toy.cpp revision 354362524a72b3fa43a6c09380b7ae3b2380cbba
1#include "llvm/Analysis/Passes.h" 2#include "llvm/Analysis/Verifier.h" 3#include "llvm/ExecutionEngine/ExecutionEngine.h" 4#include "llvm/ExecutionEngine/JIT.h" 5#include "llvm/IR/DataLayout.h" 6#include "llvm/IR/DerivedTypes.h" 7#include "llvm/IR/IRBuilder.h" 8#include "llvm/IR/LLVMContext.h" 9#include "llvm/IR/Module.h" 10#include "llvm/PassManager.h" 11#include "llvm/Support/TargetSelect.h" 12#include "llvm/Transforms/Scalar.h" 13#include <cctype> 14#include <cstdio> 15#include <map> 16#include <string> 17#include <vector> 18using namespace llvm; 19 20//===----------------------------------------------------------------------===// 21// Lexer 22//===----------------------------------------------------------------------===// 23 24// The lexer returns tokens [0-255] if it is an unknown character, otherwise one 25// of these for known things. 26enum Token { 27 tok_eof = -1, 28 29 // commands 30 tok_def = -2, tok_extern = -3, 31 32 // primary 33 tok_identifier = -4, tok_number = -5, 34 35 // control 36 tok_if = -6, tok_then = -7, tok_else = -8, 37 tok_for = -9, tok_in = -10, 38 39 // operators 40 tok_binary = -11, tok_unary = -12, 41 42 // var definition 43 tok_var = -13 44}; 45 46static std::string IdentifierStr; // Filled in if tok_identifier 47static double NumVal; // Filled in if tok_number 48 49/// gettok - Return the next token from standard input. 50static int gettok() { 51 static int LastChar = ' '; 52 53 // Skip any whitespace. 54 while (isspace(LastChar)) 55 LastChar = getchar(); 56 57 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]* 58 IdentifierStr = LastChar; 59 while (isalnum((LastChar = getchar()))) 60 IdentifierStr += LastChar; 61 62 if (IdentifierStr == "def") return tok_def; 63 if (IdentifierStr == "extern") return tok_extern; 64 if (IdentifierStr == "if") return tok_if; 65 if (IdentifierStr == "then") return tok_then; 66 if (IdentifierStr == "else") return tok_else; 67 if (IdentifierStr == "for") return tok_for; 68 if (IdentifierStr == "in") return tok_in; 69 if (IdentifierStr == "binary") return tok_binary; 70 if (IdentifierStr == "unary") return tok_unary; 71 if (IdentifierStr == "var") return tok_var; 72 return tok_identifier; 73 } 74 75 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+ 76 std::string NumStr; 77 do { 78 NumStr += LastChar; 79 LastChar = getchar(); 80 } while (isdigit(LastChar) || LastChar == '.'); 81 82 NumVal = strtod(NumStr.c_str(), 0); 83 return tok_number; 84 } 85 86 if (LastChar == '#') { 87 // Comment until end of line. 88 do LastChar = getchar(); 89 while (LastChar != EOF && LastChar != '\n' && LastChar != '\r'); 90 91 if (LastChar != EOF) 92 return gettok(); 93 } 94 95 // Check for end of file. Don't eat the EOF. 96 if (LastChar == EOF) 97 return tok_eof; 98 99 // Otherwise, just return the character as its ascii value. 100 int ThisChar = LastChar; 101 LastChar = getchar(); 102 return ThisChar; 103} 104 105//===----------------------------------------------------------------------===// 106// Abstract Syntax Tree (aka Parse Tree) 107//===----------------------------------------------------------------------===// 108 109/// ExprAST - Base class for all expression nodes. 110class ExprAST { 111public: 112 virtual ~ExprAST(); 113 virtual Value *Codegen() = 0; 114}; 115 116// Provide out-of-line definition to prevent weak vtable. 117ExprAST::~ExprAST() {} 118 119/// NumberExprAST - Expression class for numeric literals like "1.0". 120class NumberExprAST : public ExprAST { 121 double Val; 122public: 123 NumberExprAST(double val) : Val(val) {} 124 virtual Value *Codegen(); 125}; 126 127/// VariableExprAST - Expression class for referencing a variable, like "a". 128class VariableExprAST : public ExprAST { 129 std::string Name; 130public: 131 VariableExprAST(const std::string &name) : Name(name) {} 132 const std::string &getName() const { return Name; } 133 virtual Value *Codegen(); 134}; 135 136/// UnaryExprAST - Expression class for a unary operator. 137class UnaryExprAST : public ExprAST { 138 char Opcode; 139 ExprAST *Operand; 140public: 141 UnaryExprAST(char opcode, ExprAST *operand) 142 : Opcode(opcode), Operand(operand) {} 143 virtual Value *Codegen(); 144}; 145 146/// BinaryExprAST - Expression class for a binary operator. 