1 <!DOCTYPE HTML PUBLIC
"-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
6 <title>Kaleidoscope: Adding JIT and Optimizer Support
</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: Adding JIT and Optimizer Support
</div>
17 <li><a href=
"index.html">Up to Tutorial Index
</a></li>
20 <li><a href=
"#intro">Chapter
4 Introduction
</a></li>
21 <li><a href=
"#trivialconstfold">Trivial Constant Folding
</a></li>
22 <li><a href=
"#optimizerpasses">LLVM Optimization Passes
</a></li>
23 <li><a href=
"#jit">Adding a JIT Compiler
</a></li>
24 <li><a href=
"#code">Full Code Listing
</a></li>
27 <li><a href=
"LangImpl5.html">Chapter
5</a>: Extending the Language: Control
31 <div class=
"doc_author">
32 <p>Written by
<a href=
"mailto:sabre@nondot.org">Chris Lattner
</a></p>
35 <!-- *********************************************************************** -->
36 <div class=
"doc_section"><a name=
"intro">Chapter
4 Introduction
</a></div>
37 <!-- *********************************************************************** -->
39 <div class=
"doc_text">
41 <p>Welcome to Chapter
4 of the
"<a href="index.html
">Implementing a language
42 with LLVM</a>" tutorial. Chapters
1-
3 described the implementation of a simple
43 language and added support for generating LLVM IR. This chapter describes
44 two new techniques: adding optimizer support to your language, and adding JIT
45 compiler support. These additions will demonstrate how to get nice, efficient code
46 for the Kaleidoscope language.
</p>
50 <!-- *********************************************************************** -->
51 <div class=
"doc_section"><a name=
"trivialconstfold">Trivial Constant
53 <!-- *********************************************************************** -->
55 <div class=
"doc_text">
58 Our demonstration for Chapter
3 is elegant and easy to extend. Unfortunately,
59 it does not produce wonderful code. The IRBuilder, however, does give us
60 obvious optimizations when compiling simple code:
</p>
62 <div class=
"doc_code">
64 ready
> <b>def test(x)
1+
2+x;
</b>
65 Read function definition:
66 define double @test(double %x) {
68 %addtmp = add double
3.000000e+00, %x
74 <p>This code is not a literal transcription of the AST built by parsing the
77 <div class=
"doc_code">
79 ready
> <b>def test(x)
1+
2+x;
</b>
80 Read function definition:
81 define double @test(double %x) {
83 %addtmp = add double
2.000000e+00,
1.000000e+00
84 %addtmp1 = add double %addtmp, %x
90 <p>Constant folding, as seen above, in particular, is a very common and very
91 important optimization: so much so that many language implementors implement
92 constant folding support in their AST representation.
</p>
94 <p>With LLVM, you don't need this support in the AST. Since all calls to build
95 LLVM IR go through the LLVM IR builder, the builder itself checked to see if
96 there was a constant folding opportunity when you call it. If so, it just does
97 the constant fold and return the constant instead of creating an instruction.
99 <p>Well, that was easy :). In practice, we recommend always using
100 <tt>IRBuilder
</tt> when generating code like this. It has no
101 "syntactic overhead" for its use (you don't have to uglify your compiler with
102 constant checks everywhere) and it can dramatically reduce the amount of
103 LLVM IR that is generated in some cases (particular for languages with a macro
104 preprocessor or that use a lot of constants).
</p>
106 <p>On the other hand, the
<tt>IRBuilder
</tt> is limited by the fact
107 that it does all of its analysis inline with the code as it is built. If you
108 take a slightly more complex example:
</p>
110 <div class=
"doc_code">
112 ready
> <b>def test(x) (
1+
2+x)*(x+(
1+
2));
</b>
113 ready
> Read function definition:
114 define double @test(double %x) {
116 %addtmp = add double
3.000000e+00, %x
117 %addtmp1 = add double %x,
3.000000e+00
118 %multmp = mul double %addtmp, %addtmp1
124 <p>In this case, the LHS and RHS of the multiplication are the same value. We'd
125 really like to see this generate
"<tt>tmp = x+3; result = tmp*tmp;</tt>" instead
126 of computing
"<tt>x+3</tt>" twice.
