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/CodeGen/FastISel.h"
23 #include "llvm/CodeGen/SelectionDAG.h"
24 #include "llvm/CodeGen/CallingConvLower.h"
28 // X86 Specific DAG Nodes
30 // Start the numbering where the builtin ops leave off.
31 FIRST_NUMBER = ISD::BUILTIN_OP_END,
33 /// BSF - Bit scan forward.
34 /// BSR - Bit scan reverse.
38 /// SHLD, SHRD - Double shift instructions. These correspond to
39 /// X86::SHLDxx and X86::SHRDxx instructions.
43 /// FAND - Bitwise logical AND of floating point values. This corresponds
44 /// to X86::ANDPS or X86::ANDPD.
47 /// FOR - Bitwise logical OR of floating point values. This corresponds
48 /// to X86::ORPS or X86::ORPD.
51 /// FXOR - Bitwise logical XOR of floating point values. This corresponds
52 /// to X86::XORPS or X86::XORPD.
55 /// FSRL - Bitwise logical right shift of floating point values. These
56 /// corresponds to X86::PSRLDQ.
59 /// FILD, FILD_FLAG - This instruction implements SINT_TO_FP with the
60 /// integer source in memory and FP reg result. This corresponds to the
61 /// X86::FILD*m instructions. It has three inputs (token chain, address,
62 /// and source type) and two outputs (FP value and token chain). FILD_FLAG
63 /// also produces a flag).
67 /// FP_TO_INT*_IN_MEM - This instruction implements FP_TO_SINT with the
68 /// integer destination in memory and a FP reg source. This corresponds
69 /// to the X86::FIST*m instructions and the rounding mode change stuff. It
70 /// has two inputs (token chain and address) and two outputs (int value
76 /// FLD - This instruction implements an extending load to FP stack slots.
77 /// This corresponds to the X86::FLD32m / X86::FLD64m. It takes a chain
78 /// operand, ptr to load from, and a ValueType node indicating the type
82 /// FST - This instruction implements a truncating store to FP stack
83 /// slots. This corresponds to the X86::FST32m / X86::FST64m. It takes a
84 /// chain operand, value to store, address, and a ValueType to store it
88 /// CALL - These operations represent an abstract X86 call
89 /// instruction, which includes a bunch of information. In particular the
90 /// operands of these node are:
92 /// #0 - The incoming token chain
94 /// #2 - The number of arg bytes the caller pushes on the stack.
95 /// #3 - The number of arg bytes the callee pops off the stack.
96 /// #4 - The value to pass in AL/AX/EAX (optional)
97 /// #5 - The value to pass in DL/DX/EDX (optional)
99 /// The result values of these nodes are:
101 /// #0 - The outgoing token chain
102 /// #1 - The first register result value (optional)
103 /// #2 - The second register result value (optional)
107 /// RDTSC_DAG - This operation implements the lowering for
111 /// X86 compare and logical compare instructions.
114 /// X86 bit-test instructions.
117 /// X86 SetCC. Operand 0 is condition code, and operand 1 is the flag
118 /// operand produced by a CMP instruction.
121 /// X86 conditional moves. Operand 0 and operand 1 are the two values
122 /// to select from. Operand 2 is the condition code, and operand 3 is the
123 /// flag operand produced by a CMP or TEST instruction. It also writes a
127 /// X86 conditional branches. Operand 0 is the chain operand, operand 1
128 /// is the block to branch if condition is true, operand 2 is the
129 /// condition code, and operand 3 is the flag operand produced by a CMP
130 /// or TEST instruction.
133 /// Return with a flag operand. Operand 0 is the chain operand, operand
134 /// 1 is the number of bytes of stack to pop.
137 /// REP_STOS - Repeat fill, corresponds to X86::REP_STOSx.
140 /// REP_MOVS - Repeat move, corresponds to X86::REP_MOVSx.
143 /// GlobalBaseReg - On Darwin, this node represents the result of the popl
144 /// at function entry, used for PIC code.
147 /// Wrapper - A wrapper node for TargetConstantPool,
148 /// TargetExternalSymbol, and TargetGlobalAddress.
151 /// WrapperRIP - Special wrapper used under X86-64 PIC mode for RIP
152 /// relative displacements.
155 /// PEXTRB - Extract an 8-bit value from a vector and zero extend it to
156 /// i32, corresponds to X86::PEXTRB.
