1 #define MINIMAL_STDERR_OUTPUT
2 
3 #include "llvm/Analysis/Passes.h"
4 #include "llvm/ExecutionEngine/ExecutionEngine.h"
5 #include "llvm/ExecutionEngine/MCJIT.h"
6 #include "llvm/ExecutionEngine/ObjectCache.h"
7 #include "llvm/ExecutionEngine/SectionMemoryManager.h"
8 #include "llvm/IR/DataLayout.h"
9 #include "llvm/IR/DerivedTypes.h"
10 #include "llvm/IR/IRBuilder.h"
11 #include "llvm/IR/LLVMContext.h"
12 #include "llvm/IR/LegacyPassManager.h"
13 #include "llvm/IR/Module.h"
14 #include "llvm/IR/Verifier.h"
15 #include "llvm/IRReader/IRReader.h"
16 #include "llvm/Support/CommandLine.h"
17 #include "llvm/Support/FileSystem.h"
18 #include "llvm/Support/Path.h"
19 #include "llvm/Support/SourceMgr.h"
20 #include "llvm/Support/TargetSelect.h"
21 #include "llvm/Support/raw_ostream.h"
22 #include "llvm/Transforms/Scalar.h"
23 #include <cctype>
24 #include <cstdio>
25 #include <map>
26 #include <string>
27 #include <vector>
28 using namespace llvm;
29 
30 //===----------------------------------------------------------------------===//
31 // Command-line options
32 //===----------------------------------------------------------------------===//
33 
34 cl::opt<std::string>
35 InputIR("input-IR",
36         cl::desc("Specify the name of an IR file to load for function definitions"),
37         cl::value_desc("input IR file name"));
38 
39 cl::opt<bool>
40 UseObjectCache("use-object-cache",
41                cl::desc("Enable use of the MCJIT object caching"),
42                cl::init(false));
43 
44 //===----------------------------------------------------------------------===//
45 // Lexer
46 //===----------------------------------------------------------------------===//
47 
48 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
49 // of these for known things.
50 enum Token {
51   tok_eof = -1,
52 
53   // commands
54   tok_def = -2, tok_extern = -3,
55 
56   // primary
57   tok_identifier = -4, tok_number = -5,
58 
59   // control
60   tok_if = -6, tok_then = -7, tok_else = -8,
61   tok_for = -9, tok_in = -10,
62 
63   // operators
64   tok_binary = -11, tok_unary = -12,
65 
66   // var definition
67   tok_var = -13
68 };
69 
70 static std::string IdentifierStr;  // Filled in if tok_identifier
71 static double NumVal;              // Filled in if tok_number
72 
73 /// gettok - Return the next token from standard input.
gettok()74 static int gettok() {
75   static int LastChar = ' ';
76 
77   // Skip any whitespace.
78   while (isspace(LastChar))
79     LastChar = getchar();
80 
81   if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
82     IdentifierStr = LastChar;
83     while (isalnum((LastChar = getchar())))
84       IdentifierStr += LastChar;
85 
86     if (IdentifierStr == "def") return tok_def;
87     if (IdentifierStr == "extern") return tok_extern;
88     if (IdentifierStr == "if") return tok_if;
89     if (IdentifierStr == "then") return tok_then;
90     if (IdentifierStr == "else") return tok_else;
91     if (IdentifierStr == "for") return tok_for;
92     if (IdentifierStr == "in") return tok_in;
93     if (IdentifierStr == "binary") return tok_binary;
94     if (IdentifierStr == "unary") return tok_unary;
95     if (IdentifierStr == "var") return tok_var;
96     return tok_identifier;
97   }
98 
99   if (isdigit(LastChar) || LastChar == '.') {   // Number: [0-9.]+
100     std::string NumStr;
101     do {
102       NumStr += LastChar;
103       LastChar = getchar();
104     } while (isdigit(LastChar) || LastChar == '.');
105 
106     NumVal = strtod(NumStr.c_str(), 0);
107     return tok_number;
108   }
109 
110   if (LastChar == '#') {
111     // Comment until end of line.
112     do LastChar = getchar();
113     while (LastChar != EOF && LastChar != '\n' && LastChar != '\r');
114 
115     if (LastChar != EOF)
116       return gettok();
117   }
118 
119   // Check for end of file.  Don't eat the EOF.
120   if (LastChar == EOF)
121     return tok_eof;
122 
123   // Otherwise, just return the character as its ascii value.
124   int ThisChar = LastChar;
125   LastChar = getchar();
126   return ThisChar;
127 }
128 
129 //===----------------------------------------------------------------------===//
130 // Abstract Syntax Tree (aka Parse Tree)
131 //===----------------------------------------------------------------------===//
132 
133 /// ExprAST - Base class for all expression nodes.
134 class ExprAST {
135 public:
~ExprAST()136   virtual ~ExprAST() {}
137   virtual Value *Codegen() = 0;
138 };
139 
140 /// NumberExprAST - Expression class for numeric literals like "1.0".
141 class NumberExprAST : public ExprAST {
142   double Val;
143 public:
NumberExprAST(double val)144   NumberExprAST(double val) : Val(val) {}
145   virtual Value *Codegen();
146 };
147 
148 /// VariableExprAST - Expression class for referencing a variable, like "a".
149 class VariableExprAST : public ExprAST {
150   std::string Name;
151 public:
VariableExprAST(const std::string & name)152   VariableExprAST(const std::string &name) : Name(name) {}
getName() const153   const std::string &getName() const { return Name; }
154   virtual Value *Codegen();
155 };
156 
157 /// UnaryExprAST - Expression class for a unary operator.
158 class UnaryExprAST : public ExprAST {
159   char Opcode;
160   ExprAST *Operand;
161 public:
UnaryExprAST(char opcode,ExprAST * operand)162   UnaryExprAST(char opcode, ExprAST *operand)
163     : Opcode(opcode), Operand(operand) {}
164   virtual Value *Codegen();
165 };
166 
167 /// BinaryExprAST - Expression class for a binary operator.
168 class BinaryExprAST : public ExprAST {
169   char Op;
170   ExprAST *LHS, *RHS;
171 public:
BinaryExprAST(char op,ExprAST * lhs,ExprAST * rhs)172   BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs)
173     : Op(op), LHS(lhs), RHS(rhs) {}
174   virtual Value *Codegen();
175 };
176 
177 /// CallExprAST - Expression class for function calls.
178 class CallExprAST : public ExprAST {
179   std::string Callee;
180   std::vector<ExprAST*> Args;
181 public:
CallExprAST(const std::string & callee,std::vector<ExprAST * > & args)182   CallExprAST(const std::string &callee, std::vector<ExprAST*> &args)
183     : Callee(callee), Args(args) {}
184   virtual Value *Codegen();
185 };
186 
187 /// IfExprAST - Expression class for if/then/else.
188 class IfExprAST : public ExprAST {
189   ExprAST *Cond, *Then, *Else;
190 public:
IfExprAST(ExprAST * cond,ExprAST * then,ExprAST * _else)191   IfExprAST(ExprAST *cond, ExprAST *then, ExprAST *_else)
192   : Cond(cond), Then(then), Else(_else) {}
193   virtual Value *Codegen();
194 };
195 
196 /// ForExprAST - Expression class for for/in.
