xref: /freebsd/contrib/llvm-project/llvm/lib/Target/RISCV/AsmParser/RISCVAsmParser.cpp (revision 85868e8a1daeaae7a0e48effb2ea2310ae3b02c6)
1 //===-- RISCVAsmParser.cpp - Parse RISCV assembly to MCInst instructions --===//
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 #include "MCTargetDesc/RISCVAsmBackend.h"
10 #include "MCTargetDesc/RISCVMCExpr.h"
11 #include "MCTargetDesc/RISCVMCTargetDesc.h"
12 #include "MCTargetDesc/RISCVTargetStreamer.h"
13 #include "TargetInfo/RISCVTargetInfo.h"
14 #include "Utils/RISCVBaseInfo.h"
15 #include "Utils/RISCVMatInt.h"
16 #include "llvm/ADT/STLExtras.h"
17 #include "llvm/ADT/SmallVector.h"
18 #include "llvm/ADT/StringSwitch.h"
19 #include "llvm/CodeGen/Register.h"
20 #include "llvm/MC/MCAssembler.h"
21 #include "llvm/MC/MCContext.h"
22 #include "llvm/MC/MCExpr.h"
23 #include "llvm/MC/MCInst.h"
24 #include "llvm/MC/MCInstBuilder.h"
25 #include "llvm/MC/MCObjectFileInfo.h"
26 #include "llvm/MC/MCParser/MCAsmLexer.h"
27 #include "llvm/MC/MCParser/MCParsedAsmOperand.h"
28 #include "llvm/MC/MCParser/MCTargetAsmParser.h"
29 #include "llvm/MC/MCRegisterInfo.h"
30 #include "llvm/MC/MCStreamer.h"
31 #include "llvm/MC/MCSubtargetInfo.h"
32 #include "llvm/Support/Casting.h"
33 #include "llvm/Support/MathExtras.h"
34 #include "llvm/Support/TargetRegistry.h"
35 
36 #include <limits>
37 
38 using namespace llvm;
39 
40 // Include the auto-generated portion of the compress emitter.
41 #define GEN_COMPRESS_INSTR
42 #include "RISCVGenCompressInstEmitter.inc"
43 
44 namespace {
45 struct RISCVOperand;
46 
47 class RISCVAsmParser : public MCTargetAsmParser {
48   SmallVector<FeatureBitset, 4> FeatureBitStack;
49 
50   SMLoc getLoc() const { return getParser().getTok().getLoc(); }
51   bool isRV64() const { return getSTI().hasFeature(RISCV::Feature64Bit); }
52   bool isRV32E() const { return getSTI().hasFeature(RISCV::FeatureRV32E); }
53 
54   RISCVTargetStreamer &getTargetStreamer() {
55     MCTargetStreamer &TS = *getParser().getStreamer().getTargetStreamer();
56     return static_cast<RISCVTargetStreamer &>(TS);
57   }
58 
59   unsigned validateTargetOperandClass(MCParsedAsmOperand &Op,
60                                       unsigned Kind) override;
61 
62   bool generateImmOutOfRangeError(OperandVector &Operands, uint64_t ErrorInfo,
63                                   int64_t Lower, int64_t Upper, Twine Msg);
64 
65   bool MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
66                                OperandVector &Operands, MCStreamer &Out,
67                                uint64_t &ErrorInfo,
68                                bool MatchingInlineAsm) override;
69 
70   bool ParseRegister(unsigned &RegNo, SMLoc &StartLoc, SMLoc &EndLoc) override;
71 
72   bool ParseInstruction(ParseInstructionInfo &Info, StringRef Name,
73                         SMLoc NameLoc, OperandVector &Operands) override;
74 
75   bool ParseDirective(AsmToken DirectiveID) override;
76 
77   // Helper to actually emit an instruction to the MCStreamer. Also, when
78   // possible, compression of the instruction is performed.
79   void emitToStreamer(MCStreamer &S, const MCInst &Inst);
80 
81   // Helper to emit a combination of LUI, ADDI(W), and SLLI instructions that
82   // synthesize the desired immedate value into the destination register.
83   void emitLoadImm(Register DestReg, int64_t Value, MCStreamer &Out);
84 
85   // Helper to emit a combination of AUIPC and SecondOpcode. Used to implement
86   // helpers such as emitLoadLocalAddress and emitLoadAddress.
87   void emitAuipcInstPair(MCOperand DestReg, MCOperand TmpReg,
88                          const MCExpr *Symbol, RISCVMCExpr::VariantKind VKHi,
89                          unsigned SecondOpcode, SMLoc IDLoc, MCStreamer &Out);
90 
91   // Helper to emit pseudo instruction "lla" used in PC-rel addressing.
92   void emitLoadLocalAddress(MCInst &Inst, SMLoc IDLoc, MCStreamer &Out);
93 
94   // Helper to emit pseudo instruction "la" used in GOT/PC-rel addressing.
95   void emitLoadAddress(MCInst &Inst, SMLoc IDLoc, MCStreamer &Out);
96 
97   // Helper to emit pseudo instruction "la.tls.ie" used in initial-exec TLS
98   // addressing.
99   void emitLoadTLSIEAddress(MCInst &Inst, SMLoc IDLoc, MCStreamer &Out);
100 
101   // Helper to emit pseudo instruction "la.tls.gd" used in global-dynamic TLS
102   // addressing.
103   void emitLoadTLSGDAddress(MCInst &Inst, SMLoc IDLoc, MCStreamer &Out);
104 
105   // Helper to emit pseudo load/store instruction with a symbol.
106   void emitLoadStoreSymbol(MCInst &Inst, unsigned Opcode, SMLoc IDLoc,
107                            MCStreamer &Out, bool HasTmpReg);
108 
109   // Checks that a PseudoAddTPRel is using x4/tp in its second input operand.
110   // Enforcing this using a restricted register class for the second input
111   // operand of PseudoAddTPRel results in a poor diagnostic due to the fact
112   // 'add' is an overloaded mnemonic.
113   bool checkPseudoAddTPRel(MCInst &Inst, OperandVector &Operands);
114 
115   /// Helper for processing MC instructions that have been successfully matched
116   /// by MatchAndEmitInstruction. Modifications to the emitted instructions,
117   /// like the expansion of pseudo instructions (e.g., "li"), can be performed
118   /// in this method.
119   bool processInstruction(MCInst &Inst, SMLoc IDLoc, OperandVector &Operands,
120                           MCStreamer &Out);
121 
122 // Auto-generated instruction matching functions
123 #define GET_ASSEMBLER_HEADER
124 #include "RISCVGenAsmMatcher.inc"
125 
126   OperandMatchResultTy parseCSRSystemRegister(OperandVector &Operands);
127   OperandMatchResultTy parseImmediate(OperandVector &Operands);
128   OperandMatchResultTy parseRegister(OperandVector &Operands,
129                                      bool AllowParens = false);
130   OperandMatchResultTy parseMemOpBaseReg(OperandVector &Operands);
131   OperandMatchResultTy parseAtomicMemOp(OperandVector &Operands);
132   OperandMatchResultTy parseOperandWithModifier(OperandVector &Operands);
133   OperandMatchResultTy parseBareSymbol(OperandVector &Operands);
134   OperandMatchResultTy parseCallSymbol(OperandVector &Operands);
135   OperandMatchResultTy parseJALOffset(OperandVector &Operands);
136 
137   bool parseOperand(OperandVector &Operands, StringRef Mnemonic);
138 
139   bool parseDirectiveOption();
140 
141   void setFeatureBits(uint64_t Feature, StringRef FeatureString) {
142     if (!(getSTI().getFeatureBits()[Feature])) {
143       MCSubtargetInfo &STI = copySTI();
144       setAvailableFeatures(
145           ComputeAvailableFeatures(STI.ToggleFeature(FeatureString)));
146     }
147   }
148 
149   void clearFeatureBits(uint64_t Feature, StringRef FeatureString) {
150     if (getSTI().getFeatureBits()[Feature]) {
151       MCSubtargetInfo &STI = copySTI();
152       setAvailableFeatures(
153           ComputeAvailableFeatures(STI.ToggleFeature(FeatureString)));
154     }
155   }
156 
157   void pushFeatureBits() {
158     FeatureBitStack.push_back(getSTI().getFeatureBits());
159   }
160 
161   bool popFeatureBits() {
162     if (FeatureBitStack.empty())
163       return true;
164 
165     FeatureBitset FeatureBits = FeatureBitStack.pop_back_val();
166     copySTI().setFeatureBits(FeatureBits);
167     setAvailableFeatures(ComputeAvailableFeatures(FeatureBits));
168 
169     return false;
170   }
171 public:
172   enum RISCVMatchResultTy {
173     Match_Dummy = FIRST_TARGET_MATCH_RESULT_TY,
174 #define GET_OPERAND_DIAGNOSTIC_TYPES
175 #include "RISCVGenAsmMatcher.inc"
176 #undef GET_OPERAND_DIAGNOSTIC_TYPES
177   };
178 
179   static bool classifySymbolRef(const MCExpr *Expr,
180                                 RISCVMCExpr::VariantKind &Kind,
181                                 int64_t &Addend);
182 
183   RISCVAsmParser(const MCSubtargetInfo &STI, MCAsmParser &Parser,
184                  const MCInstrInfo &MII, const MCTargetOptions &Options)
185       : MCTargetAsmParser(Options, STI, MII) {
186     Parser.addAliasForDirective(".half", ".2byte");
187     Parser.addAliasForDirective(".hword", ".2byte");
188     Parser.addAliasForDirective(".word", ".4byte");
189     Parser.addAliasForDirective(".dword", ".8byte");
190     setAvailableFeatures(ComputeAvailableFeatures(STI.getFeatureBits()));
191   }
192 };
193 
194 /// RISCVOperand - Instances of this class represent a parsed machine
195 /// instruction
196 struct RISCVOperand : public MCParsedAsmOperand {
197 
198   enum class KindTy {
199     Token,
200     Register,
201     Immediate,
202     SystemRegister
203   } Kind;
204 
205   bool IsRV64;
206 
207   struct RegOp {
208     Register RegNum;
209   };
210 
211   struct ImmOp {
212     const MCExpr *Val;
213   };
214 
215   struct SysRegOp {
216     const char *Data;
217     unsigned Length;
218     unsigned Encoding;
219     // FIXME: Add the Encoding parsed fields as needed for checks,
220     // e.g.: read/write or user/supervisor/machine privileges.
