1 #define MINIMAL_STDERR_OUTPUT 2 3 #include "llvm/Analysis/Passes.h" 4 #include "llvm/ExecutionEngine/ExecutionEngine.h" 5 #include "llvm/ExecutionEngine/JIT.h" 6 #include "llvm/IR/DataLayout.h" 7 #include "llvm/IR/DerivedTypes.h" 8 #include "llvm/IR/IRBuilder.h" 9 #include "llvm/IR/LLVMContext.h" 10 #include "llvm/IR/Module.h" 11 #include "llvm/IR/Verifier.h" 12 #include "llvm/IRReader/IRReader.h" 13 #include "llvm/PassManager.h" 14 #include "llvm/Support/CommandLine.h" 15 #include "llvm/Support/SourceMgr.h" 16 #include "llvm/Support/TargetSelect.h" 17 #include "llvm/Support/raw_ostream.h" 18 #include "llvm/Transforms/Scalar.h" 19 #include <cctype> 20 #include <cstdio> 21 #include <map> 22 #include <string> 23 #include <vector> 24 25 using namespace llvm; 26 27 //===----------------------------------------------------------------------===// 28 // Command-line options 29 //===----------------------------------------------------------------------===// 30 31 namespace { 32 cl::opt<std::string> 33 InputIR("input-IR", 34 cl::desc("Specify the name of an IR file to load for function definitions"), 35 cl::value_desc("input IR file name")); 36 } // namespace 37 38 //===----------------------------------------------------------------------===// 39 // Lexer 40 //===----------------------------------------------------------------------===// 41 42 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one 43 // of these for known things. 44 enum Token { 45 tok_eof = -1, 46 47 // commands 48 tok_def = -2, tok_extern = -3, 49 50 // primary 51 tok_identifier = -4, tok_number = -5, 52 53 // control 54 tok_if = -6, tok_then = -7, tok_else = -8, 55 tok_for = -9, tok_in = -10, 56 57 // operators 58 tok_binary = -11, tok_unary = -12, 59 60 // var definition 61 tok_var = -13 62 }; 63 64 static std::string IdentifierStr; // Filled in if tok_identifier 65 static double NumVal; // Filled in if tok_number 66 67 /// gettok - Return the next token from standard input. 68 static int gettok() { 69 static int LastChar = ' '; 70 71 // Skip any whitespace. 72 while (isspace(LastChar)) 73 LastChar = getchar(); 74 75 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]* 76 IdentifierStr = LastChar; 77 while (isalnum((LastChar = getchar()))) 78 IdentifierStr += LastChar; 79 80 if (IdentifierStr == "def") return tok_def; 81 if (IdentifierStr == "extern") return tok_extern; 82 if (IdentifierStr == "if") return tok_if; 83 if (IdentifierStr == "then") return tok_then; 84 if (IdentifierStr == "else") return tok_else; 85 if (IdentifierStr == "for") return tok_for; 86 if (IdentifierStr == "in") return tok_in; 87 if (IdentifierStr == "binary") return tok_binary; 88 if (IdentifierStr == "unary") return tok_unary; 89 if (IdentifierStr == "var") return tok_var; 90 return tok_identifier; 91 } 92 93 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+ 94 std::string NumStr; 95 do { 96 NumStr += LastChar; 97 LastChar = getchar(); 98 } while (isdigit(LastChar) || LastChar == '.'); 99 100 NumVal = strtod(NumStr.c_str(), 0); 101 return tok_number; 102 } 103 104 if (LastChar == '#') { 105 // Comment until end of line. 106 do LastChar = getchar(); 107 while (LastChar != EOF && LastChar != '\n' && LastChar != '\r'); 108 109 if (LastChar != EOF) 110 return gettok(); 111 } 112 113 // Check for end of file. Don't eat the EOF. 114 if (LastChar == EOF) 115 return tok_eof; 116 117 // Otherwise, just return the character as its ascii value. 118 int ThisChar = LastChar; 119 LastChar = getchar(); 120 return ThisChar; 121 } 122 123 //===----------------------------------------------------------------------===// 124 // Abstract Syntax Tree (aka Parse Tree) 125 //===----------------------------------------------------------------------===// 126 127 /// ExprAST - Base class for all expression nodes. 128 class ExprAST { 129 public: 130 virtual ~ExprAST() {} 131 virtual Value *Codegen() = 0; 132 }; 133 134 /// NumberExprAST - Expression class for numeric literals like "1.0". 135 class NumberExprAST : public ExprAST { 136 double Val; 137 public: 138 NumberExprAST(double val) : Val(val) {} 139 virtual Value *Codegen(); 140 }; 141 142 /// VariableExprAST - Expression class for referencing a variable, like "a". 143 class VariableExprAST : public ExprAST { 144 std::string Name; 145 public: 146 VariableExprAST(const std::string &name) : Name(name) {} 147 const std::string &getName() const { return Name; } 148 virtual Value *Codegen(); 149 }; 150 151 /// UnaryExprAST - Expression class for a unary operator. 