1 //===-- X86ISelLowering.h - X86 DAG Lowering Interface ----------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file defines the interfaces that X86 uses to lower LLVM code into a
13 //===----------------------------------------------------------------------===//
15 #ifndef X86ISELLOWERING_H
16 #define X86ISELLOWERING_H
18 #include "X86Subtarget.h"
19 #include "X86RegisterInfo.h"
20 #include "X86MachineFunctionInfo.h"
21 #include "llvm/Target/TargetLowering.h"
22 #include "llvm/Target/TargetOptions.h"
23 #include "llvm/CodeGen/FastISel.h"
24 #include "llvm/CodeGen/SelectionDAG.h"
25 #include "llvm/CodeGen/CallingConvLower.h"
29 // X86 Specific DAG Nodes
31 // Start the numbering where the builtin ops leave off.
32 FIRST_NUMBER = ISD::BUILTIN_OP_END,
34 /// BSF - Bit scan forward.
35 /// BSR - Bit scan reverse.
39 /// SHLD, SHRD - Double shift instructions. These correspond to
40 /// X86::SHLDxx and X86::SHRDxx instructions.
44 /// FAND - Bitwise logical AND of floating point values. This corresponds
45 /// to X86::ANDPS or X86::ANDPD.
48 /// FOR - Bitwise logical OR of floating point values. This corresponds
49 /// to X86::ORPS or X86::ORPD.
52 /// FXOR - Bitwise logical XOR of floating point values. This corresponds
53 /// to X86::XORPS or X86::XORPD.
56 /// FSRL - Bitwise logical right shift of floating point values. These
57 /// corresponds to X86::PSRLDQ.
60 /// CALL - These operations represent an abstract X86 call
61 /// instruction, which includes a bunch of information. In particular the
62 /// operands of these node are:
64 /// #0 - The incoming token chain
66 /// #2 - The number of arg bytes the caller pushes on the stack.
67 /// #3 - The number of arg bytes the callee pops off the stack.
68 /// #4 - The value to pass in AL/AX/EAX (optional)
69 /// #5 - The value to pass in DL/DX/EDX (optional)
71 /// The result values of these nodes are:
73 /// #0 - The outgoing token chain
74 /// #1 - The first register result value (optional)
75 /// #2 - The second register result value (optional)
79 /// RDTSC_DAG - This operation implements the lowering for
83 /// X86 compare and logical compare instructions.
86 /// X86 bit-test instructions.
89 /// X86 SetCC. Operand 0 is condition code, and operand 1 is the EFLAGS
90 /// operand, usually produced by a CMP instruction.
93 // Same as SETCC except it's materialized with a sbb and the value is all
94 // one's or all zero's.
95 SETCC_CARRY, // R = carry_bit ? ~0 : 0
97 /// X86 FP SETCC, implemented with CMP{cc}SS/CMP{cc}SD.
98 /// Operands are two FP values to compare; result is a mask of
99 /// 0s or 1s. Generally DTRT for C/C++ with NaNs.
102 /// X86 MOVMSK{pd|ps}, extracts sign bits of two or four FP values,
103 /// result in an integer GPR. Needs masking for scalar result.
106 /// X86 conditional moves. Operand 0 and operand 1 are the two values
107 /// to select from. Operand 2 is the condition code, and operand 3 is the
108 /// flag operand produced by a CMP or TEST instruction. It also writes a
112 /// X86 conditional branches. Operand 0 is the chain operand, operand 1
113 /// is the block to branch if condition is true, operand 2 is the
114 /// condition code, and operand 3 is the flag operand produced by a CMP
115 /// or TEST instruction.
118 /// Return with a flag operand. Operand 0 is the chain operand, operand
119 /// 1 is the number of bytes of stack to pop.
122 /// REP_STOS - Repeat fill, corresponds to X86::REP_STOSx.
125 /// REP_MOVS - Repeat move, corresponds to X86::REP_MOVSx.
128 /// GlobalBaseReg - On Darwin, this node represents the result of the popl
129 /// at function entry, used for PIC code.
132 /// Wrapper - A wrapper node for TargetConstantPool,
133 /// TargetExternalSymbol, and TargetGlobalAddress.
136 /// WrapperRIP - Special wrapper used under X86-64 PIC mode for RIP
137 /// relative displacements.
140 /// MOVQ2DQ - Copies a 64-bit value from an MMX vector to the low word
141 /// of an XMM vector, with the high word zero filled.
