1 //==- WebAssemblyAsmParser.cpp - Assembler for WebAssembly -*- 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 /// \file
10 /// This file is part of the WebAssembly Assembler.
11 ///
12 /// It contains code to translate a parsed .s file into MCInsts.
13 ///
14 //===----------------------------------------------------------------------===//
15 
16 #include "MCTargetDesc/WebAssemblyMCTargetDesc.h"
17 #include "MCTargetDesc/WebAssemblyTargetStreamer.h"
18 #include "TargetInfo/WebAssemblyTargetInfo.h"
19 #include "WebAssembly.h"
20 #include "llvm/MC/MCContext.h"
21 #include "llvm/MC/MCExpr.h"
22 #include "llvm/MC/MCInst.h"
23 #include "llvm/MC/MCInstrInfo.h"
24 #include "llvm/MC/MCParser/MCParsedAsmOperand.h"
25 #include "llvm/MC/MCParser/MCTargetAsmParser.h"
26 #include "llvm/MC/MCSectionWasm.h"
27 #include "llvm/MC/MCStreamer.h"
28 #include "llvm/MC/MCSubtargetInfo.h"
29 #include "llvm/MC/MCSymbol.h"
30 #include "llvm/MC/MCSymbolWasm.h"
31 #include "llvm/Support/Endian.h"
32 #include "llvm/Support/TargetRegistry.h"
33 
34 using namespace llvm;
35 
36 #define DEBUG_TYPE "wasm-asm-parser"
37 
38 namespace {
39 
40 /// WebAssemblyOperand - Instances of this class represent the operands in a
41 /// parsed WASM machine instruction.
42 struct WebAssemblyOperand : public MCParsedAsmOperand {
43   enum KindTy { Token, Integer, Float, Symbol, BrList } Kind;
44 
45   SMLoc StartLoc, EndLoc;
46 
47   struct TokOp {
48     StringRef Tok;
49   };
50 
51   struct IntOp {
52     int64_t Val;
53   };
54 
55   struct FltOp {
56     double Val;
57   };
58 
59   struct SymOp {
60     const MCExpr *Exp;
61   };
62 
63   struct BrLOp {
64     std::vector<unsigned> List;
65   };
66 
67   union {
68     struct TokOp Tok;
69     struct IntOp Int;
70     struct FltOp Flt;
71     struct SymOp Sym;
72     struct BrLOp BrL;
73   };
74 
75   WebAssemblyOperand(KindTy K, SMLoc Start, SMLoc End, TokOp T)
76       : Kind(K), StartLoc(Start), EndLoc(End), Tok(T) {}
77   WebAssemblyOperand(KindTy K, SMLoc Start, SMLoc End, IntOp I)
78       : Kind(K), StartLoc(Start), EndLoc(End), Int(I) {}
79   WebAssemblyOperand(KindTy K, SMLoc Start, SMLoc End, FltOp F)
80       : Kind(K), StartLoc(Start), EndLoc(End), Flt(F) {}
81   WebAssemblyOperand(KindTy K, SMLoc Start, SMLoc End, SymOp S)
82       : Kind(K), StartLoc(Start), EndLoc(End), Sym(S) {}
83   WebAssemblyOperand(KindTy K, SMLoc Start, SMLoc End)
84       : Kind(K), StartLoc(Start), EndLoc(End), BrL() {}
85 
86   ~WebAssemblyOperand() {
87     if (isBrList())
88       BrL.~BrLOp();
89   }
90 
91   bool isToken() const override { return Kind == Token; }
92   bool isImm() const override {
93     return Kind == Integer || Kind == Float || Kind == Symbol;
94   }
95   bool isMem() const override { return false; }
96   bool isReg() const override { return false; }
97   bool isBrList() const { return Kind == BrList; }
98 
99   unsigned getReg() const override {
100     llvm_unreachable("Assembly inspects a register operand");
101     return 0;
102   }
103 
104   StringRef getToken() const {
105     assert(isToken());
106     return Tok.Tok;
107   }
108 
109   SMLoc getStartLoc() const override { return StartLoc; }
110   SMLoc getEndLoc() const override { return EndLoc; }
111 
112   void addRegOperands(MCInst &, unsigned) const {
113     // Required by the assembly matcher.
