toy.cpp revision 0a08460599eed603e469e3e16d0cf6aa33b8ba93
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 <cstdio>
14#include <map>
15#include <string>
16#include <vector>
17using namespace llvm;
18
19//===----------------------------------------------------------------------===//
20// Lexer
21//===----------------------------------------------------------------------===//
22
23// The lexer returns tokens [0-255] if it is an unknown character, otherwise one
24// of these for known things.
25enum Token {
26  tok_eof = -1,
27
28  // commands
29  tok_def = -2, tok_extern = -3,
30
31  // primary
32  tok_identifier = -4, tok_number = -5,
33
34  // control
35  tok_if = -6, tok_then = -7, tok_else = -8,
36  tok_for = -9, tok_in = -10,
37
38  // operators
39  tok_binary = -11, tok_unary = -12,
40
41  // var definition
42  tok_var = -13
43};
44
45static std::string IdentifierStr;  // Filled in if tok_identifier
46static double NumVal;              // Filled in if tok_number
47
48/// gettok - Return the next token from standard input.
49static int gettok() {
50  static int LastChar = ' ';
51
52  // Skip any whitespace.
53  while (isspace(LastChar))
54    LastChar = getchar();
55
56  if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
57    IdentifierStr = LastChar;
58    while (isalnum((LastChar = getchar())))
59      IdentifierStr += LastChar;
60
61    if (IdentifierStr == "def") return tok_def;
62    if (IdentifierStr == "extern") return tok_extern;
63    if (IdentifierStr == "if") return tok_if;
64    if (IdentifierStr == "then") return tok_then;
65    if (IdentifierStr == "else") return tok_else;
66    if (IdentifierStr == "for") return tok_for;
67    if (IdentifierStr == "in") return tok_in;
68    if (IdentifierStr == "binary") return tok_binary;
69    if (IdentifierStr == "unary") return tok_unary;
70    if (IdentifierStr == "var") return tok_var;
71    return tok_identifier;
72  }
73
74  if (isdigit(LastChar) || LastChar == '.') {   // Number: [0-9.]+
75    std::string NumStr;
76    do {
77      NumStr += LastChar;
78      LastChar = getchar();
79    } while (isdigit(LastChar) || LastChar == '.');
80
81    NumVal = strtod(NumStr.c_str(), 0);
82    return tok_number;
83  }
84
85  if (LastChar == '#') {
86    // Comment until end of line.
87    do LastChar = getchar();
88    while (LastChar != EOF && LastChar != '\n' && LastChar != '\r');
89
90    if (LastChar != EOF)
91      return gettok();
92  }
93
94  // Check for end of file.  Don't eat the EOF.
95  if (LastChar == EOF)
96    return tok_eof;
97
98  // Otherwise, just return the character as its ascii value.
99  int ThisChar = LastChar;
100  LastChar = getchar();
101  return ThisChar;
102}
103
104//===----------------------------------------------------------------------===//
105// Abstract Syntax Tree (aka Parse Tree)
106//===----------------------------------------------------------------------===//
107
108/// ExprAST - Base class for all expression nodes.
109class ExprAST {
110public:
111  virtual ~ExprAST() {}
112  virtual Value *Codegen() = 0;
113};
114
115/// NumberExprAST - Expression class for numeric literals like "1.0".
116class NumberExprAST : public ExprAST {
117  double Val;
118public:
119  NumberExprAST(double val) : Val(val) {}
120  virtual Value *Codegen();
121};
122
123/// VariableExprAST - Expression class for referencing a variable, like "a".
124class VariableExprAST : public ExprAST {
125  std::string Name;
126public:
127  VariableExprAST(const std::string &name) : Name(name) {}
128  const std::string &getName() const { return Name; }
129  virtual Value *Codegen();
130};
131
132/// UnaryExprAST - Expression class for a unary operator.
133class UnaryExprAST : public ExprAST {
134  char Opcode;
135  ExprAST *Operand;
136public:
137  UnaryExprAST(char opcode, ExprAST *operand)
138    : Opcode(opcode), Operand(operand) {}
139  virtual Value *Codegen();
140};
141
142/// BinaryExprAST - Expression class for a binary operator.
143class BinaryExprAST : public ExprAST {
144  char Op;
145  ExprAST *LHS, *RHS;
146public:
147  BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs)
148    : Op(op), LHS(lhs), RHS(rhs) {}
149  virtual Value *Codegen();
150};
151
152/// CallExprAST - Expression class for function calls.
153class CallExprAST : public ExprAST {
154  std::string Callee;
155  std::vector<ExprAST*> Args;
156public:
157  CallExprAST(const std::string &callee, std::vector<ExprAST*> &args)
158    : Callee(callee), Args(args) {}
159  virtual Value *Codegen();
160};
161
162/// IfExprAST - Expression class for if/then/else.