152 class UnaryExprAST : public ExprAST { 153 char Opcode; 154 ExprAST *Operand; 155 public: 156 UnaryExprAST(char opcode, ExprAST *operand) 157 : Opcode(opcode), Operand(operand) {} 158 virtual Value *Codegen(); 159 }; 160 161 /// BinaryExprAST - Expression class for a binary operator. 162 class BinaryExprAST : public ExprAST { 163 char Op; 164 ExprAST *LHS, *RHS; 165 public: 166 BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs) 167 : Op(op), LHS(lhs), RHS(rhs) {} 168 virtual Value *Codegen(); 169 }; 170 171 /// CallExprAST - Expression class for function calls. 172 class CallExprAST : public ExprAST { 173 std::string Callee; 174 std::vector<ExprAST*> Args; 175 public: 176 CallExprAST(const std::string &callee, std::vector<ExprAST*> &args) 177 : Callee(callee), Args(args) {} 178 virtual Value *Codegen(); 179 }; 180 181 /// IfExprAST - Expression class for if/then/else. 182 class IfExprAST : public ExprAST { 183 ExprAST *Cond, *Then, *Else; 184 public: 185 IfExprAST(ExprAST *cond, ExprAST *then, ExprAST *_else) 186 : Cond(cond), Then(then), Else(_else) {} 187 virtual Value *Codegen(); 188 }; 189 190 /// ForExprAST - Expression class for for/in. 191 class ForExprAST : public ExprAST { 192 std::string VarName; 193 ExprAST *Start, *End, *Step, *Body; 194 public: 195 ForExprAST(const std::string &varname, ExprAST *start, ExprAST *end, 196 ExprAST *step, ExprAST *body) 197 : VarName(varname), Start(start), End(end), Step(step), Body(body) {} 198 virtual Value *Codegen(); 199 }; 200 201 /// VarExprAST - Expression class for var/in 202 class VarExprAST : public ExprAST { 203 std::vector<std::pair<std::string, ExprAST*> > VarNames; 204 ExprAST *Body; 205 public: 206 VarExprAST(const std::vector<std::pair<std::string, ExprAST*> > &varnames, 207 ExprAST *body) 208 : VarNames(varnames), Body(body) {} 209 210 virtual Value *Codegen(); 211 }; 212 213 /// PrototypeAST - This class represents the "prototype" for a function, 214 /// which captures its argument names as well as if it is an operator. 215 class PrototypeAST { 216 std::string Name; 217 std::vector<std::string> Args; 218 bool isOperator; 219 unsigned Precedence; // Precedence if a binary op. 220 public: 221 PrototypeAST(const std::string &name, const std::vector<std::string> &args, 222 bool isoperator = false, unsigned prec = 0) 223 : Name(name), Args(args), isOperator(isoperator), Precedence(prec) {} 224 225 bool isUnaryOp() const { return isOperator && Args.size() == 1; } 226 bool isBinaryOp() const { return isOperator && Args.size() == 2; } 227 228 char getOperatorName() const { 229 assert(isUnaryOp() || isBinaryOp()); 230 return Name[Name.size()-1]; 231 } 232 233 unsigned getBinaryPrecedence() const { return Precedence; } 234 235 Function *Codegen(); 236 237 void CreateArgumentAllocas(Function *F); 238 }; 239 240 /// FunctionAST - This class represents a function definition itself. 241 class FunctionAST { 242 PrototypeAST *Proto; 243 ExprAST *Body; 244 public: 245 FunctionAST(PrototypeAST *proto, ExprAST *body) 246 : Proto(proto), Body(body) {} 247 248 Function *Codegen(); 249 }; 250 251 //===----------------------------------------------------------------------===// 252 // Parser 253 //===----------------------------------------------------------------------===// 254 255 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current 256 /// token the parser is looking at. getNextToken reads another token from the 257 /// lexer and updates CurTok with its results. 258 static int CurTok; 259 static int getNextToken() { 260 return CurTok = gettok(); 261 } 262 263 /// BinopPrecedence - This holds the precedence for each binary operator that is 264 /// defined. 265 static std::map<char, int> BinopPrecedence; 266 267 /// GetTokPrecedence - Get the precedence of the pending binary operator token. 268 static int GetTokPrecedence() { 269 if (!isascii(CurTok)) 270 return -1; 271 272 // Make sure it's a declared binop. 273 int TokPrec = BinopPrecedence[CurTok]; 274 if (TokPrec <= 0) return -1; 275 return TokPrec; 276 } 277 278 /// Error* - These are little helper functions for error handling. 279 ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;} 280 PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; } 281 FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; } 282 283 static ExprAST *ParseExpression(); 284 285 /// identifierexpr 286 /// ::= identifier 287 /// ::= identifier '(' expression* ')' 288 static ExprAST *ParseIdentifierExpr() { 289 std::string IdName = IdentifierStr; 290 291 getNextToken(); // eat identifier. 