1 #include "../include/KaleidoscopeJIT.h"
2 #include "llvm/ADT/APFloat.h"
3 #include "llvm/ADT/STLExtras.h"
4 #include "llvm/IR/BasicBlock.h"
5 #include "llvm/IR/Constants.h"
6 #include "llvm/IR/DerivedTypes.h"
7 #include "llvm/IR/Function.h"
8 #include "llvm/IR/IRBuilder.h"
9 #include "llvm/IR/Instructions.h"
10 #include "llvm/IR/LLVMContext.h"
11 #include "llvm/IR/LegacyPassManager.h"
12 #include "llvm/IR/Module.h"
13 #include "llvm/IR/Type.h"
14 #include "llvm/IR/Verifier.h"
15 #include "llvm/Support/TargetSelect.h"
16 #include "llvm/Target/TargetMachine.h"
17 #include "llvm/Transforms/InstCombine/InstCombine.h"
18 #include "llvm/Transforms/Scalar.h"
19 #include "llvm/Transforms/Scalar/GVN.h"
20 #include <algorithm>
21 #include <cassert>
22 #include <cctype>
23 #include <cstdint>
24 #include <cstdio>
25 #include <cstdlib>
26 #include <map>
27 #include <memory>
28 #include <string>
29 #include <vector>
30 
31 using namespace llvm;
32 using namespace llvm::orc;
33 
34 //===----------------------------------------------------------------------===//
35 // Lexer
36 //===----------------------------------------------------------------------===//
37 
38 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
39 // of these for known things.
40 enum Token {
41   tok_eof = -1,
42 
43   // commands
44   tok_def = -2,
45   tok_extern = -3,
46 
47   // primary
48   tok_identifier = -4,
49   tok_number = -5,
50 
51   // control
52   tok_if = -6,
53   tok_then = -7,
54   tok_else = -8,
55   tok_for = -9,
56   tok_in = -10
57 };
58 
59 static std::string IdentifierStr; // Filled in if tok_identifier
60 static double NumVal;             // Filled in if tok_number
61 
62 /// gettok - Return the next token from standard input.
gettok()63 static int gettok() {
64   static int LastChar = ' ';
65 
66   // Skip any whitespace.
67   while (isspace(LastChar))
68     LastChar = getchar();
69 
70   if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
71     IdentifierStr = LastChar;
72     while (isalnum((LastChar = getchar())))
73       IdentifierStr += LastChar;
74 
75     if (IdentifierStr == "def")
76       return tok_def;
77     if (IdentifierStr == "extern")
78       return tok_extern;
79     if (IdentifierStr == "if")
80       return tok_if;
81     if (IdentifierStr == "then")
82       return tok_then;
83     if (IdentifierStr == "else")
84       return tok_else;
85     if (IdentifierStr == "for")
86       return tok_for;
87     if (IdentifierStr == "in")
88       return tok_in;
89     return tok_identifier;
90   }
91 
92   if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+
93     std::string NumStr;
94     do {
95       NumStr += LastChar;
96       LastChar = getchar();
97     } while (isdigit(LastChar) || LastChar == '.');
98 
99     NumVal = strtod(NumStr.c_str(), nullptr);
100     return tok_number;
101   }
102 
103   if (LastChar == '#') {
104     // Comment until end of line.
105     do
106       LastChar = getchar();
107     while (LastChar != EOF && LastChar != '\n' && LastChar != '\r');
108 
109     if (LastChar != EOF)
110       return gettok();
111   }
112 
113   // Check for end of file.  Don't eat the EOF.
114   if (LastChar == EOF)
115     return tok_eof;
116 
117   // Otherwise, just return the character as its ascii value.
118   int ThisChar = LastChar;
119   LastChar = getchar();
120   return ThisChar;
121 }
122 
123 //===----------------------------------------------------------------------===//
124 // Abstract Syntax Tree (aka Parse Tree)
125 //===----------------------------------------------------------------------===//
126 
127 namespace {
128 
129 /// ExprAST - Base class for all expression nodes.
130 class ExprAST {
131 public:
132   virtual ~ExprAST() = default;
133 
134   virtual Value *codegen() = 0;
135 };
136 
137 /// NumberExprAST - Expression class for numeric literals like "1.0".
138 class NumberExprAST : public ExprAST {
139   double Val;
140 
141 public:
NumberExprAST(double Val)142   NumberExprAST(double Val) : Val(Val) {}
143 
144   Value *codegen() override;
145 };
146 
147 /// VariableExprAST - Expression class for referencing a variable, like "a".
