1 #include "llvm/ADT/APFloat.h"
2 #include "llvm/ADT/STLExtras.h"
3 #include "llvm/ADT/SmallVector.h"
4 #include "llvm/Analysis/Passes.h"
5 #include "llvm/IR/IRBuilder.h"
6 #include "llvm/IR/LLVMContext.h"
7 #include "llvm/IR/LegacyPassManager.h"
8 #include "llvm/IR/Metadata.h"
9 #include "llvm/IR/Module.h"
10 #include "llvm/IR/Type.h"
11 #include "llvm/IR/Verifier.h"
12 #include "llvm/Support/FileSystem.h"
13 #include "llvm/Support/TargetRegistry.h"
14 #include "llvm/Support/TargetSelect.h"
15 #include "llvm/Target/TargetMachine.h"
16 #include "llvm/Target/TargetOptions.h"
17 #include "llvm/Transforms/Scalar.h"
18 #include <cctype>
19 #include <cstdio>
20 #include <cstdlib>
21 #include <map>
22 #include <memory>
23 #include <string>
24 #include <utility>
25 #include <vector>
26
27 using namespace llvm;
28 using namespace llvm::sys;
29
30 //===----------------------------------------------------------------------===//
31 // Lexer
32 //===----------------------------------------------------------------------===//
33
34 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
35 // of these for known things.
36 enum Token {
37 tok_eof = -1,
38
39 // commands
40 tok_def = -2,
41 tok_extern = -3,
42
43 // primary
44 tok_identifier = -4,
45 tok_number = -5,
46
47 // control
48 tok_if = -6,
49 tok_then = -7,
50 tok_else = -8,
51 tok_for = -9,
52 tok_in = -10,
53
54 // operators
55 tok_binary = -11,
56 tok_unary = -12,
57
58 // var definition
59 tok_var = -13
60 };
61
62 static std::string IdentifierStr; // Filled in if tok_identifier
63 static double NumVal; // Filled in if tok_number
64
65 /// gettok - Return the next token from standard input.
gettok()66 static int gettok() {
67 static int LastChar = ' ';
68
69 // Skip any whitespace.
70 while (isspace(LastChar))
71 LastChar = getchar();
72
73 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
74 IdentifierStr = LastChar;
75 while (isalnum((LastChar = getchar())))
76 IdentifierStr += LastChar;
77
78 if (IdentifierStr == "def")
79 return tok_def;
80 if (IdentifierStr == "extern")
81 return tok_extern;
82 if (IdentifierStr == "if")
83 return tok_if;
84 if (IdentifierStr == "then")
85 return tok_then;
86 if (IdentifierStr == "else")
87 return tok_else;
88 if (IdentifierStr == "for")
89 return tok_for;
90 if (IdentifierStr == "in")
91 return tok_in;
92 if (IdentifierStr == "binary")
93 return tok_binary;
94 if (IdentifierStr == "unary")
95 return tok_unary;
96 if (IdentifierStr == "var")
97 return tok_var;
98 return tok_identifier;
99 }
100
101 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+
102 std::string NumStr;
103 do {
104 NumStr += LastChar;
105 LastChar = getchar();
106 } while (isdigit(LastChar) || LastChar == '.');
107
108 NumVal = strtod(NumStr.c_str(), nullptr);
109 return tok_number;
110 }
111
112 if (LastChar == '#') {
113 // Comment until end of line.
114 do
115 LastChar = getchar();
116 while (LastChar != EOF && LastChar != '\n' && LastChar != '\r');
117
118 if (LastChar != EOF)
119 return gettok();
120 }
121
122 // Check for end of file. Don't eat the EOF.
123 if (LastChar == EOF)
124 return tok_eof;
125
126 // Otherwise, just return the character as its ascii value.
127 int ThisChar = LastChar;
128 LastChar = getchar();
129 return ThisChar;
130 }
131
132 //===----------------------------------------------------------------------===//
133 // Abstract Syntax Tree (aka Parse Tree)
134 //===----------------------------------------------------------------------===//
135 namespace {
136 /// ExprAST - Base class for all expression nodes.
137 class ExprAST {
138 public:
~ExprAST()139 virtual ~ExprAST() {}
140 virtual Value *codegen() = 0;
141 };
142
143 /// NumberExprAST - Expression class for numeric literals like "1.0".
144 class NumberExprAST : public ExprAST {
145 double Val;
146
147 public:
NumberExprAST(double Val)148 NumberExprAST(double Val) : Val(Val) {}
149 Value *codegen() override;
150 };
151
152 /// VariableExprAST - Expression class for referencing a variable, like "a".
153 class VariableExprAST : public ExprAST {
154 std::string Name;
155
156 public:
VariableExprAST(const std::string & Name)157 VariableExprAST(const std::string &Name) : Name(Name) {}
getName() const158 const std::string &getName() const { return Name; }
159 Value *codegen() override;
160 };
161
162 /// UnaryExprAST - Expression class for a unary operator.
163 class UnaryExprAST : public ExprAST {
164 char Opcode;
165 std::unique_ptr<ExprAST> Operand;
166
167 public:
UnaryExprAST(char Opcode,std::unique_ptr<ExprAST> Operand)168 UnaryExprAST(char Opcode, std::unique_ptr<ExprAST> Operand)
169 : Opcode(Opcode), Operand(std::move(Operand)) {}
170 Value *codegen() override;
171 };
172
173 /// BinaryExprAST - Expression class for a binary operator.
