1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
6 <title>Kaleidoscope: Implementing code generation to LLVM IR</title>
7 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
8 <meta name="author" content="Chris Lattner">
9 <link rel="stylesheet" href="../llvm.css" type="text/css">
14 <div class="doc_title">Kaleidoscope: Code generation to LLVM IR</div>
16 <div class="doc_author">
17 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
20 <!-- *********************************************************************** -->
21 <div class="doc_section"><a name="intro">Part 3 Introduction</a></div>
22 <!-- *********************************************************************** -->
24 <div class="doc_text">
26 <p>Welcome to part 3 of the "<a href="index.html">Implementing a language with
27 LLVM</a>" tutorial. This chapter shows you how to transform the <a
28 href="LangImpl2.html">Abstract Syntax Tree built in Chapter 2</a> into LLVM IR.
29 This will teach you a little bit about how LLVM does things, as well as
30 demonstrate how easy it is to use. It's much more work to build a lexer and
31 parser than it is to generate LLVM IR code.
36 <!-- *********************************************************************** -->
37 <div class="doc_section"><a name="basics">Code Generation setup</a></div>
38 <!-- *********************************************************************** -->
40 <div class="doc_text">
43 In order to generate LLVM IR, we want some simple setup to get started. First,
44 we define virtual codegen methods in each AST class:</p>
46 <div class="doc_code">
48 /// ExprAST - Base class for all expression nodes.
52 virtual Value *Codegen() = 0;
55 /// NumberExprAST - Expression class for numeric literals like "1.0".
56 class NumberExprAST : public ExprAST {
59 explicit NumberExprAST(double val) : Val(val) {}
60 virtual Value *Codegen();
66 <p>The Codegen() method says to emit IR for that AST node and all things it
67 depends on, and they all return an LLVM Value object.
68 "Value" is the class used to represent a "<a
69 href="http://en.wikipedia.org/wiki/Static_single_assignment_form">Static Single
70 Assignment (SSA)</a> register" or "SSA value" in LLVM. The most distinct aspect
71 of SSA values is that their value is computed as the related instruction
72 executes, and it does not get a new value until (and if) the instruction
73 re-executes. In order words, there is no way to "change" an SSA value. For
74 more information, please read up on <a
75 href="http://en.wikipedia.org/wiki/Static_single_assignment_form">Static Single
76 Assignment</a> - the concepts are really quite natural once you grok them.</p>
79 second thing we want is an "Error" method like we used for parser, which will
80 be used to report errors found during code generation (for example, use of an
81 undeclared parameter):</p>
83 <div class="doc_code">
85 Value *ErrorV(const char *Str) { Error(Str); return 0; }
87 static Module *TheModule;
88 static LLVMBuilder Builder;
89 static std::map<std::string, Value*> NamedValues;
93 <p>The static variables will be used during code generation. <tt>TheModule</tt>
94 is the LLVM construct that contains all of the functions and global variables in
95 a chunk of code. In many ways, it is the top-level structure that the LLVM IR
96 uses to contain code.</p>
98 <p>The <tt>Builder</tt> object is a helper object that makes it easy to generate
99 LLVM instructions. The <tt>Builder</tt> keeps track of the current place to
100 insert instructions and has methods to create new instructions.</p>
102 <p>The <tt>NamedValues</tt> map keeps track of which values are defined in the
103 current scope and what their LLVM representation is. In this form of
104 Kaleidoscope, the only things that can be referenced are function parameters.
105 As such, function parameters will be in this map when generating code for their
109 With these basics in place, we can start talking about how to generate code for
110 each expression. Note that this assumes that the <tt>Builder</tt> has been set
111 up to generate code <em>into</em> something. For now, we'll assume that this
112 has already been done, and we'll just use it to emit code.
117 <!-- *********************************************************************** -->
118 <div class="doc_section"><a name="exprs">Expression Code Generation</a></div>
119 <!-- *********************************************************************** -->
121 <div class="doc_text">
123 <p>Generating LLVM code for expression nodes is very straight-forward: less
124 than 45 lines of commented code for all four of our expression nodes. First,
125 we'll do numeric literals:</p>
127 <div class="doc_code">
129 Value *NumberExprAST::Codegen() {
130 return ConstantFP::get(Type::DoubleTy, APFloat(Val));
135 <p>In the LLVM IR, numeric constants are represented with the
136 <tt>ConstantFP</tt> class, which holds the numeric value in an <tt>APFloat</tt>
137 internally (<tt>APFloat</tt> has the capability of holding floating point
138 constants of <em>A</em>rbitrary <em>P</em>recision). This code basically just
139 creates and returns a <tt>ConstantFP</tt>. Note that in the LLVM IR
140 that constants are all uniqued together and shared. For this reason, the API
141 uses "the foo::get(..)" idiom instead of "new foo(..)" or "foo::create(..).</p>
143 <div class="doc_code">
145 Value *VariableExprAST::Codegen() {
146 // Look this variable up in the function.
