LLVM API Documentation

 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages
X86AsmParser.cpp
Go to the documentation of this file.
1 //===-- X86AsmParser.cpp - Parse X86 assembly to MCInst instructions ------===//
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 
11 #include "llvm/ADT/APFloat.h"
12 #include "llvm/ADT/STLExtras.h"
13 #include "llvm/ADT/SmallString.h"
14 #include "llvm/ADT/SmallVector.h"
15 #include "llvm/ADT/StringSwitch.h"
16 #include "llvm/ADT/Twine.h"
17 #include "llvm/MC/MCContext.h"
18 #include "llvm/MC/MCExpr.h"
19 #include "llvm/MC/MCInst.h"
23 #include "llvm/MC/MCRegisterInfo.h"
24 #include "llvm/MC/MCStreamer.h"
26 #include "llvm/MC/MCSymbol.h"
28 #include "llvm/Support/SourceMgr.h"
31 
32 using namespace llvm;
33 
34 namespace {
35 struct X86Operand;
36 
37 static const char OpPrecedence[] = {
38  0, // IC_PLUS
39  0, // IC_MINUS
40  1, // IC_MULTIPLY
41  1, // IC_DIVIDE
42  2, // IC_RPAREN
43  3, // IC_LPAREN
44  0, // IC_IMM
45  0 // IC_REGISTER
46 };
47 
48 class X86AsmParser : public MCTargetAsmParser {
49  MCSubtargetInfo &STI;
50  MCAsmParser &Parser;
51  ParseInstructionInfo *InstInfo;
52 private:
53  enum InfixCalculatorTok {
54  IC_PLUS = 0,
55  IC_MINUS,
56  IC_MULTIPLY,
57  IC_DIVIDE,
58  IC_RPAREN,
59  IC_LPAREN,
60  IC_IMM,
61  IC_REGISTER
62  };
63 
64  class InfixCalculator {
65  typedef std::pair< InfixCalculatorTok, int64_t > ICToken;
66  SmallVector<InfixCalculatorTok, 4> InfixOperatorStack;
67  SmallVector<ICToken, 4> PostfixStack;
68 
69  public:
70  int64_t popOperand() {
71  assert (!PostfixStack.empty() && "Poped an empty stack!");
72  ICToken Op = PostfixStack.pop_back_val();
73  assert ((Op.first == IC_IMM || Op.first == IC_REGISTER)
74  && "Expected and immediate or register!");
75  return Op.second;
76  }
77  void pushOperand(InfixCalculatorTok Op, int64_t Val = 0) {
78  assert ((Op == IC_IMM || Op == IC_REGISTER) &&
79  "Unexpected operand!");
80  PostfixStack.push_back(std::make_pair(Op, Val));
81  }
82 
83  void popOperator() { InfixOperatorStack.pop_back(); }
84  void pushOperator(InfixCalculatorTok Op) {
85  // Push the new operator if the stack is empty.
86  if (InfixOperatorStack.empty()) {
87  InfixOperatorStack.push_back(Op);
88  return;
89  }
90 
91  // Push the new operator if it has a higher precedence than the operator
92  // on the top of the stack or the operator on the top of the stack is a
93  // left parentheses.
94  unsigned Idx = InfixOperatorStack.size() - 1;
95  InfixCalculatorTok StackOp = InfixOperatorStack[Idx];
96  if (OpPrecedence[Op] > OpPrecedence[StackOp] || StackOp == IC_LPAREN) {
97  InfixOperatorStack.push_back(Op);
98  return;
99  }
100 
101  // The operator on the top of the stack has higher precedence than the
102  // new operator.
103  unsigned ParenCount = 0;
104  while (1) {
105  // Nothing to process.
106  if (InfixOperatorStack.empty())
107  break;
108 
109  Idx = InfixOperatorStack.size() - 1;
110  StackOp = InfixOperatorStack[Idx];
111  if (!(OpPrecedence[StackOp] >= OpPrecedence[Op] || ParenCount))
112  break;
113 
114  // If we have an even parentheses count and we see a left parentheses,
115  // then stop processing.
116  if (!ParenCount && StackOp == IC_LPAREN)
117  break;
118 
119  if (StackOp == IC_RPAREN) {
120  ++ParenCount;
121  InfixOperatorStack.pop_back();
122  } else if (StackOp == IC_LPAREN) {
123  --ParenCount;
124  InfixOperatorStack.pop_back();
125  } else {
126  InfixOperatorStack.pop_back();
127  PostfixStack.push_back(std::make_pair(StackOp, 0));
128  }
129  }
130  // Push the new operator.
131  InfixOperatorStack.push_back(Op);
132  }
133  int64_t execute() {
134  // Push any remaining operators onto the postfix stack.
135  while (!InfixOperatorStack.empty()) {
136  InfixCalculatorTok StackOp = InfixOperatorStack.pop_back_val();
137  if (StackOp != IC_LPAREN && StackOp != IC_RPAREN)
138  PostfixStack.push_back(std::make_pair(StackOp, 0));
139  }
140 
141  if (PostfixStack.empty())
142  return 0;
143 
144  SmallVector<ICToken, 16> OperandStack;
145  for (unsigned i = 0, e = PostfixStack.size(); i != e; ++i) {
146  ICToken Op = PostfixStack[i];
147  if (Op.first == IC_IMM || Op.first == IC_REGISTER) {
148  OperandStack.push_back(Op);
149  } else {
150  assert (OperandStack.size() > 1 && "Too few operands.");
151  int64_t Val;
152  ICToken Op2 = OperandStack.pop_back_val();
153  ICToken Op1 = OperandStack.pop_back_val();
154  switch (Op.first) {
155  default:
156  report_fatal_error("Unexpected operator!");
157  break;
158  case IC_PLUS:
159  Val = Op1.second + Op2.second;
160  OperandStack.push_back(std::make_pair(IC_IMM, Val));
161  break;
162  case IC_MINUS:
163  Val = Op1.second - Op2.second;
164  OperandStack.push_back(std::make_pair(IC_IMM, Val));
165  break;
166  case IC_MULTIPLY:
167  assert (Op1.first == IC_IMM && Op2.first == IC_IMM &&
168  "Multiply operation with an immediate and a register!");
169  Val = Op1.second * Op2.second;
170  OperandStack.push_back(std::make_pair(IC_IMM, Val));
171  break;
172  case IC_DIVIDE:
173  assert (Op1.first == IC_IMM && Op2.first == IC_IMM &&
174  "Divide operation with an immediate and a register!");
175  assert (Op2.second != 0 && "Division by zero!");
176  Val = Op1.second / Op2.second;
177  OperandStack.push_back(std::make_pair(IC_IMM, Val));
178  break;
179  }
180  }
181  }
182  assert (OperandStack.size() == 1 && "Expected a single result.");
183  return OperandStack.pop_back_val().second;
184  }
185  };
186 
187  enum IntelExprState {
188  IES_PLUS,
189  IES_MINUS,
190  IES_MULTIPLY,
191  IES_DIVIDE,
192  IES_LBRAC,
193  IES_RBRAC,
194  IES_LPAREN,
195  IES_RPAREN,
196  IES_REGISTER,
197  IES_INTEGER,
198  IES_IDENTIFIER,
199  IES_ERROR
200  };
201 
202  class IntelExprStateMachine {
203  IntelExprState State, PrevState;
204  unsigned BaseReg, IndexReg, TmpReg, Scale;
205  int64_t Imm;
206  const MCExpr *Sym;
207  StringRef SymName;
208  bool StopOnLBrac, AddImmPrefix;
209  InfixCalculator IC;
211  public:
212  IntelExprStateMachine(int64_t imm, bool stoponlbrac, bool addimmprefix) :
213  State(IES_PLUS), PrevState(IES_ERROR), BaseReg(0), IndexReg(0), TmpReg(0),
214  Scale(1), Imm(imm), Sym(0), StopOnLBrac(stoponlbrac),
215  AddImmPrefix(addimmprefix) { Info.clear(); }
216 
217  unsigned getBaseReg() { return BaseReg; }
218  unsigned getIndexReg() { return IndexReg; }
219  unsigned getScale() { return Scale; }
220  const MCExpr *getSym() { return Sym; }
221  StringRef getSymName() { return SymName; }
222  int64_t getImm() { return Imm + IC.execute(); }
223  bool isValidEndState() {
224  return State == IES_RBRAC || State == IES_INTEGER;
225  }
226  bool getStopOnLBrac() { return StopOnLBrac; }
227  bool getAddImmPrefix() { return AddImmPrefix; }
228  bool hadError() { return State == IES_ERROR; }
229 
230  InlineAsmIdentifierInfo &getIdentifierInfo() {
231  return Info;
232  }
233 
234  void onPlus() {
235  IntelExprState CurrState = State;
236  switch (State) {
237  default:
238  State = IES_ERROR;
239  break;
240  case IES_INTEGER:
241  case IES_RPAREN:
242  case IES_REGISTER:
243  State = IES_PLUS;
244  IC.pushOperator(IC_PLUS);
245  if (CurrState == IES_REGISTER && PrevState != IES_MULTIPLY) {
246  // If we already have a BaseReg, then assume this is the IndexReg with
247  // a scale of 1.
248  if (!BaseReg) {
249  BaseReg = TmpReg;
250  } else {
251  assert (!IndexReg && "BaseReg/IndexReg already set!");
252  IndexReg = TmpReg;
253  Scale = 1;
254  }
255  }
256  break;
257  }
258  PrevState = CurrState;
259  }
260  void onMinus() {
261  IntelExprState CurrState = State;
262  switch (State) {
263  default:
264  State = IES_ERROR;
265  break;
266  case IES_PLUS:
267  case IES_MULTIPLY:
268  case IES_DIVIDE:
269  case IES_LPAREN:
270  case IES_RPAREN:
271  case IES_LBRAC:
272  case IES_RBRAC:
273  case IES_INTEGER:
274  case IES_REGISTER:
275  State = IES_MINUS;
276  // Only push the minus operator if it is not a unary operator.
277  if (!(CurrState == IES_PLUS || CurrState == IES_MINUS ||
278  CurrState == IES_MULTIPLY || CurrState == IES_DIVIDE ||
279  CurrState == IES_LPAREN || CurrState == IES_LBRAC))
280  IC.pushOperator(IC_MINUS);
281  if (CurrState == IES_REGISTER && PrevState != IES_MULTIPLY) {
282  // If we already have a BaseReg, then assume this is the IndexReg with
283  // a scale of 1.
284  if (!BaseReg) {
285  BaseReg = TmpReg;
286  } else {
287  assert (!IndexReg && "BaseReg/IndexReg already set!");
288  IndexReg = TmpReg;
289  Scale = 1;
290  }
291  }
292  break;
293  }
294  PrevState = CurrState;
295  }
296  void onRegister(unsigned Reg) {
297  IntelExprState CurrState = State;
298  switch (State) {
299  default:
300  State = IES_ERROR;
301  break;
302  case IES_PLUS:
303  case IES_LPAREN:
304  State = IES_REGISTER;
305  TmpReg = Reg;
306  IC.pushOperand(IC_REGISTER);
307  break;
308  case IES_MULTIPLY:
309  // Index Register - Scale * Register
310  if (PrevState == IES_INTEGER) {
311  assert (!IndexReg && "IndexReg already set!");
312  State = IES_REGISTER;
313  IndexReg = Reg;
314  // Get the scale and replace the 'Scale * Register' with '0'.
315  Scale = IC.popOperand();
316  IC.pushOperand(IC_IMM);
317  IC.popOperator();
318  } else {
319  State = IES_ERROR;
320  }
321  break;
322  }
323  PrevState = CurrState;
324  }
325  void onIdentifierExpr(const MCExpr *SymRef, StringRef SymRefName) {
326  PrevState = State;
327  switch (State) {
328  default:
329  State = IES_ERROR;
330  break;
331  case IES_PLUS:
332  case IES_MINUS:
333  State = IES_INTEGER;
334  Sym = SymRef;
335  SymName = SymRefName;
336  IC.pushOperand(IC_IMM);
337  break;
338  }
339  }
340  void onInteger(int64_t TmpInt) {
341  IntelExprState CurrState = State;
342  switch (State) {
343  default:
344  State = IES_ERROR;
345  break;
346  case IES_PLUS:
347  case IES_MINUS:
348  case IES_DIVIDE:
349  case IES_MULTIPLY:
350  case IES_LPAREN:
351  State = IES_INTEGER;
352  if (PrevState == IES_REGISTER && CurrState == IES_MULTIPLY) {
353  // Index Register - Register * Scale
354  assert (!IndexReg && "IndexReg already set!");
355  IndexReg = TmpReg;
356  Scale = TmpInt;
357  // Get the scale and replace the 'Register * Scale' with '0'.
358  IC.popOperator();
359  } else if ((PrevState == IES_PLUS || PrevState == IES_MINUS ||
360  PrevState == IES_MULTIPLY || PrevState == IES_DIVIDE ||
361  PrevState == IES_LPAREN || PrevState == IES_LBRAC) &&
362  CurrState == IES_MINUS) {
363  // Unary minus. No need to pop the minus operand because it was never
364  // pushed.
365  IC.pushOperand(IC_IMM, -TmpInt); // Push -Imm.
366  } else {
367  IC.pushOperand(IC_IMM, TmpInt);
368  }
369  break;
370  }
371  PrevState = CurrState;
372  }
373  void onStar() {
374  PrevState = State;
375  switch (State) {
376  default:
377  State = IES_ERROR;
378  break;
379  case IES_INTEGER:
380  case IES_REGISTER:
381  case IES_RPAREN:
382  State = IES_MULTIPLY;
383  IC.pushOperator(IC_MULTIPLY);
384  break;
385  }
386  }
387  void onDivide() {
388  PrevState = State;
389  switch (State) {
390  default:
391  State = IES_ERROR;
392  break;
393  case IES_INTEGER:
394  case IES_RPAREN:
395  State = IES_DIVIDE;
396  IC.pushOperator(IC_DIVIDE);
397  break;
398  }
399  }
400  void onLBrac() {
401  PrevState = State;
402  switch (State) {
403  default:
404  State = IES_ERROR;
405  break;
406  case IES_RBRAC:
407  State = IES_PLUS;
408  IC.pushOperator(IC_PLUS);
409  break;
410  }
411  }
412  void onRBrac() {
413  IntelExprState CurrState = State;
414  switch (State) {
415  default:
416  State = IES_ERROR;
417  break;
418  case IES_INTEGER:
419  case IES_REGISTER:
420  case IES_RPAREN:
421  State = IES_RBRAC;
422  if (CurrState == IES_REGISTER && PrevState != IES_MULTIPLY) {
423  // If we already have a BaseReg, then assume this is the IndexReg with
424  // a scale of 1.
425  if (!BaseReg) {
426  BaseReg = TmpReg;
427  } else {
428  assert (!IndexReg && "BaseReg/IndexReg already set!");
429  IndexReg = TmpReg;
430  Scale = 1;
431  }
432  }
433  break;
434  }
435  PrevState = CurrState;
436  }
437  void onLParen() {
438  IntelExprState CurrState = State;
439  switch (State) {
440  default:
441  State = IES_ERROR;
442  break;
443  case IES_PLUS:
444  case IES_MINUS:
445  case IES_MULTIPLY:
446  case IES_DIVIDE:
447  case IES_LPAREN:
448  // FIXME: We don't handle this type of unary minus, yet.
