservedMask(const MachineFunction &MF, + CallingConv::ID CC) const { + const PPCSubtarget &Subtarget = MF.getSubtarget(); + if (CC == CallingConv::AnyReg) { + if (Subtarget.hasVSX()) + return CSR_64_AllRegs_VSX_RegMask; + if (Subtarget.hasAltivec()) + return CSR_64_AllRegs_Altivec_RegMask; + return CSR_64_AllRegs_RegMask; + } - 0 - }; - if (Subtarget.isDarwinABI()) - return Subtarget.isPPC64() ? Darwin64_CalleeSavedRegs : - Darwin32_CalleeSavedRegs; + return TM.isPPC64() ? (Subtarget.hasAltivec() ? CSR_Darwin64_Altivec_RegMask + : CSR_Darwin64_RegMask) + : (Subtarget.hasAltivec() ? CSR_Darwin32_Altivec_RegMask + : CSR_Darwin32_RegMask); + + return TM.isPPC64() ? (Subtarget.hasAltivec() ? CSR_SVR464_Altivec_RegMask + : CSR_SVR464_RegMask) + : (Subtarget.hasAltivec() ? CSR_SVR432_Altivec_RegMask + : CSR_SVR432_RegMask); +} + +const uint32_t* +PPCRegisterInfo::getNoPreservedMask() const { + return CSR_NoRegs_RegMask; +} - return Subtarget.isPPC64() ? SVR4_64_CalleeSavedRegs : SVR4_CalleeSavedRegs; +void PPCRegisterInfo::adjustStackMapLiveOutMask(uint32_t *Mask) const { + for (unsigned PseudoReg : {PPC::ZERO, PPC::ZERO8, PPC::RM}) + Mask[PseudoReg / 32] &= ~(1u << (PseudoReg % 32)); } BitVector PPCRegisterInfo::getReservedRegs(const MachineFunction &MF) const { BitVector Reserved(getNumRegs()); - const PPCFrameLowering *PPCFI = - static_cast(MF.getTarget().getFrameLowering()); + const PPCSubtarget &Subtarget = MF.getSubtarget(); + const PPCFrameLowering *TFI = getFrameLowering(MF); + + // The ZERO register is not really a register, but the representation of r0 + // when used in instructions that treat r0 as the constant 0. + Reserved.set(PPC::ZERO); + Reserved.set(PPC::ZERO8); + + // The FP register is also not really a register, but is the representation + // of the frame pointer register used by ISD::FRAMEADDR. + Reserved.set(PPC::FP); + Reserved.set(PPC::FP8); + + // The BP register is also not really a register, but is the representation + // of the base pointer register used by setjmp. + Reserved.set(PPC::BP); + Reserved.set(PPC::BP8); + + // The counter registers must be reserved so that counter-based loops can + // be correctly formed (and the mtctr instructions are not DCE'd). + Reserved.set(PPC::CTR); + Reserved.set(PPC::CTR8); - Reserved.set(PPC::R0); Reserved.set(PPC::R1); Reserved.set(PPC::LR); Reserved.set(PPC::LR8); Reserved.set(PPC::RM); + if (!Subtarget.isDarwinABI() || !Subtarget.hasAltivec()) + Reserved.set(PPC::VRSAVE); + // The SVR4 ABI reserves r2 and r13 if (Subtarget.isSVR4ABI()) { Reserved.set(PPC::R2); // System-reserved register Reserved.set(PPC::R13); // Small Data Area pointer register } - // Reserve R2 on Darwin to hack around the problem of save/restore of CR - // when the stack frame is too big to address directly; we need two regs. - // This is a hack. - if (Subtarget.isDarwinABI()) { - Reserved.set(PPC::R2); - } // On PPC64, r13 is the thread pointer. Never allocate this register. - // Note that this is over conservative, as it also prevents allocation of R31 - // when the FP is not needed. - if (Subtarget.isPPC64()) { + if (TM.isPPC64()) { Reserved.set(PPC::R13); - Reserved.set(PPC::R31); - Reserved.set(PPC::X0); Reserved.set(PPC::X1); Reserved.set(PPC::X13); - Reserved.set(PPC::X31); + + if (TFI->needsFP(MF)) + Reserved.set(PPC::X31); + + if (hasBasePointer(MF)) + Reserved.set(PPC::X30); // The 64-bit SVR4 ABI reserves r2 for the TOC pointer. if (Subtarget.isSVR4ABI()) { - Reserved.set(PPC::X2); - } - // Reserve X2 on Darwin to hack around the problem of save/restore of CR - // when the stack frame is too big to address directly; we need two regs. - // This is a hack. - if (Subtarget.isDarwinABI()) { - Reserved.set(PPC::X2); + // We only reserve r2 if we need to use the TOC pointer. If we have no + // explicit uses of the TOC pointer (meaning we're a leaf function with + // no constant-pool loads, etc.) and we have no potential uses inside an + // inline asm block, then we can treat r2 has an ordinary callee-saved + // register. + const PPCFunctionInfo *FuncInfo = MF.getInfo(); + if (FuncInfo->usesTOCBasePtr() || MF.hasInlineAsm()) + Reserved.set(PPC::X2); + else + Reserved.reset(PPC::R2); } } - if (PPCFI->needsFP(MF)) + if (TFI->needsFP(MF)) Reserved.set(PPC::R31); + if (hasBasePointer(MF)) { + if (Subtarget.isSVR4ABI() && !TM.isPPC64() && + TM.getRelocationModel() == Reloc::PIC_) + Reserved.set(PPC::R29); + else + Reserved.set(PPC::R30); + } + + if (Subtarget.isSVR4ABI() && !TM.isPPC64() && + TM.getRelocationModel() == Reloc::PIC_) + Reserved.set(PPC::R30); + + // Reserve Altivec registers when Altivec is unavailable. + if (!Subtarget.hasAltivec()) + for (TargetRegisterClass::iterator I = PPC::VRRCRegClass.begin(), + IE = PPC::VRRCRegClass.end(); I != IE; ++I) + Reserved.set(*I); + return Reserved; } -unsigned -PPCRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC, - MachineFunction &MF) const { - const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); +unsigned PPCRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC, + MachineFunction &MF) const { + const PPCFrameLowering *TFI = getFrameLowering(MF); const unsigned DefaultSafety = 1; switch (RC->getID()) { default: return 0; + case PPC::G8RC_NOX0RegClassID: + case PPC::GPRC_NOR0RegClassID: case PPC::G8RCRegClassID: case PPC::GPRCRegClassID: { unsigned FP = TFI->hasFP(MF) ? 