147class BinaryExprAST : public ExprAST { 148 char Op; 149 ExprAST *LHS, *RHS; 150public: 151 BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs) 152 : Op(op), LHS(lhs), RHS(rhs) {} 153 virtual Value *Codegen(); 154}; 155 156/// CallExprAST - Expression class for function calls. 157class CallExprAST : public ExprAST { 158 std::string Callee; 159 std::vector<ExprAST*> Args; 160public: 161 CallExprAST(const std::string &callee, std::vector<ExprAST*> &args) 162 : Callee(callee), Args(args) {} 163 virtual Value *Codegen(); 164}; 165 166/// IfExprAST - Expression class for if/then/else. 167class IfExprAST : public ExprAST { 168 ExprAST *Cond, *Then, *Else; 169public: 170 IfExprAST(ExprAST *cond, ExprAST *then, ExprAST *_else) 171 : Cond(cond), Then(then), Else(_else) {} 172 virtual Value *Codegen(); 173}; 174 175/// ForExprAST - Expression class for for/in. 176class ForExprAST : public ExprAST { 177 std::string VarName; 178 ExprAST *Start, *End, *Step, *Body; 179public: 180 ForExprAST(const std::string &varname, ExprAST *start, ExprAST *end, 181 ExprAST *step, ExprAST *body) 182 : VarName(varname), Start(start), End(end), Step(step), Body(body) {} 183 virtual Value *Codegen(); 184}; 185 186/// VarExprAST - Expression class for var/in 187class VarExprAST : public ExprAST { 188 std::vector<std::pair<std::string, ExprAST*> > VarNames; 189 ExprAST *Body; 190public: 191 VarExprAST(const std::vector<std::pair<std::string, ExprAST*> > &varnames, 192 ExprAST *body) 193 : VarNames(varnames), Body(body) {} 194 195 virtual Value *Codegen(); 196}; 197 198/// PrototypeAST - This class represents the "prototype" for a function, 199/// which captures its argument names as well as if it is an operator. 200class PrototypeAST { 201 std::string Name; 202 std::vector<std::string> Args; 203 bool isOperator; 204 unsigned Precedence; // Precedence if a binary op. 205public: 206 PrototypeAST(const std::string &name, const std::vector<std::string> &args, 207 bool isoperator = false, unsigned prec = 0) 208 : Name(name), Args(args), isOperator(isoperator), Precedence(prec) {} 209 210 bool isUnaryOp() const { return isOperator && Args.size() == 1; } 211 bool isBinaryOp() const { return isOperator && Args.size() == 2; } 212 213 char getOperatorName() const { 214 assert(isUnaryOp() || isBinaryOp()); 215 return Name[Name.size()-1]; 216 } 217 218 unsigned getBinaryPrecedence() const { return Precedence; } 219 220 Function *Codegen(); 221 222 void CreateArgumentAllocas(Function *F); 223}; 224 225/// FunctionAST - This class represents a function definition itself. 226class FunctionAST { 227 PrototypeAST *Proto; 228 ExprAST *Body; 229public: 230 FunctionAST(PrototypeAST *proto, ExprAST *body) 231 : Proto(proto), Body(body) {} 232 233 Function *Codegen(); 234}; 235 236//===----------------------------------------------------------------------===// 237// Parser 238//===----------------------------------------------------------------------===// 239 240/// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current 241/// token the parser is looking at. getNextToken reads another token from the 242/// lexer and updates CurTok with its results. 243static int CurTok; 244static int getNextToken() { 245 return CurTok = gettok(); 246} 247 248/// BinopPrecedence - This holds the precedence for each binary operator that is 249/// defined. 250static std::map<char, int> BinopPrecedence; 251 252/// GetTokPrecedence - Get the precedence of the pending binary operator token. 253static int GetTokPrecedence() { 254 if (!isascii(CurTok)) 255 return -1; 256 257 // Make sure it's a declared binop. 258 int TokPrec = BinopPrecedence[CurTok]; 259 if (TokPrec <= 0) return -1; 260 return TokPrec; 261} 262 263/// Error* - These are little helper functions for error handling. 264ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;} 265PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; } 266FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; } 267 268static ExprAST *ParseExpression(); 269 270/// identifierexpr 271/// ::= identifier 272/// ::= identifier '(' expression* ')' 273static ExprAST *ParseIdentifierExpr() { 274 std::string IdName = IdentifierStr; 275 276 getNextToken(); // eat identifier. 