1 //===-- SystemZISelLowering.h - SystemZ DAG lowering interface --*- C++ -*-===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file defines the interfaces that SystemZ uses to lower LLVM code into a
11 // selection DAG.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #ifndef LLVM_LIB_TARGET_SYSTEMZ_SYSTEMZISELLOWERING_H
16 #define LLVM_LIB_TARGET_SYSTEMZ_SYSTEMZISELLOWERING_H
17 
18 #include "SystemZ.h"
19 #include "llvm/CodeGen/MachineBasicBlock.h"
20 #include "llvm/CodeGen/SelectionDAG.h"
21 #include "llvm/Target/TargetLowering.h"
22 
23 namespace llvm {
24 namespace SystemZISD {
25 enum NodeType : unsigned {
26   FIRST_NUMBER = ISD::BUILTIN_OP_END,
27 
28   // Return with a flag operand.  Operand 0 is the chain operand.
29   RET_FLAG,
30 
31   // Calls a function.  Operand 0 is the chain operand and operand 1
32   // is the target address.  The arguments start at operand 2.
33   // There is an optional glue operand at the end.
34   CALL,
35   SIBCALL,
36 
37   // TLS calls.  Like regular calls, except operand 1 is the TLS symbol.
38   // (The call target is implicitly __tls_get_offset.)
39   TLS_GDCALL,
40   TLS_LDCALL,
41 
42   // Wraps a TargetGlobalAddress that should be loaded using PC-relative
43   // accesses (LARL).  Operand 0 is the address.
44   PCREL_WRAPPER,
45 
46   // Used in cases where an offset is applied to a TargetGlobalAddress.
47   // Operand 0 is the full TargetGlobalAddress and operand 1 is a
48   // PCREL_WRAPPER for an anchor point.  This is used so that we can
49   // cheaply refer to either the full address or the anchor point
50   // as a register base.
51   PCREL_OFFSET,
52 
53   // Integer absolute.
54   IABS,
55 
56   // Integer comparisons.  There are three operands: the two values
57   // to compare, and an integer of type SystemZICMP.
58   ICMP,
59 
60   // Floating-point comparisons.  The two operands are the values to compare.
61   FCMP,
62 
63   // Test under mask.  The first operand is ANDed with the second operand
64   // and the condition codes are set on the result.  The third operand is
65   // a boolean that is true if the condition codes need to distinguish
66   // between CCMASK_TM_MIXED_MSB_0 and CCMASK_TM_MIXED_MSB_1 (which the
67   // register forms do but the memory forms don't).
68   TM,
69 
70   // Branches if a condition is true.  Operand 0 is the chain operand;
71   // operand 1 is the 4-bit condition-code mask, with bit N in
72   // big-endian order meaning "branch if CC=N"; operand 2 is the
73   // target block and operand 3 is the flag operand.
74   BR_CCMASK,
75 
76   // Selects between operand 0 and operand 1.  Operand 2 is the
77   // mask of condition-code values for which operand 0 should be
78   // chosen over operand 1; it has the same form as BR_CCMASK.
79   // Operand 3 is the flag operand.
80   SELECT_CCMASK,
81 
82   // Evaluates to the gap between the stack pointer and the
83   // base of the dynamically-allocatable area.
84   ADJDYNALLOC,
85 
86   // Extracts the value of a 32-bit access register.  Operand 0 is
87   // the number of the register.
88   EXTRACT_ACCESS,
89 
90   // Count number of bits set in operand 0 per byte.
91   POPCNT,
92 
93   // Wrappers around the ISD opcodes of the same name.  The output and
94   // first input operands are GR128s.  The trailing numbers are the
95   // widths of the second operand in bits.
96   UMUL_LOHI64,
97   SDIVREM32,
98   SDIVREM64,
99   UDIVREM32,
100   UDIVREM64,
101 
102   // Use a series of MVCs to copy bytes from one memory location to another.
103   // The operands are:
104   // - the target address
105   // - the source address
106   // - the constant length
107   //
108   // This isn't a memory opcode because we'd need to attach two
109   // MachineMemOperands rather than one.
110   MVC,
111 
112   // Like MVC, but implemented as a loop that handles X*256 bytes
113   // followed by straight-line code to handle the rest (if any).
