1 #include "llvm/Analysis/Passes.h"
2 #include "llvm/ExecutionEngine/Orc/CompileUtils.h"
3 #include "llvm/ExecutionEngine/Orc/IRCompileLayer.h"
4 #include "llvm/ExecutionEngine/Orc/LazyEmittingLayer.h"
5 #include "llvm/ExecutionEngine/Orc/ObjectLinkingLayer.h"
6 #include "llvm/IR/DataLayout.h"
7 #include "llvm/IR/DerivedTypes.h"
8 #include "llvm/IR/IRBuilder.h"
9 #include "llvm/IR/LLVMContext.h"
10 #include "llvm/IR/Module.h"
11 #include "llvm/IR/Verifier.h"
12 #include "llvm/PassManager.h"
13 #include "llvm/Support/TargetSelect.h"
14 #include "llvm/Transforms/Scalar.h"
22 //===----------------------------------------------------------------------===//
24 //===----------------------------------------------------------------------===//
26 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
27 // of these for known things.
32 tok_def = -2, tok_extern = -3,
35 tok_identifier = -4, tok_number = -5,
38 tok_if = -6, tok_then = -7, tok_else = -8,
39 tok_for = -9, tok_in = -10,
42 tok_binary = -11, tok_unary = -12,
48 static std::string IdentifierStr; // Filled in if tok_identifier
49 static double NumVal; // Filled in if tok_number
51 /// gettok - Return the next token from standard input.
53 static int LastChar = ' ';
55 // Skip any whitespace.
56 while (isspace(LastChar))
59 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
60 IdentifierStr = LastChar;
61 while (isalnum((LastChar = getchar())))
62 IdentifierStr += LastChar;
64 if (IdentifierStr == "def") return tok_def;
65 if (IdentifierStr == "extern") return tok_extern;
66 if (IdentifierStr == "if") return tok_if;
67 if (IdentifierStr == "then") return tok_then;
68 if (IdentifierStr == "else") return tok_else;
69 if (IdentifierStr == "for") return tok_for;
70 if (IdentifierStr == "in") return tok_in;
71 if (IdentifierStr == "binary") return tok_binary;
72 if (IdentifierStr == "unary") return tok_unary;
73 if (IdentifierStr == "var") return tok_var;
74 return tok_identifier;
77 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+
82 } while (isdigit(LastChar) || LastChar == '.');
84 NumVal = strtod(NumStr.c_str(), 0);
88 if (LastChar == '#') {
89 // Comment until end of line.
90 do LastChar = getchar();
91 while (LastChar != EOF && LastChar != '\n' && LastChar != '\r');
97 // Check for end of file. Don't eat the EOF.
101 // Otherwise, just return the character as its ascii value.
102 int ThisChar = LastChar;
103 LastChar = getchar();
107 //===----------------------------------------------------------------------===//
108 // Abstract Syntax Tree (aka Parse Tree)
109 //===----------------------------------------------------------------------===//
113 /// ExprAST - Base class for all expression nodes.
115 virtual ~ExprAST() {}
116 virtual Value* IRGen(IRGenContext &C) = 0;
119 /// NumberExprAST - Expression class for numeric literals like "1.0".
120 struct NumberExprAST : public ExprAST {
121 NumberExprAST(double Val) : Val(Val) {}
122 Value* IRGen(IRGenContext &C) override;
127 /// VariableExprAST - Expression class for referencing a variable, like "a".
128 struct VariableExprAST : public ExprAST {
129 VariableExprAST(std::string Name) : Name(std::move(Name)) {}
130 Value* IRGen(IRGenContext &C) override;
135 /// UnaryExprAST - Expression class for a unary operator.
136 struct UnaryExprAST : public ExprAST {
137 UnaryExprAST(char Opcode, std::unique_ptr<ExprAST> Operand)
138 : Opcode(std::move(Opcode)), Operand(std::move(Operand)) {}
140 Value* IRGen(IRGenContext &C) override;
143 std::unique_ptr<ExprAST> Operand;
146 /// BinaryExprAST - Expression class for a binary operator.
147 struct BinaryExprAST : public ExprAST {
148 BinaryExprAST(char Op, std::unique_ptr<ExprAST> LHS,
149 std::unique_ptr<ExprAST> RHS)
150 : Op(Op), LHS(std::move(LHS)), RHS(std::move(RHS)) {}
152 Value* IRGen(IRGenContext &C) override;
155 std::unique_ptr<ExprAST> LHS, RHS;
158 /// CallExprAST - Expression class for function calls.
159 struct CallExprAST : public ExprAST {
160 CallExprAST(std::string CalleeName,
161 std::vector<std::unique_ptr<ExprAST>> Args)
162 : CalleeName(std::move(CalleeName)), Args(std::move(Args)) {}
164 Value* IRGen(IRGenContext &C) override;
166 std::string CalleeName;
167 std::vector<std::unique_ptr<ExprAST>> Args;
170 /// IfExprAST - Expression class for if/then/else.