197 class ForExprAST : public ExprAST {
198   std::string VarName;
199   ExprAST *Start, *End, *Step, *Body;
200 public:
ForExprAST(const std::string & varname,ExprAST * start,ExprAST * end,ExprAST * step,ExprAST * body)201   ForExprAST(const std::string &varname, ExprAST *start, ExprAST *end,
202              ExprAST *step, ExprAST *body)
203     : VarName(varname), Start(start), End(end), Step(step), Body(body) {}
204   virtual Value *Codegen();
205 };
206 
207 /// VarExprAST - Expression class for var/in
208 class VarExprAST : public ExprAST {
209   std::vector<std::pair<std::string, ExprAST*> > VarNames;
210   ExprAST *Body;
211 public:
VarExprAST(const std::vector<std::pair<std::string,ExprAST * >> & varnames,ExprAST * body)212   VarExprAST(const std::vector<std::pair<std::string, ExprAST*> > &varnames,
213              ExprAST *body)
214   : VarNames(varnames), Body(body) {}
215 
216   virtual Value *Codegen();
217 };
218 
219 /// PrototypeAST - This class represents the "prototype" for a function,
220 /// which captures its argument names as well as if it is an operator.
221 class PrototypeAST {
222   std::string Name;
223   std::vector<std::string> Args;
224   bool isOperator;
225   unsigned Precedence;  // Precedence if a binary op.
226 public:
PrototypeAST(const std::string & name,const std::vector<std::string> & args,bool isoperator=false,unsigned prec=0)227   PrototypeAST(const std::string &name, const std::vector<std::string> &args,
228                bool isoperator = false, unsigned prec = 0)
229   : Name(name), Args(args), isOperator(isoperator), Precedence(prec) {}
230 
isUnaryOp() const231   bool isUnaryOp() const { return isOperator && Args.size() == 1; }
isBinaryOp() const232   bool isBinaryOp() const { return isOperator && Args.size() == 2; }
233 
getOperatorName() const234   char getOperatorName() const {
235     assert(isUnaryOp() || isBinaryOp());
236     return Name[Name.size()-1];
237   }
238 
getBinaryPrecedence() const239   unsigned getBinaryPrecedence() const { return Precedence; }
240 
241   Function *Codegen();
242 
243   void CreateArgumentAllocas(Function *F);
244 };
245 
246 /// FunctionAST - This class represents a function definition itself.
247 class FunctionAST {
248   PrototypeAST *Proto;
249   ExprAST *Body;
250 public:
FunctionAST(PrototypeAST * proto,ExprAST * body)251   FunctionAST(PrototypeAST *proto, ExprAST *body)
252     : Proto(proto), Body(body) {}
253 
254   Function *Codegen();
255 };
256 
257 //===----------------------------------------------------------------------===//
258 // Parser
259 //===----------------------------------------------------------------------===//
260 
261 /// CurTok/getNextToken - Provide a simple token buffer.  CurTok is the current
262 /// token the parser is looking at.  getNextToken reads another token from the
263 /// lexer and updates CurTok with its results.
264 static int CurTok;
getNextToken()265 static int getNextToken() {
266   return CurTok = gettok();
267 }
268 
269 /// BinopPrecedence - This holds the precedence for each binary operator that is
270 /// defined.
271 static std::map<char, int> BinopPrecedence;
272 
273 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
GetTokPrecedence()274 static int GetTokPrecedence() {
275   if (!isascii(CurTok))
276     return -1;
277 
278   // Make sure it's a declared binop.
279   int TokPrec = BinopPrecedence[CurTok];
280   if (TokPrec <= 0) return -1;
281   return TokPrec;
282 }
283 
284 /// Error* - These are little helper functions for error handling.
Error(const char * Str)285 ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;}
ErrorP(const char * Str)286 PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; }
ErrorF(const char * Str)287 FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; }
288 
289 static ExprAST *ParseExpression();
290 
291 /// identifierexpr
292 ///   ::= identifier
293 ///   ::= identifier '(' expression* ')'
ParseIdentifierExpr()294 static ExprAST *ParseIdentifierExpr() {
295   std::string IdName = IdentifierStr;
296 
297   getNextToken();  // eat identifier.
298 
299   if (CurTok != '(') // Simple variable ref.
300     return new VariableExprAST(IdName);
301 
302   // Call.
303   getNextToken();  // eat (
304   std::vector<ExprAST*> Args;
305   if (CurTok != ')') {
306     while (1) {
307       ExprAST *Arg = ParseExpression();
308       if (!Arg) return 0;
309       Args.push_back(Arg);
310 
311       if (CurTok == ')') break;
312 
313       if (CurTok != ',')
314         return Error("Expected ')' or ',' in argument list");
315       getNextToken();
316     }
317   }
318 
319   // Eat the ')'.
320   getNextToken();
321 
322   return new CallExprAST(IdName, Args);
323 }
324 
325 /// numberexpr ::= number
ParseNumberExpr()326 static ExprAST *ParseNumberExpr() {
327   ExprAST *Result = new NumberExprAST(NumVal);
328   getNextToken(); // consume the number
329   return Result;
330 }
331 
332 /// parenexpr ::= '(' expression ')'
ParseParenExpr()333 static ExprAST *ParseParenExpr() {
334   getNextToken();  // eat (.
335   ExprAST *V = ParseExpression();
336   if (!V) return 0;
337 
338   if (CurTok != ')')
339     return Error("expected ')'");
340   getNextToken();  // eat ).
341   return V;
342 }
343 
344 /// ifexpr ::= 'if' expression 'then' expression 'else' expression
ParseIfExpr()345 static ExprAST *ParseIfExpr() {
346   getNextToken();  // eat the if.
347 
348   // condition.
349   ExprAST *Cond = ParseExpression();
350   if (!Cond) return 0;
351 
352   if (CurTok != tok_then)
353     return Error("expected then");
354   getNextToken();  // eat the then
355 
356   ExprAST *Then = ParseExpression();
357   if (Then == 0) return 0;
358 
359   if (CurTok != tok_else)
360     return Error("expected else");
361 
362   getNextToken();
363 
364   ExprAST *Else = ParseExpression();
365   if (!Else) return 0;
366 
367   return new IfExprAST(Cond, Then, Else);
368 }
369 
370 /// forexpr ::= 'for' identifier '=' expr ',' expr (',' expr)? 'in' expression
ParseForExpr()371 static ExprAST *ParseForExpr() {
372   getNextToken();  // eat the for.
373 
374   if (CurTok != tok_identifier)
375     return Error("expected identifier after for");
376 
377   std::string IdName = IdentifierStr;
378   getNextToken();  // eat identifier.
379 
380   if (CurTok != '=')
381     return Error("expected '=' after for");
382   getNextToken();  // eat '='.
383 
384 
385   ExprAST *Start = ParseExpression();
386   if (Start == 0) return 0;
387   if (CurTok != ',')
388     return Error("expected ',' after for start value");
389   getNextToken();
390 
391   ExprAST *End = ParseExpression();
392   if (End == 0) return 0;
393 
394   // The step value is optional.
395   ExprAST *Step = 0;
396   if (CurTok == ',') {
397     getNextToken();
398     Step = ParseExpression();
399     if (Step == 0) return 0;
400   }
401 
402   if (CurTok != tok_in)
403     return Error("expected 'in' after for");
404   getNextToken();  // eat 'in'.