277 278 if (CurTok != '(') // Simple variable ref. 279 return new VariableExprAST(IdName); 280 281 // Call. 282 getNextToken(); // eat ( 283 std::vector<ExprAST*> Args; 284 if (CurTok != ')') { 285 while (1) { 286 ExprAST *Arg = ParseExpression(); 287 if (!Arg) return 0; 288 Args.push_back(Arg); 289 290 if (CurTok == ')') break; 291 292 if (CurTok != ',') 293 return Error("Expected ')' or ',' in argument list"); 294 getNextToken(); 295 } 296 } 297 298 // Eat the ')'. 299 getNextToken(); 300 301 return new CallExprAST(IdName, Args); 302} 303 304/// numberexpr ::= number 305static ExprAST *ParseNumberExpr() { 306 ExprAST *Result = new NumberExprAST(NumVal); 307 getNextToken(); // consume the number 308 return Result; 309} 310 311/// parenexpr ::= '(' expression ')' 312static ExprAST *ParseParenExpr() { 313 getNextToken(); // eat (. 314 ExprAST *V = ParseExpression(); 315 if (!V) return 0; 316 317 if (CurTok != ')') 318 return Error("expected ')'"); 319 getNextToken(); // eat ). 320 return V; 321} 322 323/// ifexpr ::= 'if' expression 'then' expression 'else' expression 324static ExprAST *ParseIfExpr() { 325 getNextToken(); // eat the if. 326 327 // condition. 328 ExprAST *Cond = ParseExpression(); 329 if (!Cond) return 0; 330 331 if (CurTok != tok_then) 332 return Error("expected then"); 333 getNextToken(); // eat the then 334 335 ExprAST *Then = ParseExpression(); 336 if (Then == 0) return 0; 337 338 if (CurTok != tok_else) 339 return Error("expected else"); 340 341 getNextToken(); 342 343 ExprAST *Else = ParseExpression(); 344 if (!Else) return 0; 345 346 return new IfExprAST(Cond, Then, Else); 347} 348 349/// forexpr ::= 'for' identifier '=' expr ',' expr (',' expr)? 'in' expression 350static ExprAST *ParseForExpr() { 351 getNextToken(); // eat the for. 352 353 if (CurTok != tok_identifier) 354 return Error("expected identifier after for"); 355 356 std::string IdName = IdentifierStr; 357 getNextToken(); // eat identifier. 358 359 if (CurTok != '=') 360 return Error("expected '=' after for"); 361 getNextToken(); // eat '='. 362 363 364 ExprAST *Start = ParseExpression(); 365 if (Start == 0) return 0; 366 if (CurTok != ',') 367 return Error("expected ',' after for start value"); 368 getNextToken(); 369 370 ExprAST *End = ParseExpression(); 371 if (End == 0) return 0; 372 373 // The step value is optional. 374 ExprAST *Step = 0; 375 if (CurTok == ',') { 376 getNextToken(); 377 Step = ParseExpression(); 378 if (Step == 0) return 0; 379 } 380 381 if (CurTok != tok_in) 382 return Error("expected 'in' after for"); 383 getNextToken(); // eat 'in'. 384 385 ExprAST *Body = ParseExpression(); 386 if (Body == 0) return 0; 387 388 return new ForExprAST(IdName, Start, End, Step, Body); 389} 390 391/// varexpr ::= 'var' identifier ('=' expression)? 392// (',' identifier ('=' expression)?)* 'in' expression 393static ExprAST *ParseVarExpr() { 394 getNextToken(); // eat the var. 395 396 std::vector<std::pair<std::string, ExprAST*> > VarNames; 397 398 // At least one variable name is required. 399 if (CurTok != tok_identifier) 400 return Error("expected identifier after var"); 401 402 while (1) { 403 std::string Name = IdentifierStr; 404 getNextToken(); // eat identifier. 405 406 // Read the optional initializer. 407 ExprAST *Init = 0; 408 if (CurTok == '=') { 409 getNextToken(); // eat the '='. 410 411 Init = ParseExpression(); 412 if (Init == 0) return 0; 413 } 414 415 VarNames.push_back(std::make_pair(Name, Init)); 416 417 // End of var list, exit loop. 418 if (CurTok != ',') break; 419 getNextToken(); // eat the ','. 420 421 if (CurTok != tok_identifier) 422 return Error("expected identifier list after var"); 423 } 424 425 // At this point, we have to have 'in'. 426 if (CurTok != tok_in) 427 return Error("expected 'in' keyword after 'var'"); 428 getNextToken(); // eat 'in'. 