X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTarget%2FARM%2FARMBaseRegisterInfo.cpp;h=419717c85a79d1d0cf6fa53e7cb859488917ef9c;hb=001f3417071d4d6b08cc0dcd1dc03f5f90fe7623;hp=8ee6ce29d98091164fe8e4f3f026444bd77fe6d4;hpb=8a8d479214745c82ef00f08d4e4f1c173b5f9ce2;p=oota-llvm.git diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.cpp b/lib/Target/ARM/ARMBaseRegisterInfo.cpp index 8ee6ce29d98..419717c85a7 100644 --- a/lib/Target/ARM/ARMBaseRegisterInfo.cpp +++ b/lib/Target/ARM/ARMBaseRegisterInfo.cpp @@ -1,4 +1,4 @@ -//===- ARMBaseRegisterInfo.cpp - ARM Register Information -------*- C++ -*-===// +//===-- ARMBaseRegisterInfo.cpp - ARM Register Information ----------------===// // // The LLVM Compiler Infrastructure // @@ -11,91 +11,132 @@ // //===----------------------------------------------------------------------===// +#include "ARMBaseRegisterInfo.h" #include "ARM.h" #include "ARMBaseInstrInfo.h" -#include "ARMBaseRegisterInfo.h" #include "ARMFrameLowering.h" -#include "ARMInstrInfo.h" #include "ARMMachineFunctionInfo.h" #include "ARMSubtarget.h" #include "MCTargetDesc/ARMAddressingModes.h" -#include "llvm/Constants.h" -#include "llvm/DerivedTypes.h" -#include "llvm/Function.h" -#include "llvm/LLVMContext.h" +#include "llvm/ADT/BitVector.h" +#include "llvm/ADT/SmallVector.h" #include "llvm/CodeGen/MachineConstantPool.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineInstrBuilder.h" #include "llvm/CodeGen/MachineRegisterInfo.h" #include "llvm/CodeGen/RegisterScavenging.h" +#include "llvm/CodeGen/VirtRegMap.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/Function.h" +#include "llvm/IR/LLVMContext.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Target/TargetFrameLowering.h" #include "llvm/Target/TargetMachine.h" #include "llvm/Target/TargetOptions.h" -#include "llvm/ADT/BitVector.h" -#include "llvm/ADT/SmallVector.h" -#include "llvm/Support/CommandLine.h" + +#define DEBUG_TYPE "arm-register-info" #define GET_REGINFO_TARGET_DESC #include "ARMGenRegisterInfo.inc" using namespace llvm; -static cl::opt -ForceAllBaseRegAlloc("arm-force-base-reg-alloc", cl::Hidden, cl::init(false), - cl::desc("Force use of virtual base registers for stack load/store")); -static cl::opt -EnableLocalStackAlloc("enable-local-stack-alloc", cl::init(true), cl::Hidden, - cl::desc("Enable pre-regalloc stack frame index allocation")); -static cl::opt -EnableBasePointer("arm-use-base-pointer", cl::Hidden, cl::init(true), - cl::desc("Enable use of a base pointer for complex stack frames")); - -ARMBaseRegisterInfo::ARMBaseRegisterInfo(const ARMBaseInstrInfo &tii, - const ARMSubtarget &sti) - : ARMGenRegisterInfo(ARM::LR), TII(tii), STI(sti), - FramePtr((STI.isTargetDarwin() || STI.isThumb()) ? ARM::R7 : ARM::R11), - BasePtr(ARM::R6) { +ARMBaseRegisterInfo::ARMBaseRegisterInfo() + : ARMGenRegisterInfo(ARM::LR, 0, 0, ARM::PC), BasePtr(ARM::R6) {} + +static unsigned getFramePointerReg(const ARMSubtarget &STI) { + if (STI.isTargetMachO()) { + if (STI.isTargetDarwin() || STI.isThumb1Only()) + return ARM::R7; + else + return ARM::R11; + } else if (STI.isTargetWindows()) + return ARM::R11; + else // ARM EABI + return STI.isThumb() ? ARM::R7 : ARM::R11; } -const unsigned* +const MCPhysReg* ARMBaseRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const { - static const unsigned CalleeSavedRegs[] = { - ARM::LR, ARM::R11, ARM::R10, ARM::R9, ARM::R8, - ARM::R7, ARM::R6, ARM::R5, ARM::R4, - - ARM::D15, ARM::D14, ARM::D13, ARM::D12, - ARM::D11, ARM::D10, ARM::D9, ARM::D8, - 0 - }; - - static const unsigned DarwinCalleeSavedRegs[] = { - // Darwin ABI deviates from ARM standard ABI. R9 is not a callee-saved - // register. - ARM::LR, ARM::R7, ARM::R6, ARM::R5, ARM::R4, - ARM::R11, ARM::R10, ARM::R8, - - ARM::D15, ARM::D14, ARM::D13, ARM::D12, - ARM::D11, ARM::D10, ARM::D9, ARM::D8, - 0 - }; - return STI.isTargetDarwin() ? DarwinCalleeSavedRegs : CalleeSavedRegs; + const ARMSubtarget &STI = MF->getSubtarget(); + const MCPhysReg *RegList = + STI.isTargetDarwin() ? CSR_iOS_SaveList : CSR_AAPCS_SaveList; + + const Function *F = MF->getFunction(); + if (F->getCallingConv() == CallingConv::GHC) { + // GHC set of callee saved regs is empty as all those regs are + // used for passing STG regs around + return CSR_NoRegs_SaveList; + } else if (F->hasFnAttribute("interrupt")) { + if (STI.isMClass()) { + // M-class CPUs have hardware which saves the registers needed to allow a + // function conforming to the AAPCS to function as a handler. + return CSR_AAPCS_SaveList; + } else if (F->getFnAttribute("interrupt").getValueAsString() == "FIQ") { + // Fast interrupt mode gives the handler a private copy of R8-R14, so less + // need to be saved to restore user-mode state. + return CSR_FIQ_SaveList; + } else { + // Generally only R13-R14 (i.e. SP, LR) are automatically preserved by + // exception handling. + return CSR_GenericInt_SaveList; + } + } + + return RegList; +} + +const uint32_t * +ARMBaseRegisterInfo::getCallPreservedMask(const MachineFunction &MF, + CallingConv::ID CC) const { + const ARMSubtarget &STI = MF.getSubtarget(); + if (CC == CallingConv::GHC) + // This is academic becase all GHC calls are (supposed to be) tail calls + return CSR_NoRegs_RegMask; + return STI.isTargetDarwin() ? CSR_iOS_RegMask : CSR_AAPCS_RegMask; +} + +const uint32_t* +ARMBaseRegisterInfo::getNoPreservedMask() const { + return CSR_NoRegs_RegMask; +} + +const uint32_t * +ARMBaseRegisterInfo::getThisReturnPreservedMask(const