405 
406   ExprAST *Body = ParseExpression();
407   if (Body == 0) return 0;
408 
409   return new ForExprAST(IdName, Start, End, Step, Body);
410 }
411 
412 /// varexpr ::= 'var' identifier ('=' expression)?
413 //                    (',' identifier ('=' expression)?)* 'in' expression
ParseVarExpr()414 static ExprAST *ParseVarExpr() {
415   getNextToken();  // eat the var.
416 
417   std::vector<std::pair<std::string, ExprAST*> > VarNames;
418 
419   // At least one variable name is required.
420   if (CurTok != tok_identifier)
421     return Error("expected identifier after var");
422 
423   while (1) {
424     std::string Name = IdentifierStr;
425     getNextToken();  // eat identifier.
426 
427     // Read the optional initializer.
428     ExprAST *Init = 0;
429     if (CurTok == '=') {
430       getNextToken(); // eat the '='.
431 
432       Init = ParseExpression();
433       if (Init == 0) return 0;
434     }
435 
436     VarNames.push_back(std::make_pair(Name, Init));
437 
438     // End of var list, exit loop.
439     if (CurTok != ',') break;
440     getNextToken(); // eat the ','.
441 
442     if (CurTok != tok_identifier)
443       return Error("expected identifier list after var");
444   }
445 
446   // At this point, we have to have 'in'.
447   if (CurTok != tok_in)
448     return Error("expected 'in' keyword after 'var'");
449   getNextToken();  // eat 'in'.
450 
451   ExprAST *Body = ParseExpression();
452   if (Body == 0) return 0;
453 
454   return new VarExprAST(VarNames, Body);
455 }
456 
457 /// primary
458 ///   ::= identifierexpr
459 ///   ::= numberexpr
460 ///   ::= parenexpr
461 ///   ::= ifexpr
462 ///   ::= forexpr
463 ///   ::= varexpr
ParsePrimary()464 static ExprAST *ParsePrimary() {
465   switch (CurTok) {
466   default: return Error("unknown token when expecting an expression");
467   case tok_identifier: return ParseIdentifierExpr();
468   case tok_number:     return ParseNumberExpr();
469   case '(':            return ParseParenExpr();
470   case tok_if:         return ParseIfExpr();
471   case tok_for:        return ParseForExpr();
472   case tok_var:        return ParseVarExpr();
473   }
474 }
475 
476 /// unary
477 ///   ::= primary
478 ///   ::= '!' unary
ParseUnary()479 static ExprAST *ParseUnary() {
480   // If the current token is not an operator, it must be a primary expr.
481   if (!isascii(CurTok) || CurTok == '(' || CurTok == ',')
482     return ParsePrimary();
483 
484   // If this is a unary operator, read it.
485   int Opc = CurTok;
486   getNextToken();
487   if (ExprAST *Operand = ParseUnary())
488     return new UnaryExprAST(Opc, Operand);
489   return 0;
490 }
491 
492 /// binoprhs
493 ///   ::= ('+' unary)*
ParseBinOpRHS(int ExprPrec,ExprAST * LHS)494 static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) {
495   // If this is a binop, find its precedence.
496   while (1) {
497     int TokPrec = GetTokPrecedence();
498 
499     // If this is a binop that binds at least as tightly as the current binop,
500     // consume it, otherwise we are done.
501     if (TokPrec < ExprPrec)
502       return LHS;
503 
504     // Okay, we know this is a binop.
505     int BinOp = CurTok;
506     getNextToken();  // eat binop
507 
508     // Parse the unary expression after the binary operator.
509     ExprAST *RHS = ParseUnary();
510     if (!RHS) return 0;
511 
512     // If BinOp binds less tightly with RHS than the operator after RHS, let
513     // the pending operator take RHS as its LHS.
514     int NextPrec = GetTokPrecedence();
515     if (TokPrec < NextPrec) {
516       RHS = ParseBinOpRHS(TokPrec+1, RHS);
517       if (RHS == 0) return 0;
518     }
519 
520     // Merge LHS/RHS.
521     LHS = new BinaryExprAST(BinOp, LHS, RHS);
522   }
523 }
524 
525 /// expression
526 ///   ::= unary binoprhs
527 ///
ParseExpression()528 static ExprAST *ParseExpression() {
529   ExprAST *LHS = ParseUnary();
530   if (!LHS) return 0;
531 
532   return ParseBinOpRHS(0, LHS);
533 }
534 
535 /// prototype
536 ///   ::= id '(' id* ')'
537 ///   ::= binary LETTER number? (id, id)
538 ///   ::= unary LETTER (id)
ParsePrototype()539 static PrototypeAST *ParsePrototype() {
540   std::string FnName;
541 
542   unsigned Kind = 0; // 0 = identifier, 1 = unary, 2 = binary.
543   unsigned BinaryPrecedence = 30;
544 
545   switch (CurTok) {
546   default:
547     return ErrorP("Expected function name in prototype");
548   case tok_identifier:
549     FnName = IdentifierStr;
550     Kind = 0;
551     getNextToken();
552     break;
553   case tok_unary:
554     getNextToken();
555     if (!isascii(CurTok))
556       return ErrorP("Expected unary operator");
557     FnName = "unary";
558     FnName += (char)CurTok;
559     Kind = 1;
560     getNextToken();
561     break;
562   case tok_binary:
563     getNextToken();
564     if (!isascii(CurTok))
565       return ErrorP("Expected binary operator");
566     FnName = "binary";
567     FnName += (char)CurTok;
568     Kind = 2;
569     getNextToken();
570 
571     // Read the precedence if present.
572     if (CurTok == tok_number) {
573       if (NumVal < 1 || NumVal > 100)
574         return ErrorP("Invalid precedecnce: must be 1..100");
575       BinaryPrecedence = (unsigned)NumVal;
576       getNextToken();
577     }
578     break;
579   }
580 
581   if (CurTok != '(')
582     return ErrorP("Expected '(' in prototype");
583 
584   std::vector<std::string> ArgNames;
585   while (getNextToken() == tok_identifier)
586     ArgNames.push_back(IdentifierStr);
587   if (CurTok != ')')
588     return ErrorP("Expected ')' in prototype");
589 
590   // success.
591   getNextToken();  // eat ')'.
592 
593   // Verify right number of names for operator.
594   if (Kind && ArgNames.size() != Kind)
595     return ErrorP("Invalid number of operands for operator");
596 
597   return new PrototypeAST(FnName, ArgNames, Kind != 0, BinaryPrecedence);
598 }
599 
600 /// definition ::= 'def' prototype expression
ParseDefinition()601 static FunctionAST *ParseDefinition() {
602   getNextToken();  // eat def.
603   PrototypeAST *Proto = ParsePrototype();
604   if (Proto == 0) return 0;
605 
606   if (ExprAST *E = ParseExpression())
607     return new FunctionAST(Proto, E);
608   return 0;
609 }
610 
611 /// toplevelexpr ::= expression
ParseTopLevelExpr()612 static FunctionAST *ParseTopLevelExpr() {
613   if (ExprAST *E = ParseExpression()) {
614     // Make an anonymous proto.