171 struct IfExprAST : public ExprAST {
172 IfExprAST(std::unique_ptr<ExprAST> Cond, std::unique_ptr<ExprAST> Then,
173 std::unique_ptr<ExprAST> Else)
174 : Cond(std::move(Cond)), Then(std::move(Then)), Else(std::move(Else)) {}
175 Value* IRGen(IRGenContext &C) override;
177 std::unique_ptr<ExprAST> Cond, Then, Else;
180 /// ForExprAST - Expression class for for/in.
181 struct ForExprAST : public ExprAST {
182 ForExprAST(std::string VarName, std::unique_ptr<ExprAST> Start,
183 std::unique_ptr<ExprAST> End, std::unique_ptr<ExprAST> Step,
184 std::unique_ptr<ExprAST> Body)
185 : VarName(std::move(VarName)), Start(std::move(Start)), End(std::move(End)),
186 Step(std::move(Step)), Body(std::move(Body)) {}
188 Value* IRGen(IRGenContext &C) override;
191 std::unique_ptr<ExprAST> Start, End, Step, Body;
194 /// VarExprAST - Expression class for var/in
195 struct VarExprAST : public ExprAST {
196 typedef std::pair<std::string, std::unique_ptr<ExprAST>> Binding;
197 typedef std::vector<Binding> BindingList;
199 VarExprAST(BindingList VarBindings, std::unique_ptr<ExprAST> Body)
200 : VarBindings(std::move(VarBindings)), Body(std::move(Body)) {}
202 Value* IRGen(IRGenContext &C) override;
204 BindingList VarBindings;
205 std::unique_ptr<ExprAST> Body;
208 /// PrototypeAST - This class represents the "prototype" for a function,
209 /// which captures its argument names as well as if it is an operator.
210 struct PrototypeAST {
211 PrototypeAST(std::string Name, std::vector<std::string> Args,
212 bool IsOperator = false, unsigned Precedence = 0)
213 : Name(std::move(Name)), Args(std::move(Args)), IsOperator(IsOperator),
214 Precedence(Precedence) {}
216 Function* IRGen(IRGenContext &C);
217 void CreateArgumentAllocas(Function *F, IRGenContext &C);
219 bool isUnaryOp() const { return IsOperator && Args.size() == 1; }
220 bool isBinaryOp() const { return IsOperator && Args.size() == 2; }
222 char getOperatorName() const {
223 assert(isUnaryOp() || isBinaryOp());
224 return Name[Name.size()-1];
228 std::vector<std::string> Args;
230 unsigned Precedence; // Precedence if a binary op.
233 /// FunctionAST - This class represents a function definition itself.
235 FunctionAST(std::unique_ptr<PrototypeAST> Proto,
236 std::unique_ptr<ExprAST> Body)
237 : Proto(std::move(Proto)), Body(std::move(Body)) {}
239 Function* IRGen(IRGenContext &C);
241 std::unique_ptr<PrototypeAST> Proto;
242 std::unique_ptr<ExprAST> Body;
245 //===----------------------------------------------------------------------===//
247 //===----------------------------------------------------------------------===//
249 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
250 /// token the parser is looking at. getNextToken reads another token from the
251 /// lexer and updates CurTok with its results.
253 static int getNextToken() {
254 return CurTok = gettok();
257 /// BinopPrecedence - This holds the precedence for each binary operator that is
259 static std::map<char, int> BinopPrecedence;
261 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
262 static int GetTokPrecedence() {
263 if (!isascii(CurTok))
266 // Make sure it's a declared binop.
267 int TokPrec = BinopPrecedence[CurTok];
268 if (TokPrec <= 0) return -1;
272 template <typename T>
273 std::unique_ptr<T> ErrorU(const char *Str) {
274 fprintf(stderr, "Error: %s\n", Str);
278 template <typename T>
279 T* ErrorP(const char *Str) {
280 fprintf(stderr, "Error: %s\n", Str);
284 static std::unique_ptr<ExprAST> ParseExpression();
288 /// ::= identifier '(' expression* ')'
289 static std::unique_ptr<ExprAST> ParseIdentifierExpr() {
290 std::string IdName = IdentifierStr;
292 getNextToken(); // eat identifier.
294 if (CurTok != '(') // Simple variable ref.
295 return llvm::make_unique<VariableExprAST>(IdName);
298 getNextToken(); // eat (
299 std::vector<std::unique_ptr<ExprAST>> Args;
302 auto Arg = ParseExpression();
303 if (!Arg) return nullptr;
304 Args.push_back(std::move(Arg));
306 if (CurTok == ')') break;
309 return ErrorU<CallExprAST>("Expected ')' or ',' in argument list");
317 return llvm::make_unique<CallExprAST>(IdName, std::move(Args));
320 /// numberexpr ::= number
321 static std::unique_ptr<NumberExprAST> ParseNumberExpr() {
322 auto Result = llvm::make_unique<NumberExprAST>(NumVal);
323 getNextToken(); // consume the number
327 /// parenexpr ::= '(' expression ')'
328 static std::unique_ptr<ExprAST> ParseParenExpr() {
329 getNextToken(); // eat (.
330 auto V = ParseExpression();
335 return ErrorU<ExprAST>("expected ')'");
336 getNextToken(); // eat ).