MachineFunction &MF, + CallingConv::ID CC) const { + const ARMSubtarget &STI = MF.getSubtarget(); + // This should return a register mask that is the same as that returned by + // getCallPreservedMask but that additionally preserves the register used for + // the first i32 argument (which must also be the register used to return a + // single i32 return value) + // + // In case that the calling convention does not use the same register for + // both or otherwise does not want to enable this optimization, the function + // should return NULL + if (CC == CallingConv::GHC) + // This is academic becase all GHC calls are (supposed to be) tail calls + return nullptr; + return STI.isTargetDarwin() ? CSR_iOS_ThisReturn_RegMask + : CSR_AAPCS_ThisReturn_RegMask; } BitVector ARMBaseRegisterInfo:: getReservedRegs(const MachineFunction &MF) const { - const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); + const ARMSubtarget &STI = MF.getSubtarget(); + const ARMFrameLowering *TFI = getFrameLowering(MF); // FIXME: avoid re-calculating this every time. BitVector Reserved(getNumRegs()); Reserved.set(ARM::SP); Reserved.set(ARM::PC); Reserved.set(ARM::FPSCR); + Reserved.set(ARM::APSR_NZCV); if (TFI->hasFP(MF)) - Reserved.set(FramePtr); + Reserved.set(getFramePointerReg(STI)); if (hasBasePointer(MF)) Reserved.set(BasePtr); // Some targets reserve R9. @@ -107,251 +148,17 @@ getReservedRegs(const MachineFunction &MF) const { for (unsigned i = 0; i != 16; ++i) Reserved.set(ARM::D16 + i); } - return Reserved; -} + const TargetRegisterClass *RC = &ARM::GPRPairRegClass; + for(TargetRegisterClass::iterator I = RC->begin(), E = RC->end(); I!=E; ++I) + for (MCSubRegIterator SI(*I, this); SI.isValid(); ++SI) + if (Reserved.test(*SI)) Reserved.set(*I); -bool ARMBaseRegisterInfo::isReservedReg(const MachineFunction &MF, - unsigned Reg) const { - const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); - - switch (Reg) { - default: break; - case ARM::SP: - case ARM::PC: - return true; - case ARM::R6: - if (hasBasePointer(MF)) - return true; - break; - case ARM::R7: - case ARM::R11: - if (FramePtr == Reg && TFI->hasFP(MF)) - return true; - break; - case ARM::R9: - return STI.isR9Reserved(); - } - - return false; + return Reserved; } const TargetRegisterClass * -ARMBaseRegisterInfo::getMatchingSuperRegClass(const TargetRegisterClass *A, - const TargetRegisterClass *B, - unsigned SubIdx) const { - switch (SubIdx) { - default: return 0; - case ARM::ssub_0: - case ARM::ssub_1: - case ARM::ssub_2: - case ARM::ssub_3: { - // S sub-registers. - if (A->getSize() == 8) { - if (B == &ARM::SPR_8RegClass) - return &ARM::DPR_8RegClass; - assert(B == &ARM::SPRRegClass && "Expecting SPR register class!"); - if (A == &ARM::DPR_8RegClass) - return A; - return &ARM::DPR_VFP2RegClass; - } - - if (A->getSize() == 16) { - if (B == &ARM::SPR_8RegClass) - return &ARM::QPR_8RegClass; - return &ARM::QPR_VFP2RegClass; - } - - if (A->getSize() == 32) { - if (B == &ARM::SPR_8RegClass) - return 0; // Do not allow coalescing! - return &ARM::QQPR_VFP2RegClass; - } - - assert(A->getSize() == 64 && "Expecting a QQQQ register class!"); - return 0; // Do not allow coalescing! - } - case ARM::dsub_0: - case ARM::dsub_1: - case ARM::dsub_2: - case ARM::dsub_3: { - // D sub-registers. - if (A->getSize() == 16) { - if (B == &ARM::DPR_VFP2RegClass) - return &ARM::QPR_VFP2RegClass; - if (B == &ARM::DPR_8RegClass) - return 0; // Do not allow coalescing! - return A; - } - - if (A->getSize() == 32) { - if (B == &ARM::DPR_VFP2RegClass) - return &ARM::QQPR_VFP2RegClass; - if (B == &ARM::DPR_8RegClass) - return 0; // Do not allow coalescing! - return A; - } - - assert(A->getSize() == 64 && "Expecting a QQQQ register class!"); - if (B != &ARM::DPRRegClass) - return 0; // Do not allow coalescing! - return A; - } - case ARM::dsub_4: - case ARM::dsub_5: - case ARM::dsub_6: - case ARM::dsub_7: { - // D sub-registers of QQQQ registers. - if (A->getSize() == 64 && B == &ARM::DPRRegClass) - return A; - return 0; // Do not allow coalescing! - } - - case ARM::qsub_0: - case ARM::qsub_1: { - // Q sub-registers. - if (A->getSize() == 32) { - if (B == &ARM::QPR_VFP2RegClass) - return &ARM::QQPR_VFP2RegClass; - if (B == &ARM::QPR_8RegClass) - return 0; // Do not allow coalescing! - return A; - } - - assert(A->getSize() == 64 && "Expecting a QQQQ register class!"); - if (B == &ARM::QPRRegClass) - return A; - return 0; // Do not allow coalescing! - } - case ARM::qsub_2: - case ARM::qsub_3: { - // Q sub-registers of QQQQ registers. - if (A->getSize() == 64 && B == &ARM::QPRRegClass) - return A; - return 0; // Do not allow coalescing! - } - } - return 0; -} - -bool -ARMBaseRegisterInfo::canCombineSubRegIndices(const TargetRegisterClass *RC, - SmallVectorImpl &SubIndices, - unsigned &NewSubIdx) const { - - unsigned Size = RC->getSize() * 8; - if (Size < 6) - return 0; - - NewSubIdx = 0; // Whole register. - unsigned NumRegs = SubIndices.size(); - if (NumRegs == 8) { - // 8 D registers -> 1 QQQQ register. - return (Size == 512 && - SubIndices[0] == ARM::dsub_0 && - SubIndices[1] == ARM::dsub_1 && - SubIndices[2] == ARM::dsub_2 && - SubIndices[3] == ARM::dsub_3 && - SubIndices[4] == ARM::dsub_4 && - SubIndices[5] == ARM::dsub_5 && - SubIndices[6] == ARM::dsub_6 && - SubIndices[7] == ARM::dsub_7); - } else if (NumRegs == 4) { - if (SubIndices[0] == ARM::qsub_0) { - // 4 Q registers -> 1 QQQQ register. - return (Size == 512 && - SubIndices[1] == ARM::qsub_1 && - SubIndices[2] == ARM::qsub_2 && - SubIndices[3] == ARM::qsub_3); - } else if (SubIndices[0] == ARM::dsub_0) { - // 4 D registers -> 1 QQ register. - if (Size >= 256 && - SubIndices[1] == ARM::dsub_1 && - SubIndices[2] == ARM::dsub_2 && - SubIndices[3] == ARM::dsub_3) { - if (Size == 512) - NewSubIdx = ARM::qqsub_0; - return true; - } - } else if (SubIndices[0] == ARM::dsub_4) { - // 4 D registers -> 1 QQ register (2nd). - if (Size == 512 && - SubIndices[1] == ARM::dsub_5 && - SubIndices[2] == ARM::dsub_6 && - SubIndices[3] == ARM::dsub_7) { - NewSubIdx = ARM::qqsub_1; - return true; - } - } else if (SubIndices[0] == ARM::ssub_0) { - // 4 S registers -> 1 Q register. - if (Size >= 128 && - SubIndices[1] == ARM::ssub_1 && - SubIndices[2] == ARM::ssub_2 && - SubIndices[3] == ARM::ssub_3) { - if (Size >= 256) - NewSubIdx = ARM::qsub_0; - return true; - } - } - } else if (NumRegs == 2) { - if (SubIndices[0] == ARM::qsub_0) { - // 2 Q registers -> 1 QQ register. - if (Size >= 256 && SubIndices[1] == ARM::qsub_1) { - if (Size == 512) - NewSubIdx = ARM::qqsub_0; - return true; - } - } else if (SubIndices[0] == ARM::qsub_2) { - // 2 Q registers -> 1 QQ register (2nd). - if (Size == 512 && SubIndices[1] == ARM::qsub_3) { - NewSubIdx = ARM::qqsub_1; - return true; - } - } else if (SubIndices[0] == ARM::dsub_0) { - // 2 D registers -> 1 Q register. - if (Size >= 128 && SubIndices[1] == ARM::dsub_1) { - if (Size >= 256) - NewSubIdx = ARM::qsub_0; - return true; - } - } else if (SubIndices[0] == ARM::dsub_2) { - // 2 D registers -> 1 Q register (2nd). - if (Size >= 256 && SubIndices[1] == ARM::dsub_3) { - NewSubIdx = ARM::qsub_1; - return true; - } - } else if (SubIndices[0] == ARM::dsub_4) { - // 2 D registers -> 1 Q register (3rd). - if (Size == 512 && SubIndices[1] == ARM::dsub_5) { - NewSubIdx = ARM::qsub_2; - return true; - } - } else if (SubIndices[0] == ARM::dsub_6) { - // 2 D registers -> 1 Q register (3rd). - if (Size == 512 && SubIndices[1] == ARM::dsub_7) { - NewSubIdx = ARM::qsub_3; - return true; - } - } else if (SubIndices[0] == ARM::ssub_0) { - // 2 S registers -> 1 D register. - if (SubIndices[1] == ARM::ssub_1) { - if (Size >= 128) - NewSubIdx = ARM::dsub_0; - return true; - } - } else if (SubIndices[0] == ARM::ssub_2) { - // 2 S registers -> 1 D register (2nd). - if (Size >= 128 && SubIndices[1] == ARM::ssub_3) { - NewSubIdx = ARM::dsub_1; - return true; - } - } - } - return false; -} - -const TargetRegisterClass* -ARMBaseRegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC) - const { +ARMBaseRegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC, + const MachineFunction &) const { const TargetRegisterClass *Super = RC; TargetRegisterClass::sc_iterator I = RC->getSuperClasses(); do { @@ -362,6 +169,7 @@ ARMBaseRegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC) case ARM::QPRRegClassID: case ARM::QQPRRegClassID: case ARM::QQQQPRRegClassID: + case ARM::GPRPairRegClassID: return Super; } Super = *I++; @@ -370,21 +178,23 @@ ARMBaseRegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC) } const TargetRegisterClass * -ARMBaseRegisterInfo::getPointerRegClass(unsigned Kind) const { - return ARM::GPRRegisterClass; +ARMBaseRegisterInfo::getPointerRegClass(const MachineFunction &MF, unsigned Kind) + const { + return &ARM::GPRRegClass; } const TargetRegisterClass * ARMBaseRegisterInfo::getCrossCopyRegClass(const TargetRegisterClass *RC) const { if (RC == &ARM::CCRRegClass) - return 0; // Can't copy CCR registers. + return &ARM::rGPRRegClass; // Can't copy CCR registers. return RC; } unsigned ARMBaseRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC, MachineFunction &MF) const { - const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); + const ARMSubtarget &STI = MF.getSubtarget(); + const ARMFrameLowering *TFI = getFrameLowering(MF); switch (RC->getID()) { default: @@ -401,158 +211,72 @@ ARMBaseRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC, } } -/// getRawAllocationOrder - Returns the register allocation order for a -/// specified register class with a target-dependent hint. -ArrayRef -ARMBaseRegisterInfo::getRawAllocationOrder(const TargetRegisterClass *RC, - unsigned HintType, unsigned HintReg, - const MachineFunction &MF) const { - const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); - // Alternative register allocation orders when favoring even / odd registers - // of register pairs. - - // No FP, R9 is available. - static const unsigned GPREven1[] = { - ARM::R0, ARM::R2, ARM::R4, ARM::R6, ARM::R8, ARM::R10, - ARM::R1, ARM::R3, ARM::R12,ARM::LR, ARM::R5, ARM::R7, - ARM::R9, ARM::R11 - }; - static const unsigned GPROdd1[] = { - ARM::R1, ARM::R3, ARM::R5, ARM::R7, ARM::R9, ARM::R11, - ARM::R0, ARM::R2, ARM::R12,ARM::LR, ARM::R4, ARM::R6, - ARM::R8, ARM::R10 - }; - - // FP is R7, R9 is available. - static const unsigned GPREven2[] = { - ARM::R0, ARM::R2, ARM::R4, ARM::R8, ARM::R10, - ARM::R1, ARM::R3, ARM::R12,ARM::LR, ARM::R5, ARM::R6, - ARM::R9, ARM::R11 - }; - static const unsigned GPROdd2[] = { - ARM::R1, ARM::R3, ARM::R5, ARM::R9, ARM::R11, - ARM::R0, ARM::R2, ARM::R12,ARM::LR, ARM::R4, ARM::R6, - ARM::R8, ARM::R10 - }; - - // FP is R11, R9 is available. - static const unsigned GPREven3[] = { - ARM::R0, ARM::R2, ARM::R4, ARM::R6, ARM::R8, - ARM::R1, ARM::R3, ARM::R10,ARM::R12,ARM::LR, ARM::R5, ARM::R7, - ARM::R9 - }; - static const unsigned