174 class BinaryExprAST : public ExprAST {
175 char Op;
176 std::unique_ptr<ExprAST> LHS, RHS;
177
178 public:
BinaryExprAST(char Op,std::unique_ptr<ExprAST> LHS,std::unique_ptr<ExprAST> RHS)179 BinaryExprAST(char Op, std::unique_ptr<ExprAST> LHS,
180 std::unique_ptr<ExprAST> RHS)
181 : Op(Op), LHS(std::move(LHS)), RHS(std::move(RHS)) {}
182 Value *codegen() override;
183 };
184
185 /// CallExprAST - Expression class for function calls.
186 class CallExprAST : public ExprAST {
187 std::string Callee;
188 std::vector<std::unique_ptr<ExprAST>> Args;
189
190 public:
CallExprAST(const std::string & Callee,std::vector<std::unique_ptr<ExprAST>> Args)191 CallExprAST(const std::string &Callee,
192 std::vector<std::unique_ptr<ExprAST>> Args)
193 : Callee(Callee), Args(std::move(Args)) {}
194 Value *codegen() override;
195 };
196
197 /// IfExprAST - Expression class for if/then/else.
198 class IfExprAST : public ExprAST {
199 std::unique_ptr<ExprAST> Cond, Then, Else;
200
201 public:
IfExprAST(std::unique_ptr<ExprAST> Cond,std::unique_ptr<ExprAST> Then,std::unique_ptr<ExprAST> Else)202 IfExprAST(std::unique_ptr<ExprAST> Cond, std::unique_ptr<ExprAST> Then,
203 std::unique_ptr<ExprAST> Else)
204 : Cond(std::move(Cond)), Then(std::move(Then)), Else(std::move(Else)) {}
205 Value *codegen() override;
206 };
207
208 /// ForExprAST - Expression class for for/in.
209 class ForExprAST : public ExprAST {
210 std::string VarName;
211 std::unique_ptr<ExprAST> Start, End, Step, Body;
212
213 public:
ForExprAST(const std::string & VarName,std::unique_ptr<ExprAST> Start,std::unique_ptr<ExprAST> End,std::unique_ptr<ExprAST> Step,std::unique_ptr<ExprAST> Body)214 ForExprAST(const std::string &VarName, std::unique_ptr<ExprAST> Start,
215 std::unique_ptr<ExprAST> End, std::unique_ptr<ExprAST> Step,
216 std::unique_ptr<ExprAST> Body)
217 : VarName(VarName), Start(std::move(Start)), End(std::move(End)),
218 Step(std::move(Step)), Body(std::move(Body)) {}
219 Value *codegen() override;
220 };
221
222 /// VarExprAST - Expression class for var/in
223 class VarExprAST : public ExprAST {
224 std::vector<std::pair<std::string, std::unique_ptr<ExprAST>>> VarNames;
225 std::unique_ptr<ExprAST> Body;
226
227 public:
VarExprAST(std::vector<std::pair<std::string,std::unique_ptr<ExprAST>>> VarNames,std::unique_ptr<ExprAST> Body)228 VarExprAST(
229 std::vector<std::pair<std::string, std::unique_ptr<ExprAST>>> VarNames,
230 std::unique_ptr<ExprAST> Body)
231 : VarNames(std::move(VarNames)), Body(std::move(Body)) {}
232 Value *codegen() override;
233 };
234
235 /// PrototypeAST - This class represents the "prototype" for a function,
236 /// which captures its name, and its argument names (thus implicitly the number
237 /// of arguments the function takes), as well as if it is an operator.
238 class PrototypeAST {
239 std::string Name;
240 std::vector<std::string> Args;
241 bool IsOperator;
242 unsigned Precedence; // Precedence if a binary op.
243
244 public:
PrototypeAST(const std::string & Name,std::vector<std::string> Args,bool IsOperator=false,unsigned Prec=0)245 PrototypeAST(const std::string &Name, std::vector<std::string> Args,
246 bool IsOperator = false, unsigned Prec = 0)
247 : Name(Name), Args(std::move(Args)), IsOperator(IsOperator),
248 Precedence(Prec) {}
249 Function *codegen();
getName() const250 const std::string &getName() const { return Name; }
251
isUnaryOp() const252 bool isUnaryOp() const { return IsOperator && Args.size() == 1; }
isBinaryOp() const253 bool isBinaryOp() const { return IsOperator && Args.size() == 2; }
254
getOperatorName() const255 char getOperatorName() const {
256 assert(isUnaryOp() || isBinaryOp());
257 return Name[Name.size() - 1];
258 }
259
getBinaryPrecedence() const260 unsigned getBinaryPrecedence() const { return Precedence; }
261 };
262
263 /// FunctionAST - This class represents a function definition itself.
264 class FunctionAST {
265 std::unique_ptr<PrototypeAST> Proto;
266 std::unique_ptr<ExprAST> Body;
267
268 public:
FunctionAST(std::unique_ptr<PrototypeAST> Proto,std::unique_ptr<ExprAST> Body)269 FunctionAST(std::unique_ptr<PrototypeAST> Proto,
270 std::unique_ptr<ExprAST> Body)
271 : Proto(std::move(Proto)), Body(std::move(Body)) {}
272 Function *codegen();
273 };
274 } // end anonymous namespace
275
276 //===----------------------------------------------------------------------===//
277 // Parser
278 //===----------------------------------------------------------------------===//
279
280 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
281 /// token the parser is looking at. getNextToken reads another token from the
282 /// lexer and updates CurTok with its results.