</p>
128 <p>Unfortunately, no amount of local analysis will be able to detect and correct
129 this. This requires two transformations: reassociation of expressions (to
130 make the add's lexically identical) and Common Subexpression Elimination (CSE)
131 to delete the redundant add instruction. Fortunately, LLVM provides a broad
132 range of optimizations that you can use, in the form of
"passes".
</p>
136 <!-- *********************************************************************** -->
137 <div class=
"doc_section"><a name=
"optimizerpasses">LLVM Optimization
139 <!-- *********************************************************************** -->
141 <div class=
"doc_text">
143 <p>LLVM provides many optimization passes, which do many different sorts of
144 things and have different tradeoffs. Unlike other systems, LLVM doesn't hold
145 to the mistaken notion that one set of optimizations is right for all languages
146 and for all situations. LLVM allows a compiler implementor to make complete
147 decisions about what optimizations to use, in which order, and in what
150 <p>As a concrete example, LLVM supports both
"whole module" passes, which look
151 across as large of body of code as they can (often a whole file, but if run
152 at link time, this can be a substantial portion of the whole program). It also
153 supports and includes
"per-function" passes which just operate on a single
154 function at a time, without looking at other functions. For more information
155 on passes and how they are run, see the
<a href=
"../WritingAnLLVMPass.html">How
156 to Write a Pass
</a> document and the
<a href=
"../Passes.html">List of LLVM
159 <p>For Kaleidoscope, we are currently generating functions on the fly, one at
160 a time, as the user types them in. We aren't shooting for the ultimate
161 optimization experience in this setting, but we also want to catch the easy and
162 quick stuff where possible. As such, we will choose to run a few per-function
163 optimizations as the user types the function in. If we wanted to make a
"static
164 Kaleidoscope compiler", we would use exactly the code we have now, except that
165 we would defer running the optimizer until the entire file has been parsed.
</p>
167 <p>In order to get per-function optimizations going, we need to set up a
168 <a href=
"../WritingAnLLVMPass.html#passmanager">FunctionPassManager
</a> to hold and
169 organize the LLVM optimizations that we want to run. Once we have that, we can
170 add a set of optimizations to run. The code looks like this:
</p>
172 <div class=
"doc_code">
174 ExistingModuleProvider *OurModuleProvider =
175 new ExistingModuleProvider(TheModule);
177 FunctionPassManager OurFPM(OurModuleProvider);
179 // Set up the optimizer pipeline. Start with registering info about how the
180 // target lays out data structures.
181 OurFPM.add(new TargetData(*TheExecutionEngine-
>getTargetData()));
182 // Do simple
"peephole" optimizations and bit-twiddling optzns.
183 OurFPM.add(createInstructionCombiningPass());
184 // Reassociate expressions.
185 OurFPM.add(createReassociatePass());
186 // Eliminate Common SubExpressions.
187 OurFPM.add(createGVNPass());
188 // Simplify the control flow graph (deleting unreachable blocks, etc).
189 OurFPM.add(createCFGSimplificationPass());
191 // Set the global so the code gen can use this.
192 TheFPM =
&OurFPM;
194 // Run the main
"interpreter loop" now.
199 <p>This code defines two objects, an
<tt>ExistingModuleProvider
</tt> and a
200 <tt>FunctionPassManager
</tt>. The former is basically a wrapper around our
201 <tt>Module
</tt> that the PassManager requires. It provides certain flexibility
202 that we're not going to take advantage of here, so I won't dive into any details
205 <p>The meat of the matter here, is the definition of
"<tt>OurFPM</tt>". It
206 requires a pointer to the
<tt>Module
</tt> (through the
<tt>ModuleProvider
</tt>)
207 to construct itself. Once it is set up, we use a series of
"add" calls to add
208 a bunch of LLVM passes. The first pass is basically boilerplate, it adds a pass
209 so that later optimizations know how the data structures in the program are
210 layed out. The
"<tt>TheExecutionEngine</tt>" variable is related to the JIT,
211 which we will get to in the next section.
</p>
213 <p>In this case, we choose to add
4 optimization passes. The passes we chose
214 here are a pretty standard set of
"cleanup" optimizations that are useful for
215 a wide variety of code. I won't delve into what they do but, believe me,
216 they are a good starting place :).