340 /// ifexpr ::= 'if' expression 'then' expression 'else' expression
341 static std::unique_ptr<ExprAST> ParseIfExpr() {
342 getNextToken(); // eat the if.
345 auto Cond = ParseExpression();
349 if (CurTok != tok_then)
350 return ErrorU<ExprAST>("expected then");
351 getNextToken(); // eat the then
353 auto Then = ParseExpression();
357 if (CurTok != tok_else)
358 return ErrorU<ExprAST>("expected else");
362 auto Else = ParseExpression();
366 return llvm::make_unique<IfExprAST>(std::move(Cond), std::move(Then),
370 /// forexpr ::= 'for' identifier '=' expr ',' expr (',' expr)? 'in' expression
371 static std::unique_ptr<ForExprAST> ParseForExpr() {
372 getNextToken(); // eat the for.
374 if (CurTok != tok_identifier)
375 return ErrorU<ForExprAST>("expected identifier after for");
377 std::string IdName = IdentifierStr;
378 getNextToken(); // eat identifier.
381 return ErrorU<ForExprAST>("expected '=' after for");
382 getNextToken(); // eat '='.
385 auto Start = ParseExpression();
389 return ErrorU<ForExprAST>("expected ',' after for start value");
392 auto End = ParseExpression();
396 // The step value is optional.
397 std::unique_ptr<ExprAST> Step;
400 Step = ParseExpression();
405 if (CurTok != tok_in)
406 return ErrorU<ForExprAST>("expected 'in' after for");
407 getNextToken(); // eat 'in'.
409 auto Body = ParseExpression();
413 return llvm::make_unique<ForExprAST>(IdName, std::move(Start), std::move(End),
414 std::move(Step), std::move(Body));
417 /// varexpr ::= 'var' identifier ('=' expression)?
418 // (',' identifier ('=' expression)?)* 'in' expression
419 static std::unique_ptr<VarExprAST> ParseVarExpr() {
420 getNextToken(); // eat the var.
422 VarExprAST::BindingList VarBindings;
424 // At least one variable name is required.
425 if (CurTok != tok_identifier)
426 return ErrorU<VarExprAST>("expected identifier after var");
429 std::string Name = IdentifierStr;
430 getNextToken(); // eat identifier.
432 // Read the optional initializer.
433 std::unique_ptr<ExprAST> Init;
435 getNextToken(); // eat the '='.
437 Init = ParseExpression();
442 VarBindings.push_back(VarExprAST::Binding(Name, std::move(Init)));
444 // End of var list, exit loop.
445 if (CurTok != ',') break;
446 getNextToken(); // eat the ','.
448 if (CurTok != tok_identifier)
449 return ErrorU<VarExprAST>("expected identifier list after var");
452 // At this point, we have to have 'in'.
453 if (CurTok != tok_in)
454 return ErrorU<VarExprAST>("expected 'in' keyword after 'var'");
455 getNextToken(); // eat 'in'.
457 auto Body = ParseExpression();
461 return llvm::make_unique<VarExprAST>(std::move(VarBindings), std::move(Body));
465 /// ::= identifierexpr
471 static std::unique_ptr<ExprAST> ParsePrimary() {
473 default: return ErrorU<ExprAST>("unknown token when expecting an expression");
474 case tok_identifier: return ParseIdentifierExpr();
475 case tok_number: return ParseNumberExpr();
476 case '(': return ParseParenExpr();
477 case tok_if: return ParseIfExpr();
478 case tok_for: return ParseForExpr();
479 case tok_var: return ParseVarExpr();
486 static std::unique_ptr<ExprAST> ParseUnary() {
487 // If the current token is not an operator, it must be a primary expr.
488 if (!isascii(CurTok) || CurTok == '(' || CurTok == ',')
489 return ParsePrimary();
491 // If this is a unary operator, read it.
494 if (auto Operand = ParseUnary())
495 return llvm::make_unique<UnaryExprAST>(Opc, std::move(Operand));
501 static std::unique_ptr<ExprAST> ParseBinOpRHS(int ExprPrec,
502 std::unique_ptr<ExprAST> LHS) {
503 // If this is a binop, find its precedence.
505 int TokPrec = GetTokPrecedence();
507 // If this is a binop that binds at least as tightly as the current binop,
508 // consume it, otherwise we are done.
509 if (TokPrec < ExprPrec)
512 // Okay, we know this is a binop.
514 getNextToken(); // eat binop
516 // Parse the unary expression after the binary operator.
517 auto RHS = ParseUnary();
521 // If BinOp binds less tightly with RHS than the operator after RHS, let
522 // the pending operator take RHS as its LHS.
523 int NextPrec = GetTokPrecedence();
524 if (TokPrec < NextPrec) {
525 RHS = ParseBinOpRHS(TokPrec+1, std::move(RHS));
531 LHS = llvm::make_unique<BinaryExprAST>(BinOp, std::move(LHS), std::move(RHS));
536 /// ::= unary binoprhs
538 static std::unique_ptr<ExprAST> ParseExpression() {
539 auto LHS = ParseUnary();
543 return ParseBinOpRHS(0, std::move(LHS));
547 /// ::= id '(' id* ')'
548 /// ::= binary LETTER number? (id, id)
549 /// ::= unary LETTER (id)
550 static std::unique_ptr<PrototypeAST> ParsePrototype() {
553 unsigned Kind = 0; // 0 = identifier, 1 = unary, 2 = binary.
