ARMISelLowering.h revision 37ed9c199ca639565f6ce88105f9e39e898d82d0
1//===-- ARMISelLowering.h - ARM 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 ARM uses to lower LLVM code into a
11// selection DAG.
12//
13//===----------------------------------------------------------------------===//
14
15#ifndef LLVM_LIB_TARGET_ARM_ARMISELLOWERING_H
16#define LLVM_LIB_TARGET_ARM_ARMISELLOWERING_H
17
18#include "MCTargetDesc/ARMBaseInfo.h"
19#include "llvm/CodeGen/CallingConvLower.h"
20#include "llvm/CodeGen/SelectionDAG.h"
21#include "llvm/Target/TargetLowering.h"
22#include <vector>
23
24namespace llvm {
25  class ARMConstantPoolValue;
26  class ARMSubtarget;
27
28  namespace ARMISD {
29    // ARM Specific DAG Nodes
30    enum NodeType {
31      // Start the numbering where the builtin ops and target ops leave off.
32      FIRST_NUMBER = ISD::BUILTIN_OP_END,
33
34      Wrapper,      // Wrapper - A wrapper node for TargetConstantPool,
35                    // TargetExternalSymbol, and TargetGlobalAddress.
36      WrapperPIC,   // WrapperPIC - A wrapper node for TargetGlobalAddress in
37                    // PIC mode.
38      WrapperJT,    // WrapperJT - A wrapper node for TargetJumpTable
39
40      // Add pseudo op to model memcpy for struct byval.
41      COPY_STRUCT_BYVAL,
42
43      CALL,         // Function call.
44      CALL_PRED,    // Function call that's predicable.
45      CALL_NOLINK,  // Function call with branch not branch-and-link.
46      tCALL,        // Thumb function call.