147 Value *V = NamedValues[Name];
148 return V ? V : ErrorV("Unknown variable name");
153 <p>References to variables is also quite simple here. In the simple version
154 of Kaleidoscope, we assume that the variable has already been emited somewhere
155 and its value is available. In practice, the only values that can be in the
156 <tt>NamedValues</tt> map are function arguments. This
157 code simply checks to see that the specified name is in the map (if not, an
158 unknown variable is being referenced) and returns the value for it.</p>
160 <div class="doc_code">
162 Value *BinaryExprAST::Codegen() {
163 Value *L = LHS->Codegen();
164 Value *R = RHS->Codegen();
165 if (L == 0 || R == 0) return 0;
168 case '+': return Builder.CreateAdd(L, R, "addtmp");
169 case '-': return Builder.CreateSub(L, R, "subtmp");
170 case '*': return Builder.CreateMul(L, R, "multmp");
172 L = Builder.CreateFCmpULT(L, R, "multmp");
173 // Convert bool 0/1 to double 0.0 or 1.0
174 return Builder.CreateUIToFP(L, Type::DoubleTy, "booltmp");
175 default: return ErrorV("invalid binary operator");
181 <p>Binary operators start to get more interesting. The basic idea here is that
182 we recursively emit code for the left-hand side of the expression, then the
183 right-hand side, then we compute the result of the binary expression. In this
184 code, we do a simple switch on the opcode to create the right LLVM instruction.
187 <p>In this example, the LLVM builder class is starting to show its value.
188 Because it knows where to insert the newly created instruction, you just have to
189 specificy what instruction to create (e.g. with <tt>CreateAdd</tt>), which
190 operands to use (<tt>L</tt> and <tt>R</tt> here) and optionally provide a name
191 for the generated instruction. One nice thing about LLVM is that the name is
192 just a hint: if there are multiple additions in a single function, the first
193 will be named "addtmp" and the second will be "autorenamed" by adding a suffix,
194 giving it a name like "addtmp42". Local value names for instructions are purely
195 optional, but it makes it much easier to read the IR dumps.</p>
197 <p><a href="../LangRef.html#instref">LLVM instructions</a> are constrained to
198 have very strict type properties: for example, the Left and Right operators of
199 an <a href="../LangRef.html#i_add">add instruction</a> have to have the same
200 type, and that the result of the add matches the operands. Because all values
201 in Kaleidoscope are doubles, this makes for very simple code for add, sub and
204 <p>On the other hand, LLVM specifies that the <a
205 href="../LangRef.html#i_fcmp">fcmp instruction</a> always returns an 'i1' value
206 (a one bit integer). However, Kaleidoscope wants the value to be a 0.0 or 1.0
207 value. In order to get these semantics, we combine the fcmp instruction with
208 a <a href="../LangRef.html#i_uitofp">uitofp instruction</a>. This instruction
209 converts its input integer into a floating point value by treating the input
210 as an unsigned value. In contrast, if we used the <a
211 href="../LangRef.html#i_sitofp">sitofp instruction</a>, the Kaleidoscope '<'
212 operator would return 0.0 and -1.0, depending on the input value.</p>
214 <div class="doc_code">
216 Value *CallExprAST::Codegen() {
217 // Look up the name in the global module table.
218 Function *CalleeF = TheModule->getFunction(Callee);
220 return ErrorV("Unknown function referenced");
222 // If argument mismatch error.