144 /// MOVDQ2Q - Copies a 64-bit value from the low word of an XMM vector
145 /// to an MMX vector. If you think this is too close to the previous
146 /// mnemonic, so do I; blame Intel.
149 /// PEXTRB - Extract an 8-bit value from a vector and zero extend it to
150 /// i32, corresponds to X86::PEXTRB.
153 /// PEXTRW - Extract a 16-bit value from a vector and zero extend it to
154 /// i32, corresponds to X86::PEXTRW.
157 /// INSERTPS - Insert any element of a 4 x float vector into any element
158 /// of a destination 4 x floatvector.
161 /// PINSRB - Insert the lower 8-bits of a 32-bit value to a vector,
162 /// corresponds to X86::PINSRB.
165 /// PINSRW - Insert the lower 16-bits of a 32-bit value to a vector,
166 /// corresponds to X86::PINSRW.
169 /// PSHUFB - Shuffle 16 8-bit values within a vector.
172 /// ANDNP - Bitwise Logical AND NOT of Packed FP values.
175 /// PSIGN - Copy integer sign.
178 /// BLEND family of opcodes
181 /// HADD - Integer horizontal add.
184 /// HSUB - Integer horizontal sub.
187 /// FHADD - Floating point horizontal add.
190 /// FHSUB - Floating point horizontal sub.
193 /// FMAX, FMIN - Floating point max and min.
197 /// FRSQRT, FRCP - Floating point reciprocal-sqrt and reciprocal
198 /// approximation. Note that these typically require refinement
199 /// in order to obtain suitable precision.
202 // TLSADDR - Thread Local Storage.
205 // TLSCALL - Thread Local Storage. When calling to an OS provided
206 // thunk at the address from an earlier relocation.
209 // EH_RETURN - Exception Handling helpers.
212 /// TC_RETURN - Tail call return.
214 /// operand #1 callee (register or absolute)
215 /// operand #2 stack adjustment
216 /// operand #3 optional in flag
219 // VZEXT_MOVL - Vector move low and zero extend.
222 // VSEXT_MOVL - Vector move low and sign extend.
225 // VSHL, VSRL - 128-bit vector logical left / right shift
228 // VSHL, VSRL, VSRA - Vector shift elements
231 // VSHLI, VSRLI, VSRAI - Vector shift elements by immediate
234 // CMPP - Vector packed double/float comparison.
237 // PCMP* - Vector integer comparisons.
240 // VPCOM, VPCOMU - XOP Vector integer comparisons.
243 // ADD, SUB, SMUL, etc. - Arithmetic operations with FLAGS results.
244 ADD, SUB, ADC, SBB, SMUL,
245 INC, DEC, OR, XOR, AND,
247 ANDN, // ANDN - Bitwise AND NOT with FLAGS results.
249 BLSI, // BLSI - Extract lowest set isolated bit
250 BLSMSK, // BLSMSK - Get mask up to lowest set bit
251 BLSR, // BLSR - Reset lowest set bit
253 UMUL, // LOW, HI, FLAGS = umul LHS, RHS
255 // MUL_IMM - X86 specific multiply by immediate.
258 // PTEST - Vector bitwise comparisons
261 // TESTP - Vector packed fp sign bitwise comparisons
264 // Several flavors of instructions with vector shuffle behaviors.
286 // PMULUDQ - Vector multiply packed unsigned doubleword integers
289 // VASTART_SAVE_XMM_REGS - Save xmm argument registers to the stack,
290 // according to %al. An operator is needed so that this can be expanded
291 // with control flow.
292 VASTART_SAVE_XMM_REGS,
294 // WIN_ALLOCA - Windows's _chkstk call to do stack probing.
297 // SEG_ALLOCA - For allocating variable amounts of stack space when using
298 // segmented stacks. Check if the current stacklet has enough space, and
299 // falls back to heap allocation if not.
308 // ATOMADD64_DAG, ATOMSUB64_DAG, ATOMOR64_DAG, ATOMAND64_DAG,
309 // ATOMXOR64_DAG, ATOMNAND64_DAG, ATOMSWAP64_DAG -
310 // Atomic 64-bit binary operations.
311 ATOMADD64_DAG = ISD::FIRST_TARGET_MEMORY_OPCODE,
319 // LCMPXCHG_DAG, LCMPXCHG8_DAG, LCMPXCHG16_DAG - Compare and swap.
324 // VZEXT_LOAD - Load, scalar_to_vector, and zero extend.
327 // FNSTCW16m - Store FP control world into i16 memory.
