docs/tutorial/LangImpl3.html

   1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
   2                       "http://www.w3.org/TR/html4/strict.dtd">
   3
   4 <html>
   5 <head>
   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">
  10 </head>
  11
  12 <body>
  13
  14 <div class="doc_title">Kaleidoscope: Code generation to LLVM IR</div>
  15
  16 <ul>
  17 <li><a href="index.html">Up to Tutorial Index</a></li>
  18 <li>Chapter 3
  19   <ol>
  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>
  26   </ol>
  27 </li>
  28 <li><a href="LangImpl4.html">Chapter 4</a>: Adding JIT and Optimizer
  29 Support</li>
  30 </ul>
  31
  32 <div class="doc_author">
  33   <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
  34 </div>
  35
  36 <!-- *********************************************************************** -->
  37 <div class="doc_section"><a name="intro">Chapter 3 Introduction</a></div>
  38 <!-- *********************************************************************** -->
  39
  40 <div class="doc_text">
  41
  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. :)
  48 </p>
  49
  50 <p><b>Please note</b>: the code in this chapter and later require LLVM 2.2 or
  51 later.  LLVM 2.1 and before will not work with it.  Also note that you need
  52 to use a version of this tutorial that matches your LLVM release: If you are
  53 using an official LLVM release, use the version of the documentation included
  54 with your release or on the <a href="http://llvm.org/releases/">llvm.org
  55 releases page</a>.</p>
  56
  57 </div>
  58
  59 <!-- *********************************************************************** -->
  60 <div class="doc_section"><a name="basics">Code Generation Setup</a></div>
  61 <!-- *********************************************************************** -->
  62
  63 <div class="doc_text">
  64
  65 <p>
  66 In order to generate LLVM IR, we want some simple setup to get started.  First
  67 we define virtual code generation (codegen) methods in each AST class:</p>
  68
  69 <div class="doc_code">
  70 <pre>
  71 /// ExprAST - Base class for all expression nodes.
  72 class ExprAST {
  73 public:
  74   virtual ~ExprAST() {}
  75   <b>virtual Value *Codegen() = 0;</b>
  76 };
  77
  78 /// NumberExprAST - Expression class for numeric literals like "1.0".
  79 class NumberExprAST : public ExprAST {
  80   double Val;
  81 public:
  82   explicit NumberExprAST(double val) : Val(val) {}
  83   <b>virtual Value *Codegen();</b>
  84 };
  85 ...
  86 </pre>
  87 </div>
  88
  89 <p>The Codegen() method says to emit IR for that AST node along with all the things it
  90 depends on, and they all return an LLVM Value object.
  91 "Value" is the class used to represent a "<a
  92 href="http://en.wikipedia.org/wiki/Static_single_assignment_form">Static Single
  93 Assignment (SSA)</a> register" or "SSA value" in LLVM.  The most distinct aspect
  94 of SSA values is that their value is computed as the related instruction
  95 executes, and it does not get a new value until (and if) the instruction
  96 re-executes.  In other words, there is no way to "change" an SSA value.  For
  97 more information, please read up on <a
  98 href="http://en.wikipedia.org/wiki/Static_single_assignment_form">Static Single
  99 Assignment</a> - the concepts are really quite natural once you grok them.</p>
 100
 101 <p>Note that instead of adding virtual methods to the ExprAST class hierarchy,
 102 it could also make sense to use a <a
 103 href="http://en.wikipedia.org/wiki/Visitor_pattern">visitor pattern</a> or some
 104 other way to model this.  Again, this tutorial won't dwell on good software
 105 engineering practices: for our purposes, adding a virtual method is
 106 simplest.</p>
 107
 108 <p>The
 109 second thing we want is an "Error" method like we used for the parser, which will
 110 be used to report errors found during code generation (for example, use of an
 111 undeclared parameter):</p>
 112
 113 <div class="doc_code">
 114 <pre>
 115 Value *ErrorV(const char *Str) { Error(Str); return 0; }
 116
 117 static Module *TheModule;
 118 static IRBuilder&lt;&gt; Builder(getGlobalContext());
 119 static std::map&lt;std::string, Value*&gt; NamedValues;
 120 </pre>
 121 </div>
 122
 123 <p>The static variables will be used during code generation.  <tt>TheModule</tt>
 124 is the LLVM construct that contains all of the functions and global variables in
 125 a chunk of code.  In many ways, it is the top-level structure that the LLVM IR
 126 uses to contain code.</p>
 127
 128 <p>The <tt>Builder</tt> object is a helper object that makes it easy to generate
 129 LLVM instructions.  Instances of the <a
 130 href="http://llvm.org/doxygen/IRBuilder_8h-source.html"><tt>IRBuilder</tt></a>
 131 class template keep track of the current place to insert instructions and has
 132 methods to create new instructions.</p>
 133
 134 <p>The <tt>NamedValues</tt> map keeps track of which values are defined in the
 135 current scope and what their LLVM representation is.  (In other words, it is a
 136 symbol table for the code).  In this form of Kaleidoscope, the only things that
 137 can be referenced are function parameters.  As such, function parameters will
 138 be in this map when generating code for their function body.</p>
 139
 140 <p>
 141 With these basics in place, we can start talking about how to generate code for
 142 each expression.  Note that this assumes that the <tt>Builder</tt> has been set
 143 up to generate code <em>into</em> something.  For now, we'll assume that this
 144 has already been done, and we'll just use it to emit code.
 145 </p>
 146
 147 </div>
 148
 149 <!-- *********************************************************************** -->
 150 <div class="doc_section"><a name="exprs">Expression Code Generation</a></div>
 151 <!-- *********************************************************************** -->
 152
 153 <div class="doc_text">
 154
 155 <p>Generating LLVM code for expression nodes is very straightforward: less
 156 than 45 lines of commented code for all four of our expression nodes.  First
 157 we'll do numeric literals:</p>
 158
 159 <div class="doc_code">
 160 <pre>
 161 Value *NumberExprAST::Codegen() {
 162   return ConstantFP::get(getGlobalContext(), APFloat(Val));
 163 }
 164 </pre>
 165 </div>
 166
 167 <p>In the LLVM IR, numeric constants are represented with the
 168 <tt>ConstantFP</tt> class, which holds the numeric value in an <tt>APFloat</tt>
 169 internally (<tt>APFloat</tt> has the capability of holding floating point
 170 constants of <em>A</em>rbitrary <em>P</em>recision).  This code basically just
 171 creates and returns a <tt>ConstantFP</tt>.  Note that in the LLVM IR
 172 that constants are all uniqued together and shared.  For this reason, the API
 173 uses "the Context.get..." idiom instead of "new foo(..)" or "foo::Create(..)".</p>
 174
 175 <div class="doc_code">
 176 <pre>
 177 Value *VariableExprAST::Codegen() {
 178   // Look this variable up in the function.