429 430 ExprAST *Body = ParseExpression(); 431 if (Body == 0) return 0; 432 433 return new VarExprAST(VarNames, Body); 434} 435 436/// primary 437/// ::= identifierexpr 438/// ::= numberexpr 439/// ::= parenexpr 440/// ::= ifexpr 441/// ::= forexpr 442/// ::= varexpr 443static ExprAST *ParsePrimary() { 444 switch (CurTok) { 445 default: return Error("unknown token when expecting an expression"); 446 case tok_identifier: return ParseIdentifierExpr(); 447 case tok_number: return ParseNumberExpr(); 448 case '(': return ParseParenExpr(); 449 case tok_if: return ParseIfExpr(); 450 case tok_for: return ParseForExpr(); 451 case tok_var: return ParseVarExpr(); 452 } 453} 454 455/// unary 456/// ::= primary 457/// ::= '!' unary 458static ExprAST *ParseUnary() { 459 // If the current token is not an operator, it must be a primary expr. 460 if (!isascii(CurTok) || CurTok == '(' || CurTok == ',') 461 return ParsePrimary(); 462 463 // If this is a unary operator, read it. 464 int Opc = CurTok; 465 getNextToken(); 466 if (ExprAST *Operand = ParseUnary()) 467 return new UnaryExprAST(Opc, Operand); 468 return 0; 469} 470 471/// binoprhs 472/// ::= ('+' unary)* 473static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) { 474 // If this is a binop, find its precedence. 475 while (1) { 476 int TokPrec = GetTokPrecedence(); 477 478 // If this is a binop that binds at least as tightly as the current binop, 479 // consume it, otherwise we are done. 480 if (TokPrec < ExprPrec) 481 return LHS; 482 483 // Okay, we know this is a binop. 484 int BinOp = CurTok; 485 getNextToken(); // eat binop 486 487 // Parse the unary expression after the binary operator. 488 ExprAST *RHS = ParseUnary(); 489 if (!RHS) return 0; 490 491 // If BinOp binds less tightly with RHS than the operator after RHS, let 492 // the pending operator take RHS as its LHS. 493 int NextPrec = GetTokPrecedence(); 494 if (TokPrec < NextPrec) { 495 RHS = ParseBinOpRHS(TokPrec+1, RHS); 496 if (RHS == 0) return 0; 497 } 498 499 // Merge LHS/RHS. 500 LHS = new BinaryExprAST(BinOp, LHS, RHS); 501 } 502} 503 504/// expression 505/// ::= unary binoprhs 506/// 507static ExprAST *ParseExpression() { 508 ExprAST *LHS = ParseUnary(); 509 if (!LHS) return 0; 510 511 return ParseBinOpRHS(0, LHS); 512} 513 514/// prototype 515/// ::= id '(' id* ')' 516/// ::= binary LETTER number? (id, id) 517/// ::= unary LETTER (id) 518static PrototypeAST *ParsePrototype() { 519 std::string FnName; 520 521 unsigned Kind = 0; // 0 = identifier, 1 = unary, 2 = binary. 522 unsigned BinaryPrecedence = 30; 523 524 switch (CurTok) { 525 default: 526 return ErrorP("Expected function name in prototype"); 527 case tok_identifier: 528 FnName = IdentifierStr; 529 Kind = 0; 530 getNextToken(); 531 break; 532 case tok_unary: 533 getNextToken(); 534 if (!isascii(CurTok)) 535 return ErrorP("Expected unary operator"); 536 FnName = "unary"; 537 FnName += (char)CurTok; 538 Kind = 1; 539 getNextToken(); 540 break; 541 case tok_binary: 542 getNextToken(); 543 if (!isascii(CurTok)) 544 return ErrorP("Expected binary operator"); 545 FnName = "binary"; 546 FnName += (char)CurTok; 547 Kind = 2; 548 getNextToken(); 549 550 // Read the precedence if present. 551 if (CurTok == tok_number) { 552 if (NumVal < 1 || NumVal > 100) 553 return ErrorP("Invalid precedecnce: must be 1..100"); 554 BinaryPrecedence = (unsigned)NumVal; 555 getNextToken(); 556 } 557 break; 558 } 559 560 if (CurTok != '(') 561 return ErrorP("Expected '(' in prototype"); 562 563 std::vector<std::string> ArgNames; 564 while (getNextToken() == tok_identifier) 565 ArgNames.push_back(IdentifierStr); 566 if (CurTok != ')') 567 return ErrorP("Expected ')' in prototype"); 568 569 // success. 570 getNextToken(); // eat ')'. 571 572 // Verify right number of names for operator. 573 if (Kind && ArgNames.size() != Kind) 574 return ErrorP("Invalid number of operands for operator"); 575 576 return new PrototypeAST(FnName, ArgNames, Kind != 0, BinaryPrecedence); 577} 578 579/// definition ::= 'def' prototype expression 580static FunctionAST *ParseDefinition() { 581 getNextToken(); // eat def. 582 PrototypeAST *Proto = ParsePrototype(); 583 if (Proto == 0) return 0; 584 585 if (ExprAST *E = ParseExpression()) 586 return new FunctionAST(Proto, E); 587 return 0; 588} 589 590/// toplevelexpr ::= expression 591static FunctionAST *ParseTopLevelExpr() { 592 if (ExprAST *E = ParseExpression()) { 593 // Make an anonymous proto. 594 PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>()); 595 return new FunctionAST(Proto, E); 596 } 597 return 0; 598} 599 600/// external ::= 'extern' prototype 601static PrototypeAST *ParseExtern() { 602 getNextToken(); // eat extern. 