114     llvm_unreachable("Assembly matcher creates register operands");
115   }
116 
117   void addImmOperands(MCInst &Inst, unsigned N) const {
118     assert(N == 1 && "Invalid number of operands!");
119     if (Kind == Integer)
120       Inst.addOperand(MCOperand::createImm(Int.Val));
121     else if (Kind == Float)
122       Inst.addOperand(MCOperand::createFPImm(Flt.Val));
123     else if (Kind == Symbol)
124       Inst.addOperand(MCOperand::createExpr(Sym.Exp));
125     else
126       llvm_unreachable("Should be immediate or symbol!");
127   }
128 
129   void addBrListOperands(MCInst &Inst, unsigned N) const {
130     assert(N == 1 && isBrList() && "Invalid BrList!");
131     for (auto Br : BrL.List)
132       Inst.addOperand(MCOperand::createImm(Br));
133   }
134 
135   void print(raw_ostream &OS) const override {
136     switch (Kind) {
137     case Token:
138       OS << "Tok:" << Tok.Tok;
139       break;
140     case Integer:
141       OS << "Int:" << Int.Val;
142       break;
143     case Float:
144       OS << "Flt:" << Flt.Val;
145       break;
146     case Symbol:
147       OS << "Sym:" << Sym.Exp;
148       break;
149     case BrList:
150       OS << "BrList:" << BrL.List.size();
151       break;
152     }
153   }
154 };
155 
156 class WebAssemblyAsmParser final : public MCTargetAsmParser {
157   MCAsmParser &Parser;
158   MCAsmLexer &Lexer;
159 
160   // Much like WebAssemblyAsmPrinter in the backend, we have to own these.
161   std::vector<std::unique_ptr<wasm::WasmSignature>> Signatures;
162 
163   // Order of labels, directives and instructions in a .s file have no
164   // syntactical enforcement. This class is a callback from the actual parser,
165   // and yet we have to be feeding data to the streamer in a very particular
166   // order to ensure a correct binary encoding that matches the regular backend
167   // (the streamer does not enforce this). This "state machine" enum helps
168   // guarantee that correct order.
169   enum ParserState {
170     FileStart,
171     Label,
172     FunctionStart,
173     FunctionLocals,
174     Instructions,
175     EndFunction,
176     DataSection,
177   } CurrentState = FileStart;
178 
179   // For ensuring blocks are properly nested.
180   enum NestingType {
181     Function,
182     Block,
183     Loop,
184     Try,
185     If,
186     Else,
187     Undefined,
188   };
189   std::vector<NestingType> NestingStack;
190 
191   // We track this to see if a .functype following a label is the same,
192   // as this is how we recognize the start of a function.
193   MCSymbol *LastLabel = nullptr;
194   MCSymbol *LastFunctionLabel = nullptr;
195 
196 public:
197   WebAssemblyAsmParser(const MCSubtargetInfo &STI, MCAsmParser &Parser,
198                        const MCInstrInfo &MII, const MCTargetOptions &Options)
199       : MCTargetAsmParser(Options, STI, MII), Parser(Parser),
200         Lexer(Parser.getLexer()) {
201     setAvailableFeatures(ComputeAvailableFeatures(STI.getFeatureBits()));
202   }
203 
204 #define GET_ASSEMBLER_HEADER
205 #include "WebAssemblyGenAsmMatcher.inc"
206 
207   // TODO: This is required to be implemented, but appears unused.
