1#include "llvm/Analysis/Passes.h"
2#include "llvm/ExecutionEngine/ExecutionEngine.h"
3#include "llvm/ExecutionEngine/MCJIT.h"
4#include "llvm/ExecutionEngine/SectionMemoryManager.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/LegacyPassManager.h"
10#include "llvm/IR/Module.h"
11#include "llvm/IR/Verifier.h"
12#include "llvm/Support/TargetSelect.h"
13#include "llvm/Transforms/Scalar.h"
14#include <cctype>
15#include <cstdio>
16#include <map>
17#include <string>
18#include <vector>
19using namespace llvm;
20
21//===----------------------------------------------------------------------===//
22// Lexer
23//===----------------------------------------------------------------------===//
24
25// The lexer returns tokens [0-255] if it is an unknown character, otherwise one
26// of these for known things.
27enum Token {
28  tok_eof = -1,
29
30  // commands
31  tok_def = -2, tok_extern = -3,
32
33  // primary
34  tok_identifier = -4, tok_number = -5,
35
36  // control
37  tok_if = -6, tok_then = -7, tok_else = -8,
38  tok_for = -9, tok_in = -10,
39
40  // operators
41  tok_binary = -11, tok_unary = -12,
42
43  // var definition
44  tok_var = -13
45};
46
47static std::string IdentifierStr;  // Filled in if tok_identifier
48static double NumVal;              // Filled in if tok_number
49
50/// gettok - Return the next token from standard input.
51static int gettok() {
52  static int LastChar = ' ';
53
54  // Skip any whitespace.
55  while (isspace(LastChar))
56    LastChar = getchar();
57
58  if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
59    IdentifierStr = LastChar;
60    while (isalnum((LastChar = getchar())))
61      IdentifierStr += LastChar;
62
63    if (IdentifierStr == "def") return tok_def;
64    if (IdentifierStr == "extern") return tok_extern;
65    if (IdentifierStr == "if") return tok_if;
66    if (IdentifierStr == "then") return tok_then;
67    if (IdentifierStr == "else") return tok_else;
68    if (IdentifierStr == "for") return tok_for;
69    if (IdentifierStr == "in") return tok_in;
70    if (IdentifierStr == "binary") return tok_binary;
71    if (IdentifierStr == "unary") return tok_unary;
72    if (IdentifierStr == "var") return tok_var;
73    return tok_identifier;
74  }
75
76  if (isdigit(LastChar) || LastChar == '.') {   // Number: [0-9.]+
77    std::string NumStr;
78    do {
79      NumStr += LastChar;
80      LastChar = getchar();
81    } while (isdigit(LastChar) || LastChar == '.');
82
83    NumVal = strtod(NumStr.c_str(), 0);
84    return tok_number;
85  }
86
87  if (LastChar == '#') {
88    // Comment until end of line.
89    do LastChar = getchar();
90    while (LastChar != EOF && LastChar != '\n' && LastChar != '\r');
91
92    if (LastChar != EOF)
93      return gettok();
94  }
95
96  // Check for end of file.  Don't eat the EOF.
97  if (LastChar == EOF)
98    return tok_eof;
99
100  // Otherwise, just return the character as its ascii value.
101  int ThisChar = LastChar;
102  LastChar = getchar();
103  return ThisChar;
104}
105
106//===----------------------------------------------------------------------===//
107// Abstract Syntax Tree (aka Parse Tree)
108//===----------------------------------------------------------------------===//
109
110/// ExprAST - Base class for all expression nodes.
111class ExprAST {
112public:
113  virtual ~ExprAST() {}
114  virtual Value *Codegen() = 0;
115};
116
117/// NumberExprAST - Expression class for numeric literals like "1.0".
118class NumberExprAST : public ExprAST {
119  double Val;
120public:
121  NumberExprAST(double val) : Val(val) {}
122  virtual Value *Codegen();
123};
124
125/// VariableExprAST - Expression class for referencing a variable, like "a".
126class VariableExprAST : public ExprAST {
127  std::string Name;
128public:
129  VariableExprAST(const std::string &name) : Name(name) {}
130  const std::string &getName() const { return Name; }
131  virtual Value *Codegen();
132};
133
134/// UnaryExprAST - Expression class for a unary operator.
135class UnaryExprAST : public ExprAST {
136  char Opcode;
137  ExprAST *Operand;
138public:
139  UnaryExprAST(char opcode, ExprAST *operand)
140    : Opcode(opcode), Operand(operand) {}
141  virtual Value *Codegen();
142};
143
144/// BinaryExprAST - Expression class for a binary operator.
145class BinaryExprAST : public ExprAST {
146  char Op;
147  ExprAST *LHS, *RHS;
148public:
149  BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs)
150    : Op(op), LHS(lhs), RHS(rhs) {}
151  virtual Value *Codegen();
152};
153
154/// CallExprAST - Expression class for function calls.
155class CallExprAST : public ExprAST {
156  std::string Callee;
157  std::vector<ExprAST*> Args;
158public:
159  CallExprAST(const std::string &callee, std::vector<ExprAST*> &args)
160    : Callee(callee), Args(args) {}
161  virtual Value *Codegen();
162};
163
164/// IfExprAST - Expression class for if/then/else.
165class IfExprAST : public ExprAST {
166  ExprAST *Cond, *Then, *Else;
167public:
168  IfExprAST(ExprAST *cond, ExprAST *then, ExprAST *_else)
169  : Cond(cond), Then(then), Else(_else) {}
170  virtual Value *Codegen();
171};
172
173/// ForExprAST - Expression class for for/in.
