1 #include "llvm/ADT/Triple.h"
2 #include "llvm/Analysis/Passes.h"
3 #include "llvm/ExecutionEngine/ExecutionEngine.h"
4 #include "llvm/ExecutionEngine/MCJIT.h"
5 #include "llvm/ExecutionEngine/SectionMemoryManager.h"
6 #include "llvm/IR/DataLayout.h"
7 #include "llvm/IR/DerivedTypes.h"
8 #include "llvm/IR/DIBuilder.h"
9 #include "llvm/IR/IRBuilder.h"
10 #include "llvm/IR/LLVMContext.h"
11 #include "llvm/IR/Module.h"
12 #include "llvm/IR/Verifier.h"
13 #include "llvm/PassManager.h"
14 #include "llvm/Support/Host.h"
15 #include "llvm/Support/TargetSelect.h"
16 #include "llvm/Transforms/Scalar.h"
25 //===----------------------------------------------------------------------===//
27 //===----------------------------------------------------------------------===//
29 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
30 // of these for known things.
57 std::string getTokName(int Tok) {
86 return std::string(1, (char)Tok);
93 static IRBuilder<> Builder(getGlobalContext());
97 std::vector<DIScope *> LexicalBlocks;
98 std::map<const PrototypeAST *, DIScope> FnScopeMap;
100 void emitLocation(ExprAST *AST);
101 DIType getDoubleTy();
104 static std::string IdentifierStr; // Filled in if tok_identifier
105 static double NumVal; // Filled in if tok_number
106 struct SourceLocation {
110 static SourceLocation CurLoc;
111 static SourceLocation LexLoc = { 1, 0 };
113 static int advance() {
114 int LastChar = getchar();
116 if (LastChar == '\n' || LastChar == '\r') {
124 /// gettok - Return the next token from standard input.
125 static int gettok() {
126 static int LastChar = ' ';
128 // Skip any whitespace.
129 while (isspace(LastChar))
130 LastChar = advance();
134 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
135 IdentifierStr = LastChar;
136 while (isalnum((LastChar = advance())))
137 IdentifierStr += LastChar;
139 if (IdentifierStr == "def")
141 if (IdentifierStr == "extern")
143 if (IdentifierStr == "if")
145 if (IdentifierStr == "then")
147 if (IdentifierStr == "else")
149 if (IdentifierStr == "for")
151 if (IdentifierStr == "in")
153 if (IdentifierStr == "binary")
155 if (IdentifierStr == "unary")
157 if (IdentifierStr == "var")
159 return tok_identifier;
162 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+
166 LastChar = advance();
167 } while (isdigit(LastChar) || LastChar == '.');
169 NumVal = strtod(NumStr.c_str(), 0);
173 if (LastChar == '#') {
174 // Comment until end of line.
176 LastChar = advance();
177 while (LastChar != EOF && LastChar != '\n' && LastChar != '\r');
183 // Check for end of file. Don't eat the EOF.
187 // Otherwise, just return the character as its ascii value.
188 int ThisChar = LastChar;
189 LastChar = advance();
193 //===----------------------------------------------------------------------===//
194 // Abstract Syntax Tree (aka Parse Tree)
195 //===----------------------------------------------------------------------===//
198 std::ostream &indent(std::ostream &O, int size) {
199 return O << std::string(size, ' ');
202 /// ExprAST - Base class for all expression nodes.
207 int getLine() const { return Loc.Line; }
208 int getCol() const { return Loc.Col; }
209 ExprAST(SourceLocation Loc = CurLoc) : Loc(Loc) {}
210 virtual std::ostream &dump(std::ostream &out, int ind) {
211 return out << ':' << getLine() << ':' << getCol() << '\n';
213 virtual ~ExprAST() {}
214 virtual Value *Codegen() = 0;
217 /// NumberExprAST - Expression class for numeric literals like "1.0".
218 class NumberExprAST : public ExprAST {
222 NumberExprAST(double val) : Val(val) {}
223 virtual std::ostream &dump(std::ostream &out, int ind) {
224 return ExprAST::dump(out << Val, ind);
226 virtual Value *Codegen();
229 /// VariableExprAST - Expression class for referencing a variable, like "a".
230 class VariableExprAST : public ExprAST {
234 VariableExprAST(SourceLocation Loc, const std::string &name)
235 : ExprAST(Loc), Name(name) {}
236 const std::string &getName() const { return Name; }
237 virtual std::ostream &dump(std::ostream &out, int ind) {
238 return ExprAST::dump(out << Name, ind);
240 virtual Value *Codegen();
243 /// UnaryExprAST - Expression class for a unary operator.
244 class UnaryExprAST : public ExprAST {
249 UnaryExprAST(char opcode, ExprAST *operand)
250 : Opcode(opcode), Operand(operand) {}
251 virtual std::ostream &dump(std::ostream &out, int ind) {
252 ExprAST::dump(out << "unary" << Opcode, ind);
253 Operand->dump(out, ind + 1);
256 virtual Value *Codegen();
259 /// BinaryExprAST - Expression class for a binary operator.
