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.
302 // WIN_FTOL - Windows's _ftol2 runtime routine to do fptoui.
311 // ATOMADD64_DAG, ATOMSUB64_DAG, ATOMOR64_DAG, ATOMAND64_DAG,
312 // ATOMXOR64_DAG, ATOMNAND64_DAG, ATOMSWAP64_DAG -
313 // Atomic 64-bit binary operations.
314 ATOMADD64_DAG = ISD::FIRST_TARGET_MEMORY_OPCODE,
322 // LCMPXCHG_DAG, LCMPXCHG8_DAG, LCMPXCHG16_DAG - Compare and swap.
327 // VZEXT_LOAD - Load, scalar_to_vector, and zero extend.
330 // FNSTCW16m - Store FP control world into i16 memory.
333 /// FP_TO_INT*_IN_MEM - This instruction implements FP_TO_SINT with the
334 /// integer destination in memory and a FP reg source. This corresponds
335 /// to the X86::FIST*m instructions and the rounding mode change stuff. It
336 /// has two inputs (token chain and address) and two outputs (int value
337 /// and token chain).
342 /// FILD, FILD_FLAG - This instruction implements SINT_TO_FP with the
343 /// integer source in memory and FP reg result. This corresponds to the
344 /// X86::FILD*m instructions. It has three inputs (token chain, address,
345 /// and source type) and two outputs (FP value and token chain). FILD_FLAG
346 /// also produces a flag).
350 /// FLD - This instruction implements an extending load to FP stack slots.
351 /// This corresponds to the X86::FLD32m / X86::FLD64m. It takes a chain
352 /// operand, ptr to load from, and a ValueType node indicating the type
356 /// FST - This instruction implements a truncating store to FP stack
357 /// slots. This corresponds to the X86::FST32m / X86::FST64m. It takes a
358 /// chain operand, value to store, address, and a ValueType to store it
362 /// VAARG_64 - This instruction grabs the address of the next argument
363 /// from a va_list. (reads and modifies the va_list in memory)
366 // WARNING: Do not add anything in the end unless you want the node to
367 // have memop! In fact, starting from ATOMADD64_DAG all opcodes will be
368 // thought as target memory ops!
372 /// Define some predicates that are used for node matching.
374 /// isVEXTRACTF128Index - Return true if the specified
375 /// EXTRACT_SUBVECTOR operand specifies a vector extract that is
376 /// suitable for input to VEXTRACTF128.
377 bool isVEXTRACTF128Index(SDNode *N);
379 /// isVINSERTF128Index - Return true if the specified
380 /// INSERT_SUBVECTOR operand specifies a subvector insert that is
381 /// suitable for input to VINSERTF128.
382 bool isVINSERTF128Index(SDNode *N);
384 /// getExtractVEXTRACTF128Immediate - Return the appropriate
385 /// immediate to extract the specified EXTRACT_SUBVECTOR index
386 /// with VEXTRACTF128 instructions.
387 unsigned getExtractVEXTRACTF128Immediate(SDNode *N);
389 /// getInsertVINSERTF128Immediate - Return the appropriate
390 /// immediate to insert at the specified INSERT_SUBVECTOR index
391 /// with VINSERTF128 instructions.
392 unsigned getInsertVINSERTF128Immediate(SDNode *N);
394 /// isZeroNode - Returns true if Elt is a constant zero or a floating point
396 bool isZeroNode(SDValue Elt);
398 /// isOffsetSuitableForCodeModel - Returns true of the given offset can be
399 /// fit into displacement field of the instruction.
400 bool isOffsetSuitableForCodeModel(int64_t Offset, CodeModel::Model M,
401 bool hasSymbolicDisplacement = true);
404 /// isCalleePop - Determines whether the callee is required to pop its
405 /// own arguments. Callee pop is necessary to support tail calls.