292 293 if (CurTok != '(') // Simple variable ref. 294 return new VariableExprAST(IdName); 295 296 // Call. 297 getNextToken(); // eat ( 298 std::vector<ExprAST*> Args; 299 if (CurTok != ')') { 300 while (1) { 301 ExprAST *Arg = ParseExpression(); 302 if (!Arg) return 0; 303 Args.push_back(Arg); 304 305 if (CurTok == ')') break; 306 307 if (CurTok != ',') 308 return Error("Expected ')' or ',' in argument list"); 309 getNextToken(); 310 } 311 } 312 313 // Eat the ')'. 314 getNextToken(); 315 316 return new CallExprAST(IdName, Args); 317 } 318 319 /// numberexpr ::= number 320 static ExprAST *ParseNumberExpr() { 321 ExprAST *Result = new NumberExprAST(NumVal); 322 getNextToken(); // consume the number 323 return Result; 324 } 325 326 /// parenexpr ::= '(' expression ')' 327 static ExprAST *ParseParenExpr() { 328 getNextToken(); // eat (. 329 ExprAST *V = ParseExpression(); 330 if (!V) return 0; 331 332 if (CurTok != ')') 333 return Error("expected ')'"); 334 getNextToken(); // eat ). 335 return V; 336 } 337 338 /// ifexpr ::= 'if' expression 'then' expression 'else' expression 339 static ExprAST *ParseIfExpr() { 340 getNextToken(); // eat the if. 341 342 // condition. 343 ExprAST *Cond = ParseExpression(); 344 if (!Cond) return 0; 345 346 if (CurTok != tok_then) 347 return Error("expected then"); 348 getNextToken(); // eat the then 349 350 ExprAST *Then = ParseExpression(); 351 if (Then == 0) return 0; 352 353 if (CurTok != tok_else) 354 return Error("expected else"); 355 356 getNextToken(); 357 358 ExprAST *Else = ParseExpression(); 359 if (!Else) return 0; 360 361 return new IfExprAST(Cond, Then, Else); 362 } 363 364 /// forexpr ::= 'for' identifier '=' expr ',' expr (',' expr)? 'in' expression 365 static ExprAST *ParseForExpr() { 366 getNextToken(); // eat the for. 367 368 if (CurTok != tok_identifier) 369 return Error("expected identifier after for"); 370 371 std::string IdName = IdentifierStr; 372 getNextToken(); // eat identifier. 373 374 if (CurTok != '=') 375 return Error("expected '=' after for"); 376 getNextToken(); // eat '='. 377 378 379 ExprAST *Start = ParseExpression(); 380 if (Start == 0) return 0; 381 if (CurTok != ',') 382 return Error("expected ',' after for start value"); 383 getNextToken(); 384 385 ExprAST *End = ParseExpression(); 386 if (End == 0) return 0; 387 388 // The step value is optional. 389 ExprAST *Step = 0; 390 if (CurTok == ',') { 391 getNextToken(); 392 Step = ParseExpression(); 393 if (Step == 0) return 0; 394 } 395 396 if (CurTok != tok_in) 397 return Error("expected 'in' after for"); 398 getNextToken(); // eat 'in'. 399 400 ExprAST *Body = ParseExpression(); 401 if (Body == 0) return 0; 402 403 return new ForExprAST(IdName, Start, End, Step, Body); 404 } 405 406 /// varexpr ::= 'var' identifier ('=' expression)? 407 // (',' identifier ('=' expression)?)* 'in' expression 408 static ExprAST *ParseVarExpr() { 409 getNextToken(); // eat the var. 410 411 std::vector<std::pair<std::string, ExprAST*> > VarNames; 412 413 // At least one variable name is required. 414 if (CurTok != tok_identifier) 415 return Error("expected identifier after var"); 416 417 while (1) { 418 std::string Name = IdentifierStr; 419 getNextToken(); // eat identifier. 420 421 // Read the optional initializer. 422 ExprAST *Init = 0; 423 if (CurTok == '=') { 424 getNextToken(); // eat the '='. 425 426 Init = ParseExpression(); 427 if (Init == 0) return 0; 428 } 429 430 VarNames.push_back(std::make_pair(Name, Init)); 431 432 // End of var list, exit loop. 433 if (CurTok != ',') break; 434 getNextToken(); // eat the ','. 435 436 if (CurTok != tok_identifier) 437 return Error("expected identifier list after var"); 438 } 439 440 // At this point, we have to have 'in'. 441 if (CurTok != tok_in) 442 return Error("expected 'in' keyword after 'var'"); 443 getNextToken(); // eat 'in'. 444 445 ExprAST *Body = ParseExpression(); 446 if (Body == 0) return 0; 447 448 return new VarExprAST(VarNames, Body); 449 } 450 451 /// primary 452 /// ::= identifierexpr 453 /// ::= numberexpr 454 /// ::= parenexpr 455 /// ::= ifexpr 456 /// ::= forexpr 457 /// ::= varexpr 458 static ExprAST *ParsePrimary() { 459 switch (CurTok) { 460 default: return Error("unknown token when expecting an expression"); 461 case tok_identifier: return ParseIdentifierExpr(); 462 case tok_number: return ParseNumberExpr(); 463 case '(': return ParseParenExpr(); 464 case tok_if: return ParseIfExpr(); 465 case tok_for: return ParseForExpr(); 466 case tok_var: return ParseVarExpr(); 467 } 468 } 469 470 /// unary 471 /// ::= primary 472 /// ::= '!' unary 473 static ExprAST *ParseUnary() { 474 // If the current token is not an operator, it must be a primary expr. 475 if (!isascii(CurTok) || CurTok == '(' || CurTok == ',') 476 return ParsePrimary(); 477 478 // If this is a unary operator, read it. 479 int Opc = CurTok; 480 getNextToken(); 481 if (ExprAST *Operand = ParseUnary()) 482 return new UnaryExprAST(Opc, Operand); 483 return 0; 484 } 485 486 /// binoprhs 487 /// ::= ('+' unary)* 488 static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) { 489 // If this is a binop, find its precedence. 