Home | History | Annotate | Download | only in PowerPC
      1 //===-- PPCISelLowering.h - PPC32 DAG Lowering Interface --------*- C++ -*-===//
      2 //
      3 //                     The LLVM Compiler Infrastructure
      4 //
      5 // This file is distributed under the University of Illinois Open Source
      6 // License. See LICENSE.TXT for details.
      7 //
      8 //===----------------------------------------------------------------------===//
      9 //
     10 // This file defines the interfaces that PPC uses to lower LLVM code into a
     11 // selection DAG.
     12 //
     13 //===----------------------------------------------------------------------===//
     14 
     15 #ifndef LLVM_TARGET_POWERPC_PPC32ISELLOWERING_H
     16 #define LLVM_TARGET_POWERPC_PPC32ISELLOWERING_H
     17 
     18 #include "PPC.h"
     19 #include "PPCInstrInfo.h"
     20 #include "PPCRegisterInfo.h"
     21 #include "llvm/CodeGen/CallingConvLower.h"
     22 #include "llvm/CodeGen/SelectionDAG.h"
     23 #include "llvm/Target/TargetLowering.h"
     24 
     25 namespace llvm {
     26   namespace PPCISD {
     27     enum NodeType {
     28       // Start the numbering where the builtin ops and target ops leave off.
     29       FIRST_NUMBER = ISD::BUILTIN_OP_END,
     30 
     31       /// FSEL - Traditional three-operand fsel node.
     32       ///
     33       FSEL,
     34 
     35       /// FCFID - The FCFID instruction, taking an f64 operand and producing
     36       /// and f64 value containing the FP representation of the integer that
     37       /// was temporarily in the f64 operand.
     38       FCFID,
     39 
     40       /// Newer FCFID[US] integer-to-floating-point conversion instructions for
     41       /// unsigned integers and single-precision outputs.
     42       FCFIDU, FCFIDS, FCFIDUS,
     43 
     44       /// FCTI[D,W]Z - The FCTIDZ and FCTIWZ instructions, taking an f32 or f64
     45       /// operand, producing an f64 value containing the integer representation
     46       /// of that FP value.
     47       FCTIDZ, FCTIWZ,
     48 
     49       /// Newer FCTI[D,W]UZ floating-point-to-integer conversion instructions for
     50       /// unsigned integers.
     51       FCTIDUZ, FCTIWUZ,
     52 
     53       /// Reciprocal estimate instructions (unary FP ops).
     54       FRE, FRSQRTE,
     55 
     56       // VMADDFP, VNMSUBFP - The VMADDFP and VNMSUBFP instructions, taking
     57       // three v4f32 operands and producing a v4f32 result.
     58       VMADDFP, VNMSUBFP,
     59 
     60       /// VPERM - The PPC VPERM Instruction.
     61       ///
     62       VPERM,
     63 
     64       /// Hi/Lo - These represent the high and low 16-bit parts of a global
     65       /// address respectively.  These nodes have two operands, the first of
     66       /// which must be a TargetGlobalAddress, and the second of which must be a
     67       /// Constant.  Selected naively, these turn into 'lis G+C' and 'li G+C',
     68       /// though these are usually folded into other nodes.
     69       Hi, Lo,
     70 
     71       TOC_ENTRY,
     72 
     73       /// The following two target-specific nodes are used for calls through
     74       /// function pointers in the 64-bit SVR4 ABI.
     75 
     76       /// Like a regular LOAD but additionally taking/producing a flag.
     77       LOAD,
     78 
     79       /// Like LOAD (taking/producing a flag), but using r2 as hard-coded
     80       /// destination.
     81       LOAD_TOC,
     82 
     83       /// OPRC, CHAIN = DYNALLOC(CHAIN, NEGSIZE, FRAME_INDEX)
     84       /// This instruction is lowered in PPCRegisterInfo::eliminateFrameIndex to
     85       /// compute an allocation on the stack.
     86       DYNALLOC,
     87 
     88       /// GlobalBaseReg - On Darwin, this node represents the result of the mflr
     89       /// at function entry, used for PIC code.
     90       GlobalBaseReg,
     91 
     92       /// These nodes represent the 32-bit PPC shifts that operate on 6-bit
     93       /// shift amounts.  These nodes are generated by the multi-precision shift
     94       /// code.
     95       SRL, SRA, SHL,
     96 
     97       /// CALL - A direct function call.
     98       /// CALL_NOP is a call with the special NOP which follows 64-bit
     99       /// SVR4 calls.
