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 flag
90 /// operand 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.
97 /// X86 conditional moves. Operand 0 and operand 1 are the two values
98 /// to select from. Operand 2 is the condition code, and operand 3 is the
99 /// flag operand produced by a CMP or TEST instruction. It also writes a
103 /// X86 conditional branches. Operand 0 is the chain operand, operand 1
104 /// is the block to branch if condition is true, operand 2 is the
105 /// condition code, and operand 3 is the flag operand produced by a CMP
106 /// or TEST instruction.
109 /// Return with a flag operand. Operand 0 is the chain operand, operand
110 /// 1 is the number of bytes of stack to pop.
113 /// REP_STOS - Repeat fill, corresponds to X86::REP_STOSx.
116 /// REP_MOVS - Repeat move, corresponds to X86::REP_MOVSx.
119 /// GlobalBaseReg - On Darwin, this node represents the result of the popl
120 /// at function entry, used for PIC code.
123 /// Wrapper - A wrapper node for TargetConstantPool,
124 /// TargetExternalSymbol, and TargetGlobalAddress.
127 /// WrapperRIP - Special wrapper used under X86-64 PIC mode for RIP
128 /// relative displacements.
131 /// MOVQ2DQ - Copies a 64-bit value from an MMX vector to the low word
132 /// of an XMM vector, with the high word zero filled.
135 /// MOVDQ2Q - Copies a 64-bit value from the low word of an XMM vector
136 /// to an MMX vector. If you think this is too close to the previous
137 /// mnemonic, so do I; blame Intel.
140 /// PEXTRB - Extract an 8-bit value from a vector and zero extend it to
141 /// i32, corresponds to X86::PEXTRB.
144 /// PEXTRW - Extract a 16-bit value from a vector and zero extend it to
145 /// i32, corresponds to X86::PEXTRW.
148 /// INSERTPS - Insert any element of a 4 x float vector into any element
149 /// of a destination 4 x floatvector.
152 /// PINSRB - Insert the lower 8-bits of a 32-bit value to a vector,
153 /// corresponds to X86::PINSRB.
156 /// PINSRW - Insert the lower 16-bits of a 32-bit value to a vector,
157 /// corresponds to X86::PINSRW.
160 /// PSHUFB - Shuffle 16 8-bit values within a vector.
163 /// FMAX, FMIN - Floating point max and min.
167 /// FRSQRT, FRCP - Floating point reciprocal-sqrt and reciprocal
168 /// approximation. Note that these typically require refinement
169 /// in order to obtain suitable precision.
172 // TLSADDR - Thread Local Storage.
175 // TLSCALL - Thread Local Storage. When calling to an OS provided
176 // thunk at the address from an earlier relocation.
179 // EH_RETURN - Exception Handling helpers.
182 /// TC_RETURN - Tail call return.
184 /// operand #1 callee (register or absolute)
185 /// operand #2 stack adjustment
186 /// operand #3 optional in flag
189 // VZEXT_MOVL - Vector move low and zero extend.
192 // VSHL, VSRL - Vector logical left / right shift.
195 // CMPPD, CMPPS - Vector double/float comparison.
196 // CMPPD, CMPPS - Vector double/float comparison.
199 // PCMP* - Vector integer comparisons.
200 PCMPEQB, PCMPEQW, PCMPEQD, PCMPEQQ,
201 PCMPGTB, PCMPGTW, PCMPGTD, PCMPGTQ,
203 // ADD, SUB, SMUL, etc. - Arithmetic operations with FLAGS results.
205 INC, DEC, OR, XOR, AND,
207 UMUL, // LOW, HI, FLAGS = umul LHS, RHS
209 // MUL_IMM - X86 specific multiply by immediate.
212 // PTEST - Vector bitwise comparisons
215 // TESTP - Vector packed fp sign bitwise comparisons
218 // Several flavors of instructions with vector shuffle behaviors.
253 // VASTART_SAVE_XMM_REGS - Save xmm argument registers to the stack,
254 // according to %al. An operator is needed so that this can be expanded
255 // with control flow.
256 VASTART_SAVE_XMM_REGS,
258 // WIN_ALLOCA - Windows's _chkstk call to do stack probing.
267 // ATOMADD64_DAG, ATOMSUB64_DAG, ATOMOR64_DAG, ATOMAND64_DAG,
268 // ATOMXOR64_DAG, ATOMNAND64_DAG, ATOMSWAP64_DAG -
269 // Atomic 64-bit binary operations.
