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>
19 <li><a href="#intro">Chapter 3 Introduction</a></li>
20 <li><a href="#basics">Code Generation setup</a></li>
21 <li><a href="#exprs">Expression Code Generation</a></li>
22 <li><a href="#funcs">Function Code Generation</a></li>
23 <li><a href="#driver">Driver Changes and Closing Thoughts</a></li>
24 <li><a href="#code">Full Code Listing</a></li>
29 <div class="doc_author">
30 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
33 <!-- *********************************************************************** -->
34 <div class="doc_section"><a name="intro">Chapter 3 Introduction</a></div>
35 <!-- *********************************************************************** -->
37 <div class="doc_text">
39 <p>Welcome to Chapter 3 of the "<a href="index.html">Implementing a language
40 with LLVM</a>" tutorial. This chapter shows you how to transform the <a
41 href="LangImpl2.html">Abstract Syntax Tree built in Chapter 2</a> into LLVM IR.
42 This will teach you a little bit about how LLVM does things, as well as
43 demonstrate how easy it is to use. It's much more work to build a lexer and
44 parser than it is to generate LLVM IR code.
49 <!-- *********************************************************************** -->
50 <div class="doc_section"><a name="basics">Code Generation setup</a></div>
51 <!-- *********************************************************************** -->
53 <div class="doc_text">
56 In order to generate LLVM IR, we want some simple setup to get started. First,
57 we define virtual codegen methods in each AST class:</p>
59 <div class="doc_code">
61 /// ExprAST - Base class for all expression nodes.
65 virtual Value *Codegen() = 0;
68 /// NumberExprAST - Expression class for numeric literals like "1.0".
69 class NumberExprAST : public ExprAST {
72 explicit NumberExprAST(double val) : Val(val) {}
73 virtual Value *Codegen();
79 <p>The Codegen() method says to emit IR for that AST node and all things it
80 depends on, and they all return an LLVM Value object.
81 "Value" is the class used to represent a "<a
82 href="http://en.wikipedia.org/wiki/Static_single_assignment_form">Static Single
83 Assignment (SSA)</a> register" or "SSA value" in LLVM. The most distinct aspect
84 of SSA values is that their value is computed as the related instruction
85 executes, and it does not get a new value until (and if) the instruction
86 re-executes. In order words, there is no way to "change" an SSA value. For
87 more information, please read up on <a
88 href="http://en.wikipedia.org/wiki/Static_single_assignment_form">Static Single
89 Assignment</a> - the concepts are really quite natural once you grok them.</p>
92 second thing we want is an "Error" method like we used for parser, which will
93 be used to report errors found during code generation (for example, use of an
94 undeclared parameter):</p>
96 <div class="doc_code">
98 Value *ErrorV(const char *Str) { Error(Str); return 0; }
100 static Module *TheModule;
101 static LLVMBuilder Builder;
102 static std::map<std::string, Value*> NamedValues;
106 <p>The static variables will be used during code generation. <tt>TheModule</tt>
107 is the LLVM construct that contains all of the functions and global variables in
108 a chunk of code. In many ways, it is the top-level structure that the LLVM IR
109 uses to contain code.</p>
111 <p>The <tt>Builder</tt> object is a helper object that makes it easy to generate
112 LLVM instructions. Instances of the <a
113 href="http://llvm.org/doxygen/LLVMBuilder_8h-source.html"><tt>LLVMBuilder</tt>
114 class</a> keeps track of the current place to
115 insert instructions and has methods to create new instructions.</p>
117 <p>The <tt>NamedValues</tt> map keeps track of which values are defined in the
118 current scope and what their LLVM representation is. In this form of
119 Kaleidoscope, the only things that can be referenced are function parameters.
120 As such, function parameters will be in this map when generating code for their
124 With these basics in place, we can start talking about how to generate code for
125 each expression. Note that this assumes that the <tt>Builder</tt> has been set
126 up to generate code <em>into</em> something. For now, we'll assume that this
127 has already been done, and we'll just use it to emit code.