208   bool ParseRegister(unsigned & /*RegNo*/, SMLoc & /*StartLoc*/,
209                      SMLoc & /*EndLoc*/) override {
210     llvm_unreachable("ParseRegister is not implemented.");
211   }
212 
213   bool error(const Twine &Msg, const AsmToken &Tok) {
214     return Parser.Error(Tok.getLoc(), Msg + Tok.getString());
215   }
216 
217   bool error(const Twine &Msg) {
218     return Parser.Error(Lexer.getTok().getLoc(), Msg);
219   }
220 
221   void addSignature(std::unique_ptr<wasm::WasmSignature> &&Sig) {
222     Signatures.push_back(std::move(Sig));
223   }
224 
225   std::pair<StringRef, StringRef> nestingString(NestingType NT) {
226     switch (NT) {
227     case Function:
228       return {"function", "end_function"};
229     case Block:
230       return {"block", "end_block"};
231     case Loop:
232       return {"loop", "end_loop"};
233     case Try:
234       return {"try", "end_try"};
235     case If:
236       return {"if", "end_if"};
237     case Else:
238       return {"else", "end_if"};
239     default:
240       llvm_unreachable("unknown NestingType");
241     }
242   }
243 
244   void push(NestingType NT) { NestingStack.push_back(NT); }
245 
246   bool pop(StringRef Ins, NestingType NT1, NestingType NT2 = Undefined) {
247     if (NestingStack.empty())
248       return error(Twine("End of block construct with no start: ") + Ins);
249     auto Top = NestingStack.back();
250     if (Top != NT1 && Top != NT2)
251       return error(Twine("Block construct type mismatch, expected: ") +
252                    nestingString(Top).second + ", instead got: " + Ins);
253     NestingStack.pop_back();
254     return false;
255   }
256 
257   bool ensureEmptyNestingStack() {
258     auto Err = !NestingStack.empty();
259     while (!NestingStack.empty()) {
260       error(Twine("Unmatched block construct(s) at function end: ") +
261             nestingString(NestingStack.back()).first);
262       NestingStack.pop_back();
263     }
264     return Err;
265   }
266 
267   bool isNext(AsmToken::TokenKind Kind) {
268     auto Ok = Lexer.is(Kind);
269     if (Ok)
270       Parser.Lex();
271     return Ok;
272   }
273 
274   bool expect(AsmToken::TokenKind Kind, const char *KindName) {
275     if (!isNext(Kind))
276       return error(std::string("Expected ") + KindName + ", instead got: ",
277                    Lexer.getTok());
278     return false;
279   }
280 
281   StringRef expectIdent() {
282     if (!Lexer.is(AsmToken::Identifier)) {
283       error("Expected identifier, got: ", Lexer.getTok());
284       return StringRef();
285     }
286     auto Name = Lexer.getTok().getString();
287     Parser.Lex();
288     return Name;
289   }
290 
291   Optional<wasm::ValType> parseType(const StringRef &Type) {
292     // FIXME: can't use StringSwitch because wasm::ValType doesn't have a
293     // "invalid" value.
294     if (Type == "i32")
295       return wasm::ValType::I32;
296     if (Type == "i64")
297       return wasm::ValType::I64;
298     if (Type == "f32")
299       return wasm::ValType::F32;
300     if (Type == "f64")
301       return wasm::ValType::F64;
302     if (Type == "v128" || Type == "i8x16" || Type == "i16x8" ||
303         Type == "i32x4" || Type == "i64x2" || Type == "f32x4" ||
304         Type == "f64x2")
305       return wasm::ValType::V128;
306     if (Type == "except_ref")
307       return wasm::ValType::EXCEPT_REF;
308     return Optional<wasm::ValType>();
309   }
310 
311   WebAssembly::ExprType parseBlockType(StringRef ID) {
312     return StringSwitch<WebAssembly::ExprType>(ID)
313         .Case("i32", WebAssembly::ExprType::I32)
314         .Case("i64", WebAssembly::ExprType::I64)
315         .Case("f32", WebAssembly::ExprType::F32)
316         .Case("f64", WebAssembly::ExprType::F64)
317         .Case("v128", WebAssembly::ExprType::V128)
