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"
17 #include "../include/KaleidoscopeJIT.h"
20 using namespace llvm::orc;
22 //===----------------------------------------------------------------------===//
24 //===----------------------------------------------------------------------===//
26 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
27 // of these for known things.
54 std::string getTokName(int Tok) {
83 return std::string(1, (char)Tok);
90 static IRBuilder<> Builder(getGlobalContext());
94 std::vector<DIScope *> LexicalBlocks;
96 void emitLocation(ExprAST *AST);
97 DIType *getDoubleTy();
100 struct SourceLocation {
104 static SourceLocation CurLoc;
105 static SourceLocation LexLoc = {1, 0};
107 static int advance() {
108 int LastChar = getchar();
110 if (LastChar == '\n' || LastChar == '\r') {
118 static std::string IdentifierStr; // Filled in if tok_identifier
119 static double NumVal; // Filled in if tok_number
121 /// gettok - Return the next token from standard input.
122 static int gettok() {
123 static int LastChar = ' ';
125 // Skip any whitespace.
126 while (isspace(LastChar))
127 LastChar = advance();
131 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
132 IdentifierStr = LastChar;
133 while (isalnum((LastChar = advance())))
134 IdentifierStr += LastChar;
136 if (IdentifierStr == "def")
138 if (IdentifierStr == "extern")
140 if (IdentifierStr == "if")
142 if (IdentifierStr == "then")
144 if (IdentifierStr == "else")
146 if (IdentifierStr == "for")
148 if (IdentifierStr == "in")
150 if (IdentifierStr == "binary")
152 if (IdentifierStr == "unary")
154 if (IdentifierStr == "var")
156 return tok_identifier;
159 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+
163 LastChar = advance();
164 } while (isdigit(LastChar) || LastChar == '.');
166 NumVal = strtod(NumStr.c_str(), 0);
170 if (LastChar == '#') {
171 // Comment until end of line.
173 LastChar = advance();
174 while (LastChar != EOF && LastChar != '\n' && LastChar != '\r');
180 // Check for end of file. Don't eat the EOF.
184 // Otherwise, just return the character as its ascii value.
185 int ThisChar = LastChar;
186 LastChar = advance();
190 //===----------------------------------------------------------------------===//
191 // Abstract Syntax Tree (aka Parse Tree)
192 //===----------------------------------------------------------------------===//
195 raw_ostream &indent(raw_ostream &O, int size) {
196 return O << std::string(size, ' ');
199 /// ExprAST - Base class for all expression nodes.
204 ExprAST(SourceLocation Loc = CurLoc) : Loc(Loc) {}
205 virtual ~ExprAST() {}
206 virtual Value *codegen() = 0;
207 int getLine() const { return Loc.Line; }
208 int getCol() const { return Loc.Col; }
209 virtual raw_ostream &dump(raw_ostream &out, int ind) {
210 return out << ':' << getLine() << ':' << getCol() << '\n';
214 /// NumberExprAST - Expression class for numeric literals like "1.0".
215 class NumberExprAST : public ExprAST {
219 NumberExprAST(double Val) : Val(Val) {}
220 raw_ostream &dump(raw_ostream &out, int ind) override {
221 return ExprAST::dump(out << Val, ind);
223 Value *codegen() override;
226 /// VariableExprAST - Expression class for referencing a variable, like "a".
227 class VariableExprAST : public ExprAST {
231 VariableExprAST(SourceLocation Loc, const std::string &Name)
232 : ExprAST(Loc), Name(Name) {}
233 const std::string &getName() const { return Name; }
234 Value *codegen() override;
235 raw_ostream &dump(raw_ostream &out, int ind) override {
236 return ExprAST::dump(out << Name, ind);
240 /// UnaryExprAST - Expression class for a unary operator.
241 class UnaryExprAST : public ExprAST {
243 std::unique_ptr<ExprAST> Operand;
246 UnaryExprAST(char Opcode, std::unique_ptr<ExprAST> Operand)
247 : Opcode(Opcode), Operand(std::move(Operand)) {}
248 Value *codegen() override;
249 raw_ostream &dump(raw_ostream &out, int ind) override {
250 ExprAST::dump(out << "unary" << Opcode, ind);
251 Operand->dump(out, ind + 1);
256 /// BinaryExprAST - Expression class for a binary operator.
257 class BinaryExprAST : public ExprAST {
259 std::unique_ptr<ExprAST> LHS, RHS;
262 BinaryExprAST(SourceLocation Loc, char Op, std::unique_ptr<ExprAST> LHS,
263 std::unique_ptr<ExprAST> RHS)
264 : ExprAST(Loc), Op(Op), LHS(std::move(LHS)), RHS(std::move(RHS)) {}
265 Value *codegen() override;
266 raw_ostream &dump(raw_ostream &out, int ind) override {
267 ExprAST::dump(out << "binary" << Op, ind);
268 LHS->dump(indent(out, ind) << "LHS:", ind + 1);
269 RHS->dump(indent(out, ind) << "RHS:", ind + 1);
274 /// CallExprAST - Expression class for function calls.