223 if (CalleeF->arg_size() != Args.size())
224 return ErrorV("Incorrect # arguments passed");
226 std::vector<Value*> ArgsV;
227 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
228 ArgsV.push_back(Args[i]->Codegen());
229 if (ArgsV.back() == 0) return 0;
232 return Builder.CreateCall(CalleeF, ArgsV.begin(), ArgsV.end(), "calltmp");
237 <p>Code generation for function calls is quite straight-forward with LLVM. The
238 code above first looks the name of the function up in the LLVM Module's symbol
239 table. Recall that the LLVM Module is the container that holds all of the
240 functions we are JIT'ing. By giving each function the same name as what the
241 user specifies, we can use the LLVM symbol table to resolve function names for
244 <p>Once we have the function to call, we recursively codegen each argument that
245 is to be passed in, and create an LLVM <a href="../LangRef.html#i_call">call
246 instruction</a>. Note that LLVM uses the native C calling conventions by
247 default, allowing these calls to call into standard library functions like
248 "sin" and "cos" with no additional effort.</p>
250 <p>This wraps up our handling of the four basic expressions that we have so far
251 in Kaleidoscope. Feel free to go in and add some more. For example, by
252 browsing the <a href="../LangRef.html">LLVM language reference</a> you'll find
253 several other interesting instructions that are really easy to plug into our
258 <!-- *********************************************************************** -->
259 <div class="doc_section"><a name="funcs">Function Code Generation</a></div>
260 <!-- *********************************************************************** -->
262 <div class="doc_text">
264 <p>Code generation for prototypes and functions has to handle a number of
265 details, which make their code less beautiful and elegant than expression code
266 generation, but they illustrate some important points. First, lets talk about
267 code generation for prototypes: this is used both for function bodies as well
268 as external function declarations. The code starts with:</p>
270 <div class="doc_code">
272 Function *PrototypeAST::Codegen() {
273 // Make the function type: double(double,double) etc.
274 std::vector<const Type*> Doubles(Args.size(), Type::DoubleTy);
275 FunctionType *FT = FunctionType::get(Type::DoubleTy, Doubles, false);
277 Function *F = new Function(FT, Function::ExternalLinkage, Name, TheModule);
281 <p>This code packs a lot of power into a few lines. The first step is to create
282 the <tt>FunctionType</tt> that should be used for a given Prototype. Since all
283 function arguments in Kaleidoscope are of type double, the first line creates
284 a vector of "N" LLVM Double types. It then uses the <tt>FunctionType::get</tt>
285 method to create a function type that takes "N" doubles as arguments, returns
286 one double as a result, and that is not vararg (the false parameter indicates
287 this). Note that Types in LLVM are uniqued just like Constants are, so you
288 don't "new" a type, you "get" it.</p>
290 <p>The final line above actually creates the function that the prototype will
291 correspond to. This indicates which type, linkage, and name to use, and which
292 module to insert into. "<a href="LangRef.html#linkage">external linkage</a>"
293 means that the function may be defined outside the current module and/or that it
294 is callable by functions outside the module. The Name passed in is the name the
295 user specified: since "<tt>TheModule</tt>" is specified, this name is registered
296 in "<tt>TheModule</tt>"s symbol table, which is used by the function call code
299 <div class="doc_code">
301 // If F conflicted, there was already something named 'Name'. If it has a
302 // body, don't allow redefinition or reextern.
303 if (F->getName() != Name) {
304 // Delete the one we just made and get the existing one.
305 F->eraseFromParent();
306 F = TheModule->getFunction(Name);
310 <p>The Module symbol table works just like the Function symbol table when it
311 comes to name conflicts: if a new function is created with a name was previously
312 added to the symbol table, it will get implicitly renamed when added to the
313 Module. The code above exploits this fact to tell if there was a previous
314 definition of this function.</p>
316 <p>In Kaleidoscope, I choose to allow redefinitions of functions in two cases:
317 first, we want to allow 'extern'ing a function more than once, so long as the
318 prototypes for the externs match (since all arguments have the same type, we
319 just have to check that the number of arguments match). Second, we want to
320 allow 'extern'ing a function and then definining a body for it. This is useful
321 when defining mutually recursive functions.</p>
323 <p>In order to implement this, the code above first checks to see if there is
324 a collision on the name of the function. If so, it deletes the function we just
325 created (by calling <tt>eraseFromParent</tt>) and then calling
326 <tt>getFunction</tt> to get the existing function with the specified name. Note
327 that many APIs in LLVM have "erase" forms and "remove" forms. The "remove" form
328 unlinks the object from its parent (e.g. a Function from a Module) and returns
329 it. The "erase" form unlinks the object and then deletes it.</p>
331 <div class="doc_code">
333 // If F already has a body, reject this.
334 if (!F->empty()) {
335 ErrorF("redefinition of function");
339 // If F took a different number of args, reject.
340 if (F->arg_size() != Args.size()) {
341 ErrorF("redefinition of function with different # args");
348 <p>In order to verify the logic above, we first check to see if the preexisting
349 function is "empty". In this case, empty means that it has no basic blocks in
350 it, which means it has no body. If it has no body, this means its a forward
351 declaration. Since we don't allow anything after a full definition of the
352 function, the code rejects this case. If the previous reference to a function
353 was an 'extern', we simply verify that the number of arguments for that
354 definition and this one match up. If not, we emit an error.</p>
356 <div class="doc_code">
358 // Set names for all arguments.