174class ForExprAST : public ExprAST {
175  std::string VarName;
176  ExprAST *Start, *End, *Step, *Body;
177public:
178  ForExprAST(const std::string &varname, ExprAST *start, ExprAST *end,
179             ExprAST *step, ExprAST *body)
180    : VarName(varname), Start(start), End(end), Step(step), Body(body) {}
181  virtual Value *Codegen();
182};
183
184/// VarExprAST - Expression class for var/in
185class VarExprAST : public ExprAST {
186  std::vector<std::pair<std::string, ExprAST*> > VarNames;
187  ExprAST *Body;
188public:
189  VarExprAST(const std::vector<std::pair<std::string, ExprAST*> > &varnames,
190             ExprAST *body)
191  : VarNames(varnames), Body(body) {}
192
193  virtual Value *Codegen();
194};
195
196/// PrototypeAST - This class represents the "prototype" for a function,
197/// which captures its argument names as well as if it is an operator.
198class PrototypeAST {
199  std::string Name;
200  std::vector<std::string> Args;
201  bool isOperator;
202  unsigned Precedence;  // Precedence if a binary op.
203public:
204  PrototypeAST(const std::string &name, const std::vector<std::string> &args,
205               bool isoperator = false, unsigned prec = 0)
206  : Name(name), Args(args), isOperator(isoperator), Precedence(prec) {}
207
208  bool isUnaryOp() const { return isOperator && Args.size() == 1; }
209  bool isBinaryOp() const { return isOperator && Args.size() == 2; }
210
211  char getOperatorName() const {
212    assert(isUnaryOp() || isBinaryOp());
213    return Name[Name.size()-1];
214  }
215
216  unsigned getBinaryPrecedence() const { return Precedence; }
217
218  Function *Codegen();
219
220  void CreateArgumentAllocas(Function *F);
221};
222
223/// FunctionAST - This class represents a function definition itself.
224class FunctionAST {
225  PrototypeAST *Proto;
226  ExprAST *Body;
227public:
228  FunctionAST(PrototypeAST *proto, ExprAST *body)
229    : Proto(proto), Body(body) {}
230
231  Function *Codegen();
232};
233
234//===----------------------------------------------------------------------===//
235// Parser
236//===----------------------------------------------------------------------===//
237
238/// CurTok/getNextToken - Provide a simple token buffer.  CurTok is the current
239/// token the parser is looking at.  getNextToken reads another token from the
240/// lexer and updates CurTok with its results.
241static int CurTok;
242static int getNextToken() {
243  return CurTok = gettok();
244}
245
246/// BinopPrecedence - This holds the precedence for each binary operator that is
247/// defined.
248static std::map<char, int> BinopPrecedence;
249
250/// GetTokPrecedence - Get the precedence of the pending binary operator token.
251static int GetTokPrecedence() {
252  if (!isascii(CurTok))
253    return -1;
254
255  // Make sure it's a declared binop.
256  int TokPrec = BinopPrecedence[CurTok];
257  if (TokPrec <= 0) return -1;
258  return TokPrec;
259}
260
261/// Error* - These are little helper functions for error handling.
262ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;}
263PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; }
264FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; }
265
266static ExprAST *ParseExpression();
267
268/// identifierexpr
269///   ::= identifier
270///   ::= identifier '(' expression* ')'
271static ExprAST *ParseIdentifierExpr() {
272  std::string IdName = IdentifierStr;
273
274  getNextToken();  // eat identifier.
275
276  if (CurTok != '(') // Simple variable ref.
277    return new VariableExprAST(IdName);
278
279  // Call.
280  getNextToken();  // eat (
281  std::vector<ExprAST*> Args;
282  if (CurTok != ')') {
283    while (1) {
284      ExprAST *Arg = ParseExpression();
285      if (!Arg) return 0;
286      Args.push_back(Arg);
287
288      if (CurTok == ')') break;
289
290      if (CurTok != ',')
291        return Error("Expected ')' or ',' in argument list");
292      getNextToken();
293    }
294  }
295
296  // Eat the ')'.
297  getNextToken();
298
299  return new CallExprAST(IdName, Args);
300}
301
302/// numberexpr ::= number
303static ExprAST *ParseNumberExpr() {
304  ExprAST *Result = new NumberExprAST(NumVal);
305  getNextToken(); // consume the number
306  return Result;
307}
308
309/// parenexpr ::= '(' expression ')'
310static ExprAST *ParseParenExpr() {
311  getNextToken();  // eat (.
312  ExprAST *V = ParseExpression();
313  if (!V) return 0;
314
315  if (CurTok != ')')
316    return Error("expected ')'");
317  getNextToken();  // eat ).
318  return V;
319}
320
321/// ifexpr ::= 'if' expression 'then' expression 'else' expression
322static ExprAST *ParseIfExpr() {
323  getNextToken();  // eat the if.
324
325  // condition.
326  ExprAST *Cond = ParseExpression();
327  if (!Cond) return 0;
328
329  if (CurTok != tok_then)
330    return Error("expected then");
331  getNextToken();  // eat the then
332
333  ExprAST *Then = ParseExpression();
334  if (Then == 0) return 0;
335
336  if (CurTok != tok_else)
337    return Error("expected else");
338
339  getNextToken();
340
341  ExprAST *Else = ParseExpression();
342  if (!Else) return 0;
343
344  return new IfExprAST(Cond, Then, Else);
345}
346
347/// forexpr ::= 'for' identifier '=' expr ',' expr (',' expr)? 'in' expression
348static ExprAST *ParseForExpr() {
349  getNextToken();  // eat the for.
350
351  if (CurTok != tok_identifier)
352    return Error("expected identifier after for");
353
354  std::string IdName = IdentifierStr;
355  getNextToken();  // eat identifier.
356
357  if (CurTok != '=')
358    return Error("expected '=' after for");
359  getNextToken();  // eat '='.
360
361
362  ExprAST *Start = ParseExpression();
363  if (Start == 0) return 0;
364  if (CurTok != ',')
365    return Error("expected ',' after for start value");
366  getNextToken();
367
368  ExprAST *End = ParseExpression();
369  if (End == 0) return 0;
370
371  // The step value is optional.