318         .Case("except_ref", WebAssembly::ExprType::ExceptRef)
319         .Case("void", WebAssembly::ExprType::Void)
320         .Default(WebAssembly::ExprType::Invalid);
321   }
322 
323   bool parseRegTypeList(SmallVectorImpl<wasm::ValType> &Types) {
324     while (Lexer.is(AsmToken::Identifier)) {
325       auto Type = parseType(Lexer.getTok().getString());
326       if (!Type)
327         return error("unknown type: ", Lexer.getTok());
328       Types.push_back(Type.getValue());
329       Parser.Lex();
330       if (!isNext(AsmToken::Comma))
331         break;
332     }
333     return false;
334   }
335 
336   void parseSingleInteger(bool IsNegative, OperandVector &Operands) {
337     auto &Int = Lexer.getTok();
338     int64_t Val = Int.getIntVal();
339     if (IsNegative)
340       Val = -Val;
341     Operands.push_back(make_unique<WebAssemblyOperand>(
342         WebAssemblyOperand::Integer, Int.getLoc(), Int.getEndLoc(),
343         WebAssemblyOperand::IntOp{Val}));
344     Parser.Lex();
345   }
346 
347   bool parseOperandStartingWithInteger(bool IsNegative, OperandVector &Operands,
348                                        StringRef InstName) {
349     parseSingleInteger(IsNegative, Operands);
350     // FIXME: there is probably a cleaner way to do this.
351     auto IsLoadStore = InstName.startswith("load") ||
352                        InstName.startswith("store");
353     auto IsAtomic = InstName.startswith("atomic");
354     if (IsLoadStore || IsAtomic) {
355       // Parse load/store operands of the form: offset:p2align=align
356       if (IsLoadStore && isNext(AsmToken::Colon)) {
357         auto Id = expectIdent();
358         if (Id != "p2align")
359           return error("Expected p2align, instead got: " + Id);
360         if (expect(AsmToken::Equal, "="))
361           return true;
362         if (!Lexer.is(AsmToken::Integer))
363           return error("Expected integer constant");
364         parseSingleInteger(false, Operands);
365       } else {
366         // Alignment not specified (or atomics, must use default alignment).
367         // We can't just call WebAssembly::GetDefaultP2Align since we don't have
368         // an opcode until after the assembly matcher, so set a default to fix
369         // up later.
370         auto Tok = Lexer.getTok();
371         Operands.push_back(make_unique<WebAssemblyOperand>(
372             WebAssemblyOperand::Integer, Tok.getLoc(), Tok.getEndLoc(),
373             WebAssemblyOperand::IntOp{-1}));
374       }
375     }
376     return false;
377   }
378 
379   void addBlockTypeOperand(OperandVector &Operands, SMLoc NameLoc,
380                            WebAssembly::ExprType BT) {
381     Operands.push_back(make_unique<WebAssemblyOperand>(
382         WebAssemblyOperand::Integer, NameLoc, NameLoc,
383         WebAssemblyOperand::IntOp{static_cast<int64_t>(BT)}));
384   }
385 
386   bool ParseInstruction(ParseInstructionInfo & /*Info*/, StringRef Name,
387                         SMLoc NameLoc, OperandVector &Operands) override {
388     // Note: Name does NOT point into the sourcecode, but to a local, so
389     // use NameLoc instead.
390     Name = StringRef(NameLoc.getPointer(), Name.size());
391 
392     // WebAssembly has instructions with / in them, which AsmLexer parses
393     // as seperate tokens, so if we find such tokens immediately adjacent (no
394     // whitespace), expand the name to include them:
395     for (;;) {
396       auto &Sep = Lexer.getTok();
397       if (Sep.getLoc().getPointer() != Name.end() ||
398           Sep.getKind() != AsmToken::Slash)
399         break;
400       // Extend name with /
401       Name = StringRef(Name.begin(), Name.size() + Sep.getString().size());
402       Parser.Lex();
403       // We must now find another identifier, or error.