330 /// FP_TO_INT*_IN_MEM - This instruction implements FP_TO_SINT with the
331 /// integer destination in memory and a FP reg source. This corresponds
332 /// to the X86::FIST*m instructions and the rounding mode change stuff. It
333 /// has two inputs (token chain and address) and two outputs (int value
334 /// and token chain).
339 /// FILD, FILD_FLAG - This instruction implements SINT_TO_FP with the
340 /// integer source in memory and FP reg result. This corresponds to the
341 /// X86::FILD*m instructions. It has three inputs (token chain, address,
342 /// and source type) and two outputs (FP value and token chain). FILD_FLAG
343 /// also produces a flag).
347 /// FLD - This instruction implements an extending load to FP stack slots.
348 /// This corresponds to the X86::FLD32m / X86::FLD64m. It takes a chain
349 /// operand, ptr to load from, and a ValueType node indicating the type
353 /// FST - This instruction implements a truncating store to FP stack
354 /// slots. This corresponds to the X86::FST32m / X86::FST64m. It takes a
355 /// chain operand, value to store, address, and a ValueType to store it
359 /// VAARG_64 - This instruction grabs the address of the next argument
360 /// from a va_list. (reads and modifies the va_list in memory)
363 // WARNING: Do not add anything in the end unless you want the node to
364 // have memop! In fact, starting from ATOMADD64_DAG all opcodes will be
365 // thought as target memory ops!
369 /// Define some predicates that are used for node matching.
371 /// isVEXTRACTF128Index - Return true if the specified
372 /// EXTRACT_SUBVECTOR operand specifies a vector extract that is
373 /// suitable for input to VEXTRACTF128.
374 bool isVEXTRACTF128Index(SDNode *N);
376 /// isVINSERTF128Index - Return true if the specified
377 /// INSERT_SUBVECTOR operand specifies a subvector insert that is
378 /// suitable for input to VINSERTF128.
379 bool isVINSERTF128Index(SDNode *N);
381 /// getExtractVEXTRACTF128Immediate - Return the appropriate
382 /// immediate to extract the specified EXTRACT_SUBVECTOR index
383 /// with VEXTRACTF128 instructions.
384 unsigned getExtractVEXTRACTF128Immediate(SDNode *N);
386 /// getInsertVINSERTF128Immediate - Return the appropriate
387 /// immediate to insert at the specified INSERT_SUBVECTOR index
388 /// with VINSERTF128 instructions.
389 unsigned getInsertVINSERTF128Immediate(SDNode *N);
391 /// isZeroNode - Returns true if Elt is a constant zero or a floating point
393 bool isZeroNode(SDValue Elt);
395 /// isOffsetSuitableForCodeModel - Returns true of the given offset can be
396 /// fit into displacement field of the instruction.
397 bool isOffsetSuitableForCodeModel(int64_t Offset, CodeModel::Model M,
398 bool hasSymbolicDisplacement = true);
401 /// isCalleePop - Determines whether the callee is required to pop its
402 /// own arguments. Callee pop is necessary to support tail calls.
403 bool isCalleePop(CallingConv::ID CallingConv,
404 bool is64Bit, bool IsVarArg, bool TailCallOpt);
407 //===--------------------------------------------------------------------===//
408 // X86TargetLowering - X86 Implementation of the TargetLowering interface
409 class X86TargetLowering : public TargetLowering {
411 explicit X86TargetLowering(X86TargetMachine &TM);
413 virtual unsigned getJumpTableEncoding() const;
415 virtual MVT getShiftAmountTy(EVT LHSTy) const { return MVT::i8; }
417 virtual const MCExpr *
418 LowerCustomJumpTableEntry(const MachineJumpTableInfo *MJTI,
419 const MachineBasicBlock *MBB, unsigned uid,
420 MCContext &Ctx) const;
422 /// getPICJumpTableRelocaBase - Returns relocation base for the given PIC
424 virtual SDValue getPICJumpTableRelocBase(SDValue Table,
425 SelectionDAG &DAG) const;
426 virtual const MCExpr *
427 getPICJumpTableRelocBaseExpr(const MachineFunction *MF,
428 unsigned JTI, MCContext &Ctx) const;
430 /// getStackPtrReg - Return the stack pointer register we are using: either
432 unsigned getStackPtrReg() const { return X86StackPtr; }
434 /// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
435 /// function arguments in the caller parameter area. For X86, aggregates
436 /// that contains are placed at 16-byte boundaries while the rest are at
437 /// 4-byte boundaries.