GPROdd3[] = { - ARM::R1, ARM::R3, ARM::R5, ARM::R6, ARM::R9, - ARM::R0, ARM::R2, ARM::R10,ARM::R12,ARM::LR, ARM::R4, ARM::R7, - ARM::R8 - }; - - // No FP, R9 is not available. - static const unsigned GPREven4[] = { - ARM::R0, ARM::R2, ARM::R4, ARM::R6, ARM::R10, - ARM::R1, ARM::R3, ARM::R12,ARM::LR, ARM::R5, ARM::R7, ARM::R8, - ARM::R11 - }; - static const unsigned GPROdd4[] = { - ARM::R1, ARM::R3, ARM::R5, ARM::R7, ARM::R11, - ARM::R0, ARM::R2, ARM::R12,ARM::LR, ARM::R4, ARM::R6, ARM::R8, - ARM::R10 - }; - - // FP is R7, R9 is not available. - static const unsigned GPREven5[] = { - ARM::R0, ARM::R2, ARM::R4, ARM::R10, - ARM::R1, ARM::R3, ARM::R12,ARM::LR, ARM::R5, ARM::R6, ARM::R8, - ARM::R11 - }; - static const unsigned GPROdd5[] = { - ARM::R1, ARM::R3, ARM::R5, ARM::R11, - ARM::R0, ARM::R2, ARM::R12,ARM::LR, ARM::R4, ARM::R6, ARM::R8, - ARM::R10 - }; - - // FP is R11, R9 is not available. - static const unsigned GPREven6[] = { - ARM::R0, ARM::R2, ARM::R4, ARM::R6, - ARM::R1, ARM::R3, ARM::R10,ARM::R12,ARM::LR, ARM::R5, ARM::R7, ARM::R8 - }; - static const unsigned GPROdd6[] = { - ARM::R1, ARM::R3, ARM::R5, ARM::R7, - ARM::R0, ARM::R2, ARM::R10,ARM::R12,ARM::LR, ARM::R4, ARM::R6, ARM::R8 - }; - - // We only support even/odd hints for GPR and rGPR. - if (RC != ARM::GPRRegisterClass && RC != ARM::rGPRRegisterClass) - return RC->getRawAllocationOrder(MF); - - if (HintType == ARMRI::RegPairEven) { - if (isPhysicalRegister(HintReg) && getRegisterPairEven(HintReg, MF) == 0) - // It's no longer possible to fulfill this hint. Return the default - // allocation order. - return RC->getRawAllocationOrder(MF); - - if (!TFI->hasFP(MF)) { - if (!STI.isR9Reserved()) - return makeArrayRef(GPREven1); - else - return makeArrayRef(GPREven4); - } else if (FramePtr == ARM::R7) { - if (!STI.isR9Reserved()) - return makeArrayRef(GPREven2); - else - return makeArrayRef(GPREven5); - } else { // FramePtr == ARM::R11 - if (!STI.isR9Reserved()) - return makeArrayRef(GPREven3); - else - return makeArrayRef(GPREven6); - } - } else if (HintType == ARMRI::RegPairOdd) { - if (isPhysicalRegister(HintReg) && getRegisterPairOdd(HintReg, MF) == 0) - // It's no longer possible to fulfill this hint. Return the default - // allocation order. - return RC->getRawAllocationOrder(MF); - - if (!TFI->hasFP(MF)) { - if (!STI.isR9Reserved()) - return makeArrayRef(GPROdd1); - else - return makeArrayRef(GPROdd4); - } else if (FramePtr == ARM::R7) { - if (!STI.isR9Reserved()) - return makeArrayRef(GPROdd2); - else - return makeArrayRef(GPROdd5); - } else { // FramePtr == ARM::R11 - if (!STI.isR9Reserved()) - return makeArrayRef(GPROdd3); - else - return makeArrayRef(GPROdd6); - } - } - return RC->getRawAllocationOrder(MF); +// Get the other register in a GPRPair. +static unsigned getPairedGPR(unsigned Reg, bool Odd, const MCRegisterInfo *RI) { + for (MCSuperRegIterator Supers(Reg, RI); Supers.isValid(); ++Supers) + if (ARM::GPRPairRegClass.contains(*Supers)) + return RI->getSubReg(*Supers, Odd ? ARM::gsub_1 : ARM::gsub_0); + return 0; } -/// ResolveRegAllocHint - Resolves the specified register allocation hint -/// to a physical register. Returns the physical register if it is successful. -unsigned -ARMBaseRegisterInfo::ResolveRegAllocHint(unsigned Type, unsigned Reg, - const MachineFunction &MF) const { - if (Reg == 0 || !isPhysicalRegister(Reg)) - return 0; - if (Type == 0) - return Reg; - else if (Type == (unsigned)ARMRI::RegPairOdd) - // Odd register. - return getRegisterPairOdd(Reg, MF); - else if (Type == (unsigned)ARMRI::RegPairEven) - // Even register. - return getRegisterPairEven(Reg, MF); - return 0; +// Resolve the RegPairEven / RegPairOdd register allocator hints. +void +ARMBaseRegisterInfo::getRegAllocationHints(unsigned VirtReg, + ArrayRef Order, + SmallVectorImpl &Hints, + const MachineFunction &MF, + const VirtRegMap *VRM, + const LiveRegMatrix *Matrix) const { + const MachineRegisterInfo &MRI = MF.getRegInfo(); + std::pair Hint = MRI.getRegAllocationHint(VirtReg); + + unsigned Odd; + switch (Hint.first) { + case ARMRI::RegPairEven: + Odd = 0; + break; + case ARMRI::RegPairOdd: + Odd = 1; + break; + default: + TargetRegisterInfo::getRegAllocationHints(VirtReg, Order, Hints, MF, VRM); + return; + } + + // This register should preferably be even (Odd == 0) or odd (Odd == 1). + // Check if the other part of the pair has already been assigned, and provide + // the paired register as the first hint. + unsigned Paired = Hint.second; + if (Paired == 0) + return; + + unsigned PairedPhys = 0; + if (TargetRegisterInfo::isPhysicalRegister(Paired)) { + PairedPhys = Paired; + } else if (VRM && VRM->hasPhys(Paired)) { + PairedPhys = getPairedGPR(VRM->getPhys(Paired), Odd, this); + } + + // First prefer the paired physreg. + if (PairedPhys && + std::find(Order.begin(), Order.end(), PairedPhys) != Order.end()) + Hints.push_back(PairedPhys); + + // Then prefer even or odd registers. + for (unsigned I = 0, E = Order.size(); I != E; ++I) { + unsigned Reg = Order[I]; + if (Reg == PairedPhys || (getEncodingValue(Reg) & 1) != Odd) + continue; + // Don't provide hints that are paired to a reserved register. + unsigned Paired = getPairedGPR(Reg, !Odd, this); + if (!Paired || MRI.isReserved(Paired)) + continue; + Hints.push_back(Reg); + } } void -ARMBaseRegisterInfo::UpdateRegAllocHint(unsigned Reg, unsigned NewReg, +ARMBaseRegisterInfo::updateRegAllocHint(unsigned Reg, unsigned NewReg, MachineFunction &MF) const { MachineRegisterInfo *MRI = &MF.getRegInfo(); std::pair Hint = MRI->getRegAllocationHint(Reg); @@ -565,43 +289,26 @@ ARMBaseRegisterInfo::UpdateRegAllocHint(unsigned Reg, unsigned NewReg, // change. unsigned OtherReg = Hint.second; Hint = MRI->getRegAllocationHint(OtherReg); - if (Hint.second == Reg) - // Make sure the pair has not already divorced. + // Make sure the pair has not already divorced. + if (Hint.second == Reg) { MRI->setRegAllocationHint(OtherReg, Hint.first, NewReg); - } -} - -bool -ARMBaseRegisterInfo::avoidWriteAfterWrite(const TargetRegisterClass *RC) const { - // CortexA9 has a Write-after-write hazard for NEON registers. - if (!STI.isCortexA9()) - return false; - - switch (RC->getID()) { - case ARM::DPRRegClassID: - case ARM::DPR_8RegClassID: - case ARM::DPR_VFP2RegClassID: - case ARM::QPRRegClassID: - case ARM::QPR_8RegClassID: - case ARM::QPR_VFP2RegClassID: - case ARM::SPRRegClassID: - case ARM::SPR_8RegClassID: - // Avoid reusing S, D, and Q registers. - // Don't increase register pressure for QQ and QQQQ. - return true; - default: - return false; + if (TargetRegisterInfo::isVirtualRegister(NewReg)) + MRI->setRegAllocationHint(NewReg, + Hint.first == (unsigned)ARMRI::RegPairOdd ? ARMRI::RegPairEven + : ARMRI::RegPairOdd, OtherReg); + } } } bool ARMBaseRegisterInfo::hasBasePointer(const MachineFunction &MF) const { const MachineFrameInfo *MFI = MF.getFrameInfo(); const ARMFunctionInfo *AFI = MF.getInfo(); + const ARMFrameLowering *TFI = getFrameLowering(MF); - if (!EnableBasePointer) - return false; - - if (needsStackRealignment(MF) && MFI->hasVarSizedObjects()) + // When outgoing call frames are so large that we adjust the stack pointer + // around the call, we can no longer use the stack pointer to reach the + // emergency spill slot. + if (needsStackRealignment(MF) && !TFI->hasReservedCallFrame(MF)) return true; // Thumb has trouble with negative offsets from the FP. Thumb2 has a limited @@ -625,25 +332,28 @@ bool ARMBaseRegisterInfo::hasBasePointer(const MachineFunction &MF) const { } bool ARMBaseRegisterInfo::canRealignStack(const MachineFunction &MF) const { - const MachineFrameInfo *MFI = MF.getFrameInfo(); + const MachineRegisterInfo *MRI = &MF.getRegInfo(); const ARMFunctionInfo *AFI = MF.getInfo(); + const ARMFrameLowering *TFI = getFrameLowering(MF); // We can't realign the stack if: // 1. Dynamic stack realignment is explicitly disabled, // 2. This is a Thumb1 function (it's not useful, so we don't bother), or // 3. There are VLAs in the function and the base pointer is disabled. - return (MF.getTarget().Options.RealignStack && !AFI->isThumb1OnlyFunction() && - (!MFI->hasVarSizedObjects() || EnableBasePointer)); -} - -bool ARMBaseRegisterInfo:: -needsStackRealignment(const MachineFunction &MF) const { - const MachineFrameInfo *MFI = MF.getFrameInfo(); - const Function *F = MF.getFunction(); - unsigned StackAlign = MF.getTarget().getFrameLowering()->getStackAlignment(); - bool requiresRealignment = ((MFI->getLocalFrameMaxAlign() > StackAlign) || - F->hasFnAttr(Attribute::StackAlignment)); - - return requiresRealignment && canRealignStack(MF); + if (!TargetRegisterInfo::canRealignStack(MF)) + return false; + if (AFI->isThumb1OnlyFunction()) + return false; + // Stack realignment requires a frame pointer. If we already started + // register allocation with frame pointer elimination, it is too late now. + if (!MRI->canReserveReg(getFramePointerReg(MF.getSubtarget()))) + return false; + // We may also need a base pointer if there are dynamic allocas or stack + // pointer adjustments around calls. + if (TFI->hasReservedCallFrame(MF)) + return true; + // A base pointer is required and allowed. Check that it isn't too late to + // reserve it. + return MRI->canReserveReg(BasePtr); } bool ARMBaseRegisterInfo:: @@ -657,129 +367,14 @@ cannotEliminateFrame(const MachineFunction &MF) const { unsigned ARMBaseRegisterInfo::getFrameRegister(const MachineFunction &MF) const { - const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); + const ARMSubtarget &STI = MF.getSubtarget(); + const ARMFrameLowering *TFI = getFrameLowering(MF); if (TFI->hasFP(MF)) - return FramePtr; + return getFramePointerReg(STI); return ARM::SP; } -unsigned ARMBaseRegisterInfo::getEHExceptionRegister() const { - llvm_unreachable("What is the exception register"); - return 0; -} - -unsigned ARMBaseRegisterInfo::getEHHandlerRegister() const { - llvm_unreachable("What is the exception handler register"); - return 0; -} - -unsigned ARMBaseRegisterInfo::getRegisterPairEven(unsigned Reg, - const MachineFunction &MF) const { - switch (Reg) { - default: break; - // Return 0 if either register of the pair is a special register. - // So no R12, etc. - case ARM::R1: return ARM::R0; - case ARM::R3: return ARM::R2; - case ARM::R5: return ARM::R4; - case ARM::R7: - return (isReservedReg(MF, ARM::R7) || isReservedReg(MF, ARM::R6)) - ? 0 : ARM::R6; - case ARM::R9: return isReservedReg(MF, ARM::R9) ? 0 :ARM::R8; - case ARM::R11: return isReservedReg(MF, ARM::R11) ? 