554 unsigned BinaryPrecedence = 30;
558 return ErrorU<PrototypeAST>("Expected function name in prototype");
560 FnName = IdentifierStr;
566 if (!isascii(CurTok))
567 return ErrorU<PrototypeAST>("Expected unary operator");
569 FnName += (char)CurTok;
575 if (!isascii(CurTok))
576 return ErrorU<PrototypeAST>("Expected binary operator");
578 FnName += (char)CurTok;
582 // Read the precedence if present.
583 if (CurTok == tok_number) {
584 if (NumVal < 1 || NumVal > 100)
585 return ErrorU<PrototypeAST>("Invalid precedecnce: must be 1..100");
586 BinaryPrecedence = (unsigned)NumVal;
593 return ErrorU<PrototypeAST>("Expected '(' in prototype");
595 std::vector<std::string> ArgNames;
596 while (getNextToken() == tok_identifier)
597 ArgNames.push_back(IdentifierStr);
599 return ErrorU<PrototypeAST>("Expected ')' in prototype");
602 getNextToken(); // eat ')'.
604 // Verify right number of names for operator.
605 if (Kind && ArgNames.size() != Kind)
606 return ErrorU<PrototypeAST>("Invalid number of operands for operator");
608 return llvm::make_unique<PrototypeAST>(FnName, std::move(ArgNames), Kind != 0,
612 /// definition ::= 'def' prototype expression
613 static std::unique_ptr<FunctionAST> ParseDefinition() {
614 getNextToken(); // eat def.
615 auto Proto = ParsePrototype();
619 if (auto Body = ParseExpression())
620 return llvm::make_unique<FunctionAST>(std::move(Proto), std::move(Body));
624 /// toplevelexpr ::= expression
625 static std::unique_ptr<FunctionAST> ParseTopLevelExpr() {
626 if (auto E = ParseExpression()) {
627 // Make an anonymous proto.
629 llvm::make_unique<PrototypeAST>("__anon_expr", std::vector<std::string>());
630 return llvm::make_unique<FunctionAST>(std::move(Proto), std::move(E));
635 /// external ::= 'extern' prototype
636 static std::unique_ptr<PrototypeAST> ParseExtern() {
637 getNextToken(); // eat extern.
638 return ParsePrototype();
641 //===----------------------------------------------------------------------===//
643 //===----------------------------------------------------------------------===//
645 // FIXME: Obviously we can do better than this
646 std::string GenerateUniqueName(const char *root)
650 sprintf(s, "%s%d", root, i++);
655 std::string MakeLegalFunctionName(std::string Name)
658 assert(!Name.empty() && "Base name must not be empty");
660 // Start with what we have
663 // Look for a numberic first character
664 if (NewName.find_first_of("0123456789") == 0) {
665 NewName.insert(0, 1, 'n');
668 // Replace illegal characters with their ASCII equivalent
669 std::string legal_elements = "_abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789";
671 while ((pos = NewName.find_first_not_of(legal_elements)) != std::string::npos) {
672 char old_c = NewName.at(pos);
674 sprintf(new_str, "%d", (int)old_c);
675 NewName = NewName.replace(pos, 1, new_str);
681 class SessionContext {
683 SessionContext(LLVMContext &C) : Context(C) {}
684 LLVMContext& getLLVMContext() const { return Context; }
685 void addPrototypeAST(std::unique_ptr<PrototypeAST> P);
686 PrototypeAST* getPrototypeAST(const std::string &Name);
687 std::map<std::string, std::unique_ptr<FunctionAST>> FunctionDefs;
689 typedef std::map<std::string, std::unique_ptr<PrototypeAST>> PrototypeMap;
690 LLVMContext &Context;
691 PrototypeMap Prototypes;
694 void SessionContext::addPrototypeAST(std::unique_ptr<PrototypeAST> P) {
695 Prototypes[P->Name] = std::move(P);
698 PrototypeAST* SessionContext::getPrototypeAST(const std::string &Name) {
699 PrototypeMap::iterator I = Prototypes.find(Name);
700 if (I != Prototypes.end())
701 return I->second.get();
708 IRGenContext(SessionContext &S)
710 M(new Module(GenerateUniqueName("jit_module_"),
711 Session.getLLVMContext())),
712 Builder(Session.getLLVMContext()) {}
714 SessionContext& getSession() { return Session; }
715 Module& getM() const { return *M; }
716 std::unique_ptr<Module> takeM() { return std::move(M); }
717 IRBuilder<>& getBuilder() { return Builder; }
718 LLVMContext& getLLVMContext() { return Session.getLLVMContext(); }
719 Function* getPrototype(const std::string &Name);
721 std::map<std::string, AllocaInst*> NamedValues;
723 SessionContext &Session;
724 std::unique_ptr<Module> M;
728 Function* IRGenContext::getPrototype(const std::string &Name) {
729 if (Function *ExistingProto = M->getFunction(Name))
730 return ExistingProto;
731 if (PrototypeAST *ProtoAST = Session.getPrototypeAST(Name))
732 return ProtoAST->IRGen(*this);
736 /// CreateEntryBlockAlloca - Create an alloca instruction in the entry block of
737 /// the function. This is used for mutable variables etc.
