Coverage Report

Created: 2019-02-23 12:57

/Users/buildslave/jenkins/workspace/clang-stage2-coverage-R/llvm/include/llvm/MC/MCParser/MCTargetAsmParser.h
Line
Count
Source (jump to first uncovered line)
1
//===- llvm/MC/MCTargetAsmParser.h - Target Assembly Parser -----*- C++ -*-===//
2
//
3
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4
// See https://llvm.org/LICENSE.txt for license information.
5
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6
//
7
//===----------------------------------------------------------------------===//
8
9
#ifndef LLVM_MC_MCPARSER_MCTARGETASMPARSER_H
10
#define LLVM_MC_MCPARSER_MCTARGETASMPARSER_H
11
12
#include "llvm/ADT/StringRef.h"
13
#include "llvm/MC/MCExpr.h"
14
#include "llvm/MC/MCInstrInfo.h"
15
#include "llvm/MC/MCParser/MCAsmLexer.h"
16
#include "llvm/MC/MCParser/MCParsedAsmOperand.h"
17
#include "llvm/MC/MCParser/MCAsmParserExtension.h"
18
#include "llvm/MC/MCTargetOptions.h"
19
#include "llvm/Support/SMLoc.h"
20
#include <cstdint>
21
#include <memory>
22
23
namespace llvm {
24
25
class MCInst;
26
class MCParsedAsmOperand;
27
class MCStreamer;
28
class MCSubtargetInfo;
29
template <typename T> class SmallVectorImpl;
30
31
using OperandVector = SmallVectorImpl<std::unique_ptr<MCParsedAsmOperand>>;
32
33
enum AsmRewriteKind {
34
  AOK_Align,          // Rewrite align as .align.
35
  AOK_EVEN,           // Rewrite even as .even.
36
  AOK_Emit,           // Rewrite _emit as .byte.
37
  AOK_Input,          // Rewrite in terms of $N.
38
  AOK_Output,         // Rewrite in terms of $N.
39
  AOK_SizeDirective,  // Add a sizing directive (e.g., dword ptr).
40
  AOK_Label,          // Rewrite local labels.
41
  AOK_EndOfStatement, // Add EndOfStatement (e.g., "\n\t").
42
  AOK_Skip,           // Skip emission (e.g., offset/type operators).
43
  AOK_IntelExpr       // SizeDirective SymDisp [BaseReg + IndexReg * Scale + ImmDisp]
44
};
45
46
const char AsmRewritePrecedence [] = {
47
  2, // AOK_Align
48
  2, // AOK_EVEN
49
  2, // AOK_Emit
50
  3, // AOK_Input
51
  3, // AOK_Output
52
  5, // AOK_SizeDirective
53
  1, // AOK_Label
54
  5, // AOK_EndOfStatement
55
  2, // AOK_Skip
56
  2  // AOK_IntelExpr
57
};
58
59
// Represnt the various parts which makes up an intel expression,
60
// used for emitting compound intel expressions
61
0
struct IntelExpr {
62
  bool NeedBracs;
63
  int64_t Imm;
64
  StringRef BaseReg;
65
  StringRef IndexReg;
66
  unsigned Scale;
67
68
  IntelExpr(bool needBracs = false) : NeedBracs(needBracs), Imm(0),
69
    BaseReg(StringRef()), IndexReg(StringRef()),
70
743
    Scale(1) {}
71
  // Compund immediate expression
72
202
  IntelExpr(int64_t imm, bool needBracs) : IntelExpr(needBracs) {
73
202
    Imm = imm;
74
202
  }
75
  // [Reg + ImmediateExpression]
76
  // We don't bother to emit an immediate expression evaluated to zero
77
  IntelExpr(StringRef reg, int64_t imm = 0, unsigned scale = 0,
78
    bool needBracs = true) :
79
202
    IntelExpr(imm, needBracs) {
80
202
    IndexReg = reg;
81
202
    if (scale)
82
3
      Scale = scale;
83
202
  }
84
  // [BaseReg + IndexReg * ScaleExpression + ImmediateExpression]
85
  IntelExpr(StringRef baseReg, StringRef indexReg, unsigned scale = 0,
86
    int64_t imm = 0, bool needBracs = true) :
87
202
    IntelExpr(indexReg, imm, scale, needBracs) {
88
202
    BaseReg = baseReg;
89
202
  }
90
420
  bool hasBaseReg() const {
91
420
    return BaseReg.size();
92
420
  }
93
366
  bool hasIndexReg() const {
94
366
    return IndexReg.size();
95
366
  }
96
215
  bool hasRegs() const {
97
215
    return hasBaseReg() || 
hasIndexReg()165
;
98
215
  }
99
0
  bool isValid() const {
100
0
    return (Scale == 1) ||
101
0
           (hasIndexReg() && (Scale == 2 || Scale == 4 || Scale == 8));
102
0
  }
103
};
104
105
struct AsmRewrite {
106
  AsmRewriteKind Kind;
107
  SMLoc Loc;
108
  unsigned Len;
109