283 static int CurTok;
getNextToken()284 static int getNextToken() { return CurTok = gettok(); }
285
286 /// BinopPrecedence - This holds the precedence for each binary operator that is
287 /// defined.
288 static std::map<char, int> BinopPrecedence;
289
290 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
GetTokPrecedence()291 static int GetTokPrecedence() {
292 if (!isascii(CurTok))
293 return -1;
294
295 // Make sure it's a declared binop.
296 int TokPrec = BinopPrecedence[CurTok];
297 if (TokPrec <= 0)
298 return -1;
299 return TokPrec;
300 }
301
302 /// LogError* - These are little helper functions for error handling.
LogError(const char * Str)303 std::unique_ptr<ExprAST> LogError(const char *Str) {
304 fprintf(stderr, "Error: %s\n", Str);
305 return nullptr;
306 }
307
LogErrorP(const char * Str)308 std::unique_ptr<PrototypeAST> LogErrorP(const char *Str) {
309 LogError(Str);
310 return nullptr;
311 }
312
313 static std::unique_ptr<ExprAST> ParseExpression();
314
315 /// numberexpr ::= number
ParseNumberExpr()316 static std::unique_ptr<ExprAST> ParseNumberExpr() {
317 auto Result = llvm::make_unique<NumberExprAST>(NumVal);
318 getNextToken(); // consume the number
319 return std::move(Result);
320 }
321
322 /// parenexpr ::= '(' expression ')'
ParseParenExpr()323 static std::unique_ptr<ExprAST> ParseParenExpr() {
324 getNextToken(); // eat (.
325 auto V = ParseExpression();
326 if (!V)
327 return nullptr;
328
329 if (CurTok != ')')
330 return LogError("expected ')'");
331 getNextToken(); // eat ).
332 return V;
333 }
334
335 /// identifierexpr
336 /// ::= identifier
337 /// ::= identifier '(' expression* ')'
ParseIdentifierExpr()338 static std::unique_ptr<ExprAST> ParseIdentifierExpr() {
339 std::string IdName = IdentifierStr;
340
341 getNextToken(); // eat identifier.
342
343 if (CurTok != '(') // Simple variable ref.
344 return llvm::make_unique<VariableExprAST>(IdName);
345
346 // Call.
347 getNextToken(); // eat (
348 std::vector<std::unique_ptr<ExprAST>> Args;
349 if (CurTok != ')') {
350 while (true) {
351 if (auto Arg = ParseExpression())
352 Args.push_back(std::move(Arg));
353 else
354 return nullptr;
355
356 if (CurTok == ')')
357 break;
358
359 if (CurTok != ',')
360 return LogError("Expected ')' or ',' in argument list");
361 getNextToken();
362 }
363 }
364
365 // Eat the ')'.
366 getNextToken();
367
368 return llvm::make_unique<CallExprAST>(IdName, std::move(Args));
369 }
370
371 /// ifexpr ::= 'if' expression 'then' expression 'else' expression
ParseIfExpr()372 static std::unique_ptr<ExprAST> ParseIfExpr() {
373 getNextToken(); // eat the if.
374
375 // condition.
376 auto Cond = ParseExpression();
377 if (!Cond)
378 return nullptr;
379
380 if (CurTok != tok_then)
381 return LogError("expected then");
382 getNextToken(); // eat the then
383
384 auto Then = ParseExpression();
385 if (!Then)
386 return nullptr;
387
388 if (CurTok != tok_else)
389 return LogError("expected else");
390
391 getNextToken();
392
393 auto Else = ParseExpression();
394 if (!Else)
395 return nullptr;
396
397 return llvm::make_unique<IfExprAST>(std::move(Cond), std::move(Then),
398 std::move(Else));
399 }
400
401 /// forexpr ::= 'for' identifier '=' expr ',' expr (',' expr)? 'in' expression
ParseForExpr()402 static std::unique_ptr<ExprAST> ParseForExpr() {
403 getNextToken(); // eat the for.
404
405 if (CurTok != tok_identifier)
406 return LogError("expected identifier after for");
407
408 std::string IdName = IdentifierStr;
409 getNextToken(); // eat identifier.
410
411 if (CurTok != '=')
412 return LogError("expected '=' after for");
413 getNextToken(); // eat '='.
414
415 auto Start = ParseExpression();
416 if (!Start)
417 return nullptr;
418 if (CurTok != ',')
419 return LogError("expected ',' after for start value");
420 getNextToken();
421
422 auto End = ParseExpression();
423 if (!End)
424 return nullptr;
425
426 // The step value is optional.
427 std::unique_ptr<ExprAST> Step;
428 if (CurTok == ',') {
429 getNextToken();
430 Step = ParseExpression();
431 if (!Step)
432 return nullptr;
433 }
434
435 if (CurTok != tok_in)
436 return LogError("expected 'in' after for");
437 getNextToken(); // eat 'in'.
438
439 auto Body = ParseExpression();
440 if (!Body)
441 return nullptr;
442
443 return llvm::make_unique<ForExprAST>(IdName, std::move(Start), std::move(End),
444 std::move(Step), std::move(Body));
445 }
446
447 /// varexpr ::= 'var' identifier ('=' expression)?
448 // (',' identifier ('=' expression)?)* 'in' expression
ParseVarExpr()449 static std::unique_ptr<ExprAST> ParseVarExpr() {
450 getNextToken(); // eat the var.
451
452 std::vector<std::pair<std::string, std::unique_ptr<ExprAST>>> VarNames;
453
454 // At least one variable name is required.