1 : 0; @@ -269,70 +270,46 @@ PPCRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC, } case PPC::F8RCRegClassID: case PPC::F4RCRegClassID: + case PPC::QFRCRegClassID: + case PPC::QSRCRegClassID: + case PPC::QBRCRegClassID: case PPC::VRRCRegClassID: + case PPC::VFRCRegClassID: + case PPC::VSLRCRegClassID: + case PPC::VSHRCRegClassID: return 32 - DefaultSafety; + case PPC::VSRCRegClassID: + case PPC::VSFRCRegClassID: + case PPC::VSSRCRegClassID: + return 64 - DefaultSafety; case PPC::CRRCRegClassID: return 8 - DefaultSafety; } } -//===----------------------------------------------------------------------===// -// Stack Frame Processing methods -//===----------------------------------------------------------------------===// - -void PPCRegisterInfo:: -eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB, - MachineBasicBlock::iterator I) const { - if (MF.getTarget().Options.GuaranteedTailCallOpt && - I->getOpcode() == PPC::ADJCALLSTACKUP) { - // Add (actually subtract) back the amount the callee popped on return. - if (int CalleeAmt = I->getOperand(1).getImm()) { - bool is64Bit = Subtarget.isPPC64(); - CalleeAmt *= -1; - unsigned StackReg = is64Bit ? PPC::X1 : PPC::R1; - unsigned TmpReg = is64Bit ? PPC::X0 : PPC::R0; - unsigned ADDIInstr = is64Bit ? PPC::ADDI8 : PPC::ADDI; - unsigned ADDInstr = is64Bit ? PPC::ADD8 : PPC::ADD4; - unsigned LISInstr = is64Bit ? PPC::LIS8 : PPC::LIS; - unsigned ORIInstr = is64Bit ? PPC::ORI8 : PPC::ORI; - MachineInstr *MI = I; - DebugLoc dl = MI->getDebugLoc(); - - if (isInt<16>(CalleeAmt)) { - BuildMI(MBB, I, dl, TII.get(ADDIInstr), StackReg) - .addReg(StackReg, RegState::Kill) - .addImm(CalleeAmt); - } else { - MachineBasicBlock::iterator MBBI = I; - BuildMI(MBB, MBBI, dl, TII.get(LISInstr), TmpReg) - .addImm(CalleeAmt >> 16); - BuildMI(MBB, MBBI, dl, TII.get(ORIInstr), TmpReg) - .addReg(TmpReg, RegState::Kill) - .addImm(CalleeAmt & 0xFFFF); - BuildMI(MBB, MBBI, dl, TII.get(ADDInstr), StackReg) - .addReg(StackReg, RegState::Kill) - .addReg(TmpReg); - } - } +const TargetRegisterClass * +PPCRegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC, + const MachineFunction &MF) const { + const PPCSubtarget &Subtarget = MF.getSubtarget(); + if (Subtarget.hasVSX()) { + // With VSX, we can inflate various sub-register classes to the full VSX + // register set. + + if (RC == &PPC::F8RCRegClass) + return &PPC::VSFRCRegClass; + else if (RC == &PPC::VRRCRegClass) + return &PPC::VSRCRegClass; + else if (RC == &PPC::F4RCRegClass && Subtarget.hasP8Vector()) + return &PPC::VSSRCRegClass; } - // Simply discard ADJCALLSTACKDOWN, ADJCALLSTACKUP instructions. - MBB.erase(I); -} -/// findScratchRegister - Find a 'free' PPC register. Try for a call-clobbered -/// register first and then a spilled callee-saved register if that fails. -static -unsigned findScratchRegister(MachineBasicBlock::iterator II, RegScavenger *RS, - const TargetRegisterClass *RC, int SPAdj) { - assert(RS && "Register scavenging must be on"); - unsigned Reg = RS->FindUnusedReg(RC); - // FIXME: move ARM callee-saved reg scan to target independent code, then - // search for already spilled CS register here. - if (Reg == 0) - Reg = RS->scavengeRegister(RC, II, SPAdj); - return Reg; + return TargetRegisterInfo::getLargestLegalSuperClass(RC, MF); } +//===----------------------------------------------------------------------===// +// Stack Frame Processing methods +//===----------------------------------------------------------------------===// + /// lowerDynamicAlloc - Generate the code for allocating an object in the /// current frame. The sequence of code with be in the general form /// @@ -340,8 +317,7 @@ unsigned findScratchRegister(MachineBasicBlock::iterator II, RegScavenger *RS, /// stwxu R0, SP, Rnegsize ; add and update the SP with the negated size /// addi Rnew, SP, \#maxCalFrameSize ; get the top of the allocation /// -void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II, - int SPAdj, RegScavenger *RS) const { +void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II) const { // Get the instruction. MachineInstr &MI = *II; // Get the instruction's basic block. @@ -350,8 +326,11 @@ void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II, MachineFunction &MF = *MBB.getParent(); // Get the frame info. MachineFrameInfo *MFI = MF.getFrameInfo(); + const PPCSubtarget &Subtarget = MF.getSubtarget(); + // Get the instruction info. + const TargetInstrInfo &TII = *Subtarget.getInstrInfo(); // Determine whether 64-bit pointers are used. - bool LP64 = Subtarget.isPPC64(); + bool LP64 = TM.isPPC64(); DebugLoc dl = MI.getDebugLoc(); // Get the maximum call stack