    100       CALL, CALL_NOP,
    101 
    102       /// CHAIN,FLAG = MTCTR(VAL, CHAIN[, INFLAG]) - Directly corresponds to a
    103       /// MTCTR instruction.
    104       MTCTR,
    105 
    106       /// CHAIN,FLAG = BCTRL(CHAIN, INFLAG) - Directly corresponds to a
    107       /// BCTRL instruction.
    108       BCTRL,
    109 
    110       /// Return with a flag operand, matched by 'blr'
    111       RET_FLAG,
    112 
    113       /// R32 = MFOCRF(CRREG, INFLAG) - Represents the MFOCRF instruction.
    114       /// This copies the bits corresponding to the specified CRREG into the
    115       /// resultant GPR.  Bits corresponding to other CR regs are undefined.
    116       MFOCRF,
    117 
    118       // FIXME: Remove these once the ANDI glue bug is fixed:
    119       /// i1 = ANDIo_1_[EQ|GT]_BIT(i32 or i64 x) - Represents the result of the
    120       /// eq or gt bit of CR0 after executing andi. x, 1. This is used to
    121       /// implement truncation of i32 or i64 to i1.
    122       ANDIo_1_EQ_BIT, ANDIo_1_GT_BIT,
    123 
    124       // EH_SJLJ_SETJMP - SjLj exception handling setjmp.
    125       EH_SJLJ_SETJMP,
    126 
    127       // EH_SJLJ_LONGJMP - SjLj exception handling longjmp.
    128       EH_SJLJ_LONGJMP,
    129 
    130       /// RESVEC = VCMP(LHS, RHS, OPC) - Represents one of the altivec VCMP*
    131       /// instructions.  For lack of better number, we use the opcode number
    132       /// encoding for the OPC field to identify the compare.  For example, 838
    133       /// is VCMPGTSH.
    134       VCMP,
    135 
    136       /// RESVEC, OUTFLAG = VCMPo(LHS, RHS, OPC) - Represents one of the
    137       /// altivec VCMP*o instructions.  For lack of better number, we use the
    138       /// opcode number encoding for the OPC field to identify the compare.  For
    139       /// example, 838 is VCMPGTSH.
    140       VCMPo,
    141 
    142       /// CHAIN = COND_BRANCH CHAIN, CRRC, OPC, DESTBB [, INFLAG] - This
    143       /// corresponds to the COND_BRANCH pseudo instruction.  CRRC is the
    144       /// condition register to branch on, OPC is the branch opcode to use (e.g.
    145       /// PPC::BLE), DESTBB is the destination block to branch to, and INFLAG is
    146       /// an optional input flag argument.
    147       COND_BRANCH,
    148 
    149       /// CHAIN = BDNZ CHAIN, DESTBB - These are used to create counter-based
    150       /// loops.
    151       BDNZ, BDZ,
    152 
    153       /// F8RC = FADDRTZ F8RC, F8RC - This is an FADD done with rounding
    154       /// towards zero.  Used only as part of the long double-to-int
    155       /// conversion sequence.
    156       FADDRTZ,
    157 
    158       /// F8RC = MFFS - This moves the FPSCR (not modeled) into the register.
    159       MFFS,
    160 
    161       /// LARX = This corresponds to PPC l{w|d}arx instrcution: load and
    162       /// reserve indexed. This is used to implement atomic operations.
    163       LARX,
    164 
    165       /// STCX = This corresponds to PPC stcx. instrcution: store conditional
    166       /// indexed. This is used to implement atomic operations.
    167       STCX,
    168 
    169       /// TC_RETURN - A tail call return.
    170       ///   operand #0 chain
    171       ///   operand #1 callee (register or absolute)
    172       ///   operand #2 stack adjustment
    173       ///   operand #3 optional in flag
    174       TC_RETURN,
    175 
    176       /// ch, gl = CR6[UN]SET ch, inglue - Toggle CR bit 6 for SVR4 vararg calls
    177       CR6SET,
    178       CR6UNSET,
    179 
    180       /// GPRC = address of _GLOBAL_OFFSET_TABLE_. Used by initial-exec TLS
    181       /// on PPC32.
    182       PPC32_GOT,
    183 
    184       /// G8RC = ADDIS_GOT_TPREL_HA %X2, Symbol - Used by the initial-exec
    185       /// TLS model, produces an ADDIS8 instruction that adds the GOT
    186       /// base to sym\@got\@tprel\@ha.
    187       ADDIS_GOT_TPREL_HA,
    188 
    189       /// G8RC = LD_GOT_TPREL_L Symbol, G8RReg - Used by the initial-exec
    190       /// TLS model, produces a LD instruction with base register G8RReg
    191       /// and offset sym\@got\@tprel\@l.  This completes the addition that
    192       /// finds the offset of "sym" relative to the thread pointer.