615     PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>());
616     return new FunctionAST(Proto, E);
617   }
618   return 0;
619 }
620 
621 /// external ::= 'extern' prototype
ParseExtern()622 static PrototypeAST *ParseExtern() {
623   getNextToken();  // eat extern.
624   return ParsePrototype();
625 }
626 
627 //===----------------------------------------------------------------------===//
628 // Quick and dirty hack
629 //===----------------------------------------------------------------------===//
630 
631 // FIXME: Obviously we can do better than this
GenerateUniqueName(const char * root)632 std::string GenerateUniqueName(const char *root)
633 {
634   static int i = 0;
635   char s[16];
636   sprintf(s, "%s%d", root, i++);
637   std::string S = s;
638   return S;
639 }
640 
MakeLegalFunctionName(std::string Name)641 std::string MakeLegalFunctionName(std::string Name)
642 {
643   std::string NewName;
644   if (!Name.length())
645       return GenerateUniqueName("anon_func_");
646 
647   // Start with what we have
648   NewName = Name;
649 
650   // Look for a numberic first character
651   if (NewName.find_first_of("0123456789") == 0) {
652     NewName.insert(0, 1, 'n');
653   }
654 
655   // Replace illegal characters with their ASCII equivalent
656   std::string legal_elements = "_abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789";
657   size_t pos;
658   while ((pos = NewName.find_first_not_of(legal_elements)) != std::string::npos) {
659     char old_c = NewName.at(pos);
660     char new_str[16];
661     sprintf(new_str, "%d", (int)old_c);
662     NewName = NewName.replace(pos, 1, new_str);
663   }
664 
665   return NewName;
666 }
667 
668 //===----------------------------------------------------------------------===//
669 // MCJIT object cache class
670 //===----------------------------------------------------------------------===//
671 
672 class MCJITObjectCache : public ObjectCache {
673 public:
MCJITObjectCache()674   MCJITObjectCache() {
675     // Set IR cache directory
676     sys::fs::current_path(CacheDir);
677     sys::path::append(CacheDir, "toy_object_cache");
678   }
679 
~MCJITObjectCache()680   virtual ~MCJITObjectCache() {
681   }
682 
notifyObjectCompiled(const Module * M,const MemoryBuffer * Obj)683   virtual void notifyObjectCompiled(const Module *M, const MemoryBuffer *Obj) {
684     // Get the ModuleID
685     const std::string ModuleID = M->getModuleIdentifier();
686 
687     // If we've flagged this as an IR file, cache it
688     if (0 == ModuleID.compare(0, 3, "IR:")) {
689       std::string IRFileName = ModuleID.substr(3);
690       SmallString<128>IRCacheFile = CacheDir;
691       sys::path::append(IRCacheFile, IRFileName);
692       if (!sys::fs::exists(CacheDir.str()) && sys::fs::create_directory(CacheDir.str())) {
693         fprintf(stderr, "Unable to create cache directory\n");
694         return;
695       }
696       std::string ErrStr;
697       raw_fd_ostream IRObjectFile(IRCacheFile.c_str(), ErrStr, raw_fd_ostream::F_Binary);
698       IRObjectFile << Obj->getBuffer();
699     }
700   }
701 
702   // MCJIT will call this function before compiling any module
703   // MCJIT takes ownership of both the MemoryBuffer object and the memory
704   // to which it refers.
getObject(const Module * M)705   virtual MemoryBuffer* getObject(const Module* M) {
706     // Get the ModuleID
707     const std::string ModuleID = M->getModuleIdentifier();
708 
709     // If we've flagged this as an IR file, cache it
710     if (0 == ModuleID.compare(0, 3, "IR:")) {
711       std::string IRFileName = ModuleID.substr(3);
712       SmallString<128> IRCacheFile = CacheDir;
713       sys::path::append(IRCacheFile, IRFileName);
714       if (!sys::fs::exists(IRCacheFile.str())) {
715         // This file isn't in our cache
716         return NULL;
717       }
718       std::unique_ptr<MemoryBuffer> IRObjectBuffer;
719       MemoryBuffer::getFile(IRCacheFile.c_str(), IRObjectBuffer, -1, false);
720       // MCJIT will want to write into this buffer, and we don't want that
721       // because the file has probably just been mmapped.  Instead we make
722       // a copy.  The filed-based buffer will be released when it goes
723       // out of scope.
724       return MemoryBuffer::getMemBufferCopy(IRObjectBuffer->getBuffer());
725     }
726 
727     return NULL;
728   }
729 
730 private:
731   SmallString<128> CacheDir;
732 };
733 
734 //===----------------------------------------------------------------------===//
735 // MCJIT helper class
736 //===----------------------------------------------------------------------===//
737 
738 class MCJITHelper
739 {
740 public:
MCJITHelper(LLVMContext & C)741   MCJITHelper(LLVMContext& C) : Context(C), OpenModule(NULL) {}
742   ~MCJITHelper();
743 
744   Function *getFunction(const std::string FnName);
745   Module *getModuleForNewFunction();
746   void *getPointerToFunction(Function* F);
747   void *getPointerToNamedFunction(const std::string &Name);
748   ExecutionEngine *compileModule(Module *M);
749   void closeCurrentModule();
750   void addModule(Module *M);
751   void dump();
752 
753 private:
754   typedef std::vector<Module*> ModuleVector;
755 
756   LLVMContext  &Context;
757   Module       *OpenModule;
758   ModuleVector  Modules;
759   std::map<Module *, ExecutionEngine *> EngineMap;
760   MCJITObjectCache OurObjectCache;
761 };
762 
763 class HelpingMemoryManager : public SectionMemoryManager
764 {
765   HelpingMemoryManager(const HelpingMemoryManager&) = delete;
766   void operator=(const HelpingMemoryManager&) = delete;
767 
768 public:
HelpingMemoryManager(MCJITHelper * Helper)769   HelpingMemoryManager(MCJITHelper *Helper) : MasterHelper(Helper) {}
~HelpingMemoryManager()770   virtual ~HelpingMemoryManager() {}
771 
772   /// This method returns the address of the specified function.
773   /// Our implementation will attempt to find functions in other
774   /// modules associated with the MCJITHelper to cross link functions
775   /// from one generated module to another.
776   ///
777   /// If \p AbortOnFailure is false and no function with the given name is
778   /// found, this function returns a null pointer. Otherwise, it prints a
779   /// message to stderr and aborts.
780   virtual void *getPointerToNamedFunction(const std::string &Name,
781                                           bool AbortOnFailure = true);
782 private:
783   MCJITHelper *MasterHelper;
784 };
785 
getPointerToNamedFunction(const std::string & Name,bool AbortOnFailure)786 void *HelpingMemoryManager::getPointerToNamedFunction(const std::string &Name,
787                                         bool AbortOnFailure)
788 {
789   // Try the standard symbol resolution first, but ask it not to abort.
790   void *pfn = SectionMemoryManager::getPointerToNamedFunction(Name, false);
791   if (pfn)
792     return pfn;
793 
794   pfn = MasterHelper->getPointerToNamedFunction(Name);
795   if (!pfn && AbortOnFailure)
796     report_fatal_error("Program used external function '" + Name +
797                         "' which could not be resolved!");
798   return pfn;
799 }
800 
~MCJITHelper()801 MCJITHelper::~MCJITHelper()
802 {
803   // Walk the vector of modules.