163class IfExprAST : public ExprAST {
164  ExprAST *Cond, *Then, *Else;
165public:
166  IfExprAST(ExprAST *cond, ExprAST *then, ExprAST *_else)
167  : Cond(cond), Then(then), Else(_else) {}
168  virtual Value *Codegen();
169};
170
171/// ForExprAST - Expression class for for/in.
172class ForExprAST : public ExprAST {
173  std::string VarName;
174  ExprAST *Start, *End, *Step, *Body;
175public:
176  ForExprAST(const std::string &varname, ExprAST *start, ExprAST *end,
177             ExprAST *step, ExprAST *body)
178    : VarName(varname), Start(start), End(end), Step(step), Body(body) {}
179  virtual Value *Codegen();
180};
181
182/// VarExprAST - Expression class for var/in
183class VarExprAST : public ExprAST {
184  std::vector<std::pair<std::string, ExprAST*> > VarNames;
185  ExprAST *Body;
186public:
187  VarExprAST(const std::vector<std::pair<std::string, ExprAST*> > &varnames,
188             ExprAST *body)
189  : VarNames(varnames), Body(body) {}
190
191  virtual Value *Codegen();
192};
193
194/// PrototypeAST - This class represents the "prototype" for a function,
195/// which captures its argument names as well as if it is an operator.
196class PrototypeAST {
197  std::string Name;
198  std::vector<std::string> Args;
199  bool isOperator;
200  unsigned Precedence;  // Precedence if a binary op.
201public:
202  PrototypeAST(const std::string &name, const std::vector<std::string> &args,
203               bool isoperator = false, unsigned prec = 0)
204  : Name(name), Args(args), isOperator(isoperator), Precedence(prec) {}
205
206  bool isUnaryOp() const { return isOperator && Args.size() == 1; }
207  bool isBinaryOp() const { return isOperator && Args.size() == 2; }
208
209  char getOperatorName() const {
210    assert(isUnaryOp() || isBinaryOp());
211    return Name[Name.size()-1];
212  }
213
214  unsigned getBinaryPrecedence() const { return Precedence; }
215
216  Function *Codegen();
217
218  void CreateArgumentAllocas(Function *F);
219};
220
221/// FunctionAST - This class represents a function definition itself.
222class FunctionAST {
223  PrototypeAST *Proto;
224  ExprAST *Body;
225public:
226  FunctionAST(PrototypeAST *proto, ExprAST *body)
227    : Proto(proto), Body(body) {}
228
229  Function *Codegen();
230};
231
232//===----------------------------------------------------------------------===//
233// Parser
234//===----------------------------------------------------------------------===//
235
236/// CurTok/getNextToken - Provide a simple token buffer.  CurTok is the current
237/// token the parser is looking at.  getNextToken reads another token from the
238/// lexer and updates CurTok with its results.
239static int CurTok;
240static int getNextToken() {
241  return CurTok = gettok();
242}
243
244/// BinopPrecedence - This holds the precedence for each binary operator that is
245/// defined.
246static std::map<char, int> BinopPrecedence;
247
248/// GetTokPrecedence - Get the precedence of the pending binary operator token.
249static int GetTokPrecedence() {
250  if (!isascii(CurTok))
251    return -1;
252
253  // Make sure it's a declared binop.
254  int TokPrec = BinopPrecedence[CurTok];
255  if (TokPrec <= 0) return -1;
256  return TokPrec;
257}
258
259/// Error* - These are little helper functions for error handling.
260ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;}
261PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; }
262FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; }
263
264static ExprAST *ParseExpression();
265
266/// identifierexpr
267///   ::= identifier
268///   ::= identifier '(' expression* ')'
269static ExprAST *ParseIdentifierExpr() {
270  std::string IdName = IdentifierStr;
271
272  getNextToken();  // eat identifier.
273
274  if (CurTok != '(') // Simple variable ref.
275    return new VariableExprAST(IdName);
276
277  // Call.
278  getNextToken();  // eat (
279  std::vector<ExprAST*> Args;
280  if (CurTok != ')') {
281    while (1) {
282      ExprAST *Arg = ParseExpression();
283      if (!Arg) return 0;
284      Args.push_back(Arg);
285
286      if (CurTok == ')') break;
287
288      if (CurTok != ',')
289        return Error("Expected ')' or ',' in argument list");
290      getNextToken();
291    }
292  }
293
294  // Eat the ')'.
295  getNextToken();
296
297  return new CallExprAST(IdName, Args);
298}
299
300/// numberexpr ::= number
301static ExprAST *ParseNumberExpr() {
302  ExprAST *Result = new NumberExprAST(NumVal);
303  getNextToken(); // consume the number
304  return Result;
305}
306
307/// parenexpr ::= '(' expression ')'
308static ExprAST *ParseParenExpr() {
309  getNextToken();  // eat (.