 179   Value *V = NamedValues[Name];
 180   return V ? V : ErrorV("Unknown variable name");
 181 }
 182 </pre>
 183 </div>
 184
 185 <p>References to variables are also quite simple using LLVM.  In the simple version
 186 of Kaleidoscope, we assume that the variable has already been emited somewhere
 187 and its value is available.  In practice, the only values that can be in the
 188 <tt>NamedValues</tt> map are function arguments.  This
 189 code simply checks to see that the specified name is in the map (if not, an
 190 unknown variable is being referenced) and returns the value for it.  In future
 191 chapters, we'll add support for <a href="LangImpl5.html#for">loop induction
 192 variables</a> in the symbol table, and for <a
 193 href="LangImpl7.html#localvars">local variables</a>.</p>
 194
 195 <div class="doc_code">
 196 <pre>
 197 Value *BinaryExprAST::Codegen() {
 198   Value *L = LHS-&gt;Codegen();
 199   Value *R = RHS-&gt;Codegen();
 200   if (L == 0 || R == 0) return 0;
 201
 202   switch (Op) {
 203   case '+': return Builder.CreateAdd(L, R, "addtmp");
 204   case '-': return Builder.CreateSub(L, R, "subtmp");
 205   case '*': return Builder.CreateMul(L, R, "multmp");
 206   case '&lt;':
 207     L = Builder.CreateFCmpULT(L, R, "cmptmp");
 208     // Convert bool 0/1 to double 0.0 or 1.0
 209     return Builder.CreateUIToFP(L, Type::getDoubleTy(getGlobalContext()),
 210                                 "booltmp");
 211   default: return ErrorV("invalid binary operator");
 212   }
 213 }
 214 </pre>
 215 </div>
 216
 217 <p>Binary operators start to get more interesting.  The basic idea here is that
 218 we recursively emit code for the left-hand side of the expression, then the
 219 right-hand side, then we compute the result of the binary expression.  In this
 220 code, we do a simple switch on the opcode to create the right LLVM instruction.
 221 </p>
 222
 223 <p>In the example above, the LLVM builder class is starting to show its value.
 224 IRBuilder knows where to insert the newly created instruction, all you have to
 225 do is specify what instruction to create (e.g. with <tt>CreateAdd</tt>), which
 226 operands to use (<tt>L</tt> and <tt>R</tt> here) and optionally provide a name
 227 for the generated instruction.</p>
 228
 229 <p>One nice thing about LLVM is that the name is just a hint.  For instance, if
 230 the code above emits multiple "addtmp" variables, LLVM will automatically
 231 provide each one with an increasing, unique numeric suffix.  Local value names
 232 for instructions are purely optional, but it makes it much easier to read the
 233 IR dumps.</p>
 234
 235 <p><a href="../LangRef.html#instref">LLVM instructions</a> are constrained by
 236 strict rules: for example, the Left and Right operators of
 237 an <a href="../LangRef.html#i_add">add instruction</a> must have the same
 238 type, and the result type of the add must match the operand types.  Because
 239 all values in Kaleidoscope are doubles, this makes for very simple code for add,
 240 sub and mul.</p>
 241
 242 <p>On the other hand, LLVM specifies that the <a
 243 href="../LangRef.html#i_fcmp">fcmp instruction</a> always returns an 'i1' value
 244 (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
 245 a <a href="../LangRef.html#i_uitofp">uitofp instruction</a>.  This instruction
 246 converts its input integer into a floating point value by treating the input
 247 as an unsigned value.  In contrast, if we used the <a
 248 href="../LangRef.html#i_sitofp">sitofp instruction</a>, the Kaleidoscope '&lt;'
 249 operator would return 0.0 and -1.0, depending on the input value.</p>
 250
 251 <div class="doc_code">
 252 <pre>
 253 Value *CallExprAST::Codegen() {
 254   // Look up the name in the global module table.
 255   Function *CalleeF = TheModule-&gt;getFunction(Callee);
 256   if (CalleeF == 0)
 257     return ErrorV("Unknown function referenced");
 258
 259   // If argument mismatch error.
 260   if (CalleeF-&gt;arg_size() != Args.size())
 261     return ErrorV("Incorrect # arguments passed");
 262
 263   std::vector&lt;Value*&gt; ArgsV;
 264   for (unsigned i = 0, e = Args.size(); i != e; ++i) {
 265     ArgsV.push_back(Args[i]-&gt;Codegen());
 266     if (ArgsV.back() == 0) return 0;
 267   }
 268
 269   return Builder.CreateCall(CalleeF, ArgsV.begin(), ArgsV.end(), "calltmp");
 270 }
 271 </pre>
 272 </div>
 273
 274 <p>Code generation for function calls is quite straightforward with LLVM.  The
 275 code above initially does a function name lookup in the LLVM Module's symbol
 276 table.  Recall that the LLVM Module is the container that holds all of the
 277 functions we are JIT'ing.  By giving each function the same name as what the
 278 user specifies, we can use the LLVM symbol table to resolve function names for
 279 us.</p>
 280
 281 <p>Once we have the function to call, we recursively codegen each argument that
 282 is to be passed in, and create an LLVM <a href="../LangRef.html#i_call">call
 283 instruction</a>.  Note that LLVM uses the native C calling conventions by
 284 default, allowing these calls to also call into standard library functions like
 285 "sin" and "cos", with no additional effort.</p>
 286
 287 <p>This wraps up our handling of the four basic expressions that we have so far
 288 in Kaleidoscope.  Feel free to go in and add some more.  For example, by
 289 browsing the <a href="../LangRef.html">LLVM language reference</a> you'll find
 290 several other interesting instructions that are really easy to plug into our
 291 basic framework.</p>
 292
 293 </div>
 294
 295 <!-- *********************************************************************** -->
 296 <div class="doc_section"><a name="funcs">Function Code Generation</a></div>
 297 <!-- *********************************************************************** -->
 298
 299 <div class="doc_text">
 300
 301 <p>Code generation for prototypes and functions must handle a number of
 302 details, which make their code less beautiful than expression code
 303 generation, but allows us to  illustrate some important points.  First, lets
 304 talk about code generation for prototypes: they are used both for function
 305 bodies and external function declarations.  The code starts with:</p>
 306
 307 <div class="doc_code">
 308 <pre>
 309 Function *PrototypeAST::Codegen() {
 310   // Make the function type:  double(double,double) etc.