  int64_t Val;
110
  StringRef Label;
111
  IntelExpr IntelExp;
112
113
public:
114
  AsmRewrite(AsmRewriteKind kind, SMLoc loc, unsigned len = 0, int64_t val = 0)
115
541
    : Kind(kind), Loc(loc), Len(len), Val(val) {}
116
  AsmRewrite(AsmRewriteKind kind, SMLoc loc, unsigned len, StringRef label)
117
43
    : AsmRewrite(kind, loc, len) { Label = label; }
118
  AsmRewrite(SMLoc loc, unsigned len, IntelExpr exp)
119
202
    : AsmRewrite(AOK_IntelExpr, loc, len) { IntelExp = exp; }
120
};
121
122
struct ParseInstructionInfo {
123
  SmallVectorImpl<AsmRewrite> *AsmRewrites = nullptr;
124
125
  ParseInstructionInfo() = default;
126
  ParseInstructionInfo(SmallVectorImpl<AsmRewrite> *rewrites)
127
471k
    : AsmRewrites(rewrites) {}
128
};
129
130
enum OperandMatchResultTy {
131
  MatchOperand_Success,  // operand matched successfully
132
  MatchOperand_NoMatch,  // operand did not match
133
  MatchOperand_ParseFail // operand matched but had errors
134
};
135
136
enum class DiagnosticPredicateTy {
137
  Match,
138
  NearMatch,
139
  NoMatch,
140
};
141
142
// When an operand is parsed, the assembler will try to iterate through a set of
143
// possible operand classes that the operand might match and call the
144
// corresponding PredicateMethod to determine that.
145
//
146
// If there are two AsmOperands that would give a specific diagnostic if there
147
// is no match, there is currently no mechanism to distinguish which operand is
148
// a closer match. The DiagnosticPredicate distinguishes between 'completely
149
// no match' and 'near match', so the assembler can decide whether to give a
150
// specific diagnostic, or use 'InvalidOperand' and continue to find a
151
// 'better matching' diagnostic.
152
//
153
// For example:
154
//    opcode opnd0, onpd1, opnd2
155
//
156
// where:
157
//    opnd2 could be an 'immediate of range [-8, 7]'
158
//    opnd2 could be a  'register + shift/extend'.
159
//
160
// If opnd2 is a valid register, but with a wrong shift/extend suffix, it makes
161
// little sense to give a diagnostic that the operand should be an immediate
162
// in range [-8, 7].
163
//
164
// This is a light-weight alternative to the 'NearMissInfo' approach
165
// below which collects *all* possible diagnostics. This alternative
166
// is optional and fully backward compatible with existing
167
// PredicateMethods that return a 'bool' (match or no match).
168
struct DiagnosticPredicate {
169
  DiagnosticPredicateTy Type;
170
171
  explicit DiagnosticPredicate(bool Match)
172
      : Type(Match ? DiagnosticPredicateTy::Match
173
2.45M
                   : DiagnosticPredicateTy::NearMatch) {}
174
346k
  DiagnosticPredicate(DiagnosticPredicateTy T) : Type(T) {}
175
  DiagnosticPredicate(const DiagnosticPredicate &) = default;
176
177
4.66k
  operator bool() const { return Type == DiagnosticPredicateTy::Match; }
178
2.79M
  bool isMatch() const { return Type == DiagnosticPredicateTy::Match; }
179
261k
  bool isNearMatch() const { return Type == DiagnosticPredicateTy::NearMatch; }
180
0
  bool isNoMatch() const { return Type == DiagnosticPredicateTy::NoMatch; }
181
};
182
183
// When matching of an assembly instruction fails, there may be multiple
184
// encodings that are close to being a match. It's often ambiguous which one
185
// the programmer intended to use, so we want to report an error which mentions
186
// each of these "near-miss" encodings. This struct contains information about
187
// one such encoding, and why it did not match the parsed instruction.
188
class NearMissInfo {
189
public:
190
  enum NearMissKind {
191
    NoNearMiss,
192
    NearMissOperand,
193
    NearMissFeature,
194
    NearMissPredicate,
195
    NearMissTooFewOperands,
196
  };
197
198
  // The encoding is valid for the parsed assembly string. This is only used
199
  // internally to the table-generated assembly matcher.