260 class BinaryExprAST : public ExprAST {
265 BinaryExprAST(SourceLocation Loc, char op, ExprAST *lhs, ExprAST *rhs)
266 : ExprAST(Loc), Op(op), LHS(lhs), RHS(rhs) {}
267 virtual std::ostream &dump(std::ostream &out, int ind) {
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);
273 virtual Value *Codegen();
276 /// CallExprAST - Expression class for function calls.
277 class CallExprAST : public ExprAST {
279 std::vector<ExprAST *> Args;
282 CallExprAST(SourceLocation Loc, const std::string &callee,
283 std::vector<ExprAST *> &args)
284 : ExprAST(Loc), Callee(callee), Args(args) {}
285 virtual std::ostream &dump(std::ostream &out, int ind) {
286 ExprAST::dump(out << "call " << Callee, ind);
287 for (ExprAST *Arg : Args)
288 Arg->dump(indent(out, ind + 1), ind + 1);
291 virtual Value *Codegen();
294 /// IfExprAST - Expression class for if/then/else.
295 class IfExprAST : public ExprAST {
296 ExprAST *Cond, *Then, *Else;
299 IfExprAST(SourceLocation Loc, ExprAST *cond, ExprAST *then, ExprAST *_else)
300 : ExprAST(Loc), Cond(cond), Then(then), Else(_else) {}
301 virtual std::ostream &dump(std::ostream &out, int ind) {
302 ExprAST::dump(out << "if", ind);
303 Cond->dump(indent(out, ind) << "Cond:", ind + 1);
304 Then->dump(indent(out, ind) << "Then:", ind + 1);
305 Else->dump(indent(out, ind) << "Else:", ind + 1);
308 virtual Value *Codegen();
311 /// ForExprAST - Expression class for for/in.
312 class ForExprAST : public ExprAST {
314 ExprAST *Start, *End, *Step, *Body;
317 ForExprAST(const std::string &varname, ExprAST *start, ExprAST *end,
318 ExprAST *step, ExprAST *body)
319 : VarName(varname), Start(start), End(end), Step(step), Body(body) {}
320 virtual std::ostream &dump(std::ostream &out, int ind) {
321 ExprAST::dump(out << "for", ind);
322 Start->dump(indent(out, ind) << "Cond:", ind + 1);
323 End->dump(indent(out, ind) << "End:", ind + 1);
324 Step->dump(indent(out, ind) << "Step:", ind + 1);
325 Body->dump(indent(out, ind) << "Body:", ind + 1);
328 virtual Value *Codegen();
331 /// VarExprAST - Expression class for var/in
332 class VarExprAST : public ExprAST {
333 std::vector<std::pair<std::string, ExprAST *> > VarNames;
337 VarExprAST(const std::vector<std::pair<std::string, ExprAST *> > &varnames,
339 : VarNames(varnames), Body(body) {}
341 virtual std::ostream &dump(std::ostream &out, int ind) {
342 ExprAST::dump(out << "var", ind);
343 for (const auto &NamedVar : VarNames)
344 NamedVar.second->dump(indent(out, ind) << NamedVar.first << ':', ind + 1);
345 Body->dump(indent(out, ind) << "Body:", ind + 1);
348 virtual Value *Codegen();
351 /// PrototypeAST - This class represents the "prototype" for a function,
352 /// which captures its argument names as well as if it is an operator.
355 std::vector<std::string> Args;
357 unsigned Precedence; // Precedence if a binary op.
361 PrototypeAST(SourceLocation Loc, const std::string &name,
362 const std::vector<std::string> &args, bool isoperator = false,
364 : Name(name), Args(args), isOperator(isoperator), Precedence(prec),
367 bool isUnaryOp() const { return isOperator && Args.size() == 1; }
368 bool isBinaryOp() const { return isOperator && Args.size() == 2; }
370 char getOperatorName() const {
371 assert(isUnaryOp() || isBinaryOp());
372 return Name[Name.size() - 1];
375 unsigned getBinaryPrecedence() const { return Precedence; }
379 void CreateArgumentAllocas(Function *F);
380 const std::vector<std::string> &getArgs() const { return Args; }
383 /// FunctionAST - This class represents a function definition itself.
389 FunctionAST(PrototypeAST *proto, ExprAST *body) : Proto(proto), Body(body) {}
391 std::ostream &dump(std::ostream &out, int ind) {
392 indent(out, ind) << "FunctionAST\n";
394 indent(out, ind) << "Body:";
395 return Body ? Body->dump(out, ind) : out << "null\n";
400 } // end anonymous namespace
402 //===----------------------------------------------------------------------===//
404 //===----------------------------------------------------------------------===//
406 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
407 /// token the parser is looking at. getNextToken reads another token from the
408 /// lexer and updates CurTok with its results.