449  if ((PrevState == IES_PLUS || PrevState == IES_MINUS ||
450  PrevState == IES_MULTIPLY || PrevState == IES_DIVIDE ||
451  PrevState == IES_LPAREN || PrevState == IES_LBRAC) &&
452  CurrState == IES_MINUS) {
453  State = IES_ERROR;
454  break;
455  }
456  State = IES_LPAREN;
457  IC.pushOperator(IC_LPAREN);
458  break;
459  }
460  PrevState = CurrState;
461  }
462  void onRParen() {
463  PrevState = State;
464  switch (State) {
465  default:
466  State = IES_ERROR;
467  break;
468  case IES_INTEGER:
469  case IES_REGISTER:
470  case IES_RPAREN:
471  State = IES_RPAREN;
472  IC.pushOperator(IC_RPAREN);
473  break;
474  }
475  }
476  };
477 
478  MCAsmParser &getParser() const { return Parser; }
479 
480  MCAsmLexer &getLexer() const { return Parser.getLexer(); }
481 
482  bool Error(SMLoc L, const Twine &Msg,
483  ArrayRef<SMRange> Ranges = None,
484  bool MatchingInlineAsm = false) {
485  if (MatchingInlineAsm) return true;
486  return Parser.Error(L, Msg, Ranges);
487  }
488 
489  X86Operand *ErrorOperand(SMLoc Loc, StringRef Msg) {
490  Error(Loc, Msg);
491  return 0;
492  }
493 
494  X86Operand *ParseOperand();
495  X86Operand *ParseATTOperand();
496  X86Operand *ParseIntelOperand();
497  X86Operand *ParseIntelOffsetOfOperator();
498  X86Operand *ParseIntelDotOperator(const MCExpr *Disp, const MCExpr *&NewDisp);
499  X86Operand *ParseIntelOperator(unsigned OpKind);
500  X86Operand *ParseIntelSegmentOverride(unsigned SegReg, SMLoc Start, unsigned Size);
501  X86Operand *ParseIntelMemOperand(int64_t ImmDisp, SMLoc StartLoc,
502  unsigned Size);
503  X86Operand *ParseIntelExpression(IntelExprStateMachine &SM, SMLoc &End);
504  X86Operand *ParseIntelBracExpression(unsigned SegReg, SMLoc Start,
505  int64_t ImmDisp, unsigned Size);
506  X86Operand *ParseIntelIdentifier(const MCExpr *&Val, StringRef &Identifier,
508  bool IsUnevaluatedOperand, SMLoc &End);
509 
510  X86Operand *ParseMemOperand(unsigned SegReg, SMLoc StartLoc);
511 
512  X86Operand *CreateMemForInlineAsm(unsigned SegReg, const MCExpr *Disp,
513  unsigned BaseReg, unsigned IndexReg,
514  unsigned Scale, SMLoc Start, SMLoc End,
515  unsigned Size, StringRef Identifier,
517 
518  bool ParseDirectiveWord(unsigned Size, SMLoc L);
519  bool ParseDirectiveCode(StringRef IDVal, SMLoc L);
520 
521  bool processInstruction(MCInst &Inst,
523 
524  bool MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
526  MCStreamer &Out, unsigned &ErrorInfo,
527  bool MatchingInlineAsm);
528 
529  /// isSrcOp - Returns true if operand is either (%rsi) or %ds:%(rsi)
530  /// in 64bit mode or (%esi) or %es:(%esi) in 32bit mode.
531  bool isSrcOp(X86Operand &Op);
532 
533  /// isDstOp - Returns true if operand is either (%rdi) or %es:(%rdi)
534  /// in 64bit mode or (%edi) or %es:(%edi) in 32bit mode.
535  bool isDstOp(X86Operand &Op);
536 
537  bool is64BitMode() const {
538  // FIXME: Can tablegen auto-generate this?
539  return (STI.getFeatureBits() & X86::Mode64Bit) != 0;
540  }
541  void SwitchMode() {
542  unsigned FB = ComputeAvailableFeatures(STI.ToggleFeature(X86::Mode64Bit));
543  setAvailableFeatures(FB);
544  }
545 
546  bool isParsingIntelSyntax() {
547  return getParser().getAssemblerDialect();
548  }
549 
550  /// @name Auto-generated Matcher Functions
551  /// {
552 
553 #define GET_ASSEMBLER_HEADER
554 #include "X86GenAsmMatcher.inc"
555 
556  /// }
557 
558 public:
559  X86AsmParser(MCSubtargetInfo &sti, MCAsmParser &parser,
560  const MCInstrInfo &MII)
561  : MCTargetAsmParser(), STI(sti), Parser(parser), InstInfo(0) {
562 
563  // Initialize the set of available features.
564  setAvailableFeatures(ComputeAvailableFeatures(STI.getFeatureBits()));
565  }
566  virtual bool ParseRegister(unsigned &RegNo, SMLoc &StartLoc, SMLoc &EndLoc);
567 
568  virtual bool ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
569  SMLoc NameLoc,
571 
572  virtual bool ParseDirective(AsmToken DirectiveID);
573 };
574 } // end anonymous namespace
575 
576 /// @name Auto-generated Match Functions
577 /// {
578 
579 static unsigned MatchRegisterName(StringRef Name);
580 
581 /// }
582 
583 static bool isImmSExti16i8Value(uint64_t Value) {
584  return (( Value <= 0x000000000000007FULL)||
585  (0x000000000000FF80ULL <= Value && Value <= 0x000000000000FFFFULL)||
586  (0xFFFFFFFFFFFFFF80ULL <= Value && Value <= 0xFFFFFFFFFFFFFFFFULL));
587 }
588 
589 static bool isImmSExti32i8Value(uint64_t Value) {
590  return (( Value <= 0x000000000000007FULL)||
591  (0x00000000FFFFFF80ULL <= Value && Value <= 0x00000000FFFFFFFFULL)||
592  (0xFFFFFFFFFFFFFF80ULL <= Value && Value <= 0xFFFFFFFFFFFFFFFFULL));
593 }
594 
595 static bool isImmZExtu32u8Value(uint64_t Value) {
596  return (Value <= 0x00000000000000FFULL);
597 }
598 
599 static bool isImmSExti64i8Value(uint64_t Value) {
600  return (( Value <= 0x000000000000007FULL)||
601  (0xFFFFFFFFFFFFFF80ULL <= Value && Value <= 0xFFFFFFFFFFFFFFFFULL));
602 }
603 
604 static bool isImmSExti64i32Value(uint64_t Value) {
605  return (( Value <= 0x000000007FFFFFFFULL)||
606  (0xFFFFFFFF80000000ULL <= Value && Value <= 0xFFFFFFFFFFFFFFFFULL));
607 }
608 namespace {
609 
610 /// X86Operand - Instances of this class represent a parsed X86 machine
611 /// instruction.
612 struct X86Operand : public MCParsedAsmOperand {
613  enum KindTy {
614  Token,
615  Register,
616  Immediate,
617  Memory
618  } Kind;
619 
620  SMLoc StartLoc, EndLoc;
621  SMLoc OffsetOfLoc;
622  StringRef SymName;
623  void *OpDecl;
624  bool AddressOf;
625 
626  struct TokOp {
627  const char *Data;
628  unsigned Length;
629  };
630 
631  struct RegOp {
632  unsigned RegNo;
633  };
634 
635  struct ImmOp {
636  const MCExpr *Val;
637  };
638 
639  struct MemOp {
640  unsigned SegReg;
641  const MCExpr *Disp;
642  unsigned BaseReg;
643  unsigned IndexReg;
644  unsigned Scale;
645  unsigned Size;
646  };
647 
648  union {
649  struct TokOp Tok;
650  struct RegOp Reg;
651  struct ImmOp Imm;
652  struct MemOp Mem;
653  };
654 
655  X86Operand(KindTy K, SMLoc Start, SMLoc End)
656  : Kind(K), StartLoc(Start), EndLoc(End) {}
657 
658  StringRef getSymName() { return SymName; }
659  void *getOpDecl() { return OpDecl; }
660 
661  /// getStartLoc - Get the location of the first token of this operand.
662  SMLoc getStartLoc() const { return StartLoc; }
663  /// getEndLoc - Get the location of the last token of this operand.
664  SMLoc getEndLoc() const { return EndLoc; }
665  /// getLocRange - Get the range between the first and last token of this
666  /// operand.
667  SMRange getLocRange() const { return SMRange(StartLoc, EndLoc); }
668  /// getOffsetOfLoc - Get the location of the offset operator.
669  SMLoc getOffsetOfLoc() const { return OffsetOfLoc; }
670 
671  virtual void print(raw_ostream &OS) const {}
672 
673  StringRef getToken() const {
674  assert(Kind == Token && "Invalid access!");
675  return StringRef(Tok.Data, Tok.Length);
676  }
677  void setTokenValue(StringRef Value) {
678  assert(Kind == Token && "Invalid access!");
679  Tok.Data = Value.data();
680  Tok.Length = Value.size();
681  }
682 
683  unsigned getReg() const {
684  assert(Kind == Register && "Invalid access!");
685  return Reg.RegNo;
686  }
687 
688  const MCExpr *getImm() const {
689  assert(Kind == Immediate && "Invalid access!");
690  return Imm.Val;
691  }
692 
693  const MCExpr *getMemDisp() const {
694  assert(Kind == Memory && "Invalid access!");
695  return Mem.Disp;
696  }
697  unsigned getMemSegReg() const {
698  assert(Kind == Memory && "Invalid access!");
699  return Mem.SegReg;
700  }
701  unsigned getMemBaseReg() const {
702  assert(Kind == Memory && "Invalid access!");
703  return Mem.BaseReg;
704  }
705  unsigned getMemIndexReg() const {
706  assert(Kind == Memory && "Invalid access!");
707  return Mem.IndexReg;
708  }
709  unsigned getMemScale() const {
710  assert(Kind == Memory && "Invalid access!");
711  return Mem.Scale;
712  }
713 
714  bool isToken() const {return Kind == Token; }
715 
716  bool isImm() const { return Kind == Immediate; }
717 
718  bool isImmSExti16i8() const {
719  if (!isImm())
720  return false;
721 
722  // If this isn't a constant expr, just assume it fits and let relaxation
723  // handle it.
724  const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
725  if (!CE)
726  return true;
727 
728  // Otherwise, check the value is in a range that makes sense for this
729  // extension.
730  return isImmSExti16i8Value(CE->getValue());
731  }
732  bool isImmSExti32i8() const {
733  if (!isImm())
734  return false;
735 
736  // If this isn't a constant expr, just assume it fits and let relaxation
737  // handle it.
738  const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
739  if (!CE)
740  return true;
741 
742  // Otherwise, check the value is in a range that makes sense for this
743  // extension.
744  return isImmSExti32i8Value(CE->getValue());
745  }
746  bool isImmZExtu32u8() const {
747  if (!isImm())
748  return false;
749 
750  // If this isn't a constant expr, just assume it fits and let relaxation
751  // handle it.
752  const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
753  if (!CE)
754  return true;
755 
756  // Otherwise, check the value is in a range that makes sense for this
757  // extension.
758  return isImmZExtu32u8Value(CE->getValue());
759  }
760  bool isImmSExti64i8() const {
761  if (!isImm())
762  return false;
763 
764  // If this isn't a constant expr, just assume it fits and let relaxation
765  // handle it.
766  const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
767  if (!CE)
768  return true;
769 
770  // Otherwise, check the value is in a range that makes sense for this
771  // extension.
772  return isImmSExti64i8Value(CE->getValue());
773  }
774  bool isImmSExti64i32() const {
775  if (!isImm())
776  return false;
777 
778  // If this isn't a constant expr, just assume it fits and let relaxation
779  // handle it.
780  const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getImm());
781  if (!CE)
782  return true;
783 
784  // Otherwise, check the value is in a range that makes sense for this
785  // extension.
786  return isImmSExti64i32Value(CE->getValue());
787  }
788 
789  bool isOffsetOf() const {
790  return OffsetOfLoc.getPointer();
791  }
792 
793  bool needAddressOf() const {
794  return AddressOf;
795  }
796 
797  bool isMem() const { return Kind == Memory; }
798  bool isMem8() const {
799  return Kind == Memory && (!Mem.Size || Mem.Size == 8);
800  }
801  bool isMem16() const {
802  return Kind == Memory && (!Mem.Size || Mem.Size == 16);
803  }
804  bool isMem32() const {
805  return Kind == Memory && (!Mem.Size || Mem.Size == 32);
806  }
807  bool isMem64() const {
808  return Kind == Memory && (!Mem.Size || Mem.Size == 64);
809  }
810  bool isMem80() const {
811  return Kind == Memory && (!Mem.Size || Mem.Size == 80);
812  }
813  bool isMem128() const {
814  return Kind == Memory && (!Mem.Size || Mem.Size == 128);
815  }
816  bool isMem256() const {
817  return Kind == Memory && (!Mem.Size || Mem.Size == 256);
818  }
819  bool isMem512() const {
820  return Kind == Memory && (!Mem.Size || Mem.Size == 512);
821  }
822 
823  bool isMemVX32() const {
824  return Kind == Memory && (!Mem.Size || Mem.Size == 32) &&
825  getMemIndexReg() >= X86::XMM0 && getMemIndexReg() <= X86::XMM15;
826  }
827  bool isMemVY32() const {
828  return Kind == Memory && (!Mem.Size || Mem.Size == 32) &&
829  getMemIndexReg() >= X86::YMM0 && getMemIndexReg() <= X86::YMM15;
830  }
831  bool isMemVX64() const {
832  return Kind == Memory && (!Mem.Size || Mem.Size == 64) &&
833  getMemIndexReg() >= X86::XMM0 && getMemIndexReg() <= X86::XMM15;
834  }
835  bool isMemVY64() const {
836  return Kind == Memory && (!Mem.Size || Mem.Size == 64) &&
837  getMemIndexReg() >= X86::YMM0 && getMemIndexReg() <= X86::YMM15;
838  }
839  bool isMemVZ32() const {
840  return Kind == Memory && (!Mem.Size || Mem.Size == 32) &&
841  getMemIndexReg() >= X86::ZMM0 && getMemIndexReg() <= X86::ZMM31;
842  }
843  bool isMemVZ64() const {
844  return Kind == Memory && (!Mem.Size || Mem.Size == 64) &&
845  getMemIndexReg() >= X86::ZMM0 && getMemIndexReg() <= X86::ZMM31;
846  }
847 
848  bool isAbsMem() const {
849  return Kind == Memory && !getMemSegReg() && !getMemBaseReg() &&
850  !getMemIndexReg() && getMemScale() == 1;
851  }
852 
853  bool isMemOffs8() const {
854  return Kind == Memory && !getMemSegReg() && !getMemBaseReg() &&
855  !getMemIndexReg() && getMemScale() == 1 && (!Mem.Size || Mem.Size == 8);
856  }
857  bool isMemOffs16() const {
858  return Kind == Memory && !getMemSegReg() && !getMemBaseReg() &&
859  !getMemIndexReg() && getMemScale() == 1 && (!Mem.Size || Mem.Size == 16);
860  }
861  bool isMemOffs32() const {
862  return Kind == Memory && !getMemSegReg() && !getMemBaseReg() &&
863  !getMemIndexReg() && getMemScale() == 1 && (!Mem.Size || Mem.Size == 32);
864  }
865  bool isMemOffs64() const {
866  return Kind == Memory && !getMemSegReg() && !getMemBaseReg() &&
867  !getMemIndexReg() && getMemScale() == 1 && (!Mem.Size || Mem.Size == 64);
868  }
869 
870  bool isReg() const { return Kind == Register; }
871 
872  bool isGR32orGR64() const {
873  return Kind == Register &&
874  (X86MCRegisterClasses[X86::GR32RegClassID].contains(getReg()) ||
875  X86MCRegisterClasses[X86::GR64RegClassID].contains(getReg()));
876  }
877 
878  void addExpr(MCInst &Inst, const MCExpr *Expr) const {
879  // Add as immediates when possible.