114   // The value of X is passed as an additional operand.
115   MVC_LOOP,
116 
117   // Similar to MVC and MVC_LOOP, but for logic operations (AND, OR, XOR).
118   NC,
119   NC_LOOP,
120   OC,
121   OC_LOOP,
122   XC,
123   XC_LOOP,
124 
125   // Use CLC to compare two blocks of memory, with the same comments
126   // as for MVC and MVC_LOOP.
127   CLC,
128   CLC_LOOP,
129 
130   // Use an MVST-based sequence to implement stpcpy().
131   STPCPY,
132 
133   // Use a CLST-based sequence to implement strcmp().  The two input operands
134   // are the addresses of the strings to compare.
135   STRCMP,
136 
137   // Use an SRST-based sequence to search a block of memory.  The first
138   // operand is the end address, the second is the start, and the third
139   // is the character to search for.  CC is set to 1 on success and 2
140   // on failure.
141   SEARCH_STRING,
142 
143   // Store the CC value in bits 29 and 28 of an integer.
144   IPM,
145 
146   // Perform a serialization operation.  (BCR 15,0 or BCR 14,0.)
147   SERIALIZE,
148 
149   // Compiler barrier only; generate a no-op.
150   MEMBARRIER,
151 
152   // Transaction begin.  The first operand is the chain, the second
153   // the TDB pointer, and the third the immediate control field.
154   // Returns chain and glue.
155   TBEGIN,
156   TBEGIN_NOFLOAT,
157 
158   // Transaction end.  Just the chain operand.  Returns chain and glue.
159   TEND,
160 
161   // Create a vector constant by filling byte N of the result with bit
162   // 15-N of the single operand.
163   BYTE_MASK,
164 
165   // Create a vector constant by replicating an element-sized RISBG-style mask.
166   // The first operand specifies the starting set bit and the second operand
167   // specifies the ending set bit.  Both operands count from the MSB of the
168   // element.
169   ROTATE_MASK,
170 
171   // Replicate a GPR scalar value into all elements of a vector.
172   REPLICATE,
173 
174   // Create a vector from two i64 GPRs.
175   JOIN_DWORDS,
176 
177   // Replicate one element of a vector into all elements.  The first operand
178   // is the vector and the second is the index of the element to replicate.
179   SPLAT,
180 
181   // Interleave elements from the high half of operand 0 and the high half
182   // of operand 1.
183   MERGE_HIGH,
184 
185   // Likewise for the low halves.
186   MERGE_LOW,
187 
188   // Concatenate the vectors in the first two operands, shift them left
189   // by the third operand, and take the first half of the result.
190   SHL_DOUBLE,
191 
192   // Take one element of the first v2i64 operand and the one element of
193   // the second v2i64 operand and concatenate them to form a v2i64 result.
194   // The third operand is a 4-bit value of the form 0A0B, where A and B
195   // are the element selectors for the first operand and second operands
196   // respectively.
197   PERMUTE_DWORDS,
198 
199   // Perform a general vector permute on vector operands 0 and 1.
200   // Each byte of operand 2 controls the corresponding byte of the result,
201   // in the same way as a byte-level VECTOR_SHUFFLE mask.
202   PERMUTE,
203 
204   // Pack vector operands 0 and 1 into a single vector with half-sized elements.
205   PACK,
206 
207   // Likewise, but saturate the result and set CC.  PACKS_CC does signed
208   // saturation and PACKLS_CC does unsigned saturation.
209   PACKS_CC,
210   PACKLS_CC,
211 
212   // Unpack the first half of vector operand 0 into double-sized elements.
213   // UNPACK_HIGH sign-extends and UNPACKL_HIGH zero-extends.
214   UNPACK_HIGH,
215   UNPACKL_HIGH,
216 
217   // Likewise for the second half.
218   UNPACK_LOW,
219   UNPACKL_LOW,
220 
221   // Shift each element of vector operand 0 by the number of bits specified
222   // by scalar operand 1.
223   VSHL_BY_SCALAR,
224   VSRL_BY_SCALAR,
225   VSRA_BY_SCALAR,
226 
227   // For each element of the output type, sum across all sub-elements of
228   // operand 0 belonging to the corresponding element, and add in the
229   // rightmost sub-element of the corresponding element of operand 1.
