X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FTarget%2FX86%2FX86InstrInfo.h;h=8da68b5701779e6bf7888d4806e3b59981a7728e;hp=f5b0cc9152727d4d3d83660b64051b1cb459e7bb;hb=7a2bdde0a0eebcd2125055e0eacaca040f0b766c;hpb=dc54d317e7a381ef8e4aca80d54ad1466bb85dda diff --git a/lib/Target/X86/X86InstrInfo.h b/lib/Target/X86/X86InstrInfo.h index f5b0cc91527..8da68b57017 100644 --- a/lib/Target/X86/X86InstrInfo.h +++ b/lib/Target/X86/X86InstrInfo.h @@ -18,13 +18,30 @@ #include "X86.h" #include "X86RegisterInfo.h" #include "llvm/ADT/DenseMap.h" -#include "llvm/Target/TargetRegisterInfo.h" namespace llvm { class X86RegisterInfo; class X86TargetMachine; namespace X86 { + // Enums for memory operand decoding. Each memory operand is represented with + // a 5 operand sequence in the form: + // [BaseReg, ScaleAmt, IndexReg, Disp, Segment] + // These enums help decode this. + enum { + AddrBaseReg = 0, + AddrScaleAmt = 1, + AddrIndexReg = 2, + AddrDisp = 3, + + /// AddrSegmentReg - The operand # of the segment in the memory operand. + AddrSegmentReg = 4, + + /// AddrNumOperands - Total number of operands in a memory reference. + AddrNumOperands = 5 + }; + + // X86 specific condition code. These correspond to X86_*_COND in // X86InstrInfo.td. They must be kept in synch. enum CondCode { @@ -55,23 +72,181 @@ namespace X86 { COND_INVALID }; - + // Turn condition code into conditional branch opcode. unsigned GetCondBranchFromCond(CondCode CC); - + /// GetOppositeBranchCondition - Return the inverse of the specified cond, /// e.g. turning COND_E to COND_NE. CondCode GetOppositeBranchCondition(X86::CondCode CC); } - + +/// X86II - This namespace holds all of the target specific flags that +/// instruction info tracks. +/// +namespace X86II { + /// Target Operand Flag enum. + enum TOF { + //===------------------------------------------------------------------===// + // X86 Specific MachineOperand flags. + + MO_NO_FLAG, + + /// MO_GOT_ABSOLUTE_ADDRESS - On a symbol operand, this represents a + /// relocation of: + /// SYMBOL_LABEL + [. - PICBASELABEL] + MO_GOT_ABSOLUTE_ADDRESS, + + /// MO_PIC_BASE_OFFSET - On a symbol operand this indicates that the + /// immediate should get the value of the symbol minus the PIC base label: + /// SYMBOL_LABEL - PICBASELABEL + MO_PIC_BASE_OFFSET, + + /// MO_GOT - On a symbol operand this indicates that the immediate is the + /// offset to the GOT entry for the symbol name from the base of the GOT. + /// + /// See the X86-64 ELF ABI supplement for more details. + /// SYMBOL_LABEL @GOT + MO_GOT, + + /// MO_GOTOFF - On a symbol operand this indicates that the immediate is + /// the offset to the location of the symbol name from the base of the GOT. + /// + /// See the X86-64 ELF ABI supplement for more details. + /// SYMBOL_LABEL @GOTOFF + MO_GOTOFF, + + /// MO_GOTPCREL - On a symbol operand this indicates that the immediate is + /// offset to the GOT entry for the symbol name from the current code + /// location. + /// + /// See the X86-64 ELF ABI supplement for more details. + /// SYMBOL_LABEL @GOTPCREL + MO_GOTPCREL, + + /// MO_PLT - On a symbol operand this indicates that the immediate is + /// offset to the PLT entry of symbol name from the current code location. + /// + /// See the X86-64 ELF ABI supplement for more details. + /// SYMBOL_LABEL @PLT + MO_PLT, + + /// MO_TLSGD - On a symbol operand this indicates that the immediate is + /// some TLS offset. + /// + /// See 'ELF Handling for Thread-Local Storage' for