880  if (const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(Expr))
882  else
883  Inst.addOperand(MCOperand::CreateExpr(Expr));
884  }
885 
886  void addRegOperands(MCInst &Inst, unsigned N) const {
887  assert(N == 1 && "Invalid number of operands!");
889  }
890 
891  static unsigned getGR32FromGR64(unsigned RegNo) {
892  switch (RegNo) {
893  default: llvm_unreachable("Unexpected register");
894  case X86::RAX: return X86::EAX;
895  case X86::RCX: return X86::ECX;
896  case X86::RDX: return X86::EDX;
897  case X86::RBX: return X86::EBX;
898  case X86::RBP: return X86::EBP;
899  case X86::RSP: return X86::ESP;
900  case X86::RSI: return X86::ESI;
901  case X86::RDI: return X86::EDI;
902  case X86::R8: return X86::R8D;
903  case X86::R9: return X86::R9D;
904  case X86::R10: return X86::R10D;
905  case X86::R11: return X86::R11D;
906  case X86::R12: return X86::R12D;
907  case X86::R13: return X86::R13D;
908  case X86::R14: return X86::R14D;
909  case X86::R15: return X86::R15D;
910  case X86::RIP: return X86::EIP;
911  }
912  }
913 
914  void addGR32orGR64Operands(MCInst &Inst, unsigned N) const {
915  assert(N == 1 && "Invalid number of operands!");
916  unsigned RegNo = getReg();
917  if (X86MCRegisterClasses[X86::GR64RegClassID].contains(RegNo))
918  RegNo = getGR32FromGR64(RegNo);
919  Inst.addOperand(MCOperand::CreateReg(RegNo));
920  }
921 
922  void addImmOperands(MCInst &Inst, unsigned N) const {
923  assert(N == 1 && "Invalid number of operands!");
924  addExpr(Inst, getImm());
925  }
926 
927  void addMemOperands(MCInst &Inst, unsigned N) const {
928  assert((N == 5) && "Invalid number of operands!");
929  Inst.addOperand(MCOperand::CreateReg(getMemBaseReg()));
930  Inst.addOperand(MCOperand::CreateImm(getMemScale()));
931  Inst.addOperand(MCOperand::CreateReg(getMemIndexReg()));
932  addExpr(Inst, getMemDisp());
933  Inst.addOperand(MCOperand::CreateReg(getMemSegReg()));
934  }
935 
936  void addAbsMemOperands(MCInst &Inst, unsigned N) const {
937  assert((N == 1) && "Invalid number of operands!");
938  // Add as immediates when possible.
939  if (const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getMemDisp()))
941  else
942  Inst.addOperand(MCOperand::CreateExpr(getMemDisp()));
943  }
944 
945  void addMemOffsOperands(MCInst &Inst, unsigned N) const {
946  assert((N == 1) && "Invalid number of operands!");
947  // Add as immediates when possible.
948  if (const MCConstantExpr *CE = dyn_cast<MCConstantExpr>(getMemDisp()))
950  else
951  Inst.addOperand(MCOperand::CreateExpr(getMemDisp()));
952  }
953 
954  static X86Operand *CreateToken(StringRef Str, SMLoc Loc) {
955  SMLoc EndLoc = SMLoc::getFromPointer(Loc.getPointer() + Str.size());
956  X86Operand *Res = new X86Operand(Token, Loc, EndLoc);
957  Res->Tok.Data = Str.data();
958  Res->Tok.Length = Str.size();
959  return Res;
960  }
961 
962  static X86Operand *CreateReg(unsigned RegNo, SMLoc StartLoc, SMLoc EndLoc,
963  bool AddressOf = false,
964  SMLoc OffsetOfLoc = SMLoc(),
965  StringRef SymName = StringRef(),
966  void *OpDecl = 0) {
967  X86Operand *Res = new X86Operand(Register, StartLoc, EndLoc);
968  Res->Reg.RegNo = RegNo;
969  Res->AddressOf = AddressOf;
970  Res->OffsetOfLoc = OffsetOfLoc;
971  Res->SymName = SymName;
972  Res->OpDecl = OpDecl;
973  return Res;
974  }
975 
976  static X86Operand *CreateImm(const MCExpr *Val, SMLoc StartLoc, SMLoc EndLoc){
977  X86Operand *Res = new X86Operand(Immediate, StartLoc, EndLoc);
978  Res->Imm.Val = Val;
979  return Res;
980  }
981 
982  /// Create an absolute memory operand.
983  static X86Operand *CreateMem(const MCExpr *Disp, SMLoc StartLoc, SMLoc EndLoc,
984  unsigned Size = 0, StringRef SymName = StringRef(),
985  void *OpDecl = 0) {
986  X86Operand *Res = new X86Operand(Memory, StartLoc, EndLoc);
987  Res->Mem.SegReg = 0;
988  Res->Mem.Disp = Disp;
989  Res->Mem.BaseReg = 0;
990  Res->Mem.IndexReg = 0;
991  Res->Mem.Scale = 1;
992  Res->Mem.Size = Size;
993  Res->SymName = SymName;
994  Res->OpDecl = OpDecl;
995  Res->AddressOf = false;
996  return Res;
997  }
998 
999  /// Create a generalized memory operand.
1000  static X86Operand *CreateMem(unsigned SegReg, const MCExpr *Disp,
1001  unsigned BaseReg, unsigned IndexReg,
1002  unsigned Scale, SMLoc StartLoc, SMLoc EndLoc,
1003  unsigned Size = 0,
1004  StringRef SymName = StringRef(),
1005  void *OpDecl = 0) {
1006  // We should never just have a displacement, that should be parsed as an
1007  // absolute memory operand.
1008  assert((SegReg || BaseReg || IndexReg) && "Invalid memory operand!");
1009 
1010  // The scale should always be one of {1,2,4,8}.
1011  assert(((Scale == 1 || Scale == 2 || Scale == 4 || Scale == 8)) &&
1012  "Invalid scale!");
1013  X86Operand *Res = new X86Operand(Memory, StartLoc, EndLoc);
1014  Res->Mem.SegReg = SegReg;
1015  Res->Mem.Disp = Disp;
1016  Res->Mem.BaseReg = BaseReg;
1017  Res->Mem.IndexReg = IndexReg;
1018  Res->Mem.Scale = Scale;
1019  Res->Mem.Size = Size;
1020  Res->SymName = SymName;
1021  Res->OpDecl = OpDecl;
1022  Res->AddressOf = false;
1023  return Res;
1024  }
1025 };
1026 
1027 } // end anonymous namespace.
1028 
1029 bool X86AsmParser::isSrcOp(X86Operand &Op) {
1030  unsigned basereg = is64BitMode() ? X86::RSI : X86::ESI;
1031 
1032  return (Op.isMem() &&
1033  (Op.Mem.SegReg == 0 || Op.Mem.SegReg == X86::DS) &&
1034  isa<MCConstantExpr>(Op.Mem.Disp) &&
1035  cast<MCConstantExpr>(Op.Mem.Disp)->getValue() == 0 &&
1036  Op.Mem.BaseReg == basereg && Op.Mem.IndexReg == 0);
1037 }
1038 
1039 bool X86AsmParser::isDstOp(X86Operand &Op) {
1040  unsigned basereg = is64BitMode() ? X86::RDI : X86::EDI;
1041 
1042  return Op.isMem() &&
1043  (Op.Mem.SegReg == 0 || Op.Mem.SegReg == X86::ES) &&
1044  isa<MCConstantExpr>(Op.Mem.Disp) &&
1045  cast<MCConstantExpr>(Op.Mem.Disp)->getValue() == 0 &&
1046  Op.Mem.BaseReg == basereg && Op.Mem.IndexReg == 0;
1047 }
1048 
1049 bool X86AsmParser::ParseRegister(unsigned &RegNo,
1050  SMLoc &StartLoc, SMLoc &EndLoc) {
1051  RegNo = 0;
1052  const AsmToken &PercentTok = Parser.getTok();
1053  StartLoc = PercentTok.getLoc();
1054 
1055  // If we encounter a %, ignore it. This code handles registers with and
1056  // without the prefix, unprefixed registers can occur in cfi directives.
1057  if (!isParsingIntelSyntax() && PercentTok.is(AsmToken::Percent))
1058  Parser.Lex(); // Eat percent token.
1059 
1060  const AsmToken &Tok = Parser.getTok();
1061  EndLoc = Tok.getEndLoc();
1062 
1063  if (Tok.isNot(AsmToken::Identifier)) {
1064  if (isParsingIntelSyntax()) return true;
1065  return Error(StartLoc, "invalid register name",
1066  SMRange(StartLoc, EndLoc));
1067  }
1068 
1069  RegNo = MatchRegisterName(Tok.getString());
1070 
1071  // If the match failed, try the register name as lowercase.
1072  if (RegNo == 0)
1073  RegNo = MatchRegisterName(Tok.getString().lower());
1074 
1075  if (!is64BitMode()) {
1076  // FIXME: This should be done using Requires<In32BitMode> and
1077  // Requires<In64BitMode> so "eiz" usage in 64-bit instructions can be also
1078  // checked.
1079  // FIXME: Check AH, CH, DH, BH cannot be used in an instruction requiring a
1080  // REX prefix.
1081  if (RegNo == X86::RIZ ||
1082  X86MCRegisterClasses[X86::GR64RegClassID].contains(RegNo) ||
1085  return Error(StartLoc, "register %"
1086  + Tok.getString() + " is only available in 64-bit mode",
1087  SMRange(StartLoc, EndLoc));
1088  }
1089 
1090  // Parse "%st" as "%st(0)" and "%st(1)", which is multiple tokens.
1091  if (RegNo == 0 && (Tok.getString() == "st" || Tok.getString() == "ST")) {
1092  RegNo = X86::ST0;
1093  Parser.Lex(); // Eat 'st'
1094 
1095  // Check to see if we have '(4)' after %st.
1096  if (getLexer().isNot(AsmToken::LParen))
1097  return false;
1098  // Lex the paren.
1099  getParser().Lex();
1100 
1101  const AsmToken &IntTok = Parser.getTok();
1102  if (IntTok.isNot(AsmToken::Integer))
1103  return Error(IntTok.getLoc(), "expected stack index");
1104  switch (IntTok.getIntVal()) {
1105  case 0: RegNo = X86::ST0; break;
1106  case 1: RegNo = X86::ST1; break;
1107  case 2: RegNo = X86::ST2; break;
1108  case 3: RegNo = X86::ST3; break;
1109  case 4: RegNo = X86::ST4; break;
1110  case 5: RegNo = X86::ST5; break;
1111  case 6: RegNo = X86::ST6; break;
1112  case 7: RegNo = X86::ST7; break;
1113  default: return Error(IntTok.getLoc(), "invalid stack index");
1114  }
1115 
1116  if (getParser().Lex().isNot(AsmToken::RParen))
1117  return Error(Parser.getTok().getLoc(), "expected ')'");
1118 
1119  EndLoc = Parser.getTok().getEndLoc();
1120  Parser.Lex(); // Eat ')'
1121  return false;
1122  }
1123 
1124  EndLoc = Parser.getTok().getEndLoc();
1125 
1126  // If this is "db[0-7]", match it as an alias
1127  // for dr[0-7].
1128  if (RegNo == 0 && Tok.getString().size() == 3 &&
1129  Tok.getString().startswith("db")) {
1130  switch (Tok.getString()[2]) {
1131  case '0': RegNo = X86::DR0; break;
1132  case '1': RegNo = X86::DR1; break;
1133  case '2': RegNo = X86::DR2; break;
1134  case '3': RegNo = X86::DR3; break;
1135  case '4': RegNo = X86::DR4; break;
1136  case '5': RegNo = X86::DR5; break;
1137  case '6': RegNo = X86::DR6; break;
1138  case '7': RegNo = X86::DR7; break;
1139  }
1140 
1141  if (RegNo != 0) {
1142  EndLoc = Parser.getTok().getEndLoc();
1143  Parser.Lex(); // Eat it.
1144  return false;
1145  }
1146  }
1147 
1148  if (RegNo == 0) {
1149  if (isParsingIntelSyntax()) return true;
1150  return Error(StartLoc, "invalid register name",
1151  SMRange(StartLoc, EndLoc));
1152  }
1153 
1154  Parser.Lex(); // Eat identifier token.
1155  return false;
1156 }
1157 
1158 X86Operand *X86AsmParser::ParseOperand() {
1159  if (isParsingIntelSyntax())
1160  return ParseIntelOperand();
1161  return ParseATTOperand();
1162 }
1163 
1164 /// getIntelMemOperandSize - Return intel memory operand size.
1165 static unsigned getIntelMemOperandSize(StringRef OpStr) {
1166  unsigned Size = StringSwitch<unsigned>(OpStr)
1167  .Cases("BYTE", "byte", 8)
1168  .Cases("WORD", "word", 16)
1169  .Cases("DWORD", "dword", 32)
1170  .Cases("QWORD", "qword", 64)
1171  .Cases("XWORD", "xword", 80)
1172  .Cases("XMMWORD", "xmmword", 128)
1173  .Cases("YMMWORD", "ymmword", 256)
1174  .Default(0);
1175  return Size;
1176 }
1177 
1178 X86Operand *
1179 X86AsmParser::CreateMemForInlineAsm(unsigned SegReg, const MCExpr *Disp,
1180  unsigned BaseReg, unsigned IndexReg,
1181  unsigned Scale, SMLoc Start, SMLoc End,
1182  unsigned Size, StringRef Identifier,
1183  InlineAsmIdentifierInfo &Info){
1184  if (isa<MCSymbolRefExpr>(Disp)) {
1185  // If this is not a VarDecl then assume it is a FuncDecl or some other label
1186  // reference. We need an 'r' constraint here, so we need to create register
1187  // operand to ensure proper matching. Just pick a GPR based on the size of
1188  // a pointer.
1189  if (!Info.IsVarDecl) {
1190  unsigned RegNo = is64BitMode() ? X86::RBX : X86::EBX;
1191  return X86Operand::CreateReg(RegNo, Start, End, /*AddressOf=*/true,
1192  SMLoc(), Identifier, Info.OpDecl);
1193  }
1194  if (!Size) {
1195  Size = Info.Type * 8; // Size is in terms of bits in this context.
1196  if (Size)
1197  InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_SizeDirective, Start,
1198  /*Len=*/0, Size));
1199  }
1200  }
1201 
1202  // When parsing inline assembly we set the base register to a non-zero value
1203  // if we don't know the actual value at this time. This is necessary to
1204  // get the matching correct in some cases.