148 class VariableExprAST : public ExprAST {
149   std::string Name;
150 
151 public:
VariableExprAST(const std::string & Name)152   VariableExprAST(const std::string &Name) : Name(Name) {}
153 
154   Value *codegen() override;
155 };
156 
157 /// BinaryExprAST - Expression class for a binary operator.
158 class BinaryExprAST : public ExprAST {
159   char Op;
160   std::unique_ptr<ExprAST> LHS, RHS;
161 
162 public:
BinaryExprAST(char Op,std::unique_ptr<ExprAST> LHS,std::unique_ptr<ExprAST> RHS)163   BinaryExprAST(char Op, std::unique_ptr<ExprAST> LHS,
164                 std::unique_ptr<ExprAST> RHS)
165       : Op(Op), LHS(std::move(LHS)), RHS(std::move(RHS)) {}
166 
167   Value *codegen() override;
168 };
169 
170 /// CallExprAST - Expression class for function calls.
171 class CallExprAST : public ExprAST {
172   std::string Callee;
173   std::vector<std::unique_ptr<ExprAST>> Args;
174 
175 public:
CallExprAST(const std::string & Callee,std::vector<std::unique_ptr<ExprAST>> Args)176   CallExprAST(const std::string &Callee,
177               std::vector<std::unique_ptr<ExprAST>> Args)
178       : Callee(Callee), Args(std::move(Args)) {}
179 
180   Value *codegen() override;
181 };
182 
183 /// IfExprAST - Expression class for if/then/else.
184 class IfExprAST : public ExprAST {
185   std::unique_ptr<ExprAST> Cond, Then, Else;
186 
187 public:
IfExprAST(std::unique_ptr<ExprAST> Cond,std::unique_ptr<ExprAST> Then,std::unique_ptr<ExprAST> Else)188   IfExprAST(std::unique_ptr<ExprAST> Cond, std::unique_ptr<ExprAST> Then,
189             std::unique_ptr<ExprAST> Else)
190       : Cond(std::move(Cond)), Then(std::move(Then)), Else(std::move(Else)) {}
191 
192   Value *codegen() override;
193 };
194 
195 /// ForExprAST - Expression class for for/in.
196 class ForExprAST : public ExprAST {
197   std::string VarName;
198   std::unique_ptr<ExprAST> Start, End, Step, Body;
199 
200 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)201   ForExprAST(const std::string &VarName, std::unique_ptr<ExprAST> Start,
202              std::unique_ptr<ExprAST> End, std::unique_ptr<ExprAST> Step,
203              std::unique_ptr<ExprAST> Body)
204       : VarName(VarName), Start(std::move(Start)), End(std::move(End)),
205         Step(std::move(Step)), Body(std::move(Body)) {}
206 
207   Value *codegen() override;
208 };
209 
210 /// PrototypeAST - This class represents the "prototype" for a function,
211 /// which captures its name, and its argument names (thus implicitly the number
212 /// of arguments the function takes).
213 class PrototypeAST {
214   std::string Name;
215   std::vector<std::string> Args;
216 
217 public:
PrototypeAST(const std::string & Name,std::vector<std::string> Args)218   PrototypeAST(const std::string &Name, std::vector<std::string> Args)
219       : Name(Name), Args(std::move(Args)) {}
220 
221   Function *codegen();
getName() const222   const std::string &getName() const { return Name; }
223 };
224 
225 /// FunctionAST - This class represents a function definition itself.
226 class FunctionAST {
227   std::unique_ptr<PrototypeAST> Proto;
228   std::unique_ptr<ExprAST> Body;
229 
230 public:
FunctionAST(std::unique_ptr<PrototypeAST> Proto,std::unique_ptr<ExprAST> Body)231   FunctionAST(std::unique_ptr<PrototypeAST> Proto,
232               std::unique_ptr<ExprAST> Body)
233       : Proto(std::move(Proto)), Body(std::move(Body)) {}
234 
235   Function *codegen();
236 };
237 
238 } // end anonymous namespace
239 
240 //===----------------------------------------------------------------------===//
241 // Parser
242 //===----------------------------------------------------------------------===//
243 
244 /// CurTok/getNextToken - Provide a simple token buffer.  CurTok is the current
245 /// token the parser is looking at.  getNextToken reads another token from the
246 /// lexer and updates CurTok with its results.