221   };
222 
223   SMLoc StartLoc, EndLoc;
224   union {
225     StringRef Tok;
226     RegOp Reg;
227     ImmOp Imm;
228     struct SysRegOp SysReg;
229   };
230 
231   RISCVOperand(KindTy K) : MCParsedAsmOperand(), Kind(K) {}
232 
233 public:
234   RISCVOperand(const RISCVOperand &o) : MCParsedAsmOperand() {
235     Kind = o.Kind;
236     IsRV64 = o.IsRV64;
237     StartLoc = o.StartLoc;
238     EndLoc = o.EndLoc;
239     switch (Kind) {
240     case KindTy::Register:
241       Reg = o.Reg;
242       break;
243     case KindTy::Immediate:
244       Imm = o.Imm;
245       break;
246     case KindTy::Token:
247       Tok = o.Tok;
248       break;
249     case KindTy::SystemRegister:
250       SysReg = o.SysReg;
251       break;
252     }
253   }
254 
255   bool isToken() const override { return Kind == KindTy::Token; }
256   bool isReg() const override { return Kind == KindTy::Register; }
257   bool isImm() const override { return Kind == KindTy::Immediate; }
258   bool isMem() const override { return false; }
259   bool isSystemRegister() const { return Kind == KindTy::SystemRegister; }
260 
261   static bool evaluateConstantImm(const MCExpr *Expr, int64_t &Imm,
262                                   RISCVMCExpr::VariantKind &VK) {
263     if (auto *RE = dyn_cast<RISCVMCExpr>(Expr)) {
264       VK = RE->getKind();
265       return RE->evaluateAsConstant(Imm);
266     }
267 
268     if (auto CE = dyn_cast<MCConstantExpr>(Expr)) {
269       VK = RISCVMCExpr::VK_RISCV_None;
270       Imm = CE->getValue();
271       return true;
272     }
273 
274     return false;
275   }
276 
277   // True if operand is a symbol with no modifiers, or a constant with no
278   // modifiers and isShiftedInt<N-1, 1>(Op).
279   template <int N> bool isBareSimmNLsb0() const {
280     int64_t Imm;
281     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
282     if (!isImm())
283       return false;
284     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
285     bool IsValid;
286     if (!IsConstantImm)
287       IsValid = RISCVAsmParser::classifySymbolRef(getImm(), VK, Imm);
288     else
289       IsValid = isShiftedInt<N - 1, 1>(Imm);
290     return IsValid && VK == RISCVMCExpr::VK_RISCV_None;
291   }
292 
293   // Predicate methods for AsmOperands defined in RISCVInstrInfo.td
294 
295   bool isBareSymbol() const {
296     int64_t Imm;
297     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
298     // Must be of 'immediate' type but not a constant.
299     if (!isImm() || evaluateConstantImm(getImm(), Imm, VK))
300       return false;
301     return RISCVAsmParser::classifySymbolRef(getImm(), VK, Imm) &&
302            VK == RISCVMCExpr::VK_RISCV_None;
303   }
304 
305   bool isCallSymbol() const {
306     int64_t Imm;
307     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
308     // Must be of 'immediate' type but not a constant.
309     if (!isImm() || evaluateConstantImm(getImm(), Imm, VK))
310       return false;
311     return RISCVAsmParser::classifySymbolRef(getImm(), VK, Imm) &&
312            (VK == RISCVMCExpr::VK_RISCV_CALL ||
313             VK == RISCVMCExpr::VK_RISCV_CALL_PLT);
314   }
315 
316   bool isTPRelAddSymbol() const {
317     int64_t Imm;
318     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
319     // Must be of 'immediate' type but not a constant.
320     if (!isImm() || evaluateConstantImm(getImm(), Imm, VK))
321       return false;
322     return RISCVAsmParser::classifySymbolRef(getImm(), VK, Imm) &&
323            VK == RISCVMCExpr::VK_RISCV_TPREL_ADD;
324   }
325 
326   bool isCSRSystemRegister() const { return isSystemRegister(); }
327 
328   /// Return true if the operand is a valid for the fence instruction e.g.
329   /// ('iorw').
330   bool isFenceArg() const {
331     if (!isImm())
332       return false;
333     const MCExpr *Val = getImm();
334     auto *SVal = dyn_cast<MCSymbolRefExpr>(Val);
335     if (!SVal || SVal->getKind() != MCSymbolRefExpr::VK_None)
336       return false;
337 
338     StringRef Str = SVal->getSymbol().getName();
339     // Letters must be unique, taken from 'iorw', and in ascending order. This
340     // holds as long as each individual character is one of 'iorw' and is
341     // greater than the previous character.
342     char Prev = '\0';
343     for (char c : Str) {
344       if (c != 'i' && c != 'o' && c != 'r' && c != 'w')
345         return false;
346       if (c <= Prev)
347         return false;
348       Prev = c;
349     }
350     return true;
351   }
352 
353   /// Return true if the operand is a valid floating point rounding mode.
354   bool isFRMArg() const {
355     if (!isImm())
356       return false;
357     const MCExpr *Val = getImm();
358     auto *SVal = dyn_cast<MCSymbolRefExpr>(Val);
359     if (!SVal || SVal->getKind() != MCSymbolRefExpr::VK_None)
360       return false;
361 
362     StringRef Str = SVal->getSymbol().getName();
363 
364     return RISCVFPRndMode::stringToRoundingMode(Str) != RISCVFPRndMode::Invalid;
365   }
366 
367   bool isImmXLenLI() const {
368     int64_t Imm;
369     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
370     if (!isImm())
371       return false;
372     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
373     if (VK == RISCVMCExpr::VK_RISCV_LO || VK == RISCVMCExpr::VK_RISCV_PCREL_LO)
374       return true;
375     // Given only Imm, ensuring that the actually specified constant is either
376     // a signed or unsigned 64-bit number is unfortunately impossible.
377     return IsConstantImm && VK == RISCVMCExpr::VK_RISCV_None &&
378            (isRV64() || (isInt<32>(Imm) || isUInt<32>(Imm)));
379   }
380 
381   bool isUImmLog2XLen() const {
382     int64_t Imm;
383     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
384     if (!isImm())
385       return false;
386     if (!evaluateConstantImm(getImm(), Imm, VK) ||
387         VK != RISCVMCExpr::VK_RISCV_None)
388       return false;
389     return (isRV64() && isUInt<6>(Imm)) || isUInt<5>(Imm);
390   }
391 
392   bool isUImmLog2XLenNonZero() const {
393     int64_t Imm;
394     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
395     if (!isImm())
396       return false;
397     if (!evaluateConstantImm(getImm(), Imm, VK) ||
398         VK != RISCVMCExpr::VK_RISCV_None)
399       return false;
400     if (Imm == 0)
401       return false;
402     return (isRV64() && isUInt<6>(Imm)) || isUInt<5>(Imm);
403   }
404 
405   bool isUImm5() const {
406     int64_t Imm;
407     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
408     if (!isImm())
409       return false;
410     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
411     return IsConstantImm && isUInt<5>(Imm) && VK == RISCVMCExpr::VK_RISCV_None;
412   }
413 
414   bool isUImm5NonZero() const {
415     int64_t Imm;
416     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
417     if (!isImm())
418       return false;
419     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
420     return IsConstantImm && isUInt<5>(Imm) && (Imm != 0) &&
421            VK == RISCVMCExpr::VK_RISCV_None;
422   }
423 
424   bool isSImm6() const {
425     if (!isImm())
426       return false;
427     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
428     int64_t Imm;
429     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
430     return IsConstantImm && isInt<6>(Imm) &&
431            VK == RISCVMCExpr::VK_RISCV_None;
432   }
433 
434   bool isSImm6NonZero() const {
435     if (!isImm())
436       return false;
437     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
438     int64_t Imm;
439     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
440     return IsConstantImm && isInt<6>(Imm) && (Imm != 0) &&
441            VK == RISCVMCExpr::VK_RISCV_None;
442   }
443 
444   bool isCLUIImm() const {
445     if (!isImm())
446       return false;
447     int64_t Imm;
448     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
449     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
450     return IsConstantImm && (Imm != 0) &&
451            (isUInt<5>(Imm) || (Imm >= 0xfffe0 && Imm <= 0xfffff)) &&
452            VK == RISCVMCExpr::VK_RISCV_None;
453   }
454 
455   bool isUImm7Lsb00() const {
456     if (!isImm())
457       return false;
458     int64_t Imm;
459     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
460     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
461     return IsConstantImm && isShiftedUInt<5, 2>(Imm) &&
462            VK == RISCVMCExpr::VK_RISCV_None;
463   }
464 
465   bool isUImm8Lsb00() const {
466     if (!isImm())
467       return false;
468     int64_t Imm;
469     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
470     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
471     return IsConstantImm && isShiftedUInt<6, 2>(Imm) &&
472            VK == RISCVMCExpr::VK_RISCV_None;
473   }
474 
475   bool isUImm8Lsb000() const {
476     if (!isImm())
477       return false;
478     int64_t Imm;
479     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
480     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
481     return IsConstantImm && isShiftedUInt<5, 3>(Imm) &&
482            VK == RISCVMCExpr::VK_RISCV_None;
483   }
484 
485   bool isSImm9Lsb0() const { return isBareSimmNLsb0<9>(); }
486 
487   bool isUImm9Lsb000() const {
488     if (!isImm())
489       return false;
490     int64_t Imm;
491     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
492     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
493     return IsConstantImm && isShiftedUInt<6, 3>(Imm) &&
494            VK == RISCVMCExpr::VK_RISCV_None;
495   }
496 
497   bool isUImm10Lsb00NonZero() const {
498     if (!isImm())
499       return false;
500     int64_t Imm;
501     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
502     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
503     return IsConstantImm && isShiftedUInt<8, 2>(Imm) && (Imm != 0) &&