size. @@ -360,10 +339,11 @@ void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II, unsigned FrameSize = MFI->getStackSize(); // Get stack alignments. - unsigned TargetAlign = MF.getTarget().getFrameLowering()->getStackAlignment(); + const PPCFrameLowering *TFI = getFrameLowering(MF); + unsigned TargetAlign = TFI->getStackAlignment(); unsigned MaxAlign = MFI->getMaxAlignment(); - if (MaxAlign > TargetAlign) - report_fatal_error("Dynamic alloca with large aligns not supported"); + assert((maxCallFrameSize & (MaxAlign-1)) == 0 && + "Maximum call-frame size not sufficiently aligned"); // Determine the previous frame's address. If FrameSize can't be // represented as 16 bits or we need special alignment, then we load the @@ -373,74 +353,77 @@ void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II, // Fortunately, a frame greater than 32K is rare. const TargetRegisterClass *G8RC = &PPC::G8RCRegClass; const TargetRegisterClass *GPRC = &PPC::GPRCRegClass; - const TargetRegisterClass *RC = LP64 ? G8RC : GPRC; - - // FIXME (64-bit): Use "findScratchRegister" - unsigned Reg; - if (requiresRegisterScavenging(MF)) - Reg = findScratchRegister(II, RS, RC, SPAdj); - else - Reg = PPC::R0; + unsigned Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC); if (MaxAlign < TargetAlign && isInt<16>(FrameSize)) { BuildMI(MBB, II, dl, TII.get(PPC::ADDI), Reg) .addReg(PPC::R31) .addImm(FrameSize); } else if (LP64) { - if (requiresRegisterScavenging(MF)) // FIXME (64-bit): Use "true" part. - BuildMI(MBB, II, dl, TII.get(PPC::LD), Reg) - .addImm(0) - .addReg(PPC::X1); - else - BuildMI(MBB, II, dl, TII.get(PPC::LD), PPC::X0) - .addImm(0) - .addReg(PPC::X1); + BuildMI(MBB, II, dl, TII.get(PPC::LD), Reg) + .addImm(0) + .addReg(PPC::X1); } else { BuildMI(MBB, II, dl, TII.get(PPC::LWZ), Reg) .addImm(0) .addReg(PPC::R1); } - + + bool KillNegSizeReg = MI.getOperand(1).isKill(); + unsigned NegSizeReg = MI.getOperand(1).getReg(); + // Grow the stack and update the stack pointer link, then determine the // address of new allocated space. if (LP64) { - if (requiresRegisterScavenging(MF)) // FIXME (64-bit): Use "true" part. - BuildMI(MBB, II, dl, TII.get(PPC::STDUX)) - .addReg(Reg, RegState::Kill) - .addReg(PPC::X1, RegState::Define) - .addReg(MI.getOperand(1).getReg()); - else - BuildMI(MBB, II, dl, TII.get(PPC::STDUX)) - .addReg(PPC::X0, RegState::Kill) - .addReg(PPC::X1, RegState::Define) - .addReg(MI.getOperand(1).getReg()); - - if (!MI.getOperand(1).isKill()) - BuildMI(MBB, II, dl, TII.get(PPC::ADDI8), MI.getOperand(0).getReg()) - .addReg(PPC::X1) - .addImm(maxCallFrameSize); - else - // Implicitly kill the register. - BuildMI(MBB, II, dl, TII.get(PPC::ADDI8), MI.getOperand(0).getReg()) - .addReg(PPC::X1) - .addImm(maxCallFrameSize) - .addReg(MI.getOperand(1).getReg(), RegState::ImplicitKill); - } else { - BuildMI(MBB, II, dl, TII.get(PPC::STWUX)) + if (MaxAlign > TargetAlign) { + unsigned UnalNegSizeReg = NegSizeReg; + NegSizeReg = MF.getRegInfo().createVirtualRegister(G8RC); + + // Unfortunately, there is no andi, only andi., and we can't insert that + // here because we might clobber cr0 while it is live. + BuildMI(MBB, II, dl, TII.get(PPC::LI8), NegSizeReg) + .addImm(~(MaxAlign-1)); + + unsigned NegSizeReg1 = NegSizeReg; + NegSizeReg = MF.getRegInfo().createVirtualRegister(G8RC); + BuildMI(MBB, II, dl, TII.get(PPC::AND8), NegSizeReg) + .addReg(UnalNegSizeReg, getKillRegState(KillNegSizeReg)) + .addReg(NegSizeReg1, RegState::Kill); + KillNegSizeReg = true; + } + + BuildMI(MBB, II, dl, TII.get(PPC::STDUX), PPC::X1) .addReg(Reg, RegState::Kill) - .addReg(PPC::R1, RegState::Define) - .addReg(MI.getOperand(1).getReg()); + .addReg(PPC::X1) + .addReg(NegSizeReg, getKillRegState(KillNegSizeReg)); + BuildMI(MBB, II, dl, TII.get(PPC::ADDI8), MI.getOperand(0).getReg()) + .addReg(PPC::X1) + .addImm(maxCallFrameSize); + } else { + if (MaxAlign > TargetAlign) { + unsigned UnalNegSizeReg = NegSizeReg; + NegSizeReg = MF.getRegInfo().createVirtualRegister(GPRC); + + // Unfortunately, there is no andi, only andi., and we can't insert that + // here because we might clobber cr0 while it is live. + BuildMI(MBB, II, dl, TII.get(PPC::LI), NegSizeReg) + .addImm(~(MaxAlign-1)); + + unsigned NegSizeReg1 = NegSizeReg; + NegSizeReg = MF.getRegInfo().createVirtualRegister(GPRC); + BuildMI(MBB, II, dl, TII.get(PPC::AND), NegSizeReg) + .addReg(UnalNegSizeReg, getKillRegState(KillNegSizeReg)) + .addReg(NegSizeReg1, RegState::Kill); + KillNegSizeReg = true; + } - if (!MI.getOperand(1).isKill()) - BuildMI(MBB, II, dl, TII.get(PPC::ADDI), MI.getOperand(0).getReg()) - .addReg(PPC::R1) - .addImm(maxCallFrameSize); - else - // Implicitly kill the register. - BuildMI(MBB, II, dl, TII.get(PPC::ADDI), MI.getOperand(0).getReg()) - .addReg(PPC::R1) - .addImm(maxCallFrameSize) - .addReg(MI.getOperand(1).getReg(), RegState::ImplicitKill); + BuildMI(MBB, II, dl, TII.get(PPC::STWUX), PPC::R1) + .addReg(Reg, RegState::Kill) + .addReg(PPC::R1) + .addReg(NegSizeReg, getKillRegState(KillNegSizeReg)); + BuildMI(MBB, II, dl, TII.get(PPC::ADDI), MI.getOperand(0).getReg()) + .addReg(PPC::R1) + .addImm(maxCallFrameSize); } // Discard