270 ATOMADD64_DAG = ISD::FIRST_TARGET_MEMORY_OPCODE,
278 // LCMPXCHG_DAG, LCMPXCHG8_DAG - Compare and swap.
282 // VZEXT_LOAD - Load, scalar_to_vector, and zero extend.
285 // FNSTCW16m - Store FP control world into i16 memory.
288 /// FP_TO_INT*_IN_MEM - This instruction implements FP_TO_SINT with the
289 /// integer destination in memory and a FP reg source. This corresponds
290 /// to the X86::FIST*m instructions and the rounding mode change stuff. It
291 /// has two inputs (token chain and address) and two outputs (int value
292 /// and token chain).
297 /// FILD, FILD_FLAG - This instruction implements SINT_TO_FP with the
298 /// integer source in memory and FP reg result. This corresponds to the
299 /// X86::FILD*m instructions. It has three inputs (token chain, address,
300 /// and source type) and two outputs (FP value and token chain). FILD_FLAG
301 /// also produces a flag).
305 /// FLD - This instruction implements an extending load to FP stack slots.
306 /// This corresponds to the X86::FLD32m / X86::FLD64m. It takes a chain
307 /// operand, ptr to load from, and a ValueType node indicating the type
311 /// FST - This instruction implements a truncating store to FP stack
312 /// slots. This corresponds to the X86::FST32m / X86::FST64m. It takes a
313 /// chain operand, value to store, address, and a ValueType to store it
317 /// VAARG_64 - This instruction grabs the address of the next argument
318 /// from a va_list. (reads and modifies the va_list in memory)
321 // WARNING: Do not add anything in the end unless you want the node to
322 // have memop! In fact, starting from ATOMADD64_DAG all opcodes will be
323 // thought as target memory ops!
327 /// Define some predicates that are used for node matching.
329 /// isPSHUFDMask - Return true if the specified VECTOR_SHUFFLE operand
330 /// specifies a shuffle of elements that is suitable for input to PSHUFD.
331 bool isPSHUFDMask(ShuffleVectorSDNode *N);
333 /// isPSHUFHWMask - Return true if the specified VECTOR_SHUFFLE operand
334 /// specifies a shuffle of elements that is suitable for input to PSHUFD.
335 bool isPSHUFHWMask(ShuffleVectorSDNode *N);
337 /// isPSHUFLWMask - Return true if the specified VECTOR_SHUFFLE operand
338 /// specifies a shuffle of elements that is suitable for input to PSHUFD.
339 bool isPSHUFLWMask(ShuffleVectorSDNode *N);
341 /// isSHUFPMask - Return true if the specified VECTOR_SHUFFLE operand
342 /// specifies a shuffle of elements that is suitable for input to SHUFP*.
343 bool isSHUFPMask(ShuffleVectorSDNode *N);
345 /// isMOVHLPSMask - Return true if the specified VECTOR_SHUFFLE operand
346 /// specifies a shuffle of elements that is suitable for input to MOVHLPS.
347 bool isMOVHLPSMask(ShuffleVectorSDNode *N);
349 /// isMOVHLPS_v_undef_Mask - Special case of isMOVHLPSMask for canonical form
350 /// of vector_shuffle v, v, <2, 3, 2, 3>, i.e. vector_shuffle v, undef,
352 bool isMOVHLPS_v_undef_Mask(ShuffleVectorSDNode *N);
354 /// isMOVLPMask - Return true if the specified VECTOR_SHUFFLE operand
355 /// specifies a shuffle of elements that is suitable for MOVLP{S|D}.
356 bool isMOVLPMask(ShuffleVectorSDNode *N);
358 /// isMOVHPMask - Return true if the specified VECTOR_SHUFFLE operand
359 /// specifies a shuffle of elements that is suitable for MOVHP{S|D}.
360 /// as well as MOVLHPS.
361 bool isMOVLHPSMask(ShuffleVectorSDNode *N);
363 /// isUNPCKLMask - Return true if the specified VECTOR_SHUFFLE operand
364 /// specifies a shuffle of elements that is suitable for input to UNPCKL.
365 bool isUNPCKLMask(ShuffleVectorSDNode *N, bool V2IsSplat = false);
367 /// isUNPCKHMask - Return true if the specified VECTOR_SHUFFLE operand
368 /// specifies a shuffle of elements that is suitable for input to UNPCKH.