159 /// PEXTRW - Extract a 16-bit value from a vector and zero extend it to
160 /// i32, corresponds to X86::PEXTRW.
163 /// INSERTPS - Insert any element of a 4 x float vector into any element
164 /// of a destination 4 x floatvector.
167 /// PINSRB - Insert the lower 8-bits of a 32-bit value to a vector,
168 /// corresponds to X86::PINSRB.
171 /// PINSRW - Insert the lower 16-bits of a 32-bit value to a vector,
172 /// corresponds to X86::PINSRW.
175 /// PSHUFB - Shuffle 16 8-bit values within a vector.
178 /// FMAX, FMIN - Floating point max and min.
182 /// FRSQRT, FRCP - Floating point reciprocal-sqrt and reciprocal
183 /// approximation. Note that these typically require refinement
184 /// in order to obtain suitable precision.
187 // TLSADDR - Thread Local Storage.
190 // SegmentBaseAddress - The address segment:0
193 // EH_RETURN - Exception Handling helpers.
196 /// TC_RETURN - Tail call return.
198 /// operand #1 callee (register or absolute)
199 /// operand #2 stack adjustment
200 /// operand #3 optional in flag
203 // LCMPXCHG_DAG, LCMPXCHG8_DAG - Compare and swap.
207 // ATOMADD64_DAG, ATOMSUB64_DAG, ATOMOR64_DAG, ATOMAND64_DAG,
208 // ATOMXOR64_DAG, ATOMNAND64_DAG, ATOMSWAP64_DAG -
209 // Atomic 64-bit binary operations.
218 // FNSTCW16m - Store FP control world into i16 memory.
221 // VZEXT_MOVL - Vector move low and zero extend.
224 // VZEXT_LOAD - Load, scalar_to_vector, and zero extend.
227 // VSHL, VSRL - Vector logical left / right shift.
230 // CMPPD, CMPPS - Vector double/float comparison.
231 // CMPPD, CMPPS - Vector double/float comparison.
234 // PCMP* - Vector integer comparisons.
235 PCMPEQB, PCMPEQW, PCMPEQD, PCMPEQQ,
236 PCMPGTB, PCMPGTW, PCMPGTD, PCMPGTQ,
238 // ADD, SUB, SMUL, UMUL, etc. - Arithmetic operations with FLAGS results.
239 ADD, SUB, SMUL, UMUL,
242 // MUL_IMM - X86 specific multiply by immediate.
245 // PTEST - Vector bitwise comparisons
250 /// Define some predicates that are used for node matching.
252 /// isPSHUFDMask - Return true if the specified VECTOR_SHUFFLE operand
253 /// specifies a shuffle of elements that is suitable for input to PSHUFD.
254 bool isPSHUFDMask(ShuffleVectorSDNode *N);
256 /// isPSHUFHWMask - Return true if the specified VECTOR_SHUFFLE operand
257 /// specifies a shuffle of elements that is suitable for input to PSHUFD.
258 bool isPSHUFHWMask(ShuffleVectorSDNode *N);
260 /// isPSHUFLWMask - Return true if the specified VECTOR_SHUFFLE operand
261 /// specifies a shuffle of elements that is suitable for input to PSHUFD.
262 bool isPSHUFLWMask(ShuffleVectorSDNode *N);
264 /// isSHUFPMask - Return true if the specified VECTOR_SHUFFLE operand
265 /// specifies a shuffle of elements that is suitable for input to SHUFP*.
266 bool isSHUFPMask(ShuffleVectorSDNode *N);
268 /// isMOVHLPSMask - Return true if the specified VECTOR_SHUFFLE operand
269 /// specifies a shuffle of elements that is suitable for input to MOVHLPS.
270 bool isMOVHLPSMask(ShuffleVectorSDNode *N);
272 /// isMOVHLPS_v_undef_Mask - Special case of isMOVHLPSMask for canonical form
273 /// of vector_shuffle v, v, <2, 3, 2, 3>, i.e. vector_shuffle v, undef,
275 bool isMOVHLPS_v_undef_Mask(ShuffleVectorSDNode *N);
277 /// isMOVLPMask - Return true if the specified VECTOR_SHUFFLE operand
278 /// specifies a shuffle of elements that is suitable for MOVLP{S|D}.