 311   std::vector&lt;const Type*&gt; Doubles(Args.size(),
 312                                    Type::getDoubleTy(getGlobalContext()));
 313   FunctionType *FT = FunctionType::get(Type::getDoubleTy(getGlobalContext()),
 314                                        Doubles, false);
 315
 316   Function *F = Function::Create(FT, Function::ExternalLinkage, Name, TheModule);
 317 </pre>
 318 </div>
 319
 320 <p>This code packs a lot of power into a few lines.  Note first that this
 321 function returns a "Function*" instead of a "Value*".  Because a "prototype"
 322 really talks about the external interface for a function (not the value computed
 323 by an expression), it makes sense for it to return the LLVM Function it
 324 corresponds to when codegen'd.</p>
 325
 326 <p>The call to <tt>Context.get</tt> creates
 327 the <tt>FunctionType</tt> that should be used for a given Prototype.  Since all
 328 function arguments in Kaleidoscope are of type double, the first line creates
 329 a vector of "N" LLVM double types.  It then uses the <tt>Context.get</tt>
 330 method to create a function type that takes "N" doubles as arguments, returns
 331 one double as a result, and that is not vararg (the false parameter indicates
 332 this).  Note that Types in LLVM are uniqued just like Constants are, so you
 333 don't "new" a type, you "get" it.</p>
 334
 335 <p>The final line above actually creates the function that the prototype will
 336 correspond to.  This indicates the type, linkage and name to use, as well as which
 337 module to insert into.  "<a href="../LangRef.html#linkage">external linkage</a>"
 338 means that the function may be defined outside the current module and/or that it
 339 is callable by functions outside the module.  The Name passed in is the name the
 340 user specified: since "<tt>TheModule</tt>" is specified, this name is registered
 341 in "<tt>TheModule</tt>"s symbol table, which is used by the function call code
 342 above.</p>
 343
 344 <div class="doc_code">
 345 <pre>
 346   // If F conflicted, there was already something named 'Name'.  If it has a
 347   // body, don't allow redefinition or reextern.
 348   if (F-&gt;getName() != Name) {
 349     // Delete the one we just made and get the existing one.
 350     F-&gt;eraseFromParent();
 351     F = TheModule-&gt;getFunction(Name);
 352 </pre>
 353 </div>
 354
 355 <p>The Module symbol table works just like the Function symbol table when it
 356 comes to name conflicts: if a new function is created with a name was previously
 357 added to the symbol table, it will get implicitly renamed when added to the
 358 Module.  The code above exploits this fact to determine if there was a previous
 359 definition of this function.</p>
 360
 361 <p>In Kaleidoscope, I choose to allow redefinitions of functions in two cases:
 362 first, we want to allow 'extern'ing a function more than once, as long as the
 363 prototypes for the externs match (since all arguments have the same type, we
 364 just have to check that the number of arguments match).  Second, we want to
 365 allow 'extern'ing a function and then definining a body for it.  This is useful
 366 when defining mutually recursive functions.</p>
 367
 368 <p>In order to implement this, the code above first checks to see if there is
 369 a collision on the name of the function.  If so, it deletes the function we just
 370 created (by calling <tt>eraseFromParent</tt>) and then calling
 371 <tt>getFunction</tt> to get the existing function with the specified name.  Note
 372 that many APIs in LLVM have "erase" forms and "remove" forms.  The "remove" form
 373 unlinks the object from its parent (e.g. a Function from a Module) and returns
 374 it.  The "erase" form unlinks the object and then deletes it.</p>
 375
 376 <div class="doc_code">
 377 <pre>
 378     // If F already has a body, reject this.
 379     if (!F-&gt;empty()) {
 380       ErrorF("redefinition of function");
 381       return 0;
 382     }
 383
 384     // If F took a different number of args, reject.
 385     if (F-&gt;arg_size() != Args.size()) {
 386       ErrorF("redefinition of function with different # args");
 387       return 0;
 388     }
 389   }
 390 </pre>
 391 </div>
 392
 393 <p>In order to verify the logic above, we first check to see if the pre-existing
 394 function is "empty".  In this case, empty means that it has no basic blocks in
 395 it, which means it has no body.  If it has no body, it is a forward
 396 declaration.  Since we don't allow anything after a full definition of the
 397 function, the code rejects this case.  If the previous reference to a function
 398 was an 'extern', we simply verify that the number of arguments for that
 399 definition and this one match up.  If not, we emit an error.</p>
 400
 401 <div class="doc_code">
 402 <pre>
 403   // Set names for all arguments.
 404   unsigned Idx = 0;
 405   for (Function::arg_iterator AI = F-&gt;arg_begin(); Idx != Args.size();
 406        ++AI, ++Idx) {
 407     AI-&gt;setName(Args[Idx]);
 408
 409     // Add arguments to variable symbol table.