455 if (CurTok != tok_identifier)
456 return LogError("expected identifier after var");
457
458 while (true) {
459 std::string Name = IdentifierStr;
460 getNextToken(); // eat identifier.
461
462 // Read the optional initializer.
463 std::unique_ptr<ExprAST> Init = nullptr;
464 if (CurTok == '=') {
465 getNextToken(); // eat the '='.
466
467 Init = ParseExpression();
468 if (!Init)
469 return nullptr;
470 }
471
472 VarNames.push_back(std::make_pair(Name, std::move(Init)));
473
474 // End of var list, exit loop.
475 if (CurTok != ',')
476 break;
477 getNextToken(); // eat the ','.
478
479 if (CurTok != tok_identifier)
480 return LogError("expected identifier list after var");
481 }
482
483 // At this point, we have to have 'in'.
484 if (CurTok != tok_in)
485 return LogError("expected 'in' keyword after 'var'");
486 getNextToken(); // eat 'in'.
487
488 auto Body = ParseExpression();
489 if (!Body)
490 return nullptr;
491
492 return llvm::make_unique<VarExprAST>(std::move(VarNames), std::move(Body));
493 }
494
495 /// primary
496 /// ::= identifierexpr
497 /// ::= numberexpr
498 /// ::= parenexpr
499 /// ::= ifexpr
500 /// ::= forexpr
501 /// ::= varexpr
ParsePrimary()502 static std::unique_ptr<ExprAST> ParsePrimary() {
503 switch (CurTok) {
504 default:
505 return LogError("unknown token when expecting an expression");
506 case tok_identifier:
507 return ParseIdentifierExpr();
508 case tok_number:
509 return ParseNumberExpr();
510 case '(':
511 return ParseParenExpr();
512 case tok_if:
513 return ParseIfExpr();
514 case tok_for:
515 return ParseForExpr();
516 case tok_var:
517 return ParseVarExpr();
518 }
519 }
520
521 /// unary
522 /// ::= primary
523 /// ::= '!' unary
ParseUnary()524 static std::unique_ptr<ExprAST> ParseUnary() {
525 // If the current token is not an operator, it must be a primary expr.
526 if (!isascii(CurTok) || CurTok == '(' || CurTok == ',')
527 return ParsePrimary();
528
529 // If this is a unary operator, read it.
530 int Opc = CurTok;
531 getNextToken();
532 if (auto Operand = ParseUnary())
533 return llvm::make_unique<UnaryExprAST>(Opc, std::move(Operand));
534 return nullptr;
535 }
536
537 /// binoprhs
538 /// ::= ('+' unary)*
ParseBinOpRHS(int ExprPrec,std::unique_ptr<ExprAST> LHS)539 static std::unique_ptr<ExprAST> ParseBinOpRHS(int ExprPrec,
540 std::unique_ptr<ExprAST> LHS) {
541 // If this is a binop, find its precedence.
542 while (true) {
543 int TokPrec = GetTokPrecedence();
544
545 // If this is a binop that binds at least as tightly as the current binop,
546 // consume it, otherwise we are done.
547 if (TokPrec < ExprPrec)
548 return LHS;
549
550 // Okay, we know this is a binop.
551 int BinOp = CurTok;
552 getNextToken(); // eat binop
553
554 // Parse the unary expression after the binary operator.
555 auto RHS = ParseUnary();
556 if (!RHS)
557 return nullptr;
558
559 // If BinOp binds less tightly with RHS than the operator after RHS, let
560 // the pending operator take RHS as its LHS.
561 int NextPrec = GetTokPrecedence();
562 if (TokPrec < NextPrec) {
563 RHS = ParseBinOpRHS(TokPrec + 1, std::move(RHS));
564 if (!RHS)
565 return nullptr;
566 }
567
568 // Merge LHS/RHS.
569 LHS =
570 llvm::make_unique<BinaryExprAST>(BinOp, std::move(LHS), std::move(RHS));
571 }
572 }
573
574 /// expression
575 /// ::= unary binoprhs
576 ///
ParseExpression()577 static std::unique_ptr<ExprAST> ParseExpression() {
578 auto LHS = ParseUnary();
579 if (!LHS)
580 return nullptr;
581
582 return ParseBinOpRHS(0, std::move(LHS));
583 }
584
585 /// prototype
586 /// ::= id '(' id* ')'
587 /// ::= binary LETTER number? (id, id)
588 /// ::= unary LETTER (id)
ParsePrototype()589 static std::unique_ptr<PrototypeAST> ParsePrototype() {
590 std::string FnName;
591
592 unsigned Kind = 0; // 0 = identifier, 1 = unary, 2 = binary.
593 unsigned BinaryPrecedence = 30;
594
595 switch (CurTok) {
596 default:
597 return LogErrorP("Expected function name in prototype");
598 case tok_identifier:
599 FnName = IdentifierStr;
600 Kind = 0;
601 getNextToken();
602 break;
603 case tok_unary:
604 getNextToken();
605 if (!isascii(CurTok))
606 return LogErrorP("Expected unary operator");
607 FnName = "unary";
608 FnName += (char)CurTok;
609 Kind = 1;
610 getNextToken();
611 break;
612 case tok_binary:
613 getNextToken();
614 if (!isascii(CurTok))
615 return LogErrorP("Expected binary operator");
616 FnName = "binary";
617 FnName += (char)CurTok;
618 Kind = 2;
619 getNextToken();
620
621 // Read the precedence if present.