404       auto &Id = Lexer.getTok();
405       if (Id.getKind() != AsmToken::Identifier ||
406           Id.getLoc().getPointer() != Name.end())
407         return error("Incomplete instruction name: ", Id);
408       Name = StringRef(Name.begin(), Name.size() + Id.getString().size());
409       Parser.Lex();
410     }
411 
412     // Now construct the name as first operand.
413     Operands.push_back(make_unique<WebAssemblyOperand>(
414         WebAssemblyOperand::Token, NameLoc, SMLoc::getFromPointer(Name.end()),
415         WebAssemblyOperand::TokOp{Name}));
416     auto NamePair = Name.split('.');
417     // If no '.', there is no type prefix.
418     auto BaseName = NamePair.second.empty() ? NamePair.first : NamePair.second;
419 
420     // If this instruction is part of a control flow structure, ensure
421     // proper nesting.
422     bool ExpectBlockType = false;
423     if (BaseName == "block") {
424       push(Block);
425       ExpectBlockType = true;
426     } else if (BaseName == "loop") {
427       push(Loop);
428       ExpectBlockType = true;
429     } else if (BaseName == "try") {
430       push(Try);
431       ExpectBlockType = true;
432     } else if (BaseName == "if") {
433       push(If);
434       ExpectBlockType = true;
435     } else if (BaseName == "else") {
436       if (pop(BaseName, If))
437         return true;
438       push(Else);
439     } else if (BaseName == "catch") {
440       if (pop(BaseName, Try))
441         return true;
442       push(Try);
443     } else if (BaseName == "end_if") {
444       if (pop(BaseName, If, Else))
445         return true;
446     } else if (BaseName == "end_try") {
447       if (pop(BaseName, Try))
448         return true;
449     } else if (BaseName == "end_loop") {
450       if (pop(BaseName, Loop))
451         return true;
452     } else if (BaseName == "end_block") {
453       if (pop(BaseName, Block))
454         return true;
455     } else if (BaseName == "end_function") {
456       CurrentState = EndFunction;
457       if (pop(BaseName, Function) || ensureEmptyNestingStack())
458         return true;
459     }
460 
461     while (Lexer.isNot(AsmToken::EndOfStatement)) {
462       auto &Tok = Lexer.getTok();
463       switch (Tok.getKind()) {
464       case AsmToken::Identifier: {
465         auto &Id = Lexer.getTok();
466         if (ExpectBlockType) {
467           // Assume this identifier is a block_type.
468           auto BT = parseBlockType(Id.getString());
469           if (BT == WebAssembly::ExprType::Invalid)
470             return error("Unknown block type: ", Id);
471           addBlockTypeOperand(Operands, NameLoc, BT);
472           Parser.Lex();
473         } else {
474           // Assume this identifier is a label.
475           const MCExpr *Val;
476           SMLoc End;
477           if (Parser.parsePrimaryExpr(Val, End))
478             return error("Cannot parse symbol: ", Lexer.getTok());
479           Operands.push_back(make_unique<WebAssemblyOperand>(
480               WebAssemblyOperand::Symbol, Id.getLoc(), Id.getEndLoc(),
481               WebAssemblyOperand::SymOp{Val}));
482         }
483         break;
484       }
485       case AsmToken::Minus:
486         Parser.Lex();
487         if (Lexer.isNot(AsmToken::Integer))
488           return error("Expected integer instead got: ", Lexer.getTok());
489         if (parseOperandStartingWithInteger(true, Operands, BaseName))
490           return true;
491         break;
492       case AsmToken::Integer:
493         if (parseOperandStartingWithInteger(false, Operands, BaseName))
494           return true;
495         break;
496       case AsmToken::Real: {
497         double Val;
498         if (Tok.getString().getAsDouble(Val, false))
499           return error("Cannot parse real: ", Tok);
500         Operands.push_back(make_unique<WebAssemblyOperand>(
501             WebAssemblyOperand::Float, Tok.getLoc(), Tok.getEndLoc(),
502             WebAssemblyOperand::FltOp{Val}));
503         Parser.Lex();
504         break;
505       }
506       case AsmToken::LCurly: {
507         Parser.Lex();
508         auto Op = make_unique<WebAssemblyOperand>(
509             WebAssemblyOperand::BrList, Tok.getLoc(), Tok.getEndLoc());
510         if (!Lexer.is(AsmToken::RCurly))
511           for (;;) {
512             Op->BrL.List.push_back(Lexer.getTok().getIntVal());
513             expect(AsmToken::Integer, "integer");
514             if (!isNext(AsmToken::Comma))
515               break;
516           }
517         expect(AsmToken::RCurly, "}");
518         Operands.push_back(std::move(Op));
519         break;
520       }
521       default:
522         return error("Unexpected token in operand: ", Tok);
523       }
524       if (Lexer.isNot(AsmToken::EndOfStatement)) {
525         if (expect(AsmToken::Comma, ","))
526           return true;
527       }
528     }
529     if (ExpectBlockType && Operands.size() == 1) {
530       // Support blocks with no operands as default to void.