 410     NamedValues[Args[Idx]] = AI;
 411   }
 412   return F;
 413 }
 414 </pre>
 415 </div>
 416
 417 <p>The last bit of code for prototypes loops over all of the arguments in the
 418 function, setting the name of the LLVM Argument objects to match, and registering
 419 the arguments in the <tt>NamedValues</tt> map for future use by the
 420 <tt>VariableExprAST</tt> AST node.  Once this is set up, it returns the Function
 421 object to the caller.  Note that we don't check for conflicting
 422 argument names here (e.g. "extern foo(a b a)").  Doing so would be very
 423 straight-forward with the mechanics we have already used above.</p>
 424
 425 <div class="doc_code">
 426 <pre>
 427 Function *FunctionAST::Codegen() {
 428   NamedValues.clear();
 429
 430   Function *TheFunction = Proto-&gt;Codegen();
 431   if (TheFunction == 0)
 432     return 0;
 433 </pre>
 434 </div>
 435
 436 <p>Code generation for function definitions starts out simply enough: we just
 437 codegen the prototype (Proto) and verify that it is ok.  We then clear out the
 438 <tt>NamedValues</tt> map to make sure that there isn't anything in it from the
 439 last function we compiled.  Code generation of the prototype ensures that there
 440 is an LLVM Function object that is ready to go for us.</p>
 441
 442 <div class="doc_code">
 443 <pre>
 444   // Create a new basic block to start insertion into.
 445   BasicBlock *BB = BasicBlock::Create(getGlobalContext(), "entry", TheFunction);
 446   Builder.SetInsertPoint(BB);
 447
 448   if (Value *RetVal = Body-&gt;Codegen()) {
 449 </pre>
 450 </div>
 451
 452 <p>Now we get to the point where the <tt>Builder</tt> is set up.  The first
 453 line creates a new <a href="http://en.wikipedia.org/wiki/Basic_block">basic
 454 block</a> (named "entry"), which is inserted into <tt>TheFunction</tt>.  The
 455 second line then tells the builder that new instructions should be inserted into
 456 the end of the new basic block.  Basic blocks in LLVM are an important part
 457 of functions that define the <a
 458 href="http://en.wikipedia.org/wiki/Control_flow_graph">Control Flow Graph</a>.
 459 Since we don't have any control flow, our functions will only contain one
 460 block at this point.  We'll fix this in <a href="LangImpl5.html">Chapter 5</a> :).</p>
 461
 462 <div class="doc_code">
 463 <pre>
 464   if (Value *RetVal = Body-&gt;Codegen()) {
 465     // Finish off the function.
 466     Builder.CreateRet(RetVal);
 467
 468     // Validate the generated code, checking for consistency.
 469     verifyFunction(*TheFunction);
 470     return TheFunction;
 471   }
 472 </pre>
 473 </div>
 474
 475 <p>Once the insertion point is set up, we call the <tt>CodeGen()</tt> method for
 476 the root expression of the function.  If no error happens, this emits code to
 477 compute the expression into the entry block and returns the value that was
 478 computed.  Assuming no error, we then create an LLVM <a
 479 href="../LangRef.html#i_ret">ret instruction</a>, which completes the function.
 480 Once the function is built, we call <tt>verifyFunction</tt>, which
 481 is provided by LLVM.  This function does a variety of consistency checks on the
 482 generated code, to determine if our compiler is doing everything right.  Using
 483 this is important: it can catch a lot of bugs.  Once the function is finished
 484 and validated, we return it.</p>
 485
 486 <div class="doc_code">
 487 <pre>
 488   // Error reading body, remove function.
 489   TheFunction-&gt;eraseFromParent();
 490   return 0;
 491 }
 492 </pre>
 493 </div>
 494
 495 <p>The only piece left here is handling of the error case.  For simplicity, we
 496 handle this by merely deleting the function we produced with the
 497 <tt>eraseFromParent</tt> method.  This allows the user to redefine a function
 498 that they incorrectly typed in before: if we didn't delete it, it would live in
 499 the symbol table, with a body, preventing future redefinition.</p>
 500
 501 <p>This code does have a bug, though.  Since the <tt>PrototypeAST::Codegen</tt>
 502 can return a previously defined forward declaration, our code can actually delete
 503 a forward declaration.  There are a number of ways to fix this bug, see what you
 504 can come up with!  Here is a testcase:</p>
 505
 506 <div class="doc_code">
 507 <pre>
 508 extern foo(a b);     # ok, defines foo.
 509 def foo(a b) c;      # error, 'c' is invalid.
 510 def bar() foo(1, 2); # error, unknown function "foo"
 511 </pre>
 512 </div>
 513
 514 </div>
 515
 516 <!-- *********************************************************************** -->
 517 <div class="doc_section"><a name="driver">Driver Changes and
 518 Closing Thoughts</a></div>
 519 <!-- *********************************************************************** -->
 520
 521 <div class="doc_text">
 522
 523 <p>
 524 For now, code generation to LLVM doesn't really get us much, except that we can
 525 look at the pretty IR calls.  The sample code inserts calls to Codegen into the
 526 "<tt>HandleDefinition</tt>", "<tt>HandleExtern</tt>" etc functions, and then
 527 dumps out the LLVM IR.  This gives a nice way to look at the LLVM IR for simple
 528 functions.  For example:
 529 </p>
 530
 531 <div class="doc_code">
 532 <pre>
 533 ready> <b>4+5</b>;
 534 Read top-level expression:
 535 define double @""() {
 536 entry:
 537         %addtmp = add double 4.000000e+00, 5.000000e+00
 538         ret double %addtmp
 539 }
 540 </pre>
 541 </div>
 542
 543 <p>Note how the parser turns the top-level expression into anonymous functions
 544 for us.  This will be handy when we add <a href="LangImpl4.html#jit">JIT
 545 support</a> in the next chapter.  Also note that the code is very literally
 546 transcribed, no optimizations are being performed.  We will
 547 <a href="LangImpl4.html#trivialconstfold">add optimizations</a> explicitly in
 548 the next chapter.</p>
 549
 550 <div class="doc_code">
 551 <pre>
 552 ready&gt; <b>def foo(a b) a*a + 2*a*b + b*b;</b>
 553 Read function definition:
 554 define double @foo(double %a, double %b) {