310  ExprAST *V = ParseExpression();
311  if (!V) return 0;
312
313  if (CurTok != ')')
314    return Error("expected ')'");
315  getNextToken();  // eat ).
316  return V;
317}
318
319/// ifexpr ::= 'if' expression 'then' expression 'else' expression
320static ExprAST *ParseIfExpr() {
321  getNextToken();  // eat the if.
322
323  // condition.
324  ExprAST *Cond = ParseExpression();
325  if (!Cond) return 0;
326
327  if (CurTok != tok_then)
328    return Error("expected then");
329  getNextToken();  // eat the then
330
331  ExprAST *Then = ParseExpression();
332  if (Then == 0) return 0;
333
334  if (CurTok != tok_else)
335    return Error("expected else");
336
337  getNextToken();
338
339  ExprAST *Else = ParseExpression();
340  if (!Else) return 0;
341
342  return new IfExprAST(Cond, Then, Else);
343}
344
345/// forexpr ::= 'for' identifier '=' expr ',' expr (',' expr)? 'in' expression
346static ExprAST *ParseForExpr() {
347  getNextToken();  // eat the for.
348
349  if (CurTok != tok_identifier)
350    return Error("expected identifier after for");
351
352  std::string IdName = IdentifierStr;
353  getNextToken();  // eat identifier.
354
355  if (CurTok != '=')
356    return Error("expected '=' after for");
357  getNextToken();  // eat '='.
358
359
360  ExprAST *Start = ParseExpression();
361  if (Start == 0) return 0;
362  if (CurTok != ',')
363    return Error("expected ',' after for start value");
364  getNextToken();
365
366  ExprAST *End = ParseExpression();
367  if (End == 0) return 0;
368
369  // The step value is optional.
370  ExprAST *Step = 0;
371  if (CurTok == ',') {
372    getNextToken();
373    Step = ParseExpression();
374    if (Step == 0) return 0;
375  }
376
377  if (CurTok != tok_in)
378    return Error("expected 'in' after for");
379  getNextToken();  // eat 'in'.
380
381  ExprAST *Body = ParseExpression();
382  if (Body == 0) return 0;
383
384  return new ForExprAST(IdName, Start, End, Step, Body);
385}
386
387/// varexpr ::= 'var' identifier ('=' expression)?
388//                    (',' identifier ('=' expression)?)* 'in' expression
389static ExprAST *ParseVarExpr() {
390  getNextToken();  // eat the var.
391
392  std::vector<std::pair<std::string, ExprAST*> > VarNames;
393
394  // At least one variable name is required.
395  if (CurTok != tok_identifier)
396    return Error("expected identifier after var");
397
398  while (1) {
399    std::string Name = IdentifierStr;
400    getNextToken();  // eat identifier.
401
402    // Read the optional initializer.
403    ExprAST *Init = 0;
404    if (CurTok == '=') {
405      getNextToken(); // eat the '='.
406
407      Init = ParseExpression();
408      if (Init == 0) return 0;
409    }
410
411    VarNames.push_back(std::make_pair(Name, Init));
412
413    // End of var list, exit loop.
414    if (CurTok != ',') break;
415    getNextToken(); // eat the ','.
416
417    if (CurTok != tok_identifier)
418      return Error("expected identifier list after var");
419  }
420
421  // At this point, we have to have 'in'.
422  if (CurTok != tok_in)
423    return Error("expected 'in' keyword after 'var'");
424  getNextToken();  // eat 'in'.
425
426  ExprAST *Body = ParseExpression();
427  if (Body == 0) return 0;
428
429  return new VarExprAST(VarNames, Body);
430}
431
432/// primary
433///   ::= identifierexpr
434///   ::= numberexpr
435///   ::= parenexpr
436///   ::= ifexpr
437///   ::= forexpr
438///   ::= varexpr
439static ExprAST *ParsePrimary() {
440  switch (CurTok) {
441  default: return Error("unknown token when expecting an expression");
442  case tok_identifier: return ParseIdentifierExpr();
443  case tok_number:     return ParseNumberExpr();
444  case '(':            return ParseParenExpr();
445  case tok_if:         return ParseIfExpr();
446  case tok_for:        return ParseForExpr();
447  case tok_var:        return ParseVarExpr();
448  }
449}
450
451/// unary
452///   ::= primary
453///   ::= '!' unary
454static ExprAST *ParseUnary() {
455  // If the current token is not an operator, it must be a primary expr.
456  if (!isascii(CurTok) || CurTok == '(' || CurTok == ',')
457    return ParsePrimary();
458
459  // If this is a unary operator, read it.
460  int Opc = CurTok;
461  getNextToken();
462  if (ExprAST *Operand = ParseUnary())
463    return new UnaryExprAST(Opc, Operand);
464  return 0;
465}
466
467/// binoprhs
468///   ::= ('+' unary)*
469static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) {
470  // If this is a binop, find its precedence.