603 return ParsePrototype(); 604} 605 606//===----------------------------------------------------------------------===// 607// Code Generation 608//===----------------------------------------------------------------------===// 609 610static Module *TheModule; 611static IRBuilder<> Builder(getGlobalContext()); 612static std::map<std::string, AllocaInst*> NamedValues; 613static FunctionPassManager *TheFPM; 614 615Value *ErrorV(const char *Str) { Error(Str); return 0; } 616 617/// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of 618/// the function. This is used for mutable variables etc. 619static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction, 620 const std::string &VarName) { 621 IRBuilder<> TmpB(&TheFunction->getEntryBlock(), 622 TheFunction->getEntryBlock().begin()); 623 return TmpB.CreateAlloca(Type::getDoubleTy(getGlobalContext()), 0, 624 VarName.c_str()); 625} 626 627Value *NumberExprAST::Codegen() { 628 return ConstantFP::get(getGlobalContext(), APFloat(Val)); 629} 630 631Value *VariableExprAST::Codegen() { 632 // Look this variable up in the function. 633 Value *V = NamedValues[Name]; 634 if (V == 0) return ErrorV("Unknown variable name"); 635 636 // Load the value. 637 return Builder.CreateLoad(V, Name.c_str()); 638} 639 640Value *UnaryExprAST::Codegen() { 641 Value *OperandV = Operand->Codegen(); 642 if (OperandV == 0) return 0; 643 644 Function *F = TheModule->getFunction(std::string("unary")+Opcode); 645 if (F == 0) 646 return ErrorV("Unknown unary operator"); 647 648 return Builder.CreateCall(F, OperandV, "unop"); 649} 650 651Value *BinaryExprAST::Codegen() { 652 // Special case '=' because we don't want to emit the LHS as an expression. 653 if (Op == '=') { 654 // Assignment requires the LHS to be an identifier. 655 VariableExprAST *LHSE = dynamic_cast<VariableExprAST*>(LHS); 656 if (!LHSE) 657 return ErrorV("destination of '=' must be a variable"); 658 // Codegen the RHS. 659 Value *Val = RHS->Codegen(); 660 if (Val == 0) return 0; 661 662 // Look up the name. 663 Value *Variable = NamedValues[LHSE->getName()]; 664 if (Variable == 0) return ErrorV("Unknown variable name"); 665 666 Builder.CreateStore(Val, Variable); 667 return Val; 668 } 669 670 Value *L = LHS->Codegen(); 671 Value *R = RHS->Codegen(); 672 if (L == 0 || R == 0) return 0; 673 674 switch (Op) { 675 case '+': return Builder.CreateFAdd(L, R, "addtmp"); 676 case '-': return Builder.CreateFSub(L, R, "subtmp"); 677 case '*': return Builder.CreateFMul(L, R, "multmp"); 678 case '<': 679 L = Builder.CreateFCmpULT(L, R, "cmptmp"); 680 // Convert bool 0/1 to double 0.0 or 1.0 681 return Builder.CreateUIToFP(L, Type::getDoubleTy(getGlobalContext()), 682 "booltmp"); 683 default: break; 684 } 685 686 // If it wasn't a builtin binary operator, it must be a user defined one. Emit 687 // a call to it. 688 Function *F = TheModule->getFunction(std::string("binary")+Op); 689 assert(F && "binary operator not found!"); 690 691 Value *Ops[] = { L, R }; 692 return Builder.CreateCall(F, Ops, "binop"); 693} 694 695Value *CallExprAST::Codegen() { 696 // Look up the name in the global module table. 697 Function *CalleeF = TheModule->getFunction(Callee); 698 if (CalleeF == 0) 699 return ErrorV("Unknown function referenced"); 700 701 // If argument mismatch error. 702 if (CalleeF->arg_size() != Args.size()) 703 return ErrorV("Incorrect # arguments passed"); 704 705 std::vector<Value*> ArgsV; 706 for (unsigned i = 0, e = Args.size(); i != e; ++i) { 707 ArgsV.push_back(Args[i]->Codegen()); 708 if (ArgsV.back() == 0) return 0; 709 } 710 711 return Builder.CreateCall(CalleeF, ArgsV, "calltmp"); 712} 713 714Value *IfExprAST::Codegen() { 715 Value *CondV = Cond->Codegen(); 716 if (CondV == 0) return 0; 717 718 // Convert condition to a bool by comparing equal to 0.0. 719 CondV = Builder.CreateFCmpONE(CondV, 720 ConstantFP::get(getGlobalContext(), APFloat(0.0)), 721 "ifcond"); 722 723 Function *TheFunction = Builder.GetInsertBlock()->getParent(); 724 725 // Create blocks for the then and else cases. Insert the 'then' block at the 726 // end of the function. 727 BasicBlock *ThenBB = BasicBlock::Create(getGlobalContext(), "then", TheFunction); 728 BasicBlock *ElseBB = BasicBlock::Create(getGlobalContext(), "else"); 729 BasicBlock *MergeBB = BasicBlock::Create(getGlobalContext(), "ifcont"); 730 731 Builder.CreateCondBr(CondV, ThenBB, ElseBB); 732 733 // Emit then value. 734 Builder.SetInsertPoint(ThenBB); 735 736 Value *ThenV = Then->Codegen(); 737 if (ThenV == 0) return 0; 738 739 Builder.CreateBr(MergeBB); 740 // Codegen of 'Then' can change the current block, update ThenBB for the PHI. 741 ThenBB = Builder.GetInsertBlock(); 742 743 // Emit else block. 