 555 entry:
 556         %multmp = mul double %a, %a
 557         %multmp1 = mul double 2.000000e+00, %a
 558         %multmp2 = mul double %multmp1, %b
 559         %addtmp = add double %multmp, %multmp2
 560         %multmp3 = mul double %b, %b
 561         %addtmp4 = add double %addtmp, %multmp3
 562         ret double %addtmp4
 563 }
 564 </pre>
 565 </div>
 566
 567 <p>This shows some simple arithmetic. Notice the striking similarity to the
 568 LLVM builder calls that we use to create the instructions.</p>
 569
 570 <div class="doc_code">
 571 <pre>
 572 ready&gt; <b>def bar(a) foo(a, 4.0) + bar(31337);</b>
 573 Read function definition:
 574 define double @bar(double %a) {
 575 entry:
 576         %calltmp = call double @foo( double %a, double 4.000000e+00 )
 577         %calltmp1 = call double @bar( double 3.133700e+04 )
 578         %addtmp = add double %calltmp, %calltmp1
 579         ret double %addtmp
 580 }
 581 </pre>
 582 </div>
 583
 584 <p>This shows some function calls.  Note that this function will take a long
 585 time to execute if you call it.  In the future we'll add conditional control
 586 flow to actually make recursion useful :).</p>
 587
 588 <div class="doc_code">
 589 <pre>
 590 ready&gt; <b>extern cos(x);</b>
 591 Read extern:
 592 declare double @cos(double)
 593
 594 ready&gt; <b>cos(1.234);</b>
 595 Read top-level expression:
 596 define double @""() {
 597 entry:
 598         %calltmp = call double @cos( double 1.234000e+00 )
 599         ret double %calltmp
 600 }
 601 </pre>
 602 </div>
 603
 604 <p>This shows an extern for the libm "cos" function, and a call to it.</p>
 605
 606
 607 <div class="doc_code">
 608 <pre>
 609 ready&gt; <b>^D</b>
 610 ; ModuleID = 'my cool jit'
 611
 612 define double @""() {
 613 entry:
 614         %addtmp = add double 4.000000e+00, 5.000000e+00
 615         ret double %addtmp
 616 }
 617
 618 define double @foo(double %a, double %b) {
 619 entry:
 620         %multmp = mul double %a, %a
 621         %multmp1 = mul double 2.000000e+00, %a
 622         %multmp2 = mul double %multmp1, %b
 623         %addtmp = add double %multmp, %multmp2
 624         %multmp3 = mul double %b, %b
 625         %addtmp4 = add double %addtmp, %multmp3
 626         ret double %addtmp4
 627 }
 628
 629 define double @bar(double %a) {
 630 entry:
 631         %calltmp = call double @foo( double %a, double 4.000000e+00 )
 632         %calltmp1 = call double @bar( double 3.133700e+04 )
 633         %addtmp = add double %calltmp, %calltmp1
 634         ret double %addtmp
 635 }
 636
 637 declare double @cos(double)
 638
 639 define double @""() {
 640 entry:
 641         %calltmp = call double @cos( double 1.234000e+00 )
 642         ret double %calltmp
 643 }
 644 </pre>
 645 </div>
 646
 647 <p>When you quit the current demo, it dumps out the IR for the entire module
 648 generated.  Here you can see the big picture with all the functions referencing
 649 each other.</p>
 650
 651 <p>This wraps up the third chapter of the Kaleidoscope tutorial.  Up next, we'll
 652 describe how to <a href="LangImpl4.html">add JIT codegen and optimizer
 653 support</a> to this so we can actually start running code!</p>
 654
 655 </div>
 656
 657
 658 <!-- *********************************************************************** -->
 659 <div class="doc_section"><a name="code">Full Code Listing</a></div>
 660 <!-- *********************************************************************** -->
 661
 662 <div class="doc_text">
 663
 664 <p>
 665 Here is the complete code listing for our running example, enhanced with the
 666 LLVM code generator.    Because this uses the LLVM libraries, we need to link
 667 them in.  To do this, we use the <a
 668 href="http://llvm.org/cmds/llvm-config.html">llvm-config</a> tool to inform
 669 our makefile/command line about which options to use:</p>
 670
 671 <div class="doc_code">
 672 <pre>
 673    # Compile
 674    g++ -g -O3 toy.cpp `llvm-config --cppflags --ldflags --libs core` -o toy
 675    # Run
 676    ./toy
 677 </pre>
 678 </div>
 679
 680 <p>Here is the code:</p>
 681
 682 <div class="doc_code">
 683 <pre>
 684 // To build this:
 685 // See example below.
 686
 687 #include "llvm/DerivedTypes.h"
 688 #include "llvm/LLVMContext.h"
 689 #include "llvm/Module.h"
 690 #include "llvm/Analysis/Verifier.h"
 691 #include "llvm/Support/IRBuilder.h"
 692 #include &lt;cstdio&gt;
 693 #include &lt;string&gt;
 694 #include &lt;map&gt;
 695 #include &lt;vector&gt;
 696 using namespace llvm;
 697
 698 //===----------------------------------------------------------------------===//
 699 // Lexer
 700 //===----------------------------------------------------------------------===//
 701
 702 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
 703 // of these for known things.
 704 enum Token {
 705   tok_eof = -1,
 706
 707   // commands
 708   tok_def = -2, tok_extern = -3,
 709
 710   // primary
 711   tok_identifier = -4, tok_number = -5,
 712 };
 713
 714 static std::string IdentifierStr;  // Filled in if tok_identifier
 715 static double NumVal;              // Filled in if tok_number
 716
 717 /// gettok - Return the next token from standard input.
 718 static int gettok() {
 719   static int LastChar = ' ';
 720
 721   // Skip any whitespace.
 722   while (isspace(LastChar))
 723     LastChar = getchar();
 724
 725   if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
 726     IdentifierStr = LastChar;
 727     while (isalnum((LastChar = getchar())))
 728       IdentifierStr += LastChar;
 729
 730     if (IdentifierStr == "def") return tok_def;
 731     if (IdentifierStr == "extern") return tok_extern;
 732     return tok_identifier;
 733   }
 734
 735   if (isdigit(LastChar) || LastChar == '.') {   // Number: [0-9.]+
 736     std::string NumStr;
 737     do {
 738       NumStr += LastChar;
 739       LastChar = getchar();
 740     } while (isdigit(LastChar) || LastChar == '.');
 741
 742     NumVal = strtod(NumStr.c_str(), 0);
 743     return tok_number;
 744   }
 745
 746   if (LastChar == '#') {
 747     // Comment until end of line.