471  while (1) {
472    int TokPrec = GetTokPrecedence();
473
474    // If this is a binop that binds at least as tightly as the current binop,
475    // consume it, otherwise we are done.
476    if (TokPrec < ExprPrec)
477      return LHS;
478
479    // Okay, we know this is a binop.
480    int BinOp = CurTok;
481    getNextToken();  // eat binop
482
483    // Parse the unary expression after the binary operator.
484    ExprAST *RHS = ParseUnary();
485    if (!RHS) return 0;
486
487    // If BinOp binds less tightly with RHS than the operator after RHS, let
488    // the pending operator take RHS as its LHS.
489    int NextPrec = GetTokPrecedence();
490    if (TokPrec < NextPrec) {
491      RHS = ParseBinOpRHS(TokPrec+1, RHS);
492      if (RHS == 0) return 0;
493    }
494
495    // Merge LHS/RHS.
496    LHS = new BinaryExprAST(BinOp, LHS, RHS);
497  }
498}
499
500/// expression
501///   ::= unary binoprhs
502///
503static ExprAST *ParseExpression() {
504  ExprAST *LHS = ParseUnary();
505  if (!LHS) return 0;
506
507  return ParseBinOpRHS(0, LHS);
508}
509
510/// prototype
511///   ::= id '(' id* ')'
512///   ::= binary LETTER number? (id, id)
513///   ::= unary LETTER (id)
514static PrototypeAST *ParsePrototype() {
515  std::string FnName;
516
517  unsigned Kind = 0; // 0 = identifier, 1 = unary, 2 = binary.
518  unsigned BinaryPrecedence = 30;
519
520  switch (CurTok) {
521  default:
522    return ErrorP("Expected function name in prototype");
523  case tok_identifier:
524    FnName = IdentifierStr;
525    Kind = 0;
526    getNextToken();
527    break;
528  case tok_unary:
529    getNextToken();
530    if (!isascii(CurTok))
531      return ErrorP("Expected unary operator");
532    FnName = "unary";
533    FnName += (char)CurTok;
534    Kind = 1;
535    getNextToken();
536    break;
537  case tok_binary:
538    getNextToken();
539    if (!isascii(CurTok))
540      return ErrorP("Expected binary operator");
541    FnName = "binary";
542    FnName += (char)CurTok;
543    Kind = 2;
544    getNextToken();
545
546    // Read the precedence if present.
547    if (CurTok == tok_number) {
548      if (NumVal < 1 || NumVal > 100)
549        return ErrorP("Invalid precedecnce: must be 1..100");
550      BinaryPrecedence = (unsigned)NumVal;
551      getNextToken();
552    }
553    break;
554  }
555
556  if (CurTok != '(')
557    return ErrorP("Expected '(' in prototype");
558
559  std::vector<std::string> ArgNames;
560  while (getNextToken() == tok_identifier)
561    ArgNames.push_back(IdentifierStr);
562  if (CurTok != ')')
563    return ErrorP("Expected ')' in prototype");
564
565  // success.
566  getNextToken();  // eat ')'.
567
568  // Verify right number of names for operator.
569  if (Kind && ArgNames.size() != Kind)
570    return ErrorP("Invalid number of operands for operator");
571
572  return new PrototypeAST(FnName, ArgNames, Kind != 0, BinaryPrecedence);
573}
574
575/// definition ::= 'def' prototype expression
576static FunctionAST *ParseDefinition() {
577  getNextToken();  // eat def.
578  PrototypeAST *Proto = ParsePrototype();
579  if (Proto == 0) return 0;
580
581  if (ExprAST *E = ParseExpression())
582    return new FunctionAST(Proto, E);
583  return 0;
584}
585
586/// toplevelexpr ::= expression
587static FunctionAST *ParseTopLevelExpr() {
588  if (ExprAST *E = ParseExpression()) {
589    // Make an anonymous proto.
590    PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>());
591    return new FunctionAST(Proto, E);
592  }
593  return 0;
594}
595
596/// external ::= 'extern' prototype
597static PrototypeAST *ParseExtern() {
598  getNextToken();  // eat extern.
599  return ParsePrototype();
600}
601
602//===----------------------------------------------------------------------===//
603// Code Generation
604//===----------------------------------------------------------------------===//
605
606static Module *TheModule;
607static IRBuilder<> Builder(getGlobalContext());
608static std::map<std::string, AllocaInst*> NamedValues;
609static FunctionPassManager *TheFPM;
610
611Value *ErrorV(const char *Str) { Error(Str); return 0; }
612
613/// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of
614/// the function.  This is used for mutable variables etc.