230   VSUM,
231 
232   // Compare integer vector operands 0 and 1 to produce the usual 0/-1
233   // vector result.  VICMPE is for equality, VICMPH for "signed greater than"
234   // and VICMPHL for "unsigned greater than".
235   VICMPE,
236   VICMPH,
237   VICMPHL,
238 
239   // Likewise, but also set the condition codes on the result.
240   VICMPES,
241   VICMPHS,
242   VICMPHLS,
243 
244   // Compare floating-point vector operands 0 and 1 to preoduce the usual 0/-1
245   // vector result.  VFCMPE is for "ordered and equal", VFCMPH for "ordered and
246   // greater than" and VFCMPHE for "ordered and greater than or equal to".
247   VFCMPE,
248   VFCMPH,
249   VFCMPHE,
250 
251   // Likewise, but also set the condition codes on the result.
252   VFCMPES,
253   VFCMPHS,
254   VFCMPHES,
255 
256   // Test floating-point data class for vectors.
257   VFTCI,
258 
259   // Extend the even f32 elements of vector operand 0 to produce a vector
260   // of f64 elements.
261   VEXTEND,
262 
263   // Round the f64 elements of vector operand 0 to f32s and store them in the
264   // even elements of the result.
265   VROUND,
266 
267   // AND the two vector operands together and set CC based on the result.
268   VTM,
269 
270   // String operations that set CC as a side-effect.
271   VFAE_CC,
272   VFAEZ_CC,
273   VFEE_CC,
274   VFEEZ_CC,
275   VFENE_CC,
276   VFENEZ_CC,
277   VISTR_CC,
278   VSTRC_CC,
279   VSTRCZ_CC,
280 
281   // Wrappers around the inner loop of an 8- or 16-bit ATOMIC_SWAP or
282   // ATOMIC_LOAD_<op>.
283   //
284   // Operand 0: the address of the containing 32-bit-aligned field
285   // Operand 1: the second operand of <op>, in the high bits of an i32
286   //            for everything except ATOMIC_SWAPW
287   // Operand 2: how many bits to rotate the i32 left to bring the first
288   //            operand into the high bits
289   // Operand 3: the negative of operand 2, for rotating the other way
290   // Operand 4: the width of the field in bits (8 or 16)
291   ATOMIC_SWAPW = ISD::FIRST_TARGET_MEMORY_OPCODE,
292   ATOMIC_LOADW_ADD,
293   ATOMIC_LOADW_SUB,
294   ATOMIC_LOADW_AND,
295   ATOMIC_LOADW_OR,
296   ATOMIC_LOADW_XOR,
297   ATOMIC_LOADW_NAND,
298   ATOMIC_LOADW_MIN,
299   ATOMIC_LOADW_MAX,
300   ATOMIC_LOADW_UMIN,
301   ATOMIC_LOADW_UMAX,
302 
303   // A wrapper around the inner loop of an ATOMIC_CMP_SWAP.
304   //
305   // Operand 0: the address of the containing 32-bit-aligned field
306   // Operand 1: the compare value, in the low bits of an i32
307   // Operand 2: the swap value, in the low bits of an i32
308   // Operand 3: how many bits to rotate the i32 left to bring the first
309   //            operand into the high bits
310   // Operand 4: the negative of operand 2, for rotating the other way
311   // Operand 5: the width of the field in bits (8 or 16)
312   ATOMIC_CMP_SWAPW,
313 
314   // Prefetch from the second operand using the 4-bit control code in
315   // the first operand.  The code is 1 for a load prefetch and 2 for
316   // a store prefetch.
317   PREFETCH
318 };
319 
320 // Return true if OPCODE is some kind of PC-relative address.
321 inline bool isPCREL(unsigned Opcode) {
322   return Opcode == PCREL_WRAPPER || Opcode == PCREL_OFFSET;
323 }
324 } // end namespace SystemZISD
325 
326 namespace SystemZICMP {
327 // Describes whether an integer comparison needs to be signed or unsigned,
328 // or whether either type is OK.