372  ExprAST *Step = 0;
373  if (CurTok == ',') {
374    getNextToken();
375    Step = ParseExpression();
376    if (Step == 0) return 0;
377  }
378
379  if (CurTok != tok_in)
380    return Error("expected 'in' after for");
381  getNextToken();  // eat 'in'.
382
383  ExprAST *Body = ParseExpression();
384  if (Body == 0) return 0;
385
386  return new ForExprAST(IdName, Start, End, Step, Body);
387}
388
389/// varexpr ::= 'var' identifier ('=' expression)?
390//                    (',' identifier ('=' expression)?)* 'in' expression
391static ExprAST *ParseVarExpr() {
392  getNextToken();  // eat the var.
393
394  std::vector<std::pair<std::string, ExprAST*> > VarNames;
395
396  // At least one variable name is required.
397  if (CurTok != tok_identifier)
398    return Error("expected identifier after var");
399
400  while (1) {
401    std::string Name = IdentifierStr;
402    getNextToken();  // eat identifier.
403
404    // Read the optional initializer.
405    ExprAST *Init = 0;
406    if (CurTok == '=') {
407      getNextToken(); // eat the '='.
408
409      Init = ParseExpression();
410      if (Init == 0) return 0;
411    }
412
413    VarNames.push_back(std::make_pair(Name, Init));
414
415    // End of var list, exit loop.
416    if (CurTok != ',') break;
417    getNextToken(); // eat the ','.
418
419    if (CurTok != tok_identifier)
420      return Error("expected identifier list after var");
421  }
422
423  // At this point, we have to have 'in'.
424  if (CurTok != tok_in)
425    return Error("expected 'in' keyword after 'var'");
426  getNextToken();  // eat 'in'.
427
428  ExprAST *Body = ParseExpression();
429  if (Body == 0) return 0;
430
431  return new VarExprAST(VarNames, Body);
432}
433
434/// primary
435///   ::= identifierexpr
436///   ::= numberexpr
437///   ::= parenexpr
438///   ::= ifexpr
439///   ::= forexpr
440///   ::= varexpr
441static ExprAST *ParsePrimary() {
442  switch (CurTok) {
443  default: return Error("unknown token when expecting an expression");
444  case tok_identifier: return ParseIdentifierExpr();
445  case tok_number:     return ParseNumberExpr();
446  case '(':            return ParseParenExpr();
447  case tok_if:         return ParseIfExpr();
448  case tok_for:        return ParseForExpr();
449  case tok_var:        return ParseVarExpr();
450  }
451}
452
453/// unary
454///   ::= primary
455///   ::= '!' unary
456static ExprAST *ParseUnary() {
457  // If the current token is not an operator, it must be a primary expr.
458  if (!isascii(CurTok) || CurTok == '(' || CurTok == ',')
459    return ParsePrimary();
460
461  // If this is a unary operator, read it.
462  int Opc = CurTok;
463  getNextToken();
464  if (ExprAST *Operand = ParseUnary())
465    return new UnaryExprAST(Opc, Operand);
466  return 0;
467}
468
469/// binoprhs
470///   ::= ('+' unary)*
471static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) {
472  // If this is a binop, find its precedence.
473  while (1) {
474    int TokPrec = GetTokPrecedence();
475
476    // If this is a binop that binds at least as tightly as the current binop,
477    // consume it, otherwise we are done.
478    if (TokPrec < ExprPrec)
479      return LHS;
480
481    // Okay, we know this is a binop.
482    int BinOp = CurTok;
483    getNextToken();  // eat binop
484
485    // Parse the unary expression after the binary operator.
486    ExprAST *RHS = ParseUnary();
487    if (!RHS) return 0;
488
489    // If BinOp binds less tightly with RHS than the operator after RHS, let
490    // the pending operator take RHS as its LHS.
491    int NextPrec = GetTokPrecedence();
492    if (TokPrec < NextPrec) {
493      RHS = ParseBinOpRHS(TokPrec+1, RHS);
494      if (RHS == 0) return 0;
495    }
496
497    // Merge LHS/RHS.
498    LHS = new BinaryExprAST(BinOp, LHS, RHS);
499  }
500}
501
502/// expression
503///   ::= unary binoprhs
504///
505static ExprAST *ParseExpression() {
506  ExprAST *LHS = ParseUnary();
507  if (!LHS) return 0;
508
509  return ParseBinOpRHS(0, LHS);
510}
511
512/// prototype
513///   ::= id '(' id* ')'
514///   ::= binary LETTER number? (id, id)
515///   ::= unary LETTER (id)
516static PrototypeAST *ParsePrototype() {
517  std::string FnName;
518
519  unsigned Kind = 0; // 0 = identifier, 1 = unary, 2 = binary.
520  unsigned BinaryPrecedence = 30;
521
522  switch (CurTok) {
523  default:
524    return ErrorP("Expected function name in prototype");
525  case tok_identifier:
526    FnName = IdentifierStr;
527    Kind = 0;
528    getNextToken();
529    break;
530  case tok_unary:
531    getNextToken();
532    if (!isascii(CurTok))
533      return ErrorP("Expected unary operator");
534    FnName = "unary";
535    FnName += (char)CurTok;
536    Kind = 1;
537    getNextToken();
538    break;
539  case tok_binary:
540    getNextToken();
541    if (!isascii(CurTok))
542      return ErrorP("Expected binary operator");
543    FnName = "binary";
544    FnName += (char)CurTok;
545    Kind = 2;
546    getNextToken();
547
548    // Read the precedence if present.
549    if (CurTok == tok_number) {
550      if (NumVal < 1 || NumVal > 100)
551        return ErrorP("Invalid precedecnce: must be 1..100");
552      BinaryPrecedence = (unsigned)NumVal;
553      getNextToken();
554    }
555    break;
556  }
557
558  if (CurTok != '(')
559    return ErrorP("Expected '(' in prototype");
560
561  std::vector<std::string> ArgNames;
562  while (getNextToken() == tok_identifier)
563    ArgNames.push_back(IdentifierStr);
564  if (CurTok != ')')
565    return ErrorP("Expected ')' in prototype");
566
567  // success.