132 <!-- *********************************************************************** -->
133 <div class="doc_section"><a name="exprs">Expression Code Generation</a></div>
134 <!-- *********************************************************************** -->
136 <div class="doc_text">
138 <p>Generating LLVM code for expression nodes is very straight-forward: less
139 than 45 lines of commented code for all four of our expression nodes. First,
140 we'll do numeric literals:</p>
142 <div class="doc_code">
144 Value *NumberExprAST::Codegen() {
145 return ConstantFP::get(Type::DoubleTy, APFloat(Val));
150 <p>In the LLVM IR, numeric constants are represented with the
151 <tt>ConstantFP</tt> class, which holds the numeric value in an <tt>APFloat</tt>
152 internally (<tt>APFloat</tt> has the capability of holding floating point
153 constants of <em>A</em>rbitrary <em>P</em>recision). This code basically just
154 creates and returns a <tt>ConstantFP</tt>. Note that in the LLVM IR
155 that constants are all uniqued together and shared. For this reason, the API
156 uses "the foo::get(..)" idiom instead of "new foo(..)" or "foo::create(..).</p>
158 <div class="doc_code">
160 Value *VariableExprAST::Codegen() {
161 // Look this variable up in the function.
162 Value *V = NamedValues[Name];
163 return V ? V : ErrorV("Unknown variable name");
168 <p>References to variables is also quite simple here. In the simple version
169 of Kaleidoscope, we assume that the variable has already been emited somewhere
170 and its value is available. In practice, the only values that can be in the
171 <tt>NamedValues</tt> map are function arguments. This
172 code simply checks to see that the specified name is in the map (if not, an
173 unknown variable is being referenced) and returns the value for it.</p>
175 <div class="doc_code">
177 Value *BinaryExprAST::Codegen() {
178 Value *L = LHS->Codegen();
179 Value *R = RHS->Codegen();
180 if (L == 0 || R == 0) return 0;
183 case '+': return Builder.CreateAdd(L, R, "addtmp");
184 case '-': return Builder.CreateSub(L, R, "subtmp");
185 case '*': return Builder.CreateMul(L, R, "multmp");
187 L = Builder.CreateFCmpULT(L, R, "multmp");
188 // Convert bool 0/1 to double 0.0 or 1.0
189 return Builder.CreateUIToFP(L, Type::DoubleTy, "booltmp");
190 default: return ErrorV("invalid binary operator");
196 <p>Binary operators start to get more interesting. The basic idea here is that
197 we recursively emit code for the left-hand side of the expression, then the
198 right-hand side, then we compute the result of the binary expression. In this
199 code, we do a simple switch on the opcode to create the right LLVM instruction.
202 <p>In this example, the LLVM builder class is starting to show its value.
203 Because it knows where to insert the newly created instruction, you just have to
204 specificy what instruction to create (e.g. with <tt>CreateAdd</tt>), which
205 operands to use (<tt>L</tt> and <tt>R</tt> here) and optionally provide a name
206 for the generated instruction. One nice thing about LLVM is that the name is
207 just a hint: if there are multiple additions in a single function, the first
208 will be named "addtmp" and the second will be "autorenamed" by adding a suffix,
209 giving it a name like "addtmp42". Local value names for instructions are purely
210 optional, but it makes it much easier to read the IR dumps.</p>
212 <p><a href="../LangRef.html#instref">LLVM instructions</a> are constrained to
213 have very strict type properties: for example, the Left and Right operators of
214 an <a href="../LangRef.html#i_add">add instruction</a> have to have the same
215 type, and that the result of the add matches the operands. Because all values
216 in Kaleidoscope are doubles, this makes for very simple code for add, sub and
219 <p>On the other hand, LLVM specifies that the <a
220 href="../LangRef.html#i_fcmp">fcmp instruction</a> always returns an 'i1' value
221 (a one bit integer). However, Kaleidoscope wants the value to be a 0.0 or 1.0
222 value. In order to get these semantics, we combine the fcmp instruction with
223 a <a href="../LangRef.html#i_uitofp">uitofp instruction</a>. This instruction
224 converts its input integer into a floating point value by treating the input
225 as an unsigned value. In contrast, if we used the <a
226 href="../LangRef.html#i_sitofp">sitofp instruction</a>, the Kaleidoscope '<'
227 operator would return 0.0 and -1.0, depending on the input value.</p>
229 <div class="doc_code">
231 Value *CallExprAST::Codegen() {
232 // Look up the name in the global module table.
233 Function *CalleeF = TheModule->getFunction(Callee);
235 return ErrorV("Unknown function referenced");
237 // If argument mismatch error.