275 class CallExprAST : public ExprAST {
277 std::vector<std::unique_ptr<ExprAST>> Args;
280 CallExprAST(SourceLocation Loc, const std::string &Callee,
281 std::vector<std::unique_ptr<ExprAST>> Args)
282 : ExprAST(Loc), Callee(Callee), Args(std::move(Args)) {}
283 Value *codegen() override;
284 raw_ostream &dump(raw_ostream &out, int ind) override {
285 ExprAST::dump(out << "call " << Callee, ind);
286 for (const auto &Arg : Args)
287 Arg->dump(indent(out, ind + 1), ind + 1);
292 /// IfExprAST - Expression class for if/then/else.
293 class IfExprAST : public ExprAST {
294 std::unique_ptr<ExprAST> Cond, Then, Else;
297 IfExprAST(SourceLocation Loc, std::unique_ptr<ExprAST> Cond,
298 std::unique_ptr<ExprAST> Then, std::unique_ptr<ExprAST> Else)
299 : ExprAST(Loc), Cond(std::move(Cond)), Then(std::move(Then)),
300 Else(std::move(Else)) {}
301 Value *codegen() override;
302 raw_ostream &dump(raw_ostream &out, int ind) override {
303 ExprAST::dump(out << "if", ind);
304 Cond->dump(indent(out, ind) << "Cond:", ind + 1);
305 Then->dump(indent(out, ind) << "Then:", ind + 1);
306 Else->dump(indent(out, ind) << "Else:", ind + 1);
311 /// ForExprAST - Expression class for for/in.
312 class ForExprAST : public ExprAST {
314 std::unique_ptr<ExprAST> Start, End, Step, Body;
317 ForExprAST(const std::string &VarName, std::unique_ptr<ExprAST> Start,
318 std::unique_ptr<ExprAST> End, std::unique_ptr<ExprAST> Step,
319 std::unique_ptr<ExprAST> Body)
320 : VarName(VarName), Start(std::move(Start)), End(std::move(End)),
321 Step(std::move(Step)), Body(std::move(Body)) {}
322 Value *codegen() override;
323 raw_ostream &dump(raw_ostream &out, int ind) override {
324 ExprAST::dump(out << "for", ind);
325 Start->dump(indent(out, ind) << "Cond:", ind + 1);
326 End->dump(indent(out, ind) << "End:", ind + 1);
327 Step->dump(indent(out, ind) << "Step:", ind + 1);
328 Body->dump(indent(out, ind) << "Body:", ind + 1);
333 /// VarExprAST - Expression class for var/in
334 class VarExprAST : public ExprAST {
335 std::vector<std::pair<std::string, std::unique_ptr<ExprAST>>> VarNames;
336 std::unique_ptr<ExprAST> Body;
340 std::vector<std::pair<std::string, std::unique_ptr<ExprAST>>> VarNames,
341 std::unique_ptr<ExprAST> Body)
342 : VarNames(std::move(VarNames)), Body(std::move(Body)) {}
343 Value *codegen() override;
344 raw_ostream &dump(raw_ostream &out, int ind) override {
345 ExprAST::dump(out << "var", ind);
346 for (const auto &NamedVar : VarNames)
347 NamedVar.second->dump(indent(out, ind) << NamedVar.first << ':', ind + 1);
348 Body->dump(indent(out, ind) << "Body:", ind + 1);
353 /// PrototypeAST - This class represents the "prototype" for a function,
354 /// which captures its name, and its argument names (thus implicitly the number
355 /// of arguments the function takes), as well as if it is an operator.
358 std::vector<std::string> Args;
360 unsigned Precedence; // Precedence if a binary op.
364 PrototypeAST(SourceLocation Loc, const std::string &Name,
365 std::vector<std::string> Args, bool IsOperator = false,
367 : Name(Name), Args(std::move(Args)), IsOperator(IsOperator),
368 Precedence(Prec), Line(Loc.Line) {}
370 const std::string &getName() const { return Name; }
372 bool isUnaryOp() const { return IsOperator && Args.size() == 1; }
373 bool isBinaryOp() const { return IsOperator && Args.size() == 2; }
375 char getOperatorName() const {
376 assert(isUnaryOp() || isBinaryOp());
377 return Name[Name.size() - 1];
380 unsigned getBinaryPrecedence() const { return Precedence; }
381 int getLine() const { return Line; }
384 /// FunctionAST - This class represents a function definition itself.