47      BRCOND,       // Conditional branch.
48      BR_JT,        // Jumptable branch.
49      BR2_JT,       // Jumptable branch (2 level - jumptable entry is a jump).
50      RET_FLAG,     // Return with a flag operand.
51      INTRET_FLAG,  // Interrupt return with an LR-offset and a flag operand.
52
53      PIC_ADD,      // Add with a PC operand and a PIC label.
54
55      CMP,          // ARM compare instructions.
56      CMN,          // ARM CMN instructions.
57      CMPZ,         // ARM compare that sets only Z flag.
58      CMPFP,        // ARM VFP compare instruction, sets FPSCR.
59      CMPFPw0,      // ARM VFP compare against zero instruction, sets FPSCR.
60      FMSTAT,       // ARM fmstat instruction.
61
62      CMOV,         // ARM conditional move instructions.
63
64      BCC_i64,
65
66      RBIT,         // ARM bitreverse instruction
67
68      FTOSI,        // FP to sint within a FP register.
69      FTOUI,        // FP to uint within a FP register.
70      SITOF,        // sint to FP within a FP register.
71      UITOF,        // uint to FP within a FP register.
72
73      SRL_FLAG,     // V,Flag = srl_flag X -> srl X, 1 + save carry out.
74      SRA_FLAG,     // V,Flag = sra_flag X -> sra X, 1 + save carry out.
75      RRX,          // V = RRX X, Flag     -> srl X, 1 + shift in carry flag.
76
77      ADDC,         // Add with carry
78      ADDE,         // Add using carry
79      SUBC,         // Sub with carry
80      SUBE,         // Sub using carry
81
82      VMOVRRD,      // double to two gprs.
83      VMOVDRR,      // Two gprs to double.
84
85      EH_SJLJ_SETJMP,         // SjLj exception handling setjmp.
86      EH_SJLJ_LONGJMP,        // SjLj exception handling longjmp.
87
88      TC_RETURN,    // Tail call return pseudo.
89
90      THREAD_POINTER,
91
92      DYN_ALLOC,    // Dynamic allocation on the stack.
93
94      MEMBARRIER_MCR, // Memory barrier (MCR)
95
96      PRELOAD,      // Preload
97
98      WIN__CHKSTK,  // Windows' __chkstk call to do stack probing.
99
100      VCEQ,         // Vector compare equal.
101      VCEQZ,        // Vector compare equal to zero.
102      VCGE,         // Vector compare greater than or equal.
103      VCGEZ,        // Vector compare greater than or equal to zero.
104      VCLEZ,        // Vector compare less than or equal to zero.
105      VCGEU,        // Vector compare unsigned greater than or equal.
106      VCGT,         // Vector compare greater than.
107      VCGTZ,        // Vector compare greater than zero.
108      VCLTZ,        // Vector compare less than zero.
109      VCGTU,        // Vector compare unsigned greater than.
110      VTST,         // Vector test bits.
111
112      // Vector shift by immediate:
113      VSHL,         // ...left
114      VSHRs,        // ...right (signed)
115      VSHRu,        // ...right (unsigned)
116
117      // Vector rounding shift by immediate:
118      VRSHRs,       // ...right (signed)
119      VRSHRu,       // ...right (unsigned)
120      VRSHRN,       // ...right narrow
121
122      // Vector saturating shift by immediate:
123      VQSHLs,       // ...left (signed)
124      VQSHLu,       // ...left (unsigned)
125      VQSHLsu,      // ...left (signed to unsigned)
126      VQSHRNs,      // ...right narrow (signed)
127      VQSHRNu,      // ...right narrow (unsigned)
128      VQSHRNsu,     // ...right narrow (signed to unsigned)
129
130      // Vector saturating rounding shift by immediate:
131      VQRSHRNs,     // ...right narrow (signed)
132      VQRSHRNu,     // ...right narrow (unsigned)
133      VQRSHRNsu,    // ...right narrow (signed to unsigned)
134
135      // Vector shift and insert:
136      VSLI,         // ...left
137      VSRI,         // ...right
138
139      // Vector get lane (VMOV scalar to ARM core register)
140      // (These are used for 8- and 16-bit element types only.)
141      VGETLANEu,    // zero-extend vector extract element
142      VGETLANEs,    // sign-extend vector extract element
143
144      // Vector move immediate and move negated immediate:
145      VMOVIMM,
146      VMVNIMM,
147
148      // Vector move f32 immediate:
149      VMOVFPIMM,
150
151      // Vector duplicate:
152      VDUP,
153      VDUPLANE,
154
155      // Vector shuffles:
156      VEXT,         // extract
157      VREV64,       // reverse elements within 64-bit doublewords
158      VREV32,       // reverse elements within 32-bit words
159      VREV16,       // reverse elements within 16-bit halfwords
160      VZIP,         // zip (interleave)
161      VUZP,         // unzip (deinterleave)
162      VTRN,         // transpose
163      VTBL1,        // 1-register shuffle with mask
164      VTBL2,        // 2-register shuffle with mask
165
166      // Vector multiply long:
167      VMULLs,       // ...signed
168      VMULLu,       // ...unsigned
169
170      UMLAL,        // 64bit Unsigned Accumulate Multiply
171      SMLAL,        // 64bit Signed Accumulate Multiply
172
173      // Operands of the standard BUILD_VECTOR node are not legalized, which
174      // is fine if BUILD_VECTORs are always lowered to shuffles or other
175      // operations, but for ARM some BUILD_VECTORs are legal as-is and their
176      // operands need to be legalized.  Define an ARM-specific version of
177      // BUILD_VECTOR for this purpose.
178      BUILD_VECTOR,
179
180      // Floating-point max and min:
181      FMAX,
182      FMIN,
183      VMAXNM,
184      VMINNM,
185
186      // Bit-field insert
187      BFI,
188
189      // Vector OR with immediate
190      VORRIMM,
191      // Vector AND with NOT of immediate
192      VBICIMM,
193
194      // Vector bitwise select
195      VBSL,
196
197      // Vector load N-element structure to all lanes:
198      VLD2DUP = ISD::FIRST_TARGET_MEMORY_OPCODE,
199      VLD3DUP,
200      VLD4DUP,
201
202      // NEON loads with post-increment base updates:
203      VLD1_UPD,
204      VLD2_UPD,
205      VLD3_UPD,
206      VLD4_UPD,
207      VLD2LN_UPD,
208      VLD3LN_UPD,
209      VLD4LN_UPD,
210      VLD2DUP_UPD,
211      VLD3DUP_UPD,
212      VLD4DUP_UPD,
213
214      // NEON stores with post-increment base updates:
215      VST1_UPD,
216      VST2_UPD,
217      VST3_UPD,
218      VST4_UPD,
219      VST2LN_UPD,
220      VST3LN_UPD,
221      VST4LN_UPD
222    };
223  }
224
225  /// Define some predicates that are used for node matching.