238 if (CalleeF->arg_size() != Args.size())
239 return ErrorV("Incorrect # arguments passed");
241 std::vector<Value*> ArgsV;
242 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
243 ArgsV.push_back(Args[i]->Codegen());
244 if (ArgsV.back() == 0) return 0;
247 return Builder.CreateCall(CalleeF, ArgsV.begin(), ArgsV.end(), "calltmp");
252 <p>Code generation for function calls is quite straight-forward with LLVM. The
253 code above first looks the name of the function up in the LLVM Module's symbol
254 table. Recall that the LLVM Module is the container that holds all of the
255 functions we are JIT'ing. By giving each function the same name as what the
256 user specifies, we can use the LLVM symbol table to resolve function names for
259 <p>Once we have the function to call, we recursively codegen each argument that
260 is to be passed in, and create an LLVM <a href="../LangRef.html#i_call">call
261 instruction</a>. Note that LLVM uses the native C calling conventions by
262 default, allowing these calls to call into standard library functions like
263 "sin" and "cos" with no additional effort.</p>
265 <p>This wraps up our handling of the four basic expressions that we have so far
266 in Kaleidoscope. Feel free to go in and add some more. For example, by
267 browsing the <a href="../LangRef.html">LLVM language reference</a> you'll find
268 several other interesting instructions that are really easy to plug into our
273 <!-- *********************************************************************** -->
274 <div class="doc_section"><a name="funcs">Function Code Generation</a></div>
275 <!-- *********************************************************************** -->
277 <div class="doc_text">
279 <p>Code generation for prototypes and functions has to handle a number of
280 details, which make their code less beautiful and elegant than expression code
281 generation, but they illustrate some important points. First, lets talk about
282 code generation for prototypes: this is used both for function bodies as well
283 as external function declarations. The code starts with:</p>
285 <div class="doc_code">
287 Function *PrototypeAST::Codegen() {
288 // Make the function type: double(double,double) etc.
289 std::vector<const Type*> Doubles(Args.size(), Type::DoubleTy);
290 FunctionType *FT = FunctionType::get(Type::DoubleTy, Doubles, false);
292 Function *F = new Function(FT, Function::ExternalLinkage, Name, TheModule);
296 <p>This code packs a lot of power into a few lines. Note first that this
297 function returns a Function* instead of a Value*. Because a "prototype" really
298 talks about the external interface for a function (not the value computed by
299 an expression), it makes sense for it to return the LLVM Function it corresponds
300 to when codegen'd.</p>
302 <p>The next step is to create
303 the <tt>FunctionType</tt> that should be used for a given Prototype. Since all
304 function arguments in Kaleidoscope are of type double, the first line creates
305 a vector of "N" LLVM Double types. It then uses the <tt>FunctionType::get</tt>
306 method to create a function type that takes "N" doubles as arguments, returns
307 one double as a result, and that is not vararg (the false parameter indicates
308 this). Note that Types in LLVM are uniqued just like Constants are, so you
309 don't "new" a type, you "get" it.</p>
311 <p>The final line above actually creates the function that the prototype will
312 correspond to. This indicates which type, linkage, and name to use, and which
313 module to insert into. "<a href="LangRef.html#linkage">external linkage</a>"
314 means that the function may be defined outside the current module and/or that it
315 is callable by functions outside the module. The Name passed in is the name the
316 user specified: since "<tt>TheModule</tt>" is specified, this name is registered
317 in "<tt>TheModule</tt>"s symbol table, which is used by the function call code
320 <div class="doc_code">
322 // If F conflicted, there was already something named 'Name'. If it has a
323 // body, don't allow redefinition or reextern.
324 if (F->getName() != Name) {
325 // Delete the one we just made and get the existing one.
326 F->eraseFromParent();
327 F = TheModule->getFunction(Name);
331 <p>The Module symbol table works just like the Function symbol table when it
332 comes to name conflicts: if a new function is created with a name was previously
333 added to the symbol table, it will get implicitly renamed when added to the
334 Module. The code above exploits this fact to tell if there was a previous
335 definition of this function.</p>
337 <p>In Kaleidoscope, I choose to allow redefinitions of functions in two cases:
338 first, we want to allow 'extern'ing a function more than once, so long as the
339 prototypes for the externs match (since all arguments have the same type, we
340 just have to check that the number of arguments match). Second, we want to
341 allow 'extern'ing a function and then definining a body for it. This is useful
342 when defining mutually recursive functions.</p>
344 <p>In order to implement this, the code above first checks to see if there is
345 a collision on the name of the function. If so, it deletes the function we just
346 created (by calling <tt>eraseFromParent</tt>) and then calling
347 <tt>getFunction</tt> to get the existing function with the specified name. Note
348 that many APIs in LLVM have "erase" forms and "remove" forms. The "remove" form
349 unlinks the object from its parent (e.g. a Function from a Module) and returns
350 it. The "erase" form unlinks the object and then deletes it.</p>
352 <div class="doc_code">
354 // If F already has a body, reject this.
355 if (!F->empty()) {
356 ErrorF("redefinition of function");
360 // If F took a different number of args, reject.