615static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction,
616                                          const std::string &VarName) {
617  IRBuilder<> TmpB(&TheFunction->getEntryBlock(),
618                 TheFunction->getEntryBlock().begin());
619  return TmpB.CreateAlloca(Type::getDoubleTy(getGlobalContext()), 0,
620                           VarName.c_str());
621}
622
623Value *NumberExprAST::Codegen() {
624  return ConstantFP::get(getGlobalContext(), APFloat(Val));
625}
626
627Value *VariableExprAST::Codegen() {
628  // Look this variable up in the function.
629  Value *V = NamedValues[Name];
630  if (V == 0) return ErrorV("Unknown variable name");
631
632  // Load the value.
633  return Builder.CreateLoad(V, Name.c_str());
634}
635
636Value *UnaryExprAST::Codegen() {
637  Value *OperandV = Operand->Codegen();
638  if (OperandV == 0) return 0;
639
640  Function *F = TheModule->getFunction(std::string("unary")+Opcode);
641  if (F == 0)
642    return ErrorV("Unknown unary operator");
643
644  return Builder.CreateCall(F, OperandV, "unop");
645}
646
647Value *BinaryExprAST::Codegen() {
648  // Special case '=' because we don't want to emit the LHS as an expression.
649  if (Op == '=') {
650    // Assignment requires the LHS to be an identifier.
651    VariableExprAST *LHSE = dynamic_cast<VariableExprAST*>(LHS);
652    if (!LHSE)
653      return ErrorV("destination of '=' must be a variable");
654    // Codegen the RHS.
655    Value *Val = RHS->Codegen();
656    if (Val == 0) return 0;
657
658    // Look up the name.
659    Value *Variable = NamedValues[LHSE->getName()];
660    if (Variable == 0) return ErrorV("Unknown variable name");
661
662    Builder.CreateStore(Val, Variable);
663    return Val;
664  }
665
666  Value *L = LHS->Codegen();
667  Value *R = RHS->Codegen();
668  if (L == 0 || R == 0) return 0;
669
670  switch (Op) {
671  case '+': return Builder.CreateFAdd(L, R, "addtmp");
672  case '-': return Builder.CreateFSub(L, R, "subtmp");
673  case '*': return Builder.CreateFMul(L, R, "multmp");
674  case '<':
675    L = Builder.CreateFCmpULT(L, R, "cmptmp");
676    // Convert bool 0/1 to double 0.0 or 1.0
677    return Builder.CreateUIToFP(L, Type::getDoubleTy(getGlobalContext()),
678                                "booltmp");
679  default: break;
680  }
681
682  // If it wasn't a builtin binary operator, it must be a user defined one. Emit
683  // a call to it.
684  Function *F = TheModule->getFunction(std::string("binary")+Op);
685  assert(F && "binary operator not found!");
686
687  Value *Ops[] = { L, R };
688  return Builder.CreateCall(F, Ops, "binop");
689}
690
691Value *CallExprAST::Codegen() {
692  // Look up the name in the global module table.
693  Function *CalleeF = TheModule->getFunction(Callee);
694  if (CalleeF == 0)
695    return ErrorV("Unknown function referenced");
696
697  // If argument mismatch error.
698  if (CalleeF->arg_size() != Args.size())
699    return ErrorV("Incorrect # arguments passed");
700
701  std::vector<Value*> ArgsV;
702  for (unsigned i = 0, e = Args.size(); i != e; ++i) {
703    ArgsV.push_back(Args[i]->Codegen());
704    if (ArgsV.back() == 0) return 0;
705  }
706
707  return Builder.CreateCall(CalleeF, ArgsV, "calltmp");
708}
709
710Value *IfExprAST::Codegen() {
711  Value *CondV = Cond->Codegen();
712  if (CondV == 0) return 0;
713
714  // Convert condition to a bool by comparing equal to 0.0.
715  CondV = Builder.CreateFCmpONE(CondV,
716                              ConstantFP::get(getGlobalContext(), APFloat(0.0)),
717                                "ifcond");
718
719  Function *TheFunction = Builder.GetInsertBlock()->getParent();
720
721  // Create blocks for the then and else cases.  Insert the 'then' block at the
722  // end of the function.
723  BasicBlock *ThenBB = BasicBlock::Create(getGlobalContext(), "then", TheFunction);
724  BasicBlock *ElseBB = BasicBlock::Create(getGlobalContext(), "else");
725  BasicBlock *MergeBB = BasicBlock::Create(getGlobalContext(), "ifcont");
726
727  Builder.CreateCondBr(CondV, ThenBB, ElseBB);
728
729  // Emit then value.
730  Builder.SetInsertPoint(ThenBB);
731
732  Value *ThenV = Then->Codegen();
733  if (ThenV == 0) return 0;
734
735  Builder.CreateBr(MergeBB);
736  // Codegen of 'Then' can change the current block, update ThenBB for the PHI.
737  ThenBB = Builder.GetInsertBlock();
738
739  // Emit else block.