622 if (CurTok == tok_number) {
623 if (NumVal < 1 || NumVal > 100)
624 return LogErrorP("Invalid precedecnce: must be 1..100");
625 BinaryPrecedence = (unsigned)NumVal;
626 getNextToken();
627 }
628 break;
629 }
630
631 if (CurTok != '(')
632 return LogErrorP("Expected '(' in prototype");
633
634 std::vector<std::string> ArgNames;
635 while (getNextToken() == tok_identifier)
636 ArgNames.push_back(IdentifierStr);
637 if (CurTok != ')')
638 return LogErrorP("Expected ')' in prototype");
639
640 // success.
641 getNextToken(); // eat ')'.
642
643 // Verify right number of names for operator.
644 if (Kind && ArgNames.size() != Kind)
645 return LogErrorP("Invalid number of operands for operator");
646
647 return llvm::make_unique<PrototypeAST>(FnName, ArgNames, Kind != 0,
648 BinaryPrecedence);
649 }
650
651 /// definition ::= 'def' prototype expression
ParseDefinition()652 static std::unique_ptr<FunctionAST> ParseDefinition() {
653 getNextToken(); // eat def.
654 auto Proto = ParsePrototype();
655 if (!Proto)
656 return nullptr;
657
658 if (auto E = ParseExpression())
659 return llvm::make_unique<FunctionAST>(std::move(Proto), std::move(E));
660 return nullptr;
661 }
662
663 /// toplevelexpr ::= expression
ParseTopLevelExpr()664 static std::unique_ptr<FunctionAST> ParseTopLevelExpr() {
665 if (auto E = ParseExpression()) {
666 // Make an anonymous proto.
667 auto Proto = llvm::make_unique<PrototypeAST>("__anon_expr",
668 std::vector<std::string>());
669 return llvm::make_unique<FunctionAST>(std::move(Proto), std::move(E));
670 }
671 return nullptr;
672 }
673
674 /// external ::= 'extern' prototype
ParseExtern()675 static std::unique_ptr<PrototypeAST> ParseExtern() {
676 getNextToken(); // eat extern.
677 return ParsePrototype();
678 }
679
680 //===----------------------------------------------------------------------===//
681 // Code Generation
682 //===----------------------------------------------------------------------===//
683
684 static LLVMContext TheContext;
685 static IRBuilder<> Builder(TheContext);
686 static std::unique_ptr<Module> TheModule;
687 static std::map<std::string, AllocaInst *> NamedValues;
688 static std::map<std::string, std::unique_ptr<PrototypeAST>> FunctionProtos;
689
LogErrorV(const char * Str)690 Value *LogErrorV(const char *Str) {
691 LogError(Str);
692 return nullptr;
693 }
694
getFunction(std::string Name)695 Function *getFunction(std::string Name) {
696 // First, see if the function has already been added to the current module.
697 if (auto *F = TheModule->getFunction(Name))
698 return F;
699
700 // If not, check whether we can codegen the declaration from some existing
701 // prototype.
702 auto FI = FunctionProtos.find(Name);
703 if (FI != FunctionProtos.end())
704 return FI->second->codegen();
705
706 // If no existing prototype exists, return null.
707 return nullptr;
708 }
709
710 /// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of
711 /// the function. This is used for mutable variables etc.
CreateEntryBlockAlloca(Function * TheFunction,const std::string & VarName)712 static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction,
713 const std::string &VarName) {
714 IRBuilder<> TmpB(&TheFunction->getEntryBlock(),
715 TheFunction->getEntryBlock().begin());
716 return TmpB.CreateAlloca(Type::getDoubleTy(TheContext), nullptr, VarName);
717 }
718
codegen()719 Value *NumberExprAST::codegen() {
720 return ConstantFP::get(TheContext, APFloat(Val));
721 }
722
codegen()723 Value *VariableExprAST::codegen() {
724 // Look this variable up in the function.
725 Value *V = NamedValues[Name];
726 if (!V)
727 return LogErrorV("Unknown variable name");
728
729 // Load the value.
730 return Builder.CreateLoad(V, Name.c_str());
731 }
732
codegen()733 Value *UnaryExprAST::codegen() {
734 Value *OperandV = Operand->codegen();
735 if (!OperandV)
736 return nullptr;
737
738 Function *F = getFunction(std::string("unary") + Opcode);
739 if (!F)
740 return LogErrorV("Unknown unary operator");
741
742 return Builder.CreateCall(F, OperandV, "unop");
743 }
744
codegen()745 Value *BinaryExprAST::codegen() {
746 // Special case '=' because we don't want to emit the LHS as an expression.
747 if (Op == '=') {
748 // Assignment requires the LHS to be an identifier.
749 // This assume we're building without RTTI because LLVM builds that way by
750 // default. If you build LLVM with RTTI this can be changed to a
751 // dynamic_cast for automatic error checking.
752 VariableExprAST *LHSE = static_cast<VariableExprAST *>(LHS.get());
753 if (!LHSE)
754 return LogErrorV("destination of '=' must be a variable");
755 // Codegen the RHS.
756 Value *Val = RHS->codegen();
757 if (!Val)
758 return nullptr;
759
760 // Look up the name.