410 static int getNextToken() { return CurTok = gettok(); }
412 /// BinopPrecedence - This holds the precedence for each binary operator that is
414 static std::map<char, int> BinopPrecedence;
416 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
417 static int GetTokPrecedence() {
418 if (!isascii(CurTok))
421 // Make sure it's a declared binop.
422 int TokPrec = BinopPrecedence[CurTok];
428 /// Error* - These are little helper functions for error handling.
429 ExprAST *Error(const char *Str) {
430 fprintf(stderr, "Error: %s\n", Str);
433 PrototypeAST *ErrorP(const char *Str) {
437 FunctionAST *ErrorF(const char *Str) {
442 static ExprAST *ParseExpression();
446 /// ::= identifier '(' expression* ')'
447 static ExprAST *ParseIdentifierExpr() {
448 std::string IdName = IdentifierStr;
450 SourceLocation LitLoc = CurLoc;
452 getNextToken(); // eat identifier.
454 if (CurTok != '(') // Simple variable ref.
455 return new VariableExprAST(LitLoc, IdName);
458 getNextToken(); // eat (
459 std::vector<ExprAST *> Args;
462 ExprAST *Arg = ParseExpression();
471 return Error("Expected ')' or ',' in argument list");
479 return new CallExprAST(LitLoc, IdName, Args);
482 /// numberexpr ::= number
483 static ExprAST *ParseNumberExpr() {
484 ExprAST *Result = new NumberExprAST(NumVal);
485 getNextToken(); // consume the number
489 /// parenexpr ::= '(' expression ')'
490 static ExprAST *ParseParenExpr() {
491 getNextToken(); // eat (.
492 ExprAST *V = ParseExpression();
497 return Error("expected ')'");
498 getNextToken(); // eat ).
502 /// ifexpr ::= 'if' expression 'then' expression 'else' expression
503 static ExprAST *ParseIfExpr() {
504 SourceLocation IfLoc = CurLoc;
506 getNextToken(); // eat the if.
509 ExprAST *Cond = ParseExpression();
513 if (CurTok != tok_then)
514 return Error("expected then");
515 getNextToken(); // eat the then
517 ExprAST *Then = ParseExpression();
521 if (CurTok != tok_else)
522 return Error("expected else");
526 ExprAST *Else = ParseExpression();
530 return new IfExprAST(IfLoc, Cond, Then, Else);
533 /// forexpr ::= 'for' identifier '=' expr ',' expr (',' expr)? 'in' expression
534 static ExprAST *ParseForExpr() {
535 getNextToken(); // eat the for.
537 if (CurTok != tok_identifier)
538 return Error("expected identifier after for");
540 std::string IdName = IdentifierStr;
541 getNextToken(); // eat identifier.
544 return Error("expected '=' after for");
545 getNextToken(); // eat '='.
547 ExprAST *Start = ParseExpression();
551 return Error("expected ',' after for start value");
554 ExprAST *End = ParseExpression();
558 // The step value is optional.
562 Step = ParseExpression();
567 if (CurTok != tok_in)
568 return Error("expected 'in' after for");
569 getNextToken(); // eat 'in'.
571 ExprAST *Body = ParseExpression();
575 return new ForExprAST(IdName, Start, End, Step, Body);
578 /// varexpr ::= 'var' identifier ('=' expression)?
579 // (',' identifier ('=' expression)?)* 'in' expression
580 static ExprAST *ParseVarExpr() {
581 getNextToken(); // eat the var.
583 std::vector<std::pair<std::string, ExprAST *> > VarNames;
585 // At least one variable name is required.
586 if (CurTok != tok_identifier)
587 return Error("expected identifier after var");
590 std::string Name = IdentifierStr;
591 getNextToken(); // eat identifier.
593 // Read the optional initializer.
596 getNextToken(); // eat the '='.
598 Init = ParseExpression();
603 VarNames.push_back(std::make_pair(Name, Init));
605 // End of var list, exit loop.
608 getNextToken(); // eat the ','.
610 if (CurTok != tok_identifier)
611 return Error("expected identifier list after var");
614 // At this point, we have to have 'in'.
615 if (CurTok != tok_in)
616 return Error("expected 'in' keyword after 'var'");
617 getNextToken(); // eat 'in'.
619 ExprAST *Body = ParseExpression();
623 return new VarExprAST(VarNames, Body);
627 /// ::= identifierexpr
633 static ExprAST *ParsePrimary() {
636 return Error("unknown token when expecting an expression");
638 return ParseIdentifierExpr();
640 return ParseNumberExpr();
642 return ParseParenExpr();
644 return ParseIfExpr();
646 return ParseForExpr();
648 return ParseVarExpr();
655 static ExprAST *ParseUnary() {
656 // If the current token is not an operator, it must be a primary expr.