</p>
218 <p>Once the PassManager is set up, we need to make use of it. We do this by
219 running it after our newly created function is constructed (in
220 <tt>FunctionAST::Codegen
</tt>), but before it is returned to the client:
</p>
222 <div class=
"doc_code">
224 if (Value *RetVal = Body-
>Codegen()) {
225 // Finish off the function.
226 Builder.CreateRet(RetVal);
228 // Validate the generated code, checking for consistency.
229 verifyFunction(*TheFunction);
231 <b>// Optimize the function.
232 TheFPM-
>run(*TheFunction);
</b>
239 <p>As you can see, this is pretty straightforward. The
240 <tt>FunctionPassManager
</tt> optimizes and updates the LLVM Function* in place,
241 improving (hopefully) its body. With this in place, we can try our test above
244 <div class=
"doc_code">
246 ready
> <b>def test(x) (
1+
2+x)*(x+(
1+
2));
</b>
247 ready
> Read function definition:
248 define double @test(double %x) {
250 %addtmp = add double %x,
3.000000e+00
251 %multmp = mul double %addtmp, %addtmp
257 <p>As expected, we now get our nicely optimized code, saving a floating point
258 add instruction from every execution of this function.
</p>
260 <p>LLVM provides a wide variety of optimizations that can be used in certain
261 circumstances. Some
<a href=
"../Passes.html">documentation about the various
262 passes
</a> is available, but it isn't very complete. Another good source of
263 ideas can come from looking at the passes that
<tt>llvm-gcc
</tt> or
264 <tt>llvm-ld
</tt> run to get started. The
"<tt>opt</tt>" tool allows you to
265 experiment with passes from the command line, so you can see if they do
268 <p>Now that we have reasonable code coming out of our front-end, lets talk about
273 <!-- *********************************************************************** -->
274 <div class=
"doc_section"><a name=
"jit">Adding a JIT Compiler
</a></div>
275 <!-- *********************************************************************** -->
277 <div class=
"doc_text">
279 <p>Code that is available in LLVM IR can have a wide variety of tools
280 applied to it. For example, you can run optimizations on it (as we did above),
281 you can dump it out in textual or binary forms, you can compile the code to an
282 assembly file (.s) for some target, or you can JIT compile it. The nice thing
283 about the LLVM IR representation is that it is the
"common currency" between
284 many different parts of the compiler.
287 <p>In this section, we'll add JIT compiler support to our interpreter. The
288 basic idea that we want for Kaleidoscope is to have the user enter function
289 bodies as they do now, but immediately evaluate the top-level expressions they
290 type in. For example, if they type in
"1 + 2;", we should evaluate and print
291 out
3. If they define a function, they should be able to call it from the
294 <p>In order to do this, we first declare and initialize the JIT. This is done
295 by adding a global variable and a call in
<tt>main
</tt>:
</p>
297 <div class=
"doc_code">
299 <b>static ExecutionEngine *TheExecutionEngine;
</b>
303 <b>// Create the JIT. This takes ownership of the module and module provider.
304 TheExecutionEngine = EngineBuilder(OurModuleProvider).create();
</b>
310 <p>This creates an abstract
"Execution Engine" which can be either a JIT
311 compiler or the LLVM interpreter. LLVM will automatically pick a JIT compiler
312 for you if one is available for your platform, otherwise it will fall back to
315 <p>Once the
<tt>ExecutionEngine
</tt> is created, the JIT is ready to be used.
316 There are a variety of APIs that are useful, but the simplest one is the
317 "<tt>getPointerToFunction(F)</tt>" method. This method JIT compiles the
318 specified LLVM Function and returns a function pointer to the generated machine
319 code. In our case, this means that we can change the code that parses a
320 top-level expression to look like this:
</p>
322 <div class=
"doc_code">
324 static void HandleTopLevelExpression() {
325 // Evaluate a top level expression into an anonymous function.
326 if (FunctionAST *F = ParseTopLevelExpr()) {
327 if (Function *LF = F-
>Codegen()) {
328 LF-
>dump(); // Dump the function for exposition purposes.
330 <b>// JIT the function, returning a function pointer.
331 void *FPtr = TheExecutionEngine-
>getPointerToFunction(LF);
333 // Cast it to the right type (takes no arguments, returns a double) so we
334 // can call it as a native function.