406 bool isCalleePop(CallingConv::ID CallingConv,
407 bool is64Bit, bool IsVarArg, bool TailCallOpt);
410 //===--------------------------------------------------------------------===//
411 // X86TargetLowering - X86 Implementation of the TargetLowering interface
412 class X86TargetLowering : public TargetLowering {
414 explicit X86TargetLowering(X86TargetMachine &TM);
416 virtual unsigned getJumpTableEncoding() const;
418 virtual MVT getShiftAmountTy(EVT LHSTy) const { return MVT::i8; }
420 virtual const MCExpr *
421 LowerCustomJumpTableEntry(const MachineJumpTableInfo *MJTI,
422 const MachineBasicBlock *MBB, unsigned uid,
423 MCContext &Ctx) const;
425 /// getPICJumpTableRelocaBase - Returns relocation base for the given PIC
427 virtual SDValue getPICJumpTableRelocBase(SDValue Table,
428 SelectionDAG &DAG) const;
429 virtual const MCExpr *
430 getPICJumpTableRelocBaseExpr(const MachineFunction *MF,
431 unsigned JTI, MCContext &Ctx) const;
433 /// getStackPtrReg - Return the stack pointer register we are using: either
435 unsigned getStackPtrReg() const { return X86StackPtr; }
437 /// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
438 /// function arguments in the caller parameter area. For X86, aggregates
439 /// that contains are placed at 16-byte boundaries while the rest are at
440 /// 4-byte boundaries.
441 virtual unsigned getByValTypeAlignment(Type *Ty) const;
443 /// getOptimalMemOpType - Returns the target specific optimal type for load
444 /// and store operations as a result of memset, memcpy, and memmove
445 /// lowering. If DstAlign is zero that means it's safe to destination
446 /// alignment can satisfy any constraint. Similarly if SrcAlign is zero it
447 /// means there isn't a need to check it against alignment requirement,
448 /// probably because the source does not need to be loaded. If
449 /// 'IsZeroVal' is true, that means it's safe to return a
450 /// non-scalar-integer type, e.g. empty string source, constant, or loaded
451 /// from memory. 'MemcpyStrSrc' indicates whether the memcpy source is
452 /// constant so it does not need to be loaded.
453 /// It returns EVT::Other if the type should be determined using generic
454 /// target-independent logic.
456 getOptimalMemOpType(uint64_t Size, unsigned DstAlign, unsigned SrcAlign,
457 bool IsZeroVal, bool MemcpyStrSrc,
458 MachineFunction &MF) const;
460 /// allowsUnalignedMemoryAccesses - Returns true if the target allows
461 /// unaligned memory accesses. of the specified type.
462 virtual bool allowsUnalignedMemoryAccesses(EVT VT) const {
466 /// LowerOperation - Provide custom lowering hooks for some operations.
468 virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
470 /// ReplaceNodeResults - Replace the results of node with an illegal result
471 /// type with new values built out of custom code.
473 virtual void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue>&Results,
474 SelectionDAG &DAG) const;
477 virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
479 /// isTypeDesirableForOp - Return true if the target has native support for
480 /// the specified value type and it is 'desirable' to use the type for the
481 /// given node type. e.g. On x86 i16 is legal, but undesirable since i16
482 /// instruction encodings are longer and some i16 instructions are slow.
483 virtual bool isTypeDesirableForOp(unsigned Opc, EVT VT) const;
485 /// isTypeDesirable - Return true if the target has native support for the
486 /// specified value type and it is 'desirable' to use the type. e.g. On x86
487 /// i16 is legal, but undesirable since i16 instruction encodings are longer
488 /// and some i16 instructions are slow.
489 virtual bool IsDesirableToPromoteOp(SDValue Op, EVT &PVT) const;
491 virtual MachineBasicBlock *
492 EmitInstrWithCustomInserter(MachineInstr *MI,
493 MachineBasicBlock *MBB) const;
496 /// getTargetNodeName - This method returns the name of a target specific
498 virtual const char *getTargetNodeName(unsigned Opcode) const;
500 /// getSetCCResultType - Return the value type to use for ISD::SETCC.
501 virtual EVT getSetCCResultType(EVT VT) const;
503 /// computeMaskedBitsForTargetNode - Determine which of the bits specified
504 /// in Mask are known to be either zero or one and return them in the
505 /// KnownZero/KnownOne bitsets.