226  namespace ARM {
227    bool isBitFieldInvertedMask(unsigned v);
228  }
229
230  //===--------------------------------------------------------------------===//
231  //  ARMTargetLowering - ARM Implementation of the TargetLowering interface
232
233  class ARMTargetLowering : public TargetLowering {
234  public:
235    explicit ARMTargetLowering(const TargetMachine &TM);
236
237    unsigned getJumpTableEncoding() const override;
238
239    SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override;
240
241    /// ReplaceNodeResults - Replace the results of node with an illegal result
242    /// type with new values built out of custom code.
243    ///
244    void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue>&Results,
245                            SelectionDAG &DAG) const override;
246
247    const char *getTargetNodeName(unsigned Opcode) const override;
248
249    bool isSelectSupported(SelectSupportKind Kind) const override {
250      // ARM does not support scalar condition selects on vectors.
251      return (Kind != ScalarCondVectorVal);
252    }
253
254    /// getSetCCResultType - Return the value type to use for ISD::SETCC.
255    EVT getSetCCResultType(LLVMContext &Context, EVT VT) const override;
256
257    MachineBasicBlock *
258      EmitInstrWithCustomInserter(MachineInstr *MI,
259                                  MachineBasicBlock *MBB) const override;
260
261    void AdjustInstrPostInstrSelection(MachineInstr *MI,
262                                       SDNode *Node) const override;
263
264    SDValue PerformCMOVCombine(SDNode *N, SelectionDAG &DAG) const;
265    SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
266
267    bool isDesirableToTransformToIntegerOp(unsigned Opc, EVT VT) const override;
268
269    /// allowsMisalignedMemoryAccesses - Returns true if the target allows
270    /// unaligned memory accesses of the specified type. Returns whether it
271    /// is "fast" by reference in the second argument.
272    bool allowsMisalignedMemoryAccesses(EVT VT, unsigned AddrSpace,
273                                        unsigned Align,
274                                        bool *Fast) const override;
275
276    EVT getOptimalMemOpType(uint64_t Size,
277                            unsigned DstAlign, unsigned SrcAlign,
278                            bool IsMemset, bool ZeroMemset,
279                            bool MemcpyStrSrc,
280                            MachineFunction &MF) const override;
281
282    using TargetLowering::isZExtFree;
283    bool isZExtFree(SDValue Val, EVT VT2) const override;
284
285    bool allowTruncateForTailCall(Type *Ty1, Type *Ty2) const override;
286
287
288    /// isLegalAddressingMode - Return true if the addressing mode represented
289    /// by AM is legal for this target, for a load/store of the specified type.
290    bool isLegalAddressingMode(const AddrMode &AM, Type *Ty) const override;
291    bool isLegalT2ScaledAddressingMode(const AddrMode &AM, EVT VT) const;
292
293    /// isLegalICmpImmediate - Return true if the specified immediate is legal
294    /// icmp immediate, that is the target has icmp instructions which can
295    /// compare a register against the immediate without having to materialize
296    /// the immediate into a register.
297    bool isLegalICmpImmediate(int64_t Imm) const override;
298
299    /// isLegalAddImmediate - Return true if the specified immediate is legal
300    /// add immediate, that is the target has add instructions which can
301    /// add a register and the immediate without having to materialize
302    /// the immediate into a register.