329 enum {
330   Any,
331   UnsignedOnly,
332   SignedOnly
333 };
334 } // end namespace SystemZICMP
335 
336 class SystemZSubtarget;
337 class SystemZTargetMachine;
338 
339 class SystemZTargetLowering : public TargetLowering {
340 public:
341   explicit SystemZTargetLowering(const TargetMachine &TM,
342                                  const SystemZSubtarget &STI);
343 
344   // Override TargetLowering.
345   MVT getScalarShiftAmountTy(const DataLayout &, EVT) const override {
346     return MVT::i32;
347   }
348   MVT getVectorIdxTy(const DataLayout &DL) const override {
349     // Only the lower 12 bits of an element index are used, so we don't
350     // want to clobber the upper 32 bits of a GPR unnecessarily.
351     return MVT::i32;
352   }
353   TargetLoweringBase::LegalizeTypeAction getPreferredVectorAction(EVT VT)
354     const override {
355     // Widen subvectors to the full width rather than promoting integer
356     // elements.  This is better because:
357     //
358     // (a) it means that we can handle the ABI for passing and returning
359     //     sub-128 vectors without having to handle them as legal types.
360     //
361     // (b) we don't have instructions to extend on load and truncate on store,
362     //     so promoting the integers is less efficient.
363     //
364     // (c) there are no multiplication instructions for the widest integer
365     //     type (v2i64).
366     if (VT.getVectorElementType().getSizeInBits() % 8 == 0)
367       return TypeWidenVector;
368     return TargetLoweringBase::getPreferredVectorAction(VT);
369   }
370   EVT getSetCCResultType(const DataLayout &DL, LLVMContext &,
371                          EVT) const override;
372   bool isFMAFasterThanFMulAndFAdd(EVT VT) const override;
373   bool isFPImmLegal(const APFloat &Imm, EVT VT) const override;
374   bool isLegalICmpImmediate(int64_t Imm) const override;
375   bool isLegalAddImmediate(int64_t Imm) const override;
376   bool isLegalAddressingMode(const DataLayout &DL, const AddrMode &AM, Type *Ty,
377                              unsigned AS) const override;
378   bool allowsMisalignedMemoryAccesses(EVT VT, unsigned AS,
379                                       unsigned Align,
380                                       bool *Fast) const override;
381   bool isTruncateFree(Type *, Type *) const override;
382   bool isTruncateFree(EVT, EVT) const override;
383   const char *getTargetNodeName(unsigned Opcode) const override;
384   std::pair<unsigned, const TargetRegisterClass *>
385   getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI,
386                                StringRef Constraint, MVT VT) const override;
387   TargetLowering::ConstraintType
388   getConstraintType(StringRef Constraint) const override;
389   TargetLowering::ConstraintWeight
390     getSingleConstraintMatchWeight(AsmOperandInfo &info,
391                                    const char *constraint) const override;
392   void LowerAsmOperandForConstraint(SDValue Op,
393                                     std::string &Constraint,
394                                     std::vector<SDValue> &Ops,
395                                     SelectionDAG &DAG) const override;
396 
397   unsigned getInlineAsmMemConstraint(StringRef ConstraintCode) const override {
398     if (ConstraintCode.size() == 1) {
399       switch(ConstraintCode[0]) {
400       default:
401         break;
402       case 'Q':
403         return InlineAsm::Constraint_Q;
404       case 'R':
405         return InlineAsm::Constraint_R;
406       case 'S':
407         return InlineAsm::Constraint_S;
408       case 'T':
409         return InlineAsm::Constraint_T;
410       }
411     }
412     return TargetLowering::getInlineAsmMemConstraint(ConstraintCode);
413   }
414 
415   /// If a physical register, this returns the register that receives the
416   /// exception address on entry to an EH pad.
417   unsigned
418   getExceptionPointerRegister(const Constant *PersonalityFn) const override {
419     return SystemZ::R6D;
420   }
421 
422   /// If a physical register, this returns the register that receives the
423   /// exception typeid on entry to a landing pad.
424   unsigned
425   getExceptionSelectorRegister(const Constant *PersonalityFn) const override {
426     return SystemZ::R7D;
427   }
428 
429   /// Override to support customized stack guard loading.