506 virtual void computeMaskedBitsForTargetNode(const SDValue Op,
509 const SelectionDAG &DAG,
510 unsigned Depth = 0) const;
512 // ComputeNumSignBitsForTargetNode - Determine the number of bits in the
513 // operation that are sign bits.
514 virtual unsigned ComputeNumSignBitsForTargetNode(SDValue Op,
515 unsigned Depth) const;
518 isGAPlusOffset(SDNode *N, const GlobalValue* &GA, int64_t &Offset) const;
520 SDValue getReturnAddressFrameIndex(SelectionDAG &DAG) const;
522 virtual bool ExpandInlineAsm(CallInst *CI) const;
524 ConstraintType getConstraintType(const std::string &Constraint) const;
526 /// Examine constraint string and operand type and determine a weight value.
527 /// The operand object must already have been set up with the operand type.
528 virtual ConstraintWeight getSingleConstraintMatchWeight(
529 AsmOperandInfo &info, const char *constraint) const;
531 virtual const char *LowerXConstraint(EVT ConstraintVT) const;
533 /// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
534 /// vector. If it is invalid, don't add anything to Ops. If hasMemory is
535 /// true it means one of the asm constraint of the inline asm instruction
536 /// being processed is 'm'.
537 virtual void LowerAsmOperandForConstraint(SDValue Op,
538 std::string &Constraint,
539 std::vector<SDValue> &Ops,
540 SelectionDAG &DAG) const;
542 /// getRegForInlineAsmConstraint - Given a physical register constraint
543 /// (e.g. {edx}), return the register number and the register class for the
544 /// register. This should only be used for C_Register constraints. On
545 /// error, this returns a register number of 0.
546 std::pair<unsigned, const TargetRegisterClass*>
547 getRegForInlineAsmConstraint(const std::string &Constraint,
550 /// isLegalAddressingMode - Return true if the addressing mode represented
551 /// by AM is legal for this target, for a load/store of the specified type.
552 virtual bool isLegalAddressingMode(const AddrMode &AM, Type *Ty)const;
554 /// isTruncateFree - Return true if it's free to truncate a value of
555 /// type Ty1 to type Ty2. e.g. On x86 it's free to truncate a i32 value in
556 /// register EAX to i16 by referencing its sub-register AX.
557 virtual bool isTruncateFree(Type *Ty1, Type *Ty2) const;
558 virtual bool isTruncateFree(EVT VT1, EVT VT2) const;
560 /// isZExtFree - Return true if any actual instruction that defines a
561 /// value of type Ty1 implicit zero-extends the value to Ty2 in the result
562 /// register. This does not necessarily include registers defined in
563 /// unknown ways, such as incoming arguments, or copies from unknown
564 /// virtual registers. Also, if isTruncateFree(Ty2, Ty1) is true, this
565 /// does not necessarily apply to truncate instructions. e.g. on x86-64,
566 /// all instructions that define 32-bit values implicit zero-extend the
567 /// result out to 64 bits.
568 virtual bool isZExtFree(Type *Ty1, Type *Ty2) const;
569 virtual bool isZExtFree(EVT VT1, EVT VT2) const;
571 /// isNarrowingProfitable - Return true if it's profitable to narrow
572 /// operations of type VT1 to VT2. e.g. on x86, it's profitable to narrow
573 /// from i32 to i8 but not from i32 to i16.
574 virtual bool isNarrowingProfitable(EVT VT1, EVT VT2) const;
576 /// isFPImmLegal - Returns true if the target can instruction select the
577 /// specified FP immediate natively. If false, the legalizer will
578 /// materialize the FP immediate as a load from a constant pool.
579 virtual bool isFPImmLegal(const APFloat &Imm, EVT VT) const;
581 /// isShuffleMaskLegal - Targets can use this to indicate that they only
582 /// support *some* VECTOR_SHUFFLE operations, those with specific masks.
583 /// By default, if a target supports the VECTOR_SHUFFLE node, all mask
584 /// values are assumed to be legal.