740  TheFunction->getBasicBlockList().push_back(ElseBB);
741  Builder.SetInsertPoint(ElseBB);
742
743  Value *ElseV = Else->Codegen();
744  if (ElseV == 0) return 0;
745
746  Builder.CreateBr(MergeBB);
747  // Codegen of 'Else' can change the current block, update ElseBB for the PHI.
748  ElseBB = Builder.GetInsertBlock();
749
750  // Emit merge block.
751  TheFunction->getBasicBlockList().push_back(MergeBB);
752  Builder.SetInsertPoint(MergeBB);
753  PHINode *PN = Builder.CreatePHI(Type::getDoubleTy(getGlobalContext()), 2,
754                                  "iftmp");
755
756  PN->addIncoming(ThenV, ThenBB);
757  PN->addIncoming(ElseV, ElseBB);
758  return PN;
759}
760
761Value *ForExprAST::Codegen() {
762  // Output this as:
763  //   var = alloca double
764  //   ...
765  //   start = startexpr
766  //   store start -> var
767  //   goto loop
768  // loop:
769  //   ...
770  //   bodyexpr
771  //   ...
772  // loopend:
773  //   step = stepexpr
774  //   endcond = endexpr
775  //
776  //   curvar = load var
777  //   nextvar = curvar + step
778  //   store nextvar -> var
779  //   br endcond, loop, endloop
780  // outloop:
781
782  Function *TheFunction = Builder.GetInsertBlock()->getParent();
783
784  // Create an alloca for the variable in the entry block.
785  AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
786
787  // Emit the start code first, without 'variable' in scope.
788  Value *StartVal = Start->Codegen();
789  if (StartVal == 0) return 0;
790
791  // Store the value into the alloca.
792  Builder.CreateStore(StartVal, Alloca);
793
794  // Make the new basic block for the loop header, inserting after current
795  // block.
796  BasicBlock *LoopBB = BasicBlock::Create(getGlobalContext(), "loop", TheFunction);
797
798  // Insert an explicit fall through from the current block to the LoopBB.
799  Builder.CreateBr(LoopBB);
800
801  // Start insertion in LoopBB.
802  Builder.SetInsertPoint(LoopBB);
803
804  // Within the loop, the variable is defined equal to the PHI node.  If it
805  // shadows an existing variable, we have to restore it, so save it now.
806  AllocaInst *OldVal = NamedValues[VarName];
807  NamedValues[VarName] = Alloca;
808
809  // Emit the body of the loop.  This, like any other expr, can change the
810  // current BB.  Note that we ignore the value computed by the body, but don't
811  // allow an error.
812  if (Body->Codegen() == 0)
813    return 0;
814
815  // Emit the step value.
816  Value *StepVal;
817  if (Step) {
818    StepVal = Step->Codegen();
819    if (StepVal == 0) return 0;
820  } else {
821    // If not specified, use 1.0.
822    StepVal = ConstantFP::get(getGlobalContext(), APFloat(1.0));
823  }
824
825  // Compute the end condition.
826  Value *EndCond = End->Codegen();
827  if (EndCond == 0) return EndCond;
828
829  // Reload, increment, and restore the alloca.  This handles the case where
830  // the body of the loop mutates the variable.
831  Value *CurVar = Builder.CreateLoad(Alloca, VarName.c_str());
832  Value *NextVar = Builder.CreateFAdd(CurVar, StepVal, "nextvar");
833  Builder.CreateStore(NextVar, Alloca);
834
835  // Convert condition to a bool by comparing equal to 0.0.
836  EndCond = Builder.CreateFCmpONE(EndCond,
837                              ConstantFP::get(getGlobalContext(), APFloat(0.0)),
838                                  "loopcond");
839
840  // Create the "after loop" block and insert it.
841  BasicBlock *AfterBB = BasicBlock::Create(getGlobalContext(), "afterloop", TheFunction);
842
843  // Insert the conditional branch into the end of LoopEndBB.
844  Builder.CreateCondBr(EndCond, LoopBB, AfterBB);
845
846  // Any new code will be inserted in AfterBB.
847  Builder.SetInsertPoint(AfterBB);
848
849  // Restore the unshadowed variable.
850  if (OldVal)
851    NamedValues[VarName] = OldVal;
852  else
853    NamedValues.erase(VarName);
854
855
856  // for expr always returns 0.0.
857  return Constant::getNullValue(Type::getDoubleTy(getGlobalContext()));
858}
859
860Value *VarExprAST::Codegen() {
861  std::vector<AllocaInst *> OldBindings;
862
863  Function *TheFunction = Builder.GetInsertBlock()->getParent();
864
865  // Register all variables and emit their initializer.
866  for (unsigned i = 0, e = VarNames.size(); i != e; ++i) {
867    const std::string &VarName = VarNames[i].first;
868    ExprAST *Init = VarNames[i].second;
869
870    // Emit the initializer before adding the variable to scope, this prevents
871    // the initializer from referencing the variable itself, and permits stuff
872    // like this:
873    //  var a = 1 in
874    //    var a = a in ...   # refers to outer 'a'.