568  getNextToken();  // eat ')'.
569
570  // Verify right number of names for operator.
571  if (Kind && ArgNames.size() != Kind)
572    return ErrorP("Invalid number of operands for operator");
573
574  return new PrototypeAST(FnName, ArgNames, Kind != 0, BinaryPrecedence);
575}
576
577/// definition ::= 'def' prototype expression
578static FunctionAST *ParseDefinition() {
579  getNextToken();  // eat def.
580  PrototypeAST *Proto = ParsePrototype();
581  if (Proto == 0) return 0;
582
583  if (ExprAST *E = ParseExpression())
584    return new FunctionAST(Proto, E);
585  return 0;
586}
587
588/// toplevelexpr ::= expression
589static FunctionAST *ParseTopLevelExpr() {
590  if (ExprAST *E = ParseExpression()) {
591    // Make an anonymous proto.
592    PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>());
593    return new FunctionAST(Proto, E);
594  }
595  return 0;
596}
597
598/// external ::= 'extern' prototype
599static PrototypeAST *ParseExtern() {
600  getNextToken();  // eat extern.
601  return ParsePrototype();
602}
603
604//===----------------------------------------------------------------------===//
605// Quick and dirty hack
606//===----------------------------------------------------------------------===//
607
608// FIXME: Obviously we can do better than this
609std::string GenerateUniqueName(const char *root)
610{
611  static int i = 0;
612  char s[16];
613  sprintf(s, "%s%d", root, i++);
614  std::string S = s;
615  return S;
616}
617
618std::string MakeLegalFunctionName(std::string Name)
619{
620  std::string NewName;
621  if (!Name.length())
622      return GenerateUniqueName("anon_func_");
623
624  // Start with what we have
625  NewName = Name;
626
627  // Look for a numberic first character
628  if (NewName.find_first_of("0123456789") == 0) {
629    NewName.insert(0, 1, 'n');
630  }
631
632  // Replace illegal characters with their ASCII equivalent
633  std::string legal_elements = "_abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789";
634  size_t pos;
635  while ((pos = NewName.find_first_not_of(legal_elements)) != std::string::npos) {
636    char old_c = NewName.at(pos);
637    char new_str[16];
638    sprintf(new_str, "%d", (int)old_c);
639    NewName = NewName.replace(pos, 1, new_str);
640  }
641
642  return NewName;
643}
644
645//===----------------------------------------------------------------------===//
646// MCJIT helper class
647//===----------------------------------------------------------------------===//
648
649class MCJITHelper
650{
651public:
652  MCJITHelper(LLVMContext& C) : Context(C), OpenModule(NULL) {}
653  ~MCJITHelper();
654
655  Function *getFunction(const std::string FnName);
656  Module *getModuleForNewFunction();
657  void *getPointerToFunction(Function* F);
658  void *getPointerToNamedFunction(const std::string &Name);
659  void dump();
660
661private:
662  typedef std::vector<Module*> ModuleVector;
663  typedef std::vector<ExecutionEngine*> EngineVector;
664
665  LLVMContext  &Context;
666  Module       *OpenModule;
667  ModuleVector  Modules;
668  EngineVector  Engines;
669};
670
671class HelpingMemoryManager : public SectionMemoryManager
672{
673  HelpingMemoryManager(const HelpingMemoryManager&) = delete;
674  void operator=(const HelpingMemoryManager&) = delete;
675
676public:
677  HelpingMemoryManager(MCJITHelper *Helper) : MasterHelper(Helper) {}
678  virtual ~HelpingMemoryManager() {}
679
680  /// This method returns the address of the specified function.
681  /// Our implementation will attempt to find functions in other
682  /// modules associated with the MCJITHelper to cross link functions
683  /// from one generated module to another.
684  ///
685  /// If \p AbortOnFailure is false and no function with the given name is
686  /// found, this function returns a null pointer. Otherwise, it prints a
687  /// message to stderr and aborts.
688  virtual void *getPointerToNamedFunction(const std::string &Name,
689                                          bool AbortOnFailure = true);
690private:
691  MCJITHelper *MasterHelper;
692};
693
694void *HelpingMemoryManager::getPointerToNamedFunction(const std::string &Name,
695                                        bool AbortOnFailure)
696{
697  // Try the standard symbol resolution first, but ask it not to abort.
698  void *pfn = SectionMemoryManager::getPointerToNamedFunction(Name, false);
699  if (pfn)
700    return pfn;
701
702  pfn = MasterHelper->getPointerToNamedFunction(Name);
703  if (!pfn && AbortOnFailure)
704    report_fatal_error("Program used external function '" + Name +
705                        "' which could not be resolved!");
706  return pfn;
707}
708
709MCJITHelper::~MCJITHelper()
710{
711  if (OpenModule)
712    delete OpenModule;
713  EngineVector::iterator begin = Engines.begin();
714  EngineVector::iterator end = Engines.end();
715  EngineVector::iterator it;
716  for (it = begin; it != end; ++it)
717    delete *it;
718}
719
720Function *MCJITHelper::getFunction(const std::string FnName) {
721  ModuleVector::iterator begin = Modules.begin();
722  ModuleVector::iterator end = Modules.end();
723  ModuleVector::iterator it;
724  for (it = begin; it != end; ++it) {
725    Function *F = (*it)->getFunction(FnName);
726    if (F) {
727      if (*it == OpenModule)
728          return F;
729
730      assert(OpenModule != NULL);
731
732      // This function is in a module that has already been JITed.
733      // We need to generate a new prototype for external linkage.
734      Function *PF = OpenModule->getFunction(FnName);
735      if (PF && !PF->empty()) {
736        ErrorF("redefinition of function across modules");
737        return 0;
738      }
739
740      // If we don't have a prototype yet, create one.