386 std::unique_ptr<PrototypeAST> Proto;
387 std::unique_ptr<ExprAST> Body;
390 FunctionAST(std::unique_ptr<PrototypeAST> Proto,
391 std::unique_ptr<ExprAST> Body)
392 : Proto(std::move(Proto)), Body(std::move(Body)) {}
394 raw_ostream &dump(raw_ostream &out, int ind) {
395 indent(out, ind) << "FunctionAST\n";
397 indent(out, ind) << "Body:";
398 return Body ? Body->dump(out, ind) : out << "null\n";
401 } // end anonymous namespace
403 //===----------------------------------------------------------------------===//
405 //===----------------------------------------------------------------------===//
407 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
408 /// token the parser is looking at. getNextToken reads another token from the
409 /// lexer and updates CurTok with its results.
411 static int getNextToken() { return CurTok = gettok(); }
413 /// BinopPrecedence - This holds the precedence for each binary operator that is
415 static std::map<char, int> BinopPrecedence;
417 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
418 static int GetTokPrecedence() {
419 if (!isascii(CurTok))
422 // Make sure it's a declared binop.
423 int TokPrec = BinopPrecedence[CurTok];
429 /// Error* - These are little helper functions for error handling.
430 std::unique_ptr<ExprAST> Error(const char *Str) {
431 fprintf(stderr, "Error: %s\n", Str);
434 std::unique_ptr<PrototypeAST> ErrorP(const char *Str) {
439 static std::unique_ptr<ExprAST> ParseExpression();
441 /// numberexpr ::= number
442 static std::unique_ptr<ExprAST> ParseNumberExpr() {
443 auto Result = llvm::make_unique<NumberExprAST>(NumVal);
444 getNextToken(); // consume the number
445 return std::move(Result);
448 /// parenexpr ::= '(' expression ')'
449 static std::unique_ptr<ExprAST> ParseParenExpr() {
450 getNextToken(); // eat (.
451 auto V = ParseExpression();
456 return Error("expected ')'");
457 getNextToken(); // eat ).
463 /// ::= identifier '(' expression* ')'
464 static std::unique_ptr<ExprAST> ParseIdentifierExpr() {
465 std::string IdName = IdentifierStr;
467 SourceLocation LitLoc = CurLoc;
469 getNextToken(); // eat identifier.
471 if (CurTok != '(') // Simple variable ref.
472 return llvm::make_unique<VariableExprAST>(LitLoc, IdName);
475 getNextToken(); // eat (
476 std::vector<std::unique_ptr<ExprAST>> Args;
479 if (auto Arg = ParseExpression())
480 Args.push_back(std::move(Arg));
488 return Error("Expected ')' or ',' in argument list");
496 return llvm::make_unique<CallExprAST>(LitLoc, IdName, std::move(Args));
499 /// ifexpr ::= 'if' expression 'then' expression 'else' expression
500 static std::unique_ptr<ExprAST> ParseIfExpr() {
501 SourceLocation IfLoc = CurLoc;
503 getNextToken(); // eat the if.
506 auto Cond = ParseExpression();
510 if (CurTok != tok_then)
511 return Error("expected then");
512 getNextToken(); // eat the then
514 auto Then = ParseExpression();
518 if (CurTok != tok_else)
519 return Error("expected else");
523 auto Else = ParseExpression();
527 return llvm::make_unique<IfExprAST>(IfLoc, std::move(Cond), std::move(Then),
531 /// forexpr ::= 'for' identifier '=' expr ',' expr (',' expr)? 'in' expression
532 static std::unique_ptr<ExprAST> ParseForExpr() {
533 getNextToken(); // eat the for.
535 if (CurTok != tok_identifier)
536 return Error("expected identifier after for");
538 std::string IdName = IdentifierStr;
539 getNextToken(); // eat identifier.
542 return Error("expected '=' after for");
543 getNextToken(); // eat '='.
545 auto Start = ParseExpression();
549 return Error("expected ',' after for start value");
552 auto End = ParseExpression();
556 // The step value is optional.
557 std::unique_ptr<ExprAST> Step;
560 Step = ParseExpression();
565 if (CurTok != tok_in)
566 return Error("expected 'in' after for");
567 getNextToken(); // eat 'in'.
569 auto Body = ParseExpression();
573 return llvm::make_unique<ForExprAST>(IdName, std::move(Start), std::move(End),
574 std::move(Step), std::move(Body));
577 /// varexpr ::= 'var' identifier ('=' expression)?
578 // (',' identifier ('=' expression)?)* 'in' expression
579 static std::unique_ptr<ExprAST> ParseVarExpr() {
580 getNextToken(); // eat the var.
582 std::vector<std::pair<std::string, std::unique_ptr<ExprAST>>> VarNames;
584 // At least one variable name is required.
585 if (CurTok != tok_identifier)
586 return Error("expected identifier after var");
589 std::string Name = IdentifierStr;
590 getNextToken(); // eat identifier.
