1 #include "llvm/Analysis/Passes.h"
2 #include "llvm/ExecutionEngine/ExecutionEngine.h"
3 #include "llvm/ExecutionEngine/MCJIT.h"
4 #include "llvm/ExecutionEngine/SectionMemoryManager.h"
5 #include "llvm/IR/DataLayout.h"
6 #include "llvm/IR/DerivedTypes.h"
7 #include "llvm/IR/IRBuilder.h"
8 #include "llvm/IR/LLVMContext.h"
9 #include "llvm/IR/Module.h"
10 #include "llvm/IR/Verifier.h"
11 #include "llvm/PassManager.h"
12 #include "llvm/Support/TargetSelect.h"
13 #include "llvm/Transforms/Scalar.h"
14 #include <cctype>
15 #include <cstdio>
16 #include <map>
17 #include <string>
18 #include <vector>
19 using namespace llvm;
20
21 //===----------------------------------------------------------------------===//
22 // Lexer
23 //===----------------------------------------------------------------------===//
24
25 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
26 // of these for known things.
27 enum Token {
28 tok_eof = -1,
29
30 // commands
31 tok_def = -2,
32 tok_extern = -3,
33
34 // primary
35 tok_identifier = -4,
36 tok_number = -5
37 };
38
39 static std::string IdentifierStr; // Filled in if tok_identifier
40 static double NumVal; // Filled in if tok_number
41
42 /// gettok - Return the next token from standard input.
gettok()43 static int gettok() {
44 static int LastChar = ' ';
45
46 // Skip any whitespace.
47 while (isspace(LastChar))
48 LastChar = getchar();
49
50 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
51 IdentifierStr = LastChar;
52 while (isalnum((LastChar = getchar())))
53 IdentifierStr += LastChar;
54
55 if (IdentifierStr == "def")
56 return tok_def;
57 if (IdentifierStr == "extern")
58 return tok_extern;
59 return tok_identifier;
60 }
61
62 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+
63 std::string NumStr;
64 do {
65 NumStr += LastChar;
66 LastChar = getchar();
67 } while (isdigit(LastChar) || LastChar == '.');
68
69 NumVal = strtod(NumStr.c_str(), 0);
70 return tok_number;
71 }
72
73 if (LastChar == '#') {
74 // Comment until end of line.
75 do
76 LastChar = getchar();
77 while (LastChar != EOF && LastChar != '\n' && LastChar != '\r');
78
79 if (LastChar != EOF)
80 return gettok();
81 }
82
83 // Check for end of file. Don't eat the EOF.
84 if (LastChar == EOF)
85 return tok_eof;
86
87 // Otherwise, just return the character as its ascii value.
88 int ThisChar = LastChar;
89 LastChar = getchar();
90 return ThisChar;
91 }
92
93 //===----------------------------------------------------------------------===//
94 // Abstract Syntax Tree (aka Parse Tree)
95 //===----------------------------------------------------------------------===//
96 namespace {
97 /// ExprAST - Base class for all expression nodes.
98 class ExprAST {
99 public:
~ExprAST()100 virtual ~ExprAST() {}
101 virtual Value *Codegen() = 0;
102 };
103
104 /// NumberExprAST - Expression class for numeric literals like "1.0".
105 class NumberExprAST : public ExprAST {
106 double Val;
107
108 public:
NumberExprAST(double val)109 NumberExprAST(double val) : Val(val) {}
110 virtual Value *Codegen();
111 };
112
113 /// VariableExprAST - Expression class for referencing a variable, like "a".
114 class VariableExprAST : public ExprAST {
115 std::string Name;
116
117 public:
VariableExprAST(const std::string & name)118 VariableExprAST(const std::string &name) : Name(name) {}
119 virtual Value *Codegen();
120 };
121
122 /// BinaryExprAST - Expression class for a binary operator.
123 class BinaryExprAST : public ExprAST {
124 char Op;
125 ExprAST *LHS, *RHS;
126
127 public:
BinaryExprAST(char op,ExprAST * lhs,ExprAST * rhs)128 BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs)
129 : Op(op), LHS(lhs), RHS(rhs) {}
130 virtual Value *Codegen();
131 };
132
133 /// CallExprAST - Expression class for function calls.