0 : ARM::R10; - - case ARM::S1: return ARM::S0; - case ARM::S3: return ARM::S2; - case ARM::S5: return ARM::S4; - case ARM::S7: return ARM::S6; - case ARM::S9: return ARM::S8; - case ARM::S11: return ARM::S10; - case ARM::S13: return ARM::S12; - case ARM::S15: return ARM::S14; - case ARM::S17: return ARM::S16; - case ARM::S19: return ARM::S18; - case ARM::S21: return ARM::S20; - case ARM::S23: return ARM::S22; - case ARM::S25: return ARM::S24; - case ARM::S27: return ARM::S26; - case ARM::S29: return ARM::S28; - case ARM::S31: return ARM::S30; - - case ARM::D1: return ARM::D0; - case ARM::D3: return ARM::D2; - case ARM::D5: return ARM::D4; - case ARM::D7: return ARM::D6; - case ARM::D9: return ARM::D8; - case ARM::D11: return ARM::D10; - case ARM::D13: return ARM::D12; - case ARM::D15: return ARM::D14; - case ARM::D17: return ARM::D16; - case ARM::D19: return ARM::D18; - case ARM::D21: return ARM::D20; - case ARM::D23: return ARM::D22; - case ARM::D25: return ARM::D24; - case ARM::D27: return ARM::D26; - case ARM::D29: return ARM::D28; - case ARM::D31: return ARM::D30; - } - - return 0; -} - -unsigned ARMBaseRegisterInfo::getRegisterPairOdd(unsigned Reg, - const MachineFunction &MF) const { - switch (Reg) { - default: break; - // Return 0 if either register of the pair is a special register. - // So no R12, etc. - case ARM::R0: return ARM::R1; - case ARM::R2: return ARM::R3; - case ARM::R4: return ARM::R5; - case ARM::R6: - return (isReservedReg(MF, ARM::R7) || isReservedReg(MF, ARM::R6)) - ? 0 : ARM::R7; - case ARM::R8: return isReservedReg(MF, ARM::R9) ? 0 :ARM::R9; - case ARM::R10: return isReservedReg(MF, ARM::R11) ? 0 : ARM::R11; - - case ARM::S0: return ARM::S1; - case ARM::S2: return ARM::S3; - case ARM::S4: return ARM::S5; - case ARM::S6: return ARM::S7; - case ARM::S8: return ARM::S9; - case ARM::S10: return ARM::S11; - case ARM::S12: return ARM::S13; - case ARM::S14: return ARM::S15; - case ARM::S16: return ARM::S17; - case ARM::S18: return ARM::S19; - case ARM::S20: return ARM::S21; - case ARM::S22: return ARM::S23; - case ARM::S24: return ARM::S25; - case ARM::S26: return ARM::S27; - case ARM::S28: return ARM::S29; - case ARM::S30: return ARM::S31; - - case ARM::D0: return ARM::D1; - case ARM::D2: return ARM::D3; - case ARM::D4: return ARM::D5; - case ARM::D6: return ARM::D7; - case ARM::D8: return ARM::D9; - case ARM::D10: return ARM::D11; - case ARM::D12: return ARM::D13; - case ARM::D14: return ARM::D15; - case ARM::D16: return ARM::D17; - case ARM::D18: return ARM::D19; - case ARM::D20: return ARM::D21; - case ARM::D22: return ARM::D23; - case ARM::D24: return ARM::D25; - case ARM::D26: return ARM::D27; - case ARM::D28: return ARM::D29; - case ARM::D30: return ARM::D31; - } - - return 0; -} - /// emitLoadConstPool - Emits a load from constpool to materialize the /// specified immediate. void ARMBaseRegisterInfo:: @@ -790,6 +385,7 @@ emitLoadConstPool(MachineBasicBlock &MBB, ARMCC::CondCodes Pred, unsigned PredReg, unsigned MIFlags) const { MachineFunction &MF = *MBB.getParent(); + const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo(); MachineConstantPool *ConstantPool = MF.getConstantPool(); const Constant *C = ConstantInt::get(Type::getInt32Ty(MF.getFunction()->getContext()), Val); @@ -808,78 +404,25 @@ requiresRegisterScavenging(const MachineFunction &MF) const { } bool ARMBaseRegisterInfo:: -requiresFrameIndexScavenging(const MachineFunction &MF) const { +trackLivenessAfterRegAlloc(const MachineFunction &MF) const { return true; } bool ARMBaseRegisterInfo:: -requiresVirtualBaseRegisters(const MachineFunction &MF) const { - return EnableLocalStackAlloc; -} - -static void -emitSPUpdate(bool isARM, - MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI, - DebugLoc dl, const ARMBaseInstrInfo &TII, - int NumBytes, - ARMCC::CondCodes Pred = ARMCC::AL, unsigned PredReg = 0) { - if (isARM) - emitARMRegPlusImmediate(MBB, MBBI, dl, ARM::SP, ARM::SP, NumBytes, - Pred, PredReg, TII); - else - emitT2RegPlusImmediate(MBB, MBBI, dl, ARM::SP, ARM::SP, NumBytes, - Pred, PredReg, TII); +requiresFrameIndexScavenging(const MachineFunction &MF) const { + return true; } - -void ARMBaseRegisterInfo:: -eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB, - MachineBasicBlock::iterator I) const { - const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); - if (!TFI->hasReservedCallFrame(MF)) { - // If we have alloca, convert as follows: - // ADJCALLSTACKDOWN -> sub, sp, sp, amount - // ADJCALLSTACKUP -> add, sp, sp, amount - MachineInstr *Old = I; - DebugLoc dl = Old->getDebugLoc(); - unsigned Amount = Old->getOperand(0).getImm(); - if (Amount != 0) { - // We need to keep the stack aligned properly. To do this, we round the - // amount of space needed for the outgoing arguments up to the next - // alignment boundary. - unsigned Align = TFI->getStackAlignment(); - Amount = (Amount+Align-1)/Align*Align; - - ARMFunctionInfo *AFI = MF.getInfo(); - assert(!AFI->isThumb1OnlyFunction() && - "This eliminateCallFramePseudoInstr does not support Thumb1!"); - bool isARM = !AFI->isThumbFunction(); - - // Replace the pseudo instruction with a new instruction... - unsigned Opc = Old->getOpcode(); - int PIdx = Old->findFirstPredOperandIdx(); - ARMCC::CondCodes Pred = (PIdx == -1) - ? ARMCC::AL : (ARMCC::CondCodes)Old->getOperand(PIdx).getImm(); - if (Opc == ARM::ADJCALLSTACKDOWN || Opc == ARM::tADJCALLSTACKDOWN) { - // Note: PredReg is operand 2 for ADJCALLSTACKDOWN. - unsigned PredReg = Old->getOperand(2).getReg(); - emitSPUpdate(isARM, MBB, I, dl, TII, -Amount, Pred, PredReg); - } else { - // Note: PredReg is operand 3 for ADJCALLSTACKUP. - unsigned PredReg = Old->getOperand(3).getReg(); - assert(Opc == ARM::ADJCALLSTACKUP || Opc == ARM::tADJCALLSTACKUP); - emitSPUpdate(isARM, MBB, I, dl, TII, Amount, Pred, PredReg); - } - } - } - MBB.erase(I); +bool ARMBaseRegisterInfo:: +requiresVirtualBaseRegisters(const MachineFunction &MF) const { + return