592 // Read the optional initializer.
593 std::unique_ptr<ExprAST> Init = nullptr;
595 getNextToken(); // eat the '='.
597 Init = ParseExpression();
602 VarNames.push_back(std::make_pair(Name, std::move(Init)));
604 // End of var list, exit loop.
607 getNextToken(); // eat the ','.
609 if (CurTok != tok_identifier)
610 return Error("expected identifier list after var");
613 // At this point, we have to have 'in'.
614 if (CurTok != tok_in)
615 return Error("expected 'in' keyword after 'var'");
616 getNextToken(); // eat 'in'.
618 auto Body = ParseExpression();
622 return llvm::make_unique<VarExprAST>(std::move(VarNames), std::move(Body));
626 /// ::= identifierexpr
632 static std::unique_ptr<ExprAST> ParsePrimary() {
635 return Error("unknown token when expecting an expression");
637 return ParseIdentifierExpr();
639 return ParseNumberExpr();
641 return ParseParenExpr();
643 return ParseIfExpr();
645 return ParseForExpr();
647 return ParseVarExpr();
654 static std::unique_ptr<ExprAST> ParseUnary() {
655 // If the current token is not an operator, it must be a primary expr.
656 if (!isascii(CurTok) || CurTok == '(' || CurTok == ',')
657 return ParsePrimary();
659 // If this is a unary operator, read it.
662 if (auto Operand = ParseUnary())
663 return llvm::make_unique<UnaryExprAST>(Opc, std::move(Operand));
669 static std::unique_ptr<ExprAST> ParseBinOpRHS(int ExprPrec,
670 std::unique_ptr<ExprAST> LHS) {
671 // If this is a binop, find its precedence.
673 int TokPrec = GetTokPrecedence();
675 // If this is a binop that binds at least as tightly as the current binop,
676 // consume it, otherwise we are done.
677 if (TokPrec < ExprPrec)
680 // Okay, we know this is a binop.
682 SourceLocation BinLoc = CurLoc;
683 getNextToken(); // eat binop
685 // Parse the unary expression after the binary operator.
686 auto RHS = ParseUnary();
690 // If BinOp binds less tightly with RHS than the operator after RHS, let
691 // the pending operator take RHS as its LHS.
692 int NextPrec = GetTokPrecedence();
693 if (TokPrec < NextPrec) {
694 RHS = ParseBinOpRHS(TokPrec + 1, std::move(RHS));
700 LHS = llvm::make_unique<BinaryExprAST>(BinLoc, BinOp, std::move(LHS),
706 /// ::= unary binoprhs
708 static std::unique_ptr<ExprAST> ParseExpression() {
709 auto LHS = ParseUnary();
713 return ParseBinOpRHS(0, std::move(LHS));
717 /// ::= id '(' id* ')'
718 /// ::= binary LETTER number? (id, id)
719 /// ::= unary LETTER (id)
720 static std::unique_ptr<PrototypeAST> ParsePrototype() {
723 SourceLocation FnLoc = CurLoc;
725 unsigned Kind = 0; // 0 = identifier, 1 = unary, 2 = binary.
726 unsigned BinaryPrecedence = 30;
730 return ErrorP("Expected function name in prototype");
732 FnName = IdentifierStr;
738 if (!isascii(CurTok))
739 return ErrorP("Expected unary operator");
741 FnName += (char)CurTok;
747 if (!isascii(CurTok))
748 return ErrorP("Expected binary operator");
750 FnName += (char)CurTok;
754 // Read the precedence if present.
755 if (CurTok == tok_number) {
756 if (NumVal < 1 || NumVal > 100)
757 return ErrorP("Invalid precedecnce: must be 1..100");
758 BinaryPrecedence = (unsigned)NumVal;
765 return ErrorP("Expected '(' in prototype");
767 std::vector<std::string> ArgNames;
768 while (getNextToken() == tok_identifier)
769 ArgNames.push_back(IdentifierStr);
771 return ErrorP("Expected ')' in prototype");
774 getNextToken(); // eat ')'.
776 // Verify right number of names for operator.
777 if (Kind && ArgNames.size() != Kind)
778 return ErrorP("Invalid number of operands for operator");
780 return llvm::make_unique<PrototypeAST>(FnLoc, FnName, ArgNames, Kind != 0,
784 /// definition ::= 'def' prototype expression
785 static std::unique_ptr<FunctionAST> ParseDefinition() {
786 getNextToken(); // eat def.
787 auto Proto = ParsePrototype();
791 if (auto E = ParseExpression())
792 return llvm::make_unique<FunctionAST>(std::move(Proto), std::move(E));
796 /// toplevelexpr ::= expression
797 static std::unique_ptr<FunctionAST> ParseTopLevelExpr() {
798 SourceLocation FnLoc = CurLoc;
799 if (auto E = ParseExpression()) {
800 // Make an anonymous proto.