360 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
362 AI->setName(Args[Idx]);
364 // Add arguments to variable symbol table.
365 NamedValues[Args[Idx]] = AI;
372 <p>The last bit of code for prototypes loops over all of the arguments in the
373 function, setting the name of the LLVM Argument objects to match and registering
374 the arguments in the <tt>NamedValues</tt> map for future use by the
375 <tt>VariableExprAST</tt> AST node. Once this is set up, it returns the Function
376 object to the caller. Note that we don't check for conflicting
377 argument names here (e.g. "extern foo(a b a)"). Doing so would be very
378 straight-forward.</p>
380 <div class="doc_code">
382 Function *FunctionAST::Codegen() {
385 Function *TheFunction = Proto->Codegen();
386 if (TheFunction == 0)
391 <p>Code generation for function definitions starts out simply enough: first we
392 codegen the prototype and verify that it is ok. We also clear out the
393 <tt>NamedValues</tt> map to make sure that there isn't anything in it from the
394 last function we compiled.</p>
396 <div class="doc_code">
398 // Create a new basic block to start insertion into.
399 BasicBlock *BB = new BasicBlock("entry", TheFunction);
400 Builder.SetInsertPoint(BB);
402 if (Value *RetVal = Body->Codegen()) {
406 <p>Now we get to the point where the <tt>Builder</tt> is set up. The first
407 line creates a new <a href="http://en.wikipedia.org/wiki/Basic_block">basic
408 block</a> (named "entry"), which is inserted into <tt>TheFunction</tt>. The
409 second line then tells the builder that new instructions should be inserted into
410 the end of the new basic block. Basic blocks in LLVM are an important part
411 of functions that define the <a
412 href="http://en.wikipedia.org/wiki/Control_flow_graph">Control Flow Graph</a>.
413 Since we don't have any control flow, our functions will only contain one
414 block so far. We'll fix this in a future installment :).</p>
416 <div class="doc_code">
418 if (Value *RetVal = Body->Codegen()) {
419 // Finish off the function.
420 Builder.CreateRet(RetVal);
422 // Validate the generated code, checking for consistency.
423 verifyFunction(*TheFunction);
429 <p>Once the insertion point is set up, we call the <tt>CodeGen()</tt> method for
430 the root expression of the function. If no error happens, this emits code to
431 compute the expression into the entry block and returns the value that was
432 computed. Assuming no error, we then create an LLVM <a
433 href="../LangRef.html#i_ret">ret instruction</a>, which completes the function.
434 Once the function is built, we call the <tt>verifyFunction</tt> function, which
435 is provided by LLVM. This function does a variety of consistency checks on the
436 generated code, to determine if our compiler is doing everything right. Using
437 this is important: it can catch a lot of bugs. Once the function is finished
438 and validated, we return it.</p>
440 <div class="doc_code">
442 // Error reading body, remove function.
443 TheFunction->eraseFromParent();
449 <p>The only piece left here is handling of the error case. For simplicity, we
450 simply handle this by deleting the function we produced with the
451 <tt>eraseFromParent</tt> method. This allows the user to redefine a function
452 that they incorrectly typed in before: if we didn't delete it, it would live in
453 the symbol table, with a body, preventing future redefinition.</p>
455 <p>This code does have a bug though. Since the <tt>PrototypeAST::Codegen</tt>
456 can return a previously defined forward declaration, this can actually delete
457 a forward declaration. There are a number of ways to fix this bug, see what you
458 can come up with! Here is a testcase:</p>
460 <div class="doc_code">
462 extern foo(a b); # ok, defines foo.
463 def foo(a b) c; # error, 'c' is invalid.