303    bool isLegalAddImmediate(int64_t Imm) const override;
304
305    /// getPreIndexedAddressParts - returns true by value, base pointer and
306    /// offset pointer and addressing mode by reference if the node's address
307    /// can be legally represented as pre-indexed load / store address.
308    bool getPreIndexedAddressParts(SDNode *N, SDValue &Base, SDValue &Offset,
309                                   ISD::MemIndexedMode &AM,
310                                   SelectionDAG &DAG) const override;
311
312    /// getPostIndexedAddressParts - returns true by value, base pointer and
313    /// offset pointer and addressing mode by reference if this node can be
314    /// combined with a load / store to form a post-indexed load / store.
315    bool getPostIndexedAddressParts(SDNode *N, SDNode *Op, SDValue &Base,
316                                    SDValue &Offset, ISD::MemIndexedMode &AM,
317                                    SelectionDAG &DAG) const override;
318
319    void computeKnownBitsForTargetNode(const SDValue Op, APInt &KnownZero,
320                                       APInt &KnownOne,
321                                       const SelectionDAG &DAG,
322                                       unsigned Depth) const override;
323
324
325    bool ExpandInlineAsm(CallInst *CI) const override;
326
327    ConstraintType
328      getConstraintType(const std::string &Constraint) const override;
329
330    /// Examine constraint string and operand type and determine a weight value.
331    /// The operand object must already have been set up with the operand type.
332    ConstraintWeight getSingleConstraintMatchWeight(
333      AsmOperandInfo &info, const char *constraint) const override;
334
335    std::pair<unsigned, const TargetRegisterClass*>
336      getRegForInlineAsmConstraint(const std::string &Constraint,
337                                   MVT VT) const override;
338
339    /// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
340    /// vector.  If it is invalid, don't add anything to Ops. If hasMemory is
341    /// true it means one of the asm constraint of the inline asm instruction
342    /// being processed is 'm'.
343    void LowerAsmOperandForConstraint(SDValue Op, std::string &Constraint,
344                                      std::vector<SDValue> &Ops,
345                                      SelectionDAG &DAG) const override;
346
347    const ARMSubtarget* getSubtarget() const {
348      return Subtarget;
349    }
350
351    /// getRegClassFor - Return the register class that should be used for the
352    /// specified value type.
353    const TargetRegisterClass *getRegClassFor(MVT VT) const override;
354
355    /// getMaximalGlobalOffset - Returns the maximal possible offset which can
356    /// be used for loads / stores from the global.
357    unsigned getMaximalGlobalOffset() const override;
358
359    /// Returns true if a cast between SrcAS and DestAS is a noop.
360    bool isNoopAddrSpaceCast(unsigned SrcAS, unsigned DestAS) const override {
361      // Addrspacecasts are always noops.
362      return true;
363    }
364
365    /// createFastISel - This method returns a target specific FastISel object,
366    /// or null if the target does not support "fast" ISel.
367    FastISel *createFastISel(FunctionLoweringInfo &funcInfo,
368                             const TargetLibraryInfo *libInfo) const override;
369
370    Sched::Preference getSchedulingPreference(SDNode *N) const override;
371
372    bool
373    isShuffleMaskLegal(const SmallVectorImpl<int> &M, EVT VT) const override;
374    bool isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const override;
375
376    /// isFPImmLegal - Returns true if the target can instruction select the
377    /// specified FP immediate natively. If false, the legalizer will
378    /// materialize the FP immediate as a load from a constant pool.
379    bool isFPImmLegal(const APFloat &Imm, EVT VT) const override;
380
381    bool getTgtMemIntrinsic(IntrinsicInfo &Info,
382                            const CallInst &I,
383                            unsigned Intrinsic) const override;
384
385    /// \brief Returns true if it is beneficial to convert a load of a constant
386    /// to just the constant itself.
387    bool shouldConvertConstantLoadToIntImm(const APInt &Imm,
388                                           Type *Ty) const override;
389
390    /// \brief Returns true if an argument of type Ty needs to be passed in a
391    /// contiguous block of registers in calling convention CallConv.