804   ModuleVector::iterator it, end;
805   for (it = Modules.begin(), end = Modules.end();
806        it != end; ++it) {
807     // See if we have an execution engine for this module.
808     std::map<Module*, ExecutionEngine*>::iterator mapIt = EngineMap.find(*it);
809     // If we have an EE, the EE owns the module so just delete the EE.
810     if (mapIt != EngineMap.end()) {
811       delete mapIt->second;
812     } else {
813       // Otherwise, we still own the module.  Delete it now.
814       delete *it;
815     }
816   }
817 }
818 
getFunction(const std::string FnName)819 Function *MCJITHelper::getFunction(const std::string FnName) {
820   ModuleVector::iterator begin = Modules.begin();
821   ModuleVector::iterator end = Modules.end();
822   ModuleVector::iterator it;
823   for (it = begin; it != end; ++it) {
824     Function *F = (*it)->getFunction(FnName);
825     if (F) {
826       if (*it == OpenModule)
827           return F;
828 
829       assert(OpenModule != NULL);
830 
831       // This function is in a module that has already been JITed.
832       // We need to generate a new prototype for external linkage.
833       Function *PF = OpenModule->getFunction(FnName);
834       if (PF && !PF->empty()) {
835         ErrorF("redefinition of function across modules");
836         return 0;
837       }
838 
839       // If we don't have a prototype yet, create one.
840       if (!PF)
841         PF = Function::Create(F->getFunctionType(),
842                                       Function::ExternalLinkage,
843                                       FnName,
844                                       OpenModule);
845       return PF;
846     }
847   }
848   return NULL;
849 }
850 
getModuleForNewFunction()851 Module *MCJITHelper::getModuleForNewFunction() {
852   // If we have a Module that hasn't been JITed, use that.
853   if (OpenModule)
854     return OpenModule;
855 
856   // Otherwise create a new Module.
857   std::string ModName = GenerateUniqueName("mcjit_module_");
858   Module *M = new Module(ModName, Context);
859   Modules.push_back(M);
860   OpenModule = M;
861   return M;
862 }
863 
getPointerToFunction(Function * F)864 void *MCJITHelper::getPointerToFunction(Function* F) {
865   // Look for this function in an existing module
866   ModuleVector::iterator begin = Modules.begin();
867   ModuleVector::iterator end = Modules.end();
868   ModuleVector::iterator it;
869   std::string FnName = F->getName();
870   for (it = begin; it != end; ++it) {
871     Function *MF = (*it)->getFunction(FnName);
872     if (MF == F) {
873       std::map<Module*, ExecutionEngine*>::iterator eeIt = EngineMap.find(*it);
874       if (eeIt != EngineMap.end()) {
875         void *P = eeIt->second->getPointerToFunction(F);
876         if (P)
877           return P;
878       } else {
879         ExecutionEngine *EE = compileModule(*it);
880         void *P = EE->getPointerToFunction(F);
881         if (P)
882           return P;
883       }
884     }
885   }
886   return NULL;
887 }
888 
closeCurrentModule()889 void MCJITHelper::closeCurrentModule() {
890   OpenModule = NULL;
891 }
892 
compileModule(Module * M)893 ExecutionEngine *MCJITHelper::compileModule(Module *M) {
894   if (M == OpenModule)
895     closeCurrentModule();
896 
897   std::string ErrStr;
898   ExecutionEngine *NewEngine = EngineBuilder(M)
899                                             .setErrorStr(&ErrStr)
900                                             .setMCJITMemoryManager(new HelpingMemoryManager(this))
901                                             .create();
902   if (!NewEngine) {
903     fprintf(stderr, "Could not create ExecutionEngine: %s\n", ErrStr.c_str());
904     exit(1);
905   }
906 
907   if (UseObjectCache)
908     NewEngine->setObjectCache(&OurObjectCache);
909 
910   // Get the ModuleID so we can identify IR input files
911   const std::string ModuleID = M->getModuleIdentifier();
912 
913   // If we've flagged this as an IR file, it doesn't need function passes run.
914   if (0 != ModuleID.compare(0, 3, "IR:")) {
915     // Create a function pass manager for this engine
916     FunctionPassManager *FPM = new FunctionPassManager(M);
917 
918     // Set up the optimizer pipeline.  Start with registering info about how the
919     // target lays out data structures.
920     FPM->add(new DataLayout(*NewEngine->getDataLayout()));
921     // Provide basic AliasAnalysis support for GVN.
922     FPM->add(createBasicAliasAnalysisPass());
923     // Promote allocas to registers.
924     FPM->add(createPromoteMemoryToRegisterPass());
925     // Do simple "peephole" optimizations and bit-twiddling optzns.
926     FPM->add(createInstructionCombiningPass());
927     // Reassociate expressions.
928     FPM->add(createReassociatePass());
929     // Eliminate Common SubExpressions.
930     FPM->add(createGVNPass());
931     // Simplify the control flow graph (deleting unreachable blocks, etc).
932     FPM->add(createCFGSimplificationPass());
933     FPM->doInitialization();
934 
935     // For each function in the module
936     Module::iterator it;
937     Module::iterator end = M->end();
938     for (it = M->begin(); it != end; ++it) {
939       // Run the FPM on this function
940       FPM->run(*it);
941     }
942 
943     // We don't need this anymore
944     delete FPM;
945   }
946 
947   // Store this engine
948   EngineMap[M] = NewEngine;
949   NewEngine->finalizeObject();
950 
951   return NewEngine;
952 }
953 
getPointerToNamedFunction(const std::string & Name)954 void *MCJITHelper::getPointerToNamedFunction(const std::string &Name)
955 {
956   // Look for the functions in our modules, compiling only as necessary
957   ModuleVector::iterator begin = Modules.begin();
958   ModuleVector::iterator end = Modules.end();
959   ModuleVector::iterator it;
960   for (it = begin; it != end; ++it) {
961     Function *F = (*it)->getFunction(Name);
962     if (F && !F->empty()) {
963       std::map<Module*, ExecutionEngine*>::iterator eeIt = EngineMap.find(*it);
964       if (eeIt != EngineMap.end()) {
965         void *P = eeIt->second->getPointerToFunction(F);
966         if (P)
967           return P;
968       } else {
969         ExecutionEngine *EE = compileModule(*it);
970         void *P = EE->getPointerToFunction(F);
971         if (P)
972           return P;
973       }
974     }
975   }
976   return NULL;
977 }
978 
addModule(Module * M)979 void MCJITHelper::addModule(Module* M) {
980   Modules.push_back(M);
981 }
982 
dump()983 void MCJITHelper::dump()
984 {
985   ModuleVector::iterator begin = Modules.begin();
986   ModuleVector::iterator end = Modules.end();
987   ModuleVector::iterator it;
988   for (it = begin; it != end; ++it)
989     (*it)->dump();
990 }
991 
992 //===----------------------------------------------------------------------===//
993 // Code Generation
994 //===----------------------------------------------------------------------===//
995 
996 static MCJITHelper *TheHelper;
997 static LLVMContext TheContext;
998 static IRBuilder<> Builder(TheContext);
999 static std::map<std::string, AllocaInst*> NamedValues;
1000 
ErrorV(const char * Str)1001 Value *ErrorV(const char *Str) { Error(Str); return 0; }
1002 
1003 /// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of
1004 /// the function.  This is used for mutable variables etc.