657 if (!isascii(CurTok) || CurTok == '(' || CurTok == ',')
658 return ParsePrimary();
660 // If this is a unary operator, read it.
663 if (ExprAST *Operand = ParseUnary())
664 return new UnaryExprAST(Opc, Operand);
670 static ExprAST *ParseBinOpRHS(int ExprPrec, 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 ExprAST *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, RHS);
700 LHS = new BinaryExprAST(BinLoc, BinOp, LHS, RHS);
705 /// ::= unary binoprhs
707 static ExprAST *ParseExpression() {
708 ExprAST *LHS = ParseUnary();
712 return ParseBinOpRHS(0, LHS);
716 /// ::= id '(' id* ')'
717 /// ::= binary LETTER number? (id, id)
718 /// ::= unary LETTER (id)
719 static PrototypeAST *ParsePrototype() {
722 SourceLocation FnLoc = CurLoc;
724 unsigned Kind = 0; // 0 = identifier, 1 = unary, 2 = binary.
725 unsigned BinaryPrecedence = 30;
729 return ErrorP("Expected function name in prototype");
731 FnName = IdentifierStr;
737 if (!isascii(CurTok))
738 return ErrorP("Expected unary operator");
740 FnName += (char)CurTok;
746 if (!isascii(CurTok))
747 return ErrorP("Expected binary operator");
749 FnName += (char)CurTok;
753 // Read the precedence if present.
754 if (CurTok == tok_number) {
755 if (NumVal < 1 || NumVal > 100)
756 return ErrorP("Invalid precedecnce: must be 1..100");
757 BinaryPrecedence = (unsigned)NumVal;
764 return ErrorP("Expected '(' in prototype");
766 std::vector<std::string> ArgNames;
767 while (getNextToken() == tok_identifier)
768 ArgNames.push_back(IdentifierStr);
770 return ErrorP("Expected ')' in prototype");
773 getNextToken(); // eat ')'.
775 // Verify right number of names for operator.
776 if (Kind && ArgNames.size() != Kind)
777 return ErrorP("Invalid number of operands for operator");
779 return new PrototypeAST(FnLoc, FnName, ArgNames, Kind != 0, BinaryPrecedence);
782 /// definition ::= 'def' prototype expression
783 static FunctionAST *ParseDefinition() {
784 getNextToken(); // eat def.
785 PrototypeAST *Proto = ParsePrototype();
789 if (ExprAST *E = ParseExpression())
790 return new FunctionAST(Proto, E);
794 /// toplevelexpr ::= expression
795 static FunctionAST *ParseTopLevelExpr() {
796 SourceLocation FnLoc = CurLoc;
797 if (ExprAST *E = ParseExpression()) {
798 // Make an anonymous proto.
799 PrototypeAST *Proto =
800 new PrototypeAST(FnLoc, "main", std::vector<std::string>());
801 return new FunctionAST(Proto, E);
806 /// external ::= 'extern' prototype
807 static PrototypeAST *ParseExtern() {
808 getNextToken(); // eat extern.
809 return ParsePrototype();
812 //===----------------------------------------------------------------------===//
813 // Debug Info Support
814 //===----------------------------------------------------------------------===//
816 static DIBuilder *DBuilder;
818 DIType DebugInfo::getDoubleTy() {
822 DblTy = DBuilder->createBasicType("double", 64, 64, dwarf::DW_ATE_float);
826 void DebugInfo::emitLocation(ExprAST *AST) {
828 return Builder.SetCurrentDebugLocation(DebugLoc());
830 if (LexicalBlocks.empty())
833 Scope = LexicalBlocks.back();
834 Builder.SetCurrentDebugLocation(
835 DebugLoc::get(AST->getLine(), AST->getCol(), DIScope(*Scope)));
838 static DICompositeType CreateFunctionType(unsigned NumArgs, DIFile Unit) {
839 SmallVector<Value *, 8> EltTys;
840 DIType DblTy = KSDbgInfo.getDoubleTy();
842 // Add the result type.
843 EltTys.push_back(DblTy);
845 for (unsigned i = 0, e = NumArgs; i != e; ++i)
846 EltTys.push_back(DblTy);
848 DITypeArray EltTypeArray = DBuilder->getOrCreateTypeArray(EltTys);
849 return DBuilder->createSubroutineType(Unit, EltTypeArray);
852 //===----------------------------------------------------------------------===//
854 //===----------------------------------------------------------------------===//
856 static Module *TheModule;
857 static std::map<std::string, AllocaInst *> NamedValues;
858 static FunctionPassManager *TheFPM;
860 Value *ErrorV(const char *Str) {
865 /// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of
866 /// the function. This is used for mutable variables etc.
