[InstCombine] Signed saturation patterns
[llvm-complete.git] / examples / Kaleidoscope / Chapter9 / toy.cpp
blob21c8993e1a064773b8d091ac4193ae8892bf3cd4
1 #include "llvm/ADT/STLExtras.h"
2 #include "llvm/Analysis/BasicAliasAnalysis.h"
3 #include "llvm/Analysis/Passes.h"
4 #include "llvm/IR/DIBuilder.h"
5 #include "llvm/IR/IRBuilder.h"
6 #include "llvm/IR/LLVMContext.h"
7 #include "llvm/IR/LegacyPassManager.h"
8 #include "llvm/IR/Module.h"
9 #include "llvm/IR/Verifier.h"
10 #include "llvm/Support/TargetSelect.h"
11 #include "llvm/Transforms/Scalar.h"
12 #include <cctype>
13 #include <cstdio>
14 #include <map>
15 #include <string>
16 #include <vector>
17 #include "../include/KaleidoscopeJIT.h"
19 using namespace llvm;
20 using namespace llvm::orc;
22 //===----------------------------------------------------------------------===//
23 // Lexer
24 //===----------------------------------------------------------------------===//
26 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
27 // of these for known things.
28 enum Token {
29 tok_eof = -1,
31 // commands
32 tok_def = -2,
33 tok_extern = -3,
35 // primary
36 tok_identifier = -4,
37 tok_number = -5,
39 // control
40 tok_if = -6,
41 tok_then = -7,
42 tok_else = -8,
43 tok_for = -9,
44 tok_in = -10,
46 // operators
47 tok_binary = -11,
48 tok_unary = -12,
50 // var definition
51 tok_var = -13
54 std::string getTokName(int Tok) {
55 switch (Tok) {
56 case tok_eof:
57 return "eof";
58 case tok_def:
59 return "def";
60 case tok_extern:
61 return "extern";
62 case tok_identifier:
63 return "identifier";
64 case tok_number:
65 return "number";
66 case tok_if:
67 return "if";
68 case tok_then:
69 return "then";
70 case tok_else:
71 return "else";
72 case tok_for:
73 return "for";
74 case tok_in:
75 return "in";
76 case tok_binary:
77 return "binary";
78 case tok_unary:
79 return "unary";
80 case tok_var:
81 return "var";
83 return std::string(1, (char)Tok);
86 namespace {
87 class PrototypeAST;
88 class ExprAST;
90 static LLVMContext TheContext;
91 static IRBuilder<> Builder(TheContext);
92 struct DebugInfo {
93 DICompileUnit *TheCU;
94 DIType *DblTy;
95 std::vector<DIScope *> LexicalBlocks;
97 void emitLocation(ExprAST *AST);
98 DIType *getDoubleTy();
99 } KSDbgInfo;
101 struct SourceLocation {
102 int Line;
103 int Col;
105 static SourceLocation CurLoc;
106 static SourceLocation LexLoc = {1, 0};
108 static int advance() {
109 int LastChar = getchar();
111 if (LastChar == '\n' || LastChar == '\r') {
112 LexLoc.Line++;
113 LexLoc.Col = 0;
114 } else
115 LexLoc.Col++;
116 return LastChar;
119 static std::string IdentifierStr; // Filled in if tok_identifier
120 static double NumVal; // Filled in if tok_number
122 /// gettok - Return the next token from standard input.
123 static int gettok() {
124 static int LastChar = ' ';
126 // Skip any whitespace.
127 while (isspace(LastChar))
128 LastChar = advance();
130 CurLoc = LexLoc;
132 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
133 IdentifierStr = LastChar;
134 while (isalnum((LastChar = advance())))
135 IdentifierStr += LastChar;
137 if (IdentifierStr == "def")
138 return tok_def;
139 if (IdentifierStr == "extern")
140 return tok_extern;
141 if (IdentifierStr == "if")
142 return tok_if;
143 if (IdentifierStr == "then")
144 return tok_then;
145 if (IdentifierStr == "else")
146 return tok_else;
147 if (IdentifierStr == "for")
148 return tok_for;
149 if (IdentifierStr == "in")
150 return tok_in;
151 if (IdentifierStr == "binary")
152 return tok_binary;
153 if (IdentifierStr == "unary")
154 return tok_unary;
155 if (IdentifierStr == "var")
156 return tok_var;
157 return tok_identifier;
160 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+
161 std::string NumStr;
162 do {
163 NumStr += LastChar;
164 LastChar = advance();
165 } while (isdigit(LastChar) || LastChar == '.');
167 NumVal = strtod(NumStr.c_str(), nullptr);
168 return tok_number;
171 if (LastChar == '#') {
172 // Comment until end of line.
174 LastChar = advance();
175 while (LastChar != EOF && LastChar != '\n' && LastChar != '\r');
177 if (LastChar != EOF)
178 return gettok();
181 // Check for end of file. Don't eat the EOF.
182 if (LastChar == EOF)
183 return tok_eof;
185 // Otherwise, just return the character as its ascii value.
186 int ThisChar = LastChar;
187 LastChar = advance();
188 return ThisChar;
191 //===----------------------------------------------------------------------===//
192 // Abstract Syntax Tree (aka Parse Tree)
193 //===----------------------------------------------------------------------===//
194 namespace {
196 raw_ostream &indent(raw_ostream &O, int size) {
197 return O << std::string(size, ' ');
200 /// ExprAST - Base class for all expression nodes.
201 class ExprAST {
202 SourceLocation Loc;
204 public:
205 ExprAST(SourceLocation Loc = CurLoc) : Loc(Loc) {}
206 virtual ~ExprAST() {}
207 virtual Value *codegen() = 0;
208 int getLine() const { return Loc.Line; }
209 int getCol() const { return Loc.Col; }
210 virtual raw_ostream &dump(raw_ostream &out, int ind) {
211 return out << ':' << getLine() << ':' << getCol() << '\n';
215 /// NumberExprAST - Expression class for numeric literals like "1.0".