738 static AllocaInst *CreateEntryBlockAlloca(Function *TheFunction,
739 const std::string &VarName) {
740 IRBuilder<> TmpB(&TheFunction->getEntryBlock(),
741 TheFunction->getEntryBlock().begin());
742 return TmpB.CreateAlloca(Type::getDoubleTy(getGlobalContext()), 0,
746 Value *NumberExprAST::IRGen(IRGenContext &C) {
747 return ConstantFP::get(C.getLLVMContext(), APFloat(Val));
750 Value *VariableExprAST::IRGen(IRGenContext &C) {
751 // Look this variable up in the function.
752 Value *V = C.NamedValues[Name];
756 sprintf(ErrStr, "Unknown variable name %s", Name.c_str());
757 return ErrorP<Value>(ErrStr);
761 return C.getBuilder().CreateLoad(V, Name.c_str());
764 Value *UnaryExprAST::IRGen(IRGenContext &C) {
765 if (Value *OperandV = Operand->IRGen(C)) {
766 std::string FnName = MakeLegalFunctionName(std::string("unary")+Opcode);
767 if (Function *F = C.getPrototype(FnName))
768 return C.getBuilder().CreateCall(F, OperandV, "unop");
769 return ErrorP<Value>("Unknown unary operator");
772 // Could not codegen operand - return null.
776 Value *BinaryExprAST::IRGen(IRGenContext &C) {
777 // Special case '=' because we don't want to emit the LHS as an expression.
779 // Assignment requires the LHS to be an identifier.
780 auto LHSVar = static_cast<VariableExprAST&>(*LHS);
782 Value *Val = RHS->IRGen(C);
783 if (!Val) return nullptr;
786 if (auto Variable = C.NamedValues[LHSVar.Name]) {
787 C.getBuilder().CreateStore(Val, Variable);
790 return ErrorP<Value>("Unknown variable name");
793 Value *L = LHS->IRGen(C);
794 Value *R = RHS->IRGen(C);
795 if (!L || !R) return nullptr;
798 case '+': return C.getBuilder().CreateFAdd(L, R, "addtmp");
799 case '-': return C.getBuilder().CreateFSub(L, R, "subtmp");
800 case '*': return C.getBuilder().CreateFMul(L, R, "multmp");
801 case '/': return C.getBuilder().CreateFDiv(L, R, "divtmp");
803 L = C.getBuilder().CreateFCmpULT(L, R, "cmptmp");
804 // Convert bool 0/1 to double 0.0 or 1.0
805 return C.getBuilder().CreateUIToFP(L, Type::getDoubleTy(getGlobalContext()),
810 // If it wasn't a builtin binary operator, it must be a user defined one. Emit
812 std::string FnName = MakeLegalFunctionName(std::string("binary")+Op);
813 if (Function *F = C.getPrototype(FnName)) {
814 Value *Ops[] = { L, R };
815 return C.getBuilder().CreateCall(F, Ops, "binop");
818 return ErrorP<Value>("Unknown binary operator");
821 Value *CallExprAST::IRGen(IRGenContext &C) {
822 // Look up the name in the global module table.
823 if (auto CalleeF = C.getPrototype(CalleeName)) {
824 // If argument mismatch error.
825 if (CalleeF->arg_size() != Args.size())
826 return ErrorP<Value>("Incorrect # arguments passed");
828 std::vector<Value*> ArgsV;
829 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
830 ArgsV.push_back(Args[i]->IRGen(C));
831 if (!ArgsV.back()) return nullptr;
834 return C.getBuilder().CreateCall(CalleeF, ArgsV, "calltmp");
837 return ErrorP<Value>("Unknown function referenced");
840 Value *IfExprAST::IRGen(IRGenContext &C) {
841 Value *CondV = Cond->IRGen(C);
842 if (!CondV) return nullptr;
844 // Convert condition to a bool by comparing equal to 0.0.
846 ConstantFP::get(C.getLLVMContext(), APFloat(0.0));
847 CondV = C.getBuilder().CreateFCmpONE(CondV, FPZero, "ifcond");
849 Function *TheFunction = C.getBuilder().GetInsertBlock()->getParent();
851 // Create blocks for the then and else cases. Insert the 'then' block at the
852 // end of the function.
853 BasicBlock *ThenBB = BasicBlock::Create(C.getLLVMContext(), "then", TheFunction);
854 BasicBlock *ElseBB = BasicBlock::Create(C.getLLVMContext(), "else");
855 BasicBlock *MergeBB = BasicBlock::Create(C.getLLVMContext(), "ifcont");
857 C.getBuilder().CreateCondBr(CondV, ThenBB, ElseBB);
860 C.getBuilder().SetInsertPoint(ThenBB);
862 Value *ThenV = Then->IRGen(C);
863 if (!ThenV) return nullptr;
865 C.getBuilder().CreateBr(MergeBB);
866 // Codegen of 'Then' can change the current block, update ThenBB for the PHI.