361 if (F->arg_size() != Args.size()) {
362 ErrorF("redefinition of function with different # args");
369 <p>In order to verify the logic above, we first check to see if the preexisting
370 function is "empty". In this case, empty means that it has no basic blocks in
371 it, which means it has no body. If it has no body, this means its a forward
372 declaration. Since we don't allow anything after a full definition of the
373 function, the code rejects this case. If the previous reference to a function
374 was an 'extern', we simply verify that the number of arguments for that
375 definition and this one match up. If not, we emit an error.</p>
377 <div class="doc_code">
379 // Set names for all arguments.
381 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
383 AI->setName(Args[Idx]);
385 // Add arguments to variable symbol table.
386 NamedValues[Args[Idx]] = AI;
393 <p>The last bit of code for prototypes loops over all of the arguments in the
394 function, setting the name of the LLVM Argument objects to match and registering
395 the arguments in the <tt>NamedValues</tt> map for future use by the
396 <tt>VariableExprAST</tt> AST node. Once this is set up, it returns the Function
397 object to the caller. Note that we don't check for conflicting
398 argument names here (e.g. "extern foo(a b a)"). Doing so would be very
399 straight-forward.</p>
401 <div class="doc_code">
403 Function *FunctionAST::Codegen() {
406 Function *TheFunction = Proto->Codegen();
407 if (TheFunction == 0)
412 <p>Code generation for function definitions starts out simply enough: first we
413 codegen the prototype and verify that it is ok. We also clear out the
414 <tt>NamedValues</tt> map to make sure that there isn't anything in it from the
415 last function we compiled.</p>
417 <div class="doc_code">
419 // Create a new basic block to start insertion into.
420 BasicBlock *BB = new BasicBlock("entry", TheFunction);
421 Builder.SetInsertPoint(BB);
423 if (Value *RetVal = Body->Codegen()) {
427 <p>Now we get to the point where the <tt>Builder</tt> is set up. The first
428 line creates a new <a href="http://en.wikipedia.org/wiki/Basic_block">basic
429 block</a> (named "entry"), which is inserted into <tt>TheFunction</tt>. The
430 second line then tells the builder that new instructions should be inserted into
431 the end of the new basic block. Basic blocks in LLVM are an important part
432 of functions that define the <a
433 href="http://en.wikipedia.org/wiki/Control_flow_graph">Control Flow Graph</a>.
434 Since we don't have any control flow, our functions will only contain one
435 block so far. We'll fix this in a future installment :).</p>
437 <div class="doc_code">
439 if (Value *RetVal = Body->Codegen()) {
440 // Finish off the function.
441 Builder.CreateRet(RetVal);
443 // Validate the generated code, checking for consistency.
444 verifyFunction(*TheFunction);
450 <p>Once the insertion point is set up, we call the <tt>CodeGen()</tt> method for
451 the root expression of the function. If no error happens, this emits code to
452 compute the expression into the entry block and returns the value that was
453 computed. Assuming no error, we then create an LLVM <a
454 href="../LangRef.html#i_ret">ret instruction</a>, which completes the function.
455 Once the function is built, we call the <tt>verifyFunction</tt> function, which
456 is provided by LLVM. This function does a variety of consistency checks on the
457 generated code, to determine if our compiler is doing everything right. Using
458 this is important: it can catch a lot of bugs. Once the function is finished
459 and validated, we return it.</p>
461 <div class="doc_code">
463 // Error reading body, remove function.
464 TheFunction->eraseFromParent();
470 <p>The only piece left here is handling of the error case. For simplicity, we
471 simply handle this by deleting the function we produced with the
472 <tt>eraseFromParent</tt> method. This allows the user to redefine a function
473 that they incorrectly typed in before: if we didn't delete it, it would live in
474 the symbol table, with a body, preventing future redefinition.</p>
476 <p>This code does have a bug though. Since the <tt>PrototypeAST::Codegen</tt>
477 can return a previously defined forward declaration, this can actually delete
478 a forward declaration. There are a number of ways to fix this bug, see what you
479 can come up with! Here is a testcase:</p>
481 <div class="doc_code">
483 extern foo(a b); # ok, defines foo.
484 def foo(a b) c; # error, 'c' is invalid.