430   bool useLoadStackGuardNode() const override {
431     return true;
432   }
433   void insertSSPDeclarations(Module &M) const override {
434   }
435 
436   MachineBasicBlock *EmitInstrWithCustomInserter(MachineInstr *MI,
437                                                  MachineBasicBlock *BB) const
438     override;
439   SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override;
440   bool allowTruncateForTailCall(Type *, Type *) const override;
441   bool mayBeEmittedAsTailCall(CallInst *CI) const override;
442   SDValue LowerFormalArguments(SDValue Chain, CallingConv::ID CallConv,
443                                bool isVarArg,
444                                const SmallVectorImpl<ISD::InputArg> &Ins,
445                                SDLoc DL, SelectionDAG &DAG,
446                                SmallVectorImpl<SDValue> &InVals) const override;
447   SDValue LowerCall(CallLoweringInfo &CLI,
448                     SmallVectorImpl<SDValue> &InVals) const override;
449 
450   bool CanLowerReturn(CallingConv::ID CallConv, MachineFunction &MF,
451                       bool isVarArg,
452                       const SmallVectorImpl<ISD::OutputArg> &Outs,
453                       LLVMContext &Context) const override;
454   SDValue LowerReturn(SDValue Chain, CallingConv::ID CallConv, bool IsVarArg,
455                       const SmallVectorImpl<ISD::OutputArg> &Outs,
456                       const SmallVectorImpl<SDValue> &OutVals,
457                       SDLoc DL, SelectionDAG &DAG) const override;
458   SDValue prepareVolatileOrAtomicLoad(SDValue Chain, SDLoc DL,
459                                       SelectionDAG &DAG) const override;
460   SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
461 
462   bool supportSwiftError() const override {
463     return true;
464   }
465 
466 private:
467   const SystemZSubtarget &Subtarget;
468 
469   // Implement LowerOperation for individual opcodes.
470   SDValue lowerSETCC(SDValue Op, SelectionDAG &DAG) const;
471   SDValue lowerBR_CC(SDValue Op, SelectionDAG &DAG) const;
472   SDValue lowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const;
473   SDValue lowerGlobalAddress(GlobalAddressSDNode *Node,
474                              SelectionDAG &DAG) const;
475   SDValue lowerTLSGetOffset(GlobalAddressSDNode *Node,
476                             SelectionDAG &DAG, unsigned Opcode,
477                             SDValue GOTOffset) const;
478   SDValue lowerThreadPointer(const SDLoc &DL, SelectionDAG &DAG) const;
479   SDValue lowerGlobalTLSAddress(GlobalAddressSDNode *Node,
480                                 SelectionDAG &DAG) const;
481   SDValue lowerBlockAddress(BlockAddressSDNode *Node,
482                             SelectionDAG &DAG) const;
483   SDValue lowerJumpTable(JumpTableSDNode *JT, SelectionDAG &DAG) const;
484   SDValue lowerConstantPool(ConstantPoolSDNode *CP, SelectionDAG &DAG) const;
485   SDValue lowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
486   SDValue lowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const;
487   SDValue lowerVASTART(SDValue Op, SelectionDAG &DAG) const;
488   SDValue lowerVACOPY(SDValue Op, SelectionDAG &DAG) const;
489   SDValue lowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG) const;
490   SDValue lowerSMUL_LOHI(SDValue Op, SelectionDAG &DAG) const;
491   SDValue lowerUMUL_LOHI(SDValue Op, SelectionDAG &DAG) const;
492   SDValue lowerSDIVREM(SDValue Op, SelectionDAG &DAG) const;
493   SDValue lowerUDIVREM(SDValue Op, SelectionDAG &DAG) const;
494   SDValue lowerBITCAST(SDValue Op, SelectionDAG &DAG) const;
495   SDValue lowerOR(SDValue Op, SelectionDAG &DAG) const;
496   SDValue lowerCTPOP(SDValue Op, SelectionDAG &DAG) const;
497   SDValue lowerATOMIC_FENCE(SDValue Op, SelectionDAG &DAG) const;
498   SDValue lowerATOMIC_LOAD(SDValue Op, SelectionDAG &DAG) const;
499   SDValue lowerATOMIC_STORE(SDValue Op, SelectionDAG &DAG) const;
500   SDValue lowerATOMIC_LOAD_OP(SDValue Op, SelectionDAG &DAG,
501                               unsigned Opcode) const;
502   SDValue lowerATOMIC_LOAD_SUB(SDValue Op, SelectionDAG &DAG) const;
503   SDValue lowerATOMIC_CMP_SWAP(SDValue Op, SelectionDAG &DAG) const;