867 ThenBB = C.getBuilder().GetInsertBlock();
870 TheFunction->getBasicBlockList().push_back(ElseBB);
871 C.getBuilder().SetInsertPoint(ElseBB);
873 Value *ElseV = Else->IRGen(C);
874 if (!ElseV) return nullptr;
876 C.getBuilder().CreateBr(MergeBB);
877 // Codegen of 'Else' can change the current block, update ElseBB for the PHI.
878 ElseBB = C.getBuilder().GetInsertBlock();
881 TheFunction->getBasicBlockList().push_back(MergeBB);
882 C.getBuilder().SetInsertPoint(MergeBB);
883 PHINode *PN = C.getBuilder().CreatePHI(Type::getDoubleTy(getGlobalContext()), 2,
886 PN->addIncoming(ThenV, ThenBB);
887 PN->addIncoming(ElseV, ElseBB);
891 Value *ForExprAST::IRGen(IRGenContext &C) {
893 // var = alloca double
896 // store start -> var
907 // nextvar = curvar + step
908 // store nextvar -> var
909 // br endcond, loop, endloop
912 Function *TheFunction = C.getBuilder().GetInsertBlock()->getParent();
914 // Create an alloca for the variable in the entry block.
915 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
917 // Emit the start code first, without 'variable' in scope.
918 Value *StartVal = Start->IRGen(C);
919 if (!StartVal) return nullptr;
921 // Store the value into the alloca.
922 C.getBuilder().CreateStore(StartVal, Alloca);
924 // Make the new basic block for the loop header, inserting after current
926 BasicBlock *LoopBB = BasicBlock::Create(getGlobalContext(), "loop", TheFunction);
928 // Insert an explicit fall through from the current block to the LoopBB.
929 C.getBuilder().CreateBr(LoopBB);
931 // Start insertion in LoopBB.
932 C.getBuilder().SetInsertPoint(LoopBB);
934 // Within the loop, the variable is defined equal to the PHI node. If it
935 // shadows an existing variable, we have to restore it, so save it now.
936 AllocaInst *OldVal = C.NamedValues[VarName];
937 C.NamedValues[VarName] = Alloca;
939 // Emit the body of the loop. This, like any other expr, can change the
940 // current BB. Note that we ignore the value computed by the body, but don't
945 // Emit the step value.
948 StepVal = Step->IRGen(C);
949 if (!StepVal) return nullptr;
951 // If not specified, use 1.0.
952 StepVal = ConstantFP::get(getGlobalContext(), APFloat(1.0));
955 // Compute the end condition.
956 Value *EndCond = End->IRGen(C);
957 if (EndCond == 0) return EndCond;
959 // Reload, increment, and restore the alloca. This handles the case where
960 // the body of the loop mutates the variable.
961 Value *CurVar = C.getBuilder().CreateLoad(Alloca, VarName.c_str());
962 Value *NextVar = C.getBuilder().CreateFAdd(CurVar, StepVal, "nextvar");
963 C.getBuilder().CreateStore(NextVar, Alloca);
965 // Convert condition to a bool by comparing equal to 0.0.
966 EndCond = C.getBuilder().CreateFCmpONE(EndCond,
967 ConstantFP::get(getGlobalContext(), APFloat(0.0)),
970 // Create the "after loop" block and insert it.
971 BasicBlock *AfterBB = BasicBlock::Create(getGlobalContext(), "afterloop", TheFunction);
973 // Insert the conditional branch into the end of LoopEndBB.
974 C.getBuilder().CreateCondBr(EndCond, LoopBB, AfterBB);
976 // Any new code will be inserted in AfterBB.
977 C.getBuilder().SetInsertPoint(AfterBB);
979 // Restore the unshadowed variable.
981 C.NamedValues[VarName] = OldVal;
983 C.NamedValues.erase(VarName);
986 // for expr always returns 0.0.
987 return Constant::getNullValue(Type::getDoubleTy(getGlobalContext()));
990 Value *VarExprAST::IRGen(IRGenContext &C) {
991 std::vector<AllocaInst *> OldBindings;
993 Function *TheFunction = C.getBuilder().GetInsertBlock()->getParent();
995 // Register all variables and emit their initializer.
996 for (unsigned i = 0, e = VarBindings.size(); i != e; ++i) {
997 auto &VarName = VarBindings[i].first;
998 auto &Init = VarBindings[i].second;
1000 // Emit the initializer before adding the variable to scope, this prevents
1001 // the initializer from referencing the variable itself, and permits stuff
1004 // var a = a in ... # refers to outer 'a'.
1007 InitVal = Init->IRGen(C);
1008 if (!InitVal) return nullptr;
1009 } else // If not specified, use 0.0.
1010 InitVal = ConstantFP::get(getGlobalContext(), APFloat(0.0));
1012 AllocaInst *Alloca = CreateEntryBlockAlloca(TheFunction, VarName);
1013 C.getBuilder().CreateStore(InitVal, Alloca);
1015 // Remember the old variable binding so that we can restore the binding when
1017 OldBindings.push_back(C.NamedValues[VarName]);
1019 // Remember this binding.