392    bool functionArgumentNeedsConsecutiveRegisters(
393        Type *Ty, CallingConv::ID CallConv, bool isVarArg) const override;
394
395    bool hasLoadLinkedStoreConditional() const override;
396    Instruction *makeDMB(IRBuilder<> &Builder, ARM_MB::MemBOpt Domain) const;
397    Value *emitLoadLinked(IRBuilder<> &Builder, Value *Addr,
398                          AtomicOrdering Ord) const override;
399    Value *emitStoreConditional(IRBuilder<> &Builder, Value *Val,
400                                Value *Addr, AtomicOrdering Ord) const override;
401
402    Instruction* emitLeadingFence(IRBuilder<> &Builder, AtomicOrdering Ord,
403                          bool IsStore, bool IsLoad) const override;
404    Instruction* emitTrailingFence(IRBuilder<> &Builder, AtomicOrdering Ord,
405                           bool IsStore, bool IsLoad) const override;
406
407    bool shouldExpandAtomicLoadInIR(LoadInst *LI) const override;
408    bool shouldExpandAtomicStoreInIR(StoreInst *SI) const override;
409    bool shouldExpandAtomicRMWInIR(AtomicRMWInst *AI) const override;
410
411    bool useLoadStackGuardNode() const override;
412
413    bool canCombineStoreAndExtract(Type *VectorTy, Value *Idx,
414                                   unsigned &Cost) const override;
415
416  protected:
417    std::pair<const TargetRegisterClass*, uint8_t>
418    findRepresentativeClass(MVT VT) const override;
419
420  private:
421    /// Subtarget - Keep a pointer to the ARMSubtarget around so that we can
422    /// make the right decision when generating code for different targets.
423    const ARMSubtarget *Subtarget;
424
425    const TargetRegisterInfo *RegInfo;
426
427    const InstrItineraryData *Itins;
428
429    /// ARMPCLabelIndex - Keep track of the number of ARM PC labels created.
430    ///
431    unsigned ARMPCLabelIndex;
432
433    void addTypeForNEON(MVT VT, MVT PromotedLdStVT, MVT PromotedBitwiseVT);
434    void addDRTypeForNEON(MVT VT);
435    void addQRTypeForNEON(MVT VT);
436    std::pair<SDValue, SDValue> getARMXALUOOp(SDValue Op, SelectionDAG &DAG, SDValue &ARMcc) const;
437
438    typedef SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPassVector;
439    void PassF64ArgInRegs(SDLoc dl, SelectionDAG &DAG,
440                          SDValue Chain, SDValue &Arg,
441                          RegsToPassVector &RegsToPass,
442                          CCValAssign &VA, CCValAssign &NextVA,
443                          SDValue &StackPtr,
444                          SmallVectorImpl<SDValue> &MemOpChains,
445                          ISD::ArgFlagsTy Flags) const;
446    SDValue GetF64FormalArgument(CCValAssign &VA, CCValAssign &NextVA,
447                                 SDValue &Root, SelectionDAG &DAG,
448                                 SDLoc dl) const;
449
450    CallingConv::ID getEffectiveCallingConv(CallingConv::ID CC,
451                                            bool isVarArg) const;
452    CCAssignFn *CCAssignFnForNode(CallingConv::ID CC, bool Return,
453                                  bool isVarArg) const;
454    SDValue LowerMemOpCallTo(SDValue Chain, SDValue StackPtr, SDValue Arg,
455                             SDLoc dl, SelectionDAG &DAG,
456                             const CCValAssign &VA,
457                             ISD::ArgFlagsTy Flags) const;
458    SDValue LowerEH_SJLJ_SETJMP(SDValue Op, SelectionDAG &DAG) const;
459    SDValue LowerEH_SJLJ_LONGJMP(SDValue Op, SelectionDAG &DAG) const;
460    SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG,
461                                    const ARMSubtarget *Subtarget) const;
462    SDValue LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const;
463    SDValue LowerGlobalAddressDarwin(SDValue Op, SelectionDAG &DAG) const;
464    SDValue LowerGlobalAddressELF(SDValue Op, SelectionDAG &DAG) const;
465    SDValue LowerGlobalAddressWindows(SDValue Op, SelectionDAG &DAG) const;
466    SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
467    SDValue LowerToTLSGeneralDynamicModel(GlobalAddressSDNode *GA,
468                                            SelectionDAG &DAG) const;
469    SDValue LowerToTLSExecModels(GlobalAddressSDNode *GA,
470                                 SelectionDAG &DAG,
471                                 TLSModel::Model model) const;
472    SDValue LowerGLOBAL_OFFSET_TABLE(SDValue Op, SelectionDAG &DAG) const;