504            VK == RISCVMCExpr::VK_RISCV_None;
505   }
506 
507   bool isSImm12() const {
508     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
509     int64_t Imm;
510     bool IsValid;
511     if (!isImm())
512       return false;
513     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
514     if (!IsConstantImm)
515       IsValid = RISCVAsmParser::classifySymbolRef(getImm(), VK, Imm);
516     else
517       IsValid = isInt<12>(Imm);
518     return IsValid && ((IsConstantImm && VK == RISCVMCExpr::VK_RISCV_None) ||
519                        VK == RISCVMCExpr::VK_RISCV_LO ||
520                        VK == RISCVMCExpr::VK_RISCV_PCREL_LO ||
521                        VK == RISCVMCExpr::VK_RISCV_TPREL_LO);
522   }
523 
524   bool isSImm12Lsb0() const { return isBareSimmNLsb0<12>(); }
525 
526   bool isSImm13Lsb0() const { return isBareSimmNLsb0<13>(); }
527 
528   bool isSImm10Lsb0000NonZero() const {
529     if (!isImm())
530       return false;
531     int64_t Imm;
532     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
533     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
534     return IsConstantImm && (Imm != 0) && isShiftedInt<6, 4>(Imm) &&
535            VK == RISCVMCExpr::VK_RISCV_None;
536   }
537 
538   bool isUImm20LUI() const {
539     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
540     int64_t Imm;
541     bool IsValid;
542     if (!isImm())
543       return false;
544     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
545     if (!IsConstantImm) {
546       IsValid = RISCVAsmParser::classifySymbolRef(getImm(), VK, Imm);
547       return IsValid && (VK == RISCVMCExpr::VK_RISCV_HI ||
548                          VK == RISCVMCExpr::VK_RISCV_TPREL_HI);
549     } else {
550       return isUInt<20>(Imm) && (VK == RISCVMCExpr::VK_RISCV_None ||
551                                  VK == RISCVMCExpr::VK_RISCV_HI ||
552                                  VK == RISCVMCExpr::VK_RISCV_TPREL_HI);
553     }
554   }
555 
556   bool isUImm20AUIPC() const {
557     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
558     int64_t Imm;
559     bool IsValid;
560     if (!isImm())
561       return false;
562     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
563     if (!IsConstantImm) {
564       IsValid = RISCVAsmParser::classifySymbolRef(getImm(), VK, Imm);
565       return IsValid && (VK == RISCVMCExpr::VK_RISCV_PCREL_HI ||
566                          VK == RISCVMCExpr::VK_RISCV_GOT_HI ||
567                          VK == RISCVMCExpr::VK_RISCV_TLS_GOT_HI ||
568                          VK == RISCVMCExpr::VK_RISCV_TLS_GD_HI);
569     } else {
570       return isUInt<20>(Imm) && (VK == RISCVMCExpr::VK_RISCV_None ||
571                                  VK == RISCVMCExpr::VK_RISCV_PCREL_HI ||
572                                  VK == RISCVMCExpr::VK_RISCV_GOT_HI ||
573                                  VK == RISCVMCExpr::VK_RISCV_TLS_GOT_HI ||
574                                  VK == RISCVMCExpr::VK_RISCV_TLS_GD_HI);
575     }
576   }
577 
578   bool isSImm21Lsb0JAL() const { return isBareSimmNLsb0<21>(); }
579 
580   bool isImmZero() const {
581     if (!isImm())
582       return false;
583     int64_t Imm;
584     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
585     bool IsConstantImm = evaluateConstantImm(getImm(), Imm, VK);
586     return IsConstantImm && (Imm == 0) && VK == RISCVMCExpr::VK_RISCV_None;
587   }
588 
589   /// getStartLoc - Gets location of the first token of this operand
590   SMLoc getStartLoc() const override { return StartLoc; }
591   /// getEndLoc - Gets location of the last token of this operand
592   SMLoc getEndLoc() const override { return EndLoc; }
593   /// True if this operand is for an RV64 instruction
594   bool isRV64() const { return IsRV64; }
595 
596   unsigned getReg() const override {
597     assert(Kind == KindTy::Register && "Invalid type access!");
598     return Reg.RegNum.id();
599   }
600 
601   StringRef getSysReg() const {
602     assert(Kind == KindTy::SystemRegister && "Invalid access!");
603     return StringRef(SysReg.Data, SysReg.Length);
604   }
605 
606   const MCExpr *getImm() const {
607     assert(Kind == KindTy::Immediate && "Invalid type access!");
608     return Imm.Val;
609   }
610 
611   StringRef getToken() const {
612     assert(Kind == KindTy::Token && "Invalid type access!");
613     return Tok;
614   }
615 
616   void print(raw_ostream &OS) const override {
617     switch (Kind) {
618     case KindTy::Immediate:
619       OS << *getImm();
620       break;
621     case KindTy::Register:
622       OS << "<register x";
623       OS << getReg() << ">";
624       break;
625     case KindTy::Token:
626       OS << "'" << getToken() << "'";
627       break;
628     case KindTy::SystemRegister:
629       OS << "<sysreg: " << getSysReg() << '>';
630       break;
631     }
632   }
633 
634   static std::unique_ptr<RISCVOperand> createToken(StringRef Str, SMLoc S,
635                                                    bool IsRV64) {
636     auto Op = std::make_unique<RISCVOperand>(KindTy::Token);
637     Op->Tok = Str;
638     Op->StartLoc = S;
639     Op->EndLoc = S;
640     Op->IsRV64 = IsRV64;
641     return Op;
642   }
643 
644   static std::unique_ptr<RISCVOperand> createReg(unsigned RegNo, SMLoc S,
645                                                  SMLoc E, bool IsRV64) {
646     auto Op = std::make_unique<RISCVOperand>(KindTy::Register);
647     Op->Reg.RegNum = RegNo;
648     Op->StartLoc = S;
649     Op->EndLoc = E;
650     Op->IsRV64 = IsRV64;
651     return Op;
652   }
653 
654   static std::unique_ptr<RISCVOperand> createImm(const MCExpr *Val, SMLoc S,
655                                                  SMLoc E, bool IsRV64) {
656     auto Op = std::make_unique<RISCVOperand>(KindTy::Immediate);
657     Op->Imm.Val = Val;
658     Op->StartLoc = S;
659     Op->EndLoc = E;
660     Op->IsRV64 = IsRV64;
661     return Op;
662   }
663 
664   static std::unique_ptr<RISCVOperand>
665   createSysReg(StringRef Str, SMLoc S, unsigned Encoding, bool IsRV64) {
666     auto Op = std::make_unique<RISCVOperand>(KindTy::SystemRegister);
667     Op->SysReg.Data = Str.data();
668     Op->SysReg.Length = Str.size();
669     Op->SysReg.Encoding = Encoding;
670     Op->StartLoc = S;
671     Op->IsRV64 = IsRV64;
672     return Op;
673   }
674 
675   void addExpr(MCInst &Inst, const MCExpr *Expr) const {
676     assert(Expr && "Expr shouldn't be null!");
677     int64_t Imm = 0;
678     RISCVMCExpr::VariantKind VK = RISCVMCExpr::VK_RISCV_None;
679     bool IsConstant = evaluateConstantImm(Expr, Imm, VK);
680 
681     if (IsConstant)
682       Inst.addOperand(MCOperand::createImm(Imm));
683     else
684       Inst.addOperand(MCOperand::createExpr(Expr));
685   }
686 
687   // Used by the TableGen Code
688   void addRegOperands(MCInst &Inst, unsigned N) const {
689     assert(N == 1 && "Invalid number of operands!");
690     Inst.addOperand(MCOperand::createReg(getReg()));
691   }
692 
693   void addImmOperands(MCInst &Inst, unsigned N) const {
694     assert(N == 1 && "Invalid number of operands!");
695     addExpr(Inst, getImm());
696   }
697 
698   void addFenceArgOperands(MCInst &Inst, unsigned N) const {
699     assert(N == 1 && "Invalid number of operands!");
700     // isFenceArg has validated the operand, meaning this cast is safe
701     auto SE = cast<MCSymbolRefExpr>(getImm());
702 
703     unsigned Imm = 0;
704     for (char c : SE->getSymbol().getName()) {
705       switch (c) {
706       default:
707         llvm_unreachable("FenceArg must contain only [iorw]");
708       case 'i': Imm |= RISCVFenceField::I; break;
709       case 'o': Imm |= RISCVFenceField::O; break;
710       case 'r': Imm |= RISCVFenceField::R; break;
711       case 'w': Imm |= RISCVFenceField::W; break;
712       }
713     }
714     Inst.addOperand(MCOperand::createImm(Imm));
715   }
716 
717   void addCSRSystemRegisterOperands(MCInst &Inst, unsigned N) const {
718     assert(N == 1 && "Invalid number of operands!");
719     Inst.addOperand(MCOperand::createImm(SysReg.Encoding));
720   }
721 
722   // Returns the rounding mode represented by this RISCVOperand. Should only
723   // be called after checking isFRMArg.
724   RISCVFPRndMode::RoundingMode getRoundingMode() const {
725     // isFRMArg has validated the operand, meaning this cast is safe.
726     auto SE = cast<MCSymbolRefExpr>(getImm());
727     RISCVFPRndMode::RoundingMode FRM =
728         RISCVFPRndMode::stringToRoundingMode(SE->getSymbol().getName());
729     assert(FRM != RISCVFPRndMode::Invalid && "Invalid rounding mode");
730     return FRM;
731   }
732 
733   void addFRMArgOperands(MCInst &Inst, unsigned N) const {
734     assert(N == 1 && "Invalid number of operands!");
735     Inst.addOperand(MCOperand::createImm(getRoundingMode()));
736   }
737 };
738 } // end anonymous namespace.
739 
740 #define GET_REGISTER_MATCHER
741 #define GET_MATCHER_IMPLEMENTATION
742 #include "RISCVGenAsmMatcher.inc"
743 
744 static Register convertFPR64ToFPR32(Register Reg) {
745   assert(Reg >= RISCV::F0_D && Reg <= RISCV::F31_D && "Invalid register");
746   return Reg - RISCV::F0_D + RISCV::F0_F;
747 }
748 
749 unsigned RISCVAsmParser::validateTargetOperandClass(MCParsedAsmOperand &AsmOp,
750                                                     unsigned Kind) {
751   RISCVOperand &Op = static_cast<RISCVOperand &>(AsmOp);
752   if (!Op.isReg())
753     return Match_InvalidOperand;
754 
755   Register Reg = Op.getReg();
756   bool IsRegFPR64 =
757       RISCVMCRegisterClasses[RISCV::FPR64RegClassID].contains(Reg);
758   bool IsRegFPR64C =
759       RISCVMCRegisterClasses[RISCV::FPR64CRegClassID].contains(Reg);
760 
761   // As the parser couldn't differentiate an FPR32 from an FPR64, coerce the
762   // register from FPR64 to FPR32 or FPR64C to FPR32C if necessary.