464 def bar() foo(1, 2); # error, unknown function "foo"
470 <!-- *********************************************************************** -->
471 <div class="doc_section"><a name="driver">Driver Changes and
472 Closing Thoughts</a></div>
473 <!-- *********************************************************************** -->
475 <div class="doc_text">
478 For now, code generation to LLVM doesn't really get us much, except that we can
479 look at the pretty IR calls. The sample code inserts calls to Codegen into the
480 "<tt>HandleDefinition</tt>", "<tt>HandleExtern</tt>" etc functions, and then
481 dumps out the LLVM IR. This gives a nice way to look at the LLVM IR for simple
482 functions. For example:
485 <div class="doc_code">
488 ready> Read top-level expression:
489 define double @""() {
491 %addtmp = add double 4.000000e+00, 5.000000e+00
497 <p>Note how the parser turns the top-level expression into anonymous functions
498 for us. This will be handy when we add JIT support in the next chapter. Also
499 note that the code is very literally transcribed, no optimizations are being
500 performed. We will add optimizations explicitly in the next chapter.</p>
502 <div class="doc_code">
504 ready> <b>def foo(a b) a*a + 2*a*b + b*b;</b>
505 ready> Read function definition:
506 define double @foo(double %a, double %b) {
508 %multmp = mul double %a, %a
509 %multmp1 = mul double 2.000000e+00, %a
510 %multmp2 = mul double %multmp1, %b
511 %addtmp = add double %multmp, %multmp2
512 %multmp3 = mul double %b, %b
513 %addtmp4 = add double %addtmp, %multmp3
519 <p>This shows some simple arithmetic. Notice the striking similarity to the
520 LLVM builder calls that we use to create the instructions.</p>
522 <div class="doc_code">
524 ready> <b>def bar(a) foo(a, 4.0) + bar(31337);</b>
525 ready> Read function definition:
526 define double @bar(double %a) {
528 %calltmp = call double @foo( double %a, double 4.000000e+00 )
529 %calltmp1 = call double @bar( double 3.133700e+04 )
530 %addtmp = add double %calltmp, %calltmp1
536 <p>This shows some function calls. Note that the runtime of this function might
537 be fairly high. In the future we'll add conditional control flow to make
538 recursion actually be useful :).</p>
540 <div class="doc_code">
542 ready> <b>extern cos(x);</b>
543 ready> Read extern:
544 declare double @cos(double)
546 ready> <b>cos(1.234);</b>
547 ready> Read top-level expression:
548 define double @""() {
550 %calltmp = call double @cos( double 1.234000e+00 )
556 <p>This shows an extern for the libm "cos" function, and a call to it.</p>
559 <div class="doc_code">
562 ; ModuleID = 'my cool jit'
564 define double @""() {
566 %addtmp = add double 4.000000e+00, 5.000000e+00
570 define double @foo(double %a, double %b) {
572 %multmp = mul double %a, %a
573 %multmp1 = mul double 2.000000e+00, %a
574 %multmp2 = mul double %multmp1, %b
575 %addtmp = add double %multmp, %multmp2
576 %multmp3 = mul double %b, %b
577 %addtmp4 = add double %addtmp, %multmp3
581 define double @bar(double %a) {
583 %calltmp = call double @foo( double %a, double 4.000000e+00 )
584 %calltmp1 = call double @bar( double 3.133700e+04 )
585 %addtmp = add double %calltmp, %calltmp1
589 declare double @cos(double)
591 define double @""() {
593 %calltmp = call double @cos( double 1.234000e+00 )
599 <p>When you quit the current demo, it dumps out the IR for the entire module
600 generated. Here you can see the big picture with all the functions referencing
603 <p>This wraps up this chapter of the Kaleidoscope tutorial. Up next we'll
604 describe how to <a href="LangImpl4.html">add JIT codegen and optimizer
605 support</a> to this so we can actually start running code!</p>
610 <!-- *********************************************************************** -->
611 <div class="doc_section"><a name="code">Full Code Listing</a></div>
612 <!-- *********************************************************************** -->
614 <div class="doc_text">
617 Here is the complete code listing for our running example, enhanced with the
618 LLVM code generator. Because this uses the LLVM libraries, we need to link
619 them in. To do this, we use the <a
620 href="http://llvm.org/cmds/llvm-config.html">llvm-config</a> tool to inform
621 our makefile/command line about which options to use:</p>
623 <div class="doc_code">
626 g++ -g toy.cpp `llvm-config --cppflags --ldflags --libs core` -o toy
632 <p>Here is the code:</p>
634 <div class="doc_code">
637 // See example below.
639 #include "llvm/DerivedTypes.h"
640 #include "llvm/Module.h"
641 #include "llvm/Analysis/Verifier.h"
642 #include "llvm/Support/LLVMBuilder.h"
643 #include <cstdio>
644 #include <string>
646 #include <vector>
647 using namespace llvm;
649 //===----------------------------------------------------------------------===//
651 //===----------------------------------------------------------------------===//
653 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
654 // of these for known things.