438 virtual unsigned getByValTypeAlignment(Type *Ty) const;
440 /// getOptimalMemOpType - Returns the target specific optimal type for load
441 /// and store operations as a result of memset, memcpy, and memmove
442 /// lowering. If DstAlign is zero that means it's safe to destination
443 /// alignment can satisfy any constraint. Similarly if SrcAlign is zero it
444 /// means there isn't a need to check it against alignment requirement,
445 /// probably because the source does not need to be loaded. If
446 /// 'IsZeroVal' is true, that means it's safe to return a
447 /// non-scalar-integer type, e.g. empty string source, constant, or loaded
448 /// from memory. 'MemcpyStrSrc' indicates whether the memcpy source is
449 /// constant so it does not need to be loaded.
450 /// It returns EVT::Other if the type should be determined using generic
451 /// target-independent logic.
453 getOptimalMemOpType(uint64_t Size, unsigned DstAlign, unsigned SrcAlign,
454 bool IsZeroVal, bool MemcpyStrSrc,
455 MachineFunction &MF) const;
457 /// allowsUnalignedMemoryAccesses - Returns true if the target allows
458 /// unaligned memory accesses. of the specified type.
459 virtual bool allowsUnalignedMemoryAccesses(EVT VT) const {
463 /// LowerOperation - Provide custom lowering hooks for some operations.
465 virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
467 /// ReplaceNodeResults - Replace the results of node with an illegal result
468 /// type with new values built out of custom code.
470 virtual void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue>&Results,
471 SelectionDAG &DAG) const;
474 virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
476 /// isTypeDesirableForOp - Return true if the target has native support for
477 /// the specified value type and it is 'desirable' to use the type for the
478 /// given node type. e.g. On x86 i16 is legal, but undesirable since i16
479 /// instruction encodings are longer and some i16 instructions are slow.
480 virtual bool isTypeDesirableForOp(unsigned Opc, EVT VT) const;
482 /// isTypeDesirable - Return true if the target has native support for the
483 /// specified value type and it is 'desirable' to use the type. e.g. On x86
484 /// i16 is legal, but undesirable since i16 instruction encodings are longer
485 /// and some i16 instructions are slow.
486 virtual bool IsDesirableToPromoteOp(SDValue Op, EVT &PVT) const;
488 virtual MachineBasicBlock *
489 EmitInstrWithCustomInserter(MachineInstr *MI,
490 MachineBasicBlock *MBB) const;
493 /// getTargetNodeName - This method returns the name of a target specific
495 virtual const char *getTargetNodeName(unsigned Opcode) const;
497 /// getSetCCResultType - Return the value type to use for ISD::SETCC.
498 virtual EVT getSetCCResultType(EVT VT) const;
500 /// computeMaskedBitsForTargetNode - Determine which of the bits specified
501 /// in Mask are known to be either zero or one and return them in the
502 /// KnownZero/KnownOne bitsets.
503 virtual void computeMaskedBitsForTargetNode(const SDValue Op,
507 const SelectionDAG &DAG,
508 unsigned Depth = 0) const;
510 // ComputeNumSignBitsForTargetNode - Determine the number of bits in the
511 // operation that are sign bits.
512 virtual unsigned ComputeNumSignBitsForTargetNode(SDValue Op,
513 unsigned Depth) const;
516 isGAPlusOffset(SDNode *N, const GlobalValue* &GA, int64_t &Offset) const;
518 SDValue getReturnAddressFrameIndex(SelectionDAG &DAG) const;
520 virtual bool ExpandInlineAsm(CallInst *CI) const;
522 ConstraintType getConstraintType(const std::string &Constraint) const;
524 /// Examine constraint string and operand type and determine a weight value.
525 /// The operand object must already have been set up with the operand type.
526 virtual ConstraintWeight getSingleConstraintMatchWeight(
527 AsmOperandInfo &info, const char *constraint) const;
529 virtual const char *LowerXConstraint(EVT ConstraintVT) const;
531 /// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
532 /// vector. If it is invalid, don't add anything to Ops. If hasMemory is
533 /// true it means one of the asm constraint of the inline asm instruction
534 /// being processed is 'm'.
535 virtual void LowerAsmOperandForConstraint(SDValue Op,
536 std::string &Constraint,
537 std::vector<SDValue> &Ops,
538 SelectionDAG &DAG) const;
540 /// getRegForInlineAsmConstraint - Given a physical register constraint
541 /// (e.g. {edx}), return the register number and the register class for the
542 /// register. This should only be used for C_Register constraints. On
543 /// error, this returns a register number of 0.