216 class NumberExprAST : public ExprAST {
217 double Val;
219 public:
220 NumberExprAST(double Val) : Val(Val) {}
221 raw_ostream &dump(raw_ostream &out, int ind) override {
222 return ExprAST::dump(out << Val, ind);
224 Value *codegen() override;
227 /// VariableExprAST - Expression class for referencing a variable, like "a".
228 class VariableExprAST : public ExprAST {
229 std::string Name;
231 public:
232 VariableExprAST(SourceLocation Loc, const std::string &Name)
233 : ExprAST(Loc), Name(Name) {}
234 const std::string &getName() const { return Name; }
235 Value *codegen() override;
236 raw_ostream &dump(raw_ostream &out, int ind) override {
237 return ExprAST::dump(out << Name, ind);
241 /// UnaryExprAST - Expression class for a unary operator.
242 class UnaryExprAST : public ExprAST {
243 char Opcode;
244 std::unique_ptr<ExprAST> Operand;
246 public:
247 UnaryExprAST(char Opcode, std::unique_ptr<ExprAST> Operand)
248 : Opcode(Opcode), Operand(std::move(Operand)) {}
249 Value *codegen() override;
250 raw_ostream &dump(raw_ostream &out, int ind) override {
251 ExprAST::dump(out << "unary" << Opcode, ind);
252 Operand->dump(out, ind + 1);
253 return out;
257 /// BinaryExprAST - Expression class for a binary operator.
258 class BinaryExprAST : public ExprAST {
259 char Op;
260 std::unique_ptr<ExprAST> LHS, RHS;
262 public:
263 BinaryExprAST(SourceLocation Loc, char Op, std::unique_ptr<ExprAST> LHS,
264 std::unique_ptr<ExprAST> RHS)
265 : ExprAST(Loc), Op(Op), LHS(std::move(LHS)), RHS(std::move(RHS)) {}
266 Value *codegen() override;
267 raw_ostream &dump(raw_ostream &out, int ind) override {
268 ExprAST::dump(out << "binary" << Op, ind);
269 LHS->dump(indent(out, ind) << "LHS:", ind + 1);
270 RHS->dump(indent(out, ind) << "RHS:", ind + 1);
271 return out;
275 /// CallExprAST - Expression class for function calls.
276 class CallExprAST : public ExprAST {
277 std::string Callee;
278 std::vector<std::unique_ptr<ExprAST>> Args;
280 public:
281 CallExprAST(SourceLocation Loc, const std::string &Callee,
282 std::vector<std::unique_ptr<ExprAST>> Args)
283 : ExprAST(Loc), Callee(Callee), Args(std::move(Args)) {}
284 Value *codegen() override;
285 raw_ostream &dump(raw_ostream &out, int ind) override {
286 ExprAST::dump(out << "call " << Callee, ind);
287 for (const auto &Arg : Args)
288 Arg->dump(indent(out, ind + 1), ind + 1);
289 return out;
293 /// IfExprAST - Expression class for if/then/else.
294 class IfExprAST : public ExprAST {
295 std::unique_ptr<ExprAST> Cond, Then, Else;
297 public:
298 IfExprAST(SourceLocation Loc, std::unique_ptr<ExprAST> Cond,
299 std::unique_ptr<ExprAST> Then, std::unique_ptr<ExprAST> Else)
300 : ExprAST(Loc), Cond(std::move(Cond)), Then(std::move(Then)),
301 Else(std::move(Else)) {}
302 Value *codegen() override;
303 raw_ostream &dump(raw_ostream &out, int ind) override {
304 ExprAST::dump(out << "if", ind);
305 Cond->dump(indent(out, ind) << "Cond:", ind + 1);
306 Then->dump(indent(out, ind) << "Then:", ind + 1);
307 Else->dump(indent(out, ind) << "Else:", ind + 1);
308 return out;
312 /// ForExprAST - Expression class for for/in.
313 class ForExprAST : public ExprAST {
314 std::string VarName;
315 std::unique_ptr<ExprAST> Start, End, Step, Body;
317 public:
318 ForExprAST(const std::string &VarName, std::unique_ptr<ExprAST> Start,
319 std::unique_ptr<ExprAST> End, std::unique_ptr<ExprAST> Step,
320 std::unique_ptr<ExprAST> Body)
321 : VarName(VarName), Start(std::move(Start)), End(std::move(End)),
322 Step(std::move(Step)), Body(std::move(Body)) {}
323 Value *codegen() override;
324 raw_ostream &dump(raw_ostream &out, int ind) override {
325 ExprAST::dump(out << "for", ind);
326 Start->dump(indent(out, ind) << "Cond:", ind + 1);
327 End->dump(indent(out, ind) << "End:", ind + 1);
328 Step->dump(indent(out, ind) << "Step:", ind + 1);
329 Body->dump(indent(out, ind) << "Body:", ind + 1);
330 return out;
334 /// VarExprAST - Expression class for var/in
335 class VarExprAST : public ExprAST {
336 std::vector<std::pair<std::string, std::unique_ptr<ExprAST>>> VarNames;
337 std::unique_ptr<ExprAST> Body;
339 public:
340 VarExprAST(
341 std::vector<std::pair<std::string, std::unique_ptr<ExprAST>>> VarNames,
342 std::unique_ptr<ExprAST> Body)
343 : VarNames(std::move(VarNames)), Body(std::move(Body)) {}
344 Value *codegen() override;
345 raw_ostream &dump(raw_ostream &out, int ind) override {
346 ExprAST::dump(out << "var", ind);
347 for (const auto &NamedVar : VarNames)
348 NamedVar.second->dump(indent(out, ind) << NamedVar.first << ':', ind + 1);
349 Body->dump(indent(out, ind) << "Body:", ind + 1);
350 return out;
354 /// PrototypeAST - This class represents the "prototype" for a function,
355 /// which captures its name, and its argument names (thus implicitly the number
356 /// of arguments the function takes), as well as if it is an operator.
357 class PrototypeAST {
358 std::string Name;
359 std::vector<std::string> Args;
360 bool IsOperator;
361 unsigned Precedence; // Precedence if a binary op.