369 bool isUNPCKHMask(ShuffleVectorSDNode *N, bool V2IsSplat = false);
371 /// isUNPCKL_v_undef_Mask - Special case of isUNPCKLMask for canonical form
372 /// of vector_shuffle v, v, <0, 4, 1, 5>, i.e. vector_shuffle v, undef,
374 bool isUNPCKL_v_undef_Mask(ShuffleVectorSDNode *N);
376 /// isUNPCKH_v_undef_Mask - Special case of isUNPCKHMask for canonical form
377 /// of vector_shuffle v, v, <2, 6, 3, 7>, i.e. vector_shuffle v, undef,
379 bool isUNPCKH_v_undef_Mask(ShuffleVectorSDNode *N);
381 /// isMOVLMask - Return true if the specified VECTOR_SHUFFLE operand
382 /// specifies a shuffle of elements that is suitable for input to MOVSS,
383 /// MOVSD, and MOVD, i.e. setting the lowest element.
384 bool isMOVLMask(ShuffleVectorSDNode *N);
386 /// isMOVSHDUPMask - Return true if the specified VECTOR_SHUFFLE operand
387 /// specifies a shuffle of elements that is suitable for input to MOVSHDUP.
388 bool isMOVSHDUPMask(ShuffleVectorSDNode *N);
390 /// isMOVSLDUPMask - Return true if the specified VECTOR_SHUFFLE operand
391 /// specifies a shuffle of elements that is suitable for input to MOVSLDUP.
392 bool isMOVSLDUPMask(ShuffleVectorSDNode *N);
394 /// isMOVDDUPMask - Return true if the specified VECTOR_SHUFFLE operand
395 /// specifies a shuffle of elements that is suitable for input to MOVDDUP.
396 bool isMOVDDUPMask(ShuffleVectorSDNode *N);
398 /// isPALIGNRMask - Return true if the specified VECTOR_SHUFFLE operand
399 /// specifies a shuffle of elements that is suitable for input to PALIGNR.
400 bool isPALIGNRMask(ShuffleVectorSDNode *N);
402 /// getShuffleSHUFImmediate - Return the appropriate immediate to shuffle
403 /// the specified isShuffleMask VECTOR_SHUFFLE mask with PSHUF* and SHUFP*
405 unsigned getShuffleSHUFImmediate(SDNode *N);
407 /// getShufflePSHUFHWImmediate - Return the appropriate immediate to shuffle
408 /// the specified VECTOR_SHUFFLE mask with PSHUFHW instruction.
409 unsigned getShufflePSHUFHWImmediate(SDNode *N);
411 /// getShufflePSHUFLWImmediate - Return the appropriate immediate to shuffle
412 /// the specified VECTOR_SHUFFLE mask with PSHUFLW instruction.
413 unsigned getShufflePSHUFLWImmediate(SDNode *N);
415 /// getShufflePALIGNRImmediate - Return the appropriate immediate to shuffle
416 /// the specified VECTOR_SHUFFLE mask with the PALIGNR instruction.
417 unsigned getShufflePALIGNRImmediate(SDNode *N);
419 /// isZeroNode - Returns true if Elt is a constant zero or a floating point
421 bool isZeroNode(SDValue Elt);
423 /// isOffsetSuitableForCodeModel - Returns true of the given offset can be
424 /// fit into displacement field of the instruction.
425 bool isOffsetSuitableForCodeModel(int64_t Offset, CodeModel::Model M,
426 bool hasSymbolicDisplacement = true);
429 //===--------------------------------------------------------------------===//
430 // X86TargetLowering - X86 Implementation of the TargetLowering interface
431 class X86TargetLowering : public TargetLowering {
433 explicit X86TargetLowering(X86TargetMachine &TM);
435 virtual unsigned getJumpTableEncoding() const;
437 virtual const MCExpr *
438 LowerCustomJumpTableEntry(const MachineJumpTableInfo *MJTI,
439 const MachineBasicBlock *MBB, unsigned uid,
440 MCContext &Ctx) const;
442 /// getPICJumpTableRelocaBase - Returns relocation base for the given PIC
444 virtual SDValue getPICJumpTableRelocBase(SDValue Table,
445 SelectionDAG &DAG) const;
446 virtual const MCExpr *
447 getPICJumpTableRelocBaseExpr(const MachineFunction *MF,
448 unsigned JTI, MCContext &Ctx) const;
450 /// getStackPtrReg - Return the stack pointer register we are using: either
452 unsigned getStackPtrReg() const { return X86StackPtr; }
454 /// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
455 /// function arguments in the caller parameter area. For X86, aggregates
456 /// that contains are placed at 16-byte boundaries while the rest are at
457 /// 4-byte boundaries.