134 class CallExprAST : public ExprAST {
135 std::string Callee;
136 std::vector<ExprAST *> Args;
137
138 public:
CallExprAST(const std::string & callee,std::vector<ExprAST * > & args)139 CallExprAST(const std::string &callee, std::vector<ExprAST *> &args)
140 : Callee(callee), Args(args) {}
141 virtual Value *Codegen();
142 };
143
144 /// PrototypeAST - This class represents the "prototype" for a function,
145 /// which captures its name, and its argument names (thus implicitly the number
146 /// of arguments the function takes).
147 class PrototypeAST {
148 std::string Name;
149 std::vector<std::string> Args;
150
151 public:
PrototypeAST(const std::string & name,const std::vector<std::string> & args)152 PrototypeAST(const std::string &name, const std::vector<std::string> &args)
153 : Name(name), Args(args) {}
154
155 Function *Codegen();
156 };
157
158 /// FunctionAST - This class represents a function definition itself.
159 class FunctionAST {
160 PrototypeAST *Proto;
161 ExprAST *Body;
162
163 public:
FunctionAST(PrototypeAST * proto,ExprAST * body)164 FunctionAST(PrototypeAST *proto, ExprAST *body) : Proto(proto), Body(body) {}
165
166 Function *Codegen();
167 };
168 } // end anonymous namespace
169
170 //===----------------------------------------------------------------------===//
171 // Parser
172 //===----------------------------------------------------------------------===//
173
174 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
175 /// token the parser is looking at. getNextToken reads another token from the
176 /// lexer and updates CurTok with its results.
177 static int CurTok;
getNextToken()178 static int getNextToken() { return CurTok = gettok(); }
179
180 /// BinopPrecedence - This holds the precedence for each binary operator that is
181 /// defined.
182 static std::map<char, int> BinopPrecedence;
183
184 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
GetTokPrecedence()185 static int GetTokPrecedence() {
186 if (!isascii(CurTok))
187 return -1;
188
189 // Make sure it's a declared binop.
190 int TokPrec = BinopPrecedence[CurTok];
191 if (TokPrec <= 0)
192 return -1;
193 return TokPrec;
194 }
195
196 /// Error* - These are little helper functions for error handling.
Error(const char * Str)197 ExprAST *Error(const char *Str) {
198 fprintf(stderr, "Error: %s\n", Str);
199 return 0;
200 }
ErrorP(const char * Str)201 PrototypeAST *ErrorP(const char *Str) {
202 Error(Str);
203 return 0;
204 }
ErrorF(const char * Str)205 FunctionAST *ErrorF(const char *Str) {
206 Error(Str);
207 return 0;
208 }
209
210 static ExprAST *ParseExpression();
211
212 /// identifierexpr
213 /// ::= identifier
214 /// ::= identifier '(' expression* ')'
ParseIdentifierExpr()215 static ExprAST *ParseIdentifierExpr() {
216 std::string IdName = IdentifierStr;
217
218 getNextToken(); // eat identifier.
219
220 if (CurTok != '(') // Simple variable ref.
221 return new VariableExprAST(IdName);
222
223 // Call.
224 getNextToken(); // eat (
225 std::vector<ExprAST *> Args;
226 if (CurTok != ')') {
227 while (1) {
228 ExprAST *Arg = ParseExpression();
229 if (!Arg)
230 return 0;
231 Args.push_back(Arg);
232
233 if (CurTok == ')')
234 break;
235
236 if (CurTok != ',')
237 return Error("Expected ')' or ',' in argument list");
238 getNextToken();
239 }
240 }
241
242 // Eat the ')'.
243 getNextToken();
244
245 return new CallExprAST(IdName, Args);
246 }
247
248 /// numberexpr ::= number
ParseNumberExpr()249 static ExprAST *ParseNumberExpr() {
250 ExprAST *Result = new NumberExprAST(NumVal);
251 getNextToken(); // consume the number
252 return Result;
253 }
254
255 /// parenexpr ::= '(' expression ')'
ParseParenExpr()256 static ExprAST *ParseParenExpr() {
257 getNextToken(); // eat (.
258 ExprAST *V = ParseExpression();
259 if (!V)
260 return 0;
261
262 if (CurTok != ')')
263 return Error("expected ')'");
264 getNextToken(); // eat ).