247 static int CurTok;
getNextToken()248 static int getNextToken() { return CurTok = gettok(); }
249 
250 /// BinopPrecedence - This holds the precedence for each binary operator that is
251 /// defined.
252 static std::map<char, int> BinopPrecedence;
253 
254 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
GetTokPrecedence()255 static int GetTokPrecedence() {
256   if (!isascii(CurTok))
257     return -1;
258 
259   // Make sure it's a declared binop.
260   int TokPrec = BinopPrecedence[CurTok];
261   if (TokPrec <= 0)
262     return -1;
263   return TokPrec;
264 }
265 
266 /// LogError* - These are little helper functions for error handling.
LogError(const char * Str)267 std::unique_ptr<ExprAST> LogError(const char *Str) {
268   fprintf(stderr, "Error: %s\n", Str);
269   return nullptr;
270 }
271 
LogErrorP(const char * Str)272 std::unique_ptr<PrototypeAST> LogErrorP(const char *Str) {
273   LogError(Str);
274   return nullptr;
275 }
276 
277 static std::unique_ptr<ExprAST> ParseExpression();
278 
279 /// numberexpr ::= number
ParseNumberExpr()280 static std::unique_ptr<ExprAST> ParseNumberExpr() {
281   auto Result = std::make_unique<NumberExprAST>(NumVal);
282   getNextToken(); // consume the number
283   return std::move(Result);
284 }
285 
286 /// parenexpr ::= '(' expression ')'
ParseParenExpr()287 static std::unique_ptr<ExprAST> ParseParenExpr() {
288   getNextToken(); // eat (.
289   auto V = ParseExpression();
290   if (!V)
291     return nullptr;
292 
293   if (CurTok != ')')
294     return LogError("expected ')'");
295   getNextToken(); // eat ).
296   return V;
297 }
298 
299 /// identifierexpr
300 ///   ::= identifier
301 ///   ::= identifier '(' expression* ')'
ParseIdentifierExpr()302 static std::unique_ptr<ExprAST> ParseIdentifierExpr() {
303   std::string IdName = IdentifierStr;
304 
305   getNextToken(); // eat identifier.
306 
307   if (CurTok != '(') // Simple variable ref.
308     return std::make_unique<VariableExprAST>(IdName);
309 
310   // Call.
311   getNextToken(); // eat (
312   std::vector<std::unique_ptr<ExprAST>> Args;
313   if (CurTok != ')') {
314     while (true) {
315       if (auto Arg = ParseExpression())
316         Args.push_back(std::move(Arg));
317       else
318         return nullptr;
319 
320       if (CurTok == ')')
321         break;
322 
323       if (CurTok != ',')
324         return LogError("Expected ')' or ',' in argument list");
325       getNextToken();
326     }
327   }
328 
329   // Eat the ')'.
330   getNextToken();
331 
332   return std::make_unique<CallExprAST>(IdName, std::move(Args));
333 }
334 
335 /// ifexpr ::= 'if' expression 'then' expression 'else' expression
ParseIfExpr()336 static std::unique_ptr<ExprAST> ParseIfExpr() {
337   getNextToken(); // eat the if.
338 
339   // condition.
340   auto Cond = ParseExpression();
341   if (!Cond)
342     return nullptr;
343 
344   if (CurTok != tok_then)
345     return LogError("expected then");
346   getNextToken(); // eat the then
347 
348   auto Then = ParseExpression();
349   if (!Then)
350     return nullptr;
351 
352   if (CurTok != tok_else)
353     return LogError("expected else");
354 
355   getNextToken();
356 
357   auto Else = ParseExpression();
358   if (!Else)
359     return nullptr;
360 
361   return std::make_unique<IfExprAST>(std::move(Cond), std::move(Then),
362                                       std::move(Else));
363 }
364 
365 /// forexpr ::= 'for' identifier '=' expr ',' expr (',' expr)? 'in' expression
ParseForExpr()366 static std::unique_ptr<ExprAST> ParseForExpr() {
367   getNextToken(); // eat the for.
368 
369   if (CurTok != tok_identifier)
370     return LogError("expected identifier after for");
371 
372   std::string IdName = IdentifierStr;
373   getNextToken(); // eat identifier.
374 
375   if (CurTok != '=')
376     return LogError("expected '=' after for");
377   getNextToken(); // eat '='.