659 tok_def = -2, tok_extern = -3,
662 tok_identifier = -4, tok_number = -5,
665 static std::string IdentifierStr; // Filled in if tok_identifier
666 static double NumVal; // Filled in if tok_number
668 /// gettok - Return the next token from standard input.
669 static int gettok() {
670 static int LastChar = ' ';
672 // Skip any whitespace.
673 while (isspace(LastChar))
674 LastChar = getchar();
676 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
677 IdentifierStr = LastChar;
678 while (isalnum((LastChar = getchar())))
679 IdentifierStr += LastChar;
681 if (IdentifierStr == "def") return tok_def;
682 if (IdentifierStr == "extern") return tok_extern;
683 return tok_identifier;
686 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+
690 LastChar = getchar();
691 } while (isdigit(LastChar) || LastChar == '.');
693 NumVal = strtod(NumStr.c_str(), 0);
697 if (LastChar == '#') {
698 // Comment until end of line.
699 do LastChar = getchar();
700 while (LastChar != EOF && LastChar != '\n' & LastChar != '\r');
706 // Check for end of file. Don't eat the EOF.
710 // Otherwise, just return the character as its ascii value.
711 int ThisChar = LastChar;
712 LastChar = getchar();
716 //===----------------------------------------------------------------------===//
717 // Abstract Syntax Tree (aka Parse Tree)
718 //===----------------------------------------------------------------------===//
720 /// ExprAST - Base class for all expression nodes.
723 virtual ~ExprAST() {}
724 virtual Value *Codegen() = 0;
727 /// NumberExprAST - Expression class for numeric literals like "1.0".
728 class NumberExprAST : public ExprAST {
731 explicit NumberExprAST(double val) : Val(val) {}
732 virtual Value *Codegen();
735 /// VariableExprAST - Expression class for referencing a variable, like "a".
736 class VariableExprAST : public ExprAST {
739 explicit VariableExprAST(const std::string &name) : Name(name) {}
740 virtual Value *Codegen();
743 /// BinaryExprAST - Expression class for a binary operator.
744 class BinaryExprAST : public ExprAST {
748 BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs)
749 : Op(op), LHS(lhs), RHS(rhs) {}
750 virtual Value *Codegen();
753 /// CallExprAST - Expression class for function calls.
754 class CallExprAST : public ExprAST {
756 std::vector<ExprAST*> Args;
758 CallExprAST(const std::string &callee, std::vector<ExprAST*> &args)
759 : Callee(callee), Args(args) {}
760 virtual Value *Codegen();
763 /// PrototypeAST - This class represents the "prototype" for a function,
764 /// which captures its argument names as well as if it is an operator.
767 std::vector<std::string> Args;
769 PrototypeAST(const std::string &name, const std::vector<std::string> &args)
770 : Name(name), Args(args) {}
775 /// FunctionAST - This class represents a function definition itself.
780 FunctionAST(PrototypeAST *proto, ExprAST *body)
781 : Proto(proto), Body(body) {}
786 //===----------------------------------------------------------------------===//
788 //===----------------------------------------------------------------------===//
790 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
791 /// token the parser it looking at. getNextToken reads another token from the
792 /// lexer and updates CurTok with its results.
794 static int getNextToken() {
795 return CurTok = gettok();
798 /// BinopPrecedence - This holds the precedence for each binary operator that is
800 static std::map<char, int> BinopPrecedence;
802 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
803 static int GetTokPrecedence() {
804 if (!isascii(CurTok))
807 // Make sure it's a declared binop.
808 int TokPrec = BinopPrecedence[CurTok];
809 if (TokPrec <= 0) return -1;
813 /// Error* - These are little helper functions for error handling.
814 ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;}
815 PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; }
816 FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; }
818 static ExprAST *ParseExpression();
822 /// ::= identifer '(' expression* ')'
823 static ExprAST *ParseIdentifierExpr() {
824 std::string IdName = IdentifierStr;
826 getNextToken(); // eat identifer.
828 if (CurTok != '(') // Simple variable ref.
829 return new VariableExprAST(IdName);
832 getNextToken(); // eat (
833 std::vector<ExprAST*> Args;
835 ExprAST *Arg = ParseExpression();
839 if (CurTok == ')') break;
842 return Error("Expected ')'");
849 return new CallExprAST(IdName, Args);
852 /// numberexpr ::= number
853 static ExprAST *ParseNumberExpr() {
854 ExprAST *Result = new NumberExprAST(NumVal);
855 getNextToken(); // consume the number
859 /// parenexpr ::= '(' expression ')'
860 static ExprAST *ParseParenExpr() {
861 getNextToken(); // eat (.