744 TheFunction->getBasicBlockList().push_back(ElseBB); 745 Builder.SetInsertPoint(ElseBB); 746 747 Value *ElseV = Else->Codegen(); 748 if (ElseV == 0) return 0; 749 750 Builder.CreateBr(MergeBB); 751 // Codegen of 'Else' can change the current block, update ElseBB for the PHI. 752 ElseBB = Builder.GetInsertBlock(); 753 754 // Emit merge block. 755 TheFunction->getBasicBlockList().push_back(MergeBB); 756 Builder.SetInsertPoint(MergeBB); 757 PHINode *PN = Builder.CreatePHI(Type::getDoubleTy(getGlobalContext()), 2, 758 "iftmp"); 759 760 PN->addIncoming(ThenV, ThenBB); 761 PN->addIncoming(ElseV, ElseBB); 762 return PN; 763} 764 765Value *ForExprAST::Codegen() { 766 // Output this as: 767 // var = alloca double 768 // ... 769 // start = startexpr 770 // store start -> var 771 // goto loop 772 // loop: 773 // ... 774 // bodyexpr 775 // ... 776 // loopend: 777 // step = stepexpr 778 // endcond = endexpr 779 // 780 // curvar = load var 781 // nextvar = curvar + step 782 // store nextvar -> var 783 // br endcond, loop, endloop 784 // outloop: 785 786 Function *TheFunction = Builder.GetInsertBlock()->getParent(); 787 788 // Create an alloca for the variable in the entry block. 789 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName); 790 791 // Emit the start code first, without 'variable' in scope. 792 Value *StartVal = Start->Codegen(); 793 if (StartVal == 0) return 0; 794 795 // Store the value into the alloca. 796 Builder.CreateStore(StartVal, Alloca); 797 798 // Make the new basic block for the loop header, inserting after current 799 // block. 800 BasicBlock *LoopBB = BasicBlock::Create(getGlobalContext(), "loop", TheFunction); 801 802 // Insert an explicit fall through from the current block to the LoopBB. 803 Builder.CreateBr(LoopBB); 804 805 // Start insertion in LoopBB. 806 Builder.SetInsertPoint(LoopBB); 807 808 // Within the loop, the variable is defined equal to the PHI node. If it 809 // shadows an existing variable, we have to restore it, so save it now. 810 AllocaInst *OldVal = NamedValues[VarName]; 811 NamedValues[VarName] = Alloca; 812 813 // Emit the body of the loop. This, like any other expr, can change the 814 // current BB. Note that we ignore the value computed by the body, but don't 815 // allow an error. 816 if (Body->Codegen() == 0) 817 return 0; 818 819 // Emit the step value. 820 Value *StepVal; 821 if (Step) { 822 StepVal = Step->Codegen(); 823 if (StepVal == 0) return 0; 824 } else { 825 // If not specified, use 1.0. 826 StepVal = ConstantFP::get(getGlobalContext(), APFloat(1.0)); 827 } 828 829 // Compute the end condition. 830 Value *EndCond = End->Codegen(); 831 if (EndCond == 0) return EndCond; 832 833 // Reload, increment, and restore the alloca. This handles the case where 834 // the body of the loop mutates the variable. 835 Value *CurVar = Builder.CreateLoad(Alloca, VarName.c_str()); 836 Value *NextVar = Builder.CreateFAdd(CurVar, StepVal, "nextvar"); 837 Builder.CreateStore(NextVar, Alloca); 838 839 // Convert condition to a bool by comparing equal to 0.0. 840 EndCond = Builder.CreateFCmpONE(EndCond, 841 ConstantFP::get(getGlobalContext(), APFloat(0.0)), 842 "loopcond"); 843 844 // Create the "after loop" block and insert it. 845 BasicBlock *AfterBB = BasicBlock::Create(getGlobalContext(), "afterloop", TheFunction); 846 847 // Insert the conditional branch into the end of LoopEndBB. 848 Builder.CreateCondBr(EndCond, LoopBB, AfterBB); 849 850 // Any new code will be inserted in AfterBB. 851 Builder.SetInsertPoint(AfterBB); 852 853 // Restore the unshadowed variable. 854 if (OldVal) 855 NamedValues[VarName] = OldVal; 856 else 857 NamedValues.erase(VarName); 858 859 860 // for expr always returns 0.0. 861 return Constant::getNullValue(Type::getDoubleTy(getGlobalContext())); 862} 863 864Value *VarExprAST::Codegen() { 865 std::vector<AllocaInst *> OldBindings; 866 867 Function *TheFunction = Builder.GetInsertBlock()->getParent(); 868 869 // Register all variables and emit their initializer. 870 for (unsigned i = 0, e = VarNames.size(); i != e; ++i) { 871 const std::string &VarName = VarNames[i].first; 872 ExprAST *Init = VarNames[i].second; 873 874 // Emit the initializer before adding the variable to scope, this prevents 875 // the initializer from referencing the variable itself, and permits stuff 876 // like this: 877 // var a = 1 in 878 // var a = a in ... # refers to outer 'a'. 