741      if (!PF)
742        PF = Function::Create(F->getFunctionType(),
743                                      Function::ExternalLinkage,
744                                      FnName,
745                                      OpenModule);
746      return PF;
747    }
748  }
749  return NULL;
750}
751
752Module *MCJITHelper::getModuleForNewFunction() {
753  // If we have a Module that hasn't been JITed, use that.
754  if (OpenModule)
755    return OpenModule;
756
757  // Otherwise create a new Module.
758  std::string ModName = GenerateUniqueName("mcjit_module_");
759  Module *M = new Module(ModName, Context);
760  Modules.push_back(M);
761  OpenModule = M;
762  return M;
763}
764
765void *MCJITHelper::getPointerToFunction(Function* F) {
766  // See if an existing instance of MCJIT has this function.
767  EngineVector::iterator begin = Engines.begin();
768  EngineVector::iterator end = Engines.end();
769  EngineVector::iterator it;
770  for (it = begin; it != end; ++it) {
771    void *P = (*it)->getPointerToFunction(F);
772    if (P)
773      return P;
774  }
775
776  // If we didn't find the function, see if we can generate it.
777  if (OpenModule) {
778    std::string ErrStr;
779    ExecutionEngine *NewEngine = EngineBuilder(OpenModule)
780                                              .setErrorStr(&ErrStr)
781                                              .setMCJITMemoryManager(new HelpingMemoryManager(this))
782                                              .create();
783    if (!NewEngine) {
784      fprintf(stderr, "Could not create ExecutionEngine: %s\n", ErrStr.c_str());
785      exit(1);
786    }
787
788    // Create a function pass manager for this engine
789    FunctionPassManager *FPM = new FunctionPassManager(OpenModule);
790
791    // Set up the optimizer pipeline.  Start with registering info about how the
792    // target lays out data structures.
793    FPM->add(new DataLayout(*NewEngine->getDataLayout()));
794    // Provide basic AliasAnalysis support for GVN.
795    FPM->add(createBasicAliasAnalysisPass());
796    // Promote allocas to registers.
797    FPM->add(createPromoteMemoryToRegisterPass());
798    // Do simple "peephole" optimizations and bit-twiddling optzns.
799    FPM->add(createInstructionCombiningPass());
800    // Reassociate expressions.
801    FPM->add(createReassociatePass());
802    // Eliminate Common SubExpressions.
803    FPM->add(createGVNPass());
804    // Simplify the control flow graph (deleting unreachable blocks, etc).
805    FPM->add(createCFGSimplificationPass());
806    FPM->doInitialization();
807
808    // For each function in the module
809    Module::iterator it;
810    Module::iterator end = OpenModule->end();
811    for (it = OpenModule->begin(); it != end; ++it) {
812      // Run the FPM on this function
813      FPM->run(*it);
814    }
815
816    // We don't need this anymore
817    delete FPM;
818
819    OpenModule = NULL;
820    Engines.push_back(NewEngine);
821    NewEngine->finalizeObject();
822    return NewEngine->getPointerToFunction(F);
823  }
824  return NULL;
825}
826
827void *MCJITHelper::getPointerToNamedFunction(const std::string &Name)
828{
829  // Look for the function in each of our execution engines.
830  EngineVector::iterator begin = Engines.begin();
831  EngineVector::iterator end = Engines.end();
832  EngineVector::iterator it;
833  for (it = begin; it != end; ++it) {
834    if (Function *F = (*it)->FindFunctionNamed(Name.c_str()))
835        return (*it)->getPointerToFunction(F);
836  }
837
838  return NULL;
839}
840
841void MCJITHelper::dump()
842{
843  ModuleVector::iterator begin = Modules.begin();
844  ModuleVector::iterator end = Modules.end();
845  ModuleVector::iterator it;
846  for (it = begin; it != end; ++it)
847    (*it)->dump();
848}
849
850//===----------------------------------------------------------------------===//
851// Code Generation
852//===----------------------------------------------------------------------===//
853
854static MCJITHelper *TheHelper;
855static LLVMContext TheContext;
856static IRBuilder<> Builder(TheContext);
857static std::map<std::string, AllocaInst*> NamedValues;
858
859Value *ErrorV(const char *Str) { Error(Str); return 0; }
860
861/// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of
862/// the function.  This is used for mutable variables etc.
863static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction,
864                                          const std::string &VarName) {
865  IRBuilder<> TmpB(&TheFunction->getEntryBlock(),
866                 TheFunction->getEntryBlock().begin());
867  return TmpB.CreateAlloca(Type::getDoubleTy(TheContext), 0, VarName.c_str());
868}
869
870Value *NumberExprAST::Codegen() {
871  return ConstantFP::get(TheContext, APFloat(Val));
872}
873
874Value *VariableExprAST::Codegen() {
875  // Look this variable up in the function.
876  Value *V = NamedValues[Name];
877  char ErrStr[256];
878  sprintf(ErrStr, "Unknown variable name %s", Name.c_str());
879  if (V == 0) return ErrorV(ErrStr);
880
881  // Load the value.
882  return Builder.CreateLoad(V, Name.c_str());
883}
884
885Value *UnaryExprAST::Codegen() {
886  Value *OperandV = Operand->Codegen();
887  if (OperandV == 0) return 0;
888
889  Function *F = TheHelper->getFunction(MakeLegalFunctionName(std::string("unary")+Opcode));
890  if (F == 0)
891    return ErrorV("Unknown unary operator");
892
893  return Builder.CreateCall(F, OperandV, "unop");
894}
895
896Value *BinaryExprAST::Codegen() {
897  // Special case '=' because we don't want to emit the LHS as an expression.
898  if (Op == '=') {
899    // Assignment requires the LHS to be an identifier.
900    VariableExprAST *LHSE = static_cast<VariableExprAST*>(LHS);
901    if (!LHSE)
902      return ErrorV("destination of '=' must be a variable");
903    // Codegen the RHS.