362 int Line;
364 public:
365 PrototypeAST(SourceLocation Loc, const std::string &Name,
366 std::vector<std::string> Args, bool IsOperator = false,
367 unsigned Prec = 0)
368 : Name(Name), Args(std::move(Args)), IsOperator(IsOperator),
369 Precedence(Prec), Line(Loc.Line) {}
370 Function *codegen();
371 const std::string &getName() const { return Name; }
373 bool isUnaryOp() const { return IsOperator && Args.size() == 1; }
374 bool isBinaryOp() const { return IsOperator && Args.size() == 2; }
376 char getOperatorName() const {
377 assert(isUnaryOp() || isBinaryOp());
378 return Name[Name.size() - 1];
381 unsigned getBinaryPrecedence() const { return Precedence; }
382 int getLine() const { return Line; }
385 /// FunctionAST - This class represents a function definition itself.
386 class FunctionAST {
387 std::unique_ptr<PrototypeAST> Proto;
388 std::unique_ptr<ExprAST> Body;
390 public:
391 FunctionAST(std::unique_ptr<PrototypeAST> Proto,
392 std::unique_ptr<ExprAST> Body)
393 : Proto(std::move(Proto)), Body(std::move(Body)) {}
394 Function *codegen();
395 raw_ostream &dump(raw_ostream &out, int ind) {
396 indent(out, ind) << "FunctionAST\n";
397 ++ind;
398 indent(out, ind) << "Body:";
399 return Body ? Body->dump(out, ind) : out << "null\n";
402 } // end anonymous namespace
404 //===----------------------------------------------------------------------===//
405 // Parser
406 //===----------------------------------------------------------------------===//
408 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
409 /// token the parser is looking at. getNextToken reads another token from the
410 /// lexer and updates CurTok with its results.
411 static int CurTok;
412 static int getNextToken() { return CurTok = gettok(); }
414 /// BinopPrecedence - This holds the precedence for each binary operator that is
415 /// defined.
416 static std::map<char, int> BinopPrecedence;
418 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
419 static int GetTokPrecedence() {
420 if (!isascii(CurTok))
421 return -1;
423 // Make sure it's a declared binop.
424 int TokPrec = BinopPrecedence[CurTok];
425 if (TokPrec <= 0)
426 return -1;
427 return TokPrec;
430 /// LogError* - These are little helper functions for error handling.
431 std::unique_ptr<ExprAST> LogError(const char *Str) {
432 fprintf(stderr, "Error: %s\n", Str);
433 return nullptr;
436 std::unique_ptr<PrototypeAST> LogErrorP(const char *Str) {
437 LogError(Str);
438 return nullptr;
441 static std::unique_ptr<ExprAST> ParseExpression();
443 /// numberexpr ::= number
444 static std::unique_ptr<ExprAST> ParseNumberExpr() {
445 auto Result = std::make_unique<NumberExprAST>(NumVal);
446 getNextToken(); // consume the number
447 return std::move(Result);
450 /// parenexpr ::= '(' expression ')'
451 static std::unique_ptr<ExprAST> ParseParenExpr() {
452 getNextToken(); // eat (.
453 auto V = ParseExpression();
454 if (!V)
455 return nullptr;
457 if (CurTok != ')')
458 return LogError("expected ')'");
459 getNextToken(); // eat ).
460 return V;
463 /// identifierexpr
464 /// ::= identifier
465 /// ::= identifier '(' expression* ')'
466 static std::unique_ptr<ExprAST> ParseIdentifierExpr() {
467 std::string IdName = IdentifierStr;
469 SourceLocation LitLoc = CurLoc;
471 getNextToken(); // eat identifier.
473 if (CurTok != '(') // Simple variable ref.
474 return std::make_unique<VariableExprAST>(LitLoc, IdName);
476 // Call.
477 getNextToken(); // eat (
478 std::vector<std::unique_ptr<ExprAST>> Args;
479 if (CurTok != ')') {
480 while (1) {
481 if (auto Arg = ParseExpression())
482 Args.push_back(std::move(Arg));
483 else
484 return nullptr;
486 if (CurTok == ')')
487 break;
489 if (CurTok != ',')
490 return LogError("Expected ')' or ',' in argument list");
491 getNextToken();
495 // Eat the ')'.
496 getNextToken();
498 return std::make_unique<CallExprAST>(LitLoc, IdName, std::move(Args));
501 /// ifexpr ::= 'if' expression 'then' expression 'else' expression
502 static std::unique_ptr<ExprAST> ParseIfExpr() {
503 SourceLocation IfLoc = CurLoc;
505 getNextToken(); // eat the if.
507 // condition.
508 auto Cond = ParseExpression();
509 if (!Cond)
510 return nullptr;
512 if (CurTok != tok_then)
513 return LogError("expected then");
514 getNextToken(); // eat the then
516 auto Then = ParseExpression();
517 if (!Then)
518 return nullptr;
520 if (CurTok != tok_else)
521 return LogError("expected else");
523 getNextToken();
525 auto Else = ParseExpression();
526 if (!Else)
527 return nullptr;
529 return std::make_unique<IfExprAST>(IfLoc, std::move(Cond), std::move(Then),
530 std::move(Else));
533 /// forexpr ::= 'for' identifier '=' expr ',' expr (',' expr)? 'in' expression
534 static std::unique_ptr<ExprAST> ParseForExpr() {
535 getNextToken(); // eat the for.
537 if (CurTok != tok_identifier)
538 return LogError("expected identifier after for");
540 std::string IdName = IdentifierStr;
541 getNextToken(); // eat identifier.
543 if (CurTok != '=')
544 return LogError("expected '=' after for");
545 getNextToken(); // eat '='.
547 auto Start = ParseExpression();
548 if (!Start)
549 return nullptr;
550 if (CurTok != ',')
551 return LogError("expected ',' after for start value");
552 getNextToken();
554 auto End = ParseExpression();
555 if (!End)
556 return nullptr;
558 // The step value is optional.