200
1.65M
  static NearMissInfo getSuccess() { return NearMissInfo(); }
201
202
  // The instruction encoding is not valid because it requires some target
203
  // features that are not currently enabled. MissingFeatures has a bit set for
204
  // each feature that the encoding needs but which is not enabled.
205
34.8k
  static NearMissInfo getMissedFeature(uint64_t MissingFeatures) {
206
34.8k
    NearMissInfo Result;
207
34.8k
    Result.Kind = NearMissFeature;
208
34.8k
    Result.Features = MissingFeatures;
209
34.8k
    return Result;
210
34.8k
  }
211
212
  // The instruction encoding is not valid because the target-specific
213
  // predicate function returned an error code. FailureCode is the
214
  // target-specific error code returned by the predicate.
215
221
  static NearMissInfo getMissedPredicate(unsigned FailureCode) {
216
221
    NearMissInfo Result;
217
221
    Result.Kind = NearMissPredicate;
218
221
    Result.PredicateError = FailureCode;
219
221
    return Result;
220
221
  }
221
222
  // The instruction encoding is not valid because one (and only one) parsed
223
  // operand is not of the correct type. OperandError is the error code
224
  // relating to the operand class expected by the encoding. OperandClass is
225
  // the type of the expected operand. Opcode is the opcode of the encoding.
226
  // OperandIndex is the index into the parsed operand list.
227
  static NearMissInfo getMissedOperand(unsigned OperandError,
228
                                       unsigned OperandClass, unsigned Opcode,
229
369k
                                       unsigned OperandIndex) {
230
369k
    NearMissInfo Result;
231
369k
    Result.Kind = NearMissOperand;
232
369k
    Result.MissedOperand.Error = OperandError;
233
369k
    Result.MissedOperand.Class = OperandClass;
234
369k
    Result.MissedOperand.Opcode = Opcode;
235
369k
    Result.MissedOperand.Index = OperandIndex;
236
369k
    return Result;
237
369k
  }
238
239
  // The instruction encoding is not valid because it expects more operands
240
  // than were parsed. OperandClass is the class of the expected operand that
241
  // was not provided. Opcode is the instruction encoding.
242
  static NearMissInfo getTooFewOperands(unsigned OperandClass,
243
823
                                        unsigned Opcode) {
244
823
    NearMissInfo Result;
245
823
    Result.Kind = NearMissTooFewOperands;
246
823
    Result.TooFewOperands.Class = OperandClass;
247
823
    Result.TooFewOperands.Opcode = Opcode;
248
823
    return Result;
249
823
  }
250
251
1.08M
  operator bool() const { return Kind != NoNearMiss; }
252
253
11.2k
  NearMissKind getKind() const { return Kind; }
254
255
  // Feature flags required by the instruction, that the current target does
256
  // not have.
257
2.40k
  uint64_t getFeatures() const {
258
2.40k
    assert(Kind == NearMissFeature);
259
2.40k
    return Features;
260
2.40k
  }
261
  // Error code returned by the target predicate when validating this
262
  // instruction encoding.
263
33
  unsigned getPredicateError() const {
264
33
    assert(Kind == NearMissPredicate);
265
33
    return PredicateError;
266
33
  }
267
  // MatchClassKind of the operand that we expected to see.
268
3.44k
  unsigned getOperandClass() const {
269
3.44k
    assert(Kind == NearMissOperand || Kind == NearMissTooFewOperands);
270
3.44k
    return MissedOperand.Class;
271
3.44k
  }
272
  // Opcode of the encoding we were trying to match.
273
0
  unsigned getOpcode() const {
274
0
    assert(Kind == NearMissOperand || Kind == NearMissTooFewOperands);
275
0
    return MissedOperand.Opcode;
276
0
  }
277
  // Error code returned when validating the operand.
278
4.93k
  unsigned getOperandError() const {
279
4.93k
    assert(Kind == NearMissOperand);
280
4.93k
    return MissedOperand.Error;
281
4.93k
  }
282
  // Index of the actual operand we were trying to match in the list of parsed
283
  // operands.
284
13.2k
  unsigned getOperandIndex() const {
285
13.2k
    assert(Kind == NearMissOperand);
286
13.2k
    return MissedOperand.Index;
287
13.2k
  }
288
289
private:
290
  NearMissKind Kind;
291
292
  // These two structs share a common prefix, so we can safely rely on the fact
293
  // that they overlap in the union.
