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>
17 <li><a href="index.html">Up to Tutorial Index</a></li>
20 <li><a href="#intro">Chapter 3 Introduction</a></li>
21 <li><a href="#basics">Code Generation Setup</a></li>
22 <li><a href="#exprs">Expression Code Generation</a></li>
23 <li><a href="#funcs">Function Code Generation</a></li>
24 <li><a href="#driver">Driver Changes and Closing Thoughts</a></li>
25 <li><a href="#code">Full Code Listing</a></li>
28 <li><a href="LangImpl4.html">Chapter 4</a>: Adding JIT and Optimizer
32 <div class="doc_author">
33 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
36 <!-- *********************************************************************** -->
37 <div class="doc_section"><a name="intro">Chapter 3 Introduction</a></div>
38 <!-- *********************************************************************** -->
40 <div class="doc_text">
42 <p>Welcome to Chapter 3 of the "<a href="index.html">Implementing a language
43 with LLVM</a>" tutorial. This chapter shows you how to transform the <a
44 href="LangImpl2.html">Abstract Syntax Tree</a>, built in Chapter 2, into LLVM IR.
45 This will teach you a little bit about how LLVM does things, as well as
46 demonstrate how easy it is to use. It's much more work to build a lexer and
47 parser than it is to generate LLVM IR code. :)
52 <!-- *********************************************************************** -->
53 <div class="doc_section"><a name="basics">Code Generation Setup</a></div>
54 <!-- *********************************************************************** -->
56 <div class="doc_text">
59 In order to generate LLVM IR, we want some simple setup to get started. First,
60 we define virtual codegen methods in each AST class:</p>
62 <div class="doc_code">
64 /// ExprAST - Base class for all expression nodes.
68 <b>virtual Value *Codegen() = 0;</b>
71 /// NumberExprAST - Expression class for numeric literals like "1.0".
72 class NumberExprAST : public ExprAST {
75 explicit NumberExprAST(double val) : Val(val) {}
76 <b>virtual Value *Codegen();</b>
82 <p>The Codegen() method says to emit IR for that AST node along with all the things it
83 depends on, and they all return an LLVM Value object.
84 "Value" is the class used to represent a "<a
85 href="http://en.wikipedia.org/wiki/Static_single_assignment_form">Static Single
86 Assignment (SSA)</a> register" or "SSA value" in LLVM. The most distinct aspect
87 of SSA values is that their value is computed as the related instruction
88 executes, and it does not get a new value until (and if) the instruction
89 re-executes. In other words, there is no way to "change" an SSA value. For
90 more information, please read up on <a
91 href="http://en.wikipedia.org/wiki/Static_single_assignment_form">Static Single
92 Assignment</a> - the concepts are really quite natural once you grok them.</p>
94 <p>Note that instead of adding virtual methods to the ExprAST class hierarchy,
95 it could also make sense to use a visitor pattern or some other way to model
96 this. Again, this tutorial won't dwell on good software engineering practices:
97 for our purposes, adding a virtual method is simplest.</p>
100 second thing we want is an "Error" method like we used for the parser, which will
101 be used to report errors found during code generation (for example, use of an
102 undeclared parameter):</p>
104 <div class="doc_code">
106 Value *ErrorV(const char *Str) { Error(Str); return 0; }
108 static Module *TheModule;
109 static LLVMBuilder Builder;
110 static std::map<std::string, Value*> NamedValues;
114 <p>The static variables will be used during code generation. <tt>TheModule</tt>
115 is the LLVM construct that contains all of the functions and global variables in
116 a chunk of code. In many ways, it is the top-level structure that the LLVM IR
117 uses to contain code.</p>
119 <p>The <tt>Builder</tt> object is a helper object that makes it easy to generate
120 LLVM instructions. Instances of the <a
121 href="http://llvm.org/doxygen/LLVMBuilder_8h-source.html"><tt>LLVMBuilder</tt>
122 class</a> keep track of the current place to
123 insert instructions and has methods to create new instructions.</p>
125 <p>The <tt>NamedValues</tt> map keeps track of which values are defined in the
126 current scope and what their LLVM representation is (in other words, it is a
127 symbol table for the code). In this form of
128 Kaleidoscope, the only things that can be referenced are function parameters.
129 As such, function parameters will be in this map when generating code for their
133 With these basics in place, we can start talking about how to generate code for
134 each expression. Note that this assumes that the <tt>Builder</tt> has been set
135 up to generate code <em>into</em> something. For now, we'll assume that this
136 has already been done, and we'll just use it to emit code.