485 def bar() foo(1, 2); # error, unknown function "foo"
491 <!-- *********************************************************************** -->
492 <div class="doc_section"><a name="driver">Driver Changes and
493 Closing Thoughts</a></div>
494 <!-- *********************************************************************** -->
496 <div class="doc_text">
499 For now, code generation to LLVM doesn't really get us much, except that we can
500 look at the pretty IR calls. The sample code inserts calls to Codegen into the
501 "<tt>HandleDefinition</tt>", "<tt>HandleExtern</tt>" etc functions, and then
502 dumps out the LLVM IR. This gives a nice way to look at the LLVM IR for simple
503 functions. For example:
506 <div class="doc_code">
509 ready> Read top-level expression:
510 define double @""() {
512 %addtmp = add double 4.000000e+00, 5.000000e+00
518 <p>Note how the parser turns the top-level expression into anonymous functions
519 for us. This will be handy when we add JIT support in the next chapter. Also
520 note that the code is very literally transcribed, no optimizations are being
521 performed. We will add optimizations explicitly in the next chapter.</p>
523 <div class="doc_code">
525 ready> <b>def foo(a b) a*a + 2*a*b + b*b;</b>
526 ready> Read function definition:
527 define double @foo(double %a, double %b) {
529 %multmp = mul double %a, %a
530 %multmp1 = mul double 2.000000e+00, %a
531 %multmp2 = mul double %multmp1, %b
532 %addtmp = add double %multmp, %multmp2
533 %multmp3 = mul double %b, %b
534 %addtmp4 = add double %addtmp, %multmp3
540 <p>This shows some simple arithmetic. Notice the striking similarity to the
541 LLVM builder calls that we use to create the instructions.</p>
543 <div class="doc_code">
545 ready> <b>def bar(a) foo(a, 4.0) + bar(31337);</b>
546 ready> Read function definition:
547 define double @bar(double %a) {
549 %calltmp = call double @foo( double %a, double 4.000000e+00 )
550 %calltmp1 = call double @bar( double 3.133700e+04 )
551 %addtmp = add double %calltmp, %calltmp1
557 <p>This shows some function calls. Note that this function will take a long
558 time to execute if you call it. In the future we'll add conditional control
559 flow to make recursion actually be useful :).</p>
561 <div class="doc_code">
563 ready> <b>extern cos(x);</b>
564 ready> Read extern:
565 declare double @cos(double)
567 ready> <b>cos(1.234);</b>
568 ready> Read top-level expression:
569 define double @""() {
571 %calltmp = call double @cos( double 1.234000e+00 )
577 <p>This shows an extern for the libm "cos" function, and a call to it.</p>
580 <div class="doc_code">
583 ; ModuleID = 'my cool jit'
585 define double @""() {
587 %addtmp = add double 4.000000e+00, 5.000000e+00
591 define double @foo(double %a, double %b) {
593 %multmp = mul double %a, %a
594 %multmp1 = mul double 2.000000e+00, %a
595 %multmp2 = mul double %multmp1, %b
596 %addtmp = add double %multmp, %multmp2
597 %multmp3 = mul double %b, %b
598 %addtmp4 = add double %addtmp, %multmp3
602 define double @bar(double %a) {
604 %calltmp = call double @foo( double %a, double 4.000000e+00 )
605 %calltmp1 = call double @bar( double 3.133700e+04 )
606 %addtmp = add double %calltmp, %calltmp1
610 declare double @cos(double)
612 define double @""() {
614 %calltmp = call double @cos( double 1.234000e+00 )
620 <p>When you quit the current demo, it dumps out the IR for the entire module
621 generated. Here you can see the big picture with all the functions referencing
624 <p>This wraps up this chapter of the Kaleidoscope tutorial. Up next we'll
625 describe how to <a href="LangImpl4.html">add JIT codegen and optimizer
626 support</a> to this so we can actually start running code!</p>
631 <!-- *********************************************************************** -->
632 <div class="doc_section"><a name="code">Full Code Listing</a></div>
633 <!-- *********************************************************************** -->
635 <div class="doc_text">
638 Here is the complete code listing for our running example, enhanced with the
639 LLVM code generator. Because this uses the LLVM libraries, we need to link
640 them in. To do this, we use the <a
641 href="http://llvm.org/cmds/llvm-config.html">llvm-config</a> tool to inform
642 our makefile/command line about which options to use:</p>
644 <div class="doc_code">
647 g++ -g toy.cpp `llvm-config --cppflags --ldflags --libs core` -o toy
653 <p>Here is the code:</p>
655 <div class="doc_code">
658 // See example below.
660 #include "llvm/DerivedTypes.h"
661 #include "llvm/Module.h"
662 #include "llvm/Analysis/Verifier.h"
663 #include "llvm/Support/LLVMBuilder.h"
664 #include <cstdio>
665 #include <string>
667 #include <vector>
668 using namespace llvm;
670 //===----------------------------------------------------------------------===//
672 //===----------------------------------------------------------------------===//
674 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
675 // of these for known things.