378 
379   auto Start = ParseExpression();
380   if (!Start)
381     return nullptr;
382   if (CurTok != ',')
383     return LogError("expected ',' after for start value");
384   getNextToken();
385 
386   auto End = ParseExpression();
387   if (!End)
388     return nullptr;
389 
390   // The step value is optional.
391   std::unique_ptr<ExprAST> Step;
392   if (CurTok == ',') {
393     getNextToken();
394     Step = ParseExpression();
395     if (!Step)
396       return nullptr;
397   }
398 
399   if (CurTok != tok_in)
400     return LogError("expected 'in' after for");
401   getNextToken(); // eat 'in'.
402 
403   auto Body = ParseExpression();
404   if (!Body)
405     return nullptr;
406 
407   return std::make_unique<ForExprAST>(IdName, std::move(Start), std::move(End),
408                                        std::move(Step), std::move(Body));
409 }
410 
411 /// primary
412 ///   ::= identifierexpr
413 ///   ::= numberexpr
414 ///   ::= parenexpr
415 ///   ::= ifexpr
416 ///   ::= forexpr
ParsePrimary()417 static std::unique_ptr<ExprAST> ParsePrimary() {
418   switch (CurTok) {
419   default:
420     return LogError("unknown token when expecting an expression");
421   case tok_identifier:
422     return ParseIdentifierExpr();
423   case tok_number:
424     return ParseNumberExpr();
425   case '(':
426     return ParseParenExpr();
427   case tok_if:
428     return ParseIfExpr();
429   case tok_for:
430     return ParseForExpr();
431   }
432 }
433 
434 /// binoprhs
435 ///   ::= ('+' primary)*
ParseBinOpRHS(int ExprPrec,std::unique_ptr<ExprAST> LHS)436 static std::unique_ptr<ExprAST> ParseBinOpRHS(int ExprPrec,
437                                               std::unique_ptr<ExprAST> LHS) {
438   // If this is a binop, find its precedence.
439   while (true) {
440     int TokPrec = GetTokPrecedence();
441 
442     // If this is a binop that binds at least as tightly as the current binop,
443     // consume it, otherwise we are done.
444     if (TokPrec < ExprPrec)
445       return LHS;
446 
447     // Okay, we know this is a binop.
448     int BinOp = CurTok;
449     getNextToken(); // eat binop
450 
451     // Parse the primary expression after the binary operator.
452     auto RHS = ParsePrimary();
453     if (!RHS)
454       return nullptr;
455 
456     // If BinOp binds less tightly with RHS than the operator after RHS, let
457     // the pending operator take RHS as its LHS.
458     int NextPrec = GetTokPrecedence();
459     if (TokPrec < NextPrec) {
460       RHS = ParseBinOpRHS(TokPrec + 1, std::move(RHS));
461       if (!RHS)
462         return nullptr;
463     }
464 
465     // Merge LHS/RHS.
466     LHS =
467         std::make_unique<BinaryExprAST>(BinOp, std::move(LHS), std::move(RHS));
468   }
469 }
470 
471 /// expression
472 ///   ::= primary binoprhs
473 ///
ParseExpression()474 static std::unique_ptr<ExprAST> ParseExpression() {
475   auto LHS = ParsePrimary();
476   if (!LHS)
477     return nullptr;
478 
479   return ParseBinOpRHS(0, std::move(LHS));
480 }
481 
482 /// prototype
483 ///   ::= id '(' id* ')'
ParsePrototype()484 static std::unique_ptr<PrototypeAST> ParsePrototype() {
485   if (CurTok != tok_identifier)
486     return LogErrorP("Expected function name in prototype");
487 
488   std::string FnName = IdentifierStr;
489   getNextToken();
490 
491   if (CurTok != '(')
492     return LogErrorP("Expected '(' in prototype");
493 
494   std::vector<std::string> ArgNames;
495   while (getNextToken() == tok_identifier)
496     ArgNames.push_back(IdentifierStr);
497   if (CurTok != ')')
498     return LogErrorP("Expected ')' in prototype");
499 
500   // success.
501   getNextToken(); // eat ')'.
502 
503   return std::make_unique<PrototypeAST>(FnName, std::move(ArgNames));
504 }
505 
506 /// definition ::= 'def' prototype expression
ParseDefinition()507 static std::unique_ptr<FunctionAST> ParseDefinition() {
508   getNextToken(); // eat def.