141 <!-- *********************************************************************** -->
142 <div class="doc_section"><a name="exprs">Expression Code Generation</a></div>
143 <!-- *********************************************************************** -->
145 <div class="doc_text">
147 <p>Generating LLVM code for expression nodes is very straightforward: less
148 than 45 lines of commented code for all four of our expression nodes. First,
149 we'll do numeric literals:</p>
151 <div class="doc_code">
153 Value *NumberExprAST::Codegen() {
154 return ConstantFP::get(Type::DoubleTy, APFloat(Val));
159 <p>In the LLVM IR, numeric constants are represented with the
160 <tt>ConstantFP</tt> class, which holds the numeric value in an <tt>APFloat</tt>
161 internally (<tt>APFloat</tt> has the capability of holding floating point
162 constants of <em>A</em>rbitrary <em>P</em>recision). This code basically just
163 creates and returns a <tt>ConstantFP</tt>. Note that in the LLVM IR
164 that constants are all uniqued together and shared. For this reason, the API
165 uses "the foo::get(..)" idiom instead of "new foo(..)" or "foo::create(..)".</p>
167 <div class="doc_code">
169 Value *VariableExprAST::Codegen() {
170 // Look this variable up in the function.
171 Value *V = NamedValues[Name];
172 return V ? V : ErrorV("Unknown variable name");
177 <p>References to variables are also quite simple using LLVM. In the simple version
178 of Kaleidoscope, we assume that the variable has already been emited somewhere
179 and its value is available. In practice, the only values that can be in the
180 <tt>NamedValues</tt> map are function arguments. This
181 code simply checks to see that the specified name is in the map (if not, an
182 unknown variable is being referenced) and returns the value for it. In future
183 chapters, we'll add support for <a href="LangImpl5.html#for">loop induction
184 variables</a> in the symbol table, and for <a
185 href="LangImpl7.html#localvars">local variables</a>.</p>
187 <div class="doc_code">
189 Value *BinaryExprAST::Codegen() {
190 Value *L = LHS->Codegen();
191 Value *R = RHS->Codegen();
192 if (L == 0 || R == 0) return 0;
195 case '+': return Builder.CreateAdd(L, R, "addtmp");
196 case '-': return Builder.CreateSub(L, R, "subtmp");
197 case '*': return Builder.CreateMul(L, R, "multmp");
199 L = Builder.CreateFCmpULT(L, R, "cmptmp");
200 // Convert bool 0/1 to double 0.0 or 1.0
201 return Builder.CreateUIToFP(L, Type::DoubleTy, "booltmp");
202 default: return ErrorV("invalid binary operator");
208 <p>Binary operators start to get more interesting. The basic idea here is that
209 we recursively emit code for the left-hand side of the expression, then the
210 right-hand side, then we compute the result of the binary expression. In this
211 code, we do a simple switch on the opcode to create the right LLVM instruction.
214 <p>In the example above, the LLVM builder class is starting to show its value.
215 LLVMBuilder knows where to insert the newly created instruction, all you have to
216 do is specify what instruction to create (e.g. with <tt>CreateAdd</tt>), which
217 operands to use (<tt>L</tt> and <tt>R</tt> here) and optionally provide a name
218 for the generated instruction. One nice thing about LLVM is that the name is
219 just a hint: if there are multiple additions in a single function, the first
220 will be named "addtmp" and the second will be "autorenamed" by adding a suffix,
221 giving it a name like "addtmp42". Local value names for instructions are purely
222 optional, but it makes it much easier to read the IR dumps.</p>
224 <p><a href="../LangRef.html#instref">LLVM instructions</a> are constrained by
225 strict rules: for example, the Left and Right operators of
226 an <a href="../LangRef.html#i_add">add instruction</a> must have the same
227 type, and the result type of the add must match the operand types. Because
228 all values in Kaleidoscope are doubles, this makes for very simple code for add,
231 <p>On the other hand, LLVM specifies that the <a
232 href="../LangRef.html#i_fcmp">fcmp instruction</a> always returns an 'i1' value
233 (a one bit integer). The problem with this is that Kaleidoscope wants the value to be a 0.0 or 1.0 value. In order to get these semantics, we combine the fcmp instruction with
234 a <a href="../LangRef.html#i_uitofp">uitofp instruction</a>. This instruction
235 converts its input integer into a floating point value by treating the input
236 as an unsigned value. In contrast, if we used the <a
237 href="../LangRef.html#i_sitofp">sitofp instruction</a>, the Kaleidoscope '<'
238 operator would return 0.0 and -1.0, depending on the input value.</p>
240 <div class="doc_code">
242 Value *CallExprAST::Codegen() {
243 // Look up the name in the global module table.
244 Function *CalleeF = TheModule->getFunction(Callee);
246 return ErrorV("Unknown function referenced");
248 // If argument mismatch error.