544 std::pair<unsigned, const TargetRegisterClass*>
545 getRegForInlineAsmConstraint(const std::string &Constraint,
548 /// isLegalAddressingMode - Return true if the addressing mode represented
549 /// by AM is legal for this target, for a load/store of the specified type.
550 virtual bool isLegalAddressingMode(const AddrMode &AM, Type *Ty)const;
552 /// isTruncateFree - Return true if it's free to truncate a value of
553 /// type Ty1 to type Ty2. e.g. On x86 it's free to truncate a i32 value in
554 /// register EAX to i16 by referencing its sub-register AX.
555 virtual bool isTruncateFree(Type *Ty1, Type *Ty2) const;
556 virtual bool isTruncateFree(EVT VT1, EVT VT2) const;
558 /// isZExtFree - Return true if any actual instruction that defines a
559 /// value of type Ty1 implicit zero-extends the value to Ty2 in the result
560 /// register. This does not necessarily include registers defined in
561 /// unknown ways, such as incoming arguments, or copies from unknown
562 /// virtual registers. Also, if isTruncateFree(Ty2, Ty1) is true, this
563 /// does not necessarily apply to truncate instructions. e.g. on x86-64,
564 /// all instructions that define 32-bit values implicit zero-extend the
565 /// result out to 64 bits.
566 virtual bool isZExtFree(Type *Ty1, Type *Ty2) const;
567 virtual bool isZExtFree(EVT VT1, EVT VT2) const;
569 /// isNarrowingProfitable - Return true if it's profitable to narrow
570 /// operations of type VT1 to VT2. e.g. on x86, it's profitable to narrow
571 /// from i32 to i8 but not from i32 to i16.
572 virtual bool isNarrowingProfitable(EVT VT1, EVT VT2) const;
574 /// isFPImmLegal - Returns true if the target can instruction select the
575 /// specified FP immediate natively. If false, the legalizer will
576 /// materialize the FP immediate as a load from a constant pool.
577 virtual bool isFPImmLegal(const APFloat &Imm, EVT VT) const;
579 /// isShuffleMaskLegal - Targets can use this to indicate that they only
580 /// support *some* VECTOR_SHUFFLE operations, those with specific masks.
581 /// By default, if a target supports the VECTOR_SHUFFLE node, all mask
582 /// values are assumed to be legal.
583 virtual bool isShuffleMaskLegal(const SmallVectorImpl<int> &Mask,
586 /// isVectorClearMaskLegal - Similar to isShuffleMaskLegal. This is
587 /// used by Targets can use this to indicate if there is a suitable
588 /// VECTOR_SHUFFLE that can be used to replace a VAND with a constant
590 virtual bool isVectorClearMaskLegal(const SmallVectorImpl<int> &Mask,
593 /// ShouldShrinkFPConstant - If true, then instruction selection should
594 /// seek to shrink the FP constant of the specified type to a smaller type
595 /// in order to save space and / or reduce runtime.
596 virtual bool ShouldShrinkFPConstant(EVT VT) const {
597 // Don't shrink FP constpool if SSE2 is available since cvtss2sd is more
598 // expensive than a straight movsd. On the other hand, it's important to
599 // shrink long double fp constant since fldt is very slow.
600 return !X86ScalarSSEf64 || VT == MVT::f80;
603 const X86Subtarget* getSubtarget() const {
607 /// isScalarFPTypeInSSEReg - Return true if the specified scalar FP type is
608 /// computed in an SSE register, not on the X87 floating point stack.
609 bool isScalarFPTypeInSSEReg(EVT VT) const {
610 return (VT == MVT::f64 && X86ScalarSSEf64) || // f64 is when SSE2
611 (VT == MVT::f32 && X86ScalarSSEf32); // f32 is when SSE1
614 /// createFastISel - This method returns a target specific FastISel object,
615 /// or null if the target does not support "fast" ISel.
616 virtual FastISel *createFastISel(FunctionLoweringInfo &funcInfo) const;
618 /// getStackCookieLocation - Return true if the target stores stack
619 /// protector cookies at a fixed offset in some non-standard address
620 /// space, and populates the address space and offset as
622 virtual bool getStackCookieLocation(unsigned &AddressSpace, unsigned &Offset) const;
624 SDValue BuildFILD(SDValue Op, EVT SrcVT, SDValue Chain, SDValue StackSlot,
625 SelectionDAG &DAG) const;
628 std::pair<const TargetRegisterClass*, uint8_t>
629 findRepresentativeClass(EVT VT) const;
632 /// Subtarget - Keep a pointer to the X86Subtarget around so that we can
633 /// make the right decision when generating code for different targets.