509   auto Proto = ParsePrototype();
510   if (!Proto)
511     return nullptr;
512 
513   if (auto E = ParseExpression())
514     return std::make_unique<FunctionAST>(std::move(Proto), std::move(E));
515   return nullptr;
516 }
517 
518 /// toplevelexpr ::= expression
ParseTopLevelExpr()519 static std::unique_ptr<FunctionAST> ParseTopLevelExpr() {
520   if (auto E = ParseExpression()) {
521     // Make an anonymous proto.
522     auto Proto = std::make_unique<PrototypeAST>("__anon_expr",
523                                                  std::vector<std::string>());
524     return std::make_unique<FunctionAST>(std::move(Proto), std::move(E));
525   }
526   return nullptr;
527 }
528 
529 /// external ::= 'extern' prototype
ParseExtern()530 static std::unique_ptr<PrototypeAST> ParseExtern() {
531   getNextToken(); // eat extern.
532   return ParsePrototype();
533 }
534 
535 //===----------------------------------------------------------------------===//
536 // Code Generation
537 //===----------------------------------------------------------------------===//
538 
539 static std::unique_ptr<LLVMContext> TheContext;
540 static std::unique_ptr<Module> TheModule;
541 static std::unique_ptr<IRBuilder<>> Builder;
542 static std::map<std::string, Value *> NamedValues;
543 static std::unique_ptr<legacy::FunctionPassManager> TheFPM;
544 static std::unique_ptr<KaleidoscopeJIT> TheJIT;
545 static std::map<std::string, std::unique_ptr<PrototypeAST>> FunctionProtos;
546 static ExitOnError ExitOnErr;
547 
LogErrorV(const char * Str)548 Value *LogErrorV(const char *Str) {
549   LogError(Str);
550   return nullptr;
551 }
552 
getFunction(std::string Name)553 Function *getFunction(std::string Name) {
554   // First, see if the function has already been added to the current module.
555   if (auto *F = TheModule->getFunction(Name))
556     return F;
557 
558   // If not, check whether we can codegen the declaration from some existing
559   // prototype.
560   auto FI = FunctionProtos.find(Name);
561   if (FI != FunctionProtos.end())
562     return FI->second->codegen();
563 
564   // If no existing prototype exists, return null.
565   return nullptr;
566 }
567 
codegen()568 Value *NumberExprAST::codegen() {
569   return ConstantFP::get(*TheContext, APFloat(Val));
570 }
571 
codegen()572 Value *VariableExprAST::codegen() {
573   // Look this variable up in the function.
574   Value *V = NamedValues[Name];
575   if (!V)
576     return LogErrorV("Unknown variable name");
577   return V;
578 }
579 
codegen()580 Value *BinaryExprAST::codegen() {
581   Value *L = LHS->codegen();
582   Value *R = RHS->codegen();
583   if (!L || !R)
584     return nullptr;
585 
586   switch (Op) {
587   case '+':
588     return Builder->CreateFAdd(L, R, "addtmp");
589   case '-':
590     return Builder->CreateFSub(L, R, "subtmp");
591   case '*':
592     return Builder->CreateFMul(L, R, "multmp");
593   case '<':
594     L = Builder->CreateFCmpULT(L, R, "cmptmp");
595     // Convert bool 0/1 to double 0.0 or 1.0
596     return Builder->CreateUIToFP(L, Type::getDoubleTy(*TheContext), "booltmp");
597   default:
598     return LogErrorV("invalid binary operator");
599   }
600 }
601 
codegen()602 Value *CallExprAST::codegen() {
603   // Look up the name in the global module table.
604   Function *CalleeF = getFunction(Callee);
605   if (!CalleeF)
606     return LogErrorV("Unknown function referenced");
607 
608   // If argument mismatch error.
609   if (CalleeF->arg_size() != Args.size())
610     return LogErrorV("Incorrect # arguments passed");
611 
612   std::vector<Value *> ArgsV;
613   for (unsigned i = 0, e = Args.size(); i != e; ++i) {
614     ArgsV.push_back(Args[i]->codegen());
615     if (!ArgsV.back())
616       return nullptr;
617   }
618 
619   return Builder->CreateCall(CalleeF, ArgsV, "calltmp");
620 }
621 
codegen()622 Value *IfExprAST::codegen() {
623   Value *CondV = Cond->codegen();
624   if (!CondV)
625     return nullptr;
626 
627   // Convert condition to a bool by comparing non-equal to 0.0.