1205  BaseReg = BaseReg ? BaseReg : 1;
1206  return X86Operand::CreateMem(SegReg, Disp, BaseReg, IndexReg, Scale, Start,
1207  End, Size, Identifier, Info.OpDecl);
1208 }
1209 
1210 static void
1212  StringRef SymName, int64_t ImmDisp,
1213  int64_t FinalImmDisp, SMLoc &BracLoc,
1214  SMLoc &StartInBrac, SMLoc &End) {
1215  // Remove the '[' and ']' from the IR string.
1216  AsmRewrites->push_back(AsmRewrite(AOK_Skip, BracLoc, 1));
1217  AsmRewrites->push_back(AsmRewrite(AOK_Skip, End, 1));
1218 
1219  // If ImmDisp is non-zero, then we parsed a displacement before the
1220  // bracketed expression (i.e., ImmDisp [ BaseReg + Scale*IndexReg + Disp])
1221  // If ImmDisp doesn't match the displacement computed by the state machine
1222  // then we have an additional displacement in the bracketed expression.
1223  if (ImmDisp != FinalImmDisp) {
1224  if (ImmDisp) {
1225  // We have an immediate displacement before the bracketed expression.
1226  // Adjust this to match the final immediate displacement.
1227  bool Found = false;
1228  for (SmallVectorImpl<AsmRewrite>::iterator I = AsmRewrites->begin(),
1229  E = AsmRewrites->end(); I != E; ++I) {
1230  if ((*I).Loc.getPointer() > BracLoc.getPointer())
1231  continue;
1232  if ((*I).Kind == AOK_ImmPrefix || (*I).Kind == AOK_Imm) {
1233  assert (!Found && "ImmDisp already rewritten.");
1234  (*I).Kind = AOK_Imm;
1235  (*I).Len = BracLoc.getPointer() - (*I).Loc.getPointer();
1236  (*I).Val = FinalImmDisp;
1237  Found = true;
1238  break;
1239  }
1240  }
1241  assert (Found && "Unable to rewrite ImmDisp.");
1242  (void)Found;
1243  } else {
1244  // We have a symbolic and an immediate displacement, but no displacement
1245  // before the bracketed expression. Put the immediate displacement
1246  // before the bracketed expression.
1247  AsmRewrites->push_back(AsmRewrite(AOK_Imm, BracLoc, 0, FinalImmDisp));
1248  }
1249  }
1250  // Remove all the ImmPrefix rewrites within the brackets.
1251  for (SmallVectorImpl<AsmRewrite>::iterator I = AsmRewrites->begin(),
1252  E = AsmRewrites->end(); I != E; ++I) {
1253  if ((*I).Loc.getPointer() < StartInBrac.getPointer())
1254  continue;
1255  if ((*I).Kind == AOK_ImmPrefix)
1256  (*I).Kind = AOK_Delete;
1257  }
1258  const char *SymLocPtr = SymName.data();
1259  // Skip everything before the symbol.
1260  if (unsigned Len = SymLocPtr - StartInBrac.getPointer()) {
1261  assert(Len > 0 && "Expected a non-negative length.");
1262  AsmRewrites->push_back(AsmRewrite(AOK_Skip, StartInBrac, Len));
1263  }
1264  // Skip everything after the symbol.
1265  if (unsigned Len = End.getPointer() - (SymLocPtr + SymName.size())) {
1266  SMLoc Loc = SMLoc::getFromPointer(SymLocPtr + SymName.size());
1267  assert(Len > 0 && "Expected a non-negative length.");
1268  AsmRewrites->push_back(AsmRewrite(AOK_Skip, Loc, Len));
1269  }
1270 }
1271 
1272 X86Operand *
1273 X86AsmParser::ParseIntelExpression(IntelExprStateMachine &SM, SMLoc &End) {
1274  const AsmToken &Tok = Parser.getTok();
1275 
1276  bool Done = false;
1277  while (!Done) {
1278  bool UpdateLocLex = true;
1279 
1280  // The period in the dot operator (e.g., [ebx].foo.bar) is parsed as an
1281  // identifier. Don't try an parse it as a register.
1282  if (Tok.getString().startswith("."))
1283  break;
1284 
1285  // If we're parsing an immediate expression, we don't expect a '['.
1286  if (SM.getStopOnLBrac() && getLexer().getKind() == AsmToken::LBrac)
1287  break;
1288 
1289  switch (getLexer().getKind()) {
1290  default: {
1291  if (SM.isValidEndState()) {
1292  Done = true;
1293  break;
1294  }
1295  return ErrorOperand(Tok.getLoc(), "Unexpected token!");
1296  }
1297  case AsmToken::EndOfStatement: {
1298  Done = true;
1299  break;
1300  }
1301  case AsmToken::Identifier: {
1302  // This could be a register or a symbolic displacement.
1303  unsigned TmpReg;
1304  const MCExpr *Val;
1305  SMLoc IdentLoc = Tok.getLoc();
1306  StringRef Identifier = Tok.getString();
1307  if(!ParseRegister(TmpReg, IdentLoc, End)) {
1308  SM.onRegister(TmpReg);
1309  UpdateLocLex = false;
1310  break;
1311  } else {
1312  if (!isParsingInlineAsm()) {
1313  if (getParser().parsePrimaryExpr(Val, End))
1314  return ErrorOperand(Tok.getLoc(), "Unexpected identifier!");
1315  } else {
1316  InlineAsmIdentifierInfo &Info = SM.getIdentifierInfo();
1317  if (X86Operand *Err = ParseIntelIdentifier(Val, Identifier, Info,
1318  /*Unevaluated*/ false, End))
1319  return Err;
1320  }
1321  SM.onIdentifierExpr(Val, Identifier);
1322  UpdateLocLex = false;
1323  break;
1324  }
1325  return ErrorOperand(Tok.getLoc(), "Unexpected identifier!");
1326  }
1327  case AsmToken::Integer:
1328  if (isParsingInlineAsm() && SM.getAddImmPrefix())
1329  InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_ImmPrefix,
1330  Tok.getLoc()));
1331  SM.onInteger(Tok.getIntVal());
1332  break;
1333  case AsmToken::Plus: SM.onPlus(); break;
1334  case AsmToken::Minus: SM.onMinus(); break;
1335  case AsmToken::Star: SM.onStar(); break;
1336  case AsmToken::Slash: SM.onDivide(); break;
1337  case AsmToken::LBrac: SM.onLBrac(); break;
1338  case AsmToken::RBrac: SM.onRBrac(); break;
1339  case AsmToken::LParen: SM.onLParen(); break;
1340  case AsmToken::RParen: SM.onRParen(); break;
1341  }
1342  if (SM.hadError())
1343  return ErrorOperand(Tok.getLoc(), "Unexpected token!");
1344 
1345  if (!Done && UpdateLocLex) {
1346  End = Tok.getLoc();
1347  Parser.Lex(); // Consume the token.
1348  }
1349  }
1350  return 0;
1351 }
1352 
1353 X86Operand *X86AsmParser::ParseIntelBracExpression(unsigned SegReg, SMLoc Start,
1354  int64_t ImmDisp,
1355  unsigned Size) {
1356  const AsmToken &Tok = Parser.getTok();
1357  SMLoc BracLoc = Tok.getLoc(), End = Tok.getEndLoc();
1358  if (getLexer().isNot(AsmToken::LBrac))
1359  return ErrorOperand(BracLoc, "Expected '[' token!");
1360  Parser.Lex(); // Eat '['
1361 
1362  SMLoc StartInBrac = Tok.getLoc();
1363  // Parse [ Symbol + ImmDisp ] and [ BaseReg + Scale*IndexReg + ImmDisp ]. We
1364  // may have already parsed an immediate displacement before the bracketed
1365  // expression.
1366  IntelExprStateMachine SM(ImmDisp, /*StopOnLBrac=*/false, /*AddImmPrefix=*/true);
1367  if (X86Operand *Err = ParseIntelExpression(SM, End))
1368  return Err;
1369 
1370  const MCExpr *Disp;
1371  if (const MCExpr *Sym = SM.getSym()) {
1372  // A symbolic displacement.
1373  Disp = Sym;
1374  if (isParsingInlineAsm())
1375  RewriteIntelBracExpression(InstInfo->AsmRewrites, SM.getSymName(),
1376  ImmDisp, SM.getImm(), BracLoc, StartInBrac,
1377  End);
1378  } else {
1379  // An immediate displacement only.
1380  Disp = MCConstantExpr::Create(SM.getImm(), getContext());
1381  }
1382 
1383  // Parse the dot operator (e.g., [ebx].foo.bar).
1384  if (Tok.getString().startswith(".")) {
1385  const MCExpr *NewDisp;
1386  if (X86Operand *Err = ParseIntelDotOperator(Disp, NewDisp))
1387  return Err;
1388 
1389  End = Tok.getEndLoc();
1390  Parser.Lex(); // Eat the field.
1391  Disp = NewDisp;
1392  }
1393 
1394  int BaseReg = SM.getBaseReg();
1395  int IndexReg = SM.getIndexReg();
1396  int Scale = SM.getScale();
1397  if (!isParsingInlineAsm()) {
1398  // handle [-42]
1399  if (!BaseReg && !IndexReg) {
1400  if (!SegReg)
1401  return X86Operand::CreateMem(Disp, Start, End, Size);
1402  else
1403  return X86Operand::CreateMem(SegReg, Disp, 0, 0, 1, Start, End, Size);
1404  }
1405  return X86Operand::CreateMem(SegReg, Disp, BaseReg, IndexReg, Scale, Start,
1406  End, Size);
1407  }
1408 
1409  InlineAsmIdentifierInfo &Info = SM.getIdentifierInfo();
1410  return CreateMemForInlineAsm(SegReg, Disp, BaseReg, IndexReg, Scale, Start,
1411  End, Size, SM.getSymName(), Info);
1412 }
1413 
1414 // Inline assembly may use variable names with namespace alias qualifiers.
1415 X86Operand *X86AsmParser::ParseIntelIdentifier(const MCExpr *&Val,
1416  StringRef &Identifier,
1418  bool IsUnevaluatedOperand,
1419  SMLoc &End) {
1420  assert (isParsingInlineAsm() && "Expected to be parsing inline assembly.");
1421  Val = 0;
1422 
1423  StringRef LineBuf(Identifier.data());
1424  SemaCallback->LookupInlineAsmIdentifier(LineBuf, Info, IsUnevaluatedOperand);
1425 
1426  const AsmToken &Tok = Parser.getTok();
1427 
1428  // Advance the token stream until the end of the current token is
1429  // after the end of what the frontend claimed.
1430  const char *EndPtr = Tok.getLoc().getPointer() + LineBuf.size();
1431  while (true) {
1432  End = Tok.getEndLoc();
1433  getLexer().Lex();
1434 
1435  assert(End.getPointer() <= EndPtr && "frontend claimed part of a token?");
1436  if (End.getPointer() == EndPtr) break;
1437  }
1438 
1439  // Create the symbol reference.
1440  Identifier = LineBuf;
1441  MCSymbol *Sym = getContext().GetOrCreateSymbol(Identifier);
1443  Val = MCSymbolRefExpr::Create(Sym, Variant, getParser().getContext());
1444  return 0;
1445 }
1446 
1447 /// \brief Parse intel style segment override.
1448 X86Operand *X86AsmParser::ParseIntelSegmentOverride(unsigned SegReg,
1449  SMLoc Start,
1450  unsigned Size) {
1451  assert(SegReg != 0 && "Tried to parse a segment override without a segment!");
1452  const AsmToken &Tok = Parser.getTok(); // Eat colon.
1453  if (Tok.isNot(AsmToken::Colon))
1454  return ErrorOperand(Tok.getLoc(), "Expected ':' token!");
1455  Parser.Lex(); // Eat ':'
1456 
1457  int64_t ImmDisp = 0;
1458  if (getLexer().is(AsmToken::Integer)) {
1459  ImmDisp = Tok.getIntVal();
1460  AsmToken ImmDispToken = Parser.Lex(); // Eat the integer.
1461 
1462  if (isParsingInlineAsm())
1463  InstInfo->AsmRewrites->push_back(
1464  AsmRewrite(AOK_ImmPrefix, ImmDispToken.getLoc()));
1465 
1466  if (getLexer().isNot(AsmToken::LBrac)) {
1467  // An immediate following a 'segment register', 'colon' token sequence can
1468  // be followed by a bracketed expression. If it isn't we know we have our
1469  // final segment override.
1470  const MCExpr *Disp = MCConstantExpr::Create(ImmDisp, getContext());
1471  return X86Operand::CreateMem(SegReg, Disp, /*BaseReg=*/0, /*IndexReg=*/0,
1472  /*Scale=*/1, Start, ImmDispToken.getEndLoc(),
1473  Size);
1474  }
1475  }
1476 
1477  if (getLexer().is(AsmToken::LBrac))
1478  return ParseIntelBracExpression(SegReg, Start, ImmDisp, Size);
1479 
1480  const MCExpr *Val;
1481  SMLoc End;
1482  if (!isParsingInlineAsm()) {
1483  if (getParser().parsePrimaryExpr(Val, End))
1484  return ErrorOperand(Tok.getLoc(), "Unexpected token!");
1485 
1486  return X86Operand::CreateMem(Val, Start, End, Size);
1487  }
1488 
1490  StringRef Identifier = Tok.getString();
1491  if (X86Operand *Err = ParseIntelIdentifier(Val, Identifier, Info,
1492  /*Unevaluated*/ false, End))
1493  return Err;
1494  return CreateMemForInlineAsm(/*SegReg=*/0, Val, /*BaseReg=*/0,/*IndexReg=*/0,
1495  /*Scale=*/1, Start, End, Size, Identifier, Info);
1496 }
1497 
1498 /// ParseIntelMemOperand - Parse intel style memory operand.
1499 X86Operand *X86AsmParser::ParseIntelMemOperand(int64_t ImmDisp, SMLoc Start,
1500  unsigned Size) {
1501  const AsmToken &Tok = Parser.getTok();
1502  SMLoc End;
1503 
1504  // Parse ImmDisp [ BaseReg + Scale*IndexReg + Disp ].
1505  if (getLexer().is(AsmToken::LBrac))
1506  return ParseIntelBracExpression(/*SegReg=*/0, Start, ImmDisp, Size);
1507 
1508  const MCExpr *Val;
1509  if (!isParsingInlineAsm()) {
1510  if (getParser().parsePrimaryExpr(Val, End))
1511  return ErrorOperand(Tok.getLoc(), "Unexpected token!");
1512 
1513  return X86Operand::CreateMem(Val, Start, End, Size);
1514  }
1515 
1517  StringRef Identifier = Tok.getString();
1518  if (X86Operand *Err = ParseIntelIdentifier(Val, Identifier, Info,
1519  /*Unevaluated*/ false, End))
1520  return Err;
1521  return CreateMemForInlineAsm(/*SegReg=*/0, Val, /*BaseReg=*/0, /*IndexReg=*/0,
1522  /*Scale=*/1, Start, End, Size, Identifier, Info);
1523 }
1524 
1525 /// Parse the '.' operator.
1526 X86Operand *X86AsmParser::ParseIntelDotOperator(const MCExpr *Disp,
1527  const MCExpr *&NewDisp) {
1528  const AsmToken &Tok = Parser.getTok();
1529  int64_t OrigDispVal, DotDispVal;
1530 
1531  // FIXME: Handle non-constant expressions.
1532  if (const MCConstantExpr *OrigDisp = dyn_cast<MCConstantExpr>(Disp))
1533  OrigDispVal = OrigDisp->getValue();
1534  else
1535  return ErrorOperand(Tok.getLoc(), "Non-constant offsets are not supported!");
1536 
1537  // Drop the '.'.