265 return V;
266 }
267
268 /// primary
269 /// ::= identifierexpr
270 /// ::= numberexpr
271 /// ::= parenexpr
ParsePrimary()272 static ExprAST *ParsePrimary() {
273 switch (CurTok) {
274 default:
275 return Error("unknown token when expecting an expression");
276 case tok_identifier:
277 return ParseIdentifierExpr();
278 case tok_number:
279 return ParseNumberExpr();
280 case '(':
281 return ParseParenExpr();
282 }
283 }
284
285 /// binoprhs
286 /// ::= ('+' primary)*
ParseBinOpRHS(int ExprPrec,ExprAST * LHS)287 static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) {
288 // If this is a binop, find its precedence.
289 while (1) {
290 int TokPrec = GetTokPrecedence();
291
292 // If this is a binop that binds at least as tightly as the current binop,
293 // consume it, otherwise we are done.
294 if (TokPrec < ExprPrec)
295 return LHS;
296
297 // Okay, we know this is a binop.
298 int BinOp = CurTok;
299 getNextToken(); // eat binop
300
301 // Parse the primary expression after the binary operator.
302 ExprAST *RHS = ParsePrimary();
303 if (!RHS)
304 return 0;
305
306 // If BinOp binds less tightly with RHS than the operator after RHS, let
307 // the pending operator take RHS as its LHS.
308 int NextPrec = GetTokPrecedence();
309 if (TokPrec < NextPrec) {
310 RHS = ParseBinOpRHS(TokPrec + 1, RHS);
311 if (RHS == 0)
312 return 0;
313 }
314
315 // Merge LHS/RHS.
316 LHS = new BinaryExprAST(BinOp, LHS, RHS);
317 }
318 }
319
320 /// expression
321 /// ::= primary binoprhs
322 ///
ParseExpression()323 static ExprAST *ParseExpression() {
324 ExprAST *LHS = ParsePrimary();
325 if (!LHS)
326 return 0;
327
328 return ParseBinOpRHS(0, LHS);
329 }
330
331 /// prototype
332 /// ::= id '(' id* ')'
ParsePrototype()333 static PrototypeAST *ParsePrototype() {
334 if (CurTok != tok_identifier)
335 return ErrorP("Expected function name in prototype");
336
337 std::string FnName = IdentifierStr;
338 getNextToken();
339
340 if (CurTok != '(')
341 return ErrorP("Expected '(' in prototype");
342
343 std::vector<std::string> ArgNames;
344 while (getNextToken() == tok_identifier)
345 ArgNames.push_back(IdentifierStr);
346 if (CurTok != ')')
347 return ErrorP("Expected ')' in prototype");
348
349 // success.
350 getNextToken(); // eat ')'.
351
352 return new PrototypeAST(FnName, ArgNames);
353 }
354
355 /// definition ::= 'def' prototype expression
ParseDefinition()356 static FunctionAST *ParseDefinition() {
357 getNextToken(); // eat def.
358 PrototypeAST *Proto = ParsePrototype();
359 if (Proto == 0)
360 return 0;
361
362 if (ExprAST *E = ParseExpression())
363 return new FunctionAST(Proto, E);
364 return 0;
365 }
366
367 /// toplevelexpr ::= expression
ParseTopLevelExpr()368 static FunctionAST *ParseTopLevelExpr() {
369 if (ExprAST *E = ParseExpression()) {
370 // Make an anonymous proto.
371 PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>());
372 return new FunctionAST(Proto, E);
373 }
374 return 0;
375 }
376
377 /// external ::= 'extern' prototype
ParseExtern()378 static PrototypeAST *ParseExtern() {
379 getNextToken(); // eat extern.
380 return ParsePrototype();
381 }
382
383 //===----------------------------------------------------------------------===//
384 // Code Generation
385 //===----------------------------------------------------------------------===//
386
387 static Module *TheModule;
388 static IRBuilder<> Builder(getGlobalContext());
389 static std::map<std::string, Value *> NamedValues;
390 static FunctionPassManager *TheFPM;
391
ErrorV(const char * Str)392 Value *ErrorV(const char *Str) {
393 Error(Str);
394 return 0;
395 }
396
Codegen()397 Value *NumberExprAST::Codegen() {
398 return ConstantFP::get(getGlobalContext(), APFloat(Val));
399 }
400
Codegen()401 Value *VariableExprAST::Codegen() {
402 // Look this variable up in the function.