504   SDValue lowerLOAD_SEQUENCE_POINT(SDValue Op, SelectionDAG &DAG) const;
505   SDValue lowerSTACKSAVE(SDValue Op, SelectionDAG &DAG) const;
506   SDValue lowerSTACKRESTORE(SDValue Op, SelectionDAG &DAG) const;
507   SDValue lowerPREFETCH(SDValue Op, SelectionDAG &DAG) const;
508   SDValue lowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) const;
509   SDValue lowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
510   SDValue lowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const;
511   SDValue lowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const;
512   SDValue lowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) const;
513   SDValue lowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
514   SDValue lowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
515   SDValue lowerExtendVectorInreg(SDValue Op, SelectionDAG &DAG,
516                                  unsigned UnpackHigh) const;
517   SDValue lowerShift(SDValue Op, SelectionDAG &DAG, unsigned ByScalar) const;
518 
519   SDValue combineExtract(SDLoc DL, EVT ElemVT, EVT VecVT, SDValue OrigOp,
520                          unsigned Index, DAGCombinerInfo &DCI,
521                          bool Force) const;
522   SDValue combineTruncateExtract(SDLoc DL, EVT TruncVT, SDValue Op,
523                                  DAGCombinerInfo &DCI) const;
524 
525   // If the last instruction before MBBI in MBB was some form of COMPARE,
526   // try to replace it with a COMPARE AND BRANCH just before MBBI.
527   // CCMask and Target are the BRC-like operands for the branch.
528   // Return true if the change was made.
529   bool convertPrevCompareToBranch(MachineBasicBlock *MBB,
530                                   MachineBasicBlock::iterator MBBI,
531                                   unsigned CCMask,
532                                   MachineBasicBlock *Target) const;
533 
534   // Implement EmitInstrWithCustomInserter for individual operation types.
535   MachineBasicBlock *emitSelect(MachineInstr *MI,
536                                 MachineBasicBlock *BB) const;
537   MachineBasicBlock *emitCondStore(MachineInstr *MI,
538                                    MachineBasicBlock *BB,
539                                    unsigned StoreOpcode, unsigned STOCOpcode,
540                                    bool Invert) const;
541   MachineBasicBlock *emitExt128(MachineInstr *MI,
542                                 MachineBasicBlock *MBB,
543                                 bool ClearEven, unsigned SubReg) const;
544   MachineBasicBlock *emitAtomicLoadBinary(MachineInstr *MI,
545                                           MachineBasicBlock *BB,
546                                           unsigned BinOpcode, unsigned BitSize,
547                                           bool Invert = false) const;
548   MachineBasicBlock *emitAtomicLoadMinMax(MachineInstr *MI,
549                                           MachineBasicBlock *MBB,
550                                           unsigned CompareOpcode,
551                                           unsigned KeepOldMask,
552                                           unsigned BitSize) const;
553   MachineBasicBlock *emitAtomicCmpSwapW(MachineInstr *MI,
554                                         MachineBasicBlock *BB) const;
555   MachineBasicBlock *emitMemMemWrapper(MachineInstr *MI,
556                                        MachineBasicBlock *BB,
557                                        unsigned Opcode) const;
558   MachineBasicBlock *emitStringWrapper(MachineInstr *MI,
559                                        MachineBasicBlock *BB,
560                                        unsigned Opcode) const;
561   MachineBasicBlock *emitTransactionBegin(MachineInstr *MI,
562                                           MachineBasicBlock *MBB,
563                                           unsigned Opcode,
564                                           bool NoFloat) const;
565   MachineBasicBlock *emitLoadAndTestCmp0(MachineInstr *MI,
566                                          MachineBasicBlock *MBB,
567                                          unsigned Opcode) const;
568 
569 };
570 } // end namespace llvm
571 
572 #endif
573