628   CondV = Builder->CreateFCmpONE(
629       CondV, ConstantFP::get(*TheContext, APFloat(0.0)), "ifcond");
630 
631   Function *TheFunction = Builder->GetInsertBlock()->getParent();
632 
633   // Create blocks for the then and else cases.  Insert the 'then' block at the
634   // end of the function.
635   BasicBlock *ThenBB = BasicBlock::Create(*TheContext, "then", TheFunction);
636   BasicBlock *ElseBB = BasicBlock::Create(*TheContext, "else");
637   BasicBlock *MergeBB = BasicBlock::Create(*TheContext, "ifcont");
638 
639   Builder->CreateCondBr(CondV, ThenBB, ElseBB);
640 
641   // Emit then value.
642   Builder->SetInsertPoint(ThenBB);
643 
644   Value *ThenV = Then->codegen();
645   if (!ThenV)
646     return nullptr;
647 
648   Builder->CreateBr(MergeBB);
649   // Codegen of 'Then' can change the current block, update ThenBB for the PHI.
650   ThenBB = Builder->GetInsertBlock();
651 
652   // Emit else block.
653   TheFunction->getBasicBlockList().push_back(ElseBB);
654   Builder->SetInsertPoint(ElseBB);
655 
656   Value *ElseV = Else->codegen();
657   if (!ElseV)
658     return nullptr;
659 
660   Builder->CreateBr(MergeBB);
661   // Codegen of 'Else' can change the current block, update ElseBB for the PHI.
662   ElseBB = Builder->GetInsertBlock();
663 
664   // Emit merge block.
665   TheFunction->getBasicBlockList().push_back(MergeBB);
666   Builder->SetInsertPoint(MergeBB);
667   PHINode *PN = Builder->CreatePHI(Type::getDoubleTy(*TheContext), 2, "iftmp");
668 
669   PN->addIncoming(ThenV, ThenBB);
670   PN->addIncoming(ElseV, ElseBB);
671   return PN;
672 }
673 
674 // Output for-loop as:
675 //   ...
676 //   start = startexpr
677 //   goto loop
678 // loop:
679 //   variable = phi [start, loopheader], [nextvariable, loopend]
680 //   ...
681 //   bodyexpr
682 //   ...
683 // loopend:
684 //   step = stepexpr
685 //   nextvariable = variable + step
686 //   endcond = endexpr
687 //   br endcond, loop, endloop
688 // outloop:
codegen()689 Value *ForExprAST::codegen() {
690   // Emit the start code first, without 'variable' in scope.
691   Value *StartVal = Start->codegen();
692   if (!StartVal)
693     return nullptr;
694 
695   // Make the new basic block for the loop header, inserting after current
696   // block.
697   Function *TheFunction = Builder->GetInsertBlock()->getParent();
698   BasicBlock *PreheaderBB = Builder->GetInsertBlock();
699   BasicBlock *LoopBB = BasicBlock::Create(*TheContext, "loop", TheFunction);
700 
701   // Insert an explicit fall through from the current block to the LoopBB.
702   Builder->CreateBr(LoopBB);
703 
704   // Start insertion in LoopBB.
705   Builder->SetInsertPoint(LoopBB);
706 
707   // Start the PHI node with an entry for Start.
708   PHINode *Variable =
709       Builder->CreatePHI(Type::getDoubleTy(*TheContext), 2, VarName);
710   Variable->addIncoming(StartVal, PreheaderBB);
711 
712   // Within the loop, the variable is defined equal to the PHI node.  If it
713   // shadows an existing variable, we have to restore it, so save it now.
714   Value *OldVal = NamedValues[VarName];
715   NamedValues[VarName] = Variable;
716 
717   // Emit the body of the loop.  This, like any other expr, can change the
718   // current BB.  Note that we ignore the value computed by the body, but don't
719   // allow an error.
720   if (!Body->codegen())
721     return nullptr;
722 
723   // Emit the step value.
724   Value *StepVal = nullptr;
725   if (Step) {
726     StepVal = Step->codegen();
727     if (!StepVal)
728       return nullptr;
729   } else {
730     // If not specified, use 1.0.
731     StepVal = ConstantFP::get(*TheContext, APFloat(1.0));
732   }
733 
734   Value *NextVar = Builder->CreateFAdd(Variable, StepVal, "nextvar");
735 
736   // Compute the end condition.
737   Value *EndCond = End->codegen();
738   if (!EndCond)
739     return nullptr;
740 
741   // Convert condition to a bool by comparing non-equal to 0.0.