294
  struct MissedOpInfo {
295
    unsigned Class;
296
    unsigned Opcode;
297
    unsigned Error;
298
    unsigned Index;
299
  };
300
301
  struct TooFewOperandsInfo {
302
    unsigned Class;
303
    unsigned Opcode;
304
  };
305
306
  union {
307
    uint64_t Features;
308
    unsigned PredicateError;
309
    MissedOpInfo MissedOperand;
310
    TooFewOperandsInfo TooFewOperands;
311
  };
312
313
2.05M
  NearMissInfo() : Kind(NoNearMiss) {}
314
};
315
316
/// MCTargetAsmParser - Generic interface to target specific assembly parsers.
317
class MCTargetAsmParser : public MCAsmParserExtension {
318
public:
319
  enum MatchResultTy {
320
    Match_InvalidOperand,
321
    Match_InvalidTiedOperand,
322
    Match_MissingFeature,
323
    Match_MnemonicFail,
324
    Match_Success,
325
    Match_NearMisses,
326
    FIRST_TARGET_MATCH_RESULT_TY
327
  };
328
329
protected: // Can only create subclasses.
330
  MCTargetAsmParser(MCTargetOptions const &, const MCSubtargetInfo &STI,
331
                    const MCInstrInfo &MII);
332
333
  /// Create a copy of STI and return a non-const reference to it.
334
  MCSubtargetInfo &copySTI();
335
336
  /// AvailableFeatures - The current set of available features.
337
  uint64_t AvailableFeatures = 0;
338
339
  /// ParsingInlineAsm - Are we parsing ms-style inline assembly?
340
  bool ParsingInlineAsm = false;
341
342
  /// SemaCallback - The Sema callback implementation.  Must be set when parsing
343
  /// ms-style inline assembly.
344
  MCAsmParserSemaCallback *SemaCallback;
345
346
  /// Set of options which affects instrumentation of inline assembly.
347
  MCTargetOptions MCOptions;
348
349
  /// Current STI.
350
  const MCSubtargetInfo *STI;
351
352
  const MCInstrInfo &MII;
353
354
public:
355
  MCTargetAsmParser(const MCTargetAsmParser &) = delete;
356
  MCTargetAsmParser &operator=(const MCTargetAsmParser &) = delete;
357
358
  ~MCTargetAsmParser() override;
359
360
  const MCSubtargetInfo &getSTI() const;
361
362
1.59M
  uint64_t getAvailableFeatures() const { return AvailableFeatures; }
363
160k
  void setAvailableFeatures(uint64_t Value) { AvailableFeatures = Value; }
364
365
890k
  bool isParsingInlineAsm () { return ParsingInlineAsm; }
366
225
  void setParsingInlineAsm (bool Value) { ParsingInlineAsm = Value; }
367
368
44.4k
  MCTargetOptions getTargetOptions() const { return MCOptions; }
369
370
225
  void setSemaCallback(MCAsmParserSemaCallback *Callback) {
371
225
    SemaCallback = Callback;
372
225
  }
373
374
  // Target-specific parsing of expression.
375
459k
  virtual bool parsePrimaryExpr(const MCExpr *&Res, SMLoc &EndLoc) {
376
459k
    return getParser().parsePrimaryExpr(Res, EndLoc);
377
459k
  }
378
379
  virtual bool ParseRegister(unsigned &RegNo, SMLoc &StartLoc,
380
                             SMLoc &EndLoc) = 0;
381
382
  /// Sets frame register corresponding to the current MachineFunction.
383
1.02k
  virtual void SetFrameRegister(unsigned RegNo) {}
384
385
  /// ParseInstruction - Parse one assembly instruction.
386
  ///
387
  /// The parser is positioned following the instruction name. The target
388
  /// specific instruction parser should parse the entire instruction and
389
  /// construct the appropriate MCInst, or emit an error. On success, the entire
390
  /// line should be parsed up to and including the end-of-statement token. On
391
  /// failure, the parser is not required to read to the end of the line.
392
  //
393
  /// \param Name - The instruction name.
394
  /// \param NameLoc - The source location of the name.
395
  /// \param Operands [out] - The list of parsed operands, this returns
396
  ///        ownership of them to the caller.
397
  /// \return True on failure.