1020 C.NamedValues[VarName] = Alloca;
1023 // Codegen the body, now that all vars are in scope.
1024 Value *BodyVal = Body->IRGen(C);
1025 if (!BodyVal) return nullptr;
1027 // Pop all our variables from scope.
1028 for (unsigned i = 0, e = VarBindings.size(); i != e; ++i)
1029 C.NamedValues[VarBindings[i].first] = OldBindings[i];
1031 // Return the body computation.
1035 Function *PrototypeAST::IRGen(IRGenContext &C) {
1036 std::string FnName = MakeLegalFunctionName(Name);
1038 // Make the function type: double(double,double) etc.
1039 std::vector<Type*> Doubles(Args.size(),
1040 Type::getDoubleTy(getGlobalContext()));
1041 FunctionType *FT = FunctionType::get(Type::getDoubleTy(getGlobalContext()),
1043 Function *F = Function::Create(FT, Function::ExternalLinkage, FnName,
1046 // If F conflicted, there was already something named 'FnName'. If it has a
1047 // body, don't allow redefinition or reextern.
1048 if (F->getName() != FnName) {
1049 // Delete the one we just made and get the existing one.
1050 F->eraseFromParent();
1051 F = C.getM().getFunction(Name);
1053 // If F already has a body, reject this.
1055 ErrorP<Function>("redefinition of function");
1059 // If F took a different number of args, reject.
1060 if (F->arg_size() != Args.size()) {
1061 ErrorP<Function>("redefinition of function with different # args");
1066 // Set names for all arguments.
1068 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
1070 AI->setName(Args[Idx]);
1075 /// CreateArgumentAllocas - Create an alloca for each argument and register the
1076 /// argument in the symbol table so that references to it will succeed.
1077 void PrototypeAST::CreateArgumentAllocas(Function *F, IRGenContext &C) {
1078 Function::arg_iterator AI = F->arg_begin();
1079 for (unsigned Idx = 0, e = Args.size(); Idx != e; ++Idx, ++AI) {
1080 // Create an alloca for this variable.
1081 AllocaInst *Alloca = CreateEntryBlockAlloca(F, Args[Idx]);
1083 // Store the initial value into the alloca.
1084 C.getBuilder().CreateStore(AI, Alloca);
1086 // Add arguments to variable symbol table.
1087 C.NamedValues[Args[Idx]] = Alloca;
1091 Function *FunctionAST::IRGen(IRGenContext &C) {
1092 C.NamedValues.clear();
1094 Function *TheFunction = Proto->IRGen(C);
1098 // If this is an operator, install it.
1099 if (Proto->isBinaryOp())
1100 BinopPrecedence[Proto->getOperatorName()] = Proto->Precedence;
1102 // Create a new basic block to start insertion into.
1103 BasicBlock *BB = BasicBlock::Create(getGlobalContext(), "entry", TheFunction);
1104 C.getBuilder().SetInsertPoint(BB);
1106 // Add all arguments to the symbol table and create their allocas.
1107 Proto->CreateArgumentAllocas(TheFunction, C);
1109 if (Value *RetVal = Body->IRGen(C)) {
1110 // Finish off the function.
1111 C.getBuilder().CreateRet(RetVal);
1113 // Validate the generated code, checking for consistency.
1114 verifyFunction(*TheFunction);
1119 // Error reading body, remove function.
1120 TheFunction->eraseFromParent();
1122 if (Proto->isBinaryOp())
1123 BinopPrecedence.erase(Proto->getOperatorName());
1127 //===----------------------------------------------------------------------===//
1128 // Top-Level parsing and JIT Driver
1129 //===----------------------------------------------------------------------===//
1131 class KaleidoscopeJIT {
1133 typedef ObjectLinkingLayer<> ObjLayerT;
1134 typedef IRCompileLayer<ObjLayerT> CompileLayerT;
1135 typedef LazyEmittingLayer<CompileLayerT> LazyEmitLayerT;
1137 typedef LazyEmitLayerT::ModuleSetHandleT ModuleHandleT;
1139 std::string Mangle(const std::string &Name) {
1140 std::string MangledName;
1142 raw_string_ostream MangledNameStream(MangledName);
1143 Mang.getNameWithPrefix(MangledNameStream, Name);
1148 KaleidoscopeJIT(SessionContext &Session)
1149 : TM(EngineBuilder().selectTarget()),
1150 Mang(TM->getDataLayout()), Session(Session),
1151 CompileLayer(ObjectLayer, SimpleCompiler(*TM)),
1152 LazyEmitLayer(CompileLayer) {}
1154 ModuleHandleT addModule(std::unique_ptr<Module> M) {
1155 if (!M->getDataLayout())
1156 M->setDataLayout(TM->getDataLayout());
1158 // The LazyEmitLayer takes lists of modules, rather than single modules, so
1159 // we'll just build a single-element list.
1160 std::vector<std::unique_ptr<Module>> S;
1161 S.push_back(std::move(M));
1163 // We need a memory manager to allocate memory and resolve symbols for this
1164 // new module. Create one that resolves symbols by looking back into the JIT.