true; } int64_t ARMBaseRegisterInfo:: getFrameIndexInstrOffset(const MachineInstr *MI, int Idx) const { const MCInstrDesc &Desc = MI->getDesc(); unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask); - int64_t InstrOffs = 0;; + int64_t InstrOffs = 0; int Scale = 1; unsigned ImmIdx = 0; switch (AddrMode) { @@ -920,7 +463,6 @@ getFrameIndexInstrOffset(const MachineInstr *MI, int Idx) const { } default: llvm_unreachable("Unsupported addressing mode!"); - break; } return InstrOffs * Scale; @@ -954,8 +496,6 @@ needsFrameBaseReg(MachineInstr *MI, int64_t Offset) const { case ARM::VLDRS: case ARM::VLDRD: case ARM::VSTRS: case ARM::VSTRD: case ARM::tSTRspi: case ARM::tLDRspi: - if (ForceAllBaseRegAlloc) - return true; break; default: return false; @@ -967,7 +507,7 @@ needsFrameBaseReg(MachineInstr *MI, int64_t Offset) const { // Note that the incoming offset is based on the SP value at function entry, // so it'll be negative. MachineFunction &MF = *MI->getParent()->getParent(); - const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering(); + const ARMFrameLowering *TFI = getFrameLowering(MF); MachineFrameInfo *MFI = MF.getFrameInfo(); ARMFunctionInfo *AFI = MF.getInfo(); @@ -983,21 +523,20 @@ needsFrameBaseReg(MachineInstr *MI, int64_t Offset) const { // 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 = -Offset; Offset += MFI->getLocalFrameSize(); // Assume that we'll have at least some spill slots allocated. // FIXME: This is a total SWAG number. We should run some statistics // and pick a real one. Offset += 128; // 128 bytes of spill slots - // If there is a frame pointer, try using it. + // If there's a frame pointer and the addressing mode allows it, try using it. // The FP is only available if there is no dynamic realignment. We // don't know for sure yet whether we'll need that, so we guess based // on whether there are any local variables that would trigger it. unsigned StackAlign = TFI->getStackAlignment(); - if (TFI->hasFP(MF) && + if (TFI->hasFP(MF) && !((MFI->getLocalFrameMaxAlign() > StackAlign) && canRealignStack(MF))) { - if (isFrameOffsetLegal(MI, FPOffset)) + if (isFrameOffsetLegal(MI, getFrameRegister(MF), FPOffset)) return false; } // If we can reference via the stack pointer, try that. @@ -1005,7 +544,7 @@ needsFrameBaseReg(MachineInstr *MI, int64_t Offset) const { // to only disallow SP relative references in the live range of // the VLA(s). In practice, it's unclear how much difference that // would make, but it may be worth doing. - if (!MFI->hasVarSizedObjects() && isFrameOffsetLegal(MI, Offset)) + if (!MFI->hasVarSizedObjects() && isFrameOffsetLegal(MI, ARM::SP, Offset)) return false; // The offset likely isn't legal, we want to allocate a virtual base register. @@ -1020,30 +559,32 @@ materializeFrameBaseRegister(MachineBasicBlock *MBB, int64_t Offset) const { ARMFunctionInfo *AFI = MBB->getParent()->getInfo(); unsigned ADDriOpc = !AFI->isThumbFunction() ? ARM::ADDri : - (AFI->isThumb1OnlyFunction() ? ARM::tADDrSPi : ARM::t2ADDri); + (AFI->isThumb1OnlyFunction() ? ARM::tADDframe : ARM::t2ADDri); MachineBasicBlock::iterator Ins = MBB->begin(); DebugLoc DL; // Defaults to "unknown" if (Ins != MBB->end()) DL = Ins->getDebugLoc(); - const MCInstrDesc &MCID = TII.get(ADDriOpc); + const MachineFunction &MF = *MBB->getParent(); MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo(); - MRI.constrainRegClass(BaseReg, TII.getRegClass(MCID, 0, this)); + const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo(); + const MCInstrDesc &MCID = TII.get(ADDriOpc); + MRI.constrainRegClass(BaseReg, TII.getRegClass(MCID, 0, this, MF)); - MachineInstrBuilder MIB = AddDefaultPred(BuildMI(*MBB, Ins, DL, MCID, BaseReg) - .addFrameIndex(FrameIdx).addImm(Offset)); + MachineInstrBuilder MIB = BuildMI(*MBB, Ins, DL, MCID, BaseReg) + .addFrameIndex(FrameIdx).addImm(Offset); if (!AFI->isThumb1OnlyFunction()) - AddDefaultCC(MIB); + AddDefaultCC(AddDefaultPred(MIB)); } -void -ARMBaseRegisterInfo::resolveFrameIndex(MachineBasicBlock::iterator I, - unsigned BaseReg, int64_t Offset) const { - MachineInstr &MI = *I; +void ARMBaseRegisterInfo::resolveFrameIndex(MachineInstr &MI, unsigned BaseReg, + int64_t Offset) const { MachineBasicBlock &MBB = *MI.getParent(); MachineFunction &MF = *MBB.getParent(); + const ARMBaseInstrInfo &TII = + *static_cast(MF.getSubtarget().getInstrInfo()); ARMFunctionInfo *AFI = MF.getInfo(); int Off = Offset; // ARM doesn't need the general 64-bit offsets unsigned i = 0; @@ -1066,7 +607,7 @@ ARMBaseRegisterInfo::resolveFrameIndex(MachineBasicBlock::iterator I, (void)Done; } -bool ARMBaseRegisterInfo::isFrameOffsetLegal(const MachineInstr *MI, +bool ARMBaseRegisterInfo::isFrameOffsetLegal(const MachineInstr *MI, unsigned BaseReg, int64_t Offset) const { const MCInstrDesc &Desc = MI->getDesc(); unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask); @@ -1110,13 +651,12 @@ bool ARMBaseRegisterInfo::isFrameOffsetLegal(const MachineInstr *MI, NumBits = 8; break; case ARMII::AddrModeT1_s: - NumBits = 5; + NumBits = (BaseReg == ARM::SP ? 