    193       LD_GOT_TPREL_L,
    194 
    195       /// G8RC = ADD_TLS G8RReg, Symbol - Used by the initial-exec TLS
    196       /// model, produces an ADD instruction that adds the contents of
    197       /// G8RReg to the thread pointer.  Symbol contains a relocation
    198       /// sym\@tls which is to be replaced by the thread pointer and
    199       /// identifies to the linker that the instruction is part of a
    200       /// TLS sequence.
    201       ADD_TLS,
    202 
    203       /// G8RC = ADDIS_TLSGD_HA %X2, Symbol - For the general-dynamic TLS
    204       /// model, produces an ADDIS8 instruction that adds the GOT base
    205       /// register to sym\@got\@tlsgd\@ha.
    206       ADDIS_TLSGD_HA,
    207 
    208       /// G8RC = ADDI_TLSGD_L G8RReg, Symbol - For the general-dynamic TLS
    209       /// model, produces an ADDI8 instruction that adds G8RReg to
    210       /// sym\@got\@tlsgd\@l.
    211       ADDI_TLSGD_L,
    212 
    213       /// G8RC = GET_TLS_ADDR %X3, Symbol - For the general-dynamic TLS
    214       /// model, produces a call to __tls_get_addr(sym\@tlsgd).
    215       GET_TLS_ADDR,
    216 
    217       /// G8RC = ADDIS_TLSLD_HA %X2, Symbol - For the local-dynamic TLS
    218       /// model, produces an ADDIS8 instruction that adds the GOT base
    219       /// register to sym\@got\@tlsld\@ha.
    220       ADDIS_TLSLD_HA,
    221 
    222       /// G8RC = ADDI_TLSLD_L G8RReg, Symbol - For the local-dynamic TLS
    223       /// model, produces an ADDI8 instruction that adds G8RReg to
    224       /// sym\@got\@tlsld\@l.
    225       ADDI_TLSLD_L,
    226 
    227       /// G8RC = GET_TLSLD_ADDR %X3, Symbol - For the local-dynamic TLS
    228       /// model, produces a call to __tls_get_addr(sym\@tlsld).
    229       GET_TLSLD_ADDR,
    230 
    231       /// G8RC = ADDIS_DTPREL_HA %X3, Symbol, Chain - For the
    232       /// local-dynamic TLS model, produces an ADDIS8 instruction
    233       /// that adds X3 to sym\@dtprel\@ha. The Chain operand is needed
    234       /// to tie this in place following a copy to %X3 from the result
    235       /// of a GET_TLSLD_ADDR.
    236       ADDIS_DTPREL_HA,
    237 
    238       /// G8RC = ADDI_DTPREL_L G8RReg, Symbol - For the local-dynamic TLS
    239       /// model, produces an ADDI8 instruction that adds G8RReg to
    240       /// sym\@got\@dtprel\@l.
    241       ADDI_DTPREL_L,
    242 
    243       /// VRRC = VADD_SPLAT Elt, EltSize - Temporary node to be expanded
    244       /// during instruction selection to optimize a BUILD_VECTOR into
    245       /// operations on splats.  This is necessary to avoid losing these
    246       /// optimizations due to constant folding.
    247       VADD_SPLAT,
    248 
    249       /// CHAIN = SC CHAIN, Imm128 - System call.  The 7-bit unsigned
    250       /// operand identifies the operating system entry point.
    251       SC,
    252 
    253       /// CHAIN = STBRX CHAIN, GPRC, Ptr, Type - This is a
    254       /// byte-swapping store instruction.  It byte-swaps the low "Type" bits of
    255       /// the GPRC input, then stores it through Ptr.  Type can be either i16 or
    256       /// i32.
    257       STBRX = ISD::FIRST_TARGET_MEMORY_OPCODE,
    258 
    259       /// GPRC, CHAIN = LBRX CHAIN, Ptr, Type - This is a
    260       /// byte-swapping load instruction.  It loads "Type" bits, byte swaps it,
    261       /// then puts it in the bottom bits of the GPRC.  TYPE can be either i16
    262       /// or i32.
    263       LBRX,
    264 
    265       /// STFIWX - The STFIWX instruction.  The first operand is an input token
    266       /// chain, then an f64 value to store, then an address to store it to.
    267       STFIWX,
    268 
    269       /// GPRC, CHAIN = LFIWAX CHAIN, Ptr - This is a floating-point
    270       /// load which sign-extends from a 32-bit integer value into the
    271       /// destination 64-bit register.