904    Value *Val = RHS->Codegen();
905    if (Val == 0) return 0;
906
907    // Look up the name.
908    Value *Variable = NamedValues[LHSE->getName()];
909    if (Variable == 0) return ErrorV("Unknown variable name");
910
911    Builder.CreateStore(Val, Variable);
912    return Val;
913  }
914
915  Value *L = LHS->Codegen();
916  Value *R = RHS->Codegen();
917  if (L == 0 || R == 0) return 0;
918
919  switch (Op) {
920  case '+': return Builder.CreateFAdd(L, R, "addtmp");
921  case '-': return Builder.CreateFSub(L, R, "subtmp");
922  case '*': return Builder.CreateFMul(L, R, "multmp");
923  case '/': return Builder.CreateFDiv(L, R, "divtmp");
924  case '<':
925    L = Builder.CreateFCmpULT(L, R, "cmptmp");
926    // Convert bool 0/1 to double 0.0 or 1.0
927    return Builder.CreateUIToFP(L, Type::getDoubleTy(TheContext), "booltmp");
928  default: break;
929  }
930
931  // If it wasn't a builtin binary operator, it must be a user defined one. Emit
932  // a call to it.
933  Function *F = TheHelper->getFunction(MakeLegalFunctionName(std::string("binary")+Op));
934  assert(F && "binary operator not found!");
935
936  Value *Ops[] = { L, R };
937  return Builder.CreateCall(F, Ops, "binop");
938}
939
940Value *CallExprAST::Codegen() {
941  // Look up the name in the global module table.
942  Function *CalleeF = TheHelper->getFunction(Callee);
943  if (CalleeF == 0)
944    return ErrorV("Unknown function referenced");
945
946  // If argument mismatch error.
947  if (CalleeF->arg_size() != Args.size())
948    return ErrorV("Incorrect # arguments passed");
949
950  std::vector<Value*> ArgsV;
951  for (unsigned i = 0, e = Args.size(); i != e; ++i) {
952    ArgsV.push_back(Args[i]->Codegen());
953    if (ArgsV.back() == 0) return 0;
954  }
955
956  return Builder.CreateCall(CalleeF, ArgsV, "calltmp");
957}
958
959Value *IfExprAST::Codegen() {
960  Value *CondV = Cond->Codegen();
961  if (CondV == 0) return 0;
962
963  // Convert condition to a bool by comparing equal to 0.0.
964  CondV = Builder.CreateFCmpONE(
965      CondV, ConstantFP::get(TheContext, APFloat(0.0)), "ifcond");
966
967  Function *TheFunction = Builder.GetInsertBlock()->getParent();
968
969  // Create blocks for the then and else cases.  Insert the 'then' block at the
970  // end of the function.
971  BasicBlock *ThenBB = BasicBlock::Create(TheContext, "then", TheFunction);
972  BasicBlock *ElseBB = BasicBlock::Create(TheContext, "else");
973  BasicBlock *MergeBB = BasicBlock::Create(TheContext, "ifcont");
974
975  Builder.CreateCondBr(CondV, ThenBB, ElseBB);
976
977  // Emit then value.
978  Builder.SetInsertPoint(ThenBB);
979
980  Value *ThenV = Then->Codegen();
981  if (ThenV == 0) return 0;
982
983  Builder.CreateBr(MergeBB);
984  // Codegen of 'Then' can change the current block, update ThenBB for the PHI.
985  ThenBB = Builder.GetInsertBlock();
986
987  // Emit else block.
988  TheFunction->getBasicBlockList().push_back(ElseBB);
989  Builder.SetInsertPoint(ElseBB);
990
991  Value *ElseV = Else->Codegen();
992  if (ElseV == 0) return 0;
993
994  Builder.CreateBr(MergeBB);
995  // Codegen of 'Else' can change the current block, update ElseBB for the PHI.
996  ElseBB = Builder.GetInsertBlock();
997
998  // Emit merge block.
999  TheFunction->getBasicBlockList().push_back(MergeBB);
1000  Builder.SetInsertPoint(MergeBB);
1001  PHINode *PN = Builder.CreatePHI(Type::getDoubleTy(TheContext), 2, "iftmp");
1002
1003  PN->addIncoming(ThenV, ThenBB);
1004  PN->addIncoming(ElseV, ElseBB);
1005  return PN;
1006}
1007
1008Value *ForExprAST::Codegen() {
1009  // Output this as:
1010  //   var = alloca double
1011  //   ...
1012  //   start = startexpr
1013  //   store start -> var
1014  //   goto loop
1015  // loop:
1016  //   ...
1017  //   bodyexpr
1018  //   ...
1019  // loopend:
1020  //   step = stepexpr
1021  //   endcond = endexpr
1022  //
1023  //   curvar = load var
1024  //   nextvar = curvar + step
1025  //   store nextvar -> var
1026  //   br endcond, loop, endloop
1027  // outloop:
1028
1029  Function *TheFunction = Builder.GetInsertBlock()->getParent();
1030
1031  // Create an alloca for the variable in the entry block.
1032  AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1033
1034  // Emit the start code first, without 'variable' in scope.
1035  Value *StartVal = Start->Codegen();
1036  if (StartVal == 0) return 0;
1037
1038  // Store the value into the alloca.
1039  Builder.CreateStore(StartVal, Alloca);
1040
1041  // Make the new basic block for the loop header, inserting after current
1042  // block.
1043  BasicBlock *LoopBB = BasicBlock::Create(TheContext, "loop", TheFunction);
1044
1045  // Insert an explicit fall through from the current block to the LoopBB.
1046  Builder.CreateBr(LoopBB);
1047
1048  // Start insertion in LoopBB.
1049  Builder.SetInsertPoint(LoopBB);
1050
1051  // Within the loop, the variable is defined equal to the PHI node.  If it
1052  // shadows an existing variable, we have to restore it, so save it now.