763   if ((IsRegFPR64 && Kind == MCK_FPR32) ||
764       (IsRegFPR64C && Kind == MCK_FPR32C)) {
765     Op.Reg.RegNum = convertFPR64ToFPR32(Reg);
766     return Match_Success;
767   }
768   return Match_InvalidOperand;
769 }
770 
771 bool RISCVAsmParser::generateImmOutOfRangeError(
772     OperandVector &Operands, uint64_t ErrorInfo, int64_t Lower, int64_t Upper,
773     Twine Msg = "immediate must be an integer in the range") {
774   SMLoc ErrorLoc = ((RISCVOperand &)*Operands[ErrorInfo]).getStartLoc();
775   return Error(ErrorLoc, Msg + " [" + Twine(Lower) + ", " + Twine(Upper) + "]");
776 }
777 
778 bool RISCVAsmParser::MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
779                                              OperandVector &Operands,
780                                              MCStreamer &Out,
781                                              uint64_t &ErrorInfo,
782                                              bool MatchingInlineAsm) {
783   MCInst Inst;
784 
785   auto Result =
786     MatchInstructionImpl(Operands, Inst, ErrorInfo, MatchingInlineAsm);
787   switch (Result) {
788   default:
789     break;
790   case Match_Success:
791     return processInstruction(Inst, IDLoc, Operands, Out);
792   case Match_MissingFeature:
793     return Error(IDLoc, "instruction use requires an option to be enabled");
794   case Match_MnemonicFail:
795     return Error(IDLoc, "unrecognized instruction mnemonic");
796   case Match_InvalidOperand: {
797     SMLoc ErrorLoc = IDLoc;
798     if (ErrorInfo != ~0U) {
799       if (ErrorInfo >= Operands.size())
800         return Error(ErrorLoc, "too few operands for instruction");
801 
802       ErrorLoc = ((RISCVOperand &)*Operands[ErrorInfo]).getStartLoc();
803       if (ErrorLoc == SMLoc())
804         ErrorLoc = IDLoc;
805     }
806     return Error(ErrorLoc, "invalid operand for instruction");
807   }
808   }
809 
810   // Handle the case when the error message is of specific type
811   // other than the generic Match_InvalidOperand, and the
812   // corresponding operand is missing.
813   if (Result > FIRST_TARGET_MATCH_RESULT_TY) {
814     SMLoc ErrorLoc = IDLoc;
815     if (ErrorInfo != ~0U && ErrorInfo >= Operands.size())
816         return Error(ErrorLoc, "too few operands for instruction");
817   }
818 
819   switch(Result) {
820   default:
821     break;
822   case Match_InvalidImmXLenLI:
823     if (isRV64()) {
824       SMLoc ErrorLoc = ((RISCVOperand &)*Operands[ErrorInfo]).getStartLoc();
825       return Error(ErrorLoc, "operand must be a constant 64-bit integer");
826     }
827     return generateImmOutOfRangeError(Operands, ErrorInfo,
828                                       std::numeric_limits<int32_t>::min(),
829                                       std::numeric_limits<uint32_t>::max());
830   case Match_InvalidImmZero: {
831     SMLoc ErrorLoc = ((RISCVOperand &)*Operands[ErrorInfo]).getStartLoc();
832     return Error(ErrorLoc, "immediate must be zero");
833   }
834   case Match_InvalidUImmLog2XLen:
835     if (isRV64())
836       return generateImmOutOfRangeError(Operands, ErrorInfo, 0, (1 << 6) - 1);
837     return generateImmOutOfRangeError(Operands, ErrorInfo, 0, (1 << 5) - 1);
838   case Match_InvalidUImmLog2XLenNonZero:
839     if (isRV64())
840       return generateImmOutOfRangeError(Operands, ErrorInfo, 1, (1 << 6) - 1);
841     return generateImmOutOfRangeError(Operands, ErrorInfo, 1, (1 << 5) - 1);
842   case Match_InvalidUImm5:
843     return generateImmOutOfRangeError(Operands, ErrorInfo, 0, (1 << 5) - 1);
844   case Match_InvalidSImm6:
845     return generateImmOutOfRangeError(Operands, ErrorInfo, -(1 << 5),
846                                       (1 << 5) - 1);
847   case Match_InvalidSImm6NonZero:
848     return generateImmOutOfRangeError(
849         Operands, ErrorInfo, -(1 << 5), (1 << 5) - 1,
850         "immediate must be non-zero in the range");
851   case Match_InvalidCLUIImm:
852     return generateImmOutOfRangeError(
853         Operands, ErrorInfo, 1, (1 << 5) - 1,
854         "immediate must be in [0xfffe0, 0xfffff] or");
855   case Match_InvalidUImm7Lsb00:
856     return generateImmOutOfRangeError(
857         Operands, ErrorInfo, 0, (1 << 7) - 4,
858         "immediate must be a multiple of 4 bytes in the range");
859   case Match_InvalidUImm8Lsb00:
860     return generateImmOutOfRangeError(
861         Operands, ErrorInfo, 0, (1 << 8) - 4,
862         "immediate must be a multiple of 4 bytes in the range");
863   case Match_InvalidUImm8Lsb000:
864     return generateImmOutOfRangeError(
865         Operands, ErrorInfo, 0, (1 << 8) - 8,
866         "immediate must be a multiple of 8 bytes in the range");
867   case Match_InvalidSImm9Lsb0:
868     return generateImmOutOfRangeError(
869         Operands, ErrorInfo, -(1 << 8), (1 << 8) - 2,
870         "immediate must be a multiple of 2 bytes in the range");
871   case Match_InvalidUImm9Lsb000:
872     return generateImmOutOfRangeError(
873         Operands, ErrorInfo, 0, (1 << 9) - 8,
874         "immediate must be a multiple of 8 bytes in the range");
875   case Match_InvalidUImm10Lsb00NonZero:
876     return generateImmOutOfRangeError(
877         Operands, ErrorInfo, 4, (1 << 10) - 4,
878         "immediate must be a multiple of 4 bytes in the range");
879   case Match_InvalidSImm10Lsb0000NonZero:
880     return generateImmOutOfRangeError(
881         Operands, ErrorInfo, -(1 << 9), (1 << 9) - 16,
882         "immediate must be a multiple of 16 bytes and non-zero in the range");
883   case Match_InvalidSImm12:
884     return generateImmOutOfRangeError(
885         Operands, ErrorInfo, -(1 << 11), (1 << 11) - 1,
886         "operand must be a symbol with %lo/%pcrel_lo/%tprel_lo modifier or an "
887         "integer in the range");
888   case Match_InvalidSImm12Lsb0:
889     return generateImmOutOfRangeError(
890         Operands, ErrorInfo, -(1 << 11), (1 << 11) - 2,
891         "immediate must be a multiple of 2 bytes in the range");
892   case Match_InvalidSImm13Lsb0:
893     return generateImmOutOfRangeError(
894         Operands, ErrorInfo, -(1 << 12), (1 << 12) - 2,
895         "immediate must be a multiple of 2 bytes in the range");
896   case Match_InvalidUImm20LUI:
897     return generateImmOutOfRangeError(Operands, ErrorInfo, 0, (1 << 20) - 1,
898                                       "operand must be a symbol with "
899                                       "%hi/%tprel_hi modifier or an integer in "
900                                       "the range");
901   case Match_InvalidUImm20AUIPC:
902     return generateImmOutOfRangeError(
903         Operands, ErrorInfo, 0, (1 << 20) - 1,
904         "operand must be a symbol with a "
905         "%pcrel_hi/%got_pcrel_hi/%tls_ie_pcrel_hi/%tls_gd_pcrel_hi modifier or "
906         "an integer in the range");
907   case Match_InvalidSImm21Lsb0JAL:
908     return generateImmOutOfRangeError(
909         Operands, ErrorInfo, -(1 << 20), (1 << 20) - 2,
910         "immediate must be a multiple of 2 bytes in the range");
911   case Match_InvalidCSRSystemRegister: {
912     return generateImmOutOfRangeError(Operands, ErrorInfo, 0, (1 << 12) - 1,
913                                       "operand must be a valid system register "
914                                       "name or an integer in the range");
915   }
916   case Match_InvalidFenceArg: {
917     SMLoc ErrorLoc = ((RISCVOperand &)*Operands[ErrorInfo]).getStartLoc();
918     return Error(
919         ErrorLoc,
920         "operand must be formed of letters selected in-order from 'iorw'");
921   }
922   case Match_InvalidFRMArg: {
923     SMLoc ErrorLoc = ((RISCVOperand &)*Operands[ErrorInfo]).getStartLoc();
924     return Error(
925         ErrorLoc,
926         "operand must be a valid floating point rounding mode mnemonic");
927   }
928   case Match_InvalidBareSymbol: {
929     SMLoc ErrorLoc = ((RISCVOperand &)*Operands[ErrorInfo]).getStartLoc();
930     return Error(ErrorLoc, "operand must be a bare symbol name");
931   }
932   case Match_InvalidCallSymbol: {
933     SMLoc ErrorLoc = ((RISCVOperand &)*Operands[ErrorInfo]).getStartLoc();
934     return Error(ErrorLoc, "operand must be a bare symbol name");
935   }
936   case Match_InvalidTPRelAddSymbol: {
937     SMLoc ErrorLoc = ((RISCVOperand &)*Operands[ErrorInfo]).getStartLoc();
938     return Error(ErrorLoc, "operand must be a symbol with %tprel_add modifier");
939   }
940   }
941 
942   llvm_unreachable("Unknown match type detected!");
943 }
944 
945 // Attempts to match Name as a register (either using the default name or
946 // alternative ABI names), setting RegNo to the matching register. Upon
947 // failure, returns true and sets RegNo to 0. If IsRV32E then registers
948 // x16-x31 will be rejected.
949 static bool matchRegisterNameHelper(bool IsRV32E, Register &RegNo,
950                                     StringRef Name) {
951   RegNo = MatchRegisterName(Name);
952   // The 32- and 64-bit FPRs have the same asm name. Check that the initial
953   // match always matches the 64-bit variant, and not the 32-bit one.
954   assert(!(RegNo >= RISCV::F0_F && RegNo <= RISCV::F31_F));
955   // The default FPR register class is based on the tablegen enum ordering.