875    Value *InitVal;
876    if (Init) {
877      InitVal = Init->Codegen();
878      if (InitVal == 0) return 0;
879    } else { // If not specified, use 0.0.
880      InitVal = ConstantFP::get(getGlobalContext(), APFloat(0.0));
881    }
882
883    AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
884    Builder.CreateStore(InitVal, Alloca);
885
886    // Remember the old variable binding so that we can restore the binding when
887    // we unrecurse.
888    OldBindings.push_back(NamedValues[VarName]);
889
890    // Remember this binding.
891    NamedValues[VarName] = Alloca;
892  }
893
894  // Codegen the body, now that all vars are in scope.
895  Value *BodyVal = Body->Codegen();
896  if (BodyVal == 0) return 0;
897
898  // Pop all our variables from scope.
899  for (unsigned i = 0, e = VarNames.size(); i != e; ++i)
900    NamedValues[VarNames[i].first] = OldBindings[i];
901
902  // Return the body computation.
903  return BodyVal;
904}
905
906Function *PrototypeAST::Codegen() {
907  // Make the function type:  double(double,double) etc.
908  std::vector<Type*> Doubles(Args.size(),
909                             Type::getDoubleTy(getGlobalContext()));
910  FunctionType *FT = FunctionType::get(Type::getDoubleTy(getGlobalContext()),
911                                       Doubles, false);
912
913  Function *F = Function::Create(FT, Function::ExternalLinkage, Name, TheModule);
914
915  // If F conflicted, there was already something named 'Name'.  If it has a
916  // body, don't allow redefinition or reextern.
917  if (F->getName() != Name) {
918    // Delete the one we just made and get the existing one.
919    F->eraseFromParent();
920    F = TheModule->getFunction(Name);
921
922    // If F already has a body, reject this.
923    if (!F->empty()) {
924      ErrorF("redefinition of function");
925      return 0;
926    }
927
928    // If F took a different number of args, reject.
929    if (F->arg_size() != Args.size()) {
930      ErrorF("redefinition of function with different # args");
931      return 0;
932    }
933  }
934
935  // Set names for all arguments.
936  unsigned Idx = 0;
937  for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
938       ++AI, ++Idx)
939    AI->setName(Args[Idx]);
940
941  return F;
942}
943
944/// CreateArgumentAllocas - Create an alloca for each argument and register the
945/// argument in the symbol table so that references to it will succeed.
946void PrototypeAST::CreateArgumentAllocas(Function *F) {
947  Function::arg_iterator AI = F->arg_begin();
948  for (unsigned Idx = 0, e = Args.size(); Idx != e; ++Idx, ++AI) {
949    // Create an alloca for this variable.
950    AllocaInst *Alloca = CreateEntryBlockAlloca(F, Args[Idx]);
951
952    // Store the initial value into the alloca.
953    Builder.CreateStore(AI, Alloca);
954
955    // Add arguments to variable symbol table.
956    NamedValues[Args[Idx]] = Alloca;
957  }
958}
959
960Function *FunctionAST::Codegen() {
961  NamedValues.clear();
962
963  Function *TheFunction = Proto->Codegen();
964  if (TheFunction == 0)
965    return 0;
966
967  // If this is an operator, install it.
968  if (Proto->isBinaryOp())
969    BinopPrecedence[Proto->getOperatorName()] = Proto->getBinaryPrecedence();
970
971  // Create a new basic block to start insertion into.
972  BasicBlock *BB = BasicBlock::Create(getGlobalContext(), "entry", TheFunction);
973  Builder.SetInsertPoint(BB);
974
975  // Add all arguments to the symbol table and create their allocas.
976  Proto->CreateArgumentAllocas(TheFunction);
977
978  if (Value *RetVal = Body->Codegen()) {
979    // Finish off the function.
980    Builder.CreateRet(RetVal);
981
982    // Validate the generated code, checking for consistency.
983    verifyFunction(*TheFunction);
984
985    // Optimize the function.
986    TheFPM->run(*TheFunction);
987
988    return TheFunction;
989  }
990
991  // Error reading body, remove function.
992  TheFunction->eraseFromParent();
993
994  if (Proto->isBinaryOp())
995    BinopPrecedence.erase(Proto->getOperatorName());
996  return 0;
997}
998
999//===----------------------------------------------------------------------===//
1000// Top-Level parsing and JIT Driver
1001//===----------------------------------------------------------------------===//
1002
1003static ExecutionEngine *TheExecutionEngine;
1004
1005static void HandleDefinition() {
1006  if (FunctionAST *F = ParseDefinition()) {
1007    if (Function *LF = F->Codegen()) {
1008      fprintf(stderr, "Read function definition:");
1009      LF->dump();
1010    }
1011  } else {
1012    // Skip token for error recovery.