867 static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction,
868 const std::string &VarName) {
869 IRBuilder<> TmpB(&TheFunction->getEntryBlock(),
870 TheFunction->getEntryBlock().begin());
871 return TmpB.CreateAlloca(Type::getDoubleTy(getGlobalContext()), 0,
875 Value *NumberExprAST::Codegen() {
876 KSDbgInfo.emitLocation(this);
877 return ConstantFP::get(getGlobalContext(), APFloat(Val));
880 Value *VariableExprAST::Codegen() {
881 // Look this variable up in the function.
882 Value *V = NamedValues[Name];
884 return ErrorV("Unknown variable name");
886 KSDbgInfo.emitLocation(this);
888 return Builder.CreateLoad(V, Name.c_str());
891 Value *UnaryExprAST::Codegen() {
892 Value *OperandV = Operand->Codegen();
896 Function *F = TheModule->getFunction(std::string("unary") + Opcode);
898 return ErrorV("Unknown unary operator");
900 KSDbgInfo.emitLocation(this);
901 return Builder.CreateCall(F, OperandV, "unop");
904 Value *BinaryExprAST::Codegen() {
905 KSDbgInfo.emitLocation(this);
907 // Special case '=' because we don't want to emit the LHS as an expression.
909 // Assignment requires the LHS to be an identifier.
910 VariableExprAST *LHSE = dynamic_cast<VariableExprAST *>(LHS);
912 return ErrorV("destination of '=' must be a variable");
914 Value *Val = RHS->Codegen();
919 Value *Variable = NamedValues[LHSE->getName()];
921 return ErrorV("Unknown variable name");
923 Builder.CreateStore(Val, Variable);
927 Value *L = LHS->Codegen();
928 Value *R = RHS->Codegen();
929 if (L == 0 || R == 0)
934 return Builder.CreateFAdd(L, R, "addtmp");
936 return Builder.CreateFSub(L, R, "subtmp");
938 return Builder.CreateFMul(L, R, "multmp");
940 L = Builder.CreateFCmpULT(L, R, "cmptmp");
941 // Convert bool 0/1 to double 0.0 or 1.0
942 return Builder.CreateUIToFP(L, Type::getDoubleTy(getGlobalContext()),
948 // If it wasn't a builtin binary operator, it must be a user defined one. Emit
950 Function *F = TheModule->getFunction(std::string("binary") + Op);
951 assert(F && "binary operator not found!");
953 Value *Ops[] = { L, R };
954 return Builder.CreateCall(F, Ops, "binop");
957 Value *CallExprAST::Codegen() {
958 KSDbgInfo.emitLocation(this);
960 // Look up the name in the global module table.
961 Function *CalleeF = TheModule->getFunction(Callee);
963 return ErrorV("Unknown function referenced");
965 // If argument mismatch error.
966 if (CalleeF->arg_size() != Args.size())
967 return ErrorV("Incorrect # arguments passed");
969 std::vector<Value *> ArgsV;
970 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
971 ArgsV.push_back(Args[i]->Codegen());
972 if (ArgsV.back() == 0)
976 return Builder.CreateCall(CalleeF, ArgsV, "calltmp");
979 Value *IfExprAST::Codegen() {
980 KSDbgInfo.emitLocation(this);
982 Value *CondV = Cond->Codegen();
986 // Convert condition to a bool by comparing equal to 0.0.
987 CondV = Builder.CreateFCmpONE(
988 CondV, ConstantFP::get(getGlobalContext(), APFloat(0.0)), "ifcond");
990 Function *TheFunction = Builder.GetInsertBlock()->getParent();
992 // Create blocks for the then and else cases. Insert the 'then' block at the
993 // end of the function.
995 BasicBlock::Create(getGlobalContext(), "then", TheFunction);
996 BasicBlock *ElseBB = BasicBlock::Create(getGlobalContext(), "else");
997 BasicBlock *MergeBB = BasicBlock::Create(getGlobalContext(), "ifcont");
999 Builder.CreateCondBr(CondV, ThenBB, ElseBB);
1002 Builder.SetInsertPoint(ThenBB);
1004 Value *ThenV = Then->Codegen();
1008 Builder.CreateBr(MergeBB);
1009 // Codegen of 'Then' can change the current block, update ThenBB for the PHI.
1010 ThenBB = Builder.GetInsertBlock();
1013 TheFunction->getBasicBlockList().push_back(ElseBB);
1014 Builder.SetInsertPoint(ElseBB);
1016 Value *ElseV = Else->Codegen();
1020 Builder.CreateBr(MergeBB);
1021 // Codegen of 'Else' can change the current block, update ElseBB for the PHI.
1022 ElseBB = Builder.GetInsertBlock();
1024 // Emit merge block.