249 if (CalleeF->arg_size() != Args.size())
250 return ErrorV("Incorrect # arguments passed");
252 std::vector<Value*> ArgsV;
253 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
254 ArgsV.push_back(Args[i]->Codegen());
255 if (ArgsV.back() == 0) return 0;
258 return Builder.CreateCall(CalleeF, ArgsV.begin(), ArgsV.end(), "calltmp");
263 <p>Code generation for function calls is quite straightforward with LLVM. The
264 code above initially does a function name lookup in the LLVM Module's symbol
265 table. Recall that the LLVM Module is the container that holds all of the
266 functions we are JIT'ing. By giving each function the same name as what the
267 user specifies, we can use the LLVM symbol table to resolve function names for
270 <p>Once we have the function to call, we recursively codegen each argument that
271 is to be passed in, and create an LLVM <a href="../LangRef.html#i_call">call
272 instruction</a>. Note that LLVM uses the native C calling conventions by
273 default, allowing these calls to also call into standard library functions like
274 "sin" and "cos", with no additional effort.</p>
276 <p>This wraps up our handling of the four basic expressions that we have so far
277 in Kaleidoscope. Feel free to go in and add some more. For example, by
278 browsing the <a href="../LangRef.html">LLVM language reference</a> you'll find
279 several other interesting instructions that are really easy to plug into our
284 <!-- *********************************************************************** -->
285 <div class="doc_section"><a name="funcs">Function Code Generation</a></div>
286 <!-- *********************************************************************** -->
288 <div class="doc_text">
290 <p>Code generation for prototypes and functions must handle a number of
291 details, which make their code less beautiful than expression code
292 generation, but allows us to illustrate some important points. First, lets
293 talk about code generation for prototypes: they are used both for function
294 bodies and external function declarations. The code starts with:</p>
296 <div class="doc_code">
298 Function *PrototypeAST::Codegen() {
299 // Make the function type: double(double,double) etc.
300 std::vector<const Type*> Doubles(Args.size(), Type::DoubleTy);
301 FunctionType *FT = FunctionType::get(Type::DoubleTy, Doubles, false);
303 Function *F = new Function(FT, Function::ExternalLinkage, Name, TheModule);
307 <p>This code packs a lot of power into a few lines. Note first that this
308 function returns a "Function*" instead of a "Value*". Because a "prototype"
309 really talks about the external interface for a function (not the value computed
310 by an expression), it makes sense for it to return the LLVM Function it
311 corresponds to when codegen'd.</p>
313 <p>The call to <tt>FunctionType::get</tt> creates
314 the <tt>FunctionType</tt> that should be used for a given Prototype. Since all
315 function arguments in Kaleidoscope are of type double, the first line creates
316 a vector of "N" LLVM double types. It then uses the <tt>FunctionType::get</tt>
317 method to create a function type that takes "N" doubles as arguments, returns
318 one double as a result, and that is not vararg (the false parameter indicates
319 this). Note that Types in LLVM are uniqued just like Constants are, so you
320 don't "new" a type, you "get" it.</p>
322 <p>The final line above actually creates the function that the prototype will
323 correspond to. This indicates the type, linkage and name to use, as well as which
324 module to insert into. "<a href="LangRef.html#linkage">external linkage</a>"
325 means that the function may be defined outside the current module and/or that it
326 is callable by functions outside the module. The Name passed in is the name the
327 user specified: since "<tt>TheModule</tt>" is specified, this name is registered
328 in "<tt>TheModule</tt>"s symbol table, which is used by the function call code
331 <div class="doc_code">
333 // If F conflicted, there was already something named 'Name'. If it has a
334 // body, don't allow redefinition or reextern.
335 if (F->getName() != Name) {
336 // Delete the one we just made and get the existing one.
337 F->eraseFromParent();
338 F = TheModule->getFunction(Name);
342 <p>The Module symbol table works just like the Function symbol table when it
343 comes to name conflicts: if a new function is created with a name was previously
344 added to the symbol table, it will get implicitly renamed when added to the
345 Module. The code above exploits this fact to determine if there was a previous
346 definition of this function.</p>
348 <p>In Kaleidoscope, I choose to allow redefinitions of functions in two cases:
349 first, we want to allow 'extern'ing a function more than once, as long as the
350 prototypes for the externs match (since all arguments have the same type, we
351 just have to check that the number of arguments match). Second, we want to
352 allow 'extern'ing a function and then definining a body for it. This is useful
353 when defining mutually recursive functions.</p>
355 <p>In order to implement this, the code above first checks to see if there is
356 a collision on the name of the function. If so, it deletes the function we just
357 created (by calling <tt>eraseFromParent</tt>) and then calling
358 <tt>getFunction</tt> to get the existing function with the specified name. Note
359 that many APIs in LLVM have "erase" forms and "remove" forms. The "remove" form
360 unlinks the object from its parent (e.g. a Function from a Module) and returns
361 it. The "erase" form unlinks the object and then deletes it.</p>
363 <div class="doc_code">
365 // If F already has a body, reject this.