490 while (1) { 491 int TokPrec = GetTokPrecedence(); 492 493 // If this is a binop that binds at least as tightly as the current binop, 494 // consume it, otherwise we are done. 495 if (TokPrec < ExprPrec) 496 return LHS; 497 498 // Okay, we know this is a binop. 499 int BinOp = CurTok; 500 getNextToken(); // eat binop 501 502 // Parse the unary expression after the binary operator. 503 ExprAST *RHS = ParseUnary(); 504 if (!RHS) return 0; 505 506 // If BinOp binds less tightly with RHS than the operator after RHS, let 507 // the pending operator take RHS as its LHS. 508 int NextPrec = GetTokPrecedence(); 509 if (TokPrec < NextPrec) { 510 RHS = ParseBinOpRHS(TokPrec+1, RHS); 511 if (RHS == 0) return 0; 512 } 513 514 // Merge LHS/RHS. 515 LHS = new BinaryExprAST(BinOp, LHS, RHS); 516 } 517 } 518 519 /// expression 520 /// ::= unary binoprhs 521 /// 522 static ExprAST *ParseExpression() { 523 ExprAST *LHS = ParseUnary(); 524 if (!LHS) return 0; 525 526 return ParseBinOpRHS(0, LHS); 527 } 528 529 /// prototype 530 /// ::= id '(' id* ')' 531 /// ::= binary LETTER number? (id, id) 532 /// ::= unary LETTER (id) 533 static PrototypeAST *ParsePrototype() { 534 std::string FnName; 535 536 unsigned Kind = 0; // 0 = identifier, 1 = unary, 2 = binary. 537 unsigned BinaryPrecedence = 30; 538 539 switch (CurTok) { 540 default: 541 return ErrorP("Expected function name in prototype"); 542 case tok_identifier: 543 FnName = IdentifierStr; 544 Kind = 0; 545 getNextToken(); 546 break; 547 case tok_unary: 548 getNextToken(); 549 if (!isascii(CurTok)) 550 return ErrorP("Expected unary operator"); 551 FnName = "unary"; 552 FnName += (char)CurTok; 553 Kind = 1; 554 getNextToken(); 555 break; 556 case tok_binary: 557 getNextToken(); 558 if (!isascii(CurTok)) 559 return ErrorP("Expected binary operator"); 560 FnName = "binary"; 561 FnName += (char)CurTok; 562 Kind = 2; 563 getNextToken(); 564 565 // Read the precedence if present. 566 if (CurTok == tok_number) { 567 if (NumVal < 1 || NumVal > 100) 568 return ErrorP("Invalid precedecnce: must be 1..100"); 569 BinaryPrecedence = (unsigned)NumVal; 570 getNextToken(); 571 } 572 break; 573 } 574 575 if (CurTok != '(') 576 return ErrorP("Expected '(' in prototype"); 577 578 std::vector<std::string> ArgNames; 579 while (getNextToken() == tok_identifier) 580 ArgNames.push_back(IdentifierStr); 581 if (CurTok != ')') 582 return ErrorP("Expected ')' in prototype"); 583 584 // success. 585 getNextToken(); // eat ')'. 586 587 // Verify right number of names for operator. 588 if (Kind && ArgNames.size() != Kind) 589 return ErrorP("Invalid number of operands for operator"); 590 591 return new PrototypeAST(FnName, ArgNames, Kind != 0, BinaryPrecedence); 592 } 593 594 /// definition ::= 'def' prototype expression 595 static FunctionAST *ParseDefinition() { 596 getNextToken(); // eat def. 597 PrototypeAST *Proto = ParsePrototype(); 598 if (Proto == 0) return 0; 599 600 if (ExprAST *E = ParseExpression()) 601 return new FunctionAST(Proto, E); 602 return 0; 603 } 604 605 /// toplevelexpr ::= expression 606 static FunctionAST *ParseTopLevelExpr() { 607 if (ExprAST *E = ParseExpression()) { 608 // Make an anonymous proto. 609 PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>()); 610 return new FunctionAST(Proto, E); 611 } 612 return 0; 613 } 614 615 /// external ::= 'extern' prototype 616 static PrototypeAST *ParseExtern() { 617 getNextToken(); // eat extern. 618 return ParsePrototype(); 619 } 620 621 //===----------------------------------------------------------------------===// 622 // Code Generation 623 //===----------------------------------------------------------------------===// 624 625 static Module *TheModule; 626 static FunctionPassManager *TheFPM; 627 static IRBuilder<> Builder(getGlobalContext()); 628 static std::map<std::string, AllocaInst*> NamedValues; 629 630 Value *ErrorV(const char *Str) { Error(Str); return 0; } 631 632 /// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of 633 /// the function. This is used for mutable variables etc. 634 static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction, 635 const std::string &VarName) { 636 IRBuilder<> TmpB(&TheFunction->getEntryBlock(), 637 TheFunction->getEntryBlock().begin()); 638 return TmpB.CreateAlloca(Type::getDoubleTy(getGlobalContext()), 0, 639 VarName.c_str()); 640 } 641 642 Value *NumberExprAST::Codegen() { 643 return ConstantFP::get(getGlobalContext(), APFloat(Val)); 644 } 645 646 Value *VariableExprAST::Codegen() { 647 // Look this variable up in the function. 