 748     do LastChar = getchar();
 749     while (LastChar != EOF &amp;&amp; LastChar != '\n' &amp;&amp; LastChar != '\r');
 750
 751     if (LastChar != EOF)
 752       return gettok();
 753   }
 754
 755   // Check for end of file.  Don't eat the EOF.
 756   if (LastChar == EOF)
 757     return tok_eof;
 758
 759   // Otherwise, just return the character as its ascii value.
 760   int ThisChar = LastChar;
 761   LastChar = getchar();
 762   return ThisChar;
 763 }
 764
 765 //===----------------------------------------------------------------------===//
 766 // Abstract Syntax Tree (aka Parse Tree)
 767 //===----------------------------------------------------------------------===//
 768
 769 /// ExprAST - Base class for all expression nodes.
 770 class ExprAST {
 771 public:
 772   virtual ~ExprAST() {}
 773   virtual Value *Codegen() = 0;
 774 };
 775
 776 /// NumberExprAST - Expression class for numeric literals like "1.0".
 777 class NumberExprAST : public ExprAST {
 778   double Val;
 779 public:
 780   explicit NumberExprAST(double val) : Val(val) {}
 781   virtual Value *Codegen();
 782 };
 783
 784 /// VariableExprAST - Expression class for referencing a variable, like "a".
 785 class VariableExprAST : public ExprAST {
 786   std::string Name;
 787 public:
 788   explicit VariableExprAST(const std::string &amp;name) : Name(name) {}
 789   virtual Value *Codegen();
 790 };
 791
 792 /// BinaryExprAST - Expression class for a binary operator.
 793 class BinaryExprAST : public ExprAST {
 794   char Op;
 795   ExprAST *LHS, *RHS;
 796 public:
 797   BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs)
 798     : Op(op), LHS(lhs), RHS(rhs) {}
 799   virtual Value *Codegen();
 800 };
 801
 802 /// CallExprAST - Expression class for function calls.
 803 class CallExprAST : public ExprAST {
 804   std::string Callee;
 805   std::vector&lt;ExprAST*&gt; Args;
 806 public:
 807   CallExprAST(const std::string &amp;callee, std::vector&lt;ExprAST*&gt; &amp;args)
 808     : Callee(callee), Args(args) {}
 809   virtual Value *Codegen();
 810 };
 811
 812 /// PrototypeAST - This class represents the "prototype" for a function,
 813 /// which captures its argument names as well as if it is an operator.
 814 class PrototypeAST {
 815   std::string Name;
 816   std::vector&lt;std::string&gt; Args;
 817 public:
 818   PrototypeAST(const std::string &amp;name, const std::vector&lt;std::string&gt; &amp;args)
 819     : Name(name), Args(args) {}
 820
 821   Function *Codegen();
 822 };
 823
 824 /// FunctionAST - This class represents a function definition itself.
 825 class FunctionAST {
 826   PrototypeAST *Proto;
 827   ExprAST *Body;
 828 public:
 829   FunctionAST(PrototypeAST *proto, ExprAST *body)
 830     : Proto(proto), Body(body) {}
 831
 832   Function *Codegen();
 833 };
 834
 835 //===----------------------------------------------------------------------===//
 836 // Parser
 837 //===----------------------------------------------------------------------===//
 838
 839 /// CurTok/getNextToken - Provide a simple token buffer.  CurTok is the current
 840 /// token the parser it looking at.  getNextToken reads another token from the
 841 /// lexer and updates CurTok with its results.
 842 static int CurTok;
 843 static int getNextToken() {
 844   return CurTok = gettok();
 845 }
 846
 847 /// BinopPrecedence - This holds the precedence for each binary operator that is
 848 /// defined.
 849 static std::map&lt;char, int&gt; BinopPrecedence;
 850
 851 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
 852 static int GetTokPrecedence() {
 853   if (!isascii(CurTok))
 854     return -1;
 855
 856   // Make sure it's a declared binop.
 857   int TokPrec = BinopPrecedence[CurTok];
 858   if (TokPrec &lt;= 0) return -1;
 859   return TokPrec;
 860 }
 861
 862 /// Error* - These are little helper functions for error handling.
 863 ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;}
 864 PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; }
 865 FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; }
 866
 867 static ExprAST *ParseExpression();
 868
 869 /// identifierexpr
 870 ///   ::= identifier
 871 ///   ::= identifier '(' expression* ')'
 872 static ExprAST *ParseIdentifierExpr() {
 873   std::string IdName = IdentifierStr;
 874
 875   getNextToken();  // eat identifier.
 876
 877   if (CurTok != '(') // Simple variable ref.
 878     return new VariableExprAST(IdName);
 879
 880   // Call.
 881   getNextToken();  // eat (
 882   std::vector&lt;ExprAST*&gt; Args;
 883   if (CurTok != ')') {
 884     while (1) {
 885       ExprAST *Arg = ParseExpression();
 886       if (!Arg) return 0;
 887       Args.push_back(Arg);
 888
 889       if (CurTok == ')') break;
 890
 891       if (CurTok != ',')
 892         return Error("Expected ')' or ',' in argument list");
 893       getNextToken();
 894     }
 895   }
 896
 897   // Eat the ')'.
 898   getNextToken();
 899
 900   return new CallExprAST(IdName, Args);
 901 }
 902
 903 /// numberexpr ::= number
 904 static ExprAST *ParseNumberExpr() {
 905   ExprAST *Result = new NumberExprAST(NumVal);
 906   getNextToken(); // consume the number
 907   return Result;
 908 }
 909
 910 /// parenexpr ::= '(' expression ')'
 911 static ExprAST *ParseParenExpr() {
 912   getNextToken();  // eat (.
 913   ExprAST *V = ParseExpression();
 914   if (!V) return 0;
 915
 916   if (CurTok != ')')
 917     return Error("expected ')'");
 918   getNextToken();  // eat ).