1053  AllocaInst *OldVal = NamedValues[VarName];
1054  NamedValues[VarName] = Alloca;
1055
1056  // Emit the body of the loop.  This, like any other expr, can change the
1057  // current BB.  Note that we ignore the value computed by the body, but don't
1058  // allow an error.
1059  if (Body->Codegen() == 0)
1060    return 0;
1061
1062  // Emit the step value.
1063  Value *StepVal;
1064  if (Step) {
1065    StepVal = Step->Codegen();
1066    if (StepVal == 0) return 0;
1067  } else {
1068    // If not specified, use 1.0.
1069    StepVal = ConstantFP::get(TheContext, APFloat(1.0));
1070  }
1071
1072  // Compute the end condition.
1073  Value *EndCond = End->Codegen();
1074  if (EndCond == 0) return EndCond;
1075
1076  // Reload, increment, and restore the alloca.  This handles the case where
1077  // the body of the loop mutates the variable.
1078  Value *CurVar = Builder.CreateLoad(Alloca, VarName.c_str());
1079  Value *NextVar = Builder.CreateFAdd(CurVar, StepVal, "nextvar");
1080  Builder.CreateStore(NextVar, Alloca);
1081
1082  // Convert condition to a bool by comparing equal to 0.0.
1083  EndCond = Builder.CreateFCmpONE(
1084      EndCond, ConstantFP::get(TheContext, APFloat(0.0)), "loopcond");
1085
1086  // Create the "after loop" block and insert it.
1087  BasicBlock *AfterBB =
1088      BasicBlock::Create(TheContext, "afterloop", TheFunction);
1089
1090  // Insert the conditional branch into the end of LoopEndBB.
1091  Builder.CreateCondBr(EndCond, LoopBB, AfterBB);
1092
1093  // Any new code will be inserted in AfterBB.
1094  Builder.SetInsertPoint(AfterBB);
1095
1096  // Restore the unshadowed variable.
1097  if (OldVal)
1098    NamedValues[VarName] = OldVal;
1099  else
1100    NamedValues.erase(VarName);
1101
1102
1103  // for expr always returns 0.0.
1104  return Constant::getNullValue(Type::getDoubleTy(TheContext));
1105}
1106
1107Value *VarExprAST::Codegen() {
1108  std::vector<AllocaInst *> OldBindings;
1109
1110  Function *TheFunction = Builder.GetInsertBlock()->getParent();
1111
1112  // Register all variables and emit their initializer.
1113  for (unsigned i = 0, e = VarNames.size(); i != e; ++i) {
1114    const std::string &VarName = VarNames[i].first;
1115    ExprAST *Init = VarNames[i].second;
1116
1117    // Emit the initializer before adding the variable to scope, this prevents
1118    // the initializer from referencing the variable itself, and permits stuff
1119    // like this:
1120    //  var a = 1 in
1121    //    var a = a in ...   # refers to outer 'a'.
1122    Value *InitVal;
1123    if (Init) {
1124      InitVal = Init->Codegen();
1125      if (InitVal == 0) return 0;
1126    } else { // If not specified, use 0.0.
1127      InitVal = ConstantFP::get(TheContext, APFloat(0.0));
1128    }
1129
1130    AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1131    Builder.CreateStore(InitVal, Alloca);
1132
1133    // Remember the old variable binding so that we can restore the binding when
1134    // we unrecurse.
1135    OldBindings.push_back(NamedValues[VarName]);
1136
1137    // Remember this binding.
1138    NamedValues[VarName] = Alloca;
1139  }
1140
1141  // Codegen the body, now that all vars are in scope.
1142  Value *BodyVal = Body->Codegen();
1143  if (BodyVal == 0) return 0;
1144
1145  // Pop all our variables from scope.
1146  for (unsigned i = 0, e = VarNames.size(); i != e; ++i)
1147    NamedValues[VarNames[i].first] = OldBindings[i];
1148
1149  // Return the body computation.
1150  return BodyVal;
1151}
1152
1153Function *PrototypeAST::Codegen() {
1154  // Make the function type:  double(double,double) etc.
1155  std::vector<Type *> Doubles(Args.size(), Type::getDoubleTy(TheContext));
1156  FunctionType *FT =
1157      FunctionType::get(Type::getDoubleTy(TheContext), Doubles, false);
1158
1159  std::string FnName = MakeLegalFunctionName(Name);
1160
1161  Module* M = TheHelper->getModuleForNewFunction();
1162
1163  Function *F = Function::Create(FT, Function::ExternalLinkage, FnName, M);
1164
1165  // If F conflicted, there was already something named 'FnName'.  If it has a
1166  // body, don't allow redefinition or reextern.
1167  if (F->getName() != FnName) {
1168    // Delete the one we just made and get the existing one.
1169    F->eraseFromParent();
1170    F = M->getFunction(Name);
1171
1172    // If F already has a body, reject this.
1173    if (!F->empty()) {
1174      ErrorF("redefinition of function");
1175      return 0;
1176    }
1177
1178    // If F took a different number of args, reject.
1179    if (F->arg_size() != Args.size()) {
1180      ErrorF("redefinition of function with different # args");
1181      return 0;
1182    }
1183  }
1184
1185  // Set names for all arguments.
1186  unsigned Idx = 0;
1187  for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
1188       ++AI, ++Idx)
1189    AI->setName(Args[Idx]);
1190
1191  return F;
1192}
1193
1194/// CreateArgumentAllocas - Create an alloca for each argument and register the
1195/// argument in the symbol table so that references to it will succeed.
1196void PrototypeAST::CreateArgumentAllocas(Function *F) {
1197  Function::arg_iterator AI = F->arg_begin();
1198  for (unsigned Idx = 0, e = Args.size(); Idx != e; ++Idx, ++AI) {
1199    // Create an alloca for this variable.