648 Value *V = NamedValues[Name]; 649 if (V == 0) return ErrorV("Unknown variable name"); 650 651 // Load the value. 652 return Builder.CreateLoad(V, Name.c_str()); 653 } 654 655 Value *UnaryExprAST::Codegen() { 656 Value *OperandV = Operand->Codegen(); 657 if (OperandV == 0) return 0; 658 #ifdef USE_MCJIT 659 Function *F = TheHelper->getFunction(MakeLegalFunctionName(std::string("unary")+Opcode)); 660 #else 661 Function *F = TheModule->getFunction(std::string("unary")+Opcode); 662 #endif 663 if (F == 0) 664 return ErrorV("Unknown unary operator"); 665 666 return Builder.CreateCall(F, OperandV, "unop"); 667 } 668 669 Value *BinaryExprAST::Codegen() { 670 // Special case '=' because we don't want to emit the LHS as an expression. 671 if (Op == '=') { 672 // Assignment requires the LHS to be an identifier. 673 // For now, I'm building without RTTI because LLVM builds that way by 674 // default and so we need to build that way to use the command line supprt. 675 // If you build LLVM with RTTI this can be changed back to a dynamic_cast. 676 VariableExprAST *LHSE = reinterpret_cast<VariableExprAST*>(LHS); 677 if (!LHSE) 678 return ErrorV("destination of '=' must be a variable"); 679 // Codegen the RHS. 680 Value *Val = RHS->Codegen(); 681 if (Val == 0) return 0; 682 683 // Look up the name. 684 Value *Variable = NamedValues[LHSE->getName()]; 685 if (Variable == 0) return ErrorV("Unknown variable name"); 686 687 Builder.CreateStore(Val, Variable); 688 return Val; 689 } 690 691 Value *L = LHS->Codegen(); 692 Value *R = RHS->Codegen(); 693 if (L == 0 || R == 0) return 0; 694 695 switch (Op) { 696 case '+': return Builder.CreateFAdd(L, R, "addtmp"); 697 case '-': return Builder.CreateFSub(L, R, "subtmp"); 698 case '*': return Builder.CreateFMul(L, R, "multmp"); 699 case '/': return Builder.CreateFDiv(L, R, "divtmp"); 700 case '<': 701 L = Builder.CreateFCmpULT(L, R, "cmptmp"); 702 // Convert bool 0/1 to double 0.0 or 1.0 703 return Builder.CreateUIToFP(L, Type::getDoubleTy(getGlobalContext()), 704 "booltmp"); 705 default: break; 706 } 707 708 // If it wasn't a builtin binary operator, it must be a user defined one. Emit 709 // a call to it. 710 Function *F = TheModule->getFunction(std::string("binary")+Op); 711 assert(F && "binary operator not found!"); 712 713 Value *Ops[] = { L, R }; 714 return Builder.CreateCall(F, Ops, "binop"); 715 } 716 717 Value *CallExprAST::Codegen() { 718 // Look up the name in the global module table. 719 Function *CalleeF = TheModule->getFunction(Callee); 720 if (CalleeF == 0) { 721 char error_str[64]; 722 sprintf(error_str, "Unknown function referenced %s", Callee.c_str()); 723 return ErrorV(error_str); 724 } 725 726 // If argument mismatch error. 727 if (CalleeF->arg_size() != Args.size()) 728 return ErrorV("Incorrect # arguments passed"); 729 730 std::vector<Value*> ArgsV; 731 for (unsigned i = 0, e = Args.size(); i != e; ++i) { 732 ArgsV.push_back(Args[i]->Codegen()); 733 if (ArgsV.back() == 0) return 0; 734 } 735 736 return Builder.CreateCall(CalleeF, ArgsV, "calltmp"); 737 } 738 739 Value *IfExprAST::Codegen() { 740 Value *CondV = Cond->Codegen(); 741 if (CondV == 0) return 0; 742 743 // Convert condition to a bool by comparing equal to 0.0. 744 CondV = Builder.CreateFCmpONE(CondV, 745 ConstantFP::get(getGlobalContext(), APFloat(0.0)), 746 "ifcond"); 747 748 Function *TheFunction = Builder.GetInsertBlock()->getParent(); 749 750 // Create blocks for the then and else cases. Insert the 'then' block at the 751 // end of the function. 752 BasicBlock *ThenBB = BasicBlock::Create(getGlobalContext(), "then", TheFunction); 753 BasicBlock *ElseBB = BasicBlock::Create(getGlobalContext(), "else"); 754 BasicBlock *MergeBB = BasicBlock::Create(getGlobalContext(), "ifcont"); 755 756 Builder.CreateCondBr(CondV, ThenBB, ElseBB); 757 758 // Emit then value. 759 Builder.SetInsertPoint(ThenBB); 760 761 Value *ThenV = Then->Codegen(); 762 if (ThenV == 0) return 0; 763 764 Builder.CreateBr(MergeBB); 765 // Codegen of 'Then' can change the current block, update ThenBB for the PHI. 766 ThenBB = Builder.GetInsertBlock(); 767 768 // Emit else block. 769 TheFunction->getBasicBlockList().push_back(ElseBB); 770 Builder.SetInsertPoint(ElseBB); 771 772 Value *ElseV = Else->Codegen(); 773 if (ElseV == 0) return 0; 774 775 Builder.CreateBr(MergeBB); 776 // Codegen of 'Else' can change the current block, update ElseBB for the PHI. 777 ElseBB = Builder.GetInsertBlock(); 778 779 // Emit merge block. 