1538  StringRef DotDispStr = Tok.getString().drop_front(1);
1539 
1540  // .Imm gets lexed as a real.
1541  if (Tok.is(AsmToken::Real)) {
1542  APInt DotDisp;
1543  DotDispStr.getAsInteger(10, DotDisp);
1544  DotDispVal = DotDisp.getZExtValue();
1545  } else if (isParsingInlineAsm() && Tok.is(AsmToken::Identifier)) {
1546  unsigned DotDisp;
1547  std::pair<StringRef, StringRef> BaseMember = DotDispStr.split('.');
1548  if (SemaCallback->LookupInlineAsmField(BaseMember.first, BaseMember.second,
1549  DotDisp))
1550  return ErrorOperand(Tok.getLoc(), "Unable to lookup field reference!");
1551  DotDispVal = DotDisp;
1552  } else
1553  return ErrorOperand(Tok.getLoc(), "Unexpected token type!");
1554 
1555  if (isParsingInlineAsm() && Tok.is(AsmToken::Identifier)) {
1556  SMLoc Loc = SMLoc::getFromPointer(DotDispStr.data());
1557  unsigned Len = DotDispStr.size();
1558  unsigned Val = OrigDispVal + DotDispVal;
1559  InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_DotOperator, Loc, Len,
1560  Val));
1561  }
1562 
1563  NewDisp = MCConstantExpr::Create(OrigDispVal + DotDispVal, getContext());
1564  return 0;
1565 }
1566 
1567 /// Parse the 'offset' operator. This operator is used to specify the
1568 /// location rather then the content of a variable.
1569 X86Operand *X86AsmParser::ParseIntelOffsetOfOperator() {
1570  const AsmToken &Tok = Parser.getTok();
1571  SMLoc OffsetOfLoc = Tok.getLoc();
1572  Parser.Lex(); // Eat offset.
1573 
1574  const MCExpr *Val;
1576  SMLoc Start = Tok.getLoc(), End;
1577  StringRef Identifier = Tok.getString();
1578  if (X86Operand *Err = ParseIntelIdentifier(Val, Identifier, Info,
1579  /*Unevaluated*/ false, End))
1580  return Err;
1581 
1582  // Don't emit the offset operator.
1583  InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Skip, OffsetOfLoc, 7));
1584 
1585  // The offset operator will have an 'r' constraint, thus we need to create
1586  // register operand to ensure proper matching. Just pick a GPR based on
1587  // the size of a pointer.
1588  unsigned RegNo = is64BitMode() ? X86::RBX : X86::EBX;
1589  return X86Operand::CreateReg(RegNo, Start, End, /*GetAddress=*/true,
1590  OffsetOfLoc, Identifier, Info.OpDecl);
1591 }
1592 
1597 };
1598 
1599 /// Parse the 'LENGTH', 'TYPE' and 'SIZE' operators. The LENGTH operator
1600 /// returns the number of elements in an array. It returns the value 1 for
1601 /// non-array variables. The SIZE operator returns the size of a C or C++
1602 /// variable. A variable's size is the product of its LENGTH and TYPE. The
1603 /// TYPE operator returns the size of a C or C++ type or variable. If the
1604 /// variable is an array, TYPE returns the size of a single element.
1605 X86Operand *X86AsmParser::ParseIntelOperator(unsigned OpKind) {
1606  const AsmToken &Tok = Parser.getTok();
1607  SMLoc TypeLoc = Tok.getLoc();
1608  Parser.Lex(); // Eat operator.
1609 
1610  const MCExpr *Val = 0;
1612  SMLoc Start = Tok.getLoc(), End;
1613  StringRef Identifier = Tok.getString();
1614  if (X86Operand *Err = ParseIntelIdentifier(Val, Identifier, Info,
1615  /*Unevaluated*/ true, End))
1616  return Err;
1617 
1618  unsigned CVal = 0;
1619  switch(OpKind) {
1620  default: llvm_unreachable("Unexpected operand kind!");
1621  case IOK_LENGTH: CVal = Info.Length; break;
1622  case IOK_SIZE: CVal = Info.Size; break;
1623  case IOK_TYPE: CVal = Info.Type; break;
1624  }
1625 
1626  // Rewrite the type operator and the C or C++ type or variable in terms of an
1627  // immediate. E.g. TYPE foo -> $$4
1628  unsigned Len = End.getPointer() - TypeLoc.getPointer();
1629  InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Imm, TypeLoc, Len, CVal));
1630 
1631  const MCExpr *Imm = MCConstantExpr::Create(CVal, getContext());
1632  return X86Operand::CreateImm(Imm, Start, End);
1633 }
1634 
1635 X86Operand *X86AsmParser::ParseIntelOperand() {
1636  const AsmToken &Tok = Parser.getTok();
1637  SMLoc Start, End;
1638 
1639  // Offset, length, type and size operators.
1640  if (isParsingInlineAsm()) {
1641  StringRef AsmTokStr = Tok.getString();
1642  if (AsmTokStr == "offset" || AsmTokStr == "OFFSET")
1643  return ParseIntelOffsetOfOperator();
1644  if (AsmTokStr == "length" || AsmTokStr == "LENGTH")
1645  return ParseIntelOperator(IOK_LENGTH);
1646  if (AsmTokStr == "size" || AsmTokStr == "SIZE")
1647  return ParseIntelOperator(IOK_SIZE);
1648  if (AsmTokStr == "type" || AsmTokStr == "TYPE")
1649  return ParseIntelOperator(IOK_TYPE);
1650  }
1651 
1652  unsigned Size = getIntelMemOperandSize(Tok.getString());
1653  if (Size) {
1654  Parser.Lex(); // Eat operand size (e.g., byte, word).
1655  if (Tok.getString() != "PTR" && Tok.getString() != "ptr")
1656  return ErrorOperand(Start, "Expected 'PTR' or 'ptr' token!");
1657  Parser.Lex(); // Eat ptr.
1658  }
1659  Start = Tok.getLoc();
1660 
1661  // Immediate.
1662  if (getLexer().is(AsmToken::Integer) || getLexer().is(AsmToken::Minus) ||
1663  getLexer().is(AsmToken::LParen)) {
1664  AsmToken StartTok = Tok;
1665  IntelExprStateMachine SM(/*Imm=*/0, /*StopOnLBrac=*/true,
1666  /*AddImmPrefix=*/false);
1667  if (X86Operand *Err = ParseIntelExpression(SM, End))
1668  return Err;
1669 
1670  int64_t Imm = SM.getImm();
1671  if (isParsingInlineAsm()) {
1672  unsigned Len = Tok.getLoc().getPointer() - Start.getPointer();
1673  if (StartTok.getString().size() == Len)
1674  // Just add a prefix if this wasn't a complex immediate expression.
1675  InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_ImmPrefix, Start));
1676  else
1677  // Otherwise, rewrite the complex expression as a single immediate.
1678  InstInfo->AsmRewrites->push_back(AsmRewrite(AOK_Imm, Start, Len, Imm));
1679  }
1680 
1681  if (getLexer().isNot(AsmToken::LBrac)) {
1682  const MCExpr *ImmExpr = MCConstantExpr::Create(Imm, getContext());
1683  return X86Operand::CreateImm(ImmExpr, Start, End);
1684  }
1685 
1686  // Only positive immediates are valid.
1687  if (Imm < 0)
1688  return ErrorOperand(Start, "expected a positive immediate displacement "
1689  "before bracketed expr.");
1690 
1691  // Parse ImmDisp [ BaseReg + Scale*IndexReg + Disp ].
1692  return ParseIntelMemOperand(Imm, Start, Size);
1693  }
1694 
1695  // Register.
1696  unsigned RegNo = 0;
1697  if (!ParseRegister(RegNo, Start, End)) {
1698  // If this is a segment register followed by a ':', then this is the start
1699  // of a segment override, otherwise this is a normal register reference.
1700  if (getLexer().isNot(AsmToken::Colon))
1701  return X86Operand::CreateReg(RegNo, Start, End);
1702 
1703  return ParseIntelSegmentOverride(/*SegReg=*/RegNo, Start, Size);
1704  }
1705 
1706  // Memory operand.
1707  return ParseIntelMemOperand(/*Disp=*/0, Start, Size);
1708 }
1709 
1710 X86Operand *X86AsmParser::ParseATTOperand() {
1711  switch (getLexer().getKind()) {
1712  default:
1713  // Parse a memory operand with no segment register.
1714  return ParseMemOperand(0, Parser.getTok().getLoc());
1715  case AsmToken::Percent: {
1716  // Read the register.
1717  unsigned RegNo;
1718  SMLoc Start, End;
1719  if (ParseRegister(RegNo, Start, End)) return 0;
1720  if (RegNo == X86::EIZ || RegNo == X86::RIZ) {
1721  Error(Start, "%eiz and %riz can only be used as index registers",
1722  SMRange(Start, End));
1723  return 0;
1724  }
1725 
1726  // If this is a segment register followed by a ':', then this is the start
1727  // of a memory reference, otherwise this is a normal register reference.
1728  if (getLexer().isNot(AsmToken::Colon))
1729  return X86Operand::CreateReg(RegNo, Start, End);
1730 
1731  getParser().Lex(); // Eat the colon.
1732  return ParseMemOperand(RegNo, Start);
1733  }
1734  case AsmToken::Dollar: {
1735  // $42 -> immediate.
1736  SMLoc Start = Parser.getTok().getLoc(), End;
1737  Parser.Lex();
1738  const MCExpr *Val;
1739  if (getParser().parseExpression(Val, End))
1740  return 0;
1741  return X86Operand::CreateImm(Val, Start, End);
1742  }
1743  }
1744 }
1745 
1746 /// ParseMemOperand: segment: disp(basereg, indexreg, scale). The '%ds:' prefix
1747 /// has already been parsed if present.
1748 X86Operand *X86AsmParser::ParseMemOperand(unsigned SegReg, SMLoc MemStart) {
1749 
1750  // We have to disambiguate a parenthesized expression "(4+5)" from the start
1751  // of a memory operand with a missing displacement "(%ebx)" or "(,%eax)". The
1752  // only way to do this without lookahead is to eat the '(' and see what is
1753  // after it.
1754  const MCExpr *Disp = MCConstantExpr::Create(0, getParser().getContext());
1755  if (getLexer().isNot(AsmToken::LParen)) {
1756  SMLoc ExprEnd;
1757  if (getParser().parseExpression(Disp, ExprEnd)) return 0;
1758 
1759  // After parsing the base expression we could either have a parenthesized
1760  // memory address or not. If not, return now. If so, eat the (.
1761  if (getLexer().isNot(AsmToken::LParen)) {
1762  // Unless we have a segment register, treat this as an immediate.
1763  if (SegReg == 0)
1764  return X86Operand::CreateMem(Disp, MemStart, ExprEnd);
1765  return X86Operand::CreateMem(SegReg, Disp, 0, 0, 1, MemStart, ExprEnd);
1766  }
1767 
1768  // Eat the '('.
1769  Parser.Lex();
1770  } else {
1771  // Okay, we have a '('. We don't know if this is an expression or not, but
1772  // so we have to eat the ( to see beyond it.
1773  SMLoc LParenLoc = Parser.getTok().getLoc();
1774  Parser.Lex(); // Eat the '('.
1775 
1776  if (getLexer().is(AsmToken::Percent) || getLexer().is(AsmToken::Comma)) {
1777  // Nothing to do here, fall into the code below with the '(' part of the
1778  // memory operand consumed.
1779  } else {
1780  SMLoc ExprEnd;
1781 
1782  // It must be an parenthesized expression, parse it now.
1783  if (getParser().parseParenExpression(Disp, ExprEnd))
1784  return 0;
1785 
1786  // After parsing the base expression we could either have a parenthesized
1787  // memory address or not. If not, return now. If so, eat the (.
1788  if (getLexer().isNot(AsmToken::LParen)) {
1789  // Unless we have a segment register, treat this as an immediate.
1790  if (SegReg == 0)
1791  return X86Operand::CreateMem(Disp, LParenLoc, ExprEnd);
1792  return X86Operand::CreateMem(SegReg, Disp, 0, 0, 1, MemStart, ExprEnd);
1793  }
1794 
1795  // Eat the '('.
1796  Parser.Lex();
1797  }
1798  }
1799 
1800  // If we reached here, then we just ate the ( of the memory operand. Process
1801  // the rest of the memory operand.
1802  unsigned BaseReg = 0, IndexReg = 0, Scale = 1;
1803  SMLoc IndexLoc;
1804 
1805  if (getLexer().is(AsmToken::Percent)) {
1806  SMLoc StartLoc, EndLoc;
1807  if (ParseRegister(BaseReg, StartLoc, EndLoc)) return 0;
1808  if (BaseReg == X86::EIZ || BaseReg == X86::RIZ) {
1809  Error(StartLoc, "eiz and riz can only be used as index registers",
1810  SMRange(StartLoc, EndLoc));
1811  return 0;
1812  }
1813  }
1814 
1815  if (getLexer().is(AsmToken::Comma)) {
1816  Parser.Lex(); // Eat the comma.
1817  IndexLoc = Parser.getTok().getLoc();
1818 
1819  // Following the comma we should have either an index register, or a scale
1820  // value. We don't support the later form, but we want to parse it
1821  // correctly.
1822  //
1823  // Not that even though it would be completely consistent to support syntax
1824  // like "1(%eax,,1)", the assembler doesn't. Use "eiz" or "riz" for this.
1825  if (getLexer().is(AsmToken::Percent)) {
1826  SMLoc L;
1827  if (ParseRegister(IndexReg, L, L)) return 0;
1828 
1829  if (getLexer().isNot(AsmToken::RParen)) {
1830  // Parse the scale amount:
1831  // ::= ',' [scale-expression]
1832  if (getLexer().isNot(AsmToken::Comma)) {
1833  Error(Parser.getTok().getLoc(),
1834  "expected comma in scale expression");
1835  return 0;
1836  }
1837  Parser.Lex(); // Eat the comma.
1838 
1839  if (getLexer().isNot(AsmToken::RParen)) {
1840  SMLoc Loc = Parser.getTok().getLoc();
1841 
1842  int64_t ScaleVal;
1843  if (getParser().parseAbsoluteExpression(ScaleVal)){
1844  Error(Loc, "expected scale expression");
1845  return 0;
1846  }
1847 
1848  // Validate the scale amount.
1849  if (ScaleVal != 1 && ScaleVal != 2 && ScaleVal != 4 && ScaleVal != 8){
1850  Error(Loc, "scale factor in address must be 1, 2, 4 or 8");
1851  return 0;
1852  }
1853  Scale = (unsigned)ScaleVal;
1854  }
1855  }
1856  } else if (getLexer().isNot(AsmToken::RParen)) {
1857  // A scale amount without an index is ignored.
1858  // index.
1859  SMLoc Loc = Parser.getTok().getLoc();
1860 
1861  int64_t Value;
1862  if (getParser().parseAbsoluteExpression(Value))
1863  return 0;
1864 
1865  if (Value != 1)
1866  Warning(Loc, "scale factor without index register is ignored");
1867  Scale = 1;
1868  }
1869  }
1870 
1871  // Ok, we've eaten the memory operand, verify we have a ')' and eat it too.