 919   return V;
 920 }
 921
 922 /// primary
 923 ///   ::= identifierexpr
 924 ///   ::= numberexpr
 925 ///   ::= parenexpr
 926 static ExprAST *ParsePrimary() {
 927   switch (CurTok) {
 928   default: return Error("unknown token when expecting an expression");
 929   case tok_identifier: return ParseIdentifierExpr();
 930   case tok_number:     return ParseNumberExpr();
 931   case '(':            return ParseParenExpr();
 932   }
 933 }
 934
 935 /// binoprhs
 936 ///   ::= ('+' primary)*
 937 static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) {
 938   // If this is a binop, find its precedence.
 939   while (1) {
 940     int TokPrec = GetTokPrecedence();
 941
 942     // If this is a binop that binds at least as tightly as the current binop,
 943     // consume it, otherwise we are done.
 944     if (TokPrec &lt; ExprPrec)
 945       return LHS;
 946
 947     // Okay, we know this is a binop.
 948     int BinOp = CurTok;
 949     getNextToken();  // eat binop
 950
 951     // Parse the primary expression after the binary operator.
 952     ExprAST *RHS = ParsePrimary();
 953     if (!RHS) return 0;
 954
 955     // If BinOp binds less tightly with RHS than the operator after RHS, let
 956     // the pending operator take RHS as its LHS.
 957     int NextPrec = GetTokPrecedence();
 958     if (TokPrec &lt; NextPrec) {
 959       RHS = ParseBinOpRHS(TokPrec+1, RHS);
 960       if (RHS == 0) return 0;
 961     }
 962
 963     // Merge LHS/RHS.
 964     LHS = new BinaryExprAST(BinOp, LHS, RHS);
 965   }
 966 }
 967
 968 /// expression
 969 ///   ::= primary binoprhs
 970 ///
 971 static ExprAST *ParseExpression() {
 972   ExprAST *LHS = ParsePrimary();
 973   if (!LHS) return 0;
 974
 975   return ParseBinOpRHS(0, LHS);
 976 }
 977
 978 /// prototype
 979 ///   ::= id '(' id* ')'
 980 static PrototypeAST *ParsePrototype() {
 981   if (CurTok != tok_identifier)
 982     return ErrorP("Expected function name in prototype");
 983
 984   std::string FnName = IdentifierStr;
 985   getNextToken();
 986
 987   if (CurTok != '(')
 988     return ErrorP("Expected '(' in prototype");
 989
 990   std::vector&lt;std::string&gt; ArgNames;
 991   while (getNextToken() == tok_identifier)
 992     ArgNames.push_back(IdentifierStr);
 993   if (CurTok != ')')
 994     return ErrorP("Expected ')' in prototype");
 995
 996   // success.
 997   getNextToken();  // eat ')'.
 998
 999   return new PrototypeAST(FnName, ArgNames);
1000 }
1001
1002 /// definition ::= 'def' prototype expression
1003 static FunctionAST *ParseDefinition() {
1004   getNextToken();  // eat def.
1005   PrototypeAST *Proto = ParsePrototype();
1006   if (Proto == 0) return 0;
1007
1008   if (ExprAST *E = ParseExpression())
1009     return new FunctionAST(Proto, E);
1010   return 0;
1011 }
1012
1013 /// toplevelexpr ::= expression
1014 static FunctionAST *ParseTopLevelExpr() {
1015   if (ExprAST *E = ParseExpression()) {
1016     // Make an anonymous proto.
1017     PrototypeAST *Proto = new PrototypeAST("", std::vector&lt;std::string&gt;());
1018     return new FunctionAST(Proto, E);
1019   }
1020   return 0;
1021 }
1022
1023 /// external ::= 'extern' prototype
1024 static PrototypeAST *ParseExtern() {
1025   getNextToken();  // eat extern.
1026   return ParsePrototype();
1027 }
1028
1029 //===----------------------------------------------------------------------===//
1030 // Code Generation
1031 //===----------------------------------------------------------------------===//
1032
1033 static Module *TheModule;
1034 static IRBuilder&lt;&gt; Builder(getGlobalContext());
1035 static std::map&lt;std::string, Value*&gt; NamedValues;
1036
1037 Value *ErrorV(const char *Str) { Error(Str); return 0; }
1038
1039 Value *NumberExprAST::Codegen() {
1040   return ConstantFP::get(getGlobalContext(), APFloat(Val));
1041 }
1042
1043 Value *VariableExprAST::Codegen() {
1044   // Look this variable up in the function.
1045   Value *V = NamedValues[Name];
1046   return V ? V : ErrorV("Unknown variable name");
1047 }
1048
1049 Value *BinaryExprAST::Codegen() {
1050   Value *L = LHS-&gt;Codegen();
1051   Value *R = RHS-&gt;Codegen();
1052   if (L == 0 || R == 0) return 0;
1053
1054   switch (Op) {
1055   case '+': return Builder.CreateAdd(L, R, "addtmp");
1056   case '-': return Builder.CreateSub(L, R, "subtmp");
1057   case '*': return Builder.CreateMul(L, R, "multmp");
1058   case '&lt;':
1059     L = Builder.CreateFCmpULT(L, R, "cmptmp");
1060     // Convert bool 0/1 to double 0.0 or 1.0
1061     return Builder.CreateUIToFP(L, Type::getDoubleTy(getGlobalContext()), "booltmp");
1062   default: return ErrorV("invalid binary operator");
1063   }
1064 }
1065
1066 Value *CallExprAST::Codegen() {
1067   // Look up the name in the global module table.
1068   Function *CalleeF = TheModule-&gt;getFunction(Callee);
1069   if (CalleeF == 0)
1070     return ErrorV("Unknown function referenced");
1071
1072   // If argument mismatch error.
1073   if (CalleeF-&gt;arg_size() != Args.size())
1074     return ErrorV("Incorrect # arguments passed");
1075
1076   std::vector&lt;Value*&gt; ArgsV;
1077   for (unsigned i = 0, e = Args.size(); i != e; ++i) {
1078     ArgsV.push_back(Args[i]-&gt;Codegen());
1079     if (ArgsV.back() == 0) return 0;
1080   }
1081
1082   return Builder.CreateCall(CalleeF, ArgsV.begin(), ArgsV.end(), "calltmp");
1083 }
1084
1085 Function *PrototypeAST::Codegen() {
1086   // Make the function type:  double(double,double) etc.