    272       LFIWAX,
    273 
    274       /// GPRC, CHAIN = LFIWZX CHAIN, Ptr - This is a floating-point
    275       /// load which zero-extends from a 32-bit integer value into the
    276       /// destination 64-bit register.
    277       LFIWZX,
    278 
    279       /// G8RC = ADDIS_TOC_HA %X2, Symbol - For medium and large code model,
    280       /// produces an ADDIS8 instruction that adds the TOC base register to
    281       /// sym\@toc\@ha.
    282       ADDIS_TOC_HA,
    283 
    284       /// G8RC = LD_TOC_L Symbol, G8RReg - For medium and large code model,
    285       /// produces a LD instruction with base register G8RReg and offset
    286       /// sym\@toc\@l. Preceded by an ADDIS_TOC_HA to form a full 32-bit offset.
    287       LD_TOC_L,
    288 
    289       /// G8RC = ADDI_TOC_L G8RReg, Symbol - For medium code model, produces
    290       /// an ADDI8 instruction that adds G8RReg to sym\@toc\@l.
    291       /// Preceded by an ADDIS_TOC_HA to form a full 32-bit offset.
    292       ADDI_TOC_L
    293     };
    294   }
    295 
    296   /// Define some predicates that are used for node matching.
    297   namespace PPC {
    298     /// isVPKUHUMShuffleMask - Return true if this is the shuffle mask for a
    299     /// VPKUHUM instruction.
    300     bool isVPKUHUMShuffleMask(ShuffleVectorSDNode *N, bool isUnary,
    301                               SelectionDAG &DAG);
    302 
    303     /// isVPKUWUMShuffleMask - Return true if this is the shuffle mask for a
    304     /// VPKUWUM instruction.
    305     bool isVPKUWUMShuffleMask(ShuffleVectorSDNode *N, bool isUnary,
    306                               SelectionDAG &DAG);
    307 
    308     /// isVMRGLShuffleMask - Return true if this is a shuffle mask suitable for
    309     /// a VRGL* instruction with the specified unit size (1,2 or 4 bytes).
    310     bool isVMRGLShuffleMask(ShuffleVectorSDNode *N, unsigned UnitSize,
    311                             bool isUnary, SelectionDAG &DAG);
    312 
    313     /// isVMRGHShuffleMask - Return true if this is a shuffle mask suitable for
    314     /// a VRGH* instruction with the specified unit size (1,2 or 4 bytes).
    315     bool isVMRGHShuffleMask(ShuffleVectorSDNode *N, unsigned UnitSize,
    316                             bool isUnary, SelectionDAG &DAG);
    317 
    318     /// isVSLDOIShuffleMask - If this is a vsldoi shuffle mask, return the shift
    319     /// amount, otherwise return -1.
    320     int isVSLDOIShuffleMask(SDNode *N, bool isUnary, SelectionDAG &DAG);
    321 
    322     /// isSplatShuffleMask - Return true if the specified VECTOR_SHUFFLE operand
    323     /// specifies a splat of a single element that is suitable for input to
    324     /// VSPLTB/VSPLTH/VSPLTW.
    325     bool isSplatShuffleMask(ShuffleVectorSDNode *N, unsigned EltSize);
    326 
    327     /// isAllNegativeZeroVector - Returns true if all elements of build_vector
    328     /// are -0.0.
    329     bool isAllNegativeZeroVector(SDNode *N);
    330 
    331     /// getVSPLTImmediate - Return the appropriate VSPLT* immediate to splat the
    332     /// specified isSplatShuffleMask VECTOR_SHUFFLE mask.
    333     unsigned getVSPLTImmediate(SDNode *N, unsigned EltSize, SelectionDAG &DAG);
    334 
    335     /// get_VSPLTI_elt - If this is a build_vector of constants which can be
    336     /// formed by using a vspltis[bhw] instruction of the specified element
    337     /// size, return the constant being splatted.  The ByteSize field indicates
    338     /// the number of bytes of each element [124] -> [bhw].
    339     SDValue get_VSPLTI_elt(SDNode *N, unsigned ByteSize, SelectionDAG &DAG);
    340   }
    341 
    342   class PPCSubtarget;
    343   class PPCTargetLowering : public TargetLowering {
    344     const PPCSubtarget &Subtarget;
    345 
    346   public:
    347     explicit PPCTargetLowering(PPCTargetMachine &TM);
    348 
    349     /// getTargetNodeName() - This method returns the name of a target specific
    350     /// DAG node.