559 std::unique_ptr<ExprAST> Step;
560 if (CurTok == ',') {
561 getNextToken();
562 Step = ParseExpression();
563 if (!Step)
564 return nullptr;
567 if (CurTok != tok_in)
568 return LogError("expected 'in' after for");
569 getNextToken(); // eat 'in'.
571 auto Body = ParseExpression();
572 if (!Body)
573 return nullptr;
575 return std::make_unique<ForExprAST>(IdName, std::move(Start), std::move(End),
576 std::move(Step), std::move(Body));
579 /// varexpr ::= 'var' identifier ('=' expression)?
580 // (',' identifier ('=' expression)?)* 'in' expression
581 static std::unique_ptr<ExprAST> ParseVarExpr() {
582 getNextToken(); // eat the var.
584 std::vector<std::pair<std::string, std::unique_ptr<ExprAST>>> VarNames;
586 // At least one variable name is required.
587 if (CurTok != tok_identifier)
588 return LogError("expected identifier after var");
590 while (1) {
591 std::string Name = IdentifierStr;
592 getNextToken(); // eat identifier.
594 // Read the optional initializer.
595 std::unique_ptr<ExprAST> Init = nullptr;
596 if (CurTok == '=') {
597 getNextToken(); // eat the '='.
599 Init = ParseExpression();
600 if (!Init)
601 return nullptr;
604 VarNames.push_back(std::make_pair(Name, std::move(Init)));
606 // End of var list, exit loop.
607 if (CurTok != ',')
608 break;
609 getNextToken(); // eat the ','.
611 if (CurTok != tok_identifier)
612 return LogError("expected identifier list after var");
615 // At this point, we have to have 'in'.
616 if (CurTok != tok_in)
617 return LogError("expected 'in' keyword after 'var'");
618 getNextToken(); // eat 'in'.
620 auto Body = ParseExpression();
621 if (!Body)
622 return nullptr;
624 return std::make_unique<VarExprAST>(std::move(VarNames), std::move(Body));
627 /// primary
628 /// ::= identifierexpr
629 /// ::= numberexpr
630 /// ::= parenexpr
631 /// ::= ifexpr
632 /// ::= forexpr
633 /// ::= varexpr
634 static std::unique_ptr<ExprAST> ParsePrimary() {
635 switch (CurTok) {
636 default:
637 return LogError("unknown token when expecting an expression");
638 case tok_identifier:
639 return ParseIdentifierExpr();
640 case tok_number:
641 return ParseNumberExpr();
642 case '(':
643 return ParseParenExpr();
644 case tok_if:
645 return ParseIfExpr();
646 case tok_for:
647 return ParseForExpr();
648 case tok_var:
649 return ParseVarExpr();
653 /// unary
654 /// ::= primary
655 /// ::= '!' unary
656 static std::unique_ptr<ExprAST> ParseUnary() {
657 // If the current token is not an operator, it must be a primary expr.
658 if (!isascii(CurTok) || CurTok == '(' || CurTok == ',')
659 return ParsePrimary();
661 // If this is a unary operator, read it.
662 int Opc = CurTok;
663 getNextToken();
664 if (auto Operand = ParseUnary())
665 return std::make_unique<UnaryExprAST>(Opc, std::move(Operand));
666 return nullptr;
669 /// binoprhs
670 /// ::= ('+' unary)*
671 static std::unique_ptr<ExprAST> ParseBinOpRHS(int ExprPrec,
672 std::unique_ptr<ExprAST> LHS) {
673 // If this is a binop, find its precedence.
674 while (1) {
675 int TokPrec = GetTokPrecedence();
677 // If this is a binop that binds at least as tightly as the current binop,
678 // consume it, otherwise we are done.
679 if (TokPrec < ExprPrec)
680 return LHS;
682 // Okay, we know this is a binop.
683 int BinOp = CurTok;
684 SourceLocation BinLoc = CurLoc;
685 getNextToken(); // eat binop
687 // Parse the unary expression after the binary operator.
688 auto RHS = ParseUnary();
689 if (!RHS)
690 return nullptr;
692 // If BinOp binds less tightly with RHS than the operator after RHS, let
693 // the pending operator take RHS as its LHS.
694 int NextPrec = GetTokPrecedence();
695 if (TokPrec < NextPrec) {
696 RHS = ParseBinOpRHS(TokPrec + 1, std::move(RHS));
697 if (!RHS)
698 return nullptr;
701 // Merge LHS/RHS.
702 LHS = std::make_unique<BinaryExprAST>(BinLoc, BinOp, std::move(LHS),
703 std::move(RHS));
707 /// expression
708 /// ::= unary binoprhs
710 static std::unique_ptr<ExprAST> ParseExpression() {
711 auto LHS = ParseUnary();
712 if (!LHS)
713 return nullptr;
715 return ParseBinOpRHS(0, std::move(LHS));
718 /// prototype
719 /// ::= id '(' id* ')'
720 /// ::= binary LETTER number? (id, id)
721 /// ::= unary LETTER (id)
722 static std::unique_ptr<PrototypeAST> ParsePrototype() {
723 std::string FnName;
725 SourceLocation FnLoc = CurLoc;
727 unsigned Kind = 0; // 0 = identifier, 1 = unary, 2 = binary.
728 unsigned BinaryPrecedence = 30;
730 switch (CurTok) {
731 default:
732 return LogErrorP("Expected function name in prototype");
733 case tok_identifier:
734 FnName = IdentifierStr;
735 Kind = 0;
736 getNextToken();
737 break;
738 case tok_unary:
739 getNextToken();
740 if (!isascii(CurTok))
741 return LogErrorP("Expected unary operator");
742 FnName = "unary";
743 FnName += (char)CurTok;
744 Kind = 1;
745 getNextToken();
746 break;
747 case tok_binary:
748 getNextToken();
749 if (!isascii(CurTok))
750 return LogErrorP("Expected binary operator");
751 FnName = "binary";
752 FnName += (char)CurTok;
753 Kind = 2;
754 getNextToken();
756 // Read the precedence if present.