more details. + /// SYMBOL_LABEL @TLSGD + MO_TLSGD, + + /// MO_GOTTPOFF - On a symbol operand this indicates that the immediate is + /// some TLS offset. + /// + /// See 'ELF Handling for Thread-Local Storage' for more details. + /// SYMBOL_LABEL @GOTTPOFF + MO_GOTTPOFF, + + /// MO_INDNTPOFF - On a symbol operand this indicates that the immediate is + /// some TLS offset. + /// + /// See 'ELF Handling for Thread-Local Storage' for more details. + /// SYMBOL_LABEL @INDNTPOFF + MO_INDNTPOFF, + + /// MO_TPOFF - On a symbol operand this indicates that the immediate is + /// some TLS offset. + /// + /// See 'ELF Handling for Thread-Local Storage' for more details. + /// SYMBOL_LABEL @TPOFF + MO_TPOFF, + + /// MO_NTPOFF - On a symbol operand this indicates that the immediate is + /// some TLS offset. + /// + /// See 'ELF Handling for Thread-Local Storage' for more details. + /// SYMBOL_LABEL @NTPOFF + MO_NTPOFF, + + /// MO_DLLIMPORT - On a symbol operand "FOO", this indicates that the + /// reference is actually to the "__imp_FOO" symbol. This is used for + /// dllimport linkage on windows. + MO_DLLIMPORT, + + /// MO_DARWIN_STUB - On a symbol operand "FOO", this indicates that the + /// reference is actually to the "FOO$stub" symbol. This is used for calls + /// and jumps to external functions on Tiger and earlier. + MO_DARWIN_STUB, + + /// MO_DARWIN_NONLAZY - On a symbol operand "FOO", this indicates that the + /// reference is actually to the "FOO$non_lazy_ptr" symbol, which is a + /// non-PIC-base-relative reference to a non-hidden dyld lazy pointer stub. + MO_DARWIN_NONLAZY, + + /// MO_DARWIN_NONLAZY_PIC_BASE - On a symbol operand "FOO", this indicates + /// that the reference is actually to "FOO$non_lazy_ptr - PICBASE", which is + /// a PIC-base-relative reference to a non-hidden dyld lazy pointer stub. + MO_DARWIN_NONLAZY_PIC_BASE, + + /// MO_DARWIN_HIDDEN_NONLAZY_PIC_BASE - On a symbol operand "FOO", this + /// indicates that the reference is actually to "FOO$non_lazy_ptr -PICBASE", + /// which is a PIC-base-relative reference to a hidden dyld lazy pointer + /// stub. + MO_DARWIN_HIDDEN_NONLAZY_PIC_BASE, + + /// MO_TLVP - On a symbol operand this indicates that the immediate is + /// some TLS offset. + /// + /// This is the TLS offset for the Darwin TLS mechanism. + MO_TLVP, + + /// MO_TLVP_PIC_BASE - On a symbol operand this indicates that the immediate + /// is some TLS offset from the picbase. + /// + /// This is the 32-bit TLS offset for Darwin TLS in PIC mode. + MO_TLVP_PIC_BASE + }; +} + +/// isGlobalStubReference - Return true if the specified TargetFlag operand is +/// a reference to a stub for a global, not the global itself. +inline static bool isGlobalStubReference(unsigned char TargetFlag) { + switch (TargetFlag) { + case X86II::MO_DLLIMPORT: // dllimport stub. + case X86II::MO_GOTPCREL: // rip-relative GOT reference. + case X86II::MO_GOT: // normal GOT reference. + case X86II::MO_DARWIN_NONLAZY_PIC_BASE: // Normal $non_lazy_ptr ref. + case X86II::MO_DARWIN_NONLAZY: // Normal $non_lazy_ptr ref. + case X86II::MO_DARWIN_HIDDEN_NONLAZY_PIC_BASE: // Hidden $non_lazy_ptr ref. + return true; + default: + return false; + } +} + +/// isGlobalRelativeToPICBase - Return true if the specified global value +/// reference is relative to a 32-bit PIC base (X86ISD::GlobalBaseReg). If this +/// is true, the addressing mode has the PIC base register added in (e.g. EBX). +inline static bool isGlobalRelativeToPICBase(unsigned char