    351     const char *getTargetNodeName(unsigned Opcode) const override;
    352 
    353     MVT getScalarShiftAmountTy(EVT LHSTy) const override { return MVT::i32; }
    354 
    355     /// getSetCCResultType - Return the ISD::SETCC ValueType
    356     EVT getSetCCResultType(LLVMContext &Context, EVT VT) const override;
    357 
    358     /// getPreIndexedAddressParts - returns true by value, base pointer and
    359     /// offset pointer and addressing mode by reference if the node's address
    360     /// can be legally represented as pre-indexed load / store address.
    361     bool getPreIndexedAddressParts(SDNode *N, SDValue &Base,
    362                                    SDValue &Offset,
    363                                    ISD::MemIndexedMode &AM,
    364                                    SelectionDAG &DAG) const override;
    365 
    366     /// SelectAddressRegReg - Given the specified addressed, check to see if it
    367     /// can be represented as an indexed [r+r] operation.  Returns false if it
    368     /// can be more efficiently represented with [r+imm].
    369     bool SelectAddressRegReg(SDValue N, SDValue &Base, SDValue &Index,
    370                              SelectionDAG &DAG) const;
    371 
    372     /// SelectAddressRegImm - Returns true if the address N can be represented
    373     /// by a base register plus a signed 16-bit displacement [r+imm], and if it
    374     /// is not better represented as reg+reg.  If Aligned is true, only accept
    375     /// displacements suitable for STD and friends, i.e. multiples of 4.
    376     bool SelectAddressRegImm(SDValue N, SDValue &Disp, SDValue &Base,
    377                              SelectionDAG &DAG, bool Aligned) const;
    378 
    379     /// SelectAddressRegRegOnly - Given the specified addressed, force it to be
    380     /// represented as an indexed [r+r] operation.
    381     bool SelectAddressRegRegOnly(SDValue N, SDValue &Base, SDValue &Index,
    382                                  SelectionDAG &DAG) const;
    383 
    384     Sched::Preference getSchedulingPreference(SDNode *N) const override;
    385 
    386     /// LowerOperation - Provide custom lowering hooks for some operations.
    387     ///
    388     SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override;
    389 
    390     /// ReplaceNodeResults - Replace the results of node with an illegal result
    391     /// type with new values built out of custom code.
    392     ///
    393     void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue>&Results,
    394                             SelectionDAG &DAG) const override;
    395 
    396     SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
    397 
    398     unsigned getRegisterByName(const char* RegName, EVT VT) const override;
    399 
    400     void computeKnownBitsForTargetNode(const SDValue Op,
    401                                        APInt &KnownZero,
    402                                        APInt &KnownOne,
    403                                        const SelectionDAG &DAG,
    404                                        unsigned Depth = 0) const override;
    405 
    406     MachineBasicBlock *
    407       EmitInstrWithCustomInserter(MachineInstr *MI,
    408                                   MachineBasicBlock *MBB) const override;
    409     MachineBasicBlock *EmitAtomicBinary(MachineInstr *MI,
    410                                         MachineBasicBlock *MBB, bool is64Bit,
    411                                         unsigned BinOpcode) const;
    412     MachineBasicBlock *EmitPartwordAtomicBinary(MachineInstr *MI,
    413                                                 MachineBasicBlock *MBB,
    414                                             bool is8bit, unsigned Opcode) const;
    415 
    416     MachineBasicBlock *emitEHSjLjSetJmp(MachineInstr *MI,
    417                                         MachineBasicBlock *MBB) const;
    418 
    419     MachineBasicBlock *emitEHSjLjLongJmp(MachineInstr *MI,
    420                                          MachineBasicBlock *MBB) const;
    421 
    422     ConstraintType
    423     getConstraintType(const std::string &Constraint) const override;
    424 
    425     /// Examine constraint string and operand type and determine a weight value.
    426     /// The operand object must already have been set up with the operand type.
    427     ConstraintWeight getSingleConstraintMatchWeight(
    428       AsmOperandInfo &info, const char *constraint) const override;
    429 
    430     std::pair<unsigned, const TargetRegisterClass*>
    431       getRegForInlineAsmConstraint(const std::string &Constraint,
    432                                    MVT VT) const override;
    433 
    434     /// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
    435     /// function arguments in the caller parameter area.  This is the actual
    436     /// alignment, not its logarithm.
    437     unsigned getByValTypeAlignment(Type *Ty) const override;
    438 
    439     /// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
    440     /// vector.  If it is invalid, don't add anything to Ops.