458 virtual unsigned getByValTypeAlignment(const Type *Ty) const;
460 /// getOptimalMemOpType - Returns the target specific optimal type for load
461 /// and store operations as a result of memset, memcpy, and memmove
462 /// lowering. If DstAlign is zero that means it's safe to destination
463 /// alignment can satisfy any constraint. Similarly if SrcAlign is zero it
464 /// means there isn't a need to check it against alignment requirement,
465 /// probably because the source does not need to be loaded. If
466 /// 'NonScalarIntSafe' is true, that means it's safe to return a
467 /// non-scalar-integer type, e.g. empty string source, constant, or loaded
468 /// from memory. 'MemcpyStrSrc' indicates whether the memcpy source is
469 /// constant so it does not need to be loaded.
470 /// It returns EVT::Other if the type should be determined using generic
471 /// target-independent logic.
473 getOptimalMemOpType(uint64_t Size, unsigned DstAlign, unsigned SrcAlign,
474 bool NonScalarIntSafe, bool MemcpyStrSrc,
475 MachineFunction &MF) const;
477 /// allowsUnalignedMemoryAccesses - Returns true if the target allows
478 /// unaligned memory accesses. of the specified type.
479 virtual bool allowsUnalignedMemoryAccesses(EVT VT) const {
483 /// LowerOperation - Provide custom lowering hooks for some operations.
485 virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const;
487 /// ReplaceNodeResults - Replace the results of node with an illegal result
488 /// type with new values built out of custom code.
490 virtual void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue>&Results,
491 SelectionDAG &DAG) const;
494 virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
496 /// isTypeDesirableForOp - Return true if the target has native support for
497 /// the specified value type and it is 'desirable' to use the type for the
498 /// given node type. e.g. On x86 i16 is legal, but undesirable since i16
499 /// instruction encodings are longer and some i16 instructions are slow.
500 virtual bool isTypeDesirableForOp(unsigned Opc, EVT VT) const;
502 /// isTypeDesirable - Return true if the target has native support for the
503 /// specified value type and it is 'desirable' to use the type. e.g. On x86
504 /// i16 is legal, but undesirable since i16 instruction encodings are longer
505 /// and some i16 instructions are slow.
506 virtual bool IsDesirableToPromoteOp(SDValue Op, EVT &PVT) const;
508 virtual MachineBasicBlock *
509 EmitInstrWithCustomInserter(MachineInstr *MI,
510 MachineBasicBlock *MBB) const;
513 /// getTargetNodeName - This method returns the name of a target specific
515 virtual const char *getTargetNodeName(unsigned Opcode) const;
517 /// getSetCCResultType - Return the ISD::SETCC ValueType
518 virtual MVT::SimpleValueType getSetCCResultType(EVT VT) const;
520 /// computeMaskedBitsForTargetNode - Determine which of the bits specified
521 /// in Mask are known to be either zero or one and return them in the
522 /// KnownZero/KnownOne bitsets.
523 virtual void computeMaskedBitsForTargetNode(const SDValue Op,
527 const SelectionDAG &DAG,
528 unsigned Depth = 0) const;
530 // ComputeNumSignBitsForTargetNode - Determine the number of bits in the
531 // operation that are sign bits.
532 virtual unsigned ComputeNumSignBitsForTargetNode(SDValue Op,
533 unsigned Depth) const;
536 isGAPlusOffset(SDNode *N, const GlobalValue* &GA, int64_t &Offset) const;
538 SDValue getReturnAddressFrameIndex(SelectionDAG &DAG) const;
540 virtual bool ExpandInlineAsm(CallInst *CI) const;
542 ConstraintType getConstraintType(const std::string &Constraint) const;
544 /// Examine constraint string and operand type and determine a weight value.
545 /// The operand object must already have been set up with the operand type.