956   static_assert(RISCV::F0_D < RISCV::F0_F, "FPR matching must be updated");
957   if (RegNo == RISCV::NoRegister)
958     RegNo = MatchRegisterAltName(Name);
959   if (IsRV32E && RegNo >= RISCV::X16 && RegNo <= RISCV::X31)
960     RegNo = RISCV::NoRegister;
961   return RegNo == RISCV::NoRegister;
962 }
963 
964 bool RISCVAsmParser::ParseRegister(unsigned &RegNo, SMLoc &StartLoc,
965                                    SMLoc &EndLoc) {
966   const AsmToken &Tok = getParser().getTok();
967   StartLoc = Tok.getLoc();
968   EndLoc = Tok.getEndLoc();
969   RegNo = 0;
970   StringRef Name = getLexer().getTok().getIdentifier();
971 
972   if (matchRegisterNameHelper(isRV32E(), (Register&)RegNo, Name))
973     return Error(StartLoc, "invalid register name");
974 
975   getParser().Lex(); // Eat identifier token.
976   return false;
977 }
978 
979 OperandMatchResultTy RISCVAsmParser::parseRegister(OperandVector &Operands,
980                                                    bool AllowParens) {
981   SMLoc FirstS = getLoc();
982   bool HadParens = false;
983   AsmToken LParen;
984 
985   // If this is an LParen and a parenthesised register name is allowed, parse it
986   // atomically.
987   if (AllowParens && getLexer().is(AsmToken::LParen)) {
988     AsmToken Buf[2];
989     size_t ReadCount = getLexer().peekTokens(Buf);
990     if (ReadCount == 2 && Buf[1].getKind() == AsmToken::RParen) {
991       HadParens = true;
992       LParen = getParser().getTok();
993       getParser().Lex(); // Eat '('
994     }
995   }
996 
997   switch (getLexer().getKind()) {
998   default:
999     if (HadParens)
1000       getLexer().UnLex(LParen);
1001     return MatchOperand_NoMatch;
1002   case AsmToken::Identifier:
1003     StringRef Name = getLexer().getTok().getIdentifier();
1004     Register RegNo;
1005     matchRegisterNameHelper(isRV32E(), RegNo, Name);
1006 
1007     if (RegNo == RISCV::NoRegister) {
1008       if (HadParens)
1009         getLexer().UnLex(LParen);
1010       return MatchOperand_NoMatch;
1011     }
1012     if (HadParens)
1013       Operands.push_back(RISCVOperand::createToken("(", FirstS, isRV64()));
1014     SMLoc S = getLoc();
1015     SMLoc E = SMLoc::getFromPointer(S.getPointer() - 1);
1016     getLexer().Lex();
1017     Operands.push_back(RISCVOperand::createReg(RegNo, S, E, isRV64()));
1018   }
1019 
1020   if (HadParens) {
1021     getParser().Lex(); // Eat ')'
1022     Operands.push_back(RISCVOperand::createToken(")", getLoc(), isRV64()));
1023   }
1024 
1025   return MatchOperand_Success;
1026 }
1027 
1028 OperandMatchResultTy
1029 RISCVAsmParser::parseCSRSystemRegister(OperandVector &Operands) {
1030   SMLoc S = getLoc();
1031   const MCExpr *Res;
1032 
1033   switch (getLexer().getKind()) {
1034   default:
1035     return MatchOperand_NoMatch;
1036   case AsmToken::LParen:
1037   case AsmToken::Minus:
1038   case AsmToken::Plus:
1039   case AsmToken::Exclaim:
1040   case AsmToken::Tilde:
1041   case AsmToken::Integer:
1042   case AsmToken::String: {
1043     if (getParser().parseExpression(Res))
1044       return MatchOperand_ParseFail;
1045 
1046     auto *CE = dyn_cast<MCConstantExpr>(Res);
1047     if (CE) {
1048       int64_t Imm = CE->getValue();
1049       if (isUInt<12>(Imm)) {
1050         auto SysReg = RISCVSysReg::lookupSysRegByEncoding(Imm);
1051         // Accept an immediate representing a named or un-named Sys Reg
1052         // if the range is valid, regardless of the required features.
1053         Operands.push_back(RISCVOperand::createSysReg(
1054             SysReg ? SysReg->Name : "", S, Imm, isRV64()));
1055         return MatchOperand_Success;
1056       }
1057     }
1058 
1059     Twine Msg = "immediate must be an integer in the range";
1060     Error(S, Msg + " [" + Twine(0) + ", " + Twine((1 << 12) - 1) + "]");
1061     return MatchOperand_ParseFail;
1062   }
1063   case AsmToken::Identifier: {
1064     StringRef Identifier;
1065     if (getParser().parseIdentifier(Identifier))
1066       return MatchOperand_ParseFail;
1067 
1068     auto SysReg = RISCVSysReg::lookupSysRegByName(Identifier);
1069     // Accept a named Sys Reg if the required features are present.
1070     if (SysReg) {
1071       if (!SysReg->haveRequiredFeatures(getSTI().getFeatureBits())) {
1072         Error(S, "system register use requires an option to be enabled");
1073         return MatchOperand_ParseFail;
1074       }
1075       Operands.push_back(RISCVOperand::createSysReg(
1076           Identifier, S, SysReg->Encoding, isRV64()));
1077       return MatchOperand_Success;
1078     }
1079 
1080     Twine Msg = "operand must be a valid system register name "
1081                 "or an integer in the range";
1082     Error(S, Msg + " [" + Twine(0) + ", " + Twine((1 << 12) - 1) + "]");
1083     return MatchOperand_ParseFail;
1084   }
1085   case AsmToken::Percent: {
1086     // Discard operand with modifier.
1087     Twine Msg = "immediate must be an integer in the range";
1088     Error(S, Msg + " [" + Twine(0) + ", " + Twine((1 << 12) - 1) + "]");
1089     return MatchOperand_ParseFail;
1090   }
1091   }
1092 
1093   return MatchOperand_NoMatch;
1094 }
1095 
1096 OperandMatchResultTy RISCVAsmParser::parseImmediate(OperandVector &Operands) {
1097   SMLoc S = getLoc();
1098   SMLoc E = SMLoc::getFromPointer(S.getPointer() - 1);
1099   const MCExpr *Res;
1100 
1101   switch (getLexer().getKind()) {
1102   default:
1103     return MatchOperand_NoMatch;
1104   case AsmToken::LParen:
1105   case AsmToken::Dot:
1106   case AsmToken::Minus:
1107   case AsmToken::Plus:
1108   case AsmToken::Exclaim:
1109   case AsmToken::Tilde:
1110   case AsmToken::Integer:
1111   case AsmToken::String:
1112   case AsmToken::Identifier:
1113     if (getParser().parseExpression(Res))
1114       return MatchOperand_ParseFail;
1115     break;
1116   case AsmToken::Percent:
1117     return parseOperandWithModifier(Operands);
1118   }
1119 
1120   Operands.push_back(RISCVOperand::createImm(Res, S, E, isRV64()));
1121   return MatchOperand_Success;
1122 }
1123 
1124 OperandMatchResultTy
1125 RISCVAsmParser::parseOperandWithModifier(OperandVector &Operands) {
1126   SMLoc S = getLoc();
1127   SMLoc E = SMLoc::getFromPointer(S.getPointer() - 1);
1128 
1129   if (getLexer().getKind() != AsmToken::Percent) {
1130     Error(getLoc(), "expected '%' for operand modifier");
1131     return MatchOperand_ParseFail;
1132   }
1133 
1134   getParser().Lex(); // Eat '%'
1135 
1136   if (getLexer().getKind() != AsmToken::Identifier) {
1137     Error(getLoc(), "expected valid identifier for operand modifier");
1138     return MatchOperand_ParseFail;
1139   }
1140   StringRef Identifier = getParser().getTok().getIdentifier();
1141   RISCVMCExpr::VariantKind VK = RISCVMCExpr::getVariantKindForName(Identifier);
1142   if (VK == RISCVMCExpr::VK_RISCV_Invalid) {
1143     Error(getLoc(), "unrecognized operand modifier");
1144     return MatchOperand_ParseFail;
1145   }
1146 
1147   getParser().Lex(); // Eat the identifier
1148   if (getLexer().getKind() != AsmToken::LParen) {
1149     Error(getLoc(), "expected '('");
1150     return MatchOperand_ParseFail;
1151   }
1152   getParser().Lex(); // Eat '('
1153 
1154   const MCExpr *SubExpr;
1155   if (getParser().parseParenExpression(SubExpr, E)) {
1156     return MatchOperand_ParseFail;
1157   }
1158 
1159   const MCExpr *ModExpr = RISCVMCExpr::create(SubExpr, VK, getContext());
1160   Operands.push_back(RISCVOperand::createImm(ModExpr, S, E, isRV64()));
1161   return MatchOperand_Success;
1162 }
1163 
1164 OperandMatchResultTy RISCVAsmParser::parseBareSymbol(OperandVector &Operands) {
1165   SMLoc S = getLoc();
1166   SMLoc E = SMLoc::getFromPointer(S.getPointer() - 1);
1167   const MCExpr *Res;
1168 
1169   if (getLexer().getKind() != AsmToken::Identifier)
1170     return MatchOperand_NoMatch;
1171 
1172   StringRef Identifier;
1173   AsmToken Tok = getLexer().getTok();
1174 
1175   if (getParser().parseIdentifier(Identifier))
1176     return MatchOperand_ParseFail;
1177 
1178   if (Identifier.consume_back("@plt")) {
1179     Error(getLoc(), "'@plt' operand not valid for instruction");
1180     return MatchOperand_ParseFail;
1181   }
1182 
1183   MCSymbol *Sym = getContext().getOrCreateSymbol(Identifier);
1184 
1185   if (Sym->isVariable()) {
1186     const MCExpr *V = Sym->getVariableValue(/*SetUsed=*/false);
1187     if (!isa<MCSymbolRefExpr>(V)) {
1188       getLexer().UnLex(Tok); // Put back if it's not a bare symbol.
