为LLVM添加制导解析

LLVM中制导（pragma信息）的添加来说，对于一些专用编译器非常重要。整个制导信息的添加，大概可以分为以下几步：

• 1、定义Token
• 2、添加对应的句柄（handler）
• 3、添加处理函数声明
• 4、添加对应handler的声明
• 5、实现对应的handler
• 6、实现处理函数

工作需要添加一些制导信息，有些地方写的比较乱，最近比较忙，有空整理下思路和逻辑，比较套路化。

大概的逻辑就是添加对应的Handler，然后在全局初始化处初始化，需要对应的Token后，解析，转到对应的ParseStmt中去解析字符串

1、定义Token

在clangincludeclangBasicTokenKinds.def文件中添加，我这里添加的是ANNOTATION类型，还有其他的类型，选择自己合适的进行添加

ANNOTATION(pragma_SIMD)

2、添加对应的句柄（handler）

在clangincludeclangParseParse.h文件中添加

std::unique_ptr<PragmaHandler> SIMDHandler;

3、添加处理函数声明

StmtResult ParseSIMDStatement() ;

4、添加对应handler的声明

在clanglibParseParsePragma.cpp中添加声明：

struct PragmaSIMDHandler : public PragmaHandler {
  PragmaSIMDHandler(const char *name) : PragmaHandler(name) {}
  void HandlePragma(Preprocessor &PP, PragmaIntroducerKind Introducer,
                    Token &FirstToken) override;
};

并且在void Parser::initializePragmaHandlers()中初始化所有的句柄

  SIMDHandler.reset(new PragmaSIMDHandler("SIMD"));
  PP.AddPragmaHandler(SIMDHandler.get());

5、实现对应的handler

void PragmaSIMDHandler::HandlePragma(Preprocessor &PP,
                                            PragmaIntroducerKind Introducer,
                                            Token &SIMDTok) {
     SourceLocation SIMDLoc = SIMDTok.getLocation();
     Token Tok;
     PP.Lex(Tok);

     MutableArrayRef<Token> Toks(PP.getPreprocessorAllocator().Allocate<Token>(1),
                                                            1);
     Toks[0].startToken();
     Toks[0].setLocation(SIMDLoc);
     Toks[0].setKind(tok::annot_pragma_SIMD);

     PP.EnterTokenStream(Toks, /*DisableMacroExpansion=*/true);      
     return;

}

需要在toolsclanglibParseParseStmt.cpp文件中的StmtResult Parser::ParseStatementOrDeclarationAfterAttributes(StmtVector &Stmts,AllowedContsructsKind Allowed, SourceLocation *TrailingElseLoc,ParsedAttributesWithRange &Attrs)函数中增加对ParseSIMDStatement的调用（这里我贴了全部的函数）

StmtResult
Parser::ParseStatementOrDeclarationAfterAttributes(StmtVector &Stmts,
          AllowedContsructsKind Allowed, SourceLocation *TrailingElseLoc,
          ParsedAttributesWithRange &Attrs) {
  const char *SemiError = nullptr;
  StmtResult Res;

  // Cases in this switch statement should fall through if the parser expects
  // the token to end in a semicolon (in which case SemiError should be set),
  // or they directly 'return;' if not.
Retry:
  tok::TokenKind Kind  = Tok.getKind();
  SourceLocation AtLoc;
  switch (Kind) {
  case tok::at: // May be a @try or @throw statement
    {
      ProhibitAttributes(Attrs); // TODO: is it correct?
      AtLoc = ConsumeToken();  // consume @
      return ParseObjCAtStatement(AtLoc);
    }

  case tok::code_completion:
    Actions.CodeCompleteOrdinaryName(getCurScope(), Sema::PCC_Statement);
    cutOffParsing();
    return StmtError();

  case tok::identifier: {
    Token Next = NextToken();
    if (Next.is(tok::colon)) { // C99 6.8.1: labeled-statement
      // identifier ':' statement
      return ParseLabeledStatement(Attrs);
    }