761 Value *Variable = NamedValues[LHSE->getName()];
762 if (!Variable)
763 return LogErrorV("Unknown variable name");
764
765 Builder.CreateStore(Val, Variable);
766 return Val;
767 }
768
769 Value *L = LHS->codegen();
770 Value *R = RHS->codegen();
771 if (!L || !R)
772 return nullptr;
773
774 switch (Op) {
775 case '+':
776 return Builder.CreateFAdd(L, R, "addtmp");
777 case '-':
778 return Builder.CreateFSub(L, R, "subtmp");
779 case '*':
780 return Builder.CreateFMul(L, R, "multmp");
781 case '<':
782 L = Builder.CreateFCmpULT(L, R, "cmptmp");
783 // Convert bool 0/1 to double 0.0 or 1.0
784 return Builder.CreateUIToFP(L, Type::getDoubleTy(TheContext), "booltmp");
785 default:
786 break;
787 }
788
789 // If it wasn't a builtin binary operator, it must be a user defined one. Emit
790 // a call to it.
791 Function *F = getFunction(std::string("binary") + Op);
792 assert(F && "binary operator not found!");
793
794 Value *Ops[] = {L, R};
795 return Builder.CreateCall(F, Ops, "binop");
796 }
797
codegen()798 Value *CallExprAST::codegen() {
799 // Look up the name in the global module table.
800 Function *CalleeF = getFunction(Callee);
801 if (!CalleeF)
802 return LogErrorV("Unknown function referenced");
803
804 // If argument mismatch error.
805 if (CalleeF->arg_size() != Args.size())
806 return LogErrorV("Incorrect # arguments passed");
807
808 std::vector<Value *> ArgsV;
809 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
810 ArgsV.push_back(Args[i]->codegen());
811 if (!ArgsV.back())
812 return nullptr;
813 }
814
815 return Builder.CreateCall(CalleeF, ArgsV, "calltmp");
816 }
817
codegen()818 Value *IfExprAST::codegen() {
819 Value *CondV = Cond->codegen();
820 if (!CondV)
821 return nullptr;
822
823 // Convert condition to a bool by comparing equal to 0.0.
824 CondV = Builder.CreateFCmpONE(
825 CondV, ConstantFP::get(TheContext, APFloat(0.0)), "ifcond");
826
827 Function *TheFunction = Builder.GetInsertBlock()->getParent();
828
829 // Create blocks for the then and else cases. Insert the 'then' block at the
830 // end of the function.
831 BasicBlock *ThenBB = BasicBlock::Create(TheContext, "then", TheFunction);
832 BasicBlock *ElseBB = BasicBlock::Create(TheContext, "else");
833 BasicBlock *MergeBB = BasicBlock::Create(TheContext, "ifcont");
834
835 Builder.CreateCondBr(CondV, ThenBB, ElseBB);
836
837 // Emit then value.
838 Builder.SetInsertPoint(ThenBB);
839
840 Value *ThenV = Then->codegen();
841 if (!ThenV)
842 return nullptr;
843
844 Builder.CreateBr(MergeBB);
845 // Codegen of 'Then' can change the current block, update ThenBB for the PHI.
846 ThenBB = Builder.GetInsertBlock();
847
848 // Emit else block.
849 TheFunction->getBasicBlockList().push_back(ElseBB);
850 Builder.SetInsertPoint(ElseBB);
851
852 Value *ElseV = Else->codegen();
853 if (!ElseV)
854 return nullptr;
855
856 Builder.CreateBr(MergeBB);
857 // Codegen of 'Else' can change the current block, update ElseBB for the PHI.
858 ElseBB = Builder.GetInsertBlock();
859
860 // Emit merge block.
861 TheFunction->getBasicBlockList().push_back(MergeBB);
862 Builder.SetInsertPoint(MergeBB);
863 PHINode *PN = Builder.CreatePHI(Type::getDoubleTy(TheContext), 2, "iftmp");
864
865 PN->addIncoming(ThenV, ThenBB);
866 PN->addIncoming(ElseV, ElseBB);
867 return PN;
868 }
869
870 // Output for-loop as:
871 // var = alloca double
872 // ...
873 // start = startexpr
874 // store start -> var
875 // goto loop
876 // loop:
877 // ...
878 // bodyexpr
879 // ...
880 // loopend:
881 // step = stepexpr
882 // endcond = endexpr
883 //
884 // curvar = load var
885 // nextvar = curvar + step
886 // store nextvar -> var
887 // br endcond, loop, endloop
888 // outloop:
codegen()889 Value *ForExprAST::codegen() {
890 Function *TheFunction = Builder.GetInsertBlock()->getParent();
891
892 // Create an alloca for the variable in the entry block.
893 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
894
895 // Emit the start code first, without 'variable' in scope.
896 Value *StartVal = Start->codegen();
897 if (!StartVal)
898 return nullptr;
899
900 // Store the value into the alloca.
901 Builder.CreateStore(StartVal, Alloca);
902
903 // Make the new basic block for the loop header, inserting after current
904 // block.
905 BasicBlock *LoopBB = BasicBlock::Create(TheContext, "loop", TheFunction);
906
907 // Insert an explicit fall through from the current block to the LoopBB.
908 Builder.CreateBr(LoopBB);
909
910 // Start insertion in LoopBB.
911 Builder.SetInsertPoint(LoopBB);
912
913 // Within the loop, the variable is defined equal to the PHI node. If it
914 // shadows an existing variable, we have to restore it, so save it now.
915 AllocaInst *OldVal = NamedValues[VarName];
916 NamedValues[VarName] = Alloca;
917
918 // Emit the body of the loop. This, like any other expr, can change the
919 // current BB. Note that we ignore the value computed by the body, but don't
920 // allow an error.
921 if (!Body->codegen())
922 return nullptr;
923
924 // Emit the step value.
925 Value *StepVal = nullptr;
926 if (Step) {
927 StepVal = Step->codegen();
928 if (!StepVal)
929 return nullptr;
930 } else {
931 // If not specified, use 1.0.