1189       return MatchOperand_NoMatch;
1190     }
1191     Res = V;
1192   } else
1193     Res = MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext());
1194 
1195   MCBinaryExpr::Opcode Opcode;
1196   switch (getLexer().getKind()) {
1197   default:
1198     Operands.push_back(RISCVOperand::createImm(Res, S, E, isRV64()));
1199     return MatchOperand_Success;
1200   case AsmToken::Plus:
1201     Opcode = MCBinaryExpr::Add;
1202     break;
1203   case AsmToken::Minus:
1204     Opcode = MCBinaryExpr::Sub;
1205     break;
1206   }
1207 
1208   const MCExpr *Expr;
1209   if (getParser().parseExpression(Expr))
1210     return MatchOperand_ParseFail;
1211   Res = MCBinaryExpr::create(Opcode, Res, Expr, getContext());
1212   Operands.push_back(RISCVOperand::createImm(Res, S, E, isRV64()));
1213   return MatchOperand_Success;
1214 }
1215 
1216 OperandMatchResultTy RISCVAsmParser::parseCallSymbol(OperandVector &Operands) {
1217   SMLoc S = getLoc();
1218   SMLoc E = SMLoc::getFromPointer(S.getPointer() - 1);
1219   const MCExpr *Res;
1220 
1221   if (getLexer().getKind() != AsmToken::Identifier)
1222     return MatchOperand_NoMatch;
1223 
1224   // Avoid parsing the register in `call rd, foo` as a call symbol.
1225   if (getLexer().peekTok().getKind() != AsmToken::EndOfStatement)
1226     return MatchOperand_NoMatch;
1227 
1228   StringRef Identifier;
1229   if (getParser().parseIdentifier(Identifier))
1230     return MatchOperand_ParseFail;
1231 
1232   RISCVMCExpr::VariantKind Kind = RISCVMCExpr::VK_RISCV_CALL;
1233   if (Identifier.consume_back("@plt"))
1234     Kind = RISCVMCExpr::VK_RISCV_CALL_PLT;
1235 
1236   MCSymbol *Sym = getContext().getOrCreateSymbol(Identifier);
1237   Res = MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext());
1238   Res = RISCVMCExpr::create(Res, Kind, getContext());
1239   Operands.push_back(RISCVOperand::createImm(Res, S, E, isRV64()));
1240   return MatchOperand_Success;
1241 }
1242 
1243 OperandMatchResultTy RISCVAsmParser::parseJALOffset(OperandVector &Operands) {
1244   // Parsing jal operands is fiddly due to the `jal foo` and `jal ra, foo`
1245   // both being acceptable forms. When parsing `jal ra, foo` this function
1246   // will be called for the `ra` register operand in an attempt to match the
1247   // single-operand alias. parseJALOffset must fail for this case. It would
1248   // seem logical to try parse the operand using parseImmediate and return
1249   // NoMatch if the next token is a comma (meaning we must be parsing a jal in
1250   // the second form rather than the first). We can't do this as there's no
1251   // way of rewinding the lexer state. Instead, return NoMatch if this operand
1252   // is an identifier and is followed by a comma.
1253   if (getLexer().is(AsmToken::Identifier) &&
1254       getLexer().peekTok().is(AsmToken::Comma))
1255     return MatchOperand_NoMatch;
1256 
1257   return parseImmediate(Operands);
1258 }
1259 
1260 OperandMatchResultTy
1261 RISCVAsmParser::parseMemOpBaseReg(OperandVector &Operands) {
1262   if (getLexer().isNot(AsmToken::LParen)) {
1263     Error(getLoc(), "expected '('");
1264     return MatchOperand_ParseFail;
1265   }
1266 
1267   getParser().Lex(); // Eat '('
1268   Operands.push_back(RISCVOperand::createToken("(", getLoc(), isRV64()));
1269 
1270   if (parseRegister(Operands) != MatchOperand_Success) {
1271     Error(getLoc(), "expected register");
1272     return MatchOperand_ParseFail;
1273   }
1274 
1275   if (getLexer().isNot(AsmToken::RParen)) {
1276     Error(getLoc(), "expected ')'");
1277     return MatchOperand_ParseFail;
1278   }
1279 
1280   getParser().Lex(); // Eat ')'
1281   Operands.push_back(RISCVOperand::createToken(")", getLoc(), isRV64()));
1282 
1283   return MatchOperand_Success;
1284 }
1285 
1286 OperandMatchResultTy RISCVAsmParser::parseAtomicMemOp(OperandVector &Operands) {
1287   // Atomic operations such as lr.w, sc.w, and amo*.w accept a "memory operand"
1288   // as one of their register operands, such as `(a0)`. This just denotes that
1289   // the register (in this case `a0`) contains a memory address.
1290   //
1291   // Normally, we would be able to parse these by putting the parens into the
1292   // instruction string. However, GNU as also accepts a zero-offset memory
1293   // operand (such as `0(a0)`), and ignores the 0. Normally this would be parsed
1294   // with parseImmediate followed by parseMemOpBaseReg, but these instructions
1295   // do not accept an immediate operand, and we do not want to add a "dummy"
1296   // operand that is silently dropped.
1297   //
1298   // Instead, we use this custom parser. This will: allow (and discard) an
1299   // offset if it is zero; require (and discard) parentheses; and add only the
1300   // parsed register operand to `Operands`.
1301   //
1302   // These operands are printed with RISCVInstPrinter::printAtomicMemOp, which
1303   // will only print the register surrounded by parentheses (which GNU as also
1304   // uses as its canonical representation for these operands).
1305   std::unique_ptr<RISCVOperand> OptionalImmOp;
1306 
1307   if (getLexer().isNot(AsmToken::LParen)) {
1308     // Parse an Integer token. We do not accept arbritrary constant expressions
1309     // in the offset field (because they may include parens, which complicates
1310     // parsing a lot).
1311     int64_t ImmVal;
1312     SMLoc ImmStart = getLoc();
1313     if (getParser().parseIntToken(ImmVal,
1314                                   "expected '(' or optional integer offset"))
1315       return MatchOperand_ParseFail;
1316 
1317     // Create a RISCVOperand for checking later (so the error messages are
1318     // nicer), but we don't add it to Operands.
1319     SMLoc ImmEnd = getLoc();
1320     OptionalImmOp =
1321         RISCVOperand::createImm(MCConstantExpr::create(ImmVal, getContext()),
1322                                 ImmStart, ImmEnd, isRV64());
1323   }
1324 
1325   if (getLexer().isNot(AsmToken::LParen)) {
1326     Error(getLoc(), OptionalImmOp ? "expected '(' after optional integer offset"
1327                                   : "expected '(' or optional integer offset");
1328     return MatchOperand_ParseFail;
1329   }
1330   getParser().Lex(); // Eat '('
1331 
1332   if (parseRegister(Operands) != MatchOperand_Success) {
1333     Error(getLoc(), "expected register");
1334     return MatchOperand_ParseFail;
1335   }
1336 
1337   if (getLexer().isNot(AsmToken::RParen)) {
1338     Error(getLoc(), "expected ')'");
1339     return MatchOperand_ParseFail;
1340   }
1341   getParser().Lex(); // Eat ')'
1342 
1343   // Deferred Handling of non-zero offsets. This makes the error messages nicer.
1344   if (OptionalImmOp && !OptionalImmOp->isImmZero()) {
1345     Error(OptionalImmOp->getStartLoc(), "optional integer offset must be 0",
1346           SMRange(OptionalImmOp->getStartLoc(), OptionalImmOp->getEndLoc()));
1347     return MatchOperand_ParseFail;
1348   }
1349 
1350   return MatchOperand_Success;
1351 }
1352 
1353 /// Looks at a token type and creates the relevant operand from this
1354 /// information, adding to Operands. If operand was parsed, returns false, else
1355 /// true.
1356 bool RISCVAsmParser::parseOperand(OperandVector &Operands, StringRef Mnemonic) {
1357   // Check if the current operand has a custom associated parser, if so, try to
1358   // custom parse the operand, or fallback to the general approach.
1359   OperandMatchResultTy Result =
1360       MatchOperandParserImpl(Operands, Mnemonic, /*ParseForAllFeatures=*/true);
1361   if (Result == MatchOperand_Success)
1362     return false;
1363   if (Result == MatchOperand_ParseFail)
1364     return true;
1365 
1366   // Attempt to parse token as a register.
1367   if (parseRegister(Operands, true) == MatchOperand_Success)
1368     return false;
1369 
1370   // Attempt to parse token as an immediate
1371   if (parseImmediate(Operands) == MatchOperand_Success) {
1372     // Parse memory base register if present
1373     if (getLexer().is(AsmToken::LParen))
1374       return parseMemOpBaseReg(Operands) != MatchOperand_Success;
1375     return false;
1376   }
1377 
1378   // Finally we have exhausted all options and must declare defeat.
1379   Error(getLoc(), "unknown operand");
1380   return true;
1381 }
1382 
1383 bool RISCVAsmParser::ParseInstruction(ParseInstructionInfo &Info,
1384                                       StringRef Name, SMLoc NameLoc,
1385                                       OperandVector &Operands) {
1386   // Ensure that if the instruction occurs when relaxation is enabled,
1387   // relocations are forced for the file. Ideally this would be done when there
1388   // is enough information to reliably determine if the instruction itself may
1389   // cause relaxations. Unfortunately instruction processing stage occurs in the
1390   // same pass as relocation emission, so it's too late to set a 'sticky bit'
1391   // for the entire file.
1392   if (getSTI().getFeatureBits()[RISCV::FeatureRelax]) {
1393     auto *Assembler = getTargetStreamer().getStreamer().getAssemblerPtr();
1394     if (Assembler != nullptr) {
1395       RISCVAsmBackend &MAB =
1396           static_cast<RISCVAsmBackend &>(Assembler->getBackend());
1397       MAB.setForceRelocs();
1398     }
1399   }
1400 
1401   // First operand is token for instruction
1402   Operands.push_back(RISCVOperand::createToken(Name, NameLoc, isRV64()));
1403 
1404   // If there are no more operands, then finish
1405   if (getLexer().is(AsmToken::EndOfStatement))
1406     return false;
1407 
1408   // Parse first operand
1409   if (parseOperand(Operands, Name))
1410     return true;
1411 
1412   // Parse until end of statement, consuming commas between operands
1413   unsigned OperandIdx = 1;
1414   while (getLexer().is(AsmToken::Comma)) {
1415     // Consume comma token
1416     getLexer().Lex();
1417 
1418     // Parse next operand
1419     if (parseOperand(Operands, Name))
1420       return true;
1421 
1422     ++OperandIdx;
1423   }
1424 
1425   if (getLexer().isNot(AsmToken::EndOfStatement)) {
1426     SMLoc Loc = getLexer().getLoc();
1427     getParser().eatToEndOfStatement();
1428     return Error(Loc, "unexpected token");
1429   }
1430 
1431   getParser().Lex(); // Consume the EndOfStatement.