366 if (!F->empty()) {
367 ErrorF("redefinition of function");
371 // If F took a different number of args, reject.
372 if (F->arg_size() != Args.size()) {
373 ErrorF("redefinition of function with different # args");
380 <p>In order to verify the logic above, we first check to see if the pre-existing
381 function is "empty". In this case, empty means that it has no basic blocks in
382 it, which means it has no body. If it has no body, it is a forward
383 declaration. Since we don't allow anything after a full definition of the
384 function, the code rejects this case. If the previous reference to a function
385 was an 'extern', we simply verify that the number of arguments for that
386 definition and this one match up. If not, we emit an error.</p>
388 <div class="doc_code">
390 // Set names for all arguments.
392 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
394 AI->setName(Args[Idx]);
396 // Add arguments to variable symbol table.
397 NamedValues[Args[Idx]] = AI;
404 <p>The last bit of code for prototypes loops over all of the arguments in the
405 function, setting the name of the LLVM Argument objects to match, and registering
406 the arguments in the <tt>NamedValues</tt> map for future use by the
407 <tt>VariableExprAST</tt> AST node. Once this is set up, it returns the Function
408 object to the caller. Note that we don't check for conflicting
409 argument names here (e.g. "extern foo(a b a)"). Doing so would be very
410 straight-forward with the mechanics we have already used above.</p>
412 <div class="doc_code">
414 Function *FunctionAST::Codegen() {
417 Function *TheFunction = Proto->Codegen();
418 if (TheFunction == 0)
423 <p>Code generation for function definitions starts out simply enough: we just
424 codegen the prototype (Proto) and verify that it is ok. We then clear out the
425 <tt>NamedValues</tt> map to make sure that there isn't anything in it from the
426 last function we compiled. Code generation of the prototype ensures that there
427 is an LLVM Function object that is ready to go for us.</p>
429 <div class="doc_code">
431 // Create a new basic block to start insertion into.
432 BasicBlock *BB = new BasicBlock("entry", TheFunction);
433 Builder.SetInsertPoint(BB);
435 if (Value *RetVal = Body->Codegen()) {
439 <p>Now we get to the point where the <tt>Builder</tt> is set up. The first
440 line creates a new <a href="http://en.wikipedia.org/wiki/Basic_block">basic
441 block</a> (named "entry"), which is inserted into <tt>TheFunction</tt>. The
442 second line then tells the builder that new instructions should be inserted into
443 the end of the new basic block. Basic blocks in LLVM are an important part
444 of functions that define the <a
445 href="http://en.wikipedia.org/wiki/Control_flow_graph">Control Flow Graph</a>.
446 Since we don't have any control flow, our functions will only contain one
447 block at this point. We'll fix this in <a href="LangImpl5.html">Chapter 5</a> :).</p>
449 <div class="doc_code">
451 if (Value *RetVal = Body->Codegen()) {
452 // Finish off the function.
453 Builder.CreateRet(RetVal);
455 // Validate the generated code, checking for consistency.
456 verifyFunction(*TheFunction);
462 <p>Once the insertion point is set up, we call the <tt>CodeGen()</tt> method for
463 the root expression of the function. If no error happens, this emits code to
464 compute the expression into the entry block and returns the value that was
465 computed. Assuming no error, we then create an LLVM <a
466 href="../LangRef.html#i_ret">ret instruction</a>, which completes the function.
467 Once the function is built, we call <tt>verifyFunction</tt>, which
468 is provided by LLVM. This function does a variety of consistency checks on the
469 generated code, to determine if our compiler is doing everything right. Using
470 this is important: it can catch a lot of bugs. Once the function is finished
471 and validated, we return it.</p>
473 <div class="doc_code">
475 // Error reading body, remove function.
476 TheFunction->eraseFromParent();
482 <p>The only piece left here is handling of the error case. For simplicity, we
483 handle this by merely deleting the function we produced with the
484 <tt>eraseFromParent</tt> method. This allows the user to redefine a function
485 that they incorrectly typed in before: if we didn't delete it, it would live in
486 the symbol table, with a body, preventing future redefinition.</p>
488 <p>This code does have a bug, though. Since the <tt>PrototypeAST::Codegen</tt>
489 can return a previously defined forward declaration, our code can actually delete
490 a forward declaration. There are a number of ways to fix this bug, see what you
491 can come up with! Here is a testcase:</p>
493 <div class="doc_code">
495 extern foo(a b); # ok, defines foo.