1872  if (getLexer().isNot(AsmToken::RParen)) {
1873  Error(Parser.getTok().getLoc(), "unexpected token in memory operand");
1874  return 0;
1875  }
1876  SMLoc MemEnd = Parser.getTok().getEndLoc();
1877  Parser.Lex(); // Eat the ')'.
1878 
1879  // If we have both a base register and an index register make sure they are
1880  // both 64-bit or 32-bit registers.
1881  // To support VSIB, IndexReg can be 128-bit or 256-bit registers.
1882  if (BaseReg != 0 && IndexReg != 0) {
1883  if (X86MCRegisterClasses[X86::GR64RegClassID].contains(BaseReg) &&
1884  (X86MCRegisterClasses[X86::GR16RegClassID].contains(IndexReg) ||
1885  X86MCRegisterClasses[X86::GR32RegClassID].contains(IndexReg)) &&
1886  IndexReg != X86::RIZ) {
1887  Error(IndexLoc, "index register is 32-bit, but base register is 64-bit");
1888  return 0;
1889  }
1890  if (X86MCRegisterClasses[X86::GR32RegClassID].contains(BaseReg) &&
1891  (X86MCRegisterClasses[X86::GR16RegClassID].contains(IndexReg) ||
1892  X86MCRegisterClasses[X86::GR64RegClassID].contains(IndexReg)) &&
1893  IndexReg != X86::EIZ){
1894  Error(IndexLoc, "index register is 64-bit, but base register is 32-bit");
1895  return 0;
1896  }
1897  }
1898 
1899  return X86Operand::CreateMem(SegReg, Disp, BaseReg, IndexReg, Scale,
1900  MemStart, MemEnd);
1901 }
1902 
1903 bool X86AsmParser::
1904 ParseInstruction(ParseInstructionInfo &Info, StringRef Name, SMLoc NameLoc,
1906  InstInfo = &Info;
1907  StringRef PatchedName = Name;
1908 
1909  // FIXME: Hack to recognize setneb as setne.
1910  if (PatchedName.startswith("set") && PatchedName.endswith("b") &&
1911  PatchedName != "setb" && PatchedName != "setnb")
1912  PatchedName = PatchedName.substr(0, Name.size()-1);
1913 
1914  // FIXME: Hack to recognize cmp<comparison code>{ss,sd,ps,pd}.
1915  const MCExpr *ExtraImmOp = 0;
1916  if ((PatchedName.startswith("cmp") || PatchedName.startswith("vcmp")) &&
1917  (PatchedName.endswith("ss") || PatchedName.endswith("sd") ||
1918  PatchedName.endswith("ps") || PatchedName.endswith("pd"))) {
1919  bool IsVCMP = PatchedName[0] == 'v';
1920  unsigned SSECCIdx = IsVCMP ? 4 : 3;
1921  unsigned SSEComparisonCode = StringSwitch<unsigned>(
1922  PatchedName.slice(SSECCIdx, PatchedName.size() - 2))
1923  .Case("eq", 0x00)
1924  .Case("lt", 0x01)
1925  .Case("le", 0x02)
1926  .Case("unord", 0x03)
1927  .Case("neq", 0x04)
1928  .Case("nlt", 0x05)
1929  .Case("nle", 0x06)
1930  .Case("ord", 0x07)
1931  /* AVX only from here */
1932  .Case("eq_uq", 0x08)
1933  .Case("nge", 0x09)
1934  .Case("ngt", 0x0A)
1935  .Case("false", 0x0B)
1936  .Case("neq_oq", 0x0C)
1937  .Case("ge", 0x0D)
1938  .Case("gt", 0x0E)
1939  .Case("true", 0x0F)
1940  .Case("eq_os", 0x10)
1941  .Case("lt_oq", 0x11)
1942  .Case("le_oq", 0x12)
1943  .Case("unord_s", 0x13)
1944  .Case("neq_us", 0x14)
1945  .Case("nlt_uq", 0x15)
1946  .Case("nle_uq", 0x16)
1947  .Case("ord_s", 0x17)
1948  .Case("eq_us", 0x18)
1949  .Case("nge_uq", 0x19)
1950  .Case("ngt_uq", 0x1A)
1951  .Case("false_os", 0x1B)
1952  .Case("neq_os", 0x1C)
1953  .Case("ge_oq", 0x1D)
1954  .Case("gt_oq", 0x1E)
1955  .Case("true_us", 0x1F)
1956  .Default(~0U);
1957  if (SSEComparisonCode != ~0U && (IsVCMP || SSEComparisonCode < 8)) {
1958  ExtraImmOp = MCConstantExpr::Create(SSEComparisonCode,
1959  getParser().getContext());
1960  if (PatchedName.endswith("ss")) {
1961  PatchedName = IsVCMP ? "vcmpss" : "cmpss";
1962  } else if (PatchedName.endswith("sd")) {
1963  PatchedName = IsVCMP ? "vcmpsd" : "cmpsd";
1964  } else if (PatchedName.endswith("ps")) {
1965  PatchedName = IsVCMP ? "vcmpps" : "cmpps";
1966  } else {
1967  assert(PatchedName.endswith("pd") && "Unexpected mnemonic!");
1968  PatchedName = IsVCMP ? "vcmppd" : "cmppd";
1969  }
1970  }
1971  }
1972 
1973  Operands.push_back(X86Operand::CreateToken(PatchedName, NameLoc));
1974 
1975  if (ExtraImmOp && !isParsingIntelSyntax())
1976  Operands.push_back(X86Operand::CreateImm(ExtraImmOp, NameLoc, NameLoc));
1977 
1978  // Determine whether this is an instruction prefix.
1979  bool isPrefix =
1980  Name == "lock" || Name == "rep" ||
1981  Name == "repe" || Name == "repz" ||
1982  Name == "repne" || Name == "repnz" ||
1983  Name == "rex64" || Name == "data16";
1984 
1985 
1986  // This does the actual operand parsing. Don't parse any more if we have a
1987  // prefix juxtaposed with an operation like "lock incl 4(%rax)", because we
1988  // just want to parse the "lock" as the first instruction and the "incl" as
1989  // the next one.
1990  if (getLexer().isNot(AsmToken::EndOfStatement) && !isPrefix) {
1991 
1992  // Parse '*' modifier.
1993  if (getLexer().is(AsmToken::Star)) {
1994  SMLoc Loc = Parser.getTok().getLoc();
1995  Operands.push_back(X86Operand::CreateToken("*", Loc));
1996  Parser.Lex(); // Eat the star.
1997  }
1998 
1999  // Read the first operand.
2000  if (X86Operand *Op = ParseOperand())
2001  Operands.push_back(Op);
2002  else {
2003  Parser.eatToEndOfStatement();
2004  return true;
2005  }
2006 
2007  while (getLexer().is(AsmToken::Comma)) {
2008  Parser.Lex(); // Eat the comma.
2009 
2010  // Parse and remember the operand.
2011  if (X86Operand *Op = ParseOperand())
2012  Operands.push_back(Op);
2013  else {
2014  Parser.eatToEndOfStatement();
2015  return true;
2016  }
2017  }
2018 
2019  if (STI.getFeatureBits() & X86::FeatureAVX512) {
2020  // Parse mask register {%k1}
2021  if (getLexer().is(AsmToken::LCurly)) {
2022  SMLoc Loc = Parser.getTok().getLoc();
2023  Operands.push_back(X86Operand::CreateToken("{", Loc));
2024  Parser.Lex(); // Eat the {
2025  if (X86Operand *Op = ParseOperand()) {
2026  Operands.push_back(Op);
2027  if (!getLexer().is(AsmToken::RCurly)) {
2028  SMLoc Loc = getLexer().getLoc();
2029  Parser.eatToEndOfStatement();
2030  return Error(Loc, "Expected } at this point");
2031  }
2032  Loc = Parser.getTok().getLoc();
2033  Operands.push_back(X86Operand::CreateToken("}", Loc));
2034  Parser.Lex(); // Eat the }
2035  } else {
2036  Parser.eatToEndOfStatement();
2037  return true;
2038  }
2039  }
2040  // Parse "zeroing non-masked" semantic {z}
2041  if (getLexer().is(AsmToken::LCurly)) {
2042  SMLoc Loc = Parser.getTok().getLoc();
2043  Operands.push_back(X86Operand::CreateToken("{z}", Loc));
2044  Parser.Lex(); // Eat the {
2045  if (!getLexer().is(AsmToken::Identifier) || getLexer().getTok().getIdentifier() != "z") {
2046  SMLoc Loc = getLexer().getLoc();
2047  Parser.eatToEndOfStatement();
2048  return Error(Loc, "Expected z at this point");
2049  }
2050  Parser.Lex(); // Eat the z
2051  if (!getLexer().is(AsmToken::RCurly)) {
2052  SMLoc Loc = getLexer().getLoc();
2053  Parser.eatToEndOfStatement();
2054  return Error(Loc, "Expected } at this point");
2055  }
2056  Parser.Lex(); // Eat the }
2057  }
2058  }
2059 
2060  if (getLexer().isNot(AsmToken::EndOfStatement)) {
2061  SMLoc Loc = getLexer().getLoc();
2062  Parser.eatToEndOfStatement();
2063  return Error(Loc, "unexpected token in argument list");
2064  }
2065  }
2066 
2067  if (getLexer().is(AsmToken::EndOfStatement))
2068  Parser.Lex(); // Consume the EndOfStatement
2069  else if (isPrefix && getLexer().is(AsmToken::Slash))
2070  Parser.Lex(); // Consume the prefix separator Slash
2071 
2072  if (ExtraImmOp && isParsingIntelSyntax())
2073  Operands.push_back(X86Operand::CreateImm(ExtraImmOp, NameLoc, NameLoc));
2074 
2075  // This is a terrible hack to handle "out[bwl]? %al, (%dx)" ->
2076  // "outb %al, %dx". Out doesn't take a memory form, but this is a widely
2077  // documented form in various unofficial manuals, so a lot of code uses it.
2078  if ((Name == "outb" || Name == "outw" || Name == "outl" || Name == "out") &&
2079  Operands.size() == 3) {
2080  X86Operand &Op = *(X86Operand*)Operands.back();
2081  if (Op.isMem() && Op.Mem.SegReg == 0 &&
2082  isa<MCConstantExpr>(Op.Mem.Disp) &&
2083  cast<MCConstantExpr>(Op.Mem.Disp)->getValue() == 0 &&
2084  Op.Mem.BaseReg == MatchRegisterName("dx") && Op.Mem.IndexReg == 0) {
2085  SMLoc Loc = Op.getEndLoc();
2086  Operands.back() = X86Operand::CreateReg(Op.Mem.BaseReg, Loc, Loc);
2087  delete &Op;
2088  }
2089  }
2090  // Same hack for "in[bwl]? (%dx), %al" -> "inb %dx, %al".
2091  if ((Name == "inb" || Name == "inw" || Name == "inl" || Name == "in") &&
2092  Operands.size() == 3) {
2093  X86Operand &Op = *(X86Operand*)Operands.begin()[1];
2094  if (Op.isMem() && Op.Mem.SegReg == 0 &&
2095  isa<MCConstantExpr>(Op.Mem.Disp) &&
2096  cast<MCConstantExpr>(Op.Mem.Disp)->getValue() == 0 &&
2097  Op.Mem.BaseReg == MatchRegisterName("dx") && Op.Mem.IndexReg == 0) {
2098  SMLoc Loc = Op.getEndLoc();
2099  Operands.begin()[1] = X86Operand::CreateReg(Op.Mem.BaseReg, Loc, Loc);
2100  delete &Op;
2101  }
2102  }
2103  // Transform "ins[bwl] %dx, %es:(%edi)" into "ins[bwl]"
2104  if (Name.startswith("ins") && Operands.size() == 3 &&
2105  (Name == "insb" || Name == "insw" || Name == "insl")) {
2106  X86Operand &Op = *(X86Operand*)Operands.begin()[1];
2107  X86Operand &Op2 = *(X86Operand*)Operands.begin()[2];
2108  if (Op.isReg() && Op.getReg() == X86::DX && isDstOp(Op2)) {
2109  Operands.pop_back();
2110  Operands.pop_back();
2111  delete &Op;
2112  delete &Op2;
2113  }
2114  }
2115 
2116  // Transform "outs[bwl] %ds:(%esi), %dx" into "out[bwl]"
2117  if (Name.startswith("outs") && Operands.size() == 3 &&
2118  (Name == "outsb" || Name == "outsw" || Name == "outsl")) {
2119  X86Operand &Op = *(X86Operand*)Operands.begin()[1];
2120  X86Operand &Op2 = *(X86Operand*)Operands.begin()[2];
2121  if (isSrcOp(Op) && Op2.isReg() && Op2.getReg() == X86::DX) {
2122  Operands.pop_back();
2123  Operands.pop_back();
2124  delete &Op;
2125  delete &Op2;
2126  }
2127  }
2128 
2129  // Transform "movs[bwl] %ds:(%esi), %es:(%edi)" into "movs[bwl]"
2130  if (Name.startswith("movs") && Operands.size() == 3 &&
2131  (Name == "movsb" || Name == "movsw" || Name == "movsl" ||
2132  (is64BitMode() && Name == "movsq"))) {
2133  X86Operand &Op = *(X86Operand*)Operands.begin()[1];
2134  X86Operand &Op2 = *(X86Operand*)Operands.begin()[2];
2135  if (isSrcOp(Op) && isDstOp(Op2)) {
2136  Operands.pop_back();
2137  Operands.pop_back();
2138  delete &Op;
2139  delete &Op2;
2140  }
2141  }
2142  // Transform "lods[bwl] %ds:(%esi),{%al,%ax,%eax,%rax}" into "lods[bwl]"
2143  if (Name.startswith("lods") && Operands.size() == 3 &&
2144  (Name == "lods" || Name == "lodsb" || Name == "lodsw" ||
2145  Name == "lodsl" || (is64BitMode() && Name == "lodsq"))) {
2146  X86Operand *Op1 = static_cast<X86Operand*>(Operands[1]);
2147  X86Operand *Op2 = static_cast<X86Operand*>(Operands[2]);
2148  if (isSrcOp(*Op1) && Op2->isReg()) {
2149  const char *ins;
2150  unsigned reg = Op2->getReg();
2151  bool isLods = Name == "lods";
2152  if (reg == X86::AL && (isLods || Name == "lodsb"))
2153  ins = "lodsb";
2154  else if (reg == X86::AX && (isLods || Name == "lodsw"))
2155  ins = "lodsw";
2156  else if (reg == X86::EAX && (isLods || Name == "lodsl"))
2157  ins = "lodsl";
2158  else if (reg == X86::RAX && (isLods || Name == "lodsq"))
2159  ins = "lodsq";
2160  else
2161  ins = NULL;
2162  if (ins != NULL) {
2163  Operands.pop_back();
2164  Operands.pop_back();
2165  delete Op1;
2166  delete Op2;
2167  if (Name != ins)
2168  static_cast<X86Operand*>(Operands[0])->setTokenValue(ins);
2169  }
2170  }
2171  }
2172  // Transform "stos[bwl] {%al,%ax,%eax,%rax},%es:(%edi)" into "stos[bwl]"
2173  if (Name.startswith("stos") && Operands.size() == 3 &&
2174  (Name == "stos" || Name == "stosb" || Name == "stosw" ||
2175  Name == "stosl" || (is64BitMode() && Name == "stosq"))) {
2176  X86Operand *Op1 = static_cast<X86Operand*>(Operands[1]);
2177  X86Operand *Op2 = static_cast<X86Operand*>(Operands[2]);
2178  if (isDstOp(*Op2) && Op1->isReg()) {
2179  const char *ins;
2180  unsigned reg = Op1->getReg();
2181  bool isStos = Name == "stos";
2182  if (reg == X86::AL && (isStos || Name == "stosb"))
2183  ins = "stosb";
2184  else if (reg == X86::AX && (isStos || Name == "stosw"))
2185  ins = "stosw";
2186  else if (reg == X86::EAX && (isStos || Name == "stosl"))
2187  ins = "stosl";
2188  else if (reg == X86::RAX && (isStos || Name == "stosq"))
2189  ins = "stosq";
2190  else
2191  ins = NULL;
2192  if (ins != NULL) {
2193  Operands.pop_back();
2194  Operands.pop_back();
2195  delete Op1;
2196  delete Op2;
2197  if (Name != ins)
2198  static_cast<X86Operand*>(Operands[0])->setTokenValue(ins);
2199  }
2200  }
2201  }
2202 
2203  // FIXME: Hack to handle recognize s{hr,ar,hl} $1, <op>. Canonicalize to
2204  // "shift <op>".