1025 TheFunction->getBasicBlockList().push_back(MergeBB);
1026 Builder.SetInsertPoint(MergeBB);
1028 Builder.CreatePHI(Type::getDoubleTy(getGlobalContext()), 2, "iftmp");
1030 PN->addIncoming(ThenV, ThenBB);
1031 PN->addIncoming(ElseV, ElseBB);
1035 Value *ForExprAST::Codegen() {
1037 // var = alloca double
1039 // start = startexpr
1040 // store start -> var
1048 // endcond = endexpr
1050 // curvar = load var
1051 // nextvar = curvar + step
1052 // store nextvar -> var
1053 // br endcond, loop, endloop
1056 Function *TheFunction = Builder.GetInsertBlock()->getParent();
1058 // Create an alloca for the variable in the entry block.
1059 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1061 KSDbgInfo.emitLocation(this);
1063 // Emit the start code first, without 'variable' in scope.
1064 Value *StartVal = Start->Codegen();
1068 // Store the value into the alloca.
1069 Builder.CreateStore(StartVal, Alloca);
1071 // Make the new basic block for the loop header, inserting after current
1073 BasicBlock *LoopBB =
1074 BasicBlock::Create(getGlobalContext(), "loop", TheFunction);
1076 // Insert an explicit fall through from the current block to the LoopBB.
1077 Builder.CreateBr(LoopBB);
1079 // Start insertion in LoopBB.
1080 Builder.SetInsertPoint(LoopBB);
1082 // Within the loop, the variable is defined equal to the PHI node. If it
1083 // shadows an existing variable, we have to restore it, so save it now.
1084 AllocaInst *OldVal = NamedValues[VarName];
1085 NamedValues[VarName] = Alloca;
1087 // Emit the body of the loop. This, like any other expr, can change the
1088 // current BB. Note that we ignore the value computed by the body, but don't
1090 if (Body->Codegen() == 0)
1093 // Emit the step value.
1096 StepVal = Step->Codegen();
1100 // If not specified, use 1.0.
1101 StepVal = ConstantFP::get(getGlobalContext(), APFloat(1.0));
1104 // Compute the end condition.
1105 Value *EndCond = End->Codegen();
1109 // Reload, increment, and restore the alloca. This handles the case where
1110 // the body of the loop mutates the variable.
1111 Value *CurVar = Builder.CreateLoad(Alloca, VarName.c_str());
1112 Value *NextVar = Builder.CreateFAdd(CurVar, StepVal, "nextvar");
1113 Builder.CreateStore(NextVar, Alloca);
1115 // Convert condition to a bool by comparing equal to 0.0.
1116 EndCond = Builder.CreateFCmpONE(
1117 EndCond, ConstantFP::get(getGlobalContext(), APFloat(0.0)), "loopcond");
1119 // Create the "after loop" block and insert it.
1120 BasicBlock *AfterBB =
1121 BasicBlock::Create(getGlobalContext(), "afterloop", TheFunction);
1123 // Insert the conditional branch into the end of LoopEndBB.
1124 Builder.CreateCondBr(EndCond, LoopBB, AfterBB);
1126 // Any new code will be inserted in AfterBB.
1127 Builder.SetInsertPoint(AfterBB);
1129 // Restore the unshadowed variable.
1131 NamedValues[VarName] = OldVal;
1133 NamedValues.erase(VarName);
1135 // for expr always returns 0.0.
1136 return Constant::getNullValue(Type::getDoubleTy(getGlobalContext()));
1139 Value *VarExprAST::Codegen() {
1140 std::vector<AllocaInst *> OldBindings;
1142 Function *TheFunction = Builder.GetInsertBlock()->getParent();
1144 // Register all variables and emit their initializer.
1145 for (unsigned i = 0, e = VarNames.size(); i != e; ++i) {
1146 const std::string &VarName = VarNames[i].first;
1147 ExprAST *Init = VarNames[i].second;
1149 // Emit the initializer before adding the variable to scope, this prevents
1150 // the initializer from referencing the variable itself, and permits stuff
1153 // var a = a in ... # refers to outer 'a'.
1156 InitVal = Init->Codegen();
1159 } else { // If not specified, use 0.0.
1160 InitVal = ConstantFP::get(getGlobalContext(), APFloat(0.0));
1163 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1164 Builder.CreateStore(InitVal, Alloca);
1166 // Remember the old variable binding so that we can restore the binding when
1168 OldBindings.push_back(NamedValues[VarName]);
1170 // Remember this binding.
1171 NamedValues[VarName] = Alloca;
1174 KSDbgInfo.emitLocation(this);
1176 // Codegen the body, now that all vars are in scope.
1177 Value *BodyVal = Body->Codegen();
1181 // Pop all our variables from scope.
1182 for (unsigned i = 0, e = VarNames.size(); i != e; ++i)
1183 NamedValues[VarNames[i].first] = OldBindings[i];
1185 // Return the body computation.
1189 Function *PrototypeAST::Codegen() {
1190 // Make the function type: double(double,double) etc.
1191 std::vector<Type *> Doubles(Args.size(),
1192 Type::getDoubleTy(getGlobalContext()));
1194 FunctionType::get(Type::getDoubleTy(getGlobalContext()), Doubles, false);
1197 Function::Create(FT, Function::ExternalLinkage, Name, TheModule);
1199 // If F conflicted, there was already something named 'Name'. If it has a
1200 // body, don't allow redefinition or reextern.