CreateEntryBlockAlloca(Function * TheFunction,const std::string & VarName)1005 static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction,
1006                                           const std::string &VarName) {
1007   IRBuilder<> TmpB(&TheFunction->getEntryBlock(),
1008                  TheFunction->getEntryBlock().begin());
1009   return TmpB.CreateAlloca(Type::getDoubleTy(TheContext), 0, VarName.c_str());
1010 }
1011 
Codegen()1012 Value *NumberExprAST::Codegen() {
1013   return ConstantFP::get(TheContext, APFloat(Val));
1014 }
1015 
Codegen()1016 Value *VariableExprAST::Codegen() {
1017   // Look this variable up in the function.
1018   Value *V = NamedValues[Name];
1019   char ErrStr[256];
1020   sprintf(ErrStr, "Unknown variable name %s", Name.c_str());
1021   if (V == 0) return ErrorV(ErrStr);
1022 
1023   // Load the value.
1024   return Builder.CreateLoad(V, Name.c_str());
1025 }
1026 
Codegen()1027 Value *UnaryExprAST::Codegen() {
1028   Value *OperandV = Operand->Codegen();
1029   if (OperandV == 0) return 0;
1030 
1031   Function *F = TheHelper->getFunction(MakeLegalFunctionName(std::string("unary")+Opcode));
1032   if (F == 0)
1033     return ErrorV("Unknown unary operator");
1034 
1035   return Builder.CreateCall(F, OperandV, "unop");
1036 }
1037 
Codegen()1038 Value *BinaryExprAST::Codegen() {
1039   // Special case '=' because we don't want to emit the LHS as an expression.
1040   if (Op == '=') {
1041     // Assignment requires the LHS to be an identifier.
1042     VariableExprAST *LHSE = static_cast<VariableExprAST*>(LHS);
1043     if (!LHSE)
1044       return ErrorV("destination of '=' must be a variable");
1045     // Codegen the RHS.
1046     Value *Val = RHS->Codegen();
1047     if (Val == 0) return 0;
1048 
1049     // Look up the name.
1050     Value *Variable = NamedValues[LHSE->getName()];
1051     if (Variable == 0) return ErrorV("Unknown variable name");
1052 
1053     Builder.CreateStore(Val, Variable);
1054     return Val;
1055   }
1056 
1057   Value *L = LHS->Codegen();
1058   Value *R = RHS->Codegen();
1059   if (L == 0 || R == 0) return 0;
1060 
1061   switch (Op) {
1062   case '+': return Builder.CreateFAdd(L, R, "addtmp");
1063   case '-': return Builder.CreateFSub(L, R, "subtmp");
1064   case '*': return Builder.CreateFMul(L, R, "multmp");
1065   case '/': return Builder.CreateFDiv(L, R, "divtmp");
1066   case '<':
1067     L = Builder.CreateFCmpULT(L, R, "cmptmp");
1068     // Convert bool 0/1 to double 0.0 or 1.0
1069     return Builder.CreateUIToFP(L, Type::getDoubleTy(TheContext), "booltmp");
1070   default: break;
1071   }
1072 
1073   // If it wasn't a builtin binary operator, it must be a user defined one. Emit
1074   // a call to it.
1075   Function *F = TheHelper->getFunction(MakeLegalFunctionName(std::string("binary")+Op));
1076   assert(F && "binary operator not found!");
1077 
1078   Value *Ops[] = { L, R };
1079   return Builder.CreateCall(F, Ops, "binop");
1080 }
1081 
Codegen()1082 Value *CallExprAST::Codegen() {
1083   // Look up the name in the global module table.
1084   Function *CalleeF = TheHelper->getFunction(Callee);
1085   if (CalleeF == 0)
1086     return ErrorV("Unknown function referenced");
1087 
1088   // If argument mismatch error.
1089   if (CalleeF->arg_size() != Args.size())
1090     return ErrorV("Incorrect # arguments passed");
1091 
1092   std::vector<Value*> ArgsV;
1093   for (unsigned i = 0, e = Args.size(); i != e; ++i) {
1094     ArgsV.push_back(Args[i]->Codegen());
1095     if (ArgsV.back() == 0) return 0;
1096   }
1097 
1098   return Builder.CreateCall(CalleeF, ArgsV, "calltmp");
1099 }
1100 
Codegen()1101 Value *IfExprAST::Codegen() {
1102   Value *CondV = Cond->Codegen();
1103   if (CondV == 0) return 0;
1104 
1105   // Convert condition to a bool by comparing equal to 0.0.
1106   CondV = Builder.CreateFCmpONE(
1107       CondV, ConstantFP::get(TheContext, APFloat(0.0)), "ifcond");
1108 
1109   Function *TheFunction = Builder.GetInsertBlock()->getParent();
1110 
1111   // Create blocks for the then and else cases.  Insert the 'then' block at the
1112   // end of the function.
1113   BasicBlock *ThenBB = BasicBlock::Create(TheContext, "then", TheFunction);
1114   BasicBlock *ElseBB = BasicBlock::Create(TheContext, "else");
1115   BasicBlock *MergeBB = BasicBlock::Create(TheContext, "ifcont");
1116 
1117   Builder.CreateCondBr(CondV, ThenBB, ElseBB);
1118 
1119   // Emit then value.
1120   Builder.SetInsertPoint(ThenBB);
1121 
1122   Value *ThenV = Then->Codegen();
1123   if (ThenV == 0) return 0;
1124 
1125   Builder.CreateBr(MergeBB);
1126   // Codegen of 'Then' can change the current block, update ThenBB for the PHI.
1127   ThenBB = Builder.GetInsertBlock();
1128 
1129   // Emit else block.
1130   TheFunction->getBasicBlockList().push_back(ElseBB);
1131   Builder.SetInsertPoint(ElseBB);
1132 
1133   Value *ElseV = Else->Codegen();
1134   if (ElseV == 0) return 0;
1135 
1136   Builder.CreateBr(MergeBB);
1137   // Codegen of 'Else' can change the current block, update ElseBB for the PHI.
1138   ElseBB = Builder.GetInsertBlock();
1139 
1140   // Emit merge block.
1141   TheFunction->getBasicBlockList().push_back(MergeBB);
1142   Builder.SetInsertPoint(MergeBB);
1143   PHINode *PN = Builder.CreatePHI(Type::getDoubleTy(TheContext), 2, "iftmp");
1144 
1145   PN->addIncoming(ThenV, ThenBB);
1146   PN->addIncoming(ElseV, ElseBB);
1147   return PN;
1148 }
1149 
Codegen()1150 Value *ForExprAST::Codegen() {
1151   // Output this as:
1152   //   var = alloca double
1153   //   ...
1154   //   start = startexpr
1155   //   store start -> var
1156   //   goto loop
1157   // loop:
1158   //   ...
1159   //   bodyexpr
1160   //   ...