279 bool isMOVLPMask(ShuffleVectorSDNode *N);
281 /// isMOVHPMask - Return true if the specified VECTOR_SHUFFLE operand
282 /// specifies a shuffle of elements that is suitable for MOVHP{S|D}.
283 /// as well as MOVLHPS.
284 bool isMOVHPMask(ShuffleVectorSDNode *N);
286 /// isUNPCKLMask - Return true if the specified VECTOR_SHUFFLE operand
287 /// specifies a shuffle of elements that is suitable for input to UNPCKL.
288 bool isUNPCKLMask(ShuffleVectorSDNode *N, bool V2IsSplat = false);
290 /// isUNPCKHMask - Return true if the specified VECTOR_SHUFFLE operand
291 /// specifies a shuffle of elements that is suitable for input to UNPCKH.
292 bool isUNPCKHMask(ShuffleVectorSDNode *N, bool V2IsSplat = false);
294 /// isUNPCKL_v_undef_Mask - Special case of isUNPCKLMask for canonical form
295 /// of vector_shuffle v, v, <0, 4, 1, 5>, i.e. vector_shuffle v, undef,
297 bool isUNPCKL_v_undef_Mask(ShuffleVectorSDNode *N);
299 /// isUNPCKH_v_undef_Mask - Special case of isUNPCKHMask for canonical form
300 /// of vector_shuffle v, v, <2, 6, 3, 7>, i.e. vector_shuffle v, undef,
302 bool isUNPCKH_v_undef_Mask(ShuffleVectorSDNode *N);
304 /// isMOVLMask - Return true if the specified VECTOR_SHUFFLE operand
305 /// specifies a shuffle of elements that is suitable for input to MOVSS,
306 /// MOVSD, and MOVD, i.e. setting the lowest element.
307 bool isMOVLMask(ShuffleVectorSDNode *N);
309 /// isMOVSHDUPMask - Return true if the specified VECTOR_SHUFFLE operand
310 /// specifies a shuffle of elements that is suitable for input to MOVSHDUP.
311 bool isMOVSHDUPMask(ShuffleVectorSDNode *N);
313 /// isMOVSLDUPMask - Return true if the specified VECTOR_SHUFFLE operand
314 /// specifies a shuffle of elements that is suitable for input to MOVSLDUP.
315 bool isMOVSLDUPMask(ShuffleVectorSDNode *N);
317 /// isMOVDDUPMask - Return true if the specified VECTOR_SHUFFLE operand
318 /// specifies a shuffle of elements that is suitable for input to MOVDDUP.
319 bool isMOVDDUPMask(ShuffleVectorSDNode *N);
321 /// getShuffleSHUFImmediate - Return the appropriate immediate to shuffle
322 /// the specified isShuffleMask VECTOR_SHUFFLE mask with PSHUF* and SHUFP*
324 unsigned getShuffleSHUFImmediate(SDNode *N);
326 /// getShufflePSHUFHWImmediate - Return the appropriate immediate to shuffle
327 /// the specified isShuffleMask VECTOR_SHUFFLE mask with PSHUFHW
329 unsigned getShufflePSHUFHWImmediate(SDNode *N);
331 /// getShufflePSHUFKWImmediate - Return the appropriate immediate to shuffle
332 /// the specified isShuffleMask VECTOR_SHUFFLE mask with PSHUFLW
334 unsigned getShufflePSHUFLWImmediate(SDNode *N);
336 /// isZeroNode - Returns true if Elt is a constant zero or a floating point
338 bool isZeroNode(SDValue Elt);
341 //===--------------------------------------------------------------------===//
342 // X86TargetLowering - X86 Implementation of the TargetLowering interface
343 class X86TargetLowering : public TargetLowering {
344 int VarArgsFrameIndex; // FrameIndex for start of varargs area.
345 int RegSaveFrameIndex; // X86-64 vararg func register save area.
346 unsigned VarArgsGPOffset; // X86-64 vararg func int reg offset.
347 unsigned VarArgsFPOffset; // X86-64 vararg func fp reg offset.
348 int BytesToPopOnReturn; // Number of arg bytes ret should pop.
349 int BytesCallerReserves; // Number of arg bytes caller makes.
352 explicit X86TargetLowering(X86TargetMachine &TM);
354 /// getPICJumpTableRelocaBase - Returns relocation base for the given PIC
356 SDValue getPICJumpTableRelocBase(SDValue Table,
357 SelectionDAG &DAG) const;
359 // Return the number of bytes that a function should pop when it returns (in
360 // addition to the space used by the return address).