TargetFlag) { + switch (TargetFlag) { + case X86II::MO_GOTOFF: // isPICStyleGOT: local global. + case X86II::MO_GOT: // isPICStyleGOT: other global. + case X86II::MO_PIC_BASE_OFFSET: // Darwin local global. + case X86II::MO_DARWIN_NONLAZY_PIC_BASE: // Darwin/32 external global. + case X86II::MO_DARWIN_HIDDEN_NONLAZY_PIC_BASE: // Darwin/32 hidden global. + case X86II::MO_TLVP: // ??? Pretty sure.. + return true; + default: + return false; + } +} + /// X86II - This namespace holds all of the target specific flags that /// instruction info tracks. /// namespace X86II { enum { //===------------------------------------------------------------------===// - // Instruction types. These are the standard/most common forms for X86 + // Instruction encodings. These are the standard/most common forms for X86 // instructions. // @@ -125,6 +300,31 @@ namespace X86II { // destinations are the same register. MRMInitReg = 32, + //// MRM_C1 - A mod/rm byte of exactly 0xC1. + MRM_C1 = 33, + MRM_C2 = 34, + MRM_C3 = 35, + MRM_C4 = 36, + MRM_C8 = 37, + MRM_C9 = 38, + MRM_E8 = 39, + MRM_F0 = 40, + MRM_F8 = 41, + MRM_F9 = 42, + MRM_D0 = 45, + MRM_D1 = 46, + + /// RawFrmImm8 - This is used for the ENTER instruction, which has two + /// immediates, the first of which is a 16-bit immediate (specified by + /// the imm encoding) and the second is a 8-bit fixed value. + RawFrmImm8 = 43, + + /// RawFrmImm16 - This is used for CALL FAR instructions, which have two + /// immediates, the first of which is a 16 or 32-bit immediate (specified by + /// the imm encoding) and the second is a 16-bit fixed value. In the AMD + /// manual, this operand is described as pntr16:32 and pntr16:16 + RawFrmImm16 = 44, + FormMask = 63, //===------------------------------------------------------------------===// @@ -147,7 +347,7 @@ namespace X86II { // set, there is no prefix byte for obtaining a multibyte opcode. // Op0Shift = 8, - Op0Mask = 0xF << Op0Shift, + Op0Mask = 0x1F << Op0Shift, // TB - TwoByte - Set if this instruction has a two byte opcode, which // starts with a 0x0F byte before the real opcode. @@ -168,8 +368,12 @@ namespace X86II { // floating point operations performed in the SSE registers. XD = 11 << Op0Shift, XS = 12 << Op0Shift, - // T8, TA - Prefix after the 0x0F prefix. + // T8, TA, A6, A7 - Prefix after the 0x0F prefix. T8 = 13 << Op0Shift, TA = 14 << Op0Shift, + A6 = 15 << Op0Shift, A7 = 16 << Op0Shift, + + // TF - Prefix before and after 0x0F + TF = 17 << Op0Shift, //===------------------------------------------------------------------===// // REX_W - REX prefixes are instruction prefixes used in 64-bit mode. @@ -177,24 +381,27 @@ namespace X86II { // etc. We only cares about REX.W and REX.R bits and only the former is // statically determined. // - REXShift = 12, + REXShift = Op0Shift + 5, REX_W = 1 << REXShift, //===------------------------------------------------------------------===// // This three-bit field describes the size of an immediate operand. Zero is // unused so that we can tell if we forgot to set a value. - ImmShift = 13, - ImmMask = 7 << ImmShift, - Imm8 = 1 << ImmShift, - Imm16 = 2 << ImmShift, - Imm32 = 3 << ImmShift, - Imm64 = 4 << ImmShift, + ImmShift = REXShift + 1, + ImmMask = 7 << ImmShift, + Imm8 = 1 << ImmShift, + Imm8PCRel = 2 << ImmShift, + Imm16 = 3 << ImmShift, + Imm16PCRel = 4 << ImmShift, + Imm32 = 5 << ImmShift, + Imm32PCRel = 6 << ImmShift, + Imm64 = 7 << ImmShift, //===------------------------------------------------------------------===// // FP