1165 auto MM = createLookasideRTDyldMM<SectionMemoryManager>(
1166 [&](const std::string &Name) -> uint64_t {
1167 // First try to find 'Name' within the JIT.
1168 if (uint64_t Addr = getMangledSymbolAddress(Name))
1171 // If we don't find 'Name' in the JIT, see if we have some AST
1173 if (!Session.FunctionDefs.count(Name))
1176 // We have AST for 'Name'. IRGen it, add it to the JIT, and
1177 // return the address for it.
1178 IRGenContext C(Session);
1180 // Take ownership of the AST: We can release the memory as
1181 // soon as we've IRGen'd it.
1182 auto FuncAST = std::move(Session.FunctionDefs[Name]);
1186 addModule(C.takeM());
1187 return getMangledSymbolAddress(Name);
1189 [](const std::string &S) { return 0; } );
1191 return LazyEmitLayer.addModuleSet(std::move(S), std::move(MM));
1194 void removeModule(ModuleHandleT H) { LazyEmitLayer.removeModuleSet(H); }
1196 uint64_t getMangledSymbolAddress(const std::string &Name) {
1197 return LazyEmitLayer.getSymbolAddress(Name, false);
1200 uint64_t getSymbolAddress(const std::string &Name) {
1201 return getMangledSymbolAddress(Mangle(Name));
1206 std::unique_ptr<TargetMachine> TM;
1208 SessionContext &Session;
1210 ObjLayerT ObjectLayer;
1211 CompileLayerT CompileLayer;
1212 LazyEmitLayerT LazyEmitLayer;
1215 static void HandleDefinition(SessionContext &S, KaleidoscopeJIT &J) {
1216 if (auto F = ParseDefinition()) {
1217 S.addPrototypeAST(llvm::make_unique<PrototypeAST>(*F->Proto));
1218 S.FunctionDefs[J.Mangle(F->Proto->Name)] = std::move(F);
1220 // Skip token for error recovery.
1225 static void HandleExtern(SessionContext &S) {
1226 if (auto P = ParseExtern())
1227 S.addPrototypeAST(std::move(P));
1229 // Skip token for error recovery.
1234 static void HandleTopLevelExpression(SessionContext &S, KaleidoscopeJIT &J) {
1235 // Evaluate a top-level expression into an anonymous function.
1236 if (auto F = ParseTopLevelExpr()) {
1238 if (auto ExprFunc = F->IRGen(C)) {
1239 #ifndef MINIMAL_STDERR_OUTPUT
1240 fprintf(stderr, "Expression function:\n");
1243 // Add the CodeGen'd module to the JIT. Keep a handle to it: We can remove
1244 // this module as soon as we've executed Function ExprFunc.
1245 auto H = J.addModule(C.takeM());
1247 // Get the address of the JIT'd function in memory.
1248 uint64_t ExprFuncAddr = J.getSymbolAddress("__anon_expr");
1250 // Cast it to the right type (takes no arguments, returns a double) so we
1251 // can call it as a native function.
1252 double (*FP)() = (double (*)())(intptr_t)ExprFuncAddr;
1253 #ifdef MINIMAL_STDERR_OUTPUT
1256 fprintf(stderr, "Evaluated to %f\n", FP());
1259 // Remove the function.
1263 // Skip token for error recovery.
1268 /// top ::= definition | external | expression | ';'
1269 static void MainLoop() {
1270 SessionContext S(getGlobalContext());
1271 KaleidoscopeJIT J(S);
1274 #ifndef MINIMAL_STDERR_OUTPUT
1275 fprintf(stderr, "ready> ");
1278 case tok_eof: return;
1279 case ';': getNextToken(); break; // ignore top-level semicolons.
1280 case tok_def: HandleDefinition(S, J); break;
1281 case tok_extern: HandleExtern(S); break;
1282 default: HandleTopLevelExpression(S, J); break;
1287 //===----------------------------------------------------------------------===//
1288 // "Library" functions that can be "extern'd" from user code.
1289 //===----------------------------------------------------------------------===//
1291 /// putchard - putchar that takes a double and returns 0.
1293 double putchard(double X) {
1298 /// printd - printf that takes a double prints it as "%f\n", returning 0.
1300 double printd(double X) {
1311 //===----------------------------------------------------------------------===//
1312 // Main driver code.
1313 //===----------------------------------------------------------------------===//
1316 InitializeNativeTarget();
1317 InitializeNativeTargetAsmPrinter();
1318 InitializeNativeTargetAsmParser();
1319 LLVMContext &Context = getGlobalContext();
1321 // Install standard binary operators.
1322 // 1 is lowest precedence.
1323 BinopPrecedence['='] = 2;
1324 BinopPrecedence['<'] = 10;
1325 BinopPrecedence['+'] = 20;
1326 BinopPrecedence['-'] = 20;
1327 BinopPrecedence['/'] = 40;
1328 BinopPrecedence['*'] = 40; // highest.
1330 // Prime the first token.
1331 #ifndef MINIMAL_STDERR_OUTPUT
1332 fprintf(stderr, "ready> ");
1336 // Run the main "interpreter loop" now.