1013    getNextToken();
1014  }
1015}
1016
1017static void HandleExtern() {
1018  if (PrototypeAST *P = ParseExtern()) {
1019    if (Function *F = P->Codegen()) {
1020      fprintf(stderr, "Read extern: ");
1021      F->dump();
1022    }
1023  } else {
1024    // Skip token for error recovery.
1025    getNextToken();
1026  }
1027}
1028
1029static void HandleTopLevelExpression() {
1030  // Evaluate a top-level expression into an anonymous function.
1031  if (FunctionAST *F = ParseTopLevelExpr()) {
1032    if (Function *LF = F->Codegen()) {
1033      // JIT the function, returning a function pointer.
1034      void *FPtr = TheExecutionEngine->getPointerToFunction(LF);
1035
1036      // Cast it to the right type (takes no arguments, returns a double) so we
1037      // can call it as a native function.
1038      double (*FP)() = (double (*)())(intptr_t)FPtr;
1039      fprintf(stderr, "Evaluated to %f\n", FP());
1040    }
1041  } else {
1042    // Skip token for error recovery.
1043    getNextToken();
1044  }
1045}
1046
1047/// top ::= definition | external | expression | ';'
1048static void MainLoop() {
1049  while (1) {
1050    fprintf(stderr, "ready> ");
1051    switch (CurTok) {
1052    case tok_eof:    return;
1053    case ';':        getNextToken(); break;  // ignore top-level semicolons.
1054    case tok_def:    HandleDefinition(); break;
1055    case tok_extern: HandleExtern(); break;
1056    default:         HandleTopLevelExpression(); break;
1057    }
1058  }
1059}
1060
1061//===----------------------------------------------------------------------===//
1062// "Library" functions that can be "extern'd" from user code.
1063//===----------------------------------------------------------------------===//
1064
1065/// putchard - putchar that takes a double and returns 0.
1066extern "C"
1067double putchard(double X) {
1068  putchar((char)X);
1069  return 0;
1070}
1071
1072/// printd - printf that takes a double prints it as "%f\n", returning 0.
1073extern "C"
1074double printd(double X) {
1075  printf("%f\n", X);
1076  return 0;
1077}
1078
1079//===----------------------------------------------------------------------===//
1080// Main driver code.
1081//===----------------------------------------------------------------------===//
1082
1083int main() {
1084  InitializeNativeTarget();
1085  LLVMContext &Context = getGlobalContext();
1086
1087  // Install standard binary operators.
1088  // 1 is lowest precedence.
1089  BinopPrecedence['='] = 2;
1090  BinopPrecedence['<'] = 10;
1091  BinopPrecedence['+'] = 20;
1092  BinopPrecedence['-'] = 20;
1093  BinopPrecedence['*'] = 40;  // highest.
1094
1095  // Prime the first token.
1096  fprintf(stderr, "ready> ");
1097  getNextToken();
1098
1099  // Make the module, which holds all the code.
1100  TheModule = new Module("my cool jit", Context);
1101
1102  // Create the JIT.  This takes ownership of the module.
1103  std::string ErrStr;
1104  TheExecutionEngine = EngineBuilder(TheModule).setErrorStr(&ErrStr).create();
1105  if (!TheExecutionEngine) {
1106    fprintf(stderr, "Could not create ExecutionEngine: %s\n", ErrStr.c_str());
1107    exit(1);
1108  }
1109
1110  FunctionPassManager OurFPM(TheModule);
1111
1112  // Set up the optimizer pipeline.  Start with registering info about how the
1113  // target lays out data structures.
1114  OurFPM.add(new DataLayout(*TheExecutionEngine->getDataLayout()));
1115  // Provide basic AliasAnalysis support for GVN.
1116  OurFPM.add(createBasicAliasAnalysisPass());
1117  // Promote allocas to registers.
1118  OurFPM.add(createPromoteMemoryToRegisterPass());
1119  // Do simple "peephole" optimizations and bit-twiddling optzns.
1120  OurFPM.add(createInstructionCombiningPass());
1121  // Reassociate expressions.
1122  OurFPM.add(createReassociatePass());
1123  // Eliminate Common SubExpressions.
1124  OurFPM.add(createGVNPass());
1125  // Simplify the control flow graph (deleting unreachable blocks, etc).
1126  OurFPM.add(createCFGSimplificationPass());
1127
1128  OurFPM.doInitialization();
1129
1130  // Set the global so the code gen can use this.
1131  TheFPM = &OurFPM;
1132
1133  // Run the main "interpreter loop" now.
1134  MainLoop();
1135
1136  TheFPM = 0;
1137
1138  // Print out all of the generated code.
1139  TheModule->dump();
1140
1141  return 0;
1142}
1143