496 def foo(a b) c; # error, 'c' is invalid.
497 def bar() foo(1, 2); # error, unknown function "foo"
503 <!-- *********************************************************************** -->
504 <div class="doc_section"><a name="driver">Driver Changes and
505 Closing Thoughts</a></div>
506 <!-- *********************************************************************** -->
508 <div class="doc_text">
511 For now, code generation to LLVM doesn't really get us much, except that we can
512 look at the pretty IR calls. The sample code inserts calls to Codegen into the
513 "<tt>HandleDefinition</tt>", "<tt>HandleExtern</tt>" etc functions, and then
514 dumps out the LLVM IR. This gives a nice way to look at the LLVM IR for simple
515 functions. For example:
518 <div class="doc_code">
521 Read top-level expression:
522 define double @""() {
524 %addtmp = add double 4.000000e+00, 5.000000e+00
530 <p>Note how the parser turns the top-level expression into anonymous functions
531 for us. This will be handy when we add <a href="LangImpl4.html#jit">JIT
532 support</a> in the next chapter. Also note that the code is very literally
533 transcribed, no optimizations are being performed. We will
534 <a href="LangImpl4.html#trivialconstfold">add optimizations</a> explicitly in
535 the next chapter.</p>
537 <div class="doc_code">
539 ready> <b>def foo(a b) a*a + 2*a*b + b*b;</b>
540 Read function definition:
541 define double @foo(double %a, double %b) {
543 %multmp = mul double %a, %a
544 %multmp1 = mul double 2.000000e+00, %a
545 %multmp2 = mul double %multmp1, %b
546 %addtmp = add double %multmp, %multmp2
547 %multmp3 = mul double %b, %b
548 %addtmp4 = add double %addtmp, %multmp3
554 <p>This shows some simple arithmetic. Notice the striking similarity to the
555 LLVM builder calls that we use to create the instructions.</p>
557 <div class="doc_code">
559 ready> <b>def bar(a) foo(a, 4.0) + bar(31337);</b>
560 Read function definition:
561 define double @bar(double %a) {
563 %calltmp = call double @foo( double %a, double 4.000000e+00 )
564 %calltmp1 = call double @bar( double 3.133700e+04 )
565 %addtmp = add double %calltmp, %calltmp1
571 <p>This shows some function calls. Note that this function will take a long
572 time to execute if you call it. In the future we'll add conditional control
573 flow to actually make recursion useful :).</p>
575 <div class="doc_code">
577 ready> <b>extern cos(x);</b>
579 declare double @cos(double)
581 ready> <b>cos(1.234);</b>
582 Read top-level expression:
583 define double @""() {
585 %calltmp = call double @cos( double 1.234000e+00 )
591 <p>This shows an extern for the libm "cos" function, and a call to it.</p>
594 <div class="doc_code">
597 ; ModuleID = 'my cool jit'
599 define double @""() {
601 %addtmp = add double 4.000000e+00, 5.000000e+00
605 define double @foo(double %a, double %b) {
607 %multmp = mul double %a, %a
608 %multmp1 = mul double 2.000000e+00, %a
609 %multmp2 = mul double %multmp1, %b
610 %addtmp = add double %multmp, %multmp2
611 %multmp3 = mul double %b, %b
612 %addtmp4 = add double %addtmp, %multmp3
616 define double @bar(double %a) {
618 %calltmp = call double @foo( double %a, double 4.000000e+00 )
619 %calltmp1 = call double @bar( double 3.133700e+04 )
620 %addtmp = add double %calltmp, %calltmp1
624 declare double @cos(double)
626 define double @""() {
628 %calltmp = call double @cos( double 1.234000e+00 )
634 <p>When you quit the current demo, it dumps out the IR for the entire module
635 generated. Here you can see the big picture with all the functions referencing
638 <p>This wraps up the third chapter of the Kaleidoscope tutorial. Up next, we'll
639 describe how to <a href="LangImpl4.html">add JIT codegen and optimizer
640 support</a> to this so we can actually start running code!</p>
645 <!-- *********************************************************************** -->
646 <div class="doc_section"><a name="code">Full Code Listing</a></div>
647 <!-- *********************************************************************** -->
649 <div class="doc_text">
652 Here is the complete code listing for our running example, enhanced with the
653 LLVM code generator. Because this uses the LLVM libraries, we need to link
654 them in. To do this, we use the <a
655 href="http://llvm.org/cmds/llvm-config.html">llvm-config</a> tool to inform
656 our makefile/command line about which options to use:</p>
658 <div class="doc_code">
661 g++ -g -O3 toy.cpp `llvm-config --cppflags --ldflags --libs core` -o toy
667 <p>Here is the code:</p>
669 <div class="doc_code">
672 // See example below.