546 virtual ConstraintWeight getSingleConstraintMatchWeight(
547 AsmOperandInfo &info, const char *constraint) const;
549 std::vector<unsigned>
550 getRegClassForInlineAsmConstraint(const std::string &Constraint,
553 virtual const char *LowerXConstraint(EVT ConstraintVT) const;
555 /// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
556 /// vector. If it is invalid, don't add anything to Ops. If hasMemory is
557 /// true it means one of the asm constraint of the inline asm instruction
558 /// being processed is 'm'.
559 virtual void LowerAsmOperandForConstraint(SDValue Op,
560 char ConstraintLetter,
561 std::vector<SDValue> &Ops,
562 SelectionDAG &DAG) const;
564 /// getRegForInlineAsmConstraint - Given a physical register constraint
565 /// (e.g. {edx}), return the register number and the register class for the
566 /// register. This should only be used for C_Register constraints. On
567 /// error, this returns a register number of 0.
568 std::pair<unsigned, const TargetRegisterClass*>
569 getRegForInlineAsmConstraint(const std::string &Constraint,
572 /// isLegalAddressingMode - Return true if the addressing mode represented
573 /// by AM is legal for this target, for a load/store of the specified type.
574 virtual bool isLegalAddressingMode(const AddrMode &AM, const Type *Ty)const;
576 /// isTruncateFree - Return true if it's free to truncate a value of
577 /// type Ty1 to type Ty2. e.g. On x86 it's free to truncate a i32 value in
578 /// register EAX to i16 by referencing its sub-register AX.
579 virtual bool isTruncateFree(const Type *Ty1, const Type *Ty2) const;
580 virtual bool isTruncateFree(EVT VT1, EVT VT2) const;
582 /// isZExtFree - Return true if any actual instruction that defines a
583 /// value of type Ty1 implicit zero-extends the value to Ty2 in the result
584 /// register. This does not necessarily include registers defined in
585 /// unknown ways, such as incoming arguments, or copies from unknown
586 /// virtual registers. Also, if isTruncateFree(Ty2, Ty1) is true, this
587 /// does not necessarily apply to truncate instructions. e.g. on x86-64,
588 /// all instructions that define 32-bit values implicit zero-extend the
589 /// result out to 64 bits.
590 virtual bool isZExtFree(const Type *Ty1, const Type *Ty2) const;
591 virtual bool isZExtFree(EVT VT1, EVT VT2) const;
593 /// isNarrowingProfitable - Return true if it's profitable to narrow
594 /// operations of type VT1 to VT2. e.g. on x86, it's profitable to narrow
595 /// from i32 to i8 but not from i32 to i16.
596 virtual bool isNarrowingProfitable(EVT VT1, EVT VT2) const;
598 /// isFPImmLegal - Returns true if the target can instruction select the
599 /// specified FP immediate natively. If false, the legalizer will
600 /// materialize the FP immediate as a load from a constant pool.
601 virtual bool isFPImmLegal(const APFloat &Imm, EVT VT) const;
603 /// isShuffleMaskLegal - Targets can use this to indicate that they only
604 /// support *some* VECTOR_SHUFFLE operations, those with specific masks.
605 /// By default, if a target supports the VECTOR_SHUFFLE node, all mask
606 /// values are assumed to be legal.
607 virtual bool isShuffleMaskLegal(const SmallVectorImpl<int> &Mask,
610 /// isVectorClearMaskLegal - Similar to isShuffleMaskLegal. This is
611 /// used by Targets can use this to indicate if there is a suitable
612 /// VECTOR_SHUFFLE that can be used to replace a VAND with a constant
614 virtual bool isVectorClearMaskLegal(const SmallVectorImpl<int> &Mask,
617 /// ShouldShrinkFPConstant - If true, then instruction selection should
618 /// seek to shrink the FP constant of the specified type to a smaller type
619 /// in order to save space and / or reduce runtime.
620 virtual bool ShouldShrinkFPConstant(EVT VT) const {
621 // Don't shrink FP constpool if SSE2 is available since cvtss2sd is more
622 // expensive than a straight movsd. On the other hand, it's important to
623 // shrink long double fp constant since fldt is very slow.
624 return !X86ScalarSSEf64 || VT == MVT::f80;
627 const X86Subtarget* getSubtarget() const {
631 /// isScalarFPTypeInSSEReg - Return true if the specified scalar FP type is
632 /// computed in an SSE register, not on the X87 floating point stack.