879 Value *InitVal; 880 if (Init) { 881 InitVal = Init->Codegen(); 882 if (InitVal == 0) return 0; 883 } else { // If not specified, use 0.0. 884 InitVal = ConstantFP::get(getGlobalContext(), APFloat(0.0)); 885 } 886 887 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName); 888 Builder.CreateStore(InitVal, Alloca); 889 890 // Remember the old variable binding so that we can restore the binding when 891 // we unrecurse. 892 OldBindings.push_back(NamedValues[VarName]); 893 894 // Remember this binding. 895 NamedValues[VarName] = Alloca; 896 } 897 898 // Codegen the body, now that all vars are in scope. 899 Value *BodyVal = Body->Codegen(); 900 if (BodyVal == 0) return 0; 901 902 // Pop all our variables from scope. 903 for (unsigned i = 0, e = VarNames.size(); i != e; ++i) 904 NamedValues[VarNames[i].first] = OldBindings[i]; 905 906 // Return the body computation. 907 return BodyVal; 908} 909 910Function *PrototypeAST::Codegen() { 911 // Make the function type: double(double,double) etc. 912 std::vector<Type*> Doubles(Args.size(), 913 Type::getDoubleTy(getGlobalContext())); 914 FunctionType *FT = FunctionType::get(Type::getDoubleTy(getGlobalContext()), 915 Doubles, false); 916 917 Function *F = Function::Create(FT, Function::ExternalLinkage, Name, TheModule); 918 919 // If F conflicted, there was already something named 'Name'. If it has a 920 // body, don't allow redefinition or reextern. 921 if (F->getName() != Name) { 922 // Delete the one we just made and get the existing one. 923 F->eraseFromParent(); 924 F = TheModule->getFunction(Name); 925 926 // If F already has a body, reject this. 927 if (!F->empty()) { 928 ErrorF("redefinition of function"); 929 return 0; 930 } 931 932 // If F took a different number of args, reject. 933 if (F->arg_size() != Args.size()) { 934 ErrorF("redefinition of function with different # args"); 935 return 0; 936 } 937 } 938 939 // Set names for all arguments. 940 unsigned Idx = 0; 941 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size(); 942 ++AI, ++Idx) 943 AI->setName(Args[Idx]); 944 945 return F; 946} 947 948/// CreateArgumentAllocas - Create an alloca for each argument and register the 949/// argument in the symbol table so that references to it will succeed. 950void PrototypeAST::CreateArgumentAllocas(Function *F) { 951 Function::arg_iterator AI = F->arg_begin(); 952 for (unsigned Idx = 0, e = Args.size(); Idx != e; ++Idx, ++AI) { 953 // Create an alloca for this variable. 954 AllocaInst *Alloca = CreateEntryBlockAlloca(F, Args[Idx]); 955 956 // Store the initial value into the alloca. 957 Builder.CreateStore(AI, Alloca); 958 959 // Add arguments to variable symbol table. 960 NamedValues[Args[Idx]] = Alloca; 961 } 962} 963 964Function *FunctionAST::Codegen() { 965 NamedValues.clear(); 966 967 Function *TheFunction = Proto->Codegen(); 968 if (TheFunction == 0) 969 return 0; 970 971 // If this is an operator, install it. 972 if (Proto->isBinaryOp()) 973 BinopPrecedence[Proto->getOperatorName()] = Proto->getBinaryPrecedence(); 974 975 // Create a new basic block to start insertion into. 976 BasicBlock *BB = BasicBlock::Create(getGlobalContext(), "entry", TheFunction); 977 Builder.SetInsertPoint(BB); 978 979 // Add all arguments to the symbol table and create their allocas. 980 Proto->CreateArgumentAllocas(TheFunction); 981 982 if (Value *RetVal = Body->Codegen()) { 983 // Finish off the function. 984 Builder.CreateRet(RetVal); 985 986 // Validate the generated code, checking for consistency. 987 verifyFunction(*TheFunction); 988 989 // Optimize the function. 990 TheFPM->run(*TheFunction); 991 992 return TheFunction; 993 } 994 995 // Error reading body, remove function. 996 TheFunction->eraseFromParent(); 997 998 if (Proto->isBinaryOp()) 999 BinopPrecedence.erase(Proto->getOperatorName()); 1000 return 0; 1001} 1002 1003//===----------------------------------------------------------------------===// 1004// Top-Level parsing and JIT Driver 1005//===----------------------------------------------------------------------===// 1006 1007static ExecutionEngine *TheExecutionEngine; 1008 1009static void HandleDefinition() { 1010 if (FunctionAST *F = ParseDefinition()) { 1011 if (Function *LF = F->Codegen()) { 1012 fprintf(stderr, "Read function definition:"); 1013 LF->dump(); 1014 } 1015 } else { 1016 // Skip token for error recovery. 