634 const X86Subtarget *Subtarget;
635 const X86RegisterInfo *RegInfo;
636 const TargetData *TD;
638 /// X86StackPtr - X86 physical register used as stack ptr.
639 unsigned X86StackPtr;
641 /// X86ScalarSSEf32, X86ScalarSSEf64 - Select between SSE or x87
642 /// floating point ops.
643 /// When SSE is available, use it for f32 operations.
644 /// When SSE2 is available, use it for f64 operations.
645 bool X86ScalarSSEf32;
646 bool X86ScalarSSEf64;
648 /// LegalFPImmediates - A list of legal fp immediates.
649 std::vector<APFloat> LegalFPImmediates;
651 /// addLegalFPImmediate - Indicate that this x86 target can instruction
652 /// select the specified FP immediate natively.
653 void addLegalFPImmediate(const APFloat& Imm) {
654 LegalFPImmediates.push_back(Imm);
657 SDValue LowerCallResult(SDValue Chain, SDValue InFlag,
658 CallingConv::ID CallConv, bool isVarArg,
659 const SmallVectorImpl<ISD::InputArg> &Ins,
660 DebugLoc dl, SelectionDAG &DAG,
661 SmallVectorImpl<SDValue> &InVals) const;
662 SDValue LowerMemArgument(SDValue Chain,
663 CallingConv::ID CallConv,
664 const SmallVectorImpl<ISD::InputArg> &ArgInfo,
665 DebugLoc dl, SelectionDAG &DAG,
666 const CCValAssign &VA, MachineFrameInfo *MFI,
668 SDValue LowerMemOpCallTo(SDValue Chain, SDValue StackPtr, SDValue Arg,
669 DebugLoc dl, SelectionDAG &DAG,
670 const CCValAssign &VA,
671 ISD::ArgFlagsTy Flags) const;
673 // Call lowering helpers.
675 /// IsEligibleForTailCallOptimization - Check whether the call is eligible
676 /// for tail call optimization. Targets which want to do tail call
677 /// optimization should implement this function.
678 bool IsEligibleForTailCallOptimization(SDValue Callee,
679 CallingConv::ID CalleeCC,
681 bool isCalleeStructRet,
682 bool isCallerStructRet,
683 const SmallVectorImpl<ISD::OutputArg> &Outs,
684 const SmallVectorImpl<SDValue> &OutVals,
685 const SmallVectorImpl<ISD::InputArg> &Ins,
686 SelectionDAG& DAG) const;
687 bool IsCalleePop(bool isVarArg, CallingConv::ID CallConv) const;
688 SDValue EmitTailCallLoadRetAddr(SelectionDAG &DAG, SDValue &OutRetAddr,
689 SDValue Chain, bool IsTailCall, bool Is64Bit,
690 int FPDiff, DebugLoc dl) const;
692 unsigned GetAlignedArgumentStackSize(unsigned StackSize,
693 SelectionDAG &DAG) const;
695 std::pair<SDValue,SDValue> FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG,
696 bool isSigned) const;
698 SDValue LowerAsSplatVectorLoad(SDValue SrcOp, EVT VT, DebugLoc dl,
699 SelectionDAG &DAG) const;
700 SDValue LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const;
701 SDValue LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) const;
702 SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const;
703 SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
704 SDValue LowerEXTRACT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) const;
705 SDValue LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
706 SDValue LowerINSERT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) const;
707 SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) const;
708 SDValue LowerEXTRACT_SUBVECTOR(SDValue Op, SelectionDAG &DAG) const;
709 SDValue LowerINSERT_SUBVECTOR(SDValue Op, SelectionDAG &DAG) const;
710 SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) const;
711 SDValue LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const;
712 SDValue LowerGlobalAddress(const GlobalValue *GV, DebugLoc dl,
713 int64_t Offset, SelectionDAG &DAG) const;
714 SDValue LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const;
715 SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
716 SDValue LowerExternalSymbol(SDValue Op, SelectionDAG &DAG) const;
717 SDValue LowerShiftParts(SDValue Op, SelectionDAG &DAG) const;
718 SDValue LowerBITCAST(SDValue op, SelectionDAG &DAG) const;
719 SDValue LowerSINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
720 SDValue LowerUINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
721 SDValue LowerUINT_TO_FP_i64(SDValue Op, SelectionDAG &DAG) const;
722 SDValue LowerUINT_TO_FP_i32(SDValue Op, SelectionDAG &DAG) const;
723 SDValue LowerFP_TO_SINT(SDValue Op, SelectionDAG &DAG) const;