the DYNALLOC instruction. @@ -456,42 +439,44 @@ void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II, /// stw rA, FI ; Store rA to the frame. /// void PPCRegisterInfo::lowerCRSpilling(MachineBasicBlock::iterator II, - unsigned FrameIndex, int SPAdj, - RegScavenger *RS) const { + unsigned FrameIndex) const { // Get the instruction. MachineInstr &MI = *II; // ; SPILL_CR , // Get the instruction's basic block. MachineBasicBlock &MBB = *MI.getParent(); + MachineFunction &MF = *MBB.getParent(); + const PPCSubtarget &Subtarget = MF.getSubtarget(); + const TargetInstrInfo &TII = *Subtarget.getInstrInfo(); DebugLoc dl = MI.getDebugLoc(); - // FIXME: Once LLVM supports creating virtual registers here, or the register - // scavenger can return multiple registers, stop using reserved registers - // here. - (void) SPAdj; - (void) RS; + bool LP64 = TM.isPPC64(); + const TargetRegisterClass *G8RC = &PPC::G8RCRegClass; + const TargetRegisterClass *GPRC = &PPC::GPRCRegClass; - bool LP64 = Subtarget.isPPC64(); - unsigned Reg = Subtarget.isDarwinABI() ? (LP64 ? PPC::X2 : PPC::R2) : - (LP64 ? PPC::X0 : PPC::R0); + unsigned Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC); unsigned SrcReg = MI.getOperand(0).getReg(); // We need to store the CR in the low 4-bits of the saved value. First, issue - // an MFCRpsued to save all of the CRBits and, if needed, kill the SrcReg. - BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::MFCR8pseud : PPC::MFCRpseud), Reg) + // an MFOCRF to save all of the CRBits and, if needed, kill the SrcReg. + BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::MFOCRF8 : PPC::MFOCRF), Reg) .addReg(SrcReg, getKillRegState(MI.getOperand(0).isKill())); // If the saved register wasn't CR0, shift the bits left so that they are in // CR0's slot. - if (SrcReg != PPC::CR0) + if (SrcReg != PPC::CR0) { + unsigned Reg1 = Reg; + Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC); + // rlwinm rA, rA, ShiftBits, 0, 31. BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::RLWINM8 : PPC::RLWINM), Reg) - .addReg(Reg, RegState::Kill) - .addImm(getPPCRegisterNumbering(SrcReg) * 4) + .addReg(Reg1, RegState::Kill) + .addImm(getEncodingValue(SrcReg) * 4) .addImm(0) .addImm(31); + } addFrameReference(BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::STW8 : PPC::STW)) - .addReg(Reg, getKillRegState(MI.getOperand(1).getImm())), + .addReg(Reg, RegState::Kill), FrameIndex); // Discard the pseudo instruction. @@ -499,23 +484,21 @@ void PPCRegisterInfo::lowerCRSpilling(MachineBasicBlock::iterator II, } void PPCRegisterInfo::lowerCRRestore(MachineBasicBlock::iterator II, - unsigned FrameIndex, int SPAdj, - RegScavenger *RS) const { + unsigned FrameIndex) const { // Get the instruction. MachineInstr &MI = *II; // ; = RESTORE_CR // Get the instruction's basic block. MachineBasicBlock &MBB = *MI.getParent(); + MachineFunction &MF = *MBB.getParent(); + const PPCSubtarget &Subtarget = MF.getSubtarget(); + const TargetInstrInfo &TII = *Subtarget.getInstrInfo(); DebugLoc dl = MI.getDebugLoc(); - // FIXME: Once LLVM supports creating virtual registers here, or the register - // scavenger can return multiple registers, stop using reserved registers - // here. - (void) SPAdj; - (void) RS; + bool LP64 = TM.isPPC64(); + const TargetRegisterClass *G8RC = &PPC::G8RCRegClass; + const TargetRegisterClass *GPRC = &PPC::GPRCRegClass; - bool LP64 = Subtarget.isPPC64(); - unsigned Reg = Subtarget.isDarwinABI() ? (LP64 ? PPC::X2 : PPC::R2) : - (LP64 ? PPC::X0 : PPC::R0); + unsigned Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC); unsigned DestReg = MI.getOperand(0).getReg(); assert(MI.definesRegister(DestReg) && "RESTORE_CR does not define its destination"); @@ -526,23 +509,223 @@ void PPCRegisterInfo::lowerCRRestore(MachineBasicBlock::iterator II, // 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 = getPPCRegisterNumbering(DestReg)*4; + unsigned Reg1 = Reg; + Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC); + + unsigned ShiftBits = getEncodingValue(DestReg)*4; // rlwinm r11, r11, 32-ShiftBits, 0, 31. - BuildMI(MBB, II, dl, TII.get(PPC::RLWINM), Reg) - .addReg(Reg).addImm(32-ShiftBits).addImm(0) + BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::RLWINM8 : PPC::RLWINM), Reg) + .addReg(Reg1, RegState::Kill).addImm(32-ShiftBits).addImm(0) .addImm(31); } - BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::MTCRF8 : PPC::MTCRF), DestReg) - .addReg(Reg); + BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::MTOCRF8 : PPC::MTOCRF), DestReg) + .addReg(Reg, RegState::Kill); + + // Discard the pseudo instruction. + MBB.erase(II); +} + +void PPCRegisterInfo::lowerCRBitSpilling(MachineBasicBlock::iterator II, + unsigned FrameIndex) const { + // Get the instruction. + MachineInstr &MI = *II; // ; SPILL_CRBIT , + // Get the instruction's basic