780 TheFunction->getBasicBlockList().push_back(MergeBB); 781 Builder.SetInsertPoint(MergeBB); 782 PHINode *PN = Builder.CreatePHI(Type::getDoubleTy(getGlobalContext()), 2, 783 "iftmp"); 784 785 PN->addIncoming(ThenV, ThenBB); 786 PN->addIncoming(ElseV, ElseBB); 787 return PN; 788 } 789 790 Value *ForExprAST::Codegen() { 791 // Output this as: 792 // var = alloca double 793 // ... 794 // start = startexpr 795 // store start -> var 796 // goto loop 797 // loop: 798 // ... 799 // bodyexpr 800 // ... 801 // loopend: 802 // step = stepexpr 803 // endcond = endexpr 804 // 805 // curvar = load var 806 // nextvar = curvar + step 807 // store nextvar -> var 808 // br endcond, loop, endloop 809 // outloop: 810 811 Function *TheFunction = Builder.GetInsertBlock()->getParent(); 812 813 // Create an alloca for the variable in the entry block. 814 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName); 815 816 // Emit the start code first, without 'variable' in scope. 817 Value *StartVal = Start->Codegen(); 818 if (StartVal == 0) return 0; 819 820 // Store the value into the alloca. 821 Builder.CreateStore(StartVal, Alloca); 822 823 // Make the new basic block for the loop header, inserting after current 824 // block. 825 BasicBlock *LoopBB = BasicBlock::Create(getGlobalContext(), "loop", TheFunction); 826 827 // Insert an explicit fall through from the current block to the LoopBB. 828 Builder.CreateBr(LoopBB); 829 830 // Start insertion in LoopBB. 831 Builder.SetInsertPoint(LoopBB); 832 833 // Within the loop, the variable is defined equal to the PHI node. If it 834 // shadows an existing variable, we have to restore it, so save it now. 835 AllocaInst *OldVal = NamedValues[VarName]; 836 NamedValues[VarName] = Alloca; 837 838 // Emit the body of the loop. This, like any other expr, can change the 839 // current BB. Note that we ignore the value computed by the body, but don't 840 // allow an error. 841 if (Body->Codegen() == 0) 842 return 0; 843 844 // Emit the step value. 845 Value *StepVal; 846 if (Step) { 847 StepVal = Step->Codegen(); 848 if (StepVal == 0) return 0; 849 } else { 850 // If not specified, use 1.0. 851 StepVal = ConstantFP::get(getGlobalContext(), APFloat(1.0)); 852 } 853 854 // Compute the end condition. 855 Value *EndCond = End->Codegen(); 856 if (EndCond == 0) return EndCond; 857 858 // Reload, increment, and restore the alloca. This handles the case where 859 // the body of the loop mutates the variable. 860 Value *CurVar = Builder.CreateLoad(Alloca, VarName.c_str()); 861 Value *NextVar = Builder.CreateFAdd(CurVar, StepVal, "nextvar"); 862 Builder.CreateStore(NextVar, Alloca); 863 864 // Convert condition to a bool by comparing equal to 0.0. 865 EndCond = Builder.CreateFCmpONE(EndCond, 866 ConstantFP::get(getGlobalContext(), APFloat(0.0)), 867 "loopcond"); 868 869 // Create the "after loop" block and insert it. 870 BasicBlock *AfterBB = BasicBlock::Create(getGlobalContext(), "afterloop", TheFunction); 871 872 // Insert the conditional branch into the end of LoopEndBB. 873 Builder.CreateCondBr(EndCond, LoopBB, AfterBB); 874 875 // Any new code will be inserted in AfterBB. 876 Builder.SetInsertPoint(AfterBB); 877 878 // Restore the unshadowed variable. 879 if (OldVal) 880 NamedValues[VarName] = OldVal; 881 else 882 NamedValues.erase(VarName); 883 884 885 // for expr always returns 0.0. 886 return Constant::getNullValue(Type::getDoubleTy(getGlobalContext())); 887 } 888 889 Value *VarExprAST::Codegen() { 890 std::vector<AllocaInst *> OldBindings; 891 892 Function *TheFunction = Builder.GetInsertBlock()->getParent(); 893 894 // Register all variables and emit their initializer. 895 for (unsigned i = 0, e = VarNames.size(); i != e; ++i) { 896 const std::string &VarName = VarNames[i].first; 897 ExprAST *Init = VarNames[i].second; 898 899 // Emit the initializer before adding the variable to scope, this prevents 900 // the initializer from referencing the variable itself, and permits stuff 901 // like this: 902 // var a = 1 in 903 // var a = a in ... # refers to outer 'a'. 904 Value *InitVal; 905 if (Init) { 906 InitVal = Init->Codegen(); 907 if (InitVal == 0) return 0; 908 } else { // If not specified, use 0.0. 909 InitVal = ConstantFP::get(getGlobalContext(), APFloat(0.0)); 910 } 911 912 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName); 913 Builder.CreateStore(InitVal, Alloca); 914 915 // Remember the old variable binding so that we can restore the binding when 916 // we unrecurse. 