585 virtual bool isShuffleMaskLegal(const SmallVectorImpl<int> &Mask,
588 /// isVectorClearMaskLegal - Similar to isShuffleMaskLegal. This is
589 /// used by Targets can use this to indicate if there is a suitable
590 /// VECTOR_SHUFFLE that can be used to replace a VAND with a constant
592 virtual bool isVectorClearMaskLegal(const SmallVectorImpl<int> &Mask,
595 /// ShouldShrinkFPConstant - If true, then instruction selection should
596 /// seek to shrink the FP constant of the specified type to a smaller type
597 /// in order to save space and / or reduce runtime.
598 virtual bool ShouldShrinkFPConstant(EVT VT) const {
599 // Don't shrink FP constpool if SSE2 is available since cvtss2sd is more
600 // expensive than a straight movsd. On the other hand, it's important to
601 // shrink long double fp constant since fldt is very slow.
602 return !X86ScalarSSEf64 || VT == MVT::f80;
605 const X86Subtarget* getSubtarget() const {
609 /// isScalarFPTypeInSSEReg - Return true if the specified scalar FP type is
610 /// computed in an SSE register, not on the X87 floating point stack.
611 bool isScalarFPTypeInSSEReg(EVT VT) const {
612 return (VT == MVT::f64 && X86ScalarSSEf64) || // f64 is when SSE2
613 (VT == MVT::f32 && X86ScalarSSEf32); // f32 is when SSE1
616 /// isTargetFTOL - Return true if the target uses the MSVC _ftol2 routine
618 bool isTargetFTOL() const {
619 return Subtarget->isTargetWindows() && !Subtarget->is64Bit();
622 /// isIntegerTypeFTOL - Return true if the MSVC _ftol2 routine should be
623 /// used for fptoui to the given type.
624 bool isIntegerTypeFTOL(EVT VT) const {
625 return isTargetFTOL() && VT == MVT::i64;
628 /// createFastISel - This method returns a target specific FastISel object,
629 /// or null if the target does not support "fast" ISel.
630 virtual FastISel *createFastISel(FunctionLoweringInfo &funcInfo) const;
632 /// getStackCookieLocation - Return true if the target stores stack
633 /// protector cookies at a fixed offset in some non-standard address
634 /// space, and populates the address space and offset as
636 virtual bool getStackCookieLocation(unsigned &AddressSpace, unsigned &Offset) const;
638 SDValue BuildFILD(SDValue Op, EVT SrcVT, SDValue Chain, SDValue StackSlot,
639 SelectionDAG &DAG) const;
642 std::pair<const TargetRegisterClass*, uint8_t>
643 findRepresentativeClass(EVT VT) const;
646 /// Subtarget - Keep a pointer to the X86Subtarget around so that we can
647 /// make the right decision when generating code for different targets.
648 const X86Subtarget *Subtarget;
649 const X86RegisterInfo *RegInfo;
650 const TargetData *TD;
652 /// X86StackPtr - X86 physical register used as stack ptr.
653 unsigned X86StackPtr;
655 /// X86ScalarSSEf32, X86ScalarSSEf64 - Select between SSE or x87
656 /// floating point ops.
657 /// When SSE is available, use it for f32 operations.
658 /// When SSE2 is available, use it for f64 operations.
659 bool X86ScalarSSEf32;
660 bool X86ScalarSSEf64;
662 /// LegalFPImmediates - A list of legal fp immediates.
663 std::vector<APFloat> LegalFPImmediates;
665 /// addLegalFPImmediate - Indicate that this x86 target can instruction
666 /// select the specified FP immediate natively.