531       addBlockTypeOperand(Operands, NameLoc, WebAssembly::ExprType::Void);
532     }
533     Parser.Lex();
534     return false;
535   }
536 
537   void onLabelParsed(MCSymbol *Symbol) override {
538     LastLabel = Symbol;
539     CurrentState = Label;
540   }
541 
542   bool parseSignature(wasm::WasmSignature *Signature) {
543     if (expect(AsmToken::LParen, "("))
544       return true;
545     if (parseRegTypeList(Signature->Params))
546       return true;
547     if (expect(AsmToken::RParen, ")"))
548       return true;
549     if (expect(AsmToken::MinusGreater, "->"))
550       return true;
551     if (expect(AsmToken::LParen, "("))
552       return true;
553     if (parseRegTypeList(Signature->Returns))
554       return true;
555     if (expect(AsmToken::RParen, ")"))
556       return true;
557     return false;
558   }
559 
560   bool CheckDataSection() {
561     if (CurrentState != DataSection) {
562       auto WS = cast<MCSectionWasm>(getStreamer().getCurrentSection().first);
563       if (WS && WS->getKind().isText())
564         return error("data directive must occur in a data segment: ",
565                      Lexer.getTok());
566     }
567     CurrentState = DataSection;
568     return false;
569   }
570 
571   // This function processes wasm-specific directives streamed to
572   // WebAssemblyTargetStreamer, all others go to the generic parser
573   // (see WasmAsmParser).
574   bool ParseDirective(AsmToken DirectiveID) override {
575     // This function has a really weird return value behavior that is different
576     // from all the other parsing functions:
577     // - return true && no tokens consumed -> don't know this directive / let
578     //   the generic parser handle it.
579     // - return true && tokens consumed -> a parsing error occurred.
580     // - return false -> processed this directive successfully.
581     assert(DirectiveID.getKind() == AsmToken::Identifier);
582     auto &Out = getStreamer();
583     auto &TOut =
584         reinterpret_cast<WebAssemblyTargetStreamer &>(*Out.getTargetStreamer());
585     auto &Ctx = Out.getContext();
586 
587     // TODO: any time we return an error, at least one token must have been
588     // consumed, otherwise this will not signal an error to the caller.
589     if (DirectiveID.getString() == ".globaltype") {
590       auto SymName = expectIdent();
591       if (SymName.empty())
592         return true;
593       if (expect(AsmToken::Comma, ","))
594         return true;
595       auto TypeTok = Lexer.getTok();
596       auto TypeName = expectIdent();
597       if (TypeName.empty())
598         return true;
599       auto Type = parseType(TypeName);
600       if (!Type)
601         return error("Unknown type in .globaltype directive: ", TypeTok);
602       // Now set this symbol with the correct type.
603       auto WasmSym = cast<MCSymbolWasm>(Ctx.getOrCreateSymbol(SymName));
604       WasmSym->setType(wasm::WASM_SYMBOL_TYPE_GLOBAL);
605       WasmSym->setGlobalType(
606           wasm::WasmGlobalType{uint8_t(Type.getValue()), true});
607       // And emit the directive again.