862 ExprAST *V = ParseExpression();
866 return Error("expected ')'");
867 getNextToken(); // eat ).
872 /// ::= identifierexpr
875 static ExprAST *ParsePrimary() {
877 default: return Error("unknown token when expecting an expression");
878 case tok_identifier: return ParseIdentifierExpr();
879 case tok_number: return ParseNumberExpr();
880 case '(': return ParseParenExpr();
885 /// ::= ('+' primary)*
886 static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) {
887 // If this is a binop, find its precedence.
889 int TokPrec = GetTokPrecedence();
891 // If this is a binop that binds at least as tightly as the current binop,
892 // consume it, otherwise we are done.
893 if (TokPrec < ExprPrec)
896 // Okay, we know this is a binop.
898 getNextToken(); // eat binop
900 // Parse the primary expression after the binary operator.
901 ExprAST *RHS = ParsePrimary();
904 // If BinOp binds less tightly with RHS than the operator after RHS, let
905 // the pending operator take RHS as its LHS.
906 int NextPrec = GetTokPrecedence();
907 if (TokPrec < NextPrec) {
908 RHS = ParseBinOpRHS(TokPrec+1, RHS);
909 if (RHS == 0) return 0;
913 LHS = new BinaryExprAST(BinOp, LHS, RHS);
918 /// ::= primary binoprhs
920 static ExprAST *ParseExpression() {
921 ExprAST *LHS = ParsePrimary();
924 return ParseBinOpRHS(0, LHS);
928 /// ::= id '(' id* ')'
929 static PrototypeAST *ParsePrototype() {
930 if (CurTok != tok_identifier)
931 return ErrorP("Expected function name in prototype");
933 std::string FnName = IdentifierStr;
937 return ErrorP("Expected '(' in prototype");
939 std::vector<std::string> ArgNames;
940 while (getNextToken() == tok_identifier)
941 ArgNames.push_back(IdentifierStr);
943 return ErrorP("Expected ')' in prototype");
946 getNextToken(); // eat ')'.
948 return new PrototypeAST(FnName, ArgNames);
951 /// definition ::= 'def' prototype expression
952 static FunctionAST *ParseDefinition() {
953 getNextToken(); // eat def.
954 PrototypeAST *Proto = ParsePrototype();
955 if (Proto == 0) return 0;
957 if (ExprAST *E = ParseExpression())
958 return new FunctionAST(Proto, E);
962 /// toplevelexpr ::= expression
963 static FunctionAST *ParseTopLevelExpr() {
964 if (ExprAST *E = ParseExpression()) {
965 // Make an anonymous proto.
966 PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>());
967 return new FunctionAST(Proto, E);
972 /// external ::= 'extern' prototype
973 static PrototypeAST *ParseExtern() {
974 getNextToken(); // eat extern.
975 return ParsePrototype();
978 //===----------------------------------------------------------------------===//
980 //===----------------------------------------------------------------------===//
982 static Module *TheModule;
983 static LLVMBuilder Builder;
984 static std::map<std::string, Value*> NamedValues;
986 Value *ErrorV(const char *Str) { Error(Str); return 0; }
988 Value *NumberExprAST::Codegen() {
989 return ConstantFP::get(Type::DoubleTy, APFloat(Val));
992 Value *VariableExprAST::Codegen() {
993 // Look this variable up in the function.
994 Value *V = NamedValues[Name];
995 return V ? V : ErrorV("Unknown variable name");
998 Value *BinaryExprAST::Codegen() {
999 Value *L = LHS->Codegen();
1000 Value *R = RHS->Codegen();
1001 if (L == 0 || R == 0) return 0;
1004 case '+': return Builder.CreateAdd(L, R, "addtmp");
1005 case '-': return Builder.CreateSub(L, R, "subtmp");
1006 case '*': return Builder.CreateMul(L, R, "multmp");
1008 L = Builder.CreateFCmpULT(L, R, "multmp");
1009 // Convert bool 0/1 to double 0.0 or 1.0
1010 return Builder.CreateUIToFP(L, Type::DoubleTy, "booltmp");
1011 default: return ErrorV("invalid binary operator");
1015 Value *CallExprAST::Codegen() {
1016 // Look up the name in the global module table.
1017 Function *CalleeF = TheModule->getFunction(Callee);
1019 return ErrorV("Unknown function referenced");
1021 // If argument mismatch error.