block. + MachineBasicBlock &MBB = *MI.getParent(); + MachineFunction &MF = *MBB.getParent(); + const PPCSubtarget &Subtarget = MF.getSubtarget(); + const TargetInstrInfo &TII = *Subtarget.getInstrInfo(); + DebugLoc dl = MI.getDebugLoc(); + + bool LP64 = TM.isPPC64(); + const TargetRegisterClass *G8RC = &PPC::G8RCRegClass; + const TargetRegisterClass *GPRC = &PPC::GPRCRegClass; + + unsigned Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC); + unsigned SrcReg = MI.getOperand(0).getReg(); + + BuildMI(MBB, II, dl, TII.get(TargetOpcode::KILL), + getCRFromCRBit(SrcReg)) + .addReg(SrcReg, getKillRegState(MI.getOperand(0).isKill())); + + BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::MFOCRF8 : PPC::MFOCRF), Reg) + .addReg(getCRFromCRBit(SrcReg)); + + // If the saved register wasn't CR0LT, shift the bits left so that the bit to + // store is the first one. Mask all but that bit. + unsigned Reg1 = Reg; + Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC); + + // rlwinm rA, rA, ShiftBits, 0, 0. + BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::RLWINM8 : PPC::RLWINM), Reg) + .addReg(Reg1, RegState::Kill) + .addImm(getEncodingValue(SrcReg)) + .addImm(0).addImm(0); + + addFrameReference(BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::STW8 : PPC::STW)) + .addReg(Reg, RegState::Kill), + FrameIndex); + + // Discard the pseudo instruction. + MBB.erase(II); +} + +void PPCRegisterInfo::lowerCRBitRestore(MachineBasicBlock::iterator II, + unsigned FrameIndex) const { + // Get the instruction. + MachineInstr &MI = *II; // ; = RESTORE_CRBIT + // Get the instruction's basic block. + MachineBasicBlock &MBB = *MI.getParent(); + MachineFunction &MF = *MBB.getParent(); + const PPCSubtarget &Subtarget = MF.getSubtarget(); + const TargetInstrInfo &TII = *Subtarget.getInstrInfo(); + DebugLoc dl = MI.getDebugLoc(); + + bool LP64 = TM.isPPC64(); + const TargetRegisterClass *G8RC = &PPC::G8RCRegClass; + const TargetRegisterClass *GPRC = &PPC::GPRCRegClass; + + unsigned Reg = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC); + unsigned DestReg = MI.getOperand(0).getReg(); + assert(MI.definesRegister(DestReg) && + "RESTORE_CRBIT does not define its destination"); + + addFrameReference(BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::LWZ8 : PPC::LWZ), + Reg), FrameIndex); + + BuildMI(MBB, II, dl, TII.get(TargetOpcode::IMPLICIT_DEF), DestReg); + + unsigned RegO = MF.getRegInfo().createVirtualRegister(LP64 ? G8RC : GPRC); + BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::MFOCRF8 : PPC::MFOCRF), RegO) + .addReg(getCRFromCRBit(DestReg)); + + unsigned ShiftBits = getEncodingValue(DestReg); + // rlwimi r11, r10, 32-ShiftBits, ..., ... + BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::RLWIMI8 : PPC::RLWIMI), RegO) + .addReg(RegO, RegState::Kill).addReg(Reg, RegState::Kill) + .addImm(ShiftBits ? 32-ShiftBits : 0) + .addImm(ShiftBits).addImm(ShiftBits); + + BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::MTOCRF8 : PPC::MTOCRF), + getCRFromCRBit(DestReg)) + .addReg(RegO, RegState::Kill) + // Make sure we have a use dependency all the way through this + // sequence of instructions. We can't have the other bits in the CR + // modified in between the mfocrf and the mtocrf. + .addReg(getCRFromCRBit(DestReg), RegState::Implicit); // Discard the pseudo instruction. MBB.erase(II); } +void PPCRegisterInfo::lowerVRSAVESpilling(MachineBasicBlock::iterator II, + unsigned FrameIndex) const { + // Get the instruction. + MachineInstr &MI = *II; // ; SPILL_VRSAVE , + // Get the instruction's basic block. + MachineBasicBlock &MBB = *MI.getParent(); + MachineFunction &MF = *MBB.getParent(); + const PPCSubtarget &Subtarget = MF.getSubtarget(); + const TargetInstrInfo &TII = *Subtarget.getInstrInfo(); + DebugLoc dl = MI.getDebugLoc(); + + const TargetRegisterClass *GPRC = &PPC::GPRCRegClass; + unsigned Reg = MF.getRegInfo().createVirtualRegister(GPRC); + unsigned SrcReg = MI.getOperand(0).getReg(); + + BuildMI(MBB, II, dl, TII.get(PPC::MFVRSAVEv), Reg) + .addReg(SrcReg, getKillRegState(MI.getOperand(0).isKill())); + + addFrameReference(BuildMI(MBB, II, dl, TII.get(PPC::STW)) + .addReg(Reg, RegState::Kill), + FrameIndex); + + // Discard the pseudo instruction. + MBB.erase(II); +} + +void PPCRegisterInfo::lowerVRSAVERestore(MachineBasicBlock::iterator II, + unsigned FrameIndex) const { + // Get the instruction. + MachineInstr &MI = *II; // ; = RESTORE_VRSAVE + // Get the instruction's basic block. + MachineBasicBlock &MBB = *MI.getParent(); + MachineFunction &MF = *MBB.getParent(); + const PPCSubtarget &Subtarget = MF.getSubtarget(); + const TargetInstrInfo &TII = *Subtarget.getInstrInfo(); + DebugLoc dl = MI.getDebugLoc(); + + const TargetRegisterClass *GPRC = &PPC::GPRCRegClass; + unsigned Reg = MF.getRegInfo().createVirtualRegister(GPRC); + unsigned DestReg = MI.getOperand(0).getReg(); + assert(MI.definesRegister(DestReg) && + "RESTORE_VRSAVE does not define its