674 #include "llvm/DerivedTypes.h"
675 #include "llvm/Module.h"
676 #include "llvm/Analysis/Verifier.h"
677 #include "llvm/Support/LLVMBuilder.h"
678 #include <cstdio>
679 #include <string>
681 #include <vector>
682 using namespace llvm;
684 //===----------------------------------------------------------------------===//
686 //===----------------------------------------------------------------------===//
688 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
689 // of these for known things.
694 tok_def = -2, tok_extern = -3,
697 tok_identifier = -4, tok_number = -5,
700 static std::string IdentifierStr; // Filled in if tok_identifier
701 static double NumVal; // Filled in if tok_number
703 /// gettok - Return the next token from standard input.
704 static int gettok() {
705 static int LastChar = ' ';
707 // Skip any whitespace.
708 while (isspace(LastChar))
709 LastChar = getchar();
711 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
712 IdentifierStr = LastChar;
713 while (isalnum((LastChar = getchar())))
714 IdentifierStr += LastChar;
716 if (IdentifierStr == "def") return tok_def;
717 if (IdentifierStr == "extern") return tok_extern;
718 return tok_identifier;
721 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+
725 LastChar = getchar();
726 } while (isdigit(LastChar) || LastChar == '.');
728 NumVal = strtod(NumStr.c_str(), 0);
732 if (LastChar == '#') {
733 // Comment until end of line.
734 do LastChar = getchar();
735 while (LastChar != EOF && LastChar != '\n' & LastChar != '\r');
741 // Check for end of file. Don't eat the EOF.
745 // Otherwise, just return the character as its ascii value.
746 int ThisChar = LastChar;
747 LastChar = getchar();
751 //===----------------------------------------------------------------------===//
752 // Abstract Syntax Tree (aka Parse Tree)
753 //===----------------------------------------------------------------------===//
755 /// ExprAST - Base class for all expression nodes.
758 virtual ~ExprAST() {}
759 virtual Value *Codegen() = 0;
762 /// NumberExprAST - Expression class for numeric literals like "1.0".
763 class NumberExprAST : public ExprAST {
766 explicit NumberExprAST(double val) : Val(val) {}
767 virtual Value *Codegen();
770 /// VariableExprAST - Expression class for referencing a variable, like "a".
771 class VariableExprAST : public ExprAST {
774 explicit VariableExprAST(const std::string &name) : Name(name) {}
775 virtual Value *Codegen();
778 /// BinaryExprAST - Expression class for a binary operator.
779 class BinaryExprAST : public ExprAST {
783 BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs)
784 : Op(op), LHS(lhs), RHS(rhs) {}
785 virtual Value *Codegen();
788 /// CallExprAST - Expression class for function calls.
789 class CallExprAST : public ExprAST {
791 std::vector<ExprAST*> Args;
793 CallExprAST(const std::string &callee, std::vector<ExprAST*> &args)
794 : Callee(callee), Args(args) {}
795 virtual Value *Codegen();
798 /// PrototypeAST - This class represents the "prototype" for a function,
799 /// which captures its argument names as well as if it is an operator.
802 std::vector<std::string> Args;
804 PrototypeAST(const std::string &name, const std::vector<std::string> &args)
805 : Name(name), Args(args) {}
810 /// FunctionAST - This class represents a function definition itself.
815 FunctionAST(PrototypeAST *proto, ExprAST *body)
816 : Proto(proto), Body(body) {}
821 //===----------------------------------------------------------------------===//
823 //===----------------------------------------------------------------------===//
825 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
826 /// token the parser it looking at. getNextToken reads another token from the
827 /// lexer and updates CurTok with its results.
829 static int getNextToken() {
830 return CurTok = gettok();
833 /// BinopPrecedence - This holds the precedence for each binary operator that is
835 static std::map<char, int> BinopPrecedence;
837 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
838 static int GetTokPrecedence() {
839 if (!isascii(CurTok))
842 // Make sure it's a declared binop.
843 int TokPrec = BinopPrecedence[CurTok];
844 if (TokPrec <= 0) return -1;
848 /// Error* - These are little helper functions for error handling.
849 ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;}
850 PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; }
851 FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; }
853 static ExprAST *ParseExpression();
857 /// ::= identifier '(' expression* ')'
858 static ExprAST *ParseIdentifierExpr() {
859 std::string IdName = IdentifierStr;
861 getNextToken(); // eat identifier.
863 if (CurTok != '(') // Simple variable ref.