633 bool isScalarFPTypeInSSEReg(EVT VT) const {
634 return (VT == MVT::f64 && X86ScalarSSEf64) || // f64 is when SSE2
635 (VT == MVT::f32 && X86ScalarSSEf32); // f32 is when SSE1
638 /// createFastISel - This method returns a target specific FastISel object,
639 /// or null if the target does not support "fast" ISel.
640 virtual FastISel *createFastISel(FunctionLoweringInfo &funcInfo) const;
642 /// getFunctionAlignment - Return the Log2 alignment of this function.
643 virtual unsigned getFunctionAlignment(const Function *F) const;
645 unsigned getRegPressureLimit(const TargetRegisterClass *RC,
646 MachineFunction &MF) const;
648 /// getStackCookieLocation - Return true if the target stores stack
649 /// protector cookies at a fixed offset in some non-standard address
650 /// space, and populates the address space and offset as
652 virtual bool getStackCookieLocation(unsigned &AddressSpace, unsigned &Offset) const;
655 std::pair<const TargetRegisterClass*, uint8_t>
656 findRepresentativeClass(EVT VT) const;
659 /// Subtarget - Keep a pointer to the X86Subtarget around so that we can
660 /// make the right decision when generating code for different targets.
661 const X86Subtarget *Subtarget;
662 const X86RegisterInfo *RegInfo;
663 const TargetData *TD;
665 /// X86StackPtr - X86 physical register used as stack ptr.
666 unsigned X86StackPtr;
668 /// X86ScalarSSEf32, X86ScalarSSEf64 - Select between SSE or x87
669 /// floating point ops.
670 /// When SSE is available, use it for f32 operations.
671 /// When SSE2 is available, use it for f64 operations.
672 bool X86ScalarSSEf32;
673 bool X86ScalarSSEf64;
675 /// LegalFPImmediates - A list of legal fp immediates.
676 std::vector<APFloat> LegalFPImmediates;
678 /// addLegalFPImmediate - Indicate that this x86 target can instruction
679 /// select the specified FP immediate natively.
680 void addLegalFPImmediate(const APFloat& Imm) {
681 LegalFPImmediates.push_back(Imm);
684 SDValue LowerCallResult(SDValue Chain, SDValue InFlag,
685 CallingConv::ID CallConv, bool isVarArg,
686 const SmallVectorImpl<ISD::InputArg> &Ins,
687 DebugLoc dl, SelectionDAG &DAG,
688 SmallVectorImpl<SDValue> &InVals) const;
689 SDValue LowerMemArgument(SDValue Chain,
690 CallingConv::ID CallConv,
691 const SmallVectorImpl<ISD::InputArg> &ArgInfo,
692 DebugLoc dl, SelectionDAG &DAG,
693 const CCValAssign &VA, MachineFrameInfo *MFI,
695 SDValue LowerMemOpCallTo(SDValue Chain, SDValue StackPtr, SDValue Arg,
696 DebugLoc dl, SelectionDAG &DAG,
697 const CCValAssign &VA,
698 ISD::ArgFlagsTy Flags) const;
700 // Call lowering helpers.
702 /// IsEligibleForTailCallOptimization - Check whether the call is eligible
703 /// for tail call optimization. Targets which want to do tail call
704 /// optimization should implement this function.