1201 if (F->getName() != Name) {
1202 // Delete the one we just made and get the existing one.
1203 F->eraseFromParent();
1204 F = TheModule->getFunction(Name);
1206 // If F already has a body, reject this.
1208 ErrorF("redefinition of function");
1212 // If F took a different number of args, reject.
1213 if (F->arg_size() != Args.size()) {
1214 ErrorF("redefinition of function with different # args");
1219 // Set names for all arguments.
1221 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
1223 AI->setName(Args[Idx]);
1225 // Create a subprogram DIE for this function.
1226 DIFile Unit = DBuilder->createFile(KSDbgInfo.TheCU.getFilename(),
1227 KSDbgInfo.TheCU.getDirectory());
1228 DIDescriptor FContext(Unit);
1229 unsigned LineNo = Line;
1230 unsigned ScopeLine = Line;
1231 DISubprogram SP = DBuilder->createFunction(
1232 FContext, Name, StringRef(), Unit, LineNo,
1233 CreateFunctionType(Args.size(), Unit), false /* internal linkage */,
1234 true /* definition */, ScopeLine, DIDescriptor::FlagPrototyped, false, F);
1236 KSDbgInfo.FnScopeMap[this] = SP;
1240 /// CreateArgumentAllocas - Create an alloca for each argument and register the
1241 /// argument in the symbol table so that references to it will succeed.
1242 void PrototypeAST::CreateArgumentAllocas(Function *F) {
1243 Function::arg_iterator AI = F->arg_begin();
1244 for (unsigned Idx = 0, e = Args.size(); Idx != e; ++Idx, ++AI) {
1245 // Create an alloca for this variable.
1246 AllocaInst *Alloca = CreateEntryBlockAlloca(F, Args[Idx]);
1248 // Create a debug descriptor for the variable.
1249 DIScope *Scope = KSDbgInfo.LexicalBlocks.back();
1250 DIFile Unit = DBuilder->createFile(KSDbgInfo.TheCU.getFilename(),
1251 KSDbgInfo.TheCU.getDirectory());
1252 DIVariable D = DBuilder->createLocalVariable(dwarf::DW_TAG_arg_variable,
1253 *Scope, Args[Idx], Unit, Line,
1254 KSDbgInfo.getDoubleTy(), Idx);
1256 Instruction *Call = DBuilder->insertDeclare(
1257 Alloca, D, DBuilder->createExpression(), Builder.GetInsertBlock());
1258 Call->setDebugLoc(DebugLoc::get(Line, 0, *Scope));
1260 // Store the initial value into the alloca.
1261 Builder.CreateStore(AI, Alloca);
1263 // Add arguments to variable symbol table.
1264 NamedValues[Args[Idx]] = Alloca;
1268 Function *FunctionAST::Codegen() {
1269 NamedValues.clear();
1271 Function *TheFunction = Proto->Codegen();
1272 if (TheFunction == 0)
1275 // Push the current scope.
1276 KSDbgInfo.LexicalBlocks.push_back(&KSDbgInfo.FnScopeMap[Proto]);
1278 // Unset the location for the prologue emission (leading instructions with no
1279 // location in a function are considered part of the prologue and the debugger
1280 // will run past them when breaking on a function)
1281 KSDbgInfo.emitLocation(nullptr);
1283 // If this is an operator, install it.
1284 if (Proto->isBinaryOp())
1285 BinopPrecedence[Proto->getOperatorName()] = Proto->getBinaryPrecedence();
1287 // Create a new basic block to start insertion into.
1288 BasicBlock *BB = BasicBlock::Create(getGlobalContext(), "entry", TheFunction);
1289 Builder.SetInsertPoint(BB);
1291 // Add all arguments to the symbol table and create their allocas.
1292 Proto->CreateArgumentAllocas(TheFunction);
1294 KSDbgInfo.emitLocation(Body);
1296 if (Value *RetVal = Body->Codegen()) {
1297 // Finish off the function.
1298 Builder.CreateRet(RetVal);
1300 // Pop off the lexical block for the function.
1301 KSDbgInfo.LexicalBlocks.pop_back();
1303 // Validate the generated code, checking for consistency.
1304 verifyFunction(*TheFunction);
1306 // Optimize the function.
1307 TheFPM->run(*TheFunction);
1312 // Error reading body, remove function.
1313 TheFunction->eraseFromParent();
1315 if (Proto->isBinaryOp())
1316 BinopPrecedence.erase(Proto->getOperatorName());
1318 // Pop off the lexical block for the function since we added it
1320 KSDbgInfo.LexicalBlocks.pop_back();
1325 //===----------------------------------------------------------------------===//
1326 // Top-Level parsing and JIT Driver
1327 //===----------------------------------------------------------------------===//
1329 static ExecutionEngine *TheExecutionEngine;
1331 static void HandleDefinition() {
1332 if (FunctionAST *F = ParseDefinition()) {
1333 if (!F->Codegen()) {
1334 fprintf(stderr, "Error reading function definition:");
1337 // Skip token for error recovery.