335 double (*FP)() = (double (*)())FPtr;
336 fprintf(stderr,
"Evaluated to %f\n", FP());
</b>
341 <p>Recall that we compile top-level expressions into a self-contained LLVM
342 function that takes no arguments and returns the computed double. Because the
343 LLVM JIT compiler matches the native platform ABI, this means that you can just
344 cast the result pointer to a function pointer of that type and call it directly.
345 This means, there is no difference between JIT compiled code and native machine
346 code that is statically linked into your application.
</p>
348 <p>With just these two changes, lets see how Kaleidoscope works now!
</p>
350 <div class=
"doc_code">
352 ready
> <b>4+
5;
</b>
353 define double @
""() {
355 ret double
9.000000e+00
358 <em>Evaluated to
9.000000</em>
362 <p>Well this looks like it is basically working. The dump of the function
363 shows the
"no argument function that always returns double" that we synthesize
364 for each top level expression that is typed in. This demonstrates very basic
365 functionality, but can we do more?
</p>
367 <div class=
"doc_code">
369 ready
> <b>def testfunc(x y) x + y*
2;
</b>
370 Read function definition:
371 define double @testfunc(double %x, double %y) {
373 %multmp = mul double %y,
2.000000e+00
374 %addtmp = add double %multmp, %x
378 ready
> <b>testfunc(
4,
10);
</b>
379 define double @
""() {
381 %calltmp = call double @testfunc( double
4.000000e+00, double
1.000000e+01 )
385 <em>Evaluated to
24.000000</em>
389 <p>This illustrates that we can now call user code, but there is something a bit subtle
390 going on here. Note that we only invoke the JIT on the anonymous functions
391 that
<em>call testfunc
</em>, but we never invoked it on
<em>testfunc
394 <p>What actually happened here is that the anonymous function was
395 JIT'd when requested. When the Kaleidoscope app calls through the function
396 pointer that is returned, the anonymous function starts executing. It ends up
397 making the call to the
"testfunc" function, and ends up in a stub that invokes
398 the JIT, lazily, on testfunc. Once the JIT finishes lazily compiling testfunc,
399 it returns and the code re-executes the call.
</p>
401 <p>In summary, the JIT will lazily JIT code, on the fly, as it is needed. The
402 JIT provides a number of other more advanced interfaces for things like freeing
403 allocated machine code, rejit'ing functions to update them, etc. However, even
404 with this simple code, we get some surprisingly powerful capabilities - check
405 this out (I removed the dump of the anonymous functions, you should get the idea
408 <div class=
"doc_code">
410 ready
> <b>extern sin(x);
</b>
412 declare double @sin(double)
414 ready
> <b>extern cos(x);
</b>
416 declare double @cos(double)
418 ready
> <b>sin(
1.0);
</b>
419 <em>Evaluated to
0.841471</em>
421 ready
> <b>def foo(x) sin(x)*sin(x) + cos(x)*cos(x);
</b>
422 Read function definition:
423 define double @foo(double %x) {
425 %calltmp = call double @sin( double %x )
426 %multmp = mul double %calltmp, %calltmp
427 %calltmp2 = call double @cos( double %x )
428 %multmp4 = mul double %calltmp2, %calltmp2
429 %addtmp = add double %multmp, %multmp4
433 ready
> <b>foo(
4.0);
</b>
434 <em>Evaluated to
1.000000</em>
438 <p>Whoa, how does the JIT know about sin and cos? The answer is surprisingly
440 example, the JIT started execution of a function and got to a function call. It
441 realized that the function was not yet JIT compiled and invoked the standard set
442 of routines to resolve the function. In this case, there is no body defined
443 for the function, so the JIT ended up calling
"<tt>dlsym("sin
")</tt>" on the
444 Kaleidoscope process itself.
445 Since
"<tt>sin</tt>" is defined within the JIT's address space, it simply
446 patches up calls in the module to call the libm version of
<tt>sin
</tt>
449 <p>The LLVM JIT provides a number of interfaces (look in the
450 <tt>ExecutionEngine.h
</tt> file) for controlling how unknown functions get
451 resolved. It allows you to establish explicit mappings between IR objects and
452 addresses (useful for LLVM global variables that you want to map to static
453 tables, for example), allows you to dynamically decide on the fly based on the
454 function name, and even allows you to have the JIT abort itself if any lazy
455 compilation is attempted.