705 bool IsEligibleForTailCallOptimization(SDValue Callee,
706 CallingConv::ID CalleeCC,
708 bool isCalleeStructRet,
709 bool isCallerStructRet,
710 const SmallVectorImpl<ISD::OutputArg> &Outs,
711 const SmallVectorImpl<SDValue> &OutVals,
712 const SmallVectorImpl<ISD::InputArg> &Ins,
713 SelectionDAG& DAG) const;
714 bool IsCalleePop(bool isVarArg, CallingConv::ID CallConv) const;
715 SDValue EmitTailCallLoadRetAddr(SelectionDAG &DAG, SDValue &OutRetAddr,
716 SDValue Chain, bool IsTailCall, bool Is64Bit,
717 int FPDiff, DebugLoc dl) const;
719 unsigned GetAlignedArgumentStackSize(unsigned StackSize,
720 SelectionDAG &DAG) const;
722 std::pair<SDValue,SDValue> FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG,
723 bool isSigned) const;
725 SDValue LowerAsSplatVectorLoad(SDValue SrcOp, EVT VT, DebugLoc dl,
726 SelectionDAG &DAG) const;
727 SDValue LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const;
728 SDValue LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) const;
729 SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const;
730 SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
731 SDValue LowerEXTRACT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) const;
732 SDValue LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
733 SDValue LowerINSERT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG) const;
734 SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) const;
735 SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) const;
736 SDValue LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const;
737 SDValue LowerGlobalAddress(const GlobalValue *GV, DebugLoc dl,
738 int64_t Offset, SelectionDAG &DAG) const;
739 SDValue LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const;
740 SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
741 SDValue LowerExternalSymbol(SDValue Op, SelectionDAG &DAG) const;
742 SDValue LowerShift(SDValue Op, SelectionDAG &DAG) const;
743 SDValue BuildFILD(SDValue Op, EVT SrcVT, SDValue Chain, SDValue StackSlot,
744 SelectionDAG &DAG) const;
745 SDValue LowerBITCAST(SDValue op, SelectionDAG &DAG) const;
746 SDValue LowerSINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
747 SDValue LowerUINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
748 SDValue LowerUINT_TO_FP_i64(SDValue Op, SelectionDAG &DAG) const;
749 SDValue LowerUINT_TO_FP_i32(SDValue Op, SelectionDAG &DAG) const;
750 SDValue LowerFP_TO_SINT(SDValue Op, SelectionDAG &DAG) const;
751 SDValue LowerFP_TO_UINT(SDValue Op, SelectionDAG &DAG) const;
752 SDValue LowerFABS(SDValue Op, SelectionDAG &DAG) const;
753 SDValue LowerFNEG(SDValue Op, SelectionDAG &DAG) const;
754 SDValue LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const;
755 SDValue LowerToBT(SDValue And, ISD::CondCode CC,
756 DebugLoc dl, SelectionDAG &DAG) const;
757 SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG) const;
758 SDValue LowerVSETCC(SDValue Op, SelectionDAG &DAG) const;
759 SDValue LowerSELECT(SDValue Op, SelectionDAG &DAG) const;
760 SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
761 SDValue LowerMEMSET(SDValue Op, SelectionDAG &DAG) const;
762 SDValue LowerJumpTable(SDValue Op, SelectionDAG &DAG) const;
763 SDValue LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const;
764 SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) const;
765 SDValue LowerVAARG(SDValue Op, SelectionDAG &DAG) const;
766 SDValue LowerVACOPY(SDValue Op, SelectionDAG &DAG) const;
767 SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
768 SDValue LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const;
769 SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
770 SDValue LowerFRAME_TO_ARGS_OFFSET(SDValue Op, SelectionDAG &DAG) const;
771 SDValue LowerEH_RETURN(SDValue Op, SelectionDAG &DAG) const;
772 SDValue LowerTRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
773 SDValue LowerFLT_ROUNDS_(SDValue Op, SelectionDAG &DAG) const;
774 SDValue LowerCTLZ(SDValue Op, SelectionDAG &DAG) const;
775 SDValue LowerCTTZ(SDValue Op, SelectionDAG &DAG) const;
776 SDValue LowerMUL_V2I64(SDValue Op, SelectionDAG &DAG) const;
777 SDValue LowerSHL(SDValue Op, SelectionDAG &DAG) const;
778 SDValue LowerXALUO(SDValue Op, SelectionDAG &DAG) const;
780 SDValue LowerCMP_SWAP(SDValue Op, SelectionDAG &DAG) const;
781 SDValue LowerLOAD_SUB(SDValue Op, SelectionDAG &DAG) const;
782 SDValue LowerREADCYCLECOUNTER(SDValue Op, SelectionDAG &DAG) const;
783 SDValue LowerMEMBARRIER(SDValue Op, SelectionDAG &DAG) const;
785 // Utility functions to help LowerVECTOR_SHUFFLE
786 SDValue LowerVECTOR_SHUFFLEv8i16(SDValue Op, SelectionDAG &DAG) const;
789 LowerFormalArguments(SDValue Chain,
790 CallingConv::ID CallConv, bool isVarArg,
791 const SmallVectorImpl<ISD::InputArg> &Ins,
792 DebugLoc dl, SelectionDAG &DAG,
793 SmallVectorImpl<SDValue> &InVals) const;
795 LowerCall(SDValue Chain, SDValue Callee,
796 CallingConv::ID CallConv, bool isVarArg, bool &isTailCall,
797 const SmallVectorImpl<ISD::OutputArg> &Outs,
798 const SmallVectorImpl<SDValue> &OutVals,
799 const SmallVectorImpl<ISD::InputArg> &Ins,
800 DebugLoc dl, SelectionDAG &DAG,
801 SmallVectorImpl<SDValue> &InVals) const;
804 LowerReturn(SDValue Chain,
805 CallingConv::ID CallConv, bool isVarArg,
806 const SmallVectorImpl<ISD::OutputArg> &Outs,
807 const SmallVectorImpl<SDValue> &OutVals,
808 DebugLoc dl, SelectionDAG &DAG) const;
810 virtual bool isUsedByReturnOnly(SDNode *N) const;
813 CanLowerReturn(CallingConv::ID CallConv, bool isVarArg,
814 const SmallVectorImpl<ISD::OutputArg> &Outs,
815 LLVMContext &Context) const;
817 void ReplaceATOMIC_BINARY_64(SDNode *N, SmallVectorImpl<SDValue> &Results,
818 SelectionDAG &DAG, unsigned NewOp) const;
820 /// Utility function to emit string processing sse4.2 instructions
821 /// that return in xmm0.