8 : 5); Scale = 4; isSigned = false; break; default: llvm_unreachable("Unsupported addressing mode!"); - break; } Offset += getFrameIndexInstrOffset(MI, i); @@ -1137,41 +677,46 @@ bool ARMBaseRegisterInfo::isFrameOffsetLegal(const MachineInstr *MI, void ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, - int SPAdj, RegScavenger *RS) const { - unsigned i = 0; + int SPAdj, unsigned FIOperandNum, + RegScavenger *RS) const { MachineInstr &MI = *II; MachineBasicBlock &MBB = *MI.getParent(); MachineFunction &MF = *MBB.getParent(); - const ARMFrameLowering *TFI = - static_cast(MF.getTarget().getFrameLowering()); + const ARMBaseInstrInfo &TII = + *static_cast(MF.getSubtarget().getInstrInfo()); + const ARMFrameLowering *TFI = getFrameLowering(MF); ARMFunctionInfo *AFI = MF.getInfo(); assert(!AFI->isThumb1OnlyFunction() && "This eliminateFrameIndex does not support Thumb1!"); - - while (!MI.getOperand(i).isFI()) { - ++i; - assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!"); - } - - int FrameIndex = MI.getOperand(i).getIndex(); + int FrameIndex = MI.getOperand(FIOperandNum).getIndex(); unsigned FrameReg; int Offset = TFI->ResolveFrameIndexReference(MF, FrameIndex, FrameReg, SPAdj); - // Special handling of dbg_value instructions. - if (MI.isDebugValue()) { - MI.getOperand(i). ChangeToRegister(FrameReg, false /*isDef*/); - MI.getOperand(i+1).ChangeToImmediate(Offset); - return; + // PEI::scavengeFrameVirtualRegs() cannot accurately track SPAdj because the + // call frame setup/destroy instructions have already been eliminated. That + // means the stack pointer cannot be used to access the emergency spill slot + // when !hasReservedCallFrame(). +#ifndef NDEBUG + if (RS && FrameReg == ARM::SP && RS->isScavengingFrameIndex(FrameIndex)){ + assert(TFI->hasReservedCallFrame(MF) && + "Cannot use SP to access the emergency spill slot in " + "functions without a reserved call frame"); + assert(!MF.getFrameInfo()->hasVarSizedObjects() && + "Cannot use SP to access the emergency spill slot in " + "functions with variable sized frame objects"); } +#endif // NDEBUG + + assert(!MI.isDebugValue() && "DBG_VALUEs should be handled in target-independent code"); // Modify MI as necessary to handle as much of 'Offset' as possible bool Done = false; if (!AFI->isThumbFunction()) - Done = rewriteARMFrameIndex(MI, i, FrameReg, Offset, TII); + Done = rewriteARMFrameIndex(MI, FIOperandNum, FrameReg, Offset, TII); else { assert(AFI->isThumb2Function()); - Done = rewriteT2FrameIndex(MI, i, FrameReg, Offset, TII); + Done = rewriteT2FrameIndex(MI, FIOperandNum, FrameReg, Offset, TII); } if (Done) return; @@ -1191,9 +736,9 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, unsigned PredReg = (PIdx == -1) ? 0 : MI.getOperand(PIdx+1).getReg(); if (Offset == 0) // Must be addrmode4/6. - MI.getOperand(i).ChangeToRegister(FrameReg, false, false, false); + MI.getOperand(FIOperandNum).ChangeToRegister(FrameReg, false, false, false); else { - ScratchReg = MF.getRegInfo().createVirtualRegister(ARM::GPRRegisterClass); + ScratchReg = MF.getRegInfo().createVirtualRegister(&ARM::GPRRegClass); if (!AFI->isThumbFunction()) emitARMRegPlusImmediate(MBB, II, MI.getDebugLoc(), ScratchReg, FrameReg, Offset, Pred, PredReg, TII); @@ -1203,6 +748,63 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, Offset, Pred, PredReg, TII); } // Update the original instruction to use the scratch register. - MI.getOperand(i).ChangeToRegister(ScratchReg, false, false, true); + MI.getOperand(FIOperandNum).ChangeToRegister(ScratchReg, false, false,true); + } +} + +bool ARMBaseRegisterInfo::shouldCoalesce(MachineInstr *MI, + const TargetRegisterClass *SrcRC, + unsigned SubReg, + const TargetRegisterClass *DstRC, + unsigned DstSubReg, + const TargetRegisterClass *NewRC) const { + auto MBB = MI->getParent(); + auto MF = MBB->getParent(); + const MachineRegisterInfo &MRI = MF->getRegInfo(); + // If not copying into a sub-register this should be ok because we shouldn't + // need to split the reg. + if (!DstSubReg) + return true; + // Small registers don't frequently cause a problem, so we can coalesce them. + if (NewRC->getSize() < 32 && DstRC->getSize() < 32 && SrcRC->getSize() < 32) + return true; + + auto NewRCWeight = + MRI.getTargetRegisterInfo()->getRegClassWeight(NewRC); + auto SrcRCWeight = + MRI.getTargetRegisterInfo()->getRegClassWeight(SrcRC); + auto DstRCWeight = + MRI.getTargetRegisterInfo()->getRegClassWeight(DstRC); + // If the source register class is more expensive than the destination, the + // coalescing is probably profitable. + if (SrcRCWeight.RegWeight > NewRCWeight.RegWeight) + return true; + if (DstRCWeight.RegWeight > NewRCWeight.RegWeight) + return true; + + // If the register allocator isn't constrained, we can always allow coalescing + // unfortunately we don't know yet if we will be constrained. + // The goal of this heuristic is to restrict how many expensive registers + // we allow to coalesce in a given basic block. + auto AFI = MF->getInfo(); + auto It = AFI->getCoalescedWeight(MBB); + + DEBUG(dbgs() << "\tARM::shouldCoalesce - Coalesced Weight: " + << It->second << "\n"); + DEBUG(dbgs() << "\tARM::shouldCoalesce - Reg Weight: " + << NewRCWeight.RegWeight << "\n"); + + // This number is the largest round number that which meets the criteria: + // (1) addresses PR18825 + // (2) generates better code in some test cases (like vldm-shed-a9.ll) + // (3) Doesn't regress any test cases (in-tree, test-suite, and SPEC) + // In practice the SizeMultiplier will only factor in for straight line code + // that uses a lot of NEON vectors, which isn't terribly common. + unsigned SizeMultiplier = MBB->size()/100; + SizeMultiplier = SizeMultiplier ? SizeMultiplier : 1; + if (It->second < NewRCWeight.WeightLimit * SizeMultiplier) { + It->second += NewRCWeight.RegWeight; + return true; } + return false; }