</p>
457 <p>One interesting application of this is that we can now extend the language
458 by writing arbitrary C++ code to implement operations. For example, if we add:
461 <div class=
"doc_code">
463 /// putchard - putchar that takes a double and returns
0.
465 double putchard(double X) {
472 <p>Now we can produce simple output to the console by using things like:
473 "<tt>extern putchard(x); putchard(120);</tt>", which prints a lowercase 'x' on
474 the console (
120 is the ASCII code for 'x'). Similar code could be used to
475 implement file I/O, console input, and many other capabilities in
478 <p>This completes the JIT and optimizer chapter of the Kaleidoscope tutorial. At
479 this point, we can compile a non-Turing-complete programming language, optimize
480 and JIT compile it in a user-driven way. Next up we'll look into
<a
481 href=
"LangImpl5.html">extending the language with control flow constructs
</a>,
482 tackling some interesting LLVM IR issues along the way.
</p>
486 <!-- *********************************************************************** -->
487 <div class=
"doc_section"><a name=
"code">Full Code Listing
</a></div>
488 <!-- *********************************************************************** -->
490 <div class=
"doc_text">
493 Here is the complete code listing for our running example, enhanced with the
494 LLVM JIT and optimizer. To build this example, use:
497 <div class=
"doc_code">
500 g++ -g toy.cpp `llvm-config --cppflags --ldflags --libs core jit native` -O3 -o toy
507 If you are compiling this on Linux, make sure to add the
"-rdynamic" option
508 as well. This makes sure that the external functions are resolved properly
511 <p>Here is the code:
</p>
513 <div class=
"doc_code">
515 #include
"llvm/DerivedTypes.h"
516 #include
"llvm/ExecutionEngine/ExecutionEngine.h"
517 #include
"llvm/LLVMContext.h"
518 #include
"llvm/Module.h"
519 #include
"llvm/ModuleProvider.h"
520 #include
"llvm/PassManager.h"
521 #include
"llvm/Analysis/Verifier.h"
522 #include
"llvm/Target/TargetData.h"
523 #include
"llvm/Transforms/Scalar.h"
524 #include
"llvm/Support/IRBuilder.h"
525 #include
<cstdio
>
526 #include
<string
>
528 #include
<vector
>
529 using namespace llvm;
531 //===----------------------------------------------------------------------===//
533 //===----------------------------------------------------------------------===//
535 // The lexer returns tokens [
0-
255] if it is an unknown character, otherwise one
536 // of these for known things.
541 tok_def = -
2, tok_extern = -
3,
544 tok_identifier = -
4, tok_number = -
5,
547 static std::string IdentifierStr; // Filled in if tok_identifier
548 static double NumVal; // Filled in if tok_number
550 /// gettok - Return the next token from standard input.
551 static int gettok() {
552 static int LastChar = ' ';
554 // Skip any whitespace.
555 while (isspace(LastChar))
556 LastChar = getchar();
558 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-
9]*
559 IdentifierStr = LastChar;
560 while (isalnum((LastChar = getchar())))
561 IdentifierStr += LastChar;
563 if (IdentifierStr ==
"def") return tok_def;
564 if (IdentifierStr ==
"extern") return tok_extern;
565 return tok_identifier;
568 if (isdigit(LastChar) || LastChar == '.') { // Number: [
0-
9.]+
572 LastChar = getchar();
573 } while (isdigit(LastChar) || LastChar == '.');
575 NumVal = strtod(NumStr.c_str(),
0);
579 if (LastChar == '#') {
580 // Comment until end of line.
581 do LastChar = getchar();
582 while (LastChar != EOF
&& LastChar != '\n'
&& LastChar != '\r');
588 // Check for end of file. Don't eat the EOF.
592 // Otherwise, just return the character as its ascii value.
593 int ThisChar = LastChar;
594 LastChar = getchar();
598 //===----------------------------------------------------------------------===//
599 // Abstract Syntax Tree (aka Parse Tree)
600 //===----------------------------------------------------------------------===//
602 /// ExprAST - Base class for all expression nodes.
605 virtual ~ExprAST() {}
606 virtual Value *Codegen() =
0;
609 /// NumberExprAST - Expression class for numeric literals like
"1.0".