Instruction Classification... Zero is non-fp instruction. // FPTypeMask - Mask for all of the FP types... - FPTypeShift = 16, + FPTypeShift = ImmShift + 3, FPTypeMask = 7 << FPTypeShift, // NotFP - The default, set for instructions that do not use FP registers. @@ -227,20 +434,158 @@ namespace X86II { SpecialFP = 7 << FPTypeShift, // Lock prefix - LOCKShift = 19, + LOCKShift = FPTypeShift + 3, LOCK = 1 << LOCKShift, // Segment override prefixes. Currently we just need ability to address // stuff in gs and fs segments. - SegOvrShift = 20, + SegOvrShift = LOCKShift + 1, SegOvrMask = 3 << SegOvrShift, FS = 1 << SegOvrShift, GS = 2 << SegOvrShift, - // Bits 22 -> 23 are unused - OpcodeShift = 24, - OpcodeMask = 0xFF << OpcodeShift + // Execution domain for SSE instructions in bits 23, 24. + // 0 in bits 23-24 means normal, non-SSE instruction. + SSEDomainShift = SegOvrShift + 2, + + OpcodeShift = SSEDomainShift + 2, + OpcodeMask = 0xFFULL << OpcodeShift, + + //===------------------------------------------------------------------===// + /// VEX - The opcode prefix used by AVX instructions + VEXShift = OpcodeShift + 8, + VEX = 1U << 0, + + /// VEX_W - Has a opcode specific functionality, but is used in the same + /// way as REX_W is for regular SSE instructions. + VEX_W = 1U << 1, + + /// VEX_4V - Used to specify an additional AVX/SSE register. Several 2 + /// address instructions in SSE are represented as 3 address ones in AVX + /// and the additional register is encoded in VEX_VVVV prefix. + VEX_4V = 1U << 2, + + /// VEX_I8IMM - Specifies that the last register used in a AVX instruction, + /// must be encoded in the i8 immediate field. This usually happens in + /// instructions with 4 operands. + VEX_I8IMM = 1U << 3, + + /// VEX_L - Stands for a bit in the VEX opcode prefix meaning the current + /// instruction uses 256-bit wide registers. This is usually auto detected + /// if a VR256 register is used, but some AVX instructions also have this + /// field marked when using a f256 memory references. + VEX_L = 1U << 4, + + /// Has3DNow0F0FOpcode - This flag indicates that the instruction uses the + /// wacky 0x0F 0x0F prefix for 3DNow! instructions. The manual documents + /// this as having a 0x0F prefix with a 0x0F opcode, and each instruction + /// storing a classifier in the imm8 field. To simplify our implementation, + /// we handle this by storeing the classifier in the opcode field and using + /// this flag to indicate that the encoder should do the wacky 3DNow! thing. + Has3DNow0F0FOpcode = 1U << 5 }; + + // getBaseOpcodeFor - This function returns the "base" X86 opcode for the + // specified machine instruction. + // + static inline unsigned char getBaseOpcodeFor(uint64_t TSFlags) { + return TSFlags >> X86II::OpcodeShift; + } + + static inline bool hasImm(uint64_t TSFlags) { + return (TSFlags & X86II::ImmMask) != 0; + } + + /// getSizeOfImm - Decode the "size of immediate" field from the TSFlags field + /// of the specified instruction. + static inline unsigned getSizeOfImm(uint64_t TSFlags) { + switch (TSFlags & X86II::ImmMask) { + default: assert(0 && "Unknown immediate size"); + case X86II::Imm8: + case X86II::Imm8PCRel: return 1; + case X86II::Imm16: + case X86II::Imm16PCRel: return 2; + case X86II::Imm32: + case X86II::Imm32PCRel: return 4; + case X86II::Imm64: return 8; + } + } + + /// isImmPCRel - Return true if the immediate of the specified instruction's + /// TSFlags indicates that it is pc