1342 static void HandleExtern() {
1343 if (PrototypeAST *P = ParseExtern()) {
1344 if (!P->Codegen()) {
1345 fprintf(stderr, "Error reading extern");
1348 // Skip token for error recovery.
1353 static void HandleTopLevelExpression() {
1354 // Evaluate a top-level expression into an anonymous function.
1355 if (FunctionAST *F = ParseTopLevelExpr()) {
1356 if (!F->Codegen()) {
1357 fprintf(stderr, "Error generating code for top level expr");
1360 // Skip token for error recovery.
1365 /// top ::= definition | external | expression | ';'
1366 static void MainLoop() {
1373 break; // ignore top-level semicolons.
1381 HandleTopLevelExpression();
1387 //===----------------------------------------------------------------------===//
1388 // "Library" functions that can be "extern'd" from user code.
1389 //===----------------------------------------------------------------------===//
1391 /// putchard - putchar that takes a double and returns 0.
1392 extern "C" double putchard(double X) {
1397 /// printd - printf that takes a double prints it as "%f\n", returning 0.
1398 extern "C" double printd(double X) {
1403 //===----------------------------------------------------------------------===//
1404 // Main driver code.
1405 //===----------------------------------------------------------------------===//
1408 InitializeNativeTarget();
1409 InitializeNativeTargetAsmPrinter();
1410 InitializeNativeTargetAsmParser();
1411 LLVMContext &Context = getGlobalContext();
1413 // Install standard binary operators.
1414 // 1 is lowest precedence.
1415 BinopPrecedence['='] = 2;
1416 BinopPrecedence['<'] = 10;
1417 BinopPrecedence['+'] = 20;
1418 BinopPrecedence['-'] = 20;
1419 BinopPrecedence['*'] = 40; // highest.
1421 // Prime the first token.
1424 // Make the module, which holds all the code.
1425 std::unique_ptr<Module> Owner = make_unique<Module>("my cool jit", Context);
1426 TheModule = Owner.get();
1428 // Add the current debug info version into the module.
1429 TheModule->addModuleFlag(Module::Warning, "Debug Info Version",
1430 DEBUG_METADATA_VERSION);
1432 // Darwin only supports dwarf2.
1433 if (Triple(sys::getProcessTriple()).isOSDarwin())
1434 TheModule->addModuleFlag(llvm::Module::Warning, "Dwarf Version", 2);
1436 // Construct the DIBuilder, we do this here because we need the module.
1437 DBuilder = new DIBuilder(*TheModule);
1439 // Create the compile unit for the module.
1440 // Currently down as "fib.ks" as a filename since we're redirecting stdin
1441 // but we'd like actual source locations.
1442 KSDbgInfo.TheCU = DBuilder->createCompileUnit(
1443 dwarf::DW_LANG_C, "fib.ks", ".", "Kaleidoscope Compiler", 0, "", 0);
1445 // Create the JIT. This takes ownership of the module.
1447 TheExecutionEngine =
1448 EngineBuilder(std::move(Owner))
1449 .setErrorStr(&ErrStr)
1450 .setMCJITMemoryManager(llvm::make_unique<SectionMemoryManager>())
1452 if (!TheExecutionEngine) {
1453 fprintf(stderr, "Could not create ExecutionEngine: %s\n", ErrStr.c_str());
1457 FunctionPassManager OurFPM(TheModule);
1459 // Set up the optimizer pipeline. Start with registering info about how the
1460 // target lays out data structures.
1461 TheModule->setDataLayout(TheExecutionEngine->getDataLayout());
1462 OurFPM.add(new DataLayoutPass());
1464 // Provide basic AliasAnalysis support for GVN.
1465 OurFPM.add(createBasicAliasAnalysisPass());
1466 // Promote allocas to registers.
1467 OurFPM.add(createPromoteMemoryToRegisterPass());
1468 // Do simple "peephole" optimizations and bit-twiddling optzns.
1469 OurFPM.add(createInstructionCombiningPass());
1470 // Reassociate expressions.
1471 OurFPM.add(createReassociatePass());
1472 // Eliminate Common SubExpressions.
1473 OurFPM.add(createGVNPass());
1474 // Simplify the control flow graph (deleting unreachable blocks, etc).
1475 OurFPM.add(createCFGSimplificationPass());
1477 OurFPM.doInitialization();
1479 // Set the global so the code gen can use this.
1482 // Run the main "interpreter loop" now.
1487 // Finalize the debug info.
1488 DBuilder->finalize();
1490 // Print out all of the generated code.