667 void addLegalFPImmediate(const APFloat& Imm) {
668 LegalFPImmediates.push_back(Imm);
671 SDValue LowerCallResult(SDValue Chain, SDValue InFlag,
672 CallingConv::ID CallConv, bool isVarArg,
673 const SmallVectorImpl<ISD::InputArg> &Ins,
674 DebugLoc dl, SelectionDAG &DAG,
675 SmallVectorImpl<SDValue> &InVals) const;
676 SDValue LowerMemArgument(SDValue Chain,
677 CallingConv::ID CallConv,
678 const SmallVectorImpl<ISD::InputArg> &ArgInfo,
679 DebugLoc dl, SelectionDAG &DAG,
680 const CCValAssign &VA, MachineFrameInfo *MFI,
682 SDValue LowerMemOpCallTo(SDValue Chain, SDValue StackPtr, SDValue Arg,
683 DebugLoc dl, SelectionDAG &DAG,
684 const CCValAssign &VA,
685 ISD::ArgFlagsTy Flags) const;
687 // Call lowering helpers.
689 /// IsEligibleForTailCallOptimization - Check whether the call is eligible
690 /// for tail call optimization. Targets which want to do tail call
691 /// optimization should implement this function.
692 bool IsEligibleForTailCallOptimization(SDValue Callee,
693 CallingConv::ID CalleeCC,
695 bool isCalleeStructRet,
696 bool isCallerStructRet,
697 const SmallVectorImpl<ISD::OutputArg> &Outs,
698 const SmallVectorImpl<SDValue> &OutVals,
699 const SmallVectorImpl<ISD::InputArg> &Ins,
700 SelectionDAG& DAG) const;
701 bool IsCalleePop(bool isVarArg, CallingConv::ID CallConv) const;
702 SDValue EmitTailCallLoadRetAddr(SelectionDAG &DAG, SDValue &OutRetAddr,
703 SDValue Chain, bool IsTailCall, bool Is64Bit,
704 int FPDiff, DebugLoc dl) const;
706 unsigned GetAlignedArgumentStackSize(unsigned StackSize,
707 SelectionDAG &DAG) const;
709 std::pair<SDValue,SDValue> FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG,
711 bool isReplace) const;
713 SDValue LowerAsSplatVectorLoad(SDValue SrcOp, EVT VT, DebugLoc dl,
714 SelectionDAG &DAG) const;
715 SDValue LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const;
716 SDValue LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) const;
717 SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const;
718 SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
719 SDValue LowerEXTRACT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) const;
720 SDValue LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
721 SDValue LowerINSERT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) const;
722 SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) const;
723 SDValue LowerEXTRACT_SUBVECTOR(SDValue Op, SelectionDAG &DAG) const;
724 SDValue LowerINSERT_SUBVECTOR(SDValue Op, SelectionDAG &DAG) const;
725 SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) const;
726 SDValue LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const;
727 SDValue LowerGlobalAddress(const GlobalValue *GV, DebugLoc dl,
728 int64_t Offset, SelectionDAG &DAG) const;
729 SDValue LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const;
730 SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
731 SDValue LowerExternalSymbol(SDValue Op, SelectionDAG &DAG) const;
732 SDValue LowerShiftParts(SDValue Op, SelectionDAG &DAG) const;
733 SDValue LowerBITCAST(SDValue op, SelectionDAG &DAG) const;
734 SDValue LowerSINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
735 SDValue LowerUINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
736 SDValue LowerUINT_TO_FP_i64(SDValue Op, SelectionDAG &DAG) const;
737 SDValue LowerUINT_TO_FP_i32(SDValue Op, SelectionDAG &DAG) const;
738 SDValue LowerFP_TO_SINT(SDValue Op, SelectionDAG &DAG) const;
739 SDValue LowerFP_TO_UINT(SDValue Op, SelectionDAG &DAG) const;
740 SDValue LowerFABS(SDValue Op, SelectionDAG &DAG) const;
741 SDValue LowerFNEG(SDValue Op, SelectionDAG &DAG) const;
742 SDValue LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const;
743 SDValue LowerFGETSIGN(SDValue Op, SelectionDAG &DAG) const;
744 SDValue LowerToBT(SDValue And, ISD::CondCode CC,
745 DebugLoc dl, SelectionDAG &DAG) const;