    441     void LowerAsmOperandForConstraint(SDValue Op,
    442                                       std::string &Constraint,
    443                                       std::vector<SDValue> &Ops,
    444                                       SelectionDAG &DAG) const override;
    445 
    446     /// isLegalAddressingMode - Return true if the addressing mode represented
    447     /// by AM is legal for this target, for a load/store of the specified type.
    448     bool isLegalAddressingMode(const AddrMode &AM, Type *Ty) const override;
    449 
    450     /// isLegalICmpImmediate - Return true if the specified immediate is legal
    451     /// icmp immediate, that is the target has icmp instructions which can
    452     /// compare a register against the immediate without having to materialize
    453     /// the immediate into a register.
    454     bool isLegalICmpImmediate(int64_t Imm) const override;
    455 
    456     /// isLegalAddImmediate - Return true if the specified immediate is legal
    457     /// add immediate, that is the target has add instructions which can
    458     /// add a register and the immediate without having to materialize
    459     /// the immediate into a register.
    460     bool isLegalAddImmediate(int64_t Imm) const override;
    461 
    462     /// isTruncateFree - Return true if it's free to truncate a value of
    463     /// type Ty1 to type Ty2. e.g. On PPC it's free to truncate a i64 value in
    464     /// register X1 to i32 by referencing its sub-register R1.
    465     bool isTruncateFree(Type *Ty1, Type *Ty2) const override;
    466     bool isTruncateFree(EVT VT1, EVT VT2) const override;
    467 
    468     /// \brief Returns true if it is beneficial to convert a load of a constant
    469     /// to just the constant itself.
    470     bool shouldConvertConstantLoadToIntImm(const APInt &Imm,
    471                                            Type *Ty) const override;
    472 
    473     bool isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const override;
    474 
    475     /// getOptimalMemOpType - Returns the target specific optimal type for load
    476     /// and store operations as a result of memset, memcpy, and memmove
    477     /// lowering. If DstAlign is zero that means it's safe to destination
    478     /// alignment can satisfy any constraint. Similarly if SrcAlign is zero it
    479     /// means there isn't a need to check it against alignment requirement,
    480     /// probably because the source does not need to be loaded. If 'IsMemset' is
    481     /// true, that means it's expanding a memset. If 'ZeroMemset' is true, that
    482     /// means it's a memset of zero. 'MemcpyStrSrc' indicates whether the memcpy
    483     /// source is constant so it does not need to be loaded.
    484     /// It returns EVT::Other if the type should be determined using generic
    485     /// target-independent logic.
    486     EVT
    487     getOptimalMemOpType(uint64_t Size, unsigned DstAlign, unsigned SrcAlign,
    488                         bool IsMemset, bool ZeroMemset, bool MemcpyStrSrc,
    489                         MachineFunction &MF) const override;
    490 
    491     /// Is unaligned memory access allowed for the given type, and is it fast
    492     /// relative to software emulation.
    493     bool allowsUnalignedMemoryAccesses(EVT VT,
    494                                        unsigned AddrSpace,
    495                                        bool *Fast = nullptr) const override;
    496 
    497     /// isFMAFasterThanFMulAndFAdd - Return true if an FMA operation is faster
    498     /// than a pair of fmul and fadd instructions. fmuladd intrinsics will be
    499     /// expanded to FMAs when this method returns true, otherwise fmuladd is
    500     /// expanded to fmul + fadd.
    501     bool isFMAFasterThanFMulAndFAdd(EVT VT) const override;
    502 
    503     // Should we expand the build vector with shuffles?
    504     bool
    505     shouldExpandBuildVectorWithShuffles(EVT VT,
    506                                         unsigned DefinedValues) const override;
    507 
    508     /// createFastISel - This method returns a target-specific FastISel object,
    509     /// or null if the target does not support "fast" instruction selection.