917 OldBindings.push_back(NamedValues[VarName]); 918 919 // Remember this binding. 920 NamedValues[VarName] = Alloca; 921 } 922 923 // Codegen the body, now that all vars are in scope. 924 Value *BodyVal = Body->Codegen(); 925 if (BodyVal == 0) return 0; 926 927 // Pop all our variables from scope. 928 for (unsigned i = 0, e = VarNames.size(); i != e; ++i) 929 NamedValues[VarNames[i].first] = OldBindings[i]; 930 931 // Return the body computation. 932 return BodyVal; 933 } 934 935 Function *PrototypeAST::Codegen() { 936 // Make the function type: double(double,double) etc. 937 std::vector<Type*> Doubles(Args.size(), 938 Type::getDoubleTy(getGlobalContext())); 939 FunctionType *FT = FunctionType::get(Type::getDoubleTy(getGlobalContext()), 940 Doubles, false); 941 942 Function *F = Function::Create(FT, Function::ExternalLinkage, Name, TheModule); 943 // If F conflicted, there was already something named 'Name'. If it has a 944 // body, don't allow redefinition or reextern. 945 if (F->getName() != Name) { 946 // Delete the one we just made and get the existing one. 947 F->eraseFromParent(); 948 F = TheModule->getFunction(Name); 949 // If F already has a body, reject this. 950 if (!F->empty()) { 951 ErrorF("redefinition of function"); 952 return 0; 953 } 954 // If F took a different number of args, reject. 955 if (F->arg_size() != Args.size()) { 956 ErrorF("redefinition of function with different # args"); 957 return 0; 958 } 959 } 960 961 // Set names for all arguments. 962 unsigned Idx = 0; 963 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size(); 964 ++AI, ++Idx) 965 AI->setName(Args[Idx]); 966 967 return F; 968 } 969 970 /// CreateArgumentAllocas - Create an alloca for each argument and register the 971 /// argument in the symbol table so that references to it will succeed. 972 void PrototypeAST::CreateArgumentAllocas(Function *F) { 973 Function::arg_iterator AI = F->arg_begin(); 974 for (unsigned Idx = 0, e = Args.size(); Idx != e; ++Idx, ++AI) { 975 // Create an alloca for this variable. 976 AllocaInst *Alloca = CreateEntryBlockAlloca(F, Args[Idx]); 977 978 // Store the initial value into the alloca. 979 Builder.CreateStore(AI, Alloca); 980 981 // Add arguments to variable symbol table. 982 NamedValues[Args[Idx]] = Alloca; 983 } 984 } 985 986 Function *FunctionAST::Codegen() { 987 NamedValues.clear(); 988 989 Function *TheFunction = Proto->Codegen(); 990 if (TheFunction == 0) 991 return 0; 992 993 // If this is an operator, install it. 994 if (Proto->isBinaryOp()) 995 BinopPrecedence[Proto->getOperatorName()] = Proto->getBinaryPrecedence(); 996 997 // Create a new basic block to start insertion into. 998 BasicBlock *BB = BasicBlock::Create(getGlobalContext(), "entry", TheFunction); 999 Builder.SetInsertPoint(BB); 1000 1001 // Add all arguments to the symbol table and create their allocas. 1002 Proto->CreateArgumentAllocas(TheFunction); 1003 1004 if (Value *RetVal = Body->Codegen()) { 1005 // Finish off the function. 1006 Builder.CreateRet(RetVal); 1007 1008 // Validate the generated code, checking for consistency. 1009 verifyFunction(*TheFunction); 1010 1011 // Optimize the function. 1012 TheFPM->run(*TheFunction); 1013 1014 return TheFunction; 1015 } 1016 1017 // Error reading body, remove function. 1018 TheFunction->eraseFromParent(); 1019 1020 if (Proto->isBinaryOp()) 1021 BinopPrecedence.erase(Proto->getOperatorName()); 1022 return 0; 1023 } 1024 1025 //===----------------------------------------------------------------------===// 1026 // Top-Level parsing and JIT Driver 1027 //===----------------------------------------------------------------------===// 1028 1029 static ExecutionEngine *TheExecutionEngine; 1030 1031 static void HandleDefinition() { 1032 if (FunctionAST *F = ParseDefinition()) { 1033 if (Function *LF = F->Codegen()) { 1034 #ifndef MINIMAL_STDERR_OUTPUT 1035 fprintf(stderr, "Read function definition:"); 1036 LF->dump(); 1037 #endif 1038 } 1039 } else { 1040 // Skip token for error recovery. 1041 getNextToken(); 1042 } 1043 } 1044 1045 static void HandleExtern() { 1046 if (PrototypeAST *P = ParseExtern()) { 1047 if (Function *F = P->Codegen()) { 1048 #ifndef MINIMAL_STDERR_OUTPUT 1049 fprintf(stderr, "Read extern: "); 1050 F->dump(); 1051 #endif 1052 } 1053 } else { 1054 // Skip token for error recovery. 1055 getNextToken(); 1056 } 1057 } 1058 1059 static void HandleTopLevelExpression() { 1060 // Evaluate a top-level expression into an anonymous function. 1061 if (FunctionAST *F = ParseTopLevelExpr()) { 1062 if (Function *LF = F->Codegen()) { 1063 // JIT the function, returning a function pointer. 