1087   std::vector&lt;const Type*&gt; Doubles(Args.size(),
1088                                    Type::getDoubleTy(getGlobalContext()));
1089   FunctionType *FT = FunctionType::get(Type::getDoubleTy(getGlobalContext()),
1090                                        Doubles, false);
1091
1092   Function *F = Function::Create(FT, Function::ExternalLinkage, Name, TheModule);
1093
1094   // If F conflicted, there was already something named 'Name'.  If it has a
1095   // body, don't allow redefinition or reextern.
1096   if (F-&gt;getName() != Name) {
1097     // Delete the one we just made and get the existing one.
1098     F-&gt;eraseFromParent();
1099     F = TheModule-&gt;getFunction(Name);
1100
1101     // If F already has a body, reject this.
1102     if (!F-&gt;empty()) {
1103       ErrorF("redefinition of function");
1104       return 0;
1105     }
1106
1107     // If F took a different number of args, reject.
1108     if (F-&gt;arg_size() != Args.size()) {
1109       ErrorF("redefinition of function with different # args");
1110       return 0;
1111     }
1112   }
1113
1114   // Set names for all arguments.
1115   unsigned Idx = 0;
1116   for (Function::arg_iterator AI = F-&gt;arg_begin(); Idx != Args.size();
1117        ++AI, ++Idx) {
1118     AI-&gt;setName(Args[Idx]);
1119
1120     // Add arguments to variable symbol table.
1121     NamedValues[Args[Idx]] = AI;
1122   }
1123
1124   return F;
1125 }
1126
1127 Function *FunctionAST::Codegen() {
1128   NamedValues.clear();
1129
1130   Function *TheFunction = Proto-&gt;Codegen();
1131   if (TheFunction == 0)
1132     return 0;
1133
1134   // Create a new basic block to start insertion into.
1135   BasicBlock *BB = BasicBlock::Create(getGlobalContext(), "entry", TheFunction);
1136   Builder.SetInsertPoint(BB);
1137
1138   if (Value *RetVal = Body-&gt;Codegen()) {
1139     // Finish off the function.
1140     Builder.CreateRet(RetVal);
1141
1142     // Validate the generated code, checking for consistency.
1143     verifyFunction(*TheFunction);
1144     return TheFunction;
1145   }
1146
1147   // Error reading body, remove function.
1148   TheFunction-&gt;eraseFromParent();
1149   return 0;
1150 }
1151
1152 //===----------------------------------------------------------------------===//
1153 // Top-Level parsing and JIT Driver
1154 //===----------------------------------------------------------------------===//
1155
1156 static void HandleDefinition() {
1157   if (FunctionAST *F = ParseDefinition()) {
1158     if (Function *LF = F-&gt;Codegen()) {
1159       fprintf(stderr, "Read function definition:");
1160       LF-&gt;dump();
1161     }
1162   } else {
1163     // Skip token for error recovery.
1164     getNextToken();
1165   }
1166 }
1167
1168 static void HandleExtern() {
1169   if (PrototypeAST *P = ParseExtern()) {
1170     if (Function *F = P-&gt;Codegen()) {
1171       fprintf(stderr, "Read extern: ");
1172       F-&gt;dump();
1173     }
1174   } else {
1175     // Skip token for error recovery.
1176     getNextToken();
1177   }
1178 }
1179
1180 static void HandleTopLevelExpression() {
1181   // Evaluate a top level expression into an anonymous function.
1182   if (FunctionAST *F = ParseTopLevelExpr()) {
1183     if (Function *LF = F-&gt;Codegen()) {
1184       fprintf(stderr, "Read top-level expression:");
1185       LF-&gt;dump();
1186     }
1187   } else {
1188     // Skip token for error recovery.
1189     getNextToken();
1190   }
1191 }
1192
1193 /// top ::= definition | external | expression | ';'
1194 static void MainLoop() {
1195   while (1) {
1196     fprintf(stderr, "ready&gt; ");
1197     switch (CurTok) {
1198     case tok_eof:    return;
1199     case ';':        getNextToken(); break;  // ignore top level semicolons.
1200     case tok_def:    HandleDefinition(); break;
1201     case tok_extern: HandleExtern(); break;
1202     default:         HandleTopLevelExpression(); break;
1203     }
1204   }
1205 }
1206
1207
1208
1209 //===----------------------------------------------------------------------===//
1210 // "Library" functions that can be "extern'd" from user code.
1211 //===----------------------------------------------------------------------===//
1212
1213 /// putchard - putchar that takes a double and returns 0.
1214 extern "C"
1215 double putchard(double X) {
1216   putchar((char)X);
1217   return 0;
1218 }
1219
1220 //===----------------------------------------------------------------------===//
1221 // Main driver code.
1222 //===----------------------------------------------------------------------===//
1223
1224 int main() {
1225   TheModule = new Module("my cool jit", getGlobalContext());
1226
1227   // Install standard binary operators.
1228   // 1 is lowest precedence.
1229   BinopPrecedence['&lt;'] = 10;
1230   BinopPrecedence['+'] = 20;
1231   BinopPrecedence['-'] = 20;
1232   BinopPrecedence['*'] = 40;  // highest.
1233
1234   // Prime the first token.
1235   fprintf(stderr, "ready&gt; ");
1236   getNextToken();
1237
1238   MainLoop();
1239   TheModule-&gt;dump();
1240   return 0;
1241 }
1242 </pre>
1243 </div>
1244 <a href="LangImpl4.html">Next: Adding JIT and Optimizer Support</a>
1245 </div>
1246
1247 <!-- *********************************************************************** -->
1248 <hr>
1249 <address>
1250   <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1251   src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1252   <a href="http://validator.w3.org/check/referer"><img
1253   src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
1254
1255   <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1256   <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
1257   Last modified: $Date: 2009-07-21 11:05:13 -0700 (Tue, 21 Jul 2009) $
1258 </address>
1259 </body>
1260 </html>