724 SDValue LowerFP_TO_UINT(SDValue Op, SelectionDAG &DAG) const;
725 SDValue LowerFABS(SDValue Op, SelectionDAG &DAG) const;
726 SDValue LowerFNEG(SDValue Op, SelectionDAG &DAG) const;
727 SDValue LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const;
728 SDValue LowerFGETSIGN(SDValue Op, SelectionDAG &DAG) const;
729 SDValue LowerToBT(SDValue And, ISD::CondCode CC,
730 DebugLoc dl, SelectionDAG &DAG) const;
731 SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG) const;
732 SDValue LowerVSETCC(SDValue Op, SelectionDAG &DAG) const;
733 SDValue LowerSELECT(SDValue Op, SelectionDAG &DAG) const;
734 SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
735 SDValue LowerMEMSET(SDValue Op, SelectionDAG &DAG) const;
736 SDValue LowerJumpTable(SDValue Op, SelectionDAG &DAG) const;
737 SDValue LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const;
738 SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) const;
739 SDValue LowerVAARG(SDValue Op, SelectionDAG &DAG) const;
740 SDValue LowerVACOPY(SDValue Op, SelectionDAG &DAG) const;
741 SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
742 SDValue LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const;
743 SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
744 SDValue LowerFRAME_TO_ARGS_OFFSET(SDValue Op, SelectionDAG &DAG) const;
745 SDValue LowerEH_RETURN(SDValue Op, SelectionDAG &DAG) const;
746 SDValue LowerINIT_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
747 SDValue LowerADJUST_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
748 SDValue LowerFLT_ROUNDS_(SDValue Op, SelectionDAG &DAG) const;
749 SDValue LowerCTLZ(SDValue Op, SelectionDAG &DAG) const;
750 SDValue LowerCTLZ_ZERO_UNDEF(SDValue Op, SelectionDAG &DAG) const;
751 SDValue LowerCTTZ(SDValue Op, SelectionDAG &DAG) const;
752 SDValue LowerADD(SDValue Op, SelectionDAG &DAG) const;
753 SDValue LowerSUB(SDValue Op, SelectionDAG &DAG) const;
754 SDValue LowerMUL(SDValue Op, SelectionDAG &DAG) const;
755 SDValue LowerShift(SDValue Op, SelectionDAG &DAG) const;
756 SDValue LowerXALUO(SDValue Op, SelectionDAG &DAG) const;
758 SDValue LowerCMP_SWAP(SDValue Op, SelectionDAG &DAG) const;
759 SDValue LowerLOAD_SUB(SDValue Op, SelectionDAG &DAG) const;
760 SDValue LowerREADCYCLECOUNTER(SDValue Op, SelectionDAG &DAG) const;
761 SDValue LowerMEMBARRIER(SDValue Op, SelectionDAG &DAG) const;
762 SDValue LowerATOMIC_FENCE(SDValue Op, SelectionDAG &DAG) const;
763 SDValue LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const;
764 SDValue PerformTruncateCombine(SDNode* N, SelectionDAG &DAG, DAGCombinerInfo &DCI) const;
766 // Utility functions to help LowerVECTOR_SHUFFLE
767 SDValue LowerVECTOR_SHUFFLEv8i16(SDValue Op, SelectionDAG &DAG) const;
770 LowerFormalArguments(SDValue Chain,
771 CallingConv::ID CallConv, bool isVarArg,
772 const SmallVectorImpl<ISD::InputArg> &Ins,
773 DebugLoc dl, SelectionDAG &DAG,
774 SmallVectorImpl<SDValue> &InVals) const;
776 LowerCall(SDValue Chain, SDValue Callee,
777 CallingConv::ID CallConv, bool isVarArg, bool &isTailCall,
778 const SmallVectorImpl<ISD::OutputArg> &Outs,
779 const SmallVectorImpl<SDValue> &OutVals,
780 const SmallVectorImpl<ISD::InputArg> &Ins,
781 DebugLoc dl, SelectionDAG &DAG,
782 SmallVectorImpl<SDValue> &InVals) const;
785 LowerReturn(SDValue Chain,
786 CallingConv::ID CallConv, bool isVarArg,
787 const SmallVectorImpl<ISD::OutputArg> &Outs,
788 const SmallVectorImpl<SDValue> &OutVals,
789 DebugLoc dl, SelectionDAG &DAG) const;
791 virtual bool isUsedByReturnOnly(SDNode *N) const;
793 virtual bool mayBeEmittedAsTailCall(CallInst *CI) const;
796 getTypeForExtArgOrReturn(LLVMContext &Context, EVT VT,
797 ISD::NodeType ExtendKind) const;
800 CanLowerReturn(CallingConv::ID CallConv, MachineFunction &MF,
802 const SmallVectorImpl<ISD::OutputArg> &Outs,
803 LLVMContext &Context) const;
805 void ReplaceATOMIC_BINARY_64(SDNode *N, SmallVectorImpl<SDValue> &Results,
806 SelectionDAG &DAG, unsigned NewOp) const;
808 /// Utility function to emit string processing sse4.2 instructions
809 /// that return in xmm0.