1064 void *FPtr = TheExecutionEngine->getPointerToFunction(LF); 1065 // Cast it to the right type (takes no arguments, returns a double) so we 1066 // can call it as a native function. 1067 double (*FP)() = (double (*)())(intptr_t)FPtr; 1068 #ifdef MINIMAL_STDERR_OUTPUT 1069 FP(); 1070 #else 1071 fprintf(stderr, "Evaluated to %f\n", FP()); 1072 #endif 1073 } 1074 } else { 1075 // Skip token for error recovery. 1076 getNextToken(); 1077 } 1078 } 1079 1080 /// top ::= definition | external | expression | ';' 1081 static void MainLoop() { 1082 while (1) { 1083 #ifndef MINIMAL_STDERR_OUTPUT 1084 fprintf(stderr, "ready> "); 1085 #endif 1086 switch (CurTok) { 1087 case tok_eof: return; 1088 case ';': getNextToken(); break; // ignore top-level semicolons. 1089 case tok_def: HandleDefinition(); break; 1090 case tok_extern: HandleExtern(); break; 1091 default: HandleTopLevelExpression(); break; 1092 } 1093 } 1094 } 1095 1096 //===----------------------------------------------------------------------===// 1097 // "Library" functions that can be "extern'd" from user code. 1098 //===----------------------------------------------------------------------===// 1099 1100 /// putchard - putchar that takes a double and returns 0. 1101 extern "C" 1102 double putchard(double X) { 1103 putchar((char)X); 1104 return 0; 1105 } 1106 1107 /// printd - printf that takes a double prints it as "%f\n", returning 0. 1108 extern "C" 1109 double printd(double X) { 1110 printf("%f", X); 1111 return 0; 1112 } 1113 1114 extern "C" 1115 double printlf() { 1116 printf("\n"); 1117 return 0; 1118 } 1119 1120 //===----------------------------------------------------------------------===// 1121 // Command line input file handlers 1122 //===----------------------------------------------------------------------===// 1123 1124 Module* parseInputIR(std::string InputFile) { 1125 SMDiagnostic Err; 1126 Module *M = ParseIRFile(InputFile, Err, getGlobalContext()); 1127 if (!M) { 1128 Err.print("IR parsing failed: ", errs()); 1129 return NULL; 1130 } 1131 1132 return M; 1133 } 1134 1135 //===----------------------------------------------------------------------===// 1136 // Main driver code. 1137 //===----------------------------------------------------------------------===// 1138 1139 int main(int argc, char **argv) { 1140 InitializeNativeTarget(); 1141 LLVMContext &Context = getGlobalContext(); 1142 1143 cl::ParseCommandLineOptions(argc, argv, 1144 "Kaleidoscope example program\n"); 1145 1146 // Install standard binary operators. 1147 // 1 is lowest precedence. 1148 BinopPrecedence['='] = 2; 1149 BinopPrecedence['<'] = 10; 1150 BinopPrecedence['+'] = 20; 1151 BinopPrecedence['-'] = 20; 1152 BinopPrecedence['/'] = 40; 1153 BinopPrecedence['*'] = 40; // highest. 1154 1155 // Make the module, which holds all the code. 1156 if (!InputIR.empty()) { 1157 TheModule = parseInputIR(InputIR); 1158 } else { 1159 TheModule = new Module("my cool jit", Context); 1160 } 1161 1162 // Create the JIT. This takes ownership of the module. 1163 std::string ErrStr; 1164 TheExecutionEngine = EngineBuilder(TheModule).setErrorStr(&ErrStr).create(); 1165 if (!TheExecutionEngine) { 1166 fprintf(stderr, "Could not create ExecutionEngine: %s\n", ErrStr.c_str()); 1167 exit(1); 1168 } 1169 1170 FunctionPassManager OurFPM(TheModule); 1171 1172 // Set up the optimizer pipeline. Start with registering info about how the 1173 // target lays out data structures. 1174 OurFPM.add(new DataLayout(*TheExecutionEngine->getDataLayout())); 1175 // Provide basic AliasAnalysis support for GVN. 1176 OurFPM.add(createBasicAliasAnalysisPass()); 1177 // Promote allocas to registers. 1178 OurFPM.add(createPromoteMemoryToRegisterPass()); 1179 // Do simple "peephole" optimizations and bit-twiddling optzns. 1180 OurFPM.add(createInstructionCombiningPass()); 1181 // Reassociate expressions. 1182 OurFPM.add(createReassociatePass()); 1183 // Eliminate Common SubExpressions. 1184 OurFPM.add(createGVNPass()); 1185 // Simplify the control flow graph (deleting unreachable blocks, etc). 1186 OurFPM.add(createCFGSimplificationPass()); 1187 1188 OurFPM.doInitialization(); 1189 1190 // Set the global so the code gen can use this. 1191 TheFPM = &OurFPM; 1192 1193 // Prime the first token. 1194 #ifndef MINIMAL_STDERR_OUTPUT 1195 fprintf(stderr, "ready> "); 1196 #endif 1197 getNextToken(); 1198 1199 // Run the main "interpreter loop" now. 1200 MainLoop(); 1201 1202 // Print out all of the generated code. 1203 TheFPM = 0; 1204 #if !defined(MINIMAL_STDERR_OUTPUT) || defined(DUMP_FINAL_MODULE) 1205 TheModule->dump(); 1206 #endif 1207 return 0; 1208 } 1209