864 return new VariableExprAST(IdName);
867 getNextToken(); // eat (
868 std::vector<ExprAST*> Args;
871 ExprAST *Arg = ParseExpression();
875 if (CurTok == ')') break;
878 return Error("Expected ')'");
886 return new CallExprAST(IdName, Args);
889 /// numberexpr ::= number
890 static ExprAST *ParseNumberExpr() {
891 ExprAST *Result = new NumberExprAST(NumVal);
892 getNextToken(); // consume the number
896 /// parenexpr ::= '(' expression ')'
897 static ExprAST *ParseParenExpr() {
898 getNextToken(); // eat (.
899 ExprAST *V = ParseExpression();
903 return Error("expected ')'");
904 getNextToken(); // eat ).
909 /// ::= identifierexpr
912 static ExprAST *ParsePrimary() {
914 default: return Error("unknown token when expecting an expression");
915 case tok_identifier: return ParseIdentifierExpr();
916 case tok_number: return ParseNumberExpr();
917 case '(': return ParseParenExpr();
922 /// ::= ('+' primary)*
923 static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) {
924 // If this is a binop, find its precedence.
926 int TokPrec = GetTokPrecedence();
928 // If this is a binop that binds at least as tightly as the current binop,
929 // consume it, otherwise we are done.
930 if (TokPrec < ExprPrec)
933 // Okay, we know this is a binop.
935 getNextToken(); // eat binop
937 // Parse the primary expression after the binary operator.
938 ExprAST *RHS = ParsePrimary();
941 // If BinOp binds less tightly with RHS than the operator after RHS, let
942 // the pending operator take RHS as its LHS.
943 int NextPrec = GetTokPrecedence();
944 if (TokPrec < NextPrec) {
945 RHS = ParseBinOpRHS(TokPrec+1, RHS);
946 if (RHS == 0) return 0;
950 LHS = new BinaryExprAST(BinOp, LHS, RHS);
955 /// ::= primary binoprhs
957 static ExprAST *ParseExpression() {
958 ExprAST *LHS = ParsePrimary();
961 return ParseBinOpRHS(0, LHS);
965 /// ::= id '(' id* ')'
966 static PrototypeAST *ParsePrototype() {
967 if (CurTok != tok_identifier)
968 return ErrorP("Expected function name in prototype");
970 std::string FnName = IdentifierStr;
974 return ErrorP("Expected '(' in prototype");
976 std::vector<std::string> ArgNames;
977 while (getNextToken() == tok_identifier)
978 ArgNames.push_back(IdentifierStr);
980 return ErrorP("Expected ')' in prototype");
983 getNextToken(); // eat ')'.
985 return new PrototypeAST(FnName, ArgNames);
988 /// definition ::= 'def' prototype expression
989 static FunctionAST *ParseDefinition() {
990 getNextToken(); // eat def.
991 PrototypeAST *Proto = ParsePrototype();
992 if (Proto == 0) return 0;
994 if (ExprAST *E = ParseExpression())
995 return new FunctionAST(Proto, E);
999 /// toplevelexpr ::= expression
1000 static FunctionAST *ParseTopLevelExpr() {
1001 if (ExprAST *E = ParseExpression()) {
1002 // Make an anonymous proto.
1003 PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>());
1004 return new FunctionAST(Proto, E);
1009 /// external ::= 'extern' prototype
1010 static PrototypeAST *ParseExtern() {
1011 getNextToken(); // eat extern.
1012 return ParsePrototype();
1015 //===----------------------------------------------------------------------===//
1017 //===----------------------------------------------------------------------===//
1019 static Module *TheModule;
1020 static LLVMBuilder Builder;
1021 static std::map<std::string, Value*> NamedValues;
1023 Value *ErrorV(const char *Str) { Error(Str); return 0; }
1025 Value *NumberExprAST::Codegen() {
1026 return ConstantFP::get(Type::DoubleTy, APFloat(Val));
1029 Value *VariableExprAST::Codegen() {
1030 // Look this variable up in the function.
1031 Value *V = NamedValues[Name];
1032 return V ? V : ErrorV("Unknown variable name");
1035 Value *BinaryExprAST::Codegen() {
1036 Value *L = LHS->Codegen();
1037 Value *R = RHS->Codegen();
1038 if (L == 0 || R == 0) return 0;
1041 case '+': return Builder.CreateAdd(L, R, "addtmp");
1042 case '-': return Builder.CreateSub(L, R, "subtmp");
1043 case '*': return Builder.CreateMul(L, R, "multmp");
1045 L = Builder.CreateFCmpULT(L, R, "cmptmp");
1046 // Convert bool 0/1 to double 0.0 or 1.0
1047 return Builder.CreateUIToFP(L, Type::DoubleTy, "booltmp");
1048 default: return ErrorV("invalid binary operator");
1052 Value *CallExprAST::Codegen() {
1053 // Look up the name in the global module table.
1054 Function *CalleeF = TheModule->getFunction(Callee);
1056 return ErrorV("Unknown function referenced");
1058 // If argument mismatch error.