610 class NumberExprAST : public ExprAST {
613 NumberExprAST(double val) : Val(val) {}
614 virtual Value *Codegen();
617 /// VariableExprAST - Expression class for referencing a variable, like
"a".
618 class VariableExprAST : public ExprAST {
621 VariableExprAST(const std::string
&name) : Name(name) {}
622 virtual Value *Codegen();
625 /// BinaryExprAST - Expression class for a binary operator.
626 class BinaryExprAST : public ExprAST {
630 BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs)
631 : Op(op), LHS(lhs), RHS(rhs) {}
632 virtual Value *Codegen();
635 /// CallExprAST - Expression class for function calls.
636 class CallExprAST : public ExprAST {
638 std::vector
<ExprAST*
> Args;
640 CallExprAST(const std::string
&callee, std::vector
<ExprAST*
> &args)
641 : Callee(callee), Args(args) {}
642 virtual Value *Codegen();
645 /// PrototypeAST - This class represents the
"prototype" for a function,
646 /// which captures its argument names as well as if it is an operator.
649 std::vector
<std::string
> Args;
651 PrototypeAST(const std::string
&name, const std::vector
<std::string
> &args)
652 : Name(name), Args(args) {}
657 /// FunctionAST - This class represents a function definition itself.
662 FunctionAST(PrototypeAST *proto, ExprAST *body)
663 : Proto(proto), Body(body) {}
668 //===----------------------------------------------------------------------===//
670 //===----------------------------------------------------------------------===//
672 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
673 /// token the parser it looking at. getNextToken reads another token from the
674 /// lexer and updates CurTok with its results.
676 static int getNextToken() {
677 return CurTok = gettok();
680 /// BinopPrecedence - This holds the precedence for each binary operator that is
682 static std::map
<char, int
> BinopPrecedence;
684 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
685 static int GetTokPrecedence() {
686 if (!isascii(CurTok))
689 // Make sure it's a declared binop.
690 int TokPrec = BinopPrecedence[CurTok];
691 if (TokPrec
<=
0) return -
1;
695 /// Error* - These are little helper functions for error handling.
696 ExprAST *Error(const char *Str) { fprintf(stderr,
"Error: %s\n", Str);return
0;}
697 PrototypeAST *ErrorP(const char *Str) { Error(Str); return
0; }
698 FunctionAST *ErrorF(const char *Str) { Error(Str); return
0; }
700 static ExprAST *ParseExpression();
704 /// ::= identifier '(' expression* ')'
705 static ExprAST *ParseIdentifierExpr() {
706 std::string IdName = IdentifierStr;
708 getNextToken(); // eat identifier.
710 if (CurTok != '(') // Simple variable ref.
711 return new VariableExprAST(IdName);
714 getNextToken(); // eat (
715 std::vector
<ExprAST*
> Args;
718 ExprAST *Arg = ParseExpression();
722 if (CurTok == ')') break;
725 return Error(
"Expected ')' or ',' in argument list");
733 return new CallExprAST(IdName, Args);
736 /// numberexpr ::= number
737 static ExprAST *ParseNumberExpr() {
738 ExprAST *Result = new NumberExprAST(NumVal);
739 getNextToken(); // consume the number
743 /// parenexpr ::= '(' expression ')'
744 static ExprAST *ParseParenExpr() {
745 getNextToken(); // eat (.
746 ExprAST *V = ParseExpression();
750 return Error(
"expected ')'");
751 getNextToken(); // eat ).
756 /// ::= identifierexpr
759 static ExprAST *ParsePrimary() {
761 default: return Error(
"unknown token when expecting an expression");
762 case tok_identifier: return ParseIdentifierExpr();
763 case tok_number: return ParseNumberExpr();
764 case '(': return ParseParenExpr();
769 /// ::= ('+' primary)*
770 static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) {
771 // If this is a binop, find its precedence.
773 int TokPrec = GetTokPrecedence();
775 // If this is a binop that binds at least as tightly as the current binop,
776 // consume it, otherwise we are done.
777 if (TokPrec
< ExprPrec)
780 // Okay, we know this is a binop.
782 getNextToken(); // eat binop
784 // Parse the primary expression after the binary operator.