362 unsigned getBytesToPopOnReturn() const { return BytesToPopOnReturn; }
364 // Return the number of bytes that the caller reserves for arguments passed
366 unsigned getBytesCallerReserves() const { return BytesCallerReserves; }
368 /// getStackPtrReg - Return the stack pointer register we are using: either
370 unsigned getStackPtrReg() const { return X86StackPtr; }
372 /// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
373 /// function arguments in the caller parameter area. For X86, aggregates
374 /// that contains are placed at 16-byte boundaries while the rest are at
375 /// 4-byte boundaries.
376 virtual unsigned getByValTypeAlignment(const Type *Ty) const;
378 /// getOptimalMemOpType - Returns the target specific optimal type for load
379 /// and store operations as a result of memset, memcpy, and memmove
380 /// lowering. It returns MVT::iAny if SelectionDAG should be responsible for
383 MVT getOptimalMemOpType(uint64_t Size, unsigned Align,
384 bool isSrcConst, bool isSrcStr,
385 SelectionDAG &DAG) const;
387 /// LowerOperation - Provide custom lowering hooks for some operations.
389 virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG);
391 /// ReplaceNodeResults - Replace the results of node with an illegal result
392 /// type with new values built out of custom code.
394 virtual void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue>&Results,
398 virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
400 virtual MachineBasicBlock *EmitInstrWithCustomInserter(MachineInstr *MI,
401 MachineBasicBlock *MBB) const;
404 /// getTargetNodeName - This method returns the name of a target specific
406 virtual const char *getTargetNodeName(unsigned Opcode) const;
408 /// getSetCCResultType - Return the ISD::SETCC ValueType
409 virtual MVT getSetCCResultType(MVT VT) const;
411 /// computeMaskedBitsForTargetNode - Determine which of the bits specified
412 /// in Mask are known to be either zero or one and return them in the
413 /// KnownZero/KnownOne bitsets.
414 virtual void computeMaskedBitsForTargetNode(const SDValue Op,
418 const SelectionDAG &DAG,
419 unsigned Depth = 0) const;
422 isGAPlusOffset(SDNode *N, GlobalValue* &GA, int64_t &Offset) const;
424 SDValue getReturnAddressFrameIndex(SelectionDAG &DAG);
426 virtual bool ExpandInlineAsm(CallInst *CI) const;
428 ConstraintType getConstraintType(const std::string &Constraint) const;
430 std::vector<unsigned>
431 getRegClassForInlineAsmConstraint(const std::string &Constraint,
434 virtual const char *LowerXConstraint(MVT ConstraintVT) const;
436 /// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
437 /// vector. If it is invalid, don't add anything to Ops. If hasMemory is
438 /// true it means one of the asm constraint of the inline asm instruction
439 /// being processed is 'm'.
440 virtual void LowerAsmOperandForConstraint(SDValue Op,
441 char ConstraintLetter,
443 std::vector<SDValue> &Ops,
444 SelectionDAG &DAG) const;
446 /// getRegForInlineAsmConstraint - Given a physical register constraint
447 /// (e.g. {edx}), return the register number and the register class for the
448 /// register. This should only be used for C_Register constraints. On
449 /// error, this returns a register number of 0.
450 std::pair<unsigned, const TargetRegisterClass*>
451 getRegForInlineAsmConstraint(const std::string &Constraint,
454 /// isLegalAddressingMode - Return true if the addressing mode represented
455 /// by AM is legal for this target, for a load/store of the specified type.
456 virtual bool isLegalAddressingMode(const AddrMode &AM, const Type *Ty)const;
458 /// isTruncateFree - Return true if it's free to truncate a value of
459 /// type Ty1 to type Ty2. e.g. On x86 it's free to truncate a i32 value in
460 /// register EAX to i16 by referencing its sub-register AX.
461 virtual bool isTruncateFree(const Type *Ty1, const Type *Ty2) const;
462 virtual bool isTruncateFree(MVT VT1, MVT VT2) const;
464 /// isZExtFree - Return true if any actual instruction that defines a
465 /// value of type Ty1 implicit zero-extends the value to Ty2 in the result
466 /// register. This does not necessarily include registers defined in
467 /// unknown ways, such as incoming arguments, or copies from unknown
468 /// virtual registers. Also, if isTruncateFree(Ty2, Ty1) is true, this
469 /// does not necessarily apply to truncate instructions. e.g. on x86-64,
470 /// all instructions that define 32-bit values implicit zero-extend the
471 /// result out to 64 bits.