757 if (CurTok == tok_number) {
758 if (NumVal < 1 || NumVal > 100)
759 return LogErrorP("Invalid precedence: must be 1..100");
760 BinaryPrecedence = (unsigned)NumVal;
761 getNextToken();
763 break;
766 if (CurTok != '(')
767 return LogErrorP("Expected '(' in prototype");
769 std::vector<std::string> ArgNames;
770 while (getNextToken() == tok_identifier)
771 ArgNames.push_back(IdentifierStr);
772 if (CurTok != ')')
773 return LogErrorP("Expected ')' in prototype");
775 // success.
776 getNextToken(); // eat ')'.
778 // Verify right number of names for operator.
779 if (Kind && ArgNames.size() != Kind)
780 return LogErrorP("Invalid number of operands for operator");
782 return std::make_unique<PrototypeAST>(FnLoc, FnName, ArgNames, Kind != 0,
783 BinaryPrecedence);
786 /// definition ::= 'def' prototype expression
787 static std::unique_ptr<FunctionAST> ParseDefinition() {
788 getNextToken(); // eat def.
789 auto Proto = ParsePrototype();
790 if (!Proto)
791 return nullptr;
793 if (auto E = ParseExpression())
794 return std::make_unique<FunctionAST>(std::move(Proto), std::move(E));
795 return nullptr;
798 /// toplevelexpr ::= expression
799 static std::unique_ptr<FunctionAST> ParseTopLevelExpr() {
800 SourceLocation FnLoc = CurLoc;
801 if (auto E = ParseExpression()) {
802 // Make an anonymous proto.
803 auto Proto = std::make_unique<PrototypeAST>(FnLoc, "__anon_expr",
804 std::vector<std::string>());
805 return std::make_unique<FunctionAST>(std::move(Proto), std::move(E));
807 return nullptr;
810 /// external ::= 'extern' prototype
811 static std::unique_ptr<PrototypeAST> ParseExtern() {
812 getNextToken(); // eat extern.
813 return ParsePrototype();
816 //===----------------------------------------------------------------------===//
817 // Debug Info Support
818 //===----------------------------------------------------------------------===//
820 static std::unique_ptr<DIBuilder> DBuilder;
822 DIType *DebugInfo::getDoubleTy() {
823 if (DblTy)
824 return DblTy;
826 DblTy = DBuilder->createBasicType("double", 64, dwarf::DW_ATE_float);
827 return DblTy;
830 void DebugInfo::emitLocation(ExprAST *AST) {
831 if (!AST)
832 return Builder.SetCurrentDebugLocation(DebugLoc());
833 DIScope *Scope;
834 if (LexicalBlocks.empty())
835 Scope = TheCU;
836 else
837 Scope = LexicalBlocks.back();
838 Builder.SetCurrentDebugLocation(
839 DebugLoc::get(AST->getLine(), AST->getCol(), Scope));
842 static DISubroutineType *CreateFunctionType(unsigned NumArgs, DIFile *Unit) {
843 SmallVector<Metadata *, 8> EltTys;
844 DIType *DblTy = KSDbgInfo.getDoubleTy();
846 // Add the result type.
847 EltTys.push_back(DblTy);
849 for (unsigned i = 0, e = NumArgs; i != e; ++i)
850 EltTys.push_back(DblTy);
852 return DBuilder->createSubroutineType(DBuilder->getOrCreateTypeArray(EltTys));
855 //===----------------------------------------------------------------------===//
856 // Code Generation
857 //===----------------------------------------------------------------------===//
859 static std::unique_ptr<Module> TheModule;
860 static std::map<std::string, AllocaInst *> NamedValues;
861 static std::unique_ptr<KaleidoscopeJIT> TheJIT;
862 static std::map<std::string, std::unique_ptr<PrototypeAST>> FunctionProtos;
864 Value *LogErrorV(const char *Str) {
865 LogError(Str);
866 return nullptr;
869 Function *getFunction(std::string Name) {
870 // First, see if the function has already been added to the current module.
871 if (auto *F = TheModule->getFunction(Name))
872 return F;
874 // If not, check whether we can codegen the declaration from some existing
875 // prototype.
876 auto FI = FunctionProtos.find(Name);
877 if (FI != FunctionProtos.end())
878 return FI->second->codegen();
880 // If no existing prototype exists, return null.
881 return nullptr;
884 /// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of
885 /// the function. This is used for mutable variables etc.
886 static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction,
887 const std::string &VarName) {
888 IRBuilder<> TmpB(&TheFunction->getEntryBlock(),
889 TheFunction->getEntryBlock().begin());
890 return TmpB.CreateAlloca(Type::getDoubleTy(TheContext), nullptr,
891 VarName.c_str());
894 Value *NumberExprAST::codegen() {
895 KSDbgInfo.emitLocation(this);
896 return ConstantFP::get(TheContext, APFloat(Val));
899 Value *VariableExprAST::codegen() {
900 // Look this variable up in the function.
901 Value *V = NamedValues[Name];
902 if (!V)
903 return LogErrorV("Unknown variable name");
905 KSDbgInfo.emitLocation(this);
906 // Load the value.
907 return Builder.CreateLoad(V, Name.c_str());
910 Value *UnaryExprAST::codegen() {
911 Value *OperandV = Operand->codegen();
912 if (!OperandV)
913 return nullptr;
915 Function *F = getFunction(std::string("unary") + Opcode);
916 if (!F)
917 return LogErrorV("Unknown unary operator");
919 KSDbgInfo.emitLocation(this);
920 return Builder.CreateCall(F, OperandV, "unop");
923 Value *BinaryExprAST::codegen() {
924 KSDbgInfo.emitLocation(this);
926 // Special case '=' because we don't want to emit the LHS as an expression.
927 if (Op == '=') {
928 // Assignment requires the LHS to be an identifier.
929 // This assume we're building without RTTI because LLVM builds that way by
930 // default. If you build LLVM with RTTI this can be changed to a
931 // dynamic_cast for automatic error checking.
932 VariableExprAST *LHSE = static_cast<VariableExprAST *>(LHS.get());
933 if (!LHSE)
934 return LogErrorV("destination of '=' must be a variable");
935 // Codegen the RHS.