680 tok_def = -2, tok_extern = -3,
683 tok_identifier = -4, tok_number = -5,
686 static std::string IdentifierStr; // Filled in if tok_identifier
687 static double NumVal; // Filled in if tok_number
689 /// gettok - Return the next token from standard input.
690 static int gettok() {
691 static int LastChar = ' ';
693 // Skip any whitespace.
694 while (isspace(LastChar))
695 LastChar = getchar();
697 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
698 IdentifierStr = LastChar;
699 while (isalnum((LastChar = getchar())))
700 IdentifierStr += LastChar;
702 if (IdentifierStr == "def") return tok_def;
703 if (IdentifierStr == "extern") return tok_extern;
704 return tok_identifier;
707 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+
711 LastChar = getchar();
712 } while (isdigit(LastChar) || LastChar == '.');
714 NumVal = strtod(NumStr.c_str(), 0);
718 if (LastChar == '#') {
719 // Comment until end of line.
720 do LastChar = getchar();
721 while (LastChar != EOF && LastChar != '\n' & LastChar != '\r');
727 // Check for end of file. Don't eat the EOF.
731 // Otherwise, just return the character as its ascii value.
732 int ThisChar = LastChar;
733 LastChar = getchar();
737 //===----------------------------------------------------------------------===//
738 // Abstract Syntax Tree (aka Parse Tree)
739 //===----------------------------------------------------------------------===//
741 /// ExprAST - Base class for all expression nodes.
744 virtual ~ExprAST() {}
745 virtual Value *Codegen() = 0;
748 /// NumberExprAST - Expression class for numeric literals like "1.0".
749 class NumberExprAST : public ExprAST {
752 explicit NumberExprAST(double val) : Val(val) {}
753 virtual Value *Codegen();
756 /// VariableExprAST - Expression class for referencing a variable, like "a".
757 class VariableExprAST : public ExprAST {
760 explicit VariableExprAST(const std::string &name) : Name(name) {}
761 virtual Value *Codegen();
764 /// BinaryExprAST - Expression class for a binary operator.
765 class BinaryExprAST : public ExprAST {
769 BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs)
770 : Op(op), LHS(lhs), RHS(rhs) {}
771 virtual Value *Codegen();
774 /// CallExprAST - Expression class for function calls.
775 class CallExprAST : public ExprAST {
777 std::vector<ExprAST*> Args;
779 CallExprAST(const std::string &callee, std::vector<ExprAST*> &args)
780 : Callee(callee), Args(args) {}
781 virtual Value *Codegen();
784 /// PrototypeAST - This class represents the "prototype" for a function,
785 /// which captures its argument names as well as if it is an operator.
788 std::vector<std::string> Args;
790 PrototypeAST(const std::string &name, const std::vector<std::string> &args)
791 : Name(name), Args(args) {}
796 /// FunctionAST - This class represents a function definition itself.
801 FunctionAST(PrototypeAST *proto, ExprAST *body)
802 : Proto(proto), Body(body) {}
807 //===----------------------------------------------------------------------===//
809 //===----------------------------------------------------------------------===//
811 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
812 /// token the parser it looking at. getNextToken reads another token from the
813 /// lexer and updates CurTok with its results.
815 static int getNextToken() {
816 return CurTok = gettok();
819 /// BinopPrecedence - This holds the precedence for each binary operator that is
821 static std::map<char, int> BinopPrecedence;
823 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
824 static int GetTokPrecedence() {
825 if (!isascii(CurTok))
828 // Make sure it's a declared binop.
829 int TokPrec = BinopPrecedence[CurTok];
830 if (TokPrec <= 0) return -1;
834 /// Error* - These are little helper functions for error handling.
835 ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;}
836 PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; }
837 FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; }
839 static ExprAST *ParseExpression();
843 /// ::= identifier '(' expression* ')'
844 static ExprAST *ParseIdentifierExpr() {
845 std::string IdName = IdentifierStr;
847 getNextToken(); // eat identifier.
849 if (CurTok != '(') // Simple variable ref.
850 return new VariableExprAST(IdName);
853 getNextToken(); // eat (
854 std::vector<ExprAST*> Args;
856 ExprAST *Arg = ParseExpression();
860 if (CurTok == ')') break;
863 return Error("Expected ')'");
870 return new CallExprAST(IdName, Args);
873 /// numberexpr ::= number
874 static ExprAST *ParseNumberExpr() {
875 ExprAST *Result = new NumberExprAST(NumVal);
876 getNextToken(); // consume the number
880 /// parenexpr ::= '(' expression ')'
881 static ExprAST *ParseParenExpr() {
882 getNextToken(); // eat (.