destination"); + + addFrameReference(BuildMI(MBB, II, dl, TII.get(PPC::LWZ), + Reg), FrameIndex); + + BuildMI(MBB, II, dl, TII.get(PPC::MTVRSAVEv), DestReg) + .addReg(Reg, RegState::Kill); + + // Discard the pseudo instruction. + MBB.erase(II); +} + +bool +PPCRegisterInfo::hasReservedSpillSlot(const MachineFunction &MF, + unsigned Reg, int &FrameIdx) const { + const PPCSubtarget &Subtarget = MF.getSubtarget(); + // For the nonvolatile condition registers (CR2, CR3, CR4) in an SVR4 + // ABI, return true to prevent allocating an additional frame slot. + // For 64-bit, the CR save area is at SP+8; the value of FrameIdx = 0 + // is arbitrary and will be subsequently ignored. For 32-bit, we have + // previously created the stack slot if needed, so return its FrameIdx. + if (Subtarget.isSVR4ABI() && PPC::CR2 <= Reg && Reg <= PPC::CR4) { + if (TM.isPPC64()) + FrameIdx = 0; + else { + const PPCFunctionInfo *FI = MF.getInfo(); + FrameIdx = FI->getCRSpillFrameIndex(); + } + return true; + } + return false; +} + +// Figure out if the offset in the instruction must be a multiple of 4. +// This is true for instructions like "STD". +static bool usesIXAddr(const MachineInstr &MI) { + unsigned OpC = MI.getOpcode(); + + switch (OpC) { + default: + return false; + case PPC::LWA: + case PPC::LWA_32: + case PPC::LD: + case PPC::STD: + return true; + } +} + +// Return the OffsetOperandNo given the FIOperandNum (and the instruction). +static unsigned getOffsetONFromFION(const MachineInstr &MI, + unsigned FIOperandNum) { + // Take into account whether it's an add or mem instruction + unsigned OffsetOperandNo = (FIOperandNum == 2) ? 1 : 2; + if (MI.isInlineAsm()) + OffsetOperandNo = FIOperandNum - 1; + else if (MI.getOpcode() == TargetOpcode::STACKMAP || + MI.getOpcode() == TargetOpcode::PATCHPOINT) + OffsetOperandNo = FIOperandNum + 1; + + return OffsetOperandNo; +} + void PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, - int SPAdj, RegScavenger *RS) const { + int SPAdj, unsigned FIOperandNum, + RegScavenger *RS) const { assert(SPAdj == 0 && "Unexpected"); // Get the instruction. @@ -551,25 +734,17 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, MachineBasicBlock &MBB = *MI.getParent(); // Get the basic block's function. MachineFunction &MF = *MBB.getParent(); + const PPCSubtarget &Subtarget = MF.getSubtarget(); + // Get the instruction info. + const TargetInstrInfo &TII = *Subtarget.getInstrInfo(); // Get the frame info. MachineFrameInfo *MFI = MF.getFrameInfo(); - const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); DebugLoc dl = MI.getDebugLoc(); - // Find out which operand is the frame index. - unsigned FIOperandNo = 0; - while (!MI.getOperand(FIOperandNo).isFI()) { - ++FIOperandNo; - assert(FIOperandNo != MI.getNumOperands() && - "Instr doesn't have FrameIndex operand!"); - } - // Take into account whether it's an add or mem instruction - unsigned OffsetOperandNo = (FIOperandNo == 2) ? 1 : 2; - if (MI.isInlineAsm()) - OffsetOperandNo = FIOperandNo-1; + unsigned OffsetOperandNo = getOffsetONFromFION(MI, FIOperandNum); // Get the frame index. - int FrameIndex = MI.getOperand(FIOperandNo).getIndex(); + int FrameIndex = MI.getOperand(FIOperandNum).getIndex(); // Get the frame pointer save index. Users of this index are primarily // DYNALLOC instructions. @@ -581,56 +756,56 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, // Special case for dynamic alloca. if (FPSI && FrameIndex == FPSI && (OpC == PPC::DYNALLOC || OpC == PPC::DYNALLOC8)) { - lowerDynamicAlloc(II, SPAdj, RS); + lowerDynamicAlloc(II); return; } - // Special case for pseudo-ops SPILL_CR and RESTORE_CR. - if (requiresRegisterScavenging(MF)) { - if (OpC == PPC::SPILL_CR) { - lowerCRSpilling(II, FrameIndex, SPAdj, RS); - return; - } else if (OpC == PPC::RESTORE_CR) { - lowerCRRestore(II, FrameIndex, SPAdj, RS); - return; - } + // Special case for pseudo-ops SPILL_CR and RESTORE_CR, etc. + if (OpC == PPC::SPILL_CR) { + lowerCRSpilling(II, FrameIndex); + return; + } else if (OpC == PPC::RESTORE_CR) { + lowerCRRestore(II, FrameIndex); + return; + } else if (OpC == PPC::SPILL_CRBIT) { + lowerCRBitSpilling(II, FrameIndex); + return; + } else if (OpC == PPC::RESTORE_CRBIT) { + lowerCRBitRestore(II, FrameIndex); + return; + } else if (OpC == PPC::SPILL_VRSAVE) { + lowerVRSAVESpilling(II, FrameIndex); + return; + } else if (OpC == PPC::RESTORE_VRSAVE) { + lowerVRSAVERestore(II, FrameIndex); + return; } // Replace the FrameIndex with base register with GPR1 (SP) or GPR31 (FP). + MI.getOperand(FIOperandNum).ChangeToRegister( + FrameIndex < 0 ? getBaseRegister(MF) : getFrameRegister(MF), false); - bool is64Bit = Subtarget.isPPC64(); - MI.getOperand(FIOperandNo).ChangeToRegister(TFI->hasFP(MF) ? - (is64Bit ? PPC::X31 : PPC::R31) : - (is64Bit ? PPC::X1 : PPC::R1), - false); + // Figure out if the offset in the instruction is shifted right two bits. + bool isIXAddr = usesIXAddr(MI); + + // If the instruction is not present in ImmToIdxMap, then it has no immediate + // form (and must be r+r). + bool noImmForm = !MI.isInlineAsm() && OpC != TargetOpcode::STACKMAP && + OpC != TargetOpcode::PATCHPOINT && !ImmToIdxMap.count(OpC); - // Figure out if the offset in the instruction is shifted right two bits. This - // is true for instructions like "STD", which the machine implicitly adds two - // low zeros to. - bool isIXAddr = false; - switch (OpC) { - case PPC::LWA: - case PPC::LD: - case PPC::STD: - case PPC::STD_32: - isIXAddr = true; - break; - } - // Now add the frame object offset to the offset from r1. int Offset = MFI->getObjectOffset(FrameIndex); - if (!isIXAddr) - Offset += MI.getOperand(OffsetOperandNo).getImm(); - else - Offset += MI.getOperand(OffsetOperandNo).getImm() << 2; + Offset += MI.getOperand(OffsetOperandNo).getImm(); // If we're not using a Frame Pointer that has been set to the value of the // SP before having the stack size subtracted from it, then add the stack size // to Offset to get the correct offset. // Naked functions have stack size 0, although getStackSize may not reflect that // because we didn't call all the pieces that compute it for naked functions. - if (!MF.getFunction()->hasFnAttr(Attribute::Naked)) - Offset += MFI->getStackSize(); + if (!MF.getFunction()->hasFnAttribute(Attribute::Naked)) { + if (!(hasBasePointer(MF) && FrameIndex < 0)) + Offset += MFI->getStackSize(); + } // If we can, encode the offset directly into the instruction. If this is a // normal PPC "ri" instruction, any 16-bit value can be safely encoded. If @@ -638,9 +813,11 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, // clear can be encoded. This is extremely uncommon, because normally you // only "std" to a stack slot that is at least 4-byte aligned, but it can // happen in invalid code. - if (isInt<16>(Offset) && (!isIXAddr || (Offset & 3) == 0)) { - if (isIXAddr) - Offset >>= 2; // The actual encoded value has the low two bits zero. + assert(OpC != PPC::DBG_VALUE && + "This should be handled in a target-independent way"); + if (!noImmForm && ((isInt<16>(Offset) && (!isIXAddr || (Offset & 3) == 0)) || + OpC == TargetOpcode::STACKMAP || + OpC == TargetOpcode::PATCHPOINT)) { MI.getOperand(OffsetOperandNo).ChangeToImmediate(Offset); return; } @@ -648,19 +825,18 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, // The offset doesn't fit into a single register, scavenge one to build the // offset in. - unsigned SReg; - if (requiresRegisterScavenging(MF)) { - const TargetRegisterClass *G8RC = &PPC::G8RCRegClass; - const TargetRegisterClass *GPRC = &PPC::GPRCRegClass; - SReg = findScratchRegister(II, RS, is64Bit ? G8RC : GPRC, SPAdj); - } else - SReg = is64Bit ? PPC::X0 : PPC::R0; + bool is64Bit = TM.isPPC64(); + const TargetRegisterClass *G8RC = &PPC::G8RCRegClass; + const TargetRegisterClass *GPRC = &PPC::GPRCRegClass; + const TargetRegisterClass *RC = is64Bit ? G8RC : GPRC; + unsigned SRegHi = MF.getRegInfo().createVirtualRegister(RC), + SReg = MF.getRegInfo().createVirtualRegister(RC); // Insert a set of rA with the full offset value before the ld, st, or add - BuildMI(MBB, II, dl, TII.get(is64Bit ? PPC::LIS8 : PPC::LIS), SReg) + BuildMI(MBB, II, dl, TII.get(is64Bit ? PPC::LIS8 : PPC::LIS), SRegHi) .addImm(Offset >> 16); BuildMI(MBB, II, dl, TII.get(is64Bit ? PPC::ORI8 : PPC::ORI), SReg) - .addReg(SReg, RegState::Kill) + .addReg(SRegHi, RegState::Kill) .addImm(Offset); // Convert into indexed form of the instruction: @@ -669,7 +845,9 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, // addi 0:rA 1:rB, 2, imm ==> add 0:rA, 1:rB, 2:r0 unsigned OperandBase; - if (OpC != TargetOpcode::INLINEASM) { + if (noImmForm) + OperandBase = 1; + else if (OpC != TargetOpcode::INLINEASM) { assert(ImmToIdxMap.count(OpC) && "No indexed form of load or store available!"); unsigned NewOpcode = ImmToIdxMap.find(OpC)->second; @@ -679,24 +857,157 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, OperandBase = OffsetOperandNo; } - unsigned StackReg = MI.getOperand(FIOperandNo).getReg(); + unsigned StackReg = MI.getOperand(FIOperandNum).getReg(); MI.getOperand(OperandBase).ChangeToRegister(StackReg, false); MI.getOperand(OperandBase + 1).ChangeToRegister(SReg, false, false, true); } unsigned PPCRegisterInfo::getFrameRegister(const MachineFunction &MF) const { - const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); + const PPCFrameLowering *TFI = getFrameLowering(MF); - if (!Subtarget.isPPC64()) + if (!TM.isPPC64()) return TFI->hasFP(MF) ? PPC::R31 : PPC::R1; else return TFI->hasFP(MF) ? PPC::X31 : PPC::X1; } -unsigned PPCRegisterInfo::getEHExceptionRegister() const { - return !Subtarget.isPPC64() ? PPC::R3 : PPC::X3; +unsigned PPCRegisterInfo::getBaseRegister(const MachineFunction &MF) const { + const PPCSubtarget &Subtarget = MF.getSubtarget(); + if (!hasBasePointer(MF)) + return getFrameRegister(MF); + + if (TM.isPPC64()) + return PPC::X30; + + if (Subtarget.isSVR4ABI() && + TM.getRelocationModel() == Reloc::PIC_) + return PPC::R29; + + return PPC::R30; +} + +bool PPCRegisterInfo::hasBasePointer(const MachineFunction &MF) const { + if (!EnableBasePointer) + return false; + if (AlwaysBasePointer) + return true; + + // If we need to realign the stack, then the stack pointer can no longer + // serve as an offset into the caller's stack space. As a result, we need a + // base pointer. + return needsStackRealignment(MF); +} + +/// Returns true if the instruction's frame index +/// reference would be better served by a base register other than FP +/// or SP. Used by LocalStackFrameAllocation to determine which frame index +/// references it should create new base registers for. +bool PPCRegisterInfo:: +needsFrameBaseReg(MachineInstr *MI, int64_t Offset) const { + assert(Offset < 0 && "Local offset must be negative"); + + // It's the load/store FI references that cause issues, as it can be difficult + // to materialize the offset if it won't fit in the literal field. Estimate + // based on the size of the local frame and some conservative assumptions + // about the rest of the stack frame (note, this is pre-regalloc, so + // we don't know everything for certain yet) whether this offset is likely + // to be out of range of the immediate. Return true if so. + + // We only generate virtual base registers for loads and stores that have + // an r+i form. Return false for everything else. + unsigned OpC = MI->getOpcode(); + if (!ImmToIdxMap.count(OpC)) + return false; + + // Don't generate a new virtual base register just to add zero to it. + if ((OpC == PPC::ADDI || OpC == PPC::ADDI8) && + MI->getOperand(2).getImm() == 0) + return false; + + MachineBasicBlock &MBB = *MI->getParent(); + MachineFunction &MF = *MBB.getParent(); + const PPCFrameLowering *TFI = getFrameLowering(MF); + unsigned StackEst = TFI->determineFrameLayout(MF, false, true); + + // If we likely don't need a stack frame, then we probably don't need a + // virtual base register either. + if (!StackEst) + return false; + + // Estimate an offset from the stack pointer. + // The incoming offset is relating to the SP at the start of the function, + // but when we access the local it'll be relative to the SP after local + // allocation, so adjust our SP-relative offset by that allocation size. + Offset += StackEst; + + // The frame pointer will point to the end of the stack, so estimate the + // offset as the difference between the object offset and the FP location. + return !isFrameOffsetLegal(MI, getBaseRegister(MF), Offset); } -unsigned PPCRegisterInfo::getEHHandlerRegister() const { - return !Subtarget.isPPC64() ? PPC::R4 : PPC::X4; +/// Insert defining instruction(s) for BaseReg to +/// be a pointer to FrameIdx at the beginning of the basic block. +void PPCRegisterInfo:: +materializeFrameBaseRegister(MachineBasicBlock *MBB, + unsigned BaseReg, int FrameIdx, + int64_t Offset) const { + unsigned ADDriOpc = TM.isPPC64() ? PPC::ADDI8 : PPC::ADDI; + + MachineBasicBlock::iterator Ins = MBB->begin(); + DebugLoc DL; // Defaults to "unknown" + if (Ins != MBB->end()) + DL = Ins->getDebugLoc(); + + const MachineFunction &MF = *MBB->getParent(); + const PPCSubtarget &Subtarget = MF.getSubtarget(); + const TargetInstrInfo &TII = *Subtarget.getInstrInfo(); + const MCInstrDesc &MCID = TII.get(ADDriOpc); + MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo(); + MRI.constrainRegClass(BaseReg, TII.getRegClass(MCID, 0, this, MF)); + + BuildMI(*MBB, Ins, DL, MCID, BaseReg) + .addFrameIndex(FrameIdx).addImm(Offset); +} + +void PPCRegisterInfo::resolveFrameIndex(MachineInstr &MI, unsigned BaseReg, + int64_t Offset) const { + unsigned FIOperandNum = 0; + while (!MI.getOperand(FIOperandNum).isFI()) { + ++FIOperandNum; + assert(FIOperandNum < MI.getNumOperands() && + "Instr doesn't have FrameIndex operand!"); + } + + MI.getOperand(FIOperandNum).ChangeToRegister(BaseReg, false); + unsigned OffsetOperandNo = getOffsetONFromFION(MI, FIOperandNum); + Offset += MI.getOperand(OffsetOperandNo).getImm(); + MI.getOperand(OffsetOperandNo).ChangeToImmediate(Offset); + + MachineBasicBlock &MBB = *MI.getParent(); + MachineFunction &MF = *MBB.getParent(); + const PPCSubtarget &Subtarget = MF.getSubtarget(); + const TargetInstrInfo &TII = *Subtarget.getInstrInfo(); + const MCInstrDesc &MCID = MI.getDesc(); + MachineRegisterInfo &MRI = MF.getRegInfo(); + MRI.constrainRegClass(BaseReg, + TII.getRegClass(MCID, FIOperandNum, this, MF)); +} + +bool PPCRegisterInfo::isFrameOffsetLegal(const MachineInstr *MI, + unsigned BaseReg, + int64_t Offset) const { + unsigned FIOperandNum = 0; + while (!MI->getOperand(FIOperandNum).isFI()) { + ++FIOperandNum; + assert(FIOperandNum < MI->getNumOperands() && + "Instr doesn't have FrameIndex operand!"); + } + + unsigned OffsetOperandNo = getOffsetONFromFION(*MI, FIOperandNum); + Offset += MI->getOperand(OffsetOperandNo).getImm(); + + return MI->getOpcode() == PPC::DBG_VALUE || // DBG_VALUE is always Reg+Imm + MI->getOpcode() == TargetOpcode::STACKMAP || + MI->getOpcode() == TargetOpcode::PATCHPOINT || + (isInt<16>(Offset) && (!usesIXAddr(*MI) || (Offset & 3) == 0)); }