936 Value *Val = RHS->codegen();
937 if (!Val)
938 return nullptr;
940 // Look up the name.
941 Value *Variable = NamedValues[LHSE->getName()];
942 if (!Variable)
943 return LogErrorV("Unknown variable name");
945 Builder.CreateStore(Val, Variable);
946 return Val;
949 Value *L = LHS->codegen();
950 Value *R = RHS->codegen();
951 if (!L || !R)
952 return nullptr;
954 switch (Op) {
955 case '+':
956 return Builder.CreateFAdd(L, R, "addtmp");
957 case '-':
958 return Builder.CreateFSub(L, R, "subtmp");
959 case '*':
960 return Builder.CreateFMul(L, R, "multmp");
961 case '<':
962 L = Builder.CreateFCmpULT(L, R, "cmptmp");
963 // Convert bool 0/1 to double 0.0 or 1.0
964 return Builder.CreateUIToFP(L, Type::getDoubleTy(TheContext), "booltmp");
965 default:
966 break;
969 // If it wasn't a builtin binary operator, it must be a user defined one. Emit
970 // a call to it.
971 Function *F = getFunction(std::string("binary") + Op);
972 assert(F && "binary operator not found!");
974 Value *Ops[] = {L, R};
975 return Builder.CreateCall(F, Ops, "binop");
978 Value *CallExprAST::codegen() {
979 KSDbgInfo.emitLocation(this);
981 // Look up the name in the global module table.
982 Function *CalleeF = getFunction(Callee);
983 if (!CalleeF)
984 return LogErrorV("Unknown function referenced");
986 // If argument mismatch error.
987 if (CalleeF->arg_size() != Args.size())
988 return LogErrorV("Incorrect # arguments passed");
990 std::vector<Value *> ArgsV;
991 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
992 ArgsV.push_back(Args[i]->codegen());
993 if (!ArgsV.back())
994 return nullptr;
997 return Builder.CreateCall(CalleeF, ArgsV, "calltmp");
1000 Value *IfExprAST::codegen() {
1001 KSDbgInfo.emitLocation(this);
1003 Value *CondV = Cond->codegen();
1004 if (!CondV)
1005 return nullptr;
1007 // Convert condition to a bool by comparing non-equal to 0.0.
1008 CondV = Builder.CreateFCmpONE(
1009 CondV, ConstantFP::get(TheContext, APFloat(0.0)), "ifcond");
1011 Function *TheFunction = Builder.GetInsertBlock()->getParent();
1013 // Create blocks for the then and else cases. Insert the 'then' block at the
1014 // end of the function.
1015 BasicBlock *ThenBB = BasicBlock::Create(TheContext, "then", TheFunction);
1016 BasicBlock *ElseBB = BasicBlock::Create(TheContext, "else");
1017 BasicBlock *MergeBB = BasicBlock::Create(TheContext, "ifcont");
1019 Builder.CreateCondBr(CondV, ThenBB, ElseBB);
1021 // Emit then value.
1022 Builder.SetInsertPoint(ThenBB);
1024 Value *ThenV = Then->codegen();
1025 if (!ThenV)
1026 return nullptr;
1028 Builder.CreateBr(MergeBB);
1029 // Codegen of 'Then' can change the current block, update ThenBB for the PHI.
1030 ThenBB = Builder.GetInsertBlock();
1032 // Emit else block.
1033 TheFunction->getBasicBlockList().push_back(ElseBB);
1034 Builder.SetInsertPoint(ElseBB);
1036 Value *ElseV = Else->codegen();
1037 if (!ElseV)
1038 return nullptr;
1040 Builder.CreateBr(MergeBB);
1041 // Codegen of 'Else' can change the current block, update ElseBB for the PHI.
1042 ElseBB = Builder.GetInsertBlock();
1044 // Emit merge block.
1045 TheFunction->getBasicBlockList().push_back(MergeBB);
1046 Builder.SetInsertPoint(MergeBB);
1047 PHINode *PN = Builder.CreatePHI(Type::getDoubleTy(TheContext), 2, "iftmp");
1049 PN->addIncoming(ThenV, ThenBB);
1050 PN->addIncoming(ElseV, ElseBB);
1051 return PN;
1054 // Output for-loop as:
1055 // var = alloca double
1056 // ...
1057 // start = startexpr
1058 // store start -> var
1059 // goto loop
1060 // loop:
1061 // ...
1062 // bodyexpr
1063 // ...
1064 // loopend:
1065 // step = stepexpr
1066 // endcond = endexpr
1068 // curvar = load var
1069 // nextvar = curvar + step
1070 // store nextvar -> var
1071 // br endcond, loop, endloop
1072 // outloop:
1073 Value *ForExprAST::codegen() {
1074 Function *TheFunction = Builder.GetInsertBlock()->getParent();
1076 // Create an alloca for the variable in the entry block.
1077 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1079 KSDbgInfo.emitLocation(this);
1081 // Emit the start code first, without 'variable' in scope.
1082 Value *StartVal = Start->codegen();
1083 if (!StartVal)
1084 return nullptr;
1086 // Store the value into the alloca.
1087 Builder.CreateStore(StartVal, Alloca);
1089 // Make the new basic block for the loop header, inserting after current
1090 // block.
1091 BasicBlock *LoopBB = BasicBlock::Create(TheContext, "loop", TheFunction);
1093 // Insert an explicit fall through from the current block to the LoopBB.
1094 Builder.CreateBr(LoopBB);
1096 // Start insertion in LoopBB.
1097 Builder.SetInsertPoint(LoopBB);
1099 // Within the loop, the variable is defined equal to the PHI node. If it
1100 // shadows an existing variable, we have to restore it, so save it now.
1101 AllocaInst *OldVal = NamedValues[VarName];
1102 NamedValues[VarName] = Alloca;
1104 // Emit the body of the loop. This, like any other expr, can change the
1105 // current BB. Note that we ignore the value computed by the body, but don't
1106 // allow an error.
1107 if (!Body->codegen())
1108 return nullptr;
1110 // Emit the step value.