    // Look up the identifier, and typo-correct it to a keyword if it's not
    // found.
    if (Next.isNot(tok::coloncolon)) {
      // Try to limit which sets of keywords should be included in typo
      // correction based on what the next token is.
      if (TryAnnotateName(/*IsAddressOfOperand*/ false,
                          llvm::make_unique<StatementFilterCCC>(Next)) ==
          ANK_Error) {
        // Handle errors here by skipping up to the next semicolon or '}', and
        // eat the semicolon if that's what stopped us.
        SkipUntil(tok::r_brace, StopAtSemi | StopBeforeMatch);
        if (Tok.is(tok::semi))
          ConsumeToken();
        return StmtError();
      }

      // If the identifier was typo-corrected, try again.
      if (Tok.isNot(tok::identifier))
        goto Retry;
    }

    // Fall through
  }

  case tok::annot_pragma_SIMD:
      return ParseSIMDStatement();

  default: {
    if ((getLangOpts().CPlusPlus || getLangOpts().MicrosoftExt ||
         Allowed == ACK_Any) &&
        isDeclarationStatement()) {
      SourceLocation DeclStart = Tok.getLocation(), DeclEnd;
      DeclGroupPtrTy Decl = ParseDeclaration(Declarator::BlockContext,
                                             DeclEnd, Attrs);
      return Actions.ActOnDeclStmt(Decl, DeclStart, DeclEnd);
    }

    if (Tok.is(tok::r_brace)) {
      Diag(Tok, diag::err_expected_statement);
      return StmtError();
    }

    return ParseExprStatement();
  }

  case tok::kw_case:                // C99 6.8.1: labeled-statement
    return ParseCaseStatement();
  case tok::kw_default:             // C99 6.8.1: labeled-statement
    return ParseDefaultStatement();

  case tok::l_brace:                // C99 6.8.2: compound-statement
    return ParseCompoundStatement();
  case tok::semi: {                 // C99 6.8.3p3: expression[opt] ';'
    bool HasLeadingEmptyMacro = Tok.hasLeadingEmptyMacro();
    return Actions.ActOnNullStmt(ConsumeToken(), HasLeadingEmptyMacro);
  }

  case tok::kw_if:                  // C99 6.8.4.1: if-statement
    return ParseIfStatement(TrailingElseLoc);
  case tok::kw_switch:              // C99 6.8.4.2: switch-statement
    return ParseSwitchStatement(TrailingElseLoc);

  case tok::kw_while:               // C99 6.8.5.1: while-statement
    return ParseWhileStatement(TrailingElseLoc);
  case tok::kw_do:                  // C99 6.8.5.2: do-statement
    Res = ParseDoStatement();
    SemiError = "do/while";
    break;
  case tok::kw_for:                 // C99 6.8.5.3: for-statement
    return ParseForStatement(TrailingElseLoc);

  case tok::kw_goto:                // C99 6.8.6.1: goto-statement
    Res = ParseGotoStatement();
    SemiError = "goto";
    break;
  case tok::kw_continue:            // C99 6.8.6.2: continue-statement
    Res = ParseContinueStatement();
    SemiError = "continue";
    break;
  case tok::kw_break:               // C99 6.8.6.3: break-statement
    Res = ParseBreakStatement();
    SemiError = "break";
    break;
  case tok::kw_return:              // C99 6.8.6.4: return-statement
    Res = ParseReturnStatement();
    SemiError = "return";
    break;
  case tok::kw_co_return:            // C++ Coroutines: co_return statement
    Res = ParseReturnStatement();
    SemiError = "co_return";
    break;

  case tok::kw_asm: {
    ProhibitAttributes(Attrs);
    bool msAsm = false;
    Res = ParseAsmStatement(msAsm);
    Res = Actions.ActOnFinishFullStmt(Res.get());
    if (msAsm) return Res;
    SemiError = "asm";
    break;
  }

  case tok::kw___if_exists:
  case tok::kw___if_not_exists:
    ProhibitAttributes(Attrs);
    ParseMicrosoftIfExistsStatement(Stmts);
    // An __if_exists block is like a compound statement, but it doesn't create
    // a new scope.
    return StmtEmpty();

  case tok::kw_try:                 // C++ 15: try-block
    return ParseCXXTryBlock();

  case tok::kw___try:
    ProhibitAttributes(Attrs); // TODO: is it correct?
    return ParseSEHTryBlock();

  case tok::kw___leave:
    Res = ParseSEHLeaveStatement();
    SemiError = "__leave";
    break;

  case tok::annot_pragma_vis:
    ProhibitAttributes(Attrs);
    HandlePragmaVisibility();
    return StmtEmpty();