473    SDValue LowerBR_JT(SDValue Op, SelectionDAG &DAG) const;
474    SDValue LowerXALUO(SDValue Op, SelectionDAG &DAG) const;
475    SDValue LowerSELECT(SDValue Op, SelectionDAG &DAG) const;
476    SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const;
477    SDValue LowerBR_CC(SDValue Op, SelectionDAG &DAG) const;
478    SDValue LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const;
479    SDValue LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const;
480    SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
481    SDValue LowerShiftRightParts(SDValue Op, SelectionDAG &DAG) const;
482    SDValue LowerShiftLeftParts(SDValue Op, SelectionDAG &DAG) const;
483    SDValue LowerFLT_ROUNDS_(SDValue Op, SelectionDAG &DAG) const;
484    SDValue LowerConstantFP(SDValue Op, SelectionDAG &DAG,
485                            const ARMSubtarget *ST) const;
486    SDValue LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
487                              const ARMSubtarget *ST) const;
488    SDValue LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const;
489    SDValue LowerDivRem(SDValue Op, SelectionDAG &DAG) const;
490    SDValue LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const;
491    SDValue LowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const;
492    SDValue LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) const;
493    SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG) const;
494    SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
495
496    unsigned getRegisterByName(const char* RegName, EVT VT) const override;
497
498    /// isFMAFasterThanFMulAndFAdd - Return true if an FMA operation is faster
499    /// than a pair of fmul and fadd instructions. fmuladd intrinsics will be
500    /// expanded to FMAs when this method returns true, otherwise fmuladd is
501    /// expanded to fmul + fadd.
502    ///
503    /// ARM supports both fused and unfused multiply-add operations; we already
504    /// lower a pair of fmul and fadd to the latter so it's not clear that there
505    /// would be a gain or that the gain would be worthwhile enough to risk
506    /// correctness bugs.
507    bool isFMAFasterThanFMulAndFAdd(EVT VT) const override { return false; }
508
509    SDValue ReconstructShuffle(SDValue Op, SelectionDAG &DAG) const;
510
511    SDValue LowerCallResult(SDValue Chain, SDValue InFlag,
512                            CallingConv::ID CallConv, bool isVarArg,
513                            const SmallVectorImpl<ISD::InputArg> &Ins,
514                            SDLoc dl, SelectionDAG &DAG,
515                            SmallVectorImpl<SDValue> &InVals,
516                            bool isThisReturn, SDValue ThisVal) const;
517
518    SDValue
519      LowerFormalArguments(SDValue Chain,
520                           CallingConv::ID CallConv, bool isVarArg,
521                           const SmallVectorImpl<ISD::InputArg> &Ins,
522                           SDLoc dl, SelectionDAG &DAG,
523                           SmallVectorImpl<SDValue> &InVals) const override;
524
525    int StoreByValRegs(CCState &CCInfo, SelectionDAG &DAG,
526                       SDLoc dl, SDValue &Chain,
527                       const Value *OrigArg,
528                       unsigned InRegsParamRecordIdx,
529                       unsigned OffsetFromOrigArg,
530                       unsigned ArgOffset,
531                       unsigned ArgSize,
532                       bool ForceMutable,
533                       unsigned ByValStoreOffset,
534                       unsigned TotalArgRegsSaveSize) const;
535
536    void VarArgStyleRegisters(CCState &CCInfo, SelectionDAG &DAG,
537                              SDLoc dl, SDValue &Chain,
538                              unsigned ArgOffset,
539                              unsigned TotalArgRegsSaveSize,
540                              bool ForceMutable = false) const;
541
542    void computeRegArea(CCState &CCInfo, MachineFunction &MF,
543                        unsigned InRegsParamRecordIdx,
544                        unsigned ArgSize,
545                        unsigned &ArgRegsSize,
546                        unsigned &ArgRegsSaveSize) const;
547
548    SDValue
549      LowerCall(TargetLowering::CallLoweringInfo &CLI,
550                SmallVectorImpl<SDValue> &InVals) const override;
551
552    /// HandleByVal - Target-specific cleanup for ByVal support.