742   EndCond = Builder->CreateFCmpONE(
743       EndCond, ConstantFP::get(*TheContext, APFloat(0.0)), "loopcond");
744 
745   // Create the "after loop" block and insert it.
746   BasicBlock *LoopEndBB = Builder->GetInsertBlock();
747   BasicBlock *AfterBB =
748       BasicBlock::Create(*TheContext, "afterloop", TheFunction);
749 
750   // Insert the conditional branch into the end of LoopEndBB.
751   Builder->CreateCondBr(EndCond, LoopBB, AfterBB);
752 
753   // Any new code will be inserted in AfterBB.
754   Builder->SetInsertPoint(AfterBB);
755 
756   // Add a new entry to the PHI node for the backedge.
757   Variable->addIncoming(NextVar, LoopEndBB);
758 
759   // Restore the unshadowed variable.
760   if (OldVal)
761     NamedValues[VarName] = OldVal;
762   else
763     NamedValues.erase(VarName);
764 
765   // for expr always returns 0.0.
766   return Constant::getNullValue(Type::getDoubleTy(*TheContext));
767 }
768 
codegen()769 Function *PrototypeAST::codegen() {
770   // Make the function type:  double(double,double) etc.
771   std::vector<Type *> Doubles(Args.size(), Type::getDoubleTy(*TheContext));
772   FunctionType *FT =
773       FunctionType::get(Type::getDoubleTy(*TheContext), Doubles, false);
774 
775   Function *F =
776       Function::Create(FT, Function::ExternalLinkage, Name, TheModule.get());
777 
778   // Set names for all arguments.
779   unsigned Idx = 0;
780   for (auto &Arg : F->args())
781     Arg.setName(Args[Idx++]);
782 
783   return F;
784 }
785 
codegen()786 Function *FunctionAST::codegen() {
787   // Transfer ownership of the prototype to the FunctionProtos map, but keep a
788   // reference to it for use below.
789   auto &P = *Proto;
790   FunctionProtos[Proto->getName()] = std::move(Proto);
791   Function *TheFunction = getFunction(P.getName());
792   if (!TheFunction)
793     return nullptr;
794 
795   // Create a new basic block to start insertion into.
796   BasicBlock *BB = BasicBlock::Create(*TheContext, "entry", TheFunction);
797   Builder->SetInsertPoint(BB);
798 
799   // Record the function arguments in the NamedValues map.
800   NamedValues.clear();
801   for (auto &Arg : TheFunction->args())
802     NamedValues[std::string(Arg.getName())] = &Arg;
803 
804   if (Value *RetVal = Body->codegen()) {
805     // Finish off the function.
806     Builder->CreateRet(RetVal);
807 
808     // Validate the generated code, checking for consistency.
809     verifyFunction(*TheFunction);
810 
811     // Run the optimizer on the function.
812     TheFPM->run(*TheFunction);
813 
814     return TheFunction;
815   }
816 
817   // Error reading body, remove function.
818   TheFunction->eraseFromParent();
819   return nullptr;
820 }
821 
822 //===----------------------------------------------------------------------===//
823 // Top-Level parsing and JIT Driver
824 //===----------------------------------------------------------------------===//
825 
InitializeModuleAndPassManager()826 static void InitializeModuleAndPassManager() {
827   // Open a new module.
828   TheContext = std::make_unique<LLVMContext>();
829   TheModule = std::make_unique<Module>("my cool jit", *TheContext);
830   TheModule->setDataLayout(TheJIT->getDataLayout());
831 
832   // Create a new builder for the module.
833   Builder = std::make_unique<IRBuilder<>>(*TheContext);
834 
835   // Create a new pass manager attached to it.
836   TheFPM = std::make_unique<legacy::FunctionPassManager>(TheModule.get());
837 
838   // Do simple "peephole" optimizations and bit-twiddling optzns.
839   TheFPM->add(createInstructionCombiningPass());
840   // Reassociate expressions.
841   TheFPM->add(createReassociatePass());
842   // Eliminate Common SubExpressions.
843   TheFPM->add(createGVNPass());
844   // Simplify the control flow graph (deleting unreachable blocks, etc).