608       TOut.emitGlobalType(WasmSym);
609       return expect(AsmToken::EndOfStatement, "EOL");
610     }
611 
612     if (DirectiveID.getString() == ".functype") {
613       // This code has to send things to the streamer similar to
614       // WebAssemblyAsmPrinter::EmitFunctionBodyStart.
615       // TODO: would be good to factor this into a common function, but the
616       // assembler and backend really don't share any common code, and this code
617       // parses the locals seperately.
618       auto SymName = expectIdent();
619       if (SymName.empty())
620         return true;
621       auto WasmSym = cast<MCSymbolWasm>(Ctx.getOrCreateSymbol(SymName));
622       if (CurrentState == Label && WasmSym == LastLabel) {
623         // This .functype indicates a start of a function.
624         if (ensureEmptyNestingStack())
625           return true;
626         CurrentState = FunctionStart;
627         LastFunctionLabel = LastLabel;
628         push(Function);
629       }
630       auto Signature = make_unique<wasm::WasmSignature>();
631       if (parseSignature(Signature.get()))
632         return true;
633       WasmSym->setSignature(Signature.get());
634       addSignature(std::move(Signature));
635       WasmSym->setType(wasm::WASM_SYMBOL_TYPE_FUNCTION);
636       TOut.emitFunctionType(WasmSym);
637       // TODO: backend also calls TOut.emitIndIdx, but that is not implemented.
638       return expect(AsmToken::EndOfStatement, "EOL");
639     }
640 
641     if (DirectiveID.getString() == ".eventtype") {
642       auto SymName = expectIdent();
643       if (SymName.empty())
644         return true;
645       auto WasmSym = cast<MCSymbolWasm>(Ctx.getOrCreateSymbol(SymName));
646       auto Signature = make_unique<wasm::WasmSignature>();
647       if (parseRegTypeList(Signature->Params))
648         return true;
649       WasmSym->setSignature(Signature.get());
650       addSignature(std::move(Signature));
651       WasmSym->setType(wasm::WASM_SYMBOL_TYPE_EVENT);
652       TOut.emitEventType(WasmSym);
653       // TODO: backend also calls TOut.emitIndIdx, but that is not implemented.
654       return expect(AsmToken::EndOfStatement, "EOL");
655     }
656 
657     if (DirectiveID.getString() == ".local") {
658       if (CurrentState != FunctionStart)
659         return error(".local directive should follow the start of a function",
660                      Lexer.getTok());
661       SmallVector<wasm::ValType, 4> Locals;
662       if (parseRegTypeList(Locals))
663         return true;
664       TOut.emitLocal(Locals);
665       CurrentState = FunctionLocals;
666       return expect(AsmToken::EndOfStatement, "EOL");
667     }
668 
669     if (DirectiveID.getString() == ".int8") {
670       if (CheckDataSection()) return true;
671       int64_t V;
672       if (Parser.parseAbsoluteExpression(V))
673         return error("Cannot parse int8 constant: ", Lexer.getTok());
674       // TODO: error if value doesn't fit?
675       Out.EmitIntValue(static_cast<uint64_t>(V), 1);
676       return expect(AsmToken::EndOfStatement, "EOL");
677     }
678 
679     if (DirectiveID.getString() == ".asciz") {
680       if (CheckDataSection()) return true;
681       std::string S;
682       if (Parser.parseEscapedString(S))
683         return error("Cannot parse string constant: ", Lexer.getTok());
684       Out.EmitBytes(StringRef(S.c_str(), S.length() + 1));
685       return expect(AsmToken::EndOfStatement, "EOL");
686     }
687 
688     return true; // We didn't process this directive.