1022 if (CalleeF->arg_size() != Args.size())
1023 return ErrorV("Incorrect # arguments passed");
1025 std::vector<Value*> ArgsV;
1026 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
1027 ArgsV.push_back(Args[i]->Codegen());
1028 if (ArgsV.back() == 0) return 0;
1031 return Builder.CreateCall(CalleeF, ArgsV.begin(), ArgsV.end(), "calltmp");
1034 Function *PrototypeAST::Codegen() {
1035 // Make the function type: double(double,double) etc.
1036 std::vector<const Type*> Doubles(Args.size(), Type::DoubleTy);
1037 FunctionType *FT = FunctionType::get(Type::DoubleTy, Doubles, false);
1039 Function *F = new Function(FT, Function::ExternalLinkage, Name, TheModule);
1041 // If F conflicted, there was already something named 'Name'. If it has a
1042 // body, don't allow redefinition or reextern.
1043 if (F->getName() != Name) {
1044 // Delete the one we just made and get the existing one.
1045 F->eraseFromParent();
1046 F = TheModule->getFunction(Name);
1048 // If F already has a body, reject this.
1049 if (!F->empty()) {
1050 ErrorF("redefinition of function");
1054 // If F took a different number of args, reject.
1055 if (F->arg_size() != Args.size()) {
1056 ErrorF("redefinition of function with different # args");
1061 // Set names for all arguments.
1063 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
1065 AI->setName(Args[Idx]);
1067 // Add arguments to variable symbol table.
1068 NamedValues[Args[Idx]] = AI;
1074 Function *FunctionAST::Codegen() {
1075 NamedValues.clear();
1077 Function *TheFunction = Proto->Codegen();
1078 if (TheFunction == 0)
1081 // Create a new basic block to start insertion into.
1082 BasicBlock *BB = new BasicBlock("entry", TheFunction);
1083 Builder.SetInsertPoint(BB);
1085 if (Value *RetVal = Body->Codegen()) {
1086 // Finish off the function.
1087 Builder.CreateRet(RetVal);
1089 // Validate the generated code, checking for consistency.
1090 verifyFunction(*TheFunction);
1094 // Error reading body, remove function.
1095 TheFunction->eraseFromParent();
1099 //===----------------------------------------------------------------------===//
1100 // Top-Level parsing and JIT Driver
1101 //===----------------------------------------------------------------------===//
1103 static void HandleDefinition() {
1104 if (FunctionAST *F = ParseDefinition()) {
1105 if (Function *LF = F->Codegen()) {
1106 fprintf(stderr, "Read function definition:");
1110 // Skip token for error recovery.
1115 static void HandleExtern() {
1116 if (PrototypeAST *P = ParseExtern()) {
1117 if (Function *F = P->Codegen()) {
1118 fprintf(stderr, "Read extern: ");
1122 // Skip token for error recovery.
1127 static void HandleTopLevelExpression() {
1128 // Evaluate a top level expression into an anonymous function.
1129 if (FunctionAST *F = ParseTopLevelExpr()) {
1130 if (Function *LF = F->Codegen()) {
1131 fprintf(stderr, "Read top-level expression:");
1135 // Skip token for error recovery.
1140 /// top ::= definition | external | expression | ';'
1141 static void MainLoop() {
1143 fprintf(stderr, "ready> ");
1145 case tok_eof: return;
1146 case ';': getNextToken(); break; // ignore top level semicolons.
1147 case tok_def: HandleDefinition(); break;
1148 case tok_extern: HandleExtern(); break;
1149 default: HandleTopLevelExpression(); break;
1156 //===----------------------------------------------------------------------===//
1157 // "Library" functions that can be "extern'd" from user code.
1158 //===----------------------------------------------------------------------===//
1160 /// putchard - putchar that takes a double and returns 0.
1162 double putchard(double X) {
1167 //===----------------------------------------------------------------------===//
1168 // Main driver code.
1169 //===----------------------------------------------------------------------===//
1172 TheModule = new Module("my cool jit");
1174 // Install standard binary operators.
1175 // 1 is lowest precedence.
1176 BinopPrecedence['<'] = 10;
1177 BinopPrecedence['+'] = 20;
1178 BinopPrecedence['-'] = 20;
1179 BinopPrecedence['*'] = 40; // highest.
1181 // Prime the first token.
1182 fprintf(stderr, "ready> ");
1186 TheModule->dump();
1193 <!-- *********************************************************************** -->
1196 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1197 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1198 <a href="http://validator.w3.org/check/referer"><img
1199 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
1201 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1202 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
1203 Last modified: $Date: 2007-10-17 11:05:13 -0700 (Wed, 17 Oct 2007) $