1161   // loopend:
1162   //   step = stepexpr
1163   //   endcond = endexpr
1164   //
1165   //   curvar = load var
1166   //   nextvar = curvar + step
1167   //   store nextvar -> var
1168   //   br endcond, loop, endloop
1169   // outloop:
1170 
1171   Function *TheFunction = Builder.GetInsertBlock()->getParent();
1172 
1173   // Create an alloca for the variable in the entry block.
1174   AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1175 
1176   // Emit the start code first, without 'variable' in scope.
1177   Value *StartVal = Start->Codegen();
1178   if (StartVal == 0) return 0;
1179 
1180   // Store the value into the alloca.
1181   Builder.CreateStore(StartVal, Alloca);
1182 
1183   // Make the new basic block for the loop header, inserting after current
1184   // block.
1185   BasicBlock *LoopBB = BasicBlock::Create(TheContext, "loop", TheFunction);
1186 
1187   // Insert an explicit fall through from the current block to the LoopBB.
1188   Builder.CreateBr(LoopBB);
1189 
1190   // Start insertion in LoopBB.
1191   Builder.SetInsertPoint(LoopBB);
1192 
1193   // Within the loop, the variable is defined equal to the PHI node.  If it
1194   // shadows an existing variable, we have to restore it, so save it now.
1195   AllocaInst *OldVal = NamedValues[VarName];
1196   NamedValues[VarName] = Alloca;
1197 
1198   // Emit the body of the loop.  This, like any other expr, can change the
1199   // current BB.  Note that we ignore the value computed by the body, but don't
1200   // allow an error.
1201   if (Body->Codegen() == 0)
1202     return 0;
1203 
1204   // Emit the step value.
1205   Value *StepVal;
1206   if (Step) {
1207     StepVal = Step->Codegen();
1208     if (StepVal == 0) return 0;
1209   } else {
1210     // If not specified, use 1.0.
1211     StepVal = ConstantFP::get(TheContext, APFloat(1.0));
1212   }
1213 
1214   // Compute the end condition.
1215   Value *EndCond = End->Codegen();
1216   if (EndCond == 0) return EndCond;
1217 
1218   // Reload, increment, and restore the alloca.  This handles the case where
1219   // the body of the loop mutates the variable.
1220   Value *CurVar = Builder.CreateLoad(Alloca, VarName.c_str());
1221   Value *NextVar = Builder.CreateFAdd(CurVar, StepVal, "nextvar");
1222   Builder.CreateStore(NextVar, Alloca);
1223 
1224   // Convert condition to a bool by comparing equal to 0.0.
1225   EndCond = Builder.CreateFCmpONE(
1226       EndCond, ConstantFP::get(TheContext, APFloat(0.0)), "loopcond");
1227 
1228   // Create the "after loop" block and insert it.
1229   BasicBlock *AfterBB =
1230       BasicBlock::Create(TheContext, "afterloop", TheFunction);
1231 
1232   // Insert the conditional branch into the end of LoopEndBB.
1233   Builder.CreateCondBr(EndCond, LoopBB, AfterBB);
1234 
1235   // Any new code will be inserted in AfterBB.
1236   Builder.SetInsertPoint(AfterBB);
1237 
1238   // Restore the unshadowed variable.
1239   if (OldVal)
1240     NamedValues[VarName] = OldVal;
1241   else
1242     NamedValues.erase(VarName);
1243 
1244 
1245   // for expr always returns 0.0.
1246   return Constant::getNullValue(Type::getDoubleTy(TheContext));
1247 }
1248 
Codegen()1249 Value *VarExprAST::Codegen() {
1250   std::vector<AllocaInst *> OldBindings;
1251 
1252   Function *TheFunction = Builder.GetInsertBlock()->getParent();
1253 
1254   // Register all variables and emit their initializer.
1255   for (unsigned i = 0, e = VarNames.size(); i != e; ++i) {
1256     const std::string &VarName = VarNames[i].first;
1257     ExprAST *Init = VarNames[i].second;
1258 
1259     // Emit the initializer before adding the variable to scope, this prevents
1260     // the initializer from referencing the variable itself, and permits stuff
1261     // like this:
1262     //  var a = 1 in
1263     //    var a = a in ...   # refers to outer 'a'.
1264     Value *InitVal;
1265     if (Init) {
1266       InitVal = Init->Codegen();
1267       if (InitVal == 0) return 0;
1268     } else { // If not specified, use 0.0.
1269       InitVal = ConstantFP::get(TheContext, APFloat(0.0));
1270     }
1271 
1272     AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1273     Builder.CreateStore(InitVal, Alloca);
1274 
1275     // Remember the old variable binding so that we can restore the binding when
1276     // we unrecurse.
1277     OldBindings.push_back(NamedValues[VarName]);
1278 
1279     // Remember this binding.
1280     NamedValues[VarName] = Alloca;
1281   }
1282 
1283   // Codegen the body, now that all vars are in scope.
1284   Value *BodyVal = Body->Codegen();
1285   if (BodyVal == 0) return 0;
1286 
1287   // Pop all our variables from scope.
1288   for (unsigned i = 0, e = VarNames.size(); i != e; ++i)
1289     NamedValues[VarNames[i].first] = OldBindings[i];
1290 
1291   // Return the body computation.
1292   return BodyVal;
1293 }
1294 
Codegen()1295 Function *PrototypeAST::Codegen() {
1296   // Make the function type:  double(double,double) etc.
1297   std::vector<Type *> Doubles(Args.size(), Type::getDoubleTy(TheContext));
1298   FunctionType *FT =
1299       FunctionType::get(Type::getDoubleTy(TheContext), Doubles, false);
1300 
1301   std::string FnName = MakeLegalFunctionName(Name);
1302 
1303   Module* M = TheHelper->getModuleForNewFunction();
1304 
1305   Function *F = Function::Create(FT, Function::ExternalLinkage, FnName, M);
1306 
1307   // If F conflicted, there was already something named 'FnName'.  If it has a
1308   // body, don't allow redefinition or reextern.
1309   if (F->getName() != FnName) {
1310     // Delete the one we just made and get the existing one.
1311     F->eraseFromParent();
1312     F = M->getFunction(Name);
1313 
1314     // If F already has a body, reject this.
1315     if (!F->empty()) {
1316       ErrorF("redefinition of function");
1317       return 0;
1318     }
1319 
1320     // If F took a different number of args, reject.
1321     if (F->arg_size() != Args.size()) {
1322       ErrorF("redefinition of function with different # args");
1323       return 0;
1324     }
1325   }
1326 
1327   // Set names for all arguments.
1328   unsigned Idx = 0;
1329   for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
1330        ++AI, ++Idx)
1331     AI->setName(Args[Idx]);
1332 
1333   return F;
1334 }
1335 
1336 /// CreateArgumentAllocas - Create an alloca for each argument and register the
1337 /// argument in the symbol table so that references to it will succeed.
CreateArgumentAllocas(Function * F)1338 void PrototypeAST::CreateArgumentAllocas(Function *F) {
1339   Function::arg_iterator AI = F->arg_begin();
1340   for (unsigned Idx = 0, e = Args.size(); Idx != e; ++Idx, ++AI) {
1341     // Create an alloca for this variable.
1342     AllocaInst *Alloca = CreateEntryBlockAlloca(F, Args[Idx]);
1343 
1344     // Store the initial value into the alloca.