  case tok::annot_pragma_pack:
    ProhibitAttributes(Attrs);
    HandlePragmaPack();
    return StmtEmpty();

  case tok::annot_pragma_msstruct:
    ProhibitAttributes(Attrs);
    HandlePragmaMSStruct();
    return StmtEmpty();

  case tok::annot_pragma_align:
    ProhibitAttributes(Attrs);
    HandlePragmaAlign();
    return StmtEmpty();

  case tok::annot_pragma_weak:
    ProhibitAttributes(Attrs);
    HandlePragmaWeak();
    return StmtEmpty();

  case tok::annot_pragma_weakalias:
    ProhibitAttributes(Attrs);
    HandlePragmaWeakAlias();
    return StmtEmpty();

  case tok::annot_pragma_redefine_extname:
    ProhibitAttributes(Attrs);
    HandlePragmaRedefineExtname();
    return StmtEmpty();

  case tok::annot_pragma_fp_contract:
    ProhibitAttributes(Attrs);
    Diag(Tok, diag::err_pragma_fp_contract_scope);
    ConsumeToken();
    return StmtError();

  case tok::annot_pragma_opencl_extension:
    ProhibitAttributes(Attrs);
    HandlePragmaOpenCLExtension();
    return StmtEmpty();

  case tok::annot_pragma_captured:
    ProhibitAttributes(Attrs);
    return HandlePragmaCaptured();

  case tok::annot_pragma_openmp:
    ProhibitAttributes(Attrs);
    return ParseOpenMPDeclarativeOrExecutableDirective(Allowed);

  case tok::annot_pragma_ms_pointers_to_members:
    ProhibitAttributes(Attrs);
    HandlePragmaMSPointersToMembers();
    return StmtEmpty();

  case tok::annot_pragma_ms_pragma:
    ProhibitAttributes(Attrs);
    HandlePragmaMSPragma();
    return StmtEmpty();

  case tok::annot_pragma_ms_vtordisp:
    ProhibitAttributes(Attrs);
    HandlePragmaMSVtorDisp();
    return StmtEmpty();

  case tok::annot_pragma_loop_hint:
    ProhibitAttributes(Attrs);
    return ParsePragmaLoopHint(Stmts, Allowed, TrailingElseLoc, Attrs);

  case tok::annot_pragma_dump:
    HandlePragmaDump();
    return StmtEmpty();
  }

  // If we reached this code, the statement must end in a semicolon.
  if (!TryConsumeToken(tok::semi) && !Res.isInvalid()) {
    // If the result was valid, then we do want to diagnose this.  Use
    // ExpectAndConsume to emit the diagnostic, even though we know it won't
    // succeed.
    ExpectAndConsume(tok::semi, diag::err_expected_semi_after_stmt, SemiError);
    // Skip until we see a } or ;, but don't eat it.
    SkipUntil(tok::r_brace, StopAtSemi | StopBeforeMatch);
  }

  return Res;
}

6、实现处理函数

StmtResult Parser::ParseSIMDStatement() {
    assert(Tok.is(tok::annot_pragma_SIMD) && "Not a SIMD stmt!");
    SourceLocation StartLoc = ConsumeToken();
    StringRef loop_var;
    if(Tok.is(tok::l_paren)){
        ConsumeAnyToken();// eat "("
    }else{
       assert("the loop Variable need be (Variable)");
    }
    IdentifierInfo *Name;
    Name=Tok.getIdentifierInfo();
    loop_var=Name->getName();
    std::cout<<loop_var.str()<<std::endl;
    ConsumeToken();
    //parse finish and end of ")"
    if(Tok.is(tok::r_paren)){
        ConsumeAnyToken();
    }else{
        assert("the loop Variable need be (Variable)");
    }
    if(Tok.is(tok::eod)){
        clang::SimdBeforeFor.insert(pair<bool,string>(true,loop_var.str()));
        ConsumeToken();
        return StmtEmpty();
    }
    return StmtEmpty();
}

这里使用了很多assert，不是一种很好的方式，因为assert只在debug中有效，暂时没有时间处理他人的历史代码，建议使用Diag(Tok, diag::err_expected) << " string or tok::comma";这种llvm内置的Diag信息处理方式，比较好一点。