    510     FastISel *createFastISel(FunctionLoweringInfo &FuncInfo,
    511                              const TargetLibraryInfo *LibInfo) const override;
    512 
    513   private:
    514     SDValue getFramePointerFrameIndex(SelectionDAG & DAG) const;
    515     SDValue getReturnAddrFrameIndex(SelectionDAG & DAG) const;
    516 
    517     bool
    518     IsEligibleForTailCallOptimization(SDValue Callee,
    519                                       CallingConv::ID CalleeCC,
    520                                       bool isVarArg,
    521                                       const SmallVectorImpl<ISD::InputArg> &Ins,
    522                                       SelectionDAG& DAG) const;
    523 
    524     SDValue EmitTailCallLoadFPAndRetAddr(SelectionDAG & DAG,
    525                                          int SPDiff,
    526                                          SDValue Chain,
    527                                          SDValue &LROpOut,
    528                                          SDValue &FPOpOut,
    529                                          bool isDarwinABI,
    530                                          SDLoc dl) const;
    531 
    532     SDValue LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const;
    533     SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
    534     SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) const;
    535     SDValue LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const;
    536     SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
    537     SDValue LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const;
    538     SDValue LowerJumpTable(SDValue Op, SelectionDAG &DAG) const;
    539     SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG) const;
    540     SDValue LowerINIT_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
    541     SDValue LowerADJUST_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
    542     SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG,
    543                          const PPCSubtarget &Subtarget) const;
    544     SDValue LowerVAARG(SDValue Op, SelectionDAG &DAG,
    545                        const PPCSubtarget &Subtarget) const;
    546     SDValue LowerVACOPY(SDValue Op, SelectionDAG &DAG,
    547                         const PPCSubtarget &Subtarget) const;
    548     SDValue LowerSTACKRESTORE(SDValue Op, SelectionDAG &DAG,
    549                                 const PPCSubtarget &Subtarget) const;
    550     SDValue LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG,
    551                                       const PPCSubtarget &Subtarget) const;
    552     SDValue LowerLOAD(SDValue Op, SelectionDAG &DAG) const;
    553     SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const;
    554     SDValue LowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const;
    555     SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const;
    556     SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG, SDLoc dl) const;
    557     SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
    558     SDValue LowerFLT_ROUNDS_(SDValue Op, SelectionDAG &DAG) const;
    559     SDValue LowerSHL_PARTS(SDValue Op, SelectionDAG &DAG) const;
    560     SDValue LowerSRL_PARTS(SDValue Op, SelectionDAG &DAG) const;
    561     SDValue LowerSRA_PARTS(SDValue Op, SelectionDAG &DAG) const;
    562     SDValue LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const;
    563     SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const;
    564     SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
    565     SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) const;
    566     SDValue LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const;
    567     SDValue LowerMUL(SDValue Op, SelectionDAG &DAG) const;
    568 
    569     SDValue LowerCallResult(SDValue Chain, SDValue InFlag,
    570                             CallingConv::ID CallConv, bool isVarArg,
    571                             const SmallVectorImpl<ISD::InputArg> &Ins,
    572                             SDLoc dl, SelectionDAG &DAG,
    573                             SmallVectorImpl<SDValue> &InVals) const;
    574     SDValue FinishCall(CallingConv::ID CallConv, SDLoc dl, bool isTailCall,
    575                        bool isVarArg,
    576                        SelectionDAG &DAG,
    577                        SmallVector<std::pair<unsigned, SDValue>, 8>
    578                          &RegsToPass,
    579                        SDValue InFlag, SDValue Chain,
    580                        SDValue &Callee,
    581                        int SPDiff, unsigned NumBytes,
    582                        const SmallVectorImpl<ISD::InputArg> &Ins,
    583                        SmallVectorImpl<SDValue> &InVals) const;
    584 
    585     SDValue
    586       LowerFormalArguments(SDValue Chain,
    587                            CallingConv::ID CallConv, bool isVarArg,
    588                            const SmallVectorImpl<ISD::InputArg> &Ins,
    589                            SDLoc dl, SelectionDAG &DAG,
    590                            SmallVectorImpl<SDValue> &InVals) const override;
    591 
    592     SDValue
    593       LowerCall(TargetLowering::CallLoweringInfo &CLI,
    594                 SmallVectorImpl<SDValue> &InVals) const override;
    595 
    596     bool
    597       CanLowerReturn(CallingConv::ID CallConv, MachineFunction &MF,
    598                    bool isVarArg,
    599                    const SmallVectorImpl<ISD::OutputArg> &Outs,