1111 Value *StepVal = nullptr;
1112 if (Step) {
1113 StepVal = Step->codegen();
1114 if (!StepVal)
1115 return nullptr;
1116 } else {
1117 // If not specified, use 1.0.
1118 StepVal = ConstantFP::get(TheContext, APFloat(1.0));
1121 // Compute the end condition.
1122 Value *EndCond = End->codegen();
1123 if (!EndCond)
1124 return nullptr;
1126 // Reload, increment, and restore the alloca. This handles the case where
1127 // the body of the loop mutates the variable.
1128 Value *CurVar = Builder.CreateLoad(Alloca, VarName.c_str());
1129 Value *NextVar = Builder.CreateFAdd(CurVar, StepVal, "nextvar");
1130 Builder.CreateStore(NextVar, Alloca);
1132 // Convert condition to a bool by comparing non-equal to 0.0.
1133 EndCond = Builder.CreateFCmpONE(
1134 EndCond, ConstantFP::get(TheContext, APFloat(0.0)), "loopcond");
1136 // Create the "after loop" block and insert it.
1137 BasicBlock *AfterBB =
1138 BasicBlock::Create(TheContext, "afterloop", TheFunction);
1140 // Insert the conditional branch into the end of LoopEndBB.
1141 Builder.CreateCondBr(EndCond, LoopBB, AfterBB);
1143 // Any new code will be inserted in AfterBB.
1144 Builder.SetInsertPoint(AfterBB);
1146 // Restore the unshadowed variable.
1147 if (OldVal)
1148 NamedValues[VarName] = OldVal;
1149 else
1150 NamedValues.erase(VarName);
1152 // for expr always returns 0.0.
1153 return Constant::getNullValue(Type::getDoubleTy(TheContext));
1156 Value *VarExprAST::codegen() {
1157 std::vector<AllocaInst *> OldBindings;
1159 Function *TheFunction = Builder.GetInsertBlock()->getParent();
1161 // Register all variables and emit their initializer.
1162 for (unsigned i = 0, e = VarNames.size(); i != e; ++i) {
1163 const std::string &VarName = VarNames[i].first;
1164 ExprAST *Init = VarNames[i].second.get();
1166 // Emit the initializer before adding the variable to scope, this prevents
1167 // the initializer from referencing the variable itself, and permits stuff
1168 // like this:
1169 // var a = 1 in
1170 // var a = a in ... # refers to outer 'a'.
1171 Value *InitVal;
1172 if (Init) {
1173 InitVal = Init->codegen();
1174 if (!InitVal)
1175 return nullptr;
1176 } else { // If not specified, use 0.0.
1177 InitVal = ConstantFP::get(TheContext, APFloat(0.0));
1180 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1181 Builder.CreateStore(InitVal, Alloca);
1183 // Remember the old variable binding so that we can restore the binding when
1184 // we unrecurse.
1185 OldBindings.push_back(NamedValues[VarName]);
1187 // Remember this binding.
1188 NamedValues[VarName] = Alloca;
1191 KSDbgInfo.emitLocation(this);
1193 // Codegen the body, now that all vars are in scope.
1194 Value *BodyVal = Body->codegen();
1195 if (!BodyVal)
1196 return nullptr;
1198 // Pop all our variables from scope.
1199 for (unsigned i = 0, e = VarNames.size(); i != e; ++i)
1200 NamedValues[VarNames[i].first] = OldBindings[i];
1202 // Return the body computation.
1203 return BodyVal;
1206 Function *PrototypeAST::codegen() {
1207 // Make the function type: double(double,double) etc.
1208 std::vector<Type *> Doubles(Args.size(), Type::getDoubleTy(TheContext));
1209 FunctionType *FT =
1210 FunctionType::get(Type::getDoubleTy(TheContext), Doubles, false);
1212 Function *F =
1213 Function::Create(FT, Function::ExternalLinkage, Name, TheModule.get());
1215 // Set names for all arguments.
1216 unsigned Idx = 0;
1217 for (auto &Arg : F->args())
1218 Arg.setName(Args[Idx++]);
1220 return F;
1223 Function *FunctionAST::codegen() {
1224 // Transfer ownership of the prototype to the FunctionProtos map, but keep a
1225 // reference to it for use below.
1226 auto &P = *Proto;
1227 FunctionProtos[Proto->getName()] = std::move(Proto);
1228 Function *TheFunction = getFunction(P.getName());
1229 if (!TheFunction)
1230 return nullptr;
1232 // If this is an operator, install it.
1233 if (P.isBinaryOp())
1234 BinopPrecedence[P.getOperatorName()] = P.getBinaryPrecedence();
1236 // Create a new basic block to start insertion into.
1237 BasicBlock *BB = BasicBlock::Create(TheContext, "entry", TheFunction);
1238 Builder.SetInsertPoint(BB);
1240 // Create a subprogram DIE for this function.
1241 DIFile *Unit = DBuilder->createFile(KSDbgInfo.TheCU->getFilename(),
1242 KSDbgInfo.TheCU->getDirectory());
1243 DIScope *FContext = Unit;
1244 unsigned LineNo = P.getLine();
1245 unsigned ScopeLine = LineNo;
1246 DISubprogram *SP = DBuilder->createFunction(
1247 FContext, P.getName(), StringRef(), Unit, LineNo,
1248 CreateFunctionType(TheFunction->arg_size(), Unit), ScopeLine,
1249 DINode::FlagPrototyped, DISubprogram::SPFlagDefinition);
1250 TheFunction->setSubprogram(SP);
1252 // Push the current scope.
1253 KSDbgInfo.LexicalBlocks.push_back(SP);
1255 // Unset the location for the prologue emission (leading instructions with no
1256 // location in a function are considered part of the prologue and the debugger
1257 // will run past them when breaking on a function)
1258 KSDbgInfo.emitLocation(nullptr);
1260 // Record the function arguments in the NamedValues map.
1261 NamedValues.clear();
1262 unsigned ArgIdx = 0;
1263 for (auto &Arg : TheFunction->args()) {
1264 // Create an alloca for this variable.