472 virtual bool isZExtFree(const Type *Ty1, const Type *Ty2) const;
473 virtual bool isZExtFree(MVT VT1, MVT VT2) const;
475 /// isNarrowingProfitable - Return true if it's profitable to narrow
476 /// operations of type VT1 to VT2. e.g. on x86, it's profitable to narrow
477 /// from i32 to i8 but not from i32 to i16.
478 virtual bool isNarrowingProfitable(MVT VT1, MVT VT2) const;
480 /// isShuffleMaskLegal - Targets can use this to indicate that they only
481 /// support *some* VECTOR_SHUFFLE operations, those with specific masks.
482 /// By default, if a target supports the VECTOR_SHUFFLE node, all mask
483 /// values are assumed to be legal.
484 virtual bool isShuffleMaskLegal(const SmallVectorImpl<int> &Mask,
487 /// isVectorClearMaskLegal - Similar to isShuffleMaskLegal. This is
488 /// used by Targets can use this to indicate if there is a suitable
489 /// VECTOR_SHUFFLE that can be used to replace a VAND with a constant
491 virtual bool isVectorClearMaskLegal(const SmallVectorImpl<int> &Mask,
494 /// ShouldShrinkFPConstant - If true, then instruction selection should
495 /// seek to shrink the FP constant of the specified type to a smaller type
496 /// in order to save space and / or reduce runtime.
497 virtual bool ShouldShrinkFPConstant(MVT VT) const {
498 // Don't shrink FP constpool if SSE2 is available since cvtss2sd is more
499 // expensive than a straight movsd. On the other hand, it's important to
500 // shrink long double fp constant since fldt is very slow.
501 return !X86ScalarSSEf64 || VT == MVT::f80;
504 /// IsEligibleForTailCallOptimization - Check whether the call is eligible
505 /// for tail call optimization. Targets which want to do tail call
506 /// optimization should implement this function.
508 IsEligibleForTailCallOptimization(SDValue Callee,
511 const SmallVectorImpl<ISD::InputArg> &Ins,
512 SelectionDAG& DAG) const;
514 virtual const X86Subtarget* getSubtarget() {
518 /// isScalarFPTypeInSSEReg - Return true if the specified scalar FP type is
519 /// computed in an SSE register, not on the X87 floating point stack.
520 bool isScalarFPTypeInSSEReg(MVT VT) const {
521 return (VT == MVT::f64 && X86ScalarSSEf64) || // f64 is when SSE2
522 (VT == MVT::f32 && X86ScalarSSEf32); // f32 is when SSE1
525 /// getWidenVectorType: given a vector type, returns the type to widen
526 /// to (e.g., v7i8 to v8i8). If the vector type is legal, it returns itself.
527 /// If there is no vector type that we want to widen to, returns MVT::Other
528 /// When and were to widen is target dependent based on the cost of
529 /// scalarizing vs using the wider vector type.
530 virtual MVT getWidenVectorType(MVT VT) const;
532 /// createFastISel - This method returns a target specific FastISel object,
533 /// or null if the target does not support "fast" ISel.
535 createFastISel(MachineFunction &mf,
536 MachineModuleInfo *mmi, DwarfWriter *dw,
537 DenseMap<const Value *, unsigned> &,
538 DenseMap<const BasicBlock *, MachineBasicBlock *> &,
539 DenseMap<const AllocaInst *, int> &
541 , SmallSet<Instruction*, 8> &
545 /// getFunctionAlignment - Return the Log2 alignment of this function.
546 virtual unsigned getFunctionAlignment(const Function *F) const;
549 /// Subtarget - Keep a pointer to the X86Subtarget around so that we can
550 /// make the right decision when generating code for different targets.
551 const X86Subtarget *Subtarget;
552 const X86RegisterInfo *RegInfo;
553 const TargetData *TD;
555 /// X86StackPtr - X86 physical register used as stack ptr.
556 unsigned X86StackPtr;
558 /// X86ScalarSSEf32, X86ScalarSSEf64 - Select between SSE or x87
559 /// floating point ops.
560 /// When SSE is available, use it for f32 operations.