689   }
690 
691   bool MatchAndEmitInstruction(SMLoc IDLoc, unsigned & /*Opcode*/,
692                                OperandVector &Operands, MCStreamer &Out,
693                                uint64_t &ErrorInfo,
694                                bool MatchingInlineAsm) override {
695     MCInst Inst;
696     unsigned MatchResult =
697         MatchInstructionImpl(Operands, Inst, ErrorInfo, MatchingInlineAsm);
698     switch (MatchResult) {
699     case Match_Success: {
700       if (CurrentState == FunctionStart) {
701         // This is the first instruction in a function, but we haven't seen
702         // a .local directive yet. The streamer requires locals to be encoded
703         // as a prelude to the instructions, so emit an empty list of locals
704         // here.
705         auto &TOut = reinterpret_cast<WebAssemblyTargetStreamer &>(
706             *Out.getTargetStreamer());
707         TOut.emitLocal(SmallVector<wasm::ValType, 0>());
708       }
709       // Fix unknown p2align operands.
710       auto Align = WebAssembly::GetDefaultP2AlignAny(Inst.getOpcode());
711       if (Align != -1U) {
712         auto &Op0 = Inst.getOperand(0);
713         if (Op0.getImm() == -1)
714           Op0.setImm(Align);
715       }
716       Out.EmitInstruction(Inst, getSTI());
717       if (CurrentState == EndFunction) {
718         onEndOfFunction();
719       } else {
720         CurrentState = Instructions;
721       }
722       return false;
723     }
724     case Match_MissingFeature:
725       return Parser.Error(
726           IDLoc, "instruction requires a WASM feature not currently enabled");
727     case Match_MnemonicFail:
728       return Parser.Error(IDLoc, "invalid instruction");
729     case Match_NearMisses:
730       return Parser.Error(IDLoc, "ambiguous instruction");
731     case Match_InvalidTiedOperand:
732     case Match_InvalidOperand: {
733       SMLoc ErrorLoc = IDLoc;
734       if (ErrorInfo != ~0ULL) {
735         if (ErrorInfo >= Operands.size())
736           return Parser.Error(IDLoc, "too few operands for instruction");
737         ErrorLoc = Operands[ErrorInfo]->getStartLoc();
738         if (ErrorLoc == SMLoc())
739           ErrorLoc = IDLoc;
740       }
741       return Parser.Error(ErrorLoc, "invalid operand for instruction");
742     }
743     }
744     llvm_unreachable("Implement any new match types added!");
745   }
746 
747   void doBeforeLabelEmit(MCSymbol *Symbol) override {
748     // Start a new section for the next function automatically, since our
749     // object writer expects each function to have its own section. This way
750     // The user can't forget this "convention".
751     auto SymName = Symbol->getName();
752     if (SymName.startswith(".L"))
753       return; // Local Symbol.
754     auto SecName = ".text." + SymName;
755     auto WS = getContext().getWasmSection(SecName, SectionKind::getText());
756     getStreamer().SwitchSection(WS);
757   }
758 
759   void onEndOfFunction() {
760     // Automatically output a .size directive, so it becomes optional for the
761     // user.
762     if (!LastFunctionLabel) return;
763     auto TempSym = getContext().createLinkerPrivateTempSymbol();
764     getStreamer().EmitLabel(TempSym);
765     auto Start = MCSymbolRefExpr::create(LastFunctionLabel, getContext());
766     auto End = MCSymbolRefExpr::create(TempSym, getContext());
767     auto Expr =
768         MCBinaryExpr::create(MCBinaryExpr::Sub, End, Start, getContext());
769     getStreamer().emitELFSize(LastFunctionLabel, Expr);
770   }
771 
772   void onEndOfFile() override { ensureEmptyNestingStack(); }
773 };
774 } // end anonymous namespace
775 
776 // Force static initialization.
777 extern "C" void LLVMInitializeWebAssemblyAsmParser() {
778   RegisterMCAsmParser<WebAssemblyAsmParser> X(getTheWebAssemblyTarget32());
779   RegisterMCAsmParser<WebAssemblyAsmParser> Y(getTheWebAssemblyTarget64());
780 }
781 
782 #define GET_REGISTER_MATCHER
783 #define GET_MATCHER_IMPLEMENTATION
784 #include "WebAssemblyGenAsmMatcher.inc"
785