932 StepVal = ConstantFP::get(TheContext, APFloat(1.0));
933 }
934
935 // Compute the end condition.
936 Value *EndCond = End->codegen();
937 if (!EndCond)
938 return nullptr;
939
940 // Reload, increment, and restore the alloca. This handles the case where
941 // the body of the loop mutates the variable.
942 Value *CurVar = Builder.CreateLoad(Alloca, VarName.c_str());
943 Value *NextVar = Builder.CreateFAdd(CurVar, StepVal, "nextvar");
944 Builder.CreateStore(NextVar, Alloca);
945
946 // Convert condition to a bool by comparing equal to 0.0.
947 EndCond = Builder.CreateFCmpONE(
948 EndCond, ConstantFP::get(TheContext, APFloat(0.0)), "loopcond");
949
950 // Create the "after loop" block and insert it.
951 BasicBlock *AfterBB =
952 BasicBlock::Create(TheContext, "afterloop", TheFunction);
953
954 // Insert the conditional branch into the end of LoopEndBB.
955 Builder.CreateCondBr(EndCond, LoopBB, AfterBB);
956
957 // Any new code will be inserted in AfterBB.
958 Builder.SetInsertPoint(AfterBB);
959
960 // Restore the unshadowed variable.
961 if (OldVal)
962 NamedValues[VarName] = OldVal;
963 else
964 NamedValues.erase(VarName);
965
966 // for expr always returns 0.0.
967 return Constant::getNullValue(Type::getDoubleTy(TheContext));
968 }
969
codegen()970 Value *VarExprAST::codegen() {
971 std::vector<AllocaInst *> OldBindings;
972
973 Function *TheFunction = Builder.GetInsertBlock()->getParent();
974
975 // Register all variables and emit their initializer.
976 for (unsigned i = 0, e = VarNames.size(); i != e; ++i) {
977 const std::string &VarName = VarNames[i].first;
978 ExprAST *Init = VarNames[i].second.get();
979
980 // Emit the initializer before adding the variable to scope, this prevents
981 // the initializer from referencing the variable itself, and permits stuff
982 // like this:
983 // var a = 1 in
984 // var a = a in ... # refers to outer 'a'.
985 Value *InitVal;
986 if (Init) {
987 InitVal = Init->codegen();
988 if (!InitVal)
989 return nullptr;
990 } else { // If not specified, use 0.0.
991 InitVal = ConstantFP::get(TheContext, APFloat(0.0));
992 }
993
994 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
995 Builder.CreateStore(InitVal, Alloca);
996
997 // Remember the old variable binding so that we can restore the binding when
998 // we unrecurse.
999 OldBindings.push_back(NamedValues[VarName]);
1000
1001 // Remember this binding.
1002 NamedValues[VarName] = Alloca;
1003 }
1004
1005 // Codegen the body, now that all vars are in scope.
1006 Value *BodyVal = Body->codegen();
1007 if (!BodyVal)
1008 return nullptr;
1009
1010 // Pop all our variables from scope.
1011 for (unsigned i = 0, e = VarNames.size(); i != e; ++i)
1012 NamedValues[VarNames[i].first] = OldBindings[i];
1013
1014 // Return the body computation.
1015 return BodyVal;
1016 }
1017
codegen()1018 Function *PrototypeAST::codegen() {
1019 // Make the function type: double(double,double) etc.
1020 std::vector<Type *> Doubles(Args.size(), Type::getDoubleTy(TheContext));
1021 FunctionType *FT =
1022 FunctionType::get(Type::getDoubleTy(TheContext), Doubles, false);
1023
1024 Function *F =
1025 Function::Create(FT, Function::ExternalLinkage, Name, TheModule.get());
1026
1027 // Set names for all arguments.
1028 unsigned Idx = 0;
1029 for (auto &Arg : F->args())
1030 Arg.setName(Args[Idx++]);
1031
1032 return F;
1033 }
1034
codegen()1035 Function *FunctionAST::codegen() {
1036 // Transfer ownership of the prototype to the FunctionProtos map, but keep a
1037 // reference to it for use below.
1038 auto &P = *Proto;
1039 FunctionProtos[Proto->getName()] = std::move(Proto);
1040 Function *TheFunction = getFunction(P.getName());
1041 if (!TheFunction)
1042 return nullptr;
1043
1044 // If this is an operator, install it.
1045 if (P.isBinaryOp())
1046 BinopPrecedence[P.getOperatorName()] = P.getBinaryPrecedence();
1047
1048 // Create a new basic block to start insertion into.
1049 BasicBlock *BB = BasicBlock::Create(TheContext, "entry", TheFunction);
1050 Builder.SetInsertPoint(BB);
1051
1052 // Record the function arguments in the NamedValues map.
1053 NamedValues.clear();
1054 for (auto &Arg : TheFunction->args()) {
1055 // Create an alloca for this variable.
1056 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, Arg.getName());
1057
1058 // Store the initial value into the alloca.
1059 Builder.CreateStore(&Arg, Alloca);
1060
1061 // Add arguments to variable symbol table.
1062 NamedValues[Arg.getName()] = Alloca;
1063 }
1064
1065 if (Value *RetVal = Body->codegen()) {
1066 // Finish off the function.
1067 Builder.CreateRet(RetVal);
1068
1069 // Validate the generated code, checking for consistency.
1070 verifyFunction(*TheFunction);
1071
1072 return TheFunction;
1073 }
1074
1075 // Error reading body, remove function.