561 /// When SSE2 is available, use it for f64 operations.
562 bool X86ScalarSSEf32;
563 bool X86ScalarSSEf64;
565 SDValue LowerCallResult(SDValue Chain, SDValue InFlag,
566 unsigned CallConv, bool isVarArg,
567 const SmallVectorImpl<ISD::InputArg> &Ins,
568 DebugLoc dl, SelectionDAG &DAG,
569 SmallVectorImpl<SDValue> &InVals);
570 SDValue LowerMemArgument(SDValue Chain,
572 const SmallVectorImpl<ISD::InputArg> &ArgInfo,
573 DebugLoc dl, SelectionDAG &DAG,
574 const CCValAssign &VA, MachineFrameInfo *MFI,
576 SDValue LowerMemOpCallTo(SDValue Chain, SDValue StackPtr, SDValue Arg,
577 DebugLoc dl, SelectionDAG &DAG,
578 const CCValAssign &VA,
579 ISD::ArgFlagsTy Flags);
581 // Call lowering helpers.
582 bool IsCalleePop(bool isVarArg, unsigned CallConv);
583 SDValue EmitTailCallLoadRetAddr(SelectionDAG &DAG, SDValue &OutRetAddr,
584 SDValue Chain, bool IsTailCall, bool Is64Bit,
585 int FPDiff, DebugLoc dl);
587 CCAssignFn *CCAssignFnForNode(unsigned CallConv) const;
588 NameDecorationStyle NameDecorationForCallConv(unsigned CallConv);
589 unsigned GetAlignedArgumentStackSize(unsigned StackSize, SelectionDAG &DAG);
591 std::pair<SDValue,SDValue> FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG,
594 SDValue LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG);
595 SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG);
596 SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG);
597 SDValue LowerEXTRACT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG);
598 SDValue LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG);
599 SDValue LowerINSERT_VECTOR_ELT_SSE4(SDValue Op, SelectionDAG &DAG);
600 SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG);
601 SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG);
602 SDValue LowerGlobalAddress(const GlobalValue *GV, DebugLoc dl,
603 int64_t Offset, SelectionDAG &DAG) const;
604 SDValue LowerGlobalAddress(SDValue Op, SelectionDAG &DAG);
605 SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG);
606 SDValue LowerExternalSymbol(SDValue Op, SelectionDAG &DAG);
607 SDValue LowerShift(SDValue Op, SelectionDAG &DAG);
608 SDValue BuildFILD(SDValue Op, MVT SrcVT, SDValue Chain, SDValue StackSlot,
610 SDValue LowerSINT_TO_FP(SDValue Op, SelectionDAG &DAG);
611 SDValue LowerUINT_TO_FP(SDValue Op, SelectionDAG &DAG);
612 SDValue LowerUINT_TO_FP_i64(SDValue Op, SelectionDAG &DAG);
613 SDValue LowerUINT_TO_FP_i32(SDValue Op, SelectionDAG &DAG);
614 SDValue LowerFP_TO_SINT(SDValue Op, SelectionDAG &DAG);
615 SDValue LowerFP_TO_UINT(SDValue Op, SelectionDAG &DAG);
616 SDValue LowerFABS(SDValue Op, SelectionDAG &DAG);
617 SDValue LowerFNEG(SDValue Op, SelectionDAG &DAG);
618 SDValue LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG);
619 SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG);
620 SDValue LowerVSETCC(SDValue Op, SelectionDAG &DAG);
621 SDValue LowerSELECT(SDValue Op, SelectionDAG &DAG);
622 SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG);
623 SDValue LowerMEMSET(SDValue Op, SelectionDAG &DAG);
624 SDValue LowerJumpTable(SDValue Op, SelectionDAG &DAG);
625 SDValue LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG);
626 SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG);
627 SDValue LowerVAARG(SDValue Op, SelectionDAG &DAG);
628 SDValue LowerVACOPY(SDValue Op, SelectionDAG &DAG);
629 SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG);
630 SDValue LowerRETURNADDR(SDValue Op, SelectionDAG &DAG);