403 Value *V = NamedValues[Name];
404 return V ? V : ErrorV("Unknown variable name");
405 }
406
Codegen()407 Value *BinaryExprAST::Codegen() {
408 Value *L = LHS->Codegen();
409 Value *R = RHS->Codegen();
410 if (L == 0 || R == 0)
411 return 0;
412
413 switch (Op) {
414 case '+':
415 return Builder.CreateFAdd(L, R, "addtmp");
416 case '-':
417 return Builder.CreateFSub(L, R, "subtmp");
418 case '*':
419 return Builder.CreateFMul(L, R, "multmp");
420 case '<':
421 L = Builder.CreateFCmpULT(L, R, "cmptmp");
422 // Convert bool 0/1 to double 0.0 or 1.0
423 return Builder.CreateUIToFP(L, Type::getDoubleTy(getGlobalContext()),
424 "booltmp");
425 default:
426 return ErrorV("invalid binary operator");
427 }
428 }
429
Codegen()430 Value *CallExprAST::Codegen() {
431 // Look up the name in the global module table.
432 Function *CalleeF = TheModule->getFunction(Callee);
433 if (CalleeF == 0)
434 return ErrorV("Unknown function referenced");
435
436 // If argument mismatch error.
437 if (CalleeF->arg_size() != Args.size())
438 return ErrorV("Incorrect # arguments passed");
439
440 std::vector<Value *> ArgsV;
441 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
442 ArgsV.push_back(Args[i]->Codegen());
443 if (ArgsV.back() == 0)
444 return 0;
445 }
446
447 return Builder.CreateCall(CalleeF, ArgsV, "calltmp");
448 }
449
Codegen()450 Function *PrototypeAST::Codegen() {
451 // Make the function type: double(double,double) etc.
452 std::vector<Type *> Doubles(Args.size(),
453 Type::getDoubleTy(getGlobalContext()));
454 FunctionType *FT =
455 FunctionType::get(Type::getDoubleTy(getGlobalContext()), Doubles, false);
456
457 Function *F =
458 Function::Create(FT, Function::ExternalLinkage, Name, TheModule);
459
460 // If F conflicted, there was already something named 'Name'. If it has a
461 // body, don't allow redefinition or reextern.
462 if (F->getName() != Name) {
463 // Delete the one we just made and get the existing one.
464 F->eraseFromParent();
465 F = TheModule->getFunction(Name);
466
467 // If F already has a body, reject this.
468 if (!F->empty()) {
469 ErrorF("redefinition of function");
470 return 0;
471 }
472
473 // If F took a different number of args, reject.
474 if (F->arg_size() != Args.size()) {
475 ErrorF("redefinition of function with different # args");
476 return 0;
477 }
478 }
479
480 // Set names for all arguments.
481 unsigned Idx = 0;
482 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
483 ++AI, ++Idx) {
484 AI->setName(Args[Idx]);
485
486 // Add arguments to variable symbol table.
487 NamedValues[Args[Idx]] = AI;
488 }
489
490 return F;
491 }
492
Codegen()493 Function *FunctionAST::Codegen() {
494 NamedValues.clear();
495
496 Function *TheFunction = Proto->Codegen();
497 if (TheFunction == 0)
498 return 0;
499
500 // Create a new basic block to start insertion into.
501 BasicBlock *BB = BasicBlock::Create(getGlobalContext(), "entry", TheFunction);
502 Builder.SetInsertPoint(BB);
503
504 if (Value *RetVal = Body->Codegen()) {
505 // Finish off the function.
506 Builder.CreateRet(RetVal);
507
508 // Validate the generated code, checking for consistency.
509 verifyFunction(*TheFunction);
510
511 // Optimize the function.
512 TheFPM->run(*TheFunction);
513
514 return TheFunction;
515 }
516
517 // Error reading body, remove function.
518 TheFunction->eraseFromParent();
519 return 0;
520 }
521
522 //===----------------------------------------------------------------------===//
523 // Top-Level parsing and JIT Driver
524 //===----------------------------------------------------------------------===//
525
526 static ExecutionEngine *TheExecutionEngine;
527
HandleDefinition()528 static void HandleDefinition() {
529 if (FunctionAST *F = ParseDefinition()) {
530 if (Function *LF = F->Codegen()) {
531 fprintf(stderr, "Read function definition:");
532 LF->dump();
533 }
534 } else {
535 // Skip token for error recovery.
536 getNextToken();
537 }
538 }
539
HandleExtern()540 static void HandleExtern() {
541 if (PrototypeAST *P = ParseExtern()) {
542 if (Function *F = P->Codegen()) {
543 fprintf(stderr, "Read extern: ");
544 F->dump();
545 }
546 } else {
547 // Skip token for error recovery.