1345     Builder.CreateStore(AI, Alloca);
1346 
1347     // Add arguments to variable symbol table.
1348     NamedValues[Args[Idx]] = Alloca;
1349   }
1350 }
1351 
Codegen()1352 Function *FunctionAST::Codegen() {
1353   NamedValues.clear();
1354 
1355   Function *TheFunction = Proto->Codegen();
1356   if (TheFunction == 0)
1357     return 0;
1358 
1359   // If this is an operator, install it.
1360   if (Proto->isBinaryOp())
1361     BinopPrecedence[Proto->getOperatorName()] = Proto->getBinaryPrecedence();
1362 
1363   // Create a new basic block to start insertion into.
1364   BasicBlock *BB = BasicBlock::Create(TheContext, "entry", TheFunction);
1365   Builder.SetInsertPoint(BB);
1366 
1367   // Add all arguments to the symbol table and create their allocas.
1368   Proto->CreateArgumentAllocas(TheFunction);
1369 
1370   if (Value *RetVal = Body->Codegen()) {
1371     // Finish off the function.
1372     Builder.CreateRet(RetVal);
1373 
1374     // Validate the generated code, checking for consistency.
1375     verifyFunction(*TheFunction);
1376 
1377     return TheFunction;
1378   }
1379 
1380   // Error reading body, remove function.
1381   TheFunction->eraseFromParent();
1382 
1383   if (Proto->isBinaryOp())
1384     BinopPrecedence.erase(Proto->getOperatorName());
1385   return 0;
1386 }
1387 
1388 //===----------------------------------------------------------------------===//
1389 // Top-Level parsing and JIT Driver
1390 //===----------------------------------------------------------------------===//
1391 
HandleDefinition()1392 static void HandleDefinition() {
1393   if (FunctionAST *F = ParseDefinition()) {
1394     TheHelper->closeCurrentModule();
1395     if (Function *LF = F->Codegen()) {
1396 #ifndef MINIMAL_STDERR_OUTPUT
1397       fprintf(stderr, "Read function definition:");
1398       LF->print(errs());
1399       fprintf(stderr, "\n");
1400 #endif
1401     }
1402   } else {
1403     // Skip token for error recovery.
1404     getNextToken();
1405   }
1406 }
1407 
HandleExtern()1408 static void HandleExtern() {
1409   if (PrototypeAST *P = ParseExtern()) {
1410     if (Function *F = P->Codegen()) {
1411 #ifndef MINIMAL_STDERR_OUTPUT
1412       fprintf(stderr, "Read extern: ");
1413       F->print(errs());
1414       fprintf(stderr, "\n");
1415 #endif
1416     }
1417   } else {
1418     // Skip token for error recovery.
1419     getNextToken();
1420   }
1421 }
1422 
HandleTopLevelExpression()1423 static void HandleTopLevelExpression() {
1424   // Evaluate a top-level expression into an anonymous function.
1425   if (FunctionAST *F = ParseTopLevelExpr()) {
1426     if (Function *LF = F->Codegen()) {
1427       // JIT the function, returning a function pointer.
1428       void *FPtr = TheHelper->getPointerToFunction(LF);
1429 
1430       // Cast it to the right type (takes no arguments, returns a double) so we
1431       // can call it as a native function.
1432       double (*FP)() = (double (*)())(intptr_t)FPtr;
1433 #ifdef MINIMAL_STDERR_OUTPUT
1434       FP();
1435 #else
1436       fprintf(stderr, "Evaluated to %f\n", FP());
1437 #endif
1438     }
1439   } else {
1440     // Skip token for error recovery.
1441     getNextToken();
1442   }
1443 }
1444 
1445 /// top ::= definition | external | expression | ';'
MainLoop()1446 static void MainLoop() {
1447   while (1) {
1448 #ifndef MINIMAL_STDERR_OUTPUT
1449     fprintf(stderr, "ready> ");
1450 #endif
1451     switch (CurTok) {
1452     case tok_eof:    return;
1453     case ';':        getNextToken(); break;  // ignore top-level semicolons.
1454     case tok_def:    HandleDefinition(); break;
1455     case tok_extern: HandleExtern(); break;
1456     default:         HandleTopLevelExpression(); break;
1457     }
1458   }
1459 }
1460 
1461 //===----------------------------------------------------------------------===//
1462 // "Library" functions that can be "extern'd" from user code.
1463 //===----------------------------------------------------------------------===//
1464 
1465 /// putchard - putchar that takes a double and returns 0.
1466 extern "C"
putchard(double X)1467 double putchard(double X) {
1468   putchar((char)X);
1469   return 0;
1470 }
1471 
1472 /// printd - printf that takes a double prints it as "%f\n", returning 0.
1473 extern "C"
printd(double X)1474 double printd(double X) {
1475   printf("%f", X);
1476   return 0;
1477 }
1478 
1479 extern "C"
printlf()1480 double printlf() {
1481   printf("\n");
1482   return 0;
1483 }
1484 
1485 //===----------------------------------------------------------------------===//
1486 // Command line input file handler
1487 //===----------------------------------------------------------------------===//
1488 
parseInputIR(std::string InputFile)1489 Module* parseInputIR(std::string InputFile) {
1490   SMDiagnostic Err;
1491   Module *M = ParseIRFile(InputFile, Err, TheContext);
1492   if (!M) {
1493     Err.print("IR parsing failed: ", errs());
1494     return NULL;
1495   }
1496 
1497   char ModID[256];
1498   sprintf(ModID, "IR:%s", InputFile.c_str());
1499   M->setModuleIdentifier(ModID);
1500 
1501   TheHelper->addModule(M);
1502   return M;
1503 }
1504 
1505 //===----------------------------------------------------------------------===//
1506 // Main driver code.
1507 //===----------------------------------------------------------------------===//
1508 
main(int argc,char ** argv)1509 int main(int argc, char **argv) {
1510   InitializeNativeTarget();
1511   InitializeNativeTargetAsmPrinter();
1512   InitializeNativeTargetAsmParser();
1513   LLVMContext &Context = TheContext;
1514 
1515   cl::ParseCommandLineOptions(argc, argv,
1516                               "Kaleidoscope example program\n");
1517 
1518   // Install standard binary operators.
1519   // 1 is lowest precedence.
1520   BinopPrecedence['='] = 2;
1521   BinopPrecedence['<'] = 10;
1522   BinopPrecedence['+'] = 20;
1523   BinopPrecedence['-'] = 20;
1524   BinopPrecedence['/'] = 40;
1525   BinopPrecedence['*'] = 40;  // highest.
1526 
1527   // Prime the first token.
1528 #ifndef MINIMAL_STDERR_OUTPUT
1529   fprintf(stderr, "ready> ");
1530 #endif
1531   getNextToken();
1532 
1533   // Make the helper, which holds all the code.
1534   TheHelper = new MCJITHelper(Context);
1535 
1536   if (!InputIR.empty()) {
1537     parseInputIR(InputIR);
1538   }
1539 
1540   // Run the main "interpreter loop" now.
1541   MainLoop();
1542 
1543 #ifndef MINIMAL_STDERR_OUTPUT
1544   // Print out all of the generated code.
1545   TheHelper->print(errs());
1546 #endif
1547 
1548   return 0;
1549 }
1550