801 auto Proto = llvm::make_unique<PrototypeAST>(FnLoc, "__anon_expr",
802 std::vector<std::string>());
803 return llvm::make_unique<FunctionAST>(std::move(Proto), std::move(E));
808 /// external ::= 'extern' prototype
809 static std::unique_ptr<PrototypeAST> ParseExtern() {
810 getNextToken(); // eat extern.
811 return ParsePrototype();
814 //===----------------------------------------------------------------------===//
815 // Debug Info Support
816 //===----------------------------------------------------------------------===//
818 static std::unique_ptr<DIBuilder> DBuilder;
820 DIType *DebugInfo::getDoubleTy() {
824 DblTy = DBuilder->createBasicType("double", 64, 64, dwarf::DW_ATE_float);
828 void DebugInfo::emitLocation(ExprAST *AST) {
830 return Builder.SetCurrentDebugLocation(DebugLoc());
832 if (LexicalBlocks.empty())
835 Scope = LexicalBlocks.back();
836 Builder.SetCurrentDebugLocation(
837 DebugLoc::get(AST->getLine(), AST->getCol(), Scope));
840 static DISubroutineType *CreateFunctionType(unsigned NumArgs, DIFile *Unit) {
841 SmallVector<Metadata *, 8> EltTys;
842 DIType *DblTy = KSDbgInfo.getDoubleTy();
844 // Add the result type.
845 EltTys.push_back(DblTy);
847 for (unsigned i = 0, e = NumArgs; i != e; ++i)
848 EltTys.push_back(DblTy);
850 return DBuilder->createSubroutineType(Unit,
851 DBuilder->getOrCreateTypeArray(EltTys));
854 //===----------------------------------------------------------------------===//
856 //===----------------------------------------------------------------------===//
858 static std::unique_ptr<Module> TheModule;
859 static std::map<std::string, AllocaInst *> NamedValues;
860 static std::unique_ptr<KaleidoscopeJIT> TheJIT;
861 static std::map<std::string, std::unique_ptr<PrototypeAST>> FunctionProtos;
863 Value *ErrorV(const char *Str) {
868 Function *getFunction(std::string Name) {
869 // First, see if the function has already been added to the current module.
870 if (auto *F = TheModule->getFunction(Name))
873 // If not, check whether we can codegen the declaration from some existing
875 auto FI = FunctionProtos.find(Name);
876 if (FI != FunctionProtos.end())
877 return FI->second->codegen();
879 // If no existing prototype exists, return null.
883 /// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of
884 /// the function. This is used for mutable variables etc.
885 static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction,
886 const std::string &VarName) {
887 IRBuilder<> TmpB(&TheFunction->getEntryBlock(),
888 TheFunction->getEntryBlock().begin());
889 return TmpB.CreateAlloca(Type::getDoubleTy(getGlobalContext()), 0,
893 Value *NumberExprAST::codegen() {
894 KSDbgInfo.emitLocation(this);
895 return ConstantFP::get(getGlobalContext(), APFloat(Val));
898 Value *VariableExprAST::codegen() {
899 // Look this variable up in the function.
900 Value *V = NamedValues[Name];
902 return ErrorV("Unknown variable name");
904 KSDbgInfo.emitLocation(this);
906 return Builder.CreateLoad(V, Name.c_str());
909 Value *UnaryExprAST::codegen() {
910 Value *OperandV = Operand->codegen();
914 Function *F = getFunction(std::string("unary") + Opcode);
916 return ErrorV("Unknown unary operator");
918 KSDbgInfo.emitLocation(this);
919 return Builder.CreateCall(F, OperandV, "unop");
922 Value *BinaryExprAST::codegen() {
923 KSDbgInfo.emitLocation(this);
925 // Special case '=' because we don't want to emit the LHS as an expression.
927 // Assignment requires the LHS to be an identifier.
928 // This assume we're building without RTTI because LLVM builds that way by
929 // default. If you build LLVM with RTTI this can be changed to a
930 // dynamic_cast for automatic error checking.
931 VariableExprAST *LHSE = static_cast<VariableExprAST *>(LHS.get());
933 return ErrorV("destination of '=' must be a variable");
935 Value *Val = RHS->codegen();
940 Value *Variable = NamedValues[LHSE->getName()];
942 return ErrorV("Unknown variable name");
944 Builder.CreateStore(Val, Variable);
948 Value *L = LHS->codegen();
949 Value *R = RHS->codegen();
955 return Builder.CreateFAdd(L, R, "addtmp");
957 return Builder.CreateFSub(L, R, "subtmp");
959 return Builder.CreateFMul(L, R, "multmp");
961 L = Builder.CreateFCmpULT(L, R, "cmptmp");
962 // Convert bool 0/1 to double 0.0 or 1.0
963 return Builder.CreateUIToFP(L, Type::getDoubleTy(getGlobalContext()),
969 // If it wasn't a builtin binary operator, it must be a user defined one. Emit
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);
984 return ErrorV("Unknown function referenced");
986 // If argument mismatch error.