548 getNextToken();
549 }
550 }
551
HandleTopLevelExpression()552 static void HandleTopLevelExpression() {
553 // Evaluate a top-level expression into an anonymous function.
554 if (FunctionAST *F = ParseTopLevelExpr()) {
555 if (Function *LF = F->Codegen()) {
556 TheExecutionEngine->finalizeObject();
557 // JIT the function, returning a function pointer.
558 void *FPtr = TheExecutionEngine->getPointerToFunction(LF);
559
560 // Cast it to the right type (takes no arguments, returns a double) so we
561 // can call it as a native function.
562 double (*FP)() = (double (*)())(intptr_t)FPtr;
563 fprintf(stderr, "Evaluated to %f\n", FP());
564 }
565 } else {
566 // Skip token for error recovery.
567 getNextToken();
568 }
569 }
570
571 /// top ::= definition | external | expression | ';'
MainLoop()572 static void MainLoop() {
573 while (1) {
574 fprintf(stderr, "ready> ");
575 switch (CurTok) {
576 case tok_eof:
577 return;
578 case ';':
579 getNextToken();
580 break; // ignore top-level semicolons.
581 case tok_def:
582 HandleDefinition();
583 break;
584 case tok_extern:
585 HandleExtern();
586 break;
587 default:
588 HandleTopLevelExpression();
589 break;
590 }
591 }
592 }
593
594 //===----------------------------------------------------------------------===//
595 // "Library" functions that can be "extern'd" from user code.
596 //===----------------------------------------------------------------------===//
597
598 /// putchard - putchar that takes a double and returns 0.
putchard(double X)599 extern "C" double putchard(double X) {
600 putchar((char)X);
601 return 0;
602 }
603
604 //===----------------------------------------------------------------------===//
605 // Main driver code.
606 //===----------------------------------------------------------------------===//
607
main()608 int main() {
609 InitializeNativeTarget();
610 InitializeNativeTargetAsmPrinter();
611 InitializeNativeTargetAsmParser();
612 LLVMContext &Context = getGlobalContext();
613
614 // Install standard binary operators.
615 // 1 is lowest precedence.
616 BinopPrecedence['<'] = 10;
617 BinopPrecedence['+'] = 20;
618 BinopPrecedence['-'] = 20;
619 BinopPrecedence['*'] = 40; // highest.
620
621 // Prime the first token.
622 fprintf(stderr, "ready> ");
623 getNextToken();
624
625 // Make the module, which holds all the code.
626 std::unique_ptr<Module> Owner = make_unique<Module>("my cool jit", Context);
627 TheModule = Owner.get();
628
629 // Create the JIT. This takes ownership of the module.
630 std::string ErrStr;
631 TheExecutionEngine =
632 EngineBuilder(std::move(Owner))
633 .setErrorStr(&ErrStr)
634 .setMCJITMemoryManager(llvm::make_unique<SectionMemoryManager>())
635 .create();
636 if (!TheExecutionEngine) {
637 fprintf(stderr, "Could not create ExecutionEngine: %s\n", ErrStr.c_str());
638 exit(1);
639 }
640
641 FunctionPassManager OurFPM(TheModule);
642
643 // Set up the optimizer pipeline. Start with registering info about how the
644 // target lays out data structures.
645 TheModule->setDataLayout(TheExecutionEngine->getDataLayout());
646 OurFPM.add(new DataLayoutPass());
647 // Provide basic AliasAnalysis support for GVN.
648 OurFPM.add(createBasicAliasAnalysisPass());
649 // Do simple "peephole" optimizations and bit-twiddling optzns.
650 OurFPM.add(createInstructionCombiningPass());
651 // Reassociate expressions.
652 OurFPM.add(createReassociatePass());
653 // Eliminate Common SubExpressions.
654 OurFPM.add(createGVNPass());
655 // Simplify the control flow graph (deleting unreachable blocks, etc).
656 OurFPM.add(createCFGSimplificationPass());
657
658 OurFPM.doInitialization();
659
660 // Set the global so the code gen can use this.
661 TheFPM = &OurFPM;
662
663 // Run the main "interpreter loop" now.
664 MainLoop();
665
666 TheFPM = 0;
667
668 // Print out all of the generated code.
669 TheModule->dump();
670
671 return 0;
672 }
673