1265 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, Arg.getName());
1267 // Create a debug descriptor for the variable.
1268 DILocalVariable *D = DBuilder->createParameterVariable(
1269 SP, Arg.getName(), ++ArgIdx, Unit, LineNo, KSDbgInfo.getDoubleTy(),
1270 true);
1272 DBuilder->insertDeclare(Alloca, D, DBuilder->createExpression(),
1273 DebugLoc::get(LineNo, 0, SP),
1274 Builder.GetInsertBlock());
1276 // Store the initial value into the alloca.
1277 Builder.CreateStore(&Arg, Alloca);
1279 // Add arguments to variable symbol table.
1280 NamedValues[Arg.getName()] = Alloca;
1283 KSDbgInfo.emitLocation(Body.get());
1285 if (Value *RetVal = Body->codegen()) {
1286 // Finish off the function.
1287 Builder.CreateRet(RetVal);
1289 // Pop off the lexical block for the function.
1290 KSDbgInfo.LexicalBlocks.pop_back();
1292 // Validate the generated code, checking for consistency.
1293 verifyFunction(*TheFunction);
1295 return TheFunction;
1298 // Error reading body, remove function.
1299 TheFunction->eraseFromParent();
1301 if (P.isBinaryOp())
1302 BinopPrecedence.erase(Proto->getOperatorName());
1304 // Pop off the lexical block for the function since we added it
1305 // unconditionally.
1306 KSDbgInfo.LexicalBlocks.pop_back();
1308 return nullptr;
1311 //===----------------------------------------------------------------------===//
1312 // Top-Level parsing and JIT Driver
1313 //===----------------------------------------------------------------------===//
1315 static void InitializeModule() {
1316 // Open a new module.
1317 TheModule = std::make_unique<Module>("my cool jit", TheContext);
1318 TheModule->setDataLayout(TheJIT->getTargetMachine().createDataLayout());
1321 static void HandleDefinition() {
1322 if (auto FnAST = ParseDefinition()) {
1323 if (!FnAST->codegen())
1324 fprintf(stderr, "Error reading function definition:");
1325 } else {
1326 // Skip token for error recovery.
1327 getNextToken();
1331 static void HandleExtern() {
1332 if (auto ProtoAST = ParseExtern()) {
1333 if (!ProtoAST->codegen())
1334 fprintf(stderr, "Error reading extern");
1335 else
1336 FunctionProtos[ProtoAST->getName()] = std::move(ProtoAST);
1337 } else {
1338 // Skip token for error recovery.
1339 getNextToken();
1343 static void HandleTopLevelExpression() {
1344 // Evaluate a top-level expression into an anonymous function.
1345 if (auto FnAST = ParseTopLevelExpr()) {
1346 if (!FnAST->codegen()) {
1347 fprintf(stderr, "Error generating code for top level expr");
1349 } else {
1350 // Skip token for error recovery.
1351 getNextToken();
1355 /// top ::= definition | external | expression | ';'
1356 static void MainLoop() {
1357 while (1) {
1358 switch (CurTok) {
1359 case tok_eof:
1360 return;
1361 case ';': // ignore top-level semicolons.
1362 getNextToken();
1363 break;
1364 case tok_def:
1365 HandleDefinition();
1366 break;
1367 case tok_extern:
1368 HandleExtern();
1369 break;
1370 default:
1371 HandleTopLevelExpression();
1372 break;
1377 //===----------------------------------------------------------------------===//
1378 // "Library" functions that can be "extern'd" from user code.
1379 //===----------------------------------------------------------------------===//
1381 #ifdef _WIN32
1382 #define DLLEXPORT __declspec(dllexport)
1383 #else
1384 #define DLLEXPORT
1385 #endif
1387 /// putchard - putchar that takes a double and returns 0.
1388 extern "C" DLLEXPORT double putchard(double X) {
1389 fputc((char)X, stderr);
1390 return 0;
1393 /// printd - printf that takes a double prints it as "%f\n", returning 0.
1394 extern "C" DLLEXPORT double printd(double X) {
1395 fprintf(stderr, "%f\n", X);
1396 return 0;
1399 //===----------------------------------------------------------------------===//
1400 // Main driver code.
1401 //===----------------------------------------------------------------------===//
1403 int main() {
1404 InitializeNativeTarget();
1405 InitializeNativeTargetAsmPrinter();
1406 InitializeNativeTargetAsmParser();
1408 // Install standard binary operators.
1409 // 1 is lowest precedence.
1410 BinopPrecedence['='] = 2;
1411 BinopPrecedence['<'] = 10;
1412 BinopPrecedence['+'] = 20;
1413 BinopPrecedence['-'] = 20;
1414 BinopPrecedence['*'] = 40; // highest.
1416 // Prime the first token.
1417 getNextToken();
1419 TheJIT = std::make_unique<KaleidoscopeJIT>();
1421 InitializeModule();
1423 // Add the current debug info version into the module.
1424 TheModule->addModuleFlag(Module::Warning, "Debug Info Version",
1425 DEBUG_METADATA_VERSION);
1427 // Darwin only supports dwarf2.
1428 if (Triple(sys::getProcessTriple()).isOSDarwin())
1429 TheModule->addModuleFlag(llvm::Module::Warning, "Dwarf Version", 2);
1431 // Construct the DIBuilder, we do this here because we need the module.
1432 DBuilder = std::make_unique<DIBuilder>(*TheModule);
1434 // Create the compile unit for the module.
1435 // Currently down as "fib.ks" as a filename since we're redirecting stdin
1436 // but we'd like actual source locations.
1437 KSDbgInfo.TheCU = DBuilder->createCompileUnit(
1438 dwarf::DW_LANG_C, DBuilder->createFile("fib.ks", "."),
1439 "Kaleidoscope Compiler", 0, "", 0);
1441 // Run the main "interpreter loop" now.
1442 MainLoop();
1444 // Finalize the debug info.
1445 DBuilder->finalize();
1447 // Print out all of the generated code.
1448 TheModule->print(errs(), nullptr);
1450 return 0;