2205  if ((Name.startswith("shr") || Name.startswith("sar") ||
2206  Name.startswith("shl") || Name.startswith("sal") ||
2207  Name.startswith("rcl") || Name.startswith("rcr") ||
2208  Name.startswith("rol") || Name.startswith("ror")) &&
2209  Operands.size() == 3) {
2210  if (isParsingIntelSyntax()) {
2211  // Intel syntax
2212  X86Operand *Op1 = static_cast<X86Operand*>(Operands[2]);
2213  if (Op1->isImm() && isa<MCConstantExpr>(Op1->getImm()) &&
2214  cast<MCConstantExpr>(Op1->getImm())->getValue() == 1) {
2215  delete Operands[2];
2216  Operands.pop_back();
2217  }
2218  } else {
2219  X86Operand *Op1 = static_cast<X86Operand*>(Operands[1]);
2220  if (Op1->isImm() && isa<MCConstantExpr>(Op1->getImm()) &&
2221  cast<MCConstantExpr>(Op1->getImm())->getValue() == 1) {
2222  delete Operands[1];
2223  Operands.erase(Operands.begin() + 1);
2224  }
2225  }
2226  }
2227 
2228  // Transforms "int $3" into "int3" as a size optimization. We can't write an
2229  // instalias with an immediate operand yet.
2230  if (Name == "int" && Operands.size() == 2) {
2231  X86Operand *Op1 = static_cast<X86Operand*>(Operands[1]);
2232  if (Op1->isImm() && isa<MCConstantExpr>(Op1->getImm()) &&
2233  cast<MCConstantExpr>(Op1->getImm())->getValue() == 3) {
2234  delete Operands[1];
2235  Operands.erase(Operands.begin() + 1);
2236  static_cast<X86Operand*>(Operands[0])->setTokenValue("int3");
2237  }
2238  }
2239 
2240  return false;
2241 }
2242 
2243 static bool convertToSExti8(MCInst &Inst, unsigned Opcode, unsigned Reg,
2244  bool isCmp) {
2245  MCInst TmpInst;
2246  TmpInst.setOpcode(Opcode);
2247  if (!isCmp)
2248  TmpInst.addOperand(MCOperand::CreateReg(Reg));
2249  TmpInst.addOperand(MCOperand::CreateReg(Reg));
2250  TmpInst.addOperand(Inst.getOperand(0));
2251  Inst = TmpInst;
2252  return true;
2253 }
2254 
2255 static bool convert16i16to16ri8(MCInst &Inst, unsigned Opcode,
2256  bool isCmp = false) {
2257  if (!Inst.getOperand(0).isImm() ||
2259  return false;
2260 
2261  return convertToSExti8(Inst, Opcode, X86::AX, isCmp);
2262 }
2263 
2264 static bool convert32i32to32ri8(MCInst &Inst, unsigned Opcode,
2265  bool isCmp = false) {
2266  if (!Inst.getOperand(0).isImm() ||
2268  return false;
2269 
2270  return convertToSExti8(Inst, Opcode, X86::EAX, isCmp);
2271 }
2272 
2273 static bool convert64i32to64ri8(MCInst &Inst, unsigned Opcode,
2274  bool isCmp = false) {
2275  if (!Inst.getOperand(0).isImm() ||
2277  return false;
2278 
2279  return convertToSExti8(Inst, Opcode, X86::RAX, isCmp);
2280 }
2281 
2282 bool X86AsmParser::
2283 processInstruction(MCInst &Inst,
2285  switch (Inst.getOpcode()) {
2286  default: return false;
2287  case X86::AND16i16: return convert16i16to16ri8(Inst, X86::AND16ri8);
2288  case X86::AND32i32: return convert32i32to32ri8(Inst, X86::AND32ri8);
2289  case X86::AND64i32: return convert64i32to64ri8(Inst, X86::AND64ri8);
2290  case X86::XOR16i16: return convert16i16to16ri8(Inst, X86::XOR16ri8);
2291  case X86::XOR32i32: return convert32i32to32ri8(Inst, X86::XOR32ri8);
2292  case X86::XOR64i32: return convert64i32to64ri8(Inst, X86::XOR64ri8);
2293  case X86::OR16i16: return convert16i16to16ri8(Inst, X86::OR16ri8);
2294  case X86::OR32i32: return convert32i32to32ri8(Inst, X86::OR32ri8);
2295  case X86::OR64i32: return convert64i32to64ri8(Inst, X86::OR64ri8);
2296  case X86::CMP16i16: return convert16i16to16ri8(Inst, X86::CMP16ri8, true);
2297  case X86::CMP32i32: return convert32i32to32ri8(Inst, X86::CMP32ri8, true);
2298  case X86::CMP64i32: return convert64i32to64ri8(Inst, X86::CMP64ri8, true);
2299  case X86::ADD16i16: return convert16i16to16ri8(Inst, X86::ADD16ri8);
2300  case X86::ADD32i32: return convert32i32to32ri8(Inst, X86::ADD32ri8);
2301  case X86::ADD64i32: return convert64i32to64ri8(Inst, X86::ADD64ri8);
2302  case X86::SUB16i16: return convert16i16to16ri8(Inst, X86::SUB16ri8);
2303  case X86::SUB32i32: return convert32i32to32ri8(Inst, X86::SUB32ri8);
2304  case X86::SUB64i32: return convert64i32to64ri8(Inst, X86::SUB64ri8);
2305  case X86::ADC16i16: return convert16i16to16ri8(Inst, X86::ADC16ri8);
2306  case X86::ADC32i32: return convert32i32to32ri8(Inst, X86::ADC32ri8);
2307  case X86::ADC64i32: return convert64i32to64ri8(Inst, X86::ADC64ri8);
2308  case X86::SBB16i16: return convert16i16to16ri8(Inst, X86::SBB16ri8);
2309  case X86::SBB32i32: return convert32i32to32ri8(Inst, X86::SBB32ri8);
2310  case X86::SBB64i32: return convert64i32to64ri8(Inst, X86::SBB64ri8);
2311  case X86::VMOVAPDrr:
2312  case X86::VMOVAPDYrr:
2313  case X86::VMOVAPSrr:
2314  case X86::VMOVAPSYrr:
2315  case X86::VMOVDQArr:
2316  case X86::VMOVDQAYrr:
2317  case X86::VMOVDQUrr:
2318  case X86::VMOVDQUYrr:
2319  case X86::VMOVUPDrr:
2320  case X86::VMOVUPDYrr:
2321  case X86::VMOVUPSrr:
2322  case X86::VMOVUPSYrr: {
2323  if (X86II::isX86_64ExtendedReg(Inst.getOperand(0).getReg()) ||
2325  return false;
2326 
2327  unsigned NewOpc;
2328  switch (Inst.getOpcode()) {
2329  default: llvm_unreachable("Invalid opcode");
2330  case X86::VMOVAPDrr: NewOpc = X86::VMOVAPDrr_REV; break;
2331  case X86::VMOVAPDYrr: NewOpc = X86::VMOVAPDYrr_REV; break;
2332  case X86::VMOVAPSrr: NewOpc = X86::VMOVAPSrr_REV; break;
2333  case X86::VMOVAPSYrr: NewOpc = X86::VMOVAPSYrr_REV; break;
2334  case X86::VMOVDQArr: NewOpc = X86::VMOVDQArr_REV; break;
2335  case X86::VMOVDQAYrr: NewOpc = X86::VMOVDQAYrr_REV; break;
2336  case X86::VMOVDQUrr: NewOpc = X86::VMOVDQUrr_REV; break;
2337  case X86::VMOVDQUYrr: NewOpc = X86::VMOVDQUYrr_REV; break;
2338  case X86::VMOVUPDrr: NewOpc = X86::VMOVUPDrr_REV; break;
2339  case X86::VMOVUPDYrr: NewOpc = X86::VMOVUPDYrr_REV; break;
2340  case X86::VMOVUPSrr: NewOpc = X86::VMOVUPSrr_REV; break;
2341  case X86::VMOVUPSYrr: NewOpc = X86::VMOVUPSYrr_REV; break;
2342  }
2343  Inst.setOpcode(NewOpc);
2344  return true;
2345  }
2346  case X86::VMOVSDrr:
2347  case X86::VMOVSSrr: {
2348  if (X86II::isX86_64ExtendedReg(Inst.getOperand(0).getReg()) ||
2350  return false;
2351  unsigned NewOpc;
2352  switch (Inst.getOpcode()) {
2353  default: llvm_unreachable("Invalid opcode");
2354  case X86::VMOVSDrr: NewOpc = X86::VMOVSDrr_REV; break;
2355  case X86::VMOVSSrr: NewOpc = X86::VMOVSSrr_REV; break;
2356  }
2357  Inst.setOpcode(NewOpc);
2358  return true;
2359  }
2360  }
2361 }
2362 
2363 static const char *getSubtargetFeatureName(unsigned Val);
2364 bool X86AsmParser::
2365 MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
2367  MCStreamer &Out, unsigned &ErrorInfo,
2368  bool MatchingInlineAsm) {
2369  assert(!Operands.empty() && "Unexpect empty operand list!");
2370  X86Operand *Op = static_cast<X86Operand*>(Operands[0]);
2371  assert(Op->isToken() && "Leading operand should always be a mnemonic!");
2372  ArrayRef<SMRange> EmptyRanges = None;
2373 
2374  // First, handle aliases that expand to multiple instructions.
2375  // FIXME: This should be replaced with a real .td file alias mechanism.
2376  // Also, MatchInstructionImpl should actually *do* the EmitInstruction
2377  // call.
2378  if (Op->getToken() == "fstsw" || Op->getToken() == "fstcw" ||
2379  Op->getToken() == "fstsww" || Op->getToken() == "fstcww" ||
2380  Op->getToken() == "finit" || Op->getToken() == "fsave" ||
2381  Op->getToken() == "fstenv" || Op->getToken() == "fclex") {
2382  MCInst Inst;
2383  Inst.setOpcode(X86::WAIT);
2384  Inst.setLoc(IDLoc);
2385  if (!MatchingInlineAsm)
2386  Out.EmitInstruction(Inst);
2387 
2388  const char *Repl =
2389  StringSwitch<const char*>(Op->getToken())
2390  .Case("finit", "fninit")
2391  .Case("fsave", "fnsave")
2392  .Case("fstcw", "fnstcw")
2393  .Case("fstcww", "fnstcw")
2394  .Case("fstenv", "fnstenv")
2395  .Case("fstsw", "fnstsw")
2396  .Case("fstsww", "fnstsw")
2397  .Case("fclex", "fnclex")
2398  .Default(0);
2399  assert(Repl && "Unknown wait-prefixed instruction");
2400  delete Operands[0];
2401  Operands[0] = X86Operand::CreateToken(Repl, IDLoc);
2402  }
2403 
2404  bool WasOriginallyInvalidOperand = false;
2405  MCInst Inst;
2406 
2407  // First, try a direct match.
2408  switch (MatchInstructionImpl(Operands, Inst,
2409  ErrorInfo, MatchingInlineAsm,
2410  isParsingIntelSyntax())) {
2411  default: break;
2412  case Match_Success:
2413  // Some instructions need post-processing to, for example, tweak which
2414  // encoding is selected. Loop on it while changes happen so the
2415  // individual transformations can chain off each other.
2416  if (!MatchingInlineAsm)
2417  while (processInstruction(Inst, Operands))
2418  ;
2419 
2420  Inst.setLoc(IDLoc);
2421  if (!MatchingInlineAsm)
2422  Out.EmitInstruction(Inst);
2423  Opcode = Inst.getOpcode();
2424  return false;
2425  case Match_MissingFeature: {
2426  assert(ErrorInfo && "Unknown missing feature!");
2427  // Special case the error message for the very common case where only
2428  // a single subtarget feature is missing.
2429  std::string Msg = "instruction requires:";
2430  unsigned Mask = 1;
2431  for (unsigned i = 0; i < (sizeof(ErrorInfo)*8-1); ++i) {
2432  if (ErrorInfo & Mask) {
2433  Msg += " ";
2434  Msg += getSubtargetFeatureName(ErrorInfo & Mask);
2435  }
2436  Mask <<= 1;
2437  }
2438  return Error(IDLoc, Msg, EmptyRanges, MatchingInlineAsm);
2439  }
2440  case Match_InvalidOperand:
2441  WasOriginallyInvalidOperand = true;
2442  break;
2443  case Match_MnemonicFail:
2444  break;
2445  }
2446 
2447  // FIXME: Ideally, we would only attempt suffix matches for things which are
2448  // valid prefixes, and we could just infer the right unambiguous
2449  // type. However, that requires substantially more matcher support than the
2450  // following hack.
2451 
2452  // Change the operand to point to a temporary token.
2453  StringRef Base = Op->getToken();
2454  SmallString<16> Tmp;
2455  Tmp += Base;
2456  Tmp += ' ';
2457  Op->setTokenValue(Tmp.str());
2458 
2459  // If this instruction starts with an 'f', then it is a floating point stack
2460  // instruction. These come in up to three forms for 32-bit, 64-bit, and
2461  // 80-bit floating point, which use the suffixes s,l,t respectively.
2462  //
2463  // Otherwise, we assume that this may be an integer instruction, which comes
2464  // in 8/16/32/64-bit forms using the b,w,l,q suffixes respectively.
2465  const char *Suffixes = Base[0] != 'f' ? "bwlq" : "slt\0";
2466 
2467  // Check for the various suffix matches.
2468  Tmp[Base.size()] = Suffixes[0];
2469  unsigned ErrorInfoIgnore;
2470  unsigned ErrorInfoMissingFeature = 0; // Init suppresses compiler warnings.
2471  unsigned Match1, Match2, Match3, Match4;
2472 
2473  Match1 = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore,
2474  MatchingInlineAsm, isParsingIntelSyntax());
2475  // If this returned as a missing feature failure, remember that.
2476  if (Match1 == Match_MissingFeature)
2477  ErrorInfoMissingFeature = ErrorInfoIgnore;
2478  Tmp[Base.size()] = Suffixes[1];
2479  Match2 = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore,
2480  MatchingInlineAsm, isParsingIntelSyntax());
2481  // If this returned as a missing feature failure, remember that.