553    void HandleByVal(CCState *, unsigned &, unsigned) const override;
554
555    /// IsEligibleForTailCallOptimization - Check whether the call is eligible
556    /// for tail call optimization. Targets which want to do tail call
557    /// optimization should implement this function.
558    bool IsEligibleForTailCallOptimization(SDValue Callee,
559                                           CallingConv::ID CalleeCC,
560                                           bool isVarArg,
561                                           bool isCalleeStructRet,
562                                           bool isCallerStructRet,
563                                    const SmallVectorImpl<ISD::OutputArg> &Outs,
564                                    const SmallVectorImpl<SDValue> &OutVals,
565                                    const SmallVectorImpl<ISD::InputArg> &Ins,
566                                           SelectionDAG& DAG) const;
567
568    bool CanLowerReturn(CallingConv::ID CallConv,
569                        MachineFunction &MF, bool isVarArg,
570                        const SmallVectorImpl<ISD::OutputArg> &Outs,
571                        LLVMContext &Context) const override;
572
573    SDValue
574      LowerReturn(SDValue Chain,
575                  CallingConv::ID CallConv, bool isVarArg,
576                  const SmallVectorImpl<ISD::OutputArg> &Outs,
577                  const SmallVectorImpl<SDValue> &OutVals,
578                  SDLoc dl, SelectionDAG &DAG) const override;
579
580    bool isUsedByReturnOnly(SDNode *N, SDValue &Chain) const override;
581
582    bool mayBeEmittedAsTailCall(CallInst *CI) const override;
583
584    SDValue getCMOV(SDLoc dl, EVT VT, SDValue FalseVal, SDValue TrueVal,
585                    SDValue ARMcc, SDValue CCR, SDValue Cmp,
586                    SelectionDAG &DAG) const;
587    SDValue getARMCmp(SDValue LHS, SDValue RHS, ISD::CondCode CC,
588                      SDValue &ARMcc, SelectionDAG &DAG, SDLoc dl) const;
589    SDValue getVFPCmp(SDValue LHS, SDValue RHS,
590                      SelectionDAG &DAG, SDLoc dl) const;
591    SDValue duplicateCmp(SDValue Cmp, SelectionDAG &DAG) const;
592
593    SDValue OptimizeVFPBrcond(SDValue Op, SelectionDAG &DAG) const;
594
595    void SetupEntryBlockForSjLj(MachineInstr *MI,
596                                MachineBasicBlock *MBB,
597                                MachineBasicBlock *DispatchBB, int FI) const;
598
599    MachineBasicBlock *EmitSjLjDispatchBlock(MachineInstr *MI,
600                                             MachineBasicBlock *MBB) const;
601
602    bool RemapAddSubWithFlags(MachineInstr *MI, MachineBasicBlock *BB) const;
603
604    MachineBasicBlock *EmitStructByval(MachineInstr *MI,
605                                       MachineBasicBlock *MBB) const;
606
607    MachineBasicBlock *EmitLowered__chkstk(MachineInstr *MI,
608                                           MachineBasicBlock *MBB) const;
609  };
610
611  enum NEONModImmType {
612    VMOVModImm,
613    VMVNModImm,
614    OtherModImm
615  };
616
617  namespace ARM {
618    FastISel *createFastISel(FunctionLoweringInfo &funcInfo,
619                             const TargetLibraryInfo *libInfo);
620  }
621}
622
623#endif  // ARMISELLOWERING_H
624