822 /// This takes the instruction to expand, the associated machine basic
823 /// block, the number of args, and whether or not the second arg is
824 /// in memory or not.
825 MachineBasicBlock *EmitPCMP(MachineInstr *BInstr, MachineBasicBlock *BB,
826 unsigned argNum, bool inMem) const;
828 /// Utility functions to emit monitor and mwait instructions. These
829 /// need to make sure that the arguments to the intrinsic are in the
830 /// correct registers.
831 MachineBasicBlock *EmitMonitor(MachineInstr *MI,
832 MachineBasicBlock *BB) const;
833 MachineBasicBlock *EmitMwait(MachineInstr *MI, MachineBasicBlock *BB) const;
835 /// Utility function to emit atomic bitwise operations (and, or, xor).
836 /// It takes the bitwise instruction to expand, the associated machine basic
837 /// block, and the associated X86 opcodes for reg/reg and reg/imm.
838 MachineBasicBlock *EmitAtomicBitwiseWithCustomInserter(
839 MachineInstr *BInstr,
840 MachineBasicBlock *BB,
847 TargetRegisterClass *RC,
848 bool invSrc = false) const;
850 MachineBasicBlock *EmitAtomicBit6432WithCustomInserter(
851 MachineInstr *BInstr,
852 MachineBasicBlock *BB,
857 bool invSrc = false) const;
859 /// Utility function to emit atomic min and max. It takes the min/max
860 /// instruction to expand, the associated basic block, and the associated
861 /// cmov opcode for moving the min or max value.
862 MachineBasicBlock *EmitAtomicMinMaxWithCustomInserter(MachineInstr *BInstr,
863 MachineBasicBlock *BB,
864 unsigned cmovOpc) const;
866 // Utility function to emit the low-level va_arg code for X86-64.
867 MachineBasicBlock *EmitVAARG64WithCustomInserter(
869 MachineBasicBlock *MBB) const;
871 /// Utility function to emit the xmm reg save portion of va_start.
872 MachineBasicBlock *EmitVAStartSaveXMMRegsWithCustomInserter(
873 MachineInstr *BInstr,
874 MachineBasicBlock *BB) const;
876 MachineBasicBlock *EmitLoweredSelect(MachineInstr *I,
877 MachineBasicBlock *BB) const;
879 MachineBasicBlock *EmitLoweredWinAlloca(MachineInstr *MI,
880 MachineBasicBlock *BB) const;
882 MachineBasicBlock *EmitLoweredTLSCall(MachineInstr *MI,
883 MachineBasicBlock *BB) const;
885 MachineBasicBlock *emitLoweredTLSAddr(MachineInstr *MI,
886 MachineBasicBlock *BB) const;
888 /// Emit nodes that will be selected as "test Op0,Op0", or something
889 /// equivalent, for use with the given x86 condition code.
890 SDValue EmitTest(SDValue Op0, unsigned X86CC, SelectionDAG &DAG) const;
892 /// Emit nodes that will be selected as "cmp Op0,Op1", or something
893 /// equivalent, for use with the given x86 condition code.
894 SDValue EmitCmp(SDValue Op0, SDValue Op1, unsigned X86CC,
895 SelectionDAG &DAG) const;
899 FastISel *createFastISel(FunctionLoweringInfo &funcInfo);
903 #endif // X86ISELLOWERING_H