2482  if (Match2 == Match_MissingFeature)
2483  ErrorInfoMissingFeature = ErrorInfoIgnore;
2484  Tmp[Base.size()] = Suffixes[2];
2485  Match3 = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore,
2486  MatchingInlineAsm, isParsingIntelSyntax());
2487  // If this returned as a missing feature failure, remember that.
2488  if (Match3 == Match_MissingFeature)
2489  ErrorInfoMissingFeature = ErrorInfoIgnore;
2490  Tmp[Base.size()] = Suffixes[3];
2491  Match4 = MatchInstructionImpl(Operands, Inst, ErrorInfoIgnore,
2492  MatchingInlineAsm, isParsingIntelSyntax());
2493  // If this returned as a missing feature failure, remember that.
2494  if (Match4 == Match_MissingFeature)
2495  ErrorInfoMissingFeature = ErrorInfoIgnore;
2496 
2497  // Restore the old token.
2498  Op->setTokenValue(Base);
2499 
2500  // If exactly one matched, then we treat that as a successful match (and the
2501  // instruction will already have been filled in correctly, since the failing
2502  // matches won't have modified it).
2503  unsigned NumSuccessfulMatches =
2504  (Match1 == Match_Success) + (Match2 == Match_Success) +
2505  (Match3 == Match_Success) + (Match4 == Match_Success);
2506  if (NumSuccessfulMatches == 1) {
2507  Inst.setLoc(IDLoc);
2508  if (!MatchingInlineAsm)
2509  Out.EmitInstruction(Inst);
2510  Opcode = Inst.getOpcode();
2511  return false;
2512  }
2513 
2514  // Otherwise, the match failed, try to produce a decent error message.
2515 
2516  // If we had multiple suffix matches, then identify this as an ambiguous
2517  // match.
2518  if (NumSuccessfulMatches > 1) {
2519  char MatchChars[4];
2520  unsigned NumMatches = 0;
2521  if (Match1 == Match_Success) MatchChars[NumMatches++] = Suffixes[0];
2522  if (Match2 == Match_Success) MatchChars[NumMatches++] = Suffixes[1];
2523  if (Match3 == Match_Success) MatchChars[NumMatches++] = Suffixes[2];
2524  if (Match4 == Match_Success) MatchChars[NumMatches++] = Suffixes[3];
2525 
2526  SmallString<126> Msg;
2527  raw_svector_ostream OS(Msg);
2528  OS << "ambiguous instructions require an explicit suffix (could be ";
2529  for (unsigned i = 0; i != NumMatches; ++i) {
2530  if (i != 0)
2531  OS << ", ";
2532  if (i + 1 == NumMatches)
2533  OS << "or ";
2534  OS << "'" << Base << MatchChars[i] << "'";
2535  }
2536  OS << ")";
2537  Error(IDLoc, OS.str(), EmptyRanges, MatchingInlineAsm);
2538  return true;
2539  }
2540 
2541  // Okay, we know that none of the variants matched successfully.
2542 
2543  // If all of the instructions reported an invalid mnemonic, then the original
2544  // mnemonic was invalid.
2545  if ((Match1 == Match_MnemonicFail) && (Match2 == Match_MnemonicFail) &&
2546  (Match3 == Match_MnemonicFail) && (Match4 == Match_MnemonicFail)) {
2547  if (!WasOriginallyInvalidOperand) {
2548  ArrayRef<SMRange> Ranges = MatchingInlineAsm ? EmptyRanges :
2549  Op->getLocRange();
2550  return Error(IDLoc, "invalid instruction mnemonic '" + Base + "'",
2551  Ranges, MatchingInlineAsm);
2552  }
2553 
2554  // Recover location info for the operand if we know which was the problem.
2555  if (ErrorInfo != ~0U) {
2556  if (ErrorInfo >= Operands.size())
2557  return Error(IDLoc, "too few operands for instruction",
2558  EmptyRanges, MatchingInlineAsm);
2559 
2560  X86Operand *Operand = (X86Operand*)Operands[ErrorInfo];
2561  if (Operand->getStartLoc().isValid()) {
2562  SMRange OperandRange = Operand->getLocRange();
2563  return Error(Operand->getStartLoc(), "invalid operand for instruction",
2564  OperandRange, MatchingInlineAsm);
2565  }
2566  }
2567 
2568  return Error(IDLoc, "invalid operand for instruction", EmptyRanges,
2569  MatchingInlineAsm);
2570  }
2571 
2572  // If one instruction matched with a missing feature, report this as a
2573  // missing feature.
2574  if ((Match1 == Match_MissingFeature) + (Match2 == Match_MissingFeature) +
2575  (Match3 == Match_MissingFeature) + (Match4 == Match_MissingFeature) == 1){
2576  std::string Msg = "instruction requires:";
2577  unsigned Mask = 1;
2578  for (unsigned i = 0; i < (sizeof(ErrorInfoMissingFeature)*8-1); ++i) {
2579  if (ErrorInfoMissingFeature & Mask) {
2580  Msg += " ";
2581  Msg += getSubtargetFeatureName(ErrorInfoMissingFeature & Mask);
2582  }
2583  Mask <<= 1;
2584  }
2585  return Error(IDLoc, Msg, EmptyRanges, MatchingInlineAsm);
2586  }
2587 
2588  // If one instruction matched with an invalid operand, report this as an
2589  // operand failure.
2590  if ((Match1 == Match_InvalidOperand) + (Match2 == Match_InvalidOperand) +
2591  (Match3 == Match_InvalidOperand) + (Match4 == Match_InvalidOperand) == 1){
2592  Error(IDLoc, "invalid operand for instruction", EmptyRanges,
2593  MatchingInlineAsm);
2594  return true;
2595  }
2596 
2597  // If all of these were an outright failure, report it in a useless way.
2598  Error(IDLoc, "unknown use of instruction mnemonic without a size suffix",
2599  EmptyRanges, MatchingInlineAsm);
2600  return true;
2601 }
2602 
2603 
2604 bool X86AsmParser::ParseDirective(AsmToken DirectiveID) {
2605  StringRef IDVal = DirectiveID.getIdentifier();
2606  if (IDVal == ".word")
2607  return ParseDirectiveWord(2, DirectiveID.getLoc());
2608  else if (IDVal.startswith(".code"))
2609  return ParseDirectiveCode(IDVal, DirectiveID.getLoc());
2610  else if (IDVal.startswith(".att_syntax")) {
2611  getParser().setAssemblerDialect(0);
2612  return false;
2613  } else if (IDVal.startswith(".intel_syntax")) {
2614  getParser().setAssemblerDialect(1);
2615  if (getLexer().isNot(AsmToken::EndOfStatement)) {
2616  if(Parser.getTok().getString() == "noprefix") {
2617  // FIXME : Handle noprefix
2618  Parser.Lex();
2619  } else
2620  return true;
2621  }
2622  return false;
2623  }
2624  return true;
2625 }
2626 
2627 /// ParseDirectiveWord
2628 /// ::= .word [ expression (, expression)* ]
2629 bool X86AsmParser::ParseDirectiveWord(unsigned Size, SMLoc L) {
2630  if (getLexer().isNot(AsmToken::EndOfStatement)) {
2631  for (;;) {
2632  const MCExpr *Value;
2633  if (getParser().parseExpression(Value))
2634  return true;
2635 
2636  getParser().getStreamer().EmitValue(Value, Size);
2637 
2638  if (getLexer().is(AsmToken::EndOfStatement))
2639  break;
2640 
2641  // FIXME: Improve diagnostic.
2642  if (getLexer().isNot(AsmToken::Comma))
2643  return Error(L, "unexpected token in directive");
2644  Parser.Lex();
2645  }
2646  }
2647 
2648  Parser.Lex();
2649  return false;
2650 }
2651 
2652 /// ParseDirectiveCode
2653 /// ::= .code32 | .code64
2654 bool X86AsmParser::ParseDirectiveCode(StringRef IDVal, SMLoc L) {
2655  if (IDVal == ".code32") {
2656  Parser.Lex();
2657  if (is64BitMode()) {
2658  SwitchMode();
2659  getParser().getStreamer().EmitAssemblerFlag(MCAF_Code32);
2660  }
2661  } else if (IDVal == ".code64") {
2662  Parser.Lex();
2663  if (!is64BitMode()) {
2664  SwitchMode();
2665  getParser().getStreamer().EmitAssemblerFlag(MCAF_Code64);
2666  }
2667  } else {
2668  return Error(L, "unexpected directive " + IDVal);
2669  }
2670 
2671  return false;
2672 }
2673 
2674 // Force static initialization.
2675 extern "C" void LLVMInitializeX86AsmParser() {
2678 }
2679 
2680 #define GET_REGISTER_MATCHER
2681 #define GET_MATCHER_IMPLEMENTATION
2682 #define GET_SUBTARGET_FEATURE_NAME
2683 #include "X86GenAsmMatcher.inc"
static bool isReg(const MCInst &MI, unsigned OpNo)
static bool isImmSExti16i8Value(uint64_t Value)
}
static bool isImmSExti32i8Value(uint64_t Value)
uint64_t getZExtValue() const
Get zero extended value.
Definition: APInt.h:1306
bool isX86_64NonExtLowByteReg(unsigned reg)
Definition: X86BaseInfo.h:695
const char * getPointer() const
Definition: SMLoc.h:33
size_t size() const
size - Get the string size.
Definition: StringRef.h:113
static MCOperand CreateReg(unsigned Reg)
Definition: MCInst.h:111
static const MCConstantExpr * Create(int64_t Value, MCContext &Ctx)
Definition: MCExpr.cpp:152
An abstraction for memory operations.
Definition: Memory.h:45
MCTargetAsmParser - Generic interface to target specific assembly parsers.
enable_if_c<!is_simple_type< Y >::value, typename cast_retty< X, const Y >::ret_type >::type dyn_cast(const Y &Val)
Definition: Casting.h:266
bool endswith(StringRef Suffix) const
Check if this string ends with the given Suffix.
Definition: StringRef.h:217
StringRef getString() const
Definition: MCAsmLexer.h:95
static MCOperand CreateExpr(const MCExpr *Val)
Definition: MCInst.h:129
StringRef substr(size_t Start, size_t N=npos) const
Definition: StringRef.h:392
bool isNot(TokenKind K) const
Definition: MCAsmLexer.h:69
virtual void EmitInstruction(const MCInst &Inst)=0
static void RewriteIntelBracExpression(SmallVectorImpl< AsmRewrite > *AsmRewrites, StringRef SymName, int64_t ImmDisp, int64_t FinalImmDisp, SMLoc &BracLoc, SMLoc &StartInBrac, SMLoc &End)
StringSwitch & Case(const char(&S)[N], const T &Value)
Definition: StringSwitch.h:55
LLVM_ATTRIBUTE_NORETURN void report_fatal_error(const char *reason, bool gen_crash_diag=true)
static bool isImmZExtu32u8Value(uint64_t Value)
void LLVMInitializeX86AsmParser()
std::pair< StringRef, StringRef > getToken(StringRef Source, StringRef Delimiters=" \t\n\v\f\r")
T LLVM_ATTRIBUTE_UNUSED_RESULT pop_back_val()
Definition: SmallVector.h:430
#define llvm_unreachable(msg)
Target TheX86_64Target
AsmToken - Target independent representation for an assembler token.
Definition: MCAsmLexer.h:21
static unsigned getIntelMemOperandSize(StringRef OpStr)
getIntelMemOperandSize - Return intel memory operand size.
static bool convert16i16to16ri8(MCInst &Inst, unsigned Opcode, bool isCmp=false)
static bool isImmSExti64i32Value(uint64_t Value)
bool LLVM_ATTRIBUTE_UNUSED_RESULT empty() const
Definition: SmallVector.h:56
unsigned getReg() const
getReg - Returns the register number.
Definition: MCInst.h:63
const char * data() const
Definition: StringRef.h:107
int64_t getIntVal() const
Definition: MCAsmLexer.h:100
bool isX86_64ExtendedReg(unsigned RegNo)
Definition: X86BaseInfo.h:660
bool isImm() const
Definition: MCInst.h:57
static const MCSymbolRefExpr * Create(const MCSymbol *Symbol, MCContext &Ctx)
Definition: MCExpr.h:270
A switch()-like statement whose cases are string literals.
Definition: StringSwitch.h:42
int64_t getValue() const
Definition: MCExpr.h:126
const MCInstrInfo & MII
SMLoc getLoc() const
Definition: MCAsmLexer.cpp:26
Target TheX86_32Target
This file declares a class to represent arbitrary precision floating point values and provide a varie...
iterator erase(iterator I)
Definition: SmallVector.h:478
void setLoc(SMLoc loc)
Definition: MCInst.h:160
static bool convert64i32to64ri8(MCInst &Inst, unsigned Opcode, bool isCmp=false)
void setOpcode(unsigned Op)
Definition: MCInst.h:157
bool startswith(StringRef Prefix) const
Check if this string starts with the given Prefix.
Definition: StringRef.h:208
StringRef drop_front(size_t N=1) const
Definition: StringRef.h:399
static unsigned MatchRegisterName(StringRef Name)
static const char * getSubtargetFeatureName(unsigned Val)
Promote Memory to Register
Definition: Mem2Reg.cpp:54
static bool isImmSExti64i8Value(uint64_t Value)
bool is(TokenKind K) const
Definition: MCAsmLexer.h:68
R Default(const T &Value) const
Definition: StringSwitch.h:111
unsigned getOpcode() const
Definition: MCInst.h:158
Class for arbitrary precision integers.
Definition: APInt.h:75
StringRef str() const
Explicit conversion to StringRef.
Definition: SmallString.h:270
static bool convertToSExti8(MCInst &Inst, unsigned Opcode, unsigned Reg, bool isCmp)
int64_t getImm() const
Definition: MCInst.h:74
.code32 (X86) / .code 32 (ARM)
Definition: MCDirectives.h:51
static SMLoc getFromPointer(const char *Ptr)
Definition: SMLoc.h:35
.code64 (X86)
Definition: MCDirectives.h:52
StringRef getIdentifier() const
Definition: MCAsmLexer.h:84
static MCOperand CreateImm(int64_t Val)
Definition: MCInst.h:117
#define I(x, y, z)
Definition: MD5.cpp:54
#define N
static bool convert32i32to32ri8(MCInst &Inst, unsigned Opcode, bool isCmp=false)
static unsigned getReg(const void *D, unsigned RC, unsigned RegNo)
static bool isMem(const MachineInstr *MI, unsigned Op)
Definition: X86InstrInfo.h:123
IntelOperatorKind
LLVM Value Representation.
Definition: Value.h:66
SMLoc getEndLoc() const
Definition: MCAsmLexer.cpp:30
StringSwitch & Cases(const char(&S0)[N0], const char(&S1)[N1], const T &Value)
Definition: StringSwitch.h:85
void addOperand(const MCOperand &Op)
Definition: MCInst.h:167
static GCMetadataPrinterRegistry::Add< OcamlGCMetadataPrinter > Y("ocaml","ocaml 3.10-compatible collector")
StringRef slice(size_t Start, size_t End) const
Definition: StringRef.h:421
Represents a location in source code.
Definition: SMLoc.h:23
static RegisterPass< NVPTXAllocaHoisting > X("alloca-hoisting","Hoisting alloca instructions in non-entry ""blocks to the entry block")
std::string lower() const
Definition: StringRef.cpp:118
const MCOperand & getOperand(unsigned i) const
Definition: MCInst.h:163