relative. + static inline unsigned isImmPCRel(uint64_t TSFlags) { + switch (TSFlags & X86II::ImmMask) { + default: assert(0 && "Unknown immediate size"); + case X86II::Imm8PCRel: + case X86II::Imm16PCRel: + case X86II::Imm32PCRel: + return true; + case X86II::Imm8: + case X86II::Imm16: + case X86II::Imm32: + case X86II::Imm64: + return false; + } + } + + /// getMemoryOperandNo - The function returns the MCInst operand # for the + /// first field of the memory operand. If the instruction doesn't have a + /// memory operand, this returns -1. + /// + /// Note that this ignores tied operands. If there is a tied register which + /// is duplicated in the MCInst (e.g. "EAX = addl EAX, [mem]") it is only + /// counted as one operand. + /// + static inline int getMemoryOperandNo(uint64_t TSFlags) { + switch (TSFlags & X86II::FormMask) { + case X86II::MRMInitReg: assert(0 && "FIXME: Remove this form"); + default: assert(0 && "Unknown FormMask value in getMemoryOperandNo!"); + case X86II::Pseudo: + case X86II::RawFrm: + case X86II::AddRegFrm: + case X86II::MRMDestReg: + case X86II::MRMSrcReg: + case X86II::RawFrmImm8: + case X86II::RawFrmImm16: + return -1; + case X86II::MRMDestMem: + return 0; + case X86II::MRMSrcMem: { + bool HasVEX_4V = (TSFlags >> X86II::VEXShift) & X86II::VEX_4V; + unsigned FirstMemOp = 1; + if (HasVEX_4V) + ++FirstMemOp;// Skip the register source (which is encoded in VEX_VVVV). + + // FIXME: Maybe lea should have its own form? This is a horrible hack. + //if (Opcode == X86::LEA64r || Opcode == X86::LEA64_32r || + // Opcode == X86::LEA16r || Opcode == X86::LEA32r) + return FirstMemOp; + } + case X86II::MRM0r: case X86II::MRM1r: + case X86II::MRM2r: case X86II::MRM3r: + case X86II::MRM4r: case X86II::MRM5r: + case X86II::MRM6r: case X86II::MRM7r: + return -1; + case X86II::MRM0m: case X86II::MRM1m: + case X86II::MRM2m: case X86II::MRM3m: + case X86II::MRM4m: case X86II::MRM5m: + case X86II::MRM6m: case X86II::MRM7m: + return 0; + case X86II::MRM_C1: + case X86II::MRM_C2: + case X86II::MRM_C3: + case X86II::MRM_C4: + case X86II::MRM_C8: + case X86II::MRM_C9: + case X86II::MRM_E8: + case X86II::MRM_F0: + case X86II::MRM_F8: + case X86II::MRM_F9: + case X86II::MRM_D0: + case X86II::MRM_D1: + return -1; + } + } } inline static bool isScale(const MachineOperand &MO) { @@ -249,7 +594,7 @@ inline static bool isScale(const MachineOperand &MO) { MO.getImm() == 4 || MO.getImm() == 8); } -inline static bool isMem(const MachineInstr *MI, unsigned Op) { +inline static bool isLeaMem(const MachineInstr *MI, unsigned Op) { if (MI->getOperand(Op).isFI()) return true; return Op+4 <= MI->getNumOperands() && MI->getOperand(Op ).isReg() && isScale(MI->getOperand(Op+1)) && @@ -260,22 +605,29 @@ inline static bool isMem(const MachineInstr *MI, unsigned Op) { MI->getOperand(Op+3).isJTI()); } +inline static bool isMem(const MachineInstr *MI, unsigned Op) { + if (MI->getOperand(Op).isFI()) return true; + return Op+5 <= MI->getNumOperands() && + MI->getOperand(Op+4).isReg() && + isLeaMem(MI, Op); +} + class X86InstrInfo : public TargetInstrInfoImpl { X86TargetMachine &TM; const X86RegisterInfo RI; - + /// RegOp2MemOpTable2Addr, RegOp2MemOpTable0, RegOp2MemOpTable1, /// RegOp2MemOpTable2 - Load / store folding opcode maps. /// - DenseMap RegOp2MemOpTable2Addr; - DenseMap RegOp2MemOpTable0; - DenseMap RegOp2MemOpTable1; - DenseMap RegOp2MemOpTable2; - + DenseMap > RegOp2MemOpTable2Addr; + DenseMap > RegOp2MemOpTable0; + DenseMap > RegOp2MemOpTable1; + DenseMap > RegOp2MemOpTable2; + /// MemOp2RegOpTable - Load / store unfolding opcode map. /// - DenseMap > MemOp2RegOpTable; - + DenseMap > MemOp2RegOpTable; + public: explicit X86InstrInfo(X86TargetMachine &tm); @@ -285,20 +637,57 @@ public: /// virtual const X86RegisterInfo &getRegisterInfo() const { return RI; } - /// Return true if the instruction is a register to register move and return - /// the source and dest operands and their sub-register indices by reference. - virtual bool isMoveInstr(const MachineInstr &MI, - unsigned &SrcReg, unsigned &DstReg, - unsigned &SrcSubIdx, unsigned &DstSubIdx) const; + /// isCoalescableExtInstr - Return true if the instruction is a "coalescable" + /// extension instruction. That is, it's like a copy where it's legal for the + /// source to overlap the destination. e.g. X86::MOVSX64rr32. If this returns + /// true, then it's expected the pre-extension value is available as a subreg + /// of the result register. This also returns the sub-register index in + /// SubIdx. + virtual bool isCoalescableExtInstr(const MachineInstr &MI, + unsigned &SrcReg, unsigned &DstReg, + unsigned &SubIdx) const; unsigned isLoadFromStackSlot(const MachineInstr *MI, int &FrameIndex) const; - unsigned isStoreToStackSlot(const MachineInstr *MI, int &FrameIndex) const; + /// isLoadFromStackSlotPostFE - Check for post-frame ptr elimination + /// stack locations as well. This uses a heuristic so it isn't + /// reliable for correctness. + unsigned isLoadFromStackSlotPostFE(const MachineInstr *MI, + int &FrameIndex) const; + + /// hasLoadFromStackSlot - If the specified machine instruction has + /// a load from a stack slot, return true along with the FrameIndex + /// of the loaded stack slot and the machine mem operand containing + /// the reference. If not, return false. Unlike + /// isLoadFromStackSlot, this returns true for any instructions that + /// loads from the stack. This is a hint only and may not catch all + /// cases. + bool hasLoadFromStackSlot(const MachineInstr *MI, + const MachineMemOperand *&MMO, + int &FrameIndex) const; - bool isReallyTriviallyReMaterializable(const MachineInstr *MI) const; + unsigned isStoreToStackSlot(const MachineInstr *MI, int &FrameIndex) const; + /// isStoreToStackSlotPostFE - Check for post-frame ptr elimination + /// stack locations as well. This uses a heuristic so it isn't + /// reliable for correctness. + unsigned isStoreToStackSlotPostFE(const MachineInstr *MI, + int &FrameIndex) const; + + /// hasStoreToStackSlot - If the specified machine instruction has a + /// store to a stack slot, return true along with the FrameIndex of + /// the loaded stack slot and the machine mem operand containing the + /// reference. If not, return false. Unlike isStoreToStackSlot, + /// this returns true for any instructions that loads from the + /// stack. This is a hint only and may not catch all cases. + bool hasStoreToStackSlot(const MachineInstr *MI, + const MachineMemOperand *&MMO, + int &FrameIndex) const; + + bool isReallyTriviallyReMaterializable(const MachineInstr *MI, + AliasAnalysis *AA) const; void reMaterialize(MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, - unsigned DestReg, const MachineInstr *Orig) const; - - bool isInvariantLoad(const MachineInstr *MI) const; + unsigned DestReg, unsigned SubIdx, + const MachineInstr *Orig, + const TargetRegisterInfo &TRI) const; /// convertToThreeAddress - This method must be implemented by targets that /// set the