631 SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG);
632 SDValue LowerFRAME_TO_ARGS_OFFSET(SDValue Op, SelectionDAG &DAG);
633 SDValue LowerEH_RETURN(SDValue Op, SelectionDAG &DAG);
634 SDValue LowerTRAMPOLINE(SDValue Op, SelectionDAG &DAG);
635 SDValue LowerFLT_ROUNDS_(SDValue Op, SelectionDAG &DAG);
636 SDValue LowerCTLZ(SDValue Op, SelectionDAG &DAG);
637 SDValue LowerCTTZ(SDValue Op, SelectionDAG &DAG);
638 SDValue LowerMUL_V2I64(SDValue Op, SelectionDAG &DAG);
639 SDValue LowerXALUO(SDValue Op, SelectionDAG &DAG);
641 SDValue LowerCMP_SWAP(SDValue Op, SelectionDAG &DAG);
642 SDValue LowerLOAD_SUB(SDValue Op, SelectionDAG &DAG);
643 SDValue LowerREADCYCLECOUNTER(SDValue Op, SelectionDAG &DAG);
646 LowerFormalArguments(SDValue Chain,
647 unsigned CallConv, bool isVarArg,
648 const SmallVectorImpl<ISD::InputArg> &Ins,
649 DebugLoc dl, SelectionDAG &DAG,
650 SmallVectorImpl<SDValue> &InVals);
652 LowerCall(SDValue Chain, SDValue Callee,
653 unsigned CallConv, bool isVarArg, bool isTailCall,
654 const SmallVectorImpl<ISD::OutputArg> &Outs,
655 const SmallVectorImpl<ISD::InputArg> &Ins,
656 DebugLoc dl, SelectionDAG &DAG,
657 SmallVectorImpl<SDValue> &InVals);
660 LowerReturn(SDValue Chain,
661 unsigned CallConv, bool isVarArg,
662 const SmallVectorImpl<ISD::OutputArg> &Outs,
663 DebugLoc dl, SelectionDAG &DAG);
665 void ReplaceATOMIC_BINARY_64(SDNode *N, SmallVectorImpl<SDValue> &Results,
666 SelectionDAG &DAG, unsigned NewOp);
668 SDValue EmitTargetCodeForMemset(SelectionDAG &DAG, DebugLoc dl,
670 SDValue Dst, SDValue Src,
671 SDValue Size, unsigned Align,
672 const Value *DstSV, uint64_t DstSVOff);
673 SDValue EmitTargetCodeForMemcpy(SelectionDAG &DAG, DebugLoc dl,
675 SDValue Dst, SDValue Src,
676 SDValue Size, unsigned Align,
678 const Value *DstSV, uint64_t DstSVOff,
679 const Value *SrcSV, uint64_t SrcSVOff);
681 /// Utility function to emit atomic bitwise operations (and, or, xor).
682 // It takes the bitwise instruction to expand, the associated machine basic
683 // block, and the associated X86 opcodes for reg/reg and reg/imm.
684 MachineBasicBlock *EmitAtomicBitwiseWithCustomInserter(
685 MachineInstr *BInstr,
686 MachineBasicBlock *BB,
694 TargetRegisterClass *RC,
695 bool invSrc = false) const;
697 MachineBasicBlock *EmitAtomicBit6432WithCustomInserter(
698 MachineInstr *BInstr,
699 MachineBasicBlock *BB,
704 bool invSrc = false) const;
706 /// Utility function to emit atomic min and max. It takes the min/max
707 /// instruction to expand, the associated basic block, and the associated
708 /// cmov opcode for moving the min or max value.
709 MachineBasicBlock *EmitAtomicMinMaxWithCustomInserter(MachineInstr *BInstr,
710 MachineBasicBlock *BB,
711 unsigned cmovOpc) const;
713 /// Emit nodes that will be selected as "test Op0,Op0", or something
714 /// equivalent, for use with the given x86 condition code.
715 SDValue EmitTest(SDValue Op0, unsigned X86CC, SelectionDAG &DAG);
717 /// Emit nodes that will be selected as "cmp Op0,Op1", or something
718 /// equivalent, for use with the given x86 condition code.
719 SDValue EmitCmp(SDValue Op0, SDValue Op1, unsigned X86CC,
724 FastISel *createFastISel(MachineFunction &mf,
725 MachineModuleInfo *mmi, DwarfWriter *dw,
726 DenseMap<const Value *, unsigned> &,
727 DenseMap<const BasicBlock *, MachineBasicBlock *> &,
728 DenseMap<const AllocaInst *, int> &
730 , SmallSet<Instruction*, 8> &
736 #endif // X86ISELLOWERING_H