    600                    LLVMContext &Context) const override;
    601 
    602     SDValue
    603       LowerReturn(SDValue Chain,
    604                   CallingConv::ID CallConv, bool isVarArg,
    605                   const SmallVectorImpl<ISD::OutputArg> &Outs,
    606                   const SmallVectorImpl<SDValue> &OutVals,
    607                   SDLoc dl, SelectionDAG &DAG) const override;
    608 
    609     SDValue
    610       extendArgForPPC64(ISD::ArgFlagsTy Flags, EVT ObjectVT, SelectionDAG &DAG,
    611                         SDValue ArgVal, SDLoc dl) const;
    612 
    613     SDValue
    614       LowerFormalArguments_Darwin(SDValue Chain,
    615                                   CallingConv::ID CallConv, bool isVarArg,
    616                                   const SmallVectorImpl<ISD::InputArg> &Ins,
    617                                   SDLoc dl, SelectionDAG &DAG,
    618                                   SmallVectorImpl<SDValue> &InVals) const;
    619     SDValue
    620       LowerFormalArguments_64SVR4(SDValue Chain,
    621                                   CallingConv::ID CallConv, bool isVarArg,
    622                                   const SmallVectorImpl<ISD::InputArg> &Ins,
    623                                   SDLoc dl, SelectionDAG &DAG,
    624                                   SmallVectorImpl<SDValue> &InVals) const;
    625     SDValue
    626       LowerFormalArguments_32SVR4(SDValue Chain,
    627                                   CallingConv::ID CallConv, bool isVarArg,
    628                                   const SmallVectorImpl<ISD::InputArg> &Ins,
    629                                   SDLoc dl, SelectionDAG &DAG,
    630                                   SmallVectorImpl<SDValue> &InVals) const;
    631 
    632     SDValue
    633       createMemcpyOutsideCallSeq(SDValue Arg, SDValue PtrOff,
    634                                  SDValue CallSeqStart, ISD::ArgFlagsTy Flags,
    635                                  SelectionDAG &DAG, SDLoc dl) const;
    636 
    637     SDValue
    638       LowerCall_Darwin(SDValue Chain, SDValue Callee,
    639                        CallingConv::ID CallConv,
    640                        bool isVarArg, bool isTailCall,
    641                        const SmallVectorImpl<ISD::OutputArg> &Outs,
    642                        const SmallVectorImpl<SDValue> &OutVals,
    643                        const SmallVectorImpl<ISD::InputArg> &Ins,
    644                        SDLoc dl, SelectionDAG &DAG,
    645                        SmallVectorImpl<SDValue> &InVals) const;
    646     SDValue
    647       LowerCall_64SVR4(SDValue Chain, SDValue Callee,
    648                        CallingConv::ID CallConv,
    649                        bool isVarArg, bool isTailCall,
    650                        const SmallVectorImpl<ISD::OutputArg> &Outs,
    651                        const SmallVectorImpl<SDValue> &OutVals,
    652                        const SmallVectorImpl<ISD::InputArg> &Ins,
    653                        SDLoc dl, SelectionDAG &DAG,
    654                        SmallVectorImpl<SDValue> &InVals) const;
    655     SDValue
    656     LowerCall_32SVR4(SDValue Chain, SDValue Callee, CallingConv::ID CallConv,
    657                      bool isVarArg, bool isTailCall,
    658                      const SmallVectorImpl<ISD::OutputArg> &Outs,
    659                      const SmallVectorImpl<SDValue> &OutVals,
    660                      const SmallVectorImpl<ISD::InputArg> &Ins,
    661                      SDLoc dl, SelectionDAG &DAG,
    662                      SmallVectorImpl<SDValue> &InVals) const;
    663 
    664     SDValue lowerEH_SJLJ_SETJMP(SDValue Op, SelectionDAG &DAG) const;
    665     SDValue lowerEH_SJLJ_LONGJMP(SDValue Op, SelectionDAG &DAG) const;
    666 
    667     SDValue DAGCombineExtBoolTrunc(SDNode *N, DAGCombinerInfo &DCI) const;
    668     SDValue DAGCombineTruncBoolExt(SDNode *N, DAGCombinerInfo &DCI) const;
    669     SDValue DAGCombineFastRecip(SDValue Op, DAGCombinerInfo &DCI) const;
    670     SDValue DAGCombineFastRecipFSQRT(SDValue Op, DAGCombinerInfo &DCI) const;
    671 
    672     CCAssignFn *useFastISelCCs(unsigned Flag) const;
    673   };
    674 
    675   namespace PPC {
    676     FastISel *createFastISel(FunctionLoweringInfo &FuncInfo,
    677                              const TargetLibraryInfo *LibInfo);
    678   }
    679 
    680   bool CC_PPC32_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
    681                                   CCValAssign::LocInfo &LocInfo,
    682                                   ISD::ArgFlagsTy &ArgFlags,
    683                                   CCState &State);
    684 
    685   bool CC_PPC32_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
    686                                          MVT &LocVT,
    687                                          CCValAssign::LocInfo &LocInfo,
    688                                          ISD::ArgFlagsTy &ArgFlags,
    689                                          CCState &State);
    690 
    691   bool CC_PPC32_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
    692                                            MVT &LocVT,
    693                                            CCValAssign::LocInfo &LocInfo,
    694                                            ISD::ArgFlagsTy &ArgFlags,
    695                                            CCState &State);
    696 }
    697 
    698 #endif   // LLVM_TARGET_POWERPC_PPC32ISELLOWERING_H
    699