883 ExprAST *V = ParseExpression();
887 return Error("expected ')'");
888 getNextToken(); // eat ).
893 /// ::= identifierexpr
896 static ExprAST *ParsePrimary() {
898 default: return Error("unknown token when expecting an expression");
899 case tok_identifier: return ParseIdentifierExpr();
900 case tok_number: return ParseNumberExpr();
901 case '(': return ParseParenExpr();
906 /// ::= ('+' primary)*
907 static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) {
908 // If this is a binop, find its precedence.
910 int TokPrec = GetTokPrecedence();
912 // If this is a binop that binds at least as tightly as the current binop,
913 // consume it, otherwise we are done.
914 if (TokPrec < ExprPrec)
917 // Okay, we know this is a binop.
919 getNextToken(); // eat binop
921 // Parse the primary expression after the binary operator.
922 ExprAST *RHS = ParsePrimary();
925 // If BinOp binds less tightly with RHS than the operator after RHS, let
926 // the pending operator take RHS as its LHS.
927 int NextPrec = GetTokPrecedence();
928 if (TokPrec < NextPrec) {
929 RHS = ParseBinOpRHS(TokPrec+1, RHS);
930 if (RHS == 0) return 0;
934 LHS = new BinaryExprAST(BinOp, LHS, RHS);
939 /// ::= primary binoprhs
941 static ExprAST *ParseExpression() {
942 ExprAST *LHS = ParsePrimary();
945 return ParseBinOpRHS(0, LHS);
949 /// ::= id '(' id* ')'
950 static PrototypeAST *ParsePrototype() {
951 if (CurTok != tok_identifier)
952 return ErrorP("Expected function name in prototype");
954 std::string FnName = IdentifierStr;
958 return ErrorP("Expected '(' in prototype");
960 std::vector<std::string> ArgNames;
961 while (getNextToken() == tok_identifier)
962 ArgNames.push_back(IdentifierStr);
964 return ErrorP("Expected ')' in prototype");
967 getNextToken(); // eat ')'.
969 return new PrototypeAST(FnName, ArgNames);
972 /// definition ::= 'def' prototype expression
973 static FunctionAST *ParseDefinition() {
974 getNextToken(); // eat def.
975 PrototypeAST *Proto = ParsePrototype();
976 if (Proto == 0) return 0;
978 if (ExprAST *E = ParseExpression())
979 return new FunctionAST(Proto, E);
983 /// toplevelexpr ::= expression
984 static FunctionAST *ParseTopLevelExpr() {
985 if (ExprAST *E = ParseExpression()) {
986 // Make an anonymous proto.
987 PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>());
988 return new FunctionAST(Proto, E);
993 /// external ::= 'extern' prototype
994 static PrototypeAST *ParseExtern() {
995 getNextToken(); // eat extern.
996 return ParsePrototype();
999 //===----------------------------------------------------------------------===//
1001 //===----------------------------------------------------------------------===//
1003 static Module *TheModule;
1004 static LLVMBuilder Builder;
1005 static std::map<std::string, Value*> NamedValues;
1007 Value *ErrorV(const char *Str) { Error(Str); return 0; }
1009 Value *NumberExprAST::Codegen() {
1010 return ConstantFP::get(Type::DoubleTy, APFloat(Val));
1013 Value *VariableExprAST::Codegen() {
1014 // Look this variable up in the function.
1015 Value *V = NamedValues[Name];
1016 return V ? V : ErrorV("Unknown variable name");
1019 Value *BinaryExprAST::Codegen() {
1020 Value *L = LHS->Codegen();
1021 Value *R = RHS->Codegen();
1022 if (L == 0 || R == 0) return 0;
1025 case '+': return Builder.CreateAdd(L, R, "addtmp");
1026 case '-': return Builder.CreateSub(L, R, "subtmp");
1027 case '*': return Builder.CreateMul(L, R, "multmp");
1029 L = Builder.CreateFCmpULT(L, R, "multmp");
1030 // Convert bool 0/1 to double 0.0 or 1.0
1031 return Builder.CreateUIToFP(L, Type::DoubleTy, "booltmp");
1032 default: return ErrorV("invalid binary operator");
1036 Value *CallExprAST::Codegen() {
1037 // Look up the name in the global module table.
1038 Function *CalleeF = TheModule->getFunction(Callee);
1040 return ErrorV("Unknown function referenced");
1042 // If argument mismatch error.