746 SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG) const;
747 SDValue LowerVSETCC(SDValue Op, SelectionDAG &DAG) const;
748 SDValue LowerSELECT(SDValue Op, SelectionDAG &DAG) const;
749 SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
750 SDValue LowerMEMSET(SDValue Op, SelectionDAG &DAG) const;
751 SDValue LowerJumpTable(SDValue Op, SelectionDAG &DAG) const;
752 SDValue LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const;
753 SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) const;
754 SDValue LowerVAARG(SDValue Op, SelectionDAG &DAG) const;
755 SDValue LowerVACOPY(SDValue Op, SelectionDAG &DAG) const;
756 SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
757 SDValue LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const;
758 SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
759 SDValue LowerFRAME_TO_ARGS_OFFSET(SDValue Op, SelectionDAG &DAG) const;
760 SDValue LowerEH_RETURN(SDValue Op, SelectionDAG &DAG) const;
761 SDValue LowerINIT_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
762 SDValue LowerADJUST_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
763 SDValue LowerFLT_ROUNDS_(SDValue Op, SelectionDAG &DAG) const;
764 SDValue LowerCTLZ(SDValue Op, SelectionDAG &DAG) const;
765 SDValue LowerCTLZ_ZERO_UNDEF(SDValue Op, SelectionDAG &DAG) const;
766 SDValue LowerCTTZ(SDValue Op, SelectionDAG &DAG) const;
767 SDValue LowerADD(SDValue Op, SelectionDAG &DAG) const;
768 SDValue LowerSUB(SDValue Op, SelectionDAG &DAG) const;
769 SDValue LowerMUL(SDValue Op, SelectionDAG &DAG) const;
770 SDValue LowerShift(SDValue Op, SelectionDAG &DAG) const;
771 SDValue LowerXALUO(SDValue Op, SelectionDAG &DAG) const;
773 SDValue LowerCMP_SWAP(SDValue Op, SelectionDAG &DAG) const;
774 SDValue LowerLOAD_SUB(SDValue Op, SelectionDAG &DAG) const;
775 SDValue LowerREADCYCLECOUNTER(SDValue Op, SelectionDAG &DAG) const;
776 SDValue LowerMEMBARRIER(SDValue Op, SelectionDAG &DAG) const;
777 SDValue LowerATOMIC_FENCE(SDValue Op, SelectionDAG &DAG) const;
778 SDValue LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const;
779 SDValue PerformTruncateCombine(SDNode* N, SelectionDAG &DAG, DAGCombinerInfo &DCI) const;
781 // Utility functions to help LowerVECTOR_SHUFFLE
782 SDValue LowerVECTOR_SHUFFLEv8i16(SDValue Op, SelectionDAG &DAG) const;
783 SDValue LowerVectorBroadcast(SDValue &Op, SelectionDAG &DAG) const;
784 SDValue NormalizeVectorShuffle(SDValue Op, SelectionDAG &DAG) const;
787 LowerFormalArguments(SDValue Chain,
788 CallingConv::ID CallConv, bool isVarArg,
789 const SmallVectorImpl<ISD::InputArg> &Ins,
790 DebugLoc dl, SelectionDAG &DAG,
791 SmallVectorImpl<SDValue> &InVals) const;
793 LowerCall(SDValue Chain, SDValue Callee, CallingConv::ID CallConv,
794 bool isVarArg, bool doesNotRet, bool &isTailCall,
795 const SmallVectorImpl<ISD::OutputArg> &Outs,
796 const SmallVectorImpl<SDValue> &OutVals,
797 const SmallVectorImpl<ISD::InputArg> &Ins,
798 DebugLoc dl, SelectionDAG &DAG,
799 SmallVectorImpl<SDValue> &InVals) const;
802 LowerReturn(SDValue Chain,
803 CallingConv::ID CallConv, bool isVarArg,
804 const SmallVectorImpl<ISD::OutputArg> &Outs,
805 const SmallVectorImpl<SDValue> &OutVals,
806 DebugLoc dl, SelectionDAG &DAG) const;
808 virtual bool isUsedByReturnOnly(SDNode *N, SDValue &Chain) const;
810 virtual bool mayBeEmittedAsTailCall(CallInst *CI) const;
813 getTypeForExtArgOrReturn(LLVMContext &Context, EVT VT,
814 ISD::NodeType ExtendKind) const;
817 CanLowerReturn(CallingConv::ID CallConv, MachineFunction &MF,
819 const SmallVectorImpl<ISD::OutputArg> &Outs,
820 LLVMContext &Context) const;
822 void ReplaceATOMIC_BINARY_64(SDNode *N, SmallVectorImpl<SDValue> &Results,
823 SelectionDAG &DAG, unsigned NewOp) const;
825 /// Utility function to emit string processing sse4.2 instructions
826 /// that return in xmm0.