398
  virtual bool ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
399
                                SMLoc NameLoc, OperandVector &Operands) = 0;
400
  virtual bool ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
401
466k
                                AsmToken Token, OperandVector &Operands) {
402
466k
    return ParseInstruction(Info, Name, Token.getLoc(), Operands);
403
466k
  }
404
405
  /// ParseDirective - Parse a target specific assembler directive
406
  ///
407
  /// The parser is positioned following the directive name.  The target
408
  /// specific directive parser should parse the entire directive doing or
409
  /// recording any target specific work, or return true and do nothing if the
410
  /// directive is not target specific. If the directive is specific for
411
  /// the target, the entire line is parsed up to and including the
412
  /// end-of-statement token and false is returned.
413
  ///
414
  /// \param DirectiveID - the identifier token of the directive.
415
  virtual bool ParseDirective(AsmToken DirectiveID) = 0;
416
417
  /// MatchAndEmitInstruction - Recognize a series of operands of a parsed
418
  /// instruction as an actual MCInst and emit it to the specified MCStreamer.
419
  /// This returns false on success and returns true on failure to match.
420
  ///
421
  /// On failure, the target parser is responsible for emitting a diagnostic
422
  /// explaining the match failure.
423
  virtual bool MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
424
                                       OperandVector &Operands, MCStreamer &Out,
425
                                       uint64_t &ErrorInfo,
426
                                       bool MatchingInlineAsm) = 0;
427
428
  /// Allows targets to let registers opt out of clobber lists.
429
0
  virtual bool OmitRegisterFromClobberLists(unsigned RegNo) { return false; }
430
431
  /// Allow a target to add special case operand matching for things that
432
  /// tblgen doesn't/can't handle effectively. For example, literal
433
  /// immediates on ARM. TableGen expects a token operand, but the parser
434
  /// will recognize them as immediates.
435
  virtual unsigned validateTargetOperandClass(MCParsedAsmOperand &Op,
436
1.08M
                                              unsigned Kind) {
437
1.08M
    return Match_InvalidOperand;
438
1.08M
  }
439
440
  /// Validate the instruction match against any complex target predicates
441
  /// before rendering any operands to it.
442
  virtual unsigned
443
474k
  checkEarlyTargetMatchPredicate(MCInst &Inst, const OperandVector &Operands) {
444
474k
    return Match_Success;
445
474k
  }
446
447
  /// checkTargetMatchPredicate - Validate the instruction match against
448
  /// any complex target predicates not expressible via match classes.
449
202k
  virtual unsigned checkTargetMatchPredicate(MCInst &Inst) {
450
202k
    return Match_Success;
451
202k
  }
452
453
  virtual void convertToMapAndConstraints(unsigned Kind,
454
                                          const OperandVector &Operands) = 0;
455
456
  /// Returns whether two registers are equal and is used by the tied-operands
457
  /// checks in the AsmMatcher. This method can be overridden allow e.g. a
458
  /// sub- or super-register as the tied operand.
459
  virtual bool regsEqual(const MCParsedAsmOperand &Op1,
460
2.07k
                         const MCParsedAsmOperand &Op2) const {
461
2.07k
    assert(Op1.isReg() && Op2.isReg() && "Operands not all regs");
462
2.07k
    return Op1.getReg() == Op2.getReg();
463
2.07k
  }
464
465
  // Return whether this parser uses assignment statements with equals tokens
466
618
  virtual bool equalIsAsmAssignment() { return true; };
467
  // Return whether this start of statement identifier is a label
468
23.5k
  virtual bool isLabel(AsmToken &Token) { return true; };
469
  // Return whether this parser accept star as start of statement
470
0
  virtual bool starIsStartOfStatement() { return false; };
471
472
  virtual const MCExpr *applyModifierToExpr(const MCExpr *E,
473
                                            MCSymbolRefExpr::VariantKind,
474
26
                                            MCContext &Ctx) {
475
26
    return nullptr;
476
26
  }
477
478
  // For actions that have to be performed before a label is emitted
479
20.3k
  virtual void doBeforeLabelEmit(MCSymbol *Symbol) {}
480
  
481
20.3k
  virtual void onLabelParsed(MCSymbol *Symbol) {}
482
483
  /// Ensure that all previously parsed instructions have been emitted to the
484
  /// output streamer, if the target does not emit them immediately.
485
676k
  virtual void flushPendingInstructions(MCStreamer &Out) {}
486
487
  virtual const MCExpr *createTargetUnaryExpr(const MCExpr *E,
488
                                              AsmToken::TokenKind OperatorToken,
489
0
                                              MCContext &Ctx) {
490
0
    return nullptr;
491
0
  }
492
493
  // For any checks or cleanups at the end of parsing.
494
15.5k
  virtual void onEndOfFile() {}
495
};
496
497
} // end namespace llvm
498
499
#endif // LLVM_MC_MCPARSER_MCTARGETASMPARSER_H