1043 if (CalleeF->arg_size() != Args.size())
1044 return ErrorV("Incorrect # arguments passed");
1046 std::vector<Value*> ArgsV;
1047 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
1048 ArgsV.push_back(Args[i]->Codegen());
1049 if (ArgsV.back() == 0) return 0;
1052 return Builder.CreateCall(CalleeF, ArgsV.begin(), ArgsV.end(), "calltmp");
1055 Function *PrototypeAST::Codegen() {
1056 // Make the function type: double(double,double) etc.
1057 std::vector<const Type*> Doubles(Args.size(), Type::DoubleTy);
1058 FunctionType *FT = FunctionType::get(Type::DoubleTy, Doubles, false);
1060 Function *F = new Function(FT, Function::ExternalLinkage, Name, TheModule);
1062 // If F conflicted, there was already something named 'Name'. If it has a
1063 // body, don't allow redefinition or reextern.
1064 if (F->getName() != Name) {
1065 // Delete the one we just made and get the existing one.
1066 F->eraseFromParent();
1067 F = TheModule->getFunction(Name);
1069 // If F already has a body, reject this.
1070 if (!F->empty()) {
1071 ErrorF("redefinition of function");
1075 // If F took a different number of args, reject.
1076 if (F->arg_size() != Args.size()) {
1077 ErrorF("redefinition of function with different # args");
1082 // Set names for all arguments.
1084 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
1086 AI->setName(Args[Idx]);
1088 // Add arguments to variable symbol table.
1089 NamedValues[Args[Idx]] = AI;
1095 Function *FunctionAST::Codegen() {
1096 NamedValues.clear();
1098 Function *TheFunction = Proto->Codegen();
1099 if (TheFunction == 0)
1102 // Create a new basic block to start insertion into.
1103 BasicBlock *BB = new BasicBlock("entry", TheFunction);
1104 Builder.SetInsertPoint(BB);
1106 if (Value *RetVal = Body->Codegen()) {
1107 // Finish off the function.
1108 Builder.CreateRet(RetVal);
1110 // Validate the generated code, checking for consistency.
1111 verifyFunction(*TheFunction);
1115 // Error reading body, remove function.
1116 TheFunction->eraseFromParent();
1120 //===----------------------------------------------------------------------===//
1121 // Top-Level parsing and JIT Driver
1122 //===----------------------------------------------------------------------===//
1124 static void HandleDefinition() {
1125 if (FunctionAST *F = ParseDefinition()) {
1126 if (Function *LF = F->Codegen()) {
1127 fprintf(stderr, "Read function definition:");
1131 // Skip token for error recovery.
1136 static void HandleExtern() {
1137 if (PrototypeAST *P = ParseExtern()) {
1138 if (Function *F = P->Codegen()) {
1139 fprintf(stderr, "Read extern: ");
1143 // Skip token for error recovery.
1148 static void HandleTopLevelExpression() {
1149 // Evaluate a top level expression into an anonymous function.
1150 if (FunctionAST *F = ParseTopLevelExpr()) {
1151 if (Function *LF = F->Codegen()) {
1152 fprintf(stderr, "Read top-level expression:");
1156 // Skip token for error recovery.
1161 /// top ::= definition | external | expression | ';'
1162 static void MainLoop() {
1164 fprintf(stderr, "ready> ");
1166 case tok_eof: return;
1167 case ';': getNextToken(); break; // ignore top level semicolons.
1168 case tok_def: HandleDefinition(); break;
1169 case tok_extern: HandleExtern(); break;
1170 default: HandleTopLevelExpression(); break;
1177 //===----------------------------------------------------------------------===//
1178 // "Library" functions that can be "extern'd" from user code.
1179 //===----------------------------------------------------------------------===//
1181 /// putchard - putchar that takes a double and returns 0.
1183 double putchard(double X) {
1188 //===----------------------------------------------------------------------===//
1189 // Main driver code.
1190 //===----------------------------------------------------------------------===//
1193 TheModule = new Module("my cool jit");
1195 // Install standard binary operators.
1196 // 1 is lowest precedence.
1197 BinopPrecedence['<'] = 10;
1198 BinopPrecedence['+'] = 20;
1199 BinopPrecedence['-'] = 20;
1200 BinopPrecedence['*'] = 40; // highest.
1202 // Prime the first token.
1203 fprintf(stderr, "ready> ");
1207 TheModule->dump();
1214 <!-- *********************************************************************** -->
1217 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1218 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1219 <a href="http://validator.w3.org/check/referer"><img
1220 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
1222 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1223 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
1224 Last modified: $Date: 2007-10-17 11:05:13 -0700 (Wed, 17 Oct 2007) $