827 /// This takes the instruction to expand, the associated machine basic
828 /// block, the number of args, and whether or not the second arg is
829 /// in memory or not.
830 MachineBasicBlock *EmitPCMP(MachineInstr *BInstr, MachineBasicBlock *BB,
831 unsigned argNum, bool inMem) const;
833 /// Utility functions to emit monitor and mwait instructions. These
834 /// need to make sure that the arguments to the intrinsic are in the
835 /// correct registers.
836 MachineBasicBlock *EmitMonitor(MachineInstr *MI,
837 MachineBasicBlock *BB) const;
838 MachineBasicBlock *EmitMwait(MachineInstr *MI, MachineBasicBlock *BB) const;
840 /// Utility function to emit atomic bitwise operations (and, or, xor).
841 /// It takes the bitwise instruction to expand, the associated machine basic
842 /// block, and the associated X86 opcodes for reg/reg and reg/imm.
843 MachineBasicBlock *EmitAtomicBitwiseWithCustomInserter(
844 MachineInstr *BInstr,
845 MachineBasicBlock *BB,
852 const TargetRegisterClass *RC,
853 bool invSrc = false) const;
855 MachineBasicBlock *EmitAtomicBit6432WithCustomInserter(
856 MachineInstr *BInstr,
857 MachineBasicBlock *BB,
862 bool invSrc = false) const;
864 /// Utility function to emit atomic min and max. It takes the min/max
865 /// instruction to expand, the associated basic block, and the associated
866 /// cmov opcode for moving the min or max value.
867 MachineBasicBlock *EmitAtomicMinMaxWithCustomInserter(MachineInstr *BInstr,
868 MachineBasicBlock *BB,
869 unsigned cmovOpc) const;
871 // Utility function to emit the low-level va_arg code for X86-64.
872 MachineBasicBlock *EmitVAARG64WithCustomInserter(
874 MachineBasicBlock *MBB) const;
876 /// Utility function to emit the xmm reg save portion of va_start.
877 MachineBasicBlock *EmitVAStartSaveXMMRegsWithCustomInserter(
878 MachineInstr *BInstr,
879 MachineBasicBlock *BB) const;
881 MachineBasicBlock *EmitLoweredSelect(MachineInstr *I,
882 MachineBasicBlock *BB) const;
884 MachineBasicBlock *EmitLoweredWinAlloca(MachineInstr *MI,
885 MachineBasicBlock *BB) const;
887 MachineBasicBlock *EmitLoweredSegAlloca(MachineInstr *MI,
888 MachineBasicBlock *BB,
891 MachineBasicBlock *EmitLoweredTLSCall(MachineInstr *MI,
892 MachineBasicBlock *BB) const;
894 MachineBasicBlock *emitLoweredTLSAddr(MachineInstr *MI,
895 MachineBasicBlock *BB) const;
897 /// Emit nodes that will be selected as "test Op0,Op0", or something
898 /// equivalent, for use with the given x86 condition code.
899 SDValue EmitTest(SDValue Op0, unsigned X86CC, SelectionDAG &DAG) const;
901 /// Emit nodes that will be selected as "cmp Op0,Op1", or something
902 /// equivalent, for use with the given x86 condition code.
903 SDValue EmitCmp(SDValue Op0, SDValue Op1, unsigned X86CC,
904 SelectionDAG &DAG) const;
908 FastISel *createFastISel(FunctionLoweringInfo &funcInfo);
912 #endif // X86ISELLOWERING_H