1432   return false;
1433 }
1434 
1435 bool RISCVAsmParser::classifySymbolRef(const MCExpr *Expr,
1436                                        RISCVMCExpr::VariantKind &Kind,
1437                                        int64_t &Addend) {
1438   Kind = RISCVMCExpr::VK_RISCV_None;
1439   Addend = 0;
1440 
1441   if (const RISCVMCExpr *RE = dyn_cast<RISCVMCExpr>(Expr)) {
1442     Kind = RE->getKind();
1443     Expr = RE->getSubExpr();
1444   }
1445 
1446   // It's a simple symbol reference or constant with no addend.
1447   if (isa<MCConstantExpr>(Expr) || isa<MCSymbolRefExpr>(Expr))
1448     return true;
1449 
1450   const MCBinaryExpr *BE = dyn_cast<MCBinaryExpr>(Expr);
1451   if (!BE)
1452     return false;
1453 
1454   if (!isa<MCSymbolRefExpr>(BE->getLHS()))
1455     return false;
1456 
1457   if (BE->getOpcode() != MCBinaryExpr::Add &&
1458       BE->getOpcode() != MCBinaryExpr::Sub)
1459     return false;
1460 
1461   // We are able to support the subtraction of two symbol references
1462   if (BE->getOpcode() == MCBinaryExpr::Sub &&
1463       isa<MCSymbolRefExpr>(BE->getRHS()))
1464     return true;
1465 
1466   // See if the addend is a constant, otherwise there's more going
1467   // on here than we can deal with.
1468   auto AddendExpr = dyn_cast<MCConstantExpr>(BE->getRHS());
1469   if (!AddendExpr)
1470     return false;
1471 
1472   Addend = AddendExpr->getValue();
1473   if (BE->getOpcode() == MCBinaryExpr::Sub)
1474     Addend = -Addend;
1475 
1476   // It's some symbol reference + a constant addend
1477   return Kind != RISCVMCExpr::VK_RISCV_Invalid;
1478 }
1479 
1480 bool RISCVAsmParser::ParseDirective(AsmToken DirectiveID) {
1481   // This returns false if this function recognizes the directive
1482   // regardless of whether it is successfully handles or reports an
1483   // error. Otherwise it returns true to give the generic parser a
1484   // chance at recognizing it.
1485   StringRef IDVal = DirectiveID.getString();
1486 
1487   if (IDVal == ".option")
1488     return parseDirectiveOption();
1489 
1490   return true;
1491 }
1492 
1493 bool RISCVAsmParser::parseDirectiveOption() {
1494   MCAsmParser &Parser = getParser();
1495   // Get the option token.
1496   AsmToken Tok = Parser.getTok();
1497   // At the moment only identifiers are supported.
1498   if (Tok.isNot(AsmToken::Identifier))
1499     return Error(Parser.getTok().getLoc(),
1500                  "unexpected token, expected identifier");
1501 
1502   StringRef Option = Tok.getIdentifier();
1503 
1504   if (Option == "push") {
1505     getTargetStreamer().emitDirectiveOptionPush();
1506 
1507     Parser.Lex();
1508     if (Parser.getTok().isNot(AsmToken::EndOfStatement))
1509       return Error(Parser.getTok().getLoc(),
1510                    "unexpected token, expected end of statement");
1511 
1512     pushFeatureBits();
1513     return false;
1514   }
1515 
1516   if (Option == "pop") {
1517     SMLoc StartLoc = Parser.getTok().getLoc();
1518     getTargetStreamer().emitDirectiveOptionPop();
1519 
1520     Parser.Lex();
1521     if (Parser.getTok().isNot(AsmToken::EndOfStatement))
1522       return Error(Parser.getTok().getLoc(),
1523                    "unexpected token, expected end of statement");
1524 
1525     if (popFeatureBits())
1526       return Error(StartLoc, ".option pop with no .option push");
1527 
1528     return false;
1529   }
1530 
1531   if (Option == "rvc") {
1532     getTargetStreamer().emitDirectiveOptionRVC();
1533 
1534     Parser.Lex();
1535     if (Parser.getTok().isNot(AsmToken::EndOfStatement))
1536       return Error(Parser.getTok().getLoc(),
1537                    "unexpected token, expected end of statement");
1538 
1539     setFeatureBits(RISCV::FeatureStdExtC, "c");
1540     return false;
1541   }
1542 
1543   if (Option == "norvc") {
1544     getTargetStreamer().emitDirectiveOptionNoRVC();
1545 
1546     Parser.Lex();
1547     if (Parser.getTok().isNot(AsmToken::EndOfStatement))
1548       return Error(Parser.getTok().getLoc(),
1549                    "unexpected token, expected end of statement");
1550 
1551     clearFeatureBits(RISCV::FeatureStdExtC, "c");
1552     return false;
1553   }
1554 
1555   if (Option == "relax") {
1556     getTargetStreamer().emitDirectiveOptionRelax();
1557 
1558     Parser.Lex();
1559     if (Parser.getTok().isNot(AsmToken::EndOfStatement))
1560       return Error(Parser.getTok().getLoc(),
1561                    "unexpected token, expected end of statement");
1562 
1563     setFeatureBits(RISCV::FeatureRelax, "relax");
1564     return false;
1565   }
1566 
1567   if (Option == "norelax") {
1568     getTargetStreamer().emitDirectiveOptionNoRelax();
1569 
1570     Parser.Lex();
1571     if (Parser.getTok().isNot(AsmToken::EndOfStatement))
1572       return Error(Parser.getTok().getLoc(),
1573                    "unexpected token, expected end of statement");
1574 
1575     clearFeatureBits(RISCV::FeatureRelax, "relax");
1576     return false;
1577   }
1578 
1579   // Unknown option.
1580   Warning(Parser.getTok().getLoc(),
1581           "unknown option, expected 'push', 'pop', 'rvc', 'norvc', 'relax' or "
1582           "'norelax'");
1583   Parser.eatToEndOfStatement();
1584   return false;
1585 }
1586 
1587 void RISCVAsmParser::emitToStreamer(MCStreamer &S, const MCInst &Inst) {
1588   MCInst CInst;
1589   bool Res = compressInst(CInst, Inst, getSTI(), S.getContext());
1590   CInst.setLoc(Inst.getLoc());
1591   S.EmitInstruction((Res ? CInst : Inst), getSTI());
1592 }
1593 
1594 void RISCVAsmParser::emitLoadImm(Register DestReg, int64_t Value,
1595                                  MCStreamer &Out) {
1596   RISCVMatInt::InstSeq Seq;
1597   RISCVMatInt::generateInstSeq(Value, isRV64(), Seq);
1598 
1599   Register SrcReg = RISCV::X0;
1600   for (RISCVMatInt::Inst &Inst : Seq) {
1601     if (Inst.Opc == RISCV::LUI) {
1602       emitToStreamer(
1603           Out, MCInstBuilder(RISCV::LUI).addReg(DestReg).addImm(Inst.Imm));
1604     } else {
1605       emitToStreamer(
1606           Out, MCInstBuilder(Inst.Opc).addReg(DestReg).addReg(SrcReg).addImm(
1607                    Inst.Imm));
1608     }
1609 
1610     // Only the first instruction has X0 as its source.