1200    AllocaInst *Alloca = CreateEntryBlockAlloca(F, Args[Idx]);
1201
1202    // Store the initial value into the alloca.
1203    Builder.CreateStore(AI, Alloca);
1204
1205    // Add arguments to variable symbol table.
1206    NamedValues[Args[Idx]] = Alloca;
1207  }
1208}
1209
1210Function *FunctionAST::Codegen() {
1211  NamedValues.clear();
1212
1213  Function *TheFunction = Proto->Codegen();
1214  if (TheFunction == 0)
1215    return 0;
1216
1217  // If this is an operator, install it.
1218  if (Proto->isBinaryOp())
1219    BinopPrecedence[Proto->getOperatorName()] = Proto->getBinaryPrecedence();
1220
1221  // Create a new basic block to start insertion into.
1222  BasicBlock *BB = BasicBlock::Create(TheContext, "entry", TheFunction);
1223  Builder.SetInsertPoint(BB);
1224
1225  // Add all arguments to the symbol table and create their allocas.
1226  Proto->CreateArgumentAllocas(TheFunction);
1227
1228  if (Value *RetVal = Body->Codegen()) {
1229    // Finish off the function.
1230    Builder.CreateRet(RetVal);
1231
1232    // Validate the generated code, checking for consistency.
1233    verifyFunction(*TheFunction);
1234
1235    return TheFunction;
1236  }
1237
1238  // Error reading body, remove function.
1239  TheFunction->eraseFromParent();
1240
1241  if (Proto->isBinaryOp())
1242    BinopPrecedence.erase(Proto->getOperatorName());
1243  return 0;
1244}
1245
1246//===----------------------------------------------------------------------===//
1247// Top-Level parsing and JIT Driver
1248//===----------------------------------------------------------------------===//
1249
1250static void HandleDefinition() {
1251  if (FunctionAST *F = ParseDefinition()) {
1252    if (Function *LF = F->Codegen()) {
1253#ifndef MINIMAL_STDERR_OUTPUT
1254      fprintf(stderr, "Read function definition:");
1255      LF->dump();
1256#endif
1257    }
1258  } else {
1259    // Skip token for error recovery.
1260    getNextToken();
1261  }
1262}
1263
1264static void HandleExtern() {
1265  if (PrototypeAST *P = ParseExtern()) {
1266    if (Function *F = P->Codegen()) {
1267#ifndef MINIMAL_STDERR_OUTPUT
1268      fprintf(stderr, "Read extern: ");
1269      F->dump();
1270#endif
1271    }
1272  } else {
1273    // Skip token for error recovery.
1274    getNextToken();
1275  }
1276}
1277
1278static void HandleTopLevelExpression() {
1279  // Evaluate a top-level expression into an anonymous function.
1280  if (FunctionAST *F = ParseTopLevelExpr()) {
1281    if (Function *LF = F->Codegen()) {
1282      // JIT the function, returning a function pointer.
1283      void *FPtr = TheHelper->getPointerToFunction(LF);
1284
1285      // Cast it to the right type (takes no arguments, returns a double) so we
1286      // can call it as a native function.
1287      double (*FP)() = (double (*)())(intptr_t)FPtr;
1288#ifdef MINIMAL_STDERR_OUTPUT
1289      FP();
1290#else
1291      fprintf(stderr, "Evaluated to %f\n", FP());
1292#endif
1293    }
1294  } else {
1295    // Skip token for error recovery.
1296    getNextToken();
1297  }
1298}
1299
1300/// top ::= definition | external | expression | ';'
1301static void MainLoop() {
1302  while (1) {
1303#ifndef MINIMAL_STDERR_OUTPUT
1304    fprintf(stderr, "ready> ");
1305#endif
1306    switch (CurTok) {
1307    case tok_eof:    return;
1308    case ';':        getNextToken(); break;  // ignore top-level semicolons.
1309    case tok_def:    HandleDefinition(); break;
1310    case tok_extern: HandleExtern(); break;
1311    default:         HandleTopLevelExpression(); break;
1312    }
1313  }
1314}
1315
1316//===----------------------------------------------------------------------===//
1317// "Library" functions that can be "extern'd" from user code.
1318//===----------------------------------------------------------------------===//
1319
1320/// putchard - putchar that takes a double and returns 0.
1321extern "C"
1322double putchard(double X) {
1323  putchar((char)X);
1324  return 0;
1325}
1326
1327/// printd - printf that takes a double prints it as "%f\n", returning 0.
1328extern "C"
1329double printd(double X) {
1330  printf("%f", X);
1331  return 0;
1332}
1333
1334extern "C"
1335double printlf() {
1336  printf("\n");
1337  return 0;
1338}
1339
1340//===----------------------------------------------------------------------===//
1341// Main driver code.
1342//===----------------------------------------------------------------------===//
1343
1344int main() {
1345  InitializeNativeTarget();
1346  InitializeNativeTargetAsmPrinter();
1347  InitializeNativeTargetAsmParser();
1348  LLVMContext &Context = TheContext;
1349
1350  // Install standard binary operators.
1351  // 1 is lowest precedence.
1352  BinopPrecedence['='] = 2;
1353  BinopPrecedence['<'] = 10;
1354  BinopPrecedence['+'] = 20;
1355  BinopPrecedence['-'] = 20;
1356  BinopPrecedence['/'] = 40;
1357  BinopPrecedence['*'] = 40;  // highest.
1358
1359  // Prime the first token.
1360#ifndef MINIMAL_STDERR_OUTPUT
1361  fprintf(stderr, "ready> ");
1362#endif
1363  getNextToken();
1364
1365  // Make the helper, which holds all the code.
1366  TheHelper = new MCJITHelper(Context);
1367
1368  // Run the main "interpreter loop" now.
1369  MainLoop();
1370
1371#ifndef MINIMAL_STDERR_OUTPUT
1372  // Print out all of the generated code.
1373  TheHelper->dump();
1374#endif
1375
1376  return 0;
1377}
1378