845   TheFPM->add(createCFGSimplificationPass());
846 
847   TheFPM->doInitialization();
848 }
849 
HandleDefinition()850 static void HandleDefinition() {
851   if (auto FnAST = ParseDefinition()) {
852     if (auto *FnIR = FnAST->codegen()) {
853       fprintf(stderr, "Read function definition:");
854       FnIR->print(errs());
855       fprintf(stderr, "\n");
856       ExitOnErr(TheJIT->addModule(
857           ThreadSafeModule(std::move(TheModule), std::move(TheContext))));
858       InitializeModuleAndPassManager();
859     }
860   } else {
861     // Skip token for error recovery.
862     getNextToken();
863   }
864 }
865 
HandleExtern()866 static void HandleExtern() {
867   if (auto ProtoAST = ParseExtern()) {
868     if (auto *FnIR = ProtoAST->codegen()) {
869       fprintf(stderr, "Read extern: ");
870       FnIR->print(errs());
871       fprintf(stderr, "\n");
872       FunctionProtos[ProtoAST->getName()] = std::move(ProtoAST);
873     }
874   } else {
875     // Skip token for error recovery.
876     getNextToken();
877   }
878 }
879 
HandleTopLevelExpression()880 static void HandleTopLevelExpression() {
881   // Evaluate a top-level expression into an anonymous function.
882   if (auto FnAST = ParseTopLevelExpr()) {
883     if (FnAST->codegen()) {
884       // Create a ResourceTracker to track JIT'd memory allocated to our
885       // anonymous expression -- that way we can free it after executing.
886       auto RT = TheJIT->getMainJITDylib().createResourceTracker();
887 
888       auto TSM = ThreadSafeModule(std::move(TheModule), std::move(TheContext));
889       ExitOnErr(TheJIT->addModule(std::move(TSM), RT));
890       InitializeModuleAndPassManager();
891 
892       // Search the JIT for the __anon_expr symbol.
893       auto ExprSymbol = ExitOnErr(TheJIT->lookup("__anon_expr"));
894 
895       // Get the symbol's address and cast it to the right type (takes no
896       // arguments, returns a double) so we can call it as a native function.
897       double (*FP)() = (double (*)())(intptr_t)ExprSymbol.getAddress();
898       fprintf(stderr, "Evaluated to %f\n", FP());
899 
900       // Delete the anonymous expression module from the JIT.
901       ExitOnErr(RT->remove());
902     }
903   } else {
904     // Skip token for error recovery.
905     getNextToken();
906   }
907 }
908 
909 /// top ::= definition | external | expression | ';'
MainLoop()910 static void MainLoop() {
911   while (true) {
912     fprintf(stderr, "ready> ");
913     switch (CurTok) {
914     case tok_eof:
915       return;
916     case ';': // ignore top-level semicolons.
917       getNextToken();
918       break;
919     case tok_def:
920       HandleDefinition();
921       break;
922     case tok_extern:
923       HandleExtern();
924       break;
925     default:
926       HandleTopLevelExpression();
927       break;
928     }
929   }
930 }
931 
932 //===----------------------------------------------------------------------===//
933 // "Library" functions that can be "extern'd" from user code.
934 //===----------------------------------------------------------------------===//
935 
936 #ifdef _WIN32
937 #define DLLEXPORT __declspec(dllexport)
938 #else
939 #define DLLEXPORT
940 #endif
941 
942 /// putchard - putchar that takes a double and returns 0.
putchard(double X)943 extern "C" DLLEXPORT double putchard(double X) {
944   fputc((char)X, stderr);
945   return 0;
946 }
947 
948 /// printd - printf that takes a double prints it as "%f\n", returning 0.
printd(double X)949 extern "C" DLLEXPORT double printd(double X) {
950   fprintf(stderr, "%f\n", X);
951   return 0;
952 }
953 
954 //===----------------------------------------------------------------------===//
955 // Main driver code.
956 //===----------------------------------------------------------------------===//
957 
main()958 int main() {
959   InitializeNativeTarget();
960   InitializeNativeTargetAsmPrinter();
961   InitializeNativeTargetAsmParser();
962 
963   // Install standard binary operators.
964   // 1 is lowest precedence.
965   BinopPrecedence['<'] = 10;
966   BinopPrecedence['+'] = 20;
967   BinopPrecedence['-'] = 20;
968   BinopPrecedence['*'] = 40; // highest.
969 
970   // Prime the first token.
971   fprintf(stderr, "ready> ");
972   getNextToken();
973 
974   TheJIT = ExitOnErr(KaleidoscopeJIT::Create());
975 
976   InitializeModuleAndPassManager();
977 
978   // Run the main "interpreter loop" now.
979   MainLoop();
980 
981   return 0;
982 }
983