1076 TheFunction->eraseFromParent();
1077
1078 if (P.isBinaryOp())
1079 BinopPrecedence.erase(Proto->getOperatorName());
1080 return nullptr;
1081 }
1082
1083 //===----------------------------------------------------------------------===//
1084 // Top-Level parsing and JIT Driver
1085 //===----------------------------------------------------------------------===//
1086
InitializeModuleAndPassManager()1087 static void InitializeModuleAndPassManager() {
1088 // Open a new module.
1089 TheModule = llvm::make_unique<Module>("my cool jit", TheContext);
1090 }
1091
HandleDefinition()1092 static void HandleDefinition() {
1093 if (auto FnAST = ParseDefinition()) {
1094 if (auto *FnIR = FnAST->codegen()) {
1095 fprintf(stderr, "Read function definition:");
1096 FnIR->dump();
1097 }
1098 } else {
1099 // Skip token for error recovery.
1100 getNextToken();
1101 }
1102 }
1103
HandleExtern()1104 static void HandleExtern() {
1105 if (auto ProtoAST = ParseExtern()) {
1106 if (auto *FnIR = ProtoAST->codegen()) {
1107 fprintf(stderr, "Read extern: ");
1108 FnIR->dump();
1109 FunctionProtos[ProtoAST->getName()] = std::move(ProtoAST);
1110 }
1111 } else {
1112 // Skip token for error recovery.
1113 getNextToken();
1114 }
1115 }
1116
HandleTopLevelExpression()1117 static void HandleTopLevelExpression() {
1118 // Evaluate a top-level expression into an anonymous function.
1119 if (auto FnAST = ParseTopLevelExpr()) {
1120 FnAST->codegen();
1121 } else {
1122 // Skip token for error recovery.
1123 getNextToken();
1124 }
1125 }
1126
1127 /// top ::= definition | external | expression | ';'
MainLoop()1128 static void MainLoop() {
1129 while (true) {
1130 switch (CurTok) {
1131 case tok_eof:
1132 return;
1133 case ';': // ignore top-level semicolons.
1134 getNextToken();
1135 break;
1136 case tok_def:
1137 HandleDefinition();
1138 break;
1139 case tok_extern:
1140 HandleExtern();
1141 break;
1142 default:
1143 HandleTopLevelExpression();
1144 break;
1145 }
1146 }
1147 }
1148
1149 //===----------------------------------------------------------------------===//
1150 // "Library" functions that can be "extern'd" from user code.
1151 //===----------------------------------------------------------------------===//
1152
1153 /// putchard - putchar that takes a double and returns 0.
putchard(double X)1154 extern "C" double putchard(double X) {
1155 fputc((char)X, stderr);
1156 return 0;
1157 }
1158
1159 /// printd - printf that takes a double prints it as "%f\n", returning 0.
printd(double X)1160 extern "C" double printd(double X) {
1161 fprintf(stderr, "%f\n", X);
1162 return 0;
1163 }
1164
1165 //===----------------------------------------------------------------------===//
1166 // Main driver code.
1167 //===----------------------------------------------------------------------===//
1168
main()1169 int main() {
1170 // Install standard binary operators.
1171 // 1 is lowest precedence.
1172 BinopPrecedence['<'] = 10;
1173 BinopPrecedence['+'] = 20;
1174 BinopPrecedence['-'] = 20;
1175 BinopPrecedence['*'] = 40; // highest.
1176
1177 // Prime the first token.
1178 fprintf(stderr, "ready> ");
1179 getNextToken();
1180
1181 InitializeModuleAndPassManager();
1182
1183 // Run the main "interpreter loop" now.
1184 MainLoop();
1185
1186 // Initialize the target registry etc.
1187 InitializeAllTargetInfos();
1188 InitializeAllTargets();
1189 InitializeAllTargetMCs();
1190 InitializeAllAsmParsers();
1191 InitializeAllAsmPrinters();
1192
1193 auto TargetTriple = sys::getDefaultTargetTriple();
1194 TheModule->setTargetTriple(TargetTriple);
1195
1196 std::string Error;
1197 auto Target = TargetRegistry::lookupTarget(TargetTriple, Error);
1198
1199 // Print an error and exit if we couldn't find the requested target.
1200 // This generally occurs if we've forgotten to initialise the
1201 // TargetRegistry or we have a bogus target triple.
1202 if (!Target) {
1203 errs() << Error;
1204 return 1;
1205 }
1206
1207 auto CPU = "generic";
1208 auto Features = "";
1209
1210 TargetOptions opt;
1211 auto RM = Optional<Reloc::Model>();
1212 auto TheTargetMachine =
1213 Target->createTargetMachine(TargetTriple, CPU, Features, opt, RM);
1214
1215 TheModule->setDataLayout(TheTargetMachine->createDataLayout());
1216
1217 auto Filename = "output.o";
1218 std::error_code EC;
1219 raw_fd_ostream dest(Filename, EC, sys::fs::F_None);
1220
1221 if (EC) {
1222 errs() << "Could not open file: " << EC.message();
1223 return 1;
1224 }
1225
1226 legacy::PassManager pass;
1227 auto FileType = TargetMachine::CGFT_ObjectFile;
1228
1229 if (TheTargetMachine->addPassesToEmitFile(pass, dest, FileType)) {
1230 errs() << "TheTargetMachine can't emit a file of this type";
1231 return 1;
1232 }
1233
1234 pass.run(*TheModule);
1235 dest.flush();
1236
1237 outs() << "Wrote " << Filename << "\n";
1238
1239 return 0;
1240 }
1241