987 if (CalleeF->arg_size() != Args.size())
988 return ErrorV("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());
997 return Builder.CreateCall(CalleeF, ArgsV, "calltmp");
1000 Value *IfExprAST::codegen() {
1001 KSDbgInfo.emitLocation(this);
1003 Value *CondV = Cond->codegen();
1007 // Convert condition to a bool by comparing equal to 0.0.
1008 CondV = Builder.CreateFCmpONE(
1009 CondV, ConstantFP::get(getGlobalContext(), 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 =
1016 BasicBlock::Create(getGlobalContext(), "then", TheFunction);
1017 BasicBlock *ElseBB = BasicBlock::Create(getGlobalContext(), "else");
1018 BasicBlock *MergeBB = BasicBlock::Create(getGlobalContext(), "ifcont");
1020 Builder.CreateCondBr(CondV, ThenBB, ElseBB);
1023 Builder.SetInsertPoint(ThenBB);
1025 Value *ThenV = Then->codegen();
1029 Builder.CreateBr(MergeBB);
1030 // Codegen of 'Then' can change the current block, update ThenBB for the PHI.
1031 ThenBB = Builder.GetInsertBlock();
1034 TheFunction->getBasicBlockList().push_back(ElseBB);
1035 Builder.SetInsertPoint(ElseBB);
1037 Value *ElseV = Else->codegen();
1041 Builder.CreateBr(MergeBB);
1042 // Codegen of 'Else' can change the current block, update ElseBB for the PHI.
1043 ElseBB = Builder.GetInsertBlock();
1045 // Emit merge block.
1046 TheFunction->getBasicBlockList().push_back(MergeBB);
1047 Builder.SetInsertPoint(MergeBB);
1049 Builder.CreatePHI(Type::getDoubleTy(getGlobalContext()), 2, "iftmp");
1051 PN->addIncoming(ThenV, ThenBB);
1052 PN->addIncoming(ElseV, ElseBB);
1056 // Output for-loop as:
1057 // var = alloca double
1059 // start = startexpr
1060 // store start -> var
1068 // endcond = endexpr
1070 // curvar = load var
1071 // nextvar = curvar + step
1072 // store nextvar -> var
1073 // br endcond, loop, endloop
1075 Value *ForExprAST::codegen() {
1076 Function *TheFunction = Builder.GetInsertBlock()->getParent();
1078 // Create an alloca for the variable in the entry block.
1079 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1081 KSDbgInfo.emitLocation(this);
1083 // Emit the start code first, without 'variable' in scope.
1084 Value *StartVal = Start->codegen();
1088 // Store the value into the alloca.
1089 Builder.CreateStore(StartVal, Alloca);
1091 // Make the new basic block for the loop header, inserting after current
1093 BasicBlock *LoopBB =
1094 BasicBlock::Create(getGlobalContext(), "loop", TheFunction);
1096 // Insert an explicit fall through from the current block to the LoopBB.
1097 Builder.CreateBr(LoopBB);
1099 // Start insertion in LoopBB.
1100 Builder.SetInsertPoint(LoopBB);
1102 // Within the loop, the variable is defined equal to the PHI node. If it
1103 // shadows an existing variable, we have to restore it, so save it now.
1104 AllocaInst *OldVal = NamedValues[VarName];
1105 NamedValues[VarName] = Alloca;
1107 // Emit the body of the loop. This, like any other expr, can change the
1108 // current BB. Note that we ignore the value computed by the body, but don't
1110 if (!Body->codegen())
1113 // Emit the step value.
1114 Value *StepVal = nullptr;
1116 StepVal = Step->codegen();
1120 // If not specified, use 1.0.
1121 StepVal = ConstantFP::get(getGlobalContext(), APFloat(1.0));
1124 // Compute the end condition.
1125 Value *EndCond = End->codegen();
1129 // Reload, increment, and restore the alloca. This handles the case where
1130 // the body of the loop mutates the variable.
1131 Value *CurVar = Builder.CreateLoad(Alloca, VarName.c_str());
1132 Value *NextVar = Builder.CreateFAdd(CurVar, StepVal, "nextvar");
1133 Builder.CreateStore(NextVar, Alloca);
1135 // Convert condition to a bool by comparing equal to 0.0.
1136 EndCond = Builder.CreateFCmpONE(
1137 EndCond, ConstantFP::get(getGlobalContext(), APFloat(0.0)), "loopcond");
1139 // Create the "after loop" block and insert it.