M_CONVERTIBLE_TO_3_ADDR flag. When this flag is set, the target @@ -328,40 +717,43 @@ public: virtual unsigned RemoveBranch(MachineBasicBlock &MBB) const; virtual unsigned InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB, MachineBasicBlock *FBB, - const SmallVectorImpl &Cond) const; - virtual bool copyRegToReg(MachineBasicBlock &MBB, - MachineBasicBlock::iterator MI, - unsigned DestReg, unsigned SrcReg, - const TargetRegisterClass *DestRC, - const TargetRegisterClass *SrcRC) const; + const SmallVectorImpl &Cond, + DebugLoc DL) const; + virtual void copyPhysReg(MachineBasicBlock &MBB, + MachineBasicBlock::iterator MI, DebugLoc DL, + unsigned DestReg, unsigned SrcReg, + bool KillSrc) const; virtual void storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, unsigned SrcReg, bool isKill, int FrameIndex, - const TargetRegisterClass *RC) const; + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI) const; virtual void storeRegToAddr(MachineFunction &MF, unsigned SrcReg, bool isKill, SmallVectorImpl &Addr, const TargetRegisterClass *RC, + MachineInstr::mmo_iterator MMOBegin, + MachineInstr::mmo_iterator MMOEnd, SmallVectorImpl &NewMIs) const; virtual void loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator MI, unsigned DestReg, int FrameIndex, - const TargetRegisterClass *RC) const; + const TargetRegisterClass *RC, + const TargetRegisterInfo *TRI) const; virtual void loadRegFromAddr(MachineFunction &MF, unsigned DestReg, SmallVectorImpl &Addr, const TargetRegisterClass *RC, + MachineInstr::mmo_iterator MMOBegin, + MachineInstr::mmo_iterator MMOEnd, SmallVectorImpl &NewMIs) const; - - virtual bool spillCalleeSavedRegisters(MachineBasicBlock &MBB, - MachineBasicBlock::iterator MI, - const std::vector &CSI) const; - - virtual bool restoreCalleeSavedRegisters(MachineBasicBlock &MBB, - MachineBasicBlock::iterator MI, - const std::vector &CSI) const; - + virtual + MachineInstr *emitFrameIndexDebugValue(MachineFunction &MF, + int FrameIx, uint64_t Offset, + const MDNode *MDPtr, + DebugLoc DL) const; + /// foldMemoryOperand - If this target supports it, fold a load or store of /// the specified stack slot into the specified machine instruction for the /// specified operand(s). If this is possible, the target should perform the @@ -399,11 +791,35 @@ public: /// getOpcodeAfterMemoryUnfold - Returns the opcode of the would be new /// instruction after load / store are unfolded from an instruction of the /// specified opcode. It returns zero if the specified unfolding is not - /// possible. + /// possible. If LoadRegIndex is non-null, it is filled in with the operand + /// index of the operand which will hold the register holding the loaded + /// value. virtual unsigned getOpcodeAfterMemoryUnfold(unsigned Opc, - bool UnfoldLoad, bool UnfoldStore) const; - - virtual bool BlockHasNoFallThrough(const MachineBasicBlock &MBB) const; + bool UnfoldLoad, bool UnfoldStore, + unsigned *LoadRegIndex = 0) const; + + /// areLoadsFromSameBasePtr - This is used by the pre-regalloc scheduler + /// to determine if two loads are loading from the same base address. It + /// should only return true if the base pointers are the same and the + /// only differences between the two addresses are the offset. It also returns + /// the offsets by reference. + virtual bool areLoadsFromSameBasePtr(SDNode *Load1, SDNode *Load2, + int64_t &Offset1, int64_t &Offset2) const; + + /// shouldScheduleLoadsNear - This is a used by the pre-regalloc scheduler to + /// determine (in conjunction with areLoadsFromSameBasePtr) if two loads should + /// be scheduled togther. On some targets if two loads are loading from + /// addresses in the same cache line, it's better if they are scheduled + /// together. This function takes two integers that represent the load offsets + /// from the common base address. It returns true if it decides it's desirable + /// to schedule the two loads together. "NumLoads" is the number of loads that + /// have already been scheduled after Load1. + virtual bool shouldScheduleLoadsNear(SDNode *Load1, SDNode *Load2, + int64_t Offset1, int64_t Offset2, + unsigned NumLoads) const; + + virtual void getNoopForMachoTarget(MCInst &NopInst) const; + virtual bool ReverseBranchCondition(SmallVectorImpl &Cond) const; @@ -411,28 +827,19 @@ public: /// instruction that defines the specified register class. bool isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const; - // getBaseOpcodeFor - This function returns the "base" X86 opcode for the - // specified machine instruction. - // - unsigned char getBaseOpcodeFor(const TargetInstrDesc *TID) const { - return TID->TSFlags >> X86II::OpcodeShift; - } - unsigned char getBaseOpcodeFor(unsigned Opcode) const { - return getBaseOpcodeFor(&get(Opcode)); - } - static bool isX86_64NonExtLowByteReg(unsigned reg) { return (reg == X86::SPL || reg == X86::BPL || reg == X86::SIL || reg == X86::DIL); } - - static unsigned sizeOfImm(const TargetInstrDesc *Desc); - static bool isX86_64ExtendedReg(const MachineOperand &MO); - static unsigned determineREX(const MachineInstr &MI); - /// GetInstSize - Returns the size of the specified MachineInstr. - /// - virtual unsigned GetInstSizeInBytes(const MachineInstr *MI) const; + static bool isX86_64ExtendedReg(const MachineOperand &MO) { + if (!MO.isReg()) return false; + return isX86_64ExtendedReg(MO.getReg()); + } + + /// isX86_64ExtendedReg - Is the MachineOperand a x86-64 extended (r8 or + /// higher) register? e.g. r8, xmm8, xmm13, etc. + static bool isX86_64ExtendedReg(unsigned RegNo); /// getGlobalBaseReg - Return a virtual register initialized with the /// the global base register value. Output instructions required to @@ -440,11 +847,36 @@ public: /// unsigned getGlobalBaseReg(MachineFunction *MF) const; -private: + /// GetSSEDomain - Return the SSE execution domain of MI as the first element, + /// and a bitmask of possible arguments to SetSSEDomain ase the second. + std::pair GetSSEDomain(const MachineInstr *MI) const; + + /// SetSSEDomain - Set the SSEDomain of MI. + void SetSSEDomain(MachineInstr *MI, unsigned Domain) const; + MachineInstr* foldMemoryOperandImpl(MachineFunction &MF, MachineInstr* MI, unsigned OpNum, - const SmallVectorImpl &MOs) const; + const SmallVectorImpl &MOs, + unsigned Size, unsigned Alignment) const; + + bool isHighLatencyDef(int opc) const; + + bool hasHighOperandLatency(const InstrItineraryData *ItinData, + const MachineRegisterInfo *MRI, + const MachineInstr *DefMI, unsigned DefIdx, + const MachineInstr *UseMI, unsigned UseIdx) const; + +private: + MachineInstr * convertToThreeAddressWithLEA(unsigned MIOpc, + MachineFunction::iterator &MFI, + MachineBasicBlock::iterator &MBBI, + LiveVariables *LV) const; + + /// isFrameOperand - Return true and the FrameIndex if the specified + /// operand and follow operands form a reference to the stack frame. + bool isFrameOperand(const MachineInstr *MI, unsigned int Op, + int &FrameIndex) const; }; } // End llvm namespace