1017 getNextToken(); 1018 } 1019} 1020 1021static void HandleExtern() { 1022 if (PrototypeAST *P = ParseExtern()) { 1023 if (Function *F = P->Codegen()) { 1024 fprintf(stderr, "Read extern: "); 1025 F->dump(); 1026 } 1027 } else { 1028 // Skip token for error recovery. 1029 getNextToken(); 1030 } 1031} 1032 1033static void HandleTopLevelExpression() { 1034 // Evaluate a top-level expression into an anonymous function. 1035 if (FunctionAST *F = ParseTopLevelExpr()) { 1036 if (Function *LF = F->Codegen()) { 1037 // JIT the function, returning a function pointer. 1038 void *FPtr = TheExecutionEngine->getPointerToFunction(LF); 1039 1040 // Cast it to the right type (takes no arguments, returns a double) so we 1041 // can call it as a native function. 1042 double (*FP)() = (double (*)())(intptr_t)FPtr; 1043 fprintf(stderr, "Evaluated to %f\n", FP()); 1044 } 1045 } else { 1046 // Skip token for error recovery. 1047 getNextToken(); 1048 } 1049} 1050 1051/// top ::= definition | external | expression | ';' 1052static void MainLoop() { 1053 while (1) { 1054 fprintf(stderr, "ready> "); 1055 switch (CurTok) { 1056 case tok_eof: return; 1057 case ';': getNextToken(); break; // ignore top-level semicolons. 1058 case tok_def: HandleDefinition(); break; 1059 case tok_extern: HandleExtern(); break; 1060 default: HandleTopLevelExpression(); break; 1061 } 1062 } 1063} 1064 1065//===----------------------------------------------------------------------===// 1066// "Library" functions that can be "extern'd" from user code. 1067//===----------------------------------------------------------------------===// 1068 1069/// putchard - putchar that takes a double and returns 0. 1070extern "C" 1071double putchard(double X) { 1072 putchar((char)X); 1073 return 0; 1074} 1075 1076/// printd - printf that takes a double prints it as "%f\n", returning 0. 1077extern "C" 1078double printd(double X) { 1079 printf("%f\n", X); 1080 return 0; 1081} 1082 1083//===----------------------------------------------------------------------===// 1084// Main driver code. 1085//===----------------------------------------------------------------------===// 1086 1087int main() { 1088 InitializeNativeTarget(); 1089 LLVMContext &Context = getGlobalContext(); 1090 1091 // Install standard binary operators. 1092 // 1 is lowest precedence. 1093 BinopPrecedence['='] = 2; 1094 BinopPrecedence['<'] = 10; 1095 BinopPrecedence['+'] = 20; 1096 BinopPrecedence['-'] = 20; 1097 BinopPrecedence['*'] = 40; // highest. 1098 1099 // Prime the first token. 1100 fprintf(stderr, "ready> "); 1101 getNextToken(); 1102 1103 // Make the module, which holds all the code. 1104 TheModule = new Module("my cool jit", Context); 1105 1106 // Create the JIT. This takes ownership of the module. 1107 std::string ErrStr; 1108 TheExecutionEngine = EngineBuilder(TheModule).setErrorStr(&ErrStr).create(); 1109 if (!TheExecutionEngine) { 1110 fprintf(stderr, "Could not create ExecutionEngine: %s\n", ErrStr.c_str()); 1111 exit(1); 1112 } 1113 1114 FunctionPassManager OurFPM(TheModule); 1115 1116 // Set up the optimizer pipeline. Start with registering info about how the 1117 // target lays out data structures. 1118 OurFPM.add(new DataLayout(*TheExecutionEngine->getDataLayout())); 1119 // Provide basic AliasAnalysis support for GVN. 1120 OurFPM.add(createBasicAliasAnalysisPass()); 1121 // Promote allocas to registers. 1122 OurFPM.add(createPromoteMemoryToRegisterPass()); 1123 // Do simple "peephole" optimizations and bit-twiddling optzns. 1124 OurFPM.add(createInstructionCombiningPass()); 1125 // Reassociate expressions. 1126 OurFPM.add(createReassociatePass()); 1127 // Eliminate Common SubExpressions. 1128 OurFPM.add(createGVNPass()); 1129 // Simplify the control flow graph (deleting unreachable blocks, etc). 1130 OurFPM.add(createCFGSimplificationPass()); 1131 1132 OurFPM.doInitialization(); 1133 1134 // Set the global so the code gen can use this. 1135 TheFPM = &OurFPM; 1136 1137 // Run the main "interpreter loop" now. 1138 MainLoop(); 1139 1140 TheFPM = 0; 1141 1142 // Print out all of the generated code. 1143 TheModule->dump(); 1144 1145 return 0; 1146} 1147