1140 BasicBlock *AfterBB =
1141 BasicBlock::Create(getGlobalContext(), "afterloop", TheFunction);
1143 // Insert the conditional branch into the end of LoopEndBB.
1144 Builder.CreateCondBr(EndCond, LoopBB, AfterBB);
1146 // Any new code will be inserted in AfterBB.
1147 Builder.SetInsertPoint(AfterBB);
1149 // Restore the unshadowed variable.
1151 NamedValues[VarName] = OldVal;
1153 NamedValues.erase(VarName);
1155 // for expr always returns 0.0.
1156 return Constant::getNullValue(Type::getDoubleTy(getGlobalContext()));
1159 Value *VarExprAST::codegen() {
1160 std::vector<AllocaInst *> OldBindings;
1162 Function *TheFunction = Builder.GetInsertBlock()->getParent();
1164 // Register all variables and emit their initializer.
1165 for (unsigned i = 0, e = VarNames.size(); i != e; ++i) {
1166 const std::string &VarName = VarNames[i].first;
1167 ExprAST *Init = VarNames[i].second.get();
1169 // Emit the initializer before adding the variable to scope, this prevents
1170 // the initializer from referencing the variable itself, and permits stuff
1173 // var a = a in ... # refers to outer 'a'.
1176 InitVal = Init->codegen();
1179 } else { // If not specified, use 0.0.
1180 InitVal = ConstantFP::get(getGlobalContext(), APFloat(0.0));
1183 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1184 Builder.CreateStore(InitVal, Alloca);
1186 // Remember the old variable binding so that we can restore the binding when
1188 OldBindings.push_back(NamedValues[VarName]);
1190 // Remember this binding.
1191 NamedValues[VarName] = Alloca;
1194 KSDbgInfo.emitLocation(this);
1196 // Codegen the body, now that all vars are in scope.
1197 Value *BodyVal = Body->codegen();
1201 // Pop all our variables from scope.
1202 for (unsigned i = 0, e = VarNames.size(); i != e; ++i)
1203 NamedValues[VarNames[i].first] = OldBindings[i];
1205 // Return the body computation.
1209 Function *PrototypeAST::codegen() {
1210 // Make the function type: double(double,double) etc.
1211 std::vector<Type *> Doubles(Args.size(),
1212 Type::getDoubleTy(getGlobalContext()));
1214 FunctionType::get(Type::getDoubleTy(getGlobalContext()), Doubles, false);
1217 Function::Create(FT, Function::ExternalLinkage, Name, TheModule.get());
1219 // Set names for all arguments.
1221 for (auto &Arg : F->args())
1222 Arg.setName(Args[Idx++]);
1227 Function *FunctionAST::codegen() {
1228 // Transfer ownership of the prototype to the FunctionProtos map, but keep a
1229 // reference to it for use below.
1231 FunctionProtos[Proto->getName()] = std::move(Proto);
1232 Function *TheFunction = getFunction(P.getName());
1236 // If this is an operator, install it.
1238 BinopPrecedence[P.getOperatorName()] = P.getBinaryPrecedence();
1240 // Create a new basic block to start insertion into.
1241 BasicBlock *BB = BasicBlock::Create(getGlobalContext(), "entry", TheFunction);
1242 Builder.SetInsertPoint(BB);
1244 // Create a subprogram DIE for this function.
1245 DIFile *Unit = DBuilder->createFile(KSDbgInfo.TheCU->getFilename(),
1246 KSDbgInfo.TheCU->getDirectory());
1247 DIScope *FContext = Unit;
1248 unsigned LineNo = P.getLine();
1249 unsigned ScopeLine = LineNo;
1250 DISubprogram *SP = DBuilder->createFunction(
1251 FContext, P.getName(), StringRef(), Unit, LineNo,
1252 CreateFunctionType(TheFunction->arg_size(), Unit),
1253 false /* internal linkage */, true /* definition */, ScopeLine,
1254 DINode::FlagPrototyped, false, TheFunction);
1256 // Push the current scope.
1257 KSDbgInfo.LexicalBlocks.push_back(SP);
1259 // Unset the location for the prologue emission (leading instructions with no
1260 // location in a function are considered part of the prologue and the debugger
1261 // will run past them when breaking on a function)
1262 KSDbgInfo.emitLocation(nullptr);
1264 // Record the function arguments in the NamedValues map.
1265 NamedValues.clear();
1266 unsigned ArgIdx = 0;
1267 for (auto &Arg : TheFunction->args()) {
1268 // Create an alloca for this variable.
1269 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, Arg.getName());
1271 // Create a debug descriptor for the variable.