810 /// This takes the instruction to expand, the associated machine basic
811 /// block, the number of args, and whether or not the second arg is
812 /// in memory or not.
813 MachineBasicBlock *EmitPCMP(MachineInstr *BInstr, MachineBasicBlock *BB,
814 unsigned argNum, bool inMem) const;
816 /// Utility functions to emit monitor and mwait instructions. These
817 /// need to make sure that the arguments to the intrinsic are in the
818 /// correct registers.
819 MachineBasicBlock *EmitMonitor(MachineInstr *MI,
820 MachineBasicBlock *BB) const;
821 MachineBasicBlock *EmitMwait(MachineInstr *MI, MachineBasicBlock *BB) const;
823 /// Utility function to emit atomic bitwise operations (and, or, xor).
824 /// It takes the bitwise instruction to expand, the associated machine basic
825 /// block, and the associated X86 opcodes for reg/reg and reg/imm.
826 MachineBasicBlock *EmitAtomicBitwiseWithCustomInserter(
827 MachineInstr *BInstr,
828 MachineBasicBlock *BB,
835 TargetRegisterClass *RC,
836 bool invSrc = false) const;
838 MachineBasicBlock *EmitAtomicBit6432WithCustomInserter(
839 MachineInstr *BInstr,
840 MachineBasicBlock *BB,
845 bool invSrc = false) const;
847 /// Utility function to emit atomic min and max. It takes the min/max
848 /// instruction to expand, the associated basic block, and the associated
849 /// cmov opcode for moving the min or max value.
850 MachineBasicBlock *EmitAtomicMinMaxWithCustomInserter(MachineInstr *BInstr,
851 MachineBasicBlock *BB,
852 unsigned cmovOpc) const;
854 // Utility function to emit the low-level va_arg code for X86-64.
855 MachineBasicBlock *EmitVAARG64WithCustomInserter(
857 MachineBasicBlock *MBB) const;
859 /// Utility function to emit the xmm reg save portion of va_start.
860 MachineBasicBlock *EmitVAStartSaveXMMRegsWithCustomInserter(
861 MachineInstr *BInstr,
862 MachineBasicBlock *BB) const;
864 MachineBasicBlock *EmitLoweredSelect(MachineInstr *I,
865 MachineBasicBlock *BB) const;
867 MachineBasicBlock *EmitLoweredWinAlloca(MachineInstr *MI,
868 MachineBasicBlock *BB) const;
870 MachineBasicBlock *EmitLoweredSegAlloca(MachineInstr *MI,
871 MachineBasicBlock *BB,
874 MachineBasicBlock *EmitLoweredTLSCall(MachineInstr *MI,
875 MachineBasicBlock *BB) const;
877 MachineBasicBlock *emitLoweredTLSAddr(MachineInstr *MI,
878 MachineBasicBlock *BB) const;
880 /// Emit nodes that will be selected as "test Op0,Op0", or something
881 /// equivalent, for use with the given x86 condition code.
882 SDValue EmitTest(SDValue Op0, unsigned X86CC, SelectionDAG &DAG) const;
884 /// Emit nodes that will be selected as "cmp Op0,Op1", or something
885 /// equivalent, for use with the given x86 condition code.
886 SDValue EmitCmp(SDValue Op0, SDValue Op1, unsigned X86CC,
887 SelectionDAG &DAG) const;
891 FastISel *createFastISel(FunctionLoweringInfo &funcInfo);
895 #endif // X86ISELLOWERING_H