1611     SrcReg = DestReg;
1612   }
1613 }
1614 
1615 void RISCVAsmParser::emitAuipcInstPair(MCOperand DestReg, MCOperand TmpReg,
1616                                        const MCExpr *Symbol,
1617                                        RISCVMCExpr::VariantKind VKHi,
1618                                        unsigned SecondOpcode, SMLoc IDLoc,
1619                                        MCStreamer &Out) {
1620   // A pair of instructions for PC-relative addressing; expands to
1621   //   TmpLabel: AUIPC TmpReg, VKHi(symbol)
1622   //             OP DestReg, TmpReg, %pcrel_lo(TmpLabel)
1623   MCContext &Ctx = getContext();
1624 
1625   MCSymbol *TmpLabel = Ctx.createTempSymbol(
1626       "pcrel_hi", /* AlwaysAddSuffix */ true, /* CanBeUnnamed */ false);
1627   Out.EmitLabel(TmpLabel);
1628 
1629   const RISCVMCExpr *SymbolHi = RISCVMCExpr::create(Symbol, VKHi, Ctx);
1630   emitToStreamer(
1631       Out, MCInstBuilder(RISCV::AUIPC).addOperand(TmpReg).addExpr(SymbolHi));
1632 
1633   const MCExpr *RefToLinkTmpLabel =
1634       RISCVMCExpr::create(MCSymbolRefExpr::create(TmpLabel, Ctx),
1635                           RISCVMCExpr::VK_RISCV_PCREL_LO, Ctx);
1636 
1637   emitToStreamer(Out, MCInstBuilder(SecondOpcode)
1638                           .addOperand(DestReg)
1639                           .addOperand(TmpReg)
1640                           .addExpr(RefToLinkTmpLabel));
1641 }
1642 
1643 void RISCVAsmParser::emitLoadLocalAddress(MCInst &Inst, SMLoc IDLoc,
1644                                           MCStreamer &Out) {
1645   // The load local address pseudo-instruction "lla" is used in PC-relative
1646   // addressing of local symbols:
1647   //   lla rdest, symbol
1648   // expands to
1649   //   TmpLabel: AUIPC rdest, %pcrel_hi(symbol)
1650   //             ADDI rdest, rdest, %pcrel_lo(TmpLabel)
1651   MCOperand DestReg = Inst.getOperand(0);
1652   const MCExpr *Symbol = Inst.getOperand(1).getExpr();
1653   emitAuipcInstPair(DestReg, DestReg, Symbol, RISCVMCExpr::VK_RISCV_PCREL_HI,
1654                     RISCV::ADDI, IDLoc, Out);
1655 }
1656 
1657 void RISCVAsmParser::emitLoadAddress(MCInst &Inst, SMLoc IDLoc,
1658                                      MCStreamer &Out) {
1659   // The load address pseudo-instruction "la" is used in PC-relative and
1660   // GOT-indirect addressing of global symbols:
1661   //   la rdest, symbol
1662   // expands to either (for non-PIC)
1663   //   TmpLabel: AUIPC rdest, %pcrel_hi(symbol)
1664   //             ADDI rdest, rdest, %pcrel_lo(TmpLabel)
1665   // or (for PIC)
1666   //   TmpLabel: AUIPC rdest, %got_pcrel_hi(symbol)
1667   //             Lx rdest, %pcrel_lo(TmpLabel)(rdest)
1668   MCOperand DestReg = Inst.getOperand(0);
1669   const MCExpr *Symbol = Inst.getOperand(1).getExpr();
1670   unsigned SecondOpcode;
1671   RISCVMCExpr::VariantKind VKHi;
1672   // FIXME: Should check .option (no)pic when implemented
1673   if (getContext().getObjectFileInfo()->isPositionIndependent()) {
1674     SecondOpcode = isRV64() ? RISCV::LD : RISCV::LW;
1675     VKHi = RISCVMCExpr::VK_RISCV_GOT_HI;
1676   } else {
1677     SecondOpcode = RISCV::ADDI;
1678     VKHi = RISCVMCExpr::VK_RISCV_PCREL_HI;
1679   }
1680   emitAuipcInstPair(DestReg, DestReg, Symbol, VKHi, SecondOpcode, IDLoc, Out);
1681 }
1682 
1683 void RISCVAsmParser::emitLoadTLSIEAddress(MCInst &Inst, SMLoc IDLoc,
1684                                           MCStreamer &Out) {
1685   // The load TLS IE address pseudo-instruction "la.tls.ie" is used in
1686   // initial-exec TLS model addressing of global symbols:
1687   //   la.tls.ie rdest, symbol
1688   // expands to
1689   //   TmpLabel: AUIPC rdest, %tls_ie_pcrel_hi(symbol)
1690   //             Lx rdest, %pcrel_lo(TmpLabel)(rdest)
1691   MCOperand DestReg = Inst.getOperand(0);
1692   const MCExpr *Symbol = Inst.getOperand(1).getExpr();
1693   unsigned SecondOpcode = isRV64() ? RISCV::LD : RISCV::LW;
1694   emitAuipcInstPair(DestReg, DestReg, Symbol, RISCVMCExpr::VK_RISCV_TLS_GOT_HI,
1695                     SecondOpcode, IDLoc, Out);
1696 }
1697 
1698 void RISCVAsmParser::emitLoadTLSGDAddress(MCInst &Inst, SMLoc IDLoc,
1699                                           MCStreamer &Out) {
1700   // The load TLS GD address pseudo-instruction "la.tls.gd" is used in
1701   // global-dynamic TLS model addressing of global symbols:
1702   //   la.tls.gd rdest, symbol
1703   // expands to
1704   //   TmpLabel: AUIPC rdest, %tls_gd_pcrel_hi(symbol)
1705   //             ADDI rdest, rdest, %pcrel_lo(TmpLabel)
1706   MCOperand DestReg = Inst.getOperand(0);
1707   const MCExpr *Symbol = Inst.getOperand(1).getExpr();
1708   emitAuipcInstPair(DestReg, DestReg, Symbol, RISCVMCExpr::VK_RISCV_TLS_GD_HI,
1709                     RISCV::ADDI, IDLoc, Out);
1710 }
1711 
1712 void RISCVAsmParser::emitLoadStoreSymbol(MCInst &Inst, unsigned Opcode,
1713                                          SMLoc IDLoc, MCStreamer &Out,
1714                                          bool HasTmpReg) {
1715   // The load/store pseudo-instruction does a pc-relative load with
1716   // a symbol.
1717   //
1718   // The expansion looks like this
1719   //
1720   //   TmpLabel: AUIPC tmp, %pcrel_hi(symbol)
1721   //             [S|L]X    rd, %pcrel_lo(TmpLabel)(tmp)
1722   MCOperand DestReg = Inst.getOperand(0);
1723   unsigned SymbolOpIdx = HasTmpReg ? 2 : 1;
1724   unsigned TmpRegOpIdx = HasTmpReg ? 1 : 0;
1725   MCOperand TmpReg = Inst.getOperand(TmpRegOpIdx);
1726   const MCExpr *Symbol = Inst.getOperand(SymbolOpIdx).getExpr();
1727   emitAuipcInstPair(DestReg, TmpReg, Symbol, RISCVMCExpr::VK_RISCV_PCREL_HI,
1728                     Opcode, IDLoc, Out);
1729 }
1730 
1731 bool RISCVAsmParser::checkPseudoAddTPRel(MCInst &Inst,
1732                                          OperandVector &Operands) {
1733   assert(Inst.getOpcode() == RISCV::PseudoAddTPRel && "Invalid instruction");
1734   assert(Inst.getOperand(2).isReg() && "Unexpected second operand kind");
1735   if (Inst.getOperand(2).getReg() != RISCV::X4) {
1736     SMLoc ErrorLoc = ((RISCVOperand &)*Operands[3]).getStartLoc();
1737     return Error(ErrorLoc, "the second input operand must be tp/x4 when using "
1738                            "%tprel_add modifier");
1739   }
1740 
1741   return false;
1742 }
1743 
1744 bool RISCVAsmParser::processInstruction(MCInst &Inst, SMLoc IDLoc,
1745                                         OperandVector &Operands,
1746                                         MCStreamer &Out) {
1747   Inst.setLoc(IDLoc);
1748 
1749   switch (Inst.getOpcode()) {
1750   default:
1751     break;
1752   case RISCV::PseudoLI: {
1753     Register Reg = Inst.getOperand(0).getReg();
1754     const MCOperand &Op1 = Inst.getOperand(1);
1755     if (Op1.isExpr()) {
1756       // We must have li reg, %lo(sym) or li reg, %pcrel_lo(sym) or similar.
1757       // Just convert to an addi. This allows compatibility with gas.
1758       emitToStreamer(Out, MCInstBuilder(RISCV::ADDI)
1759                               .addReg(Reg)
1760                               .addReg(RISCV::X0)
1761                               .addExpr(Op1.getExpr()));
1762       return false;
1763     }
1764     int64_t Imm = Inst.getOperand(1).getImm();
1765     // On RV32 the immediate here can either be a signed or an unsigned
1766     // 32-bit number. Sign extension has to be performed to ensure that Imm
1767     // represents the expected signed 64-bit number.
1768     if (!isRV64())
1769       Imm = SignExtend64<32>(Imm);
1770     emitLoadImm(Reg, Imm, Out);
1771     return false;
1772   }
1773   case RISCV::PseudoLLA:
1774     emitLoadLocalAddress(Inst, IDLoc, Out);
1775     return false;
1776   case RISCV::PseudoLA:
1777     emitLoadAddress(Inst, IDLoc, Out);
1778     return false;
1779   case RISCV::PseudoLA_TLS_IE:
1780     emitLoadTLSIEAddress(Inst, IDLoc, Out);
1781     return false;
1782   case RISCV::PseudoLA_TLS_GD:
1783     emitLoadTLSGDAddress(Inst, IDLoc, Out);
1784     return false;
1785   case RISCV::PseudoLB:
1786     emitLoadStoreSymbol(Inst, RISCV::LB, IDLoc, Out, /*HasTmpReg=*/false);
1787     return false;
1788   case RISCV::PseudoLBU:
1789     emitLoadStoreSymbol(Inst, RISCV::LBU, IDLoc, Out, /*HasTmpReg=*/false);
1790     return false;
1791   case RISCV::PseudoLH:
1792     emitLoadStoreSymbol(Inst, RISCV::LH, IDLoc, Out, /*HasTmpReg=*/false);
1793     return false;
1794   case RISCV::PseudoLHU:
1795     emitLoadStoreSymbol(Inst, RISCV::LHU, IDLoc, Out, /*HasTmpReg=*/false);
1796     return false;
1797   case RISCV::PseudoLW:
1798     emitLoadStoreSymbol(Inst, RISCV::LW, IDLoc, Out, /*HasTmpReg=*/false);
1799     return false;
1800   case RISCV::PseudoLWU:
1801     emitLoadStoreSymbol(Inst, RISCV::LWU, IDLoc, Out, /*HasTmpReg=*/false);
1802     return false;
1803   case RISCV::PseudoLD:
1804     emitLoadStoreSymbol(Inst, RISCV::LD, IDLoc, Out, /*HasTmpReg=*/false);
1805     return false;
1806   case RISCV::PseudoFLW:
1807     emitLoadStoreSymbol(Inst, RISCV::FLW, IDLoc, Out, /*HasTmpReg=*/true);
1808     return false;
1809   case RISCV::PseudoFLD:
1810     emitLoadStoreSymbol(Inst, RISCV::FLD, IDLoc, Out, /*HasTmpReg=*/true);
1811     return false;
1812   case RISCV::PseudoSB:
1813     emitLoadStoreSymbol(Inst, RISCV::SB, IDLoc, Out, /*HasTmpReg=*/true);
1814     return false;
1815   case RISCV::PseudoSH:
1816     emitLoadStoreSymbol(Inst, RISCV::SH, IDLoc, Out, /*HasTmpReg=*/true);
1817     return false;
1818   case RISCV::PseudoSW:
1819     emitLoadStoreSymbol(Inst, RISCV::SW, IDLoc, Out, /*HasTmpReg=*/true);
1820     return false;
1821   case RISCV::PseudoSD:
1822     emitLoadStoreSymbol(Inst, RISCV::SD, IDLoc, Out, /*HasTmpReg=*/true);
1823     return false;
1824   case RISCV::PseudoFSW:
1825     emitLoadStoreSymbol(Inst, RISCV::FSW, IDLoc, Out, /*HasTmpReg=*/true);
1826     return false;
1827   case RISCV::PseudoFSD:
1828     emitLoadStoreSymbol(Inst, RISCV::FSD, IDLoc, Out, /*HasTmpReg=*/true);
1829     return false;
1830   case RISCV::PseudoAddTPRel:
1831     if (checkPseudoAddTPRel(Inst, Operands))
1832       return true;
1833     break;
1834   }
1835 
1836   emitToStreamer(Out, Inst);
1837   return false;
1838 }
1839 
1840 extern "C" void LLVMInitializeRISCVAsmParser() {
1841   RegisterMCAsmParser<RISCVAsmParser> X(getTheRISCV32Target());
1842   RegisterMCAsmParser<RISCVAsmParser> Y(getTheRISCV64Target());
1843 }
1844