1272 DILocalVariable *D = DBuilder->createParameterVariable(
1273 SP, Arg.getName(), ++ArgIdx, Unit, LineNo, KSDbgInfo.getDoubleTy(),
1276 DBuilder->insertDeclare(Alloca, D, DBuilder->createExpression(),
1277 DebugLoc::get(LineNo, 0, SP),
1278 Builder.GetInsertBlock());
1280 // Store the initial value into the alloca.
1281 Builder.CreateStore(&Arg, Alloca);
1283 // Add arguments to variable symbol table.
1284 NamedValues[Arg.getName()] = Alloca;
1287 KSDbgInfo.emitLocation(Body.get());
1289 if (Value *RetVal = Body->codegen()) {
1290 // Finish off the function.
1291 Builder.CreateRet(RetVal);
1293 // Pop off the lexical block for the function.
1294 KSDbgInfo.LexicalBlocks.pop_back();
1296 // Validate the generated code, checking for consistency.
1297 verifyFunction(*TheFunction);
1302 // Error reading body, remove function.
1303 TheFunction->eraseFromParent();
1306 BinopPrecedence.erase(Proto->getOperatorName());
1308 // Pop off the lexical block for the function since we added it
1310 KSDbgInfo.LexicalBlocks.pop_back();
1315 //===----------------------------------------------------------------------===//
1316 // Top-Level parsing and JIT Driver
1317 //===----------------------------------------------------------------------===//
1319 static void InitializeModule() {
1320 // Open a new module.
1321 TheModule = llvm::make_unique<Module>("my cool jit", getGlobalContext());
1322 TheModule->setDataLayout(TheJIT->getTargetMachine().createDataLayout());
1325 static void HandleDefinition() {
1326 if (auto FnAST = ParseDefinition()) {
1327 if (!FnAST->codegen())
1328 fprintf(stderr, "Error reading function definition:");
1330 // Skip token for error recovery.
1335 static void HandleExtern() {
1336 if (auto ProtoAST = ParseExtern()) {
1337 if (!ProtoAST->codegen())
1338 fprintf(stderr, "Error reading extern");
1340 FunctionProtos[ProtoAST->getName()] = std::move(ProtoAST);
1342 // Skip token for error recovery.
1347 static void HandleTopLevelExpression() {
1348 // Evaluate a top-level expression into an anonymous function.
1349 if (auto FnAST = ParseTopLevelExpr()) {
1350 if (!FnAST->codegen()) {
1351 fprintf(stderr, "Error generating code for top level expr");
1354 // Skip token for error recovery.
1359 /// top ::= definition | external | expression | ';'
1360 static void MainLoop() {
1365 case ';': // ignore top-level semicolons.
1375 HandleTopLevelExpression();
1381 //===----------------------------------------------------------------------===//
1382 // "Library" functions that can be "extern'd" from user code.
1383 //===----------------------------------------------------------------------===//
1385 /// putchard - putchar that takes a double and returns 0.
1386 extern "C" double putchard(double X) {
1387 fputc((char)X, stderr);
1391 /// printd - printf that takes a double prints it as "%f\n", returning 0.
1392 extern "C" double printd(double X) {
1393 fprintf(stderr, "%f\n", X);
1397 //===----------------------------------------------------------------------===//
1398 // Main driver code.
1399 //===----------------------------------------------------------------------===//
1402 InitializeNativeTarget();
1403 InitializeNativeTargetAsmPrinter();
1404 InitializeNativeTargetAsmParser();
1406 // Install standard binary operators.
1407 // 1 is lowest precedence.
1408 BinopPrecedence['='] = 2;
1409 BinopPrecedence['<'] = 10;
1410 BinopPrecedence['+'] = 20;
1411 BinopPrecedence['-'] = 20;
1412 BinopPrecedence['*'] = 40; // highest.
1414 // Prime the first token.
1417 TheJIT = llvm::make_unique<KaleidoscopeJIT>();
1421 // Add the current debug info version into the module.
1422 TheModule->addModuleFlag(Module::Warning, "Debug Info Version",
1423 DEBUG_METADATA_VERSION);
1425 // Darwin only supports dwarf2.
1426 if (Triple(sys::getProcessTriple()).isOSDarwin())
1427 TheModule->addModuleFlag(llvm::Module::Warning, "Dwarf Version", 2);
1429 // Construct the DIBuilder, we do this here because we need the module.
1430 DBuilder = llvm::make_unique<DIBuilder>(*TheModule);
1432 // Create the compile unit for the module.
1433 // Currently down as "fib.ks" as a filename since we're redirecting stdin
1434 // but we'd like actual source locations.
1435 KSDbgInfo.TheCU = DBuilder->createCompileUnit(
1436 dwarf::DW_LANG_C, "fib.ks", ".", "Kaleidoscope Compiler", 0, "", 0);
1438 // Run the main "interpreter loop" now.
1441 // Finalize the debug info.
1442 DBuilder->finalize();
1444 // Print out all of the generated code.