1059 if (CalleeF->arg_size() != Args.size())
1060 return ErrorV("Incorrect # arguments passed");
1062 std::vector<Value*> ArgsV;
1063 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
1064 ArgsV.push_back(Args[i]->Codegen());
1065 if (ArgsV.back() == 0) return 0;
1068 return Builder.CreateCall(CalleeF, ArgsV.begin(), ArgsV.end(), "calltmp");
1071 Function *PrototypeAST::Codegen() {
1072 // Make the function type: double(double,double) etc.
1073 std::vector<const Type*> Doubles(Args.size(), Type::DoubleTy);
1074 FunctionType *FT = FunctionType::get(Type::DoubleTy, Doubles, false);
1076 Function *F = new Function(FT, Function::ExternalLinkage, Name, TheModule);
1078 // If F conflicted, there was already something named 'Name'. If it has a
1079 // body, don't allow redefinition or reextern.
1080 if (F->getName() != Name) {
1081 // Delete the one we just made and get the existing one.
1082 F->eraseFromParent();
1083 F = TheModule->getFunction(Name);
1085 // If F already has a body, reject this.
1086 if (!F->empty()) {
1087 ErrorF("redefinition of function");
1091 // If F took a different number of args, reject.
1092 if (F->arg_size() != Args.size()) {
1093 ErrorF("redefinition of function with different # args");
1098 // Set names for all arguments.
1100 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
1102 AI->setName(Args[Idx]);
1104 // Add arguments to variable symbol table.
1105 NamedValues[Args[Idx]] = AI;
1111 Function *FunctionAST::Codegen() {
1112 NamedValues.clear();
1114 Function *TheFunction = Proto->Codegen();
1115 if (TheFunction == 0)
1118 // Create a new basic block to start insertion into.
1119 BasicBlock *BB = new BasicBlock("entry", TheFunction);
1120 Builder.SetInsertPoint(BB);
1122 if (Value *RetVal = Body->Codegen()) {
1123 // Finish off the function.
1124 Builder.CreateRet(RetVal);
1126 // Validate the generated code, checking for consistency.
1127 verifyFunction(*TheFunction);
1131 // Error reading body, remove function.
1132 TheFunction->eraseFromParent();
1136 //===----------------------------------------------------------------------===//
1137 // Top-Level parsing and JIT Driver
1138 //===----------------------------------------------------------------------===//
1140 static void HandleDefinition() {
1141 if (FunctionAST *F = ParseDefinition()) {
1142 if (Function *LF = F->Codegen()) {
1143 fprintf(stderr, "Read function definition:");
1147 // Skip token for error recovery.
1152 static void HandleExtern() {
1153 if (PrototypeAST *P = ParseExtern()) {
1154 if (Function *F = P->Codegen()) {
1155 fprintf(stderr, "Read extern: ");
1159 // Skip token for error recovery.
1164 static void HandleTopLevelExpression() {
1165 // Evaluate a top level expression into an anonymous function.
1166 if (FunctionAST *F = ParseTopLevelExpr()) {
1167 if (Function *LF = F->Codegen()) {
1168 fprintf(stderr, "Read top-level expression:");
1172 // Skip token for error recovery.
1177 /// top ::= definition | external | expression | ';'
1178 static void MainLoop() {
1180 fprintf(stderr, "ready> ");
1182 case tok_eof: return;
1183 case ';': getNextToken(); break; // ignore top level semicolons.
1184 case tok_def: HandleDefinition(); break;
1185 case tok_extern: HandleExtern(); break;
1186 default: HandleTopLevelExpression(); break;
1193 //===----------------------------------------------------------------------===//
1194 // "Library" functions that can be "extern'd" from user code.
1195 //===----------------------------------------------------------------------===//
1197 /// putchard - putchar that takes a double and returns 0.
1199 double putchard(double X) {
1204 //===----------------------------------------------------------------------===//
1205 // Main driver code.
1206 //===----------------------------------------------------------------------===//
1209 TheModule = new Module("my cool jit");
1211 // Install standard binary operators.
1212 // 1 is lowest precedence.
1213 BinopPrecedence['<'] = 10;
1214 BinopPrecedence['+'] = 20;
1215 BinopPrecedence['-'] = 20;
1216 BinopPrecedence['*'] = 40; // highest.
1218 // Prime the first token.
1219 fprintf(stderr, "ready> ");
1223 TheModule->dump();
1230 <!-- *********************************************************************** -->
1233 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1234 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1235 <a href="http://validator.w3.org/check/referer"><img
1236 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
1238 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1239 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
1240 Last modified: $Date: 2007-10-17 11:05:13 -0700 (Wed, 17 Oct 2007) $