785 ExprAST *RHS = ParsePrimary();
788 // If BinOp binds less tightly with RHS than the operator after RHS, let
789 // the pending operator take RHS as its LHS.
790 int NextPrec = GetTokPrecedence();
791 if (TokPrec
< NextPrec) {
792 RHS = ParseBinOpRHS(TokPrec+
1, RHS);
793 if (RHS ==
0) return
0;
797 LHS = new BinaryExprAST(BinOp, LHS, RHS);
802 /// ::= primary binoprhs
804 static ExprAST *ParseExpression() {
805 ExprAST *LHS = ParsePrimary();
808 return ParseBinOpRHS(
0, LHS);
812 /// ::= id '(' id* ')'
813 static PrototypeAST *ParsePrototype() {
814 if (CurTok != tok_identifier)
815 return ErrorP(
"Expected function name in prototype");
817 std::string FnName = IdentifierStr;
821 return ErrorP(
"Expected '(' in prototype");
823 std::vector
<std::string
> ArgNames;
824 while (getNextToken() == tok_identifier)
825 ArgNames.push_back(IdentifierStr);
827 return ErrorP(
"Expected ')' in prototype");
830 getNextToken(); // eat ')'.
832 return new PrototypeAST(FnName, ArgNames);
835 /// definition ::= 'def' prototype expression
836 static FunctionAST *ParseDefinition() {
837 getNextToken(); // eat def.
838 PrototypeAST *Proto = ParsePrototype();
839 if (Proto ==
0) return
0;
841 if (ExprAST *E = ParseExpression())
842 return new FunctionAST(Proto, E);
846 /// toplevelexpr ::= expression
847 static FunctionAST *ParseTopLevelExpr() {
848 if (ExprAST *E = ParseExpression()) {
849 // Make an anonymous proto.
850 PrototypeAST *Proto = new PrototypeAST(
"", std::vector
<std::string
>());
851 return new FunctionAST(Proto, E);
856 /// external ::= 'extern' prototype
857 static PrototypeAST *ParseExtern() {
858 getNextToken(); // eat extern.
859 return ParsePrototype();
862 //===----------------------------------------------------------------------===//
864 //===----------------------------------------------------------------------===//
866 static Module *TheModule;
867 static IRBuilder
<> Builder(getGlobalContext());
868 static std::map
<std::string, Value*
> NamedValues;
869 static FunctionPassManager *TheFPM;
871 Value *ErrorV(const char *Str) { Error(Str); return
0; }
873 Value *NumberExprAST::Codegen() {
874 return ConstantFP::get(getGlobalContext(), APFloat(Val));
877 Value *VariableExprAST::Codegen() {
878 // Look this variable up in the function.
879 Value *V = NamedValues[Name];
880 return V ? V : ErrorV(
"Unknown variable name");
883 Value *BinaryExprAST::Codegen() {
884 Value *L = LHS-
>Codegen();
885 Value *R = RHS-
>Codegen();
886 if (L ==
0 || R ==
0) return
0;
889 case '+': return Builder.CreateAdd(L, R,
"addtmp");
890 case '-': return Builder.CreateSub(L, R,
"subtmp");
891 case '*': return Builder.CreateMul(L, R,
"multmp");
893 L = Builder.CreateFCmpULT(L, R,
"cmptmp");
894 // Convert bool
0/
1 to double
0.0 or
1.0
895 return Builder.CreateUIToFP(L, Type::getDoubleTy(getGlobalContext()),
"booltmp");
896 default: return ErrorV(
"invalid binary operator");
900 Value *CallExprAST::Codegen() {
901 // Look up the name in the global module table.
902 Function *CalleeF = TheModule-
>getFunction(Callee);
904 return ErrorV(
"Unknown function referenced");
906 // If argument mismatch error.
907 if (CalleeF-
>arg_size() != Args.size())
908 return ErrorV(
"Incorrect # arguments passed");
910 std::vector
<Value*
> ArgsV;
911 for (unsigned i =
0, e = Args.size(); i != e; ++i) {
912 ArgsV.push_back(Args[i]-
>Codegen());
913 if (ArgsV.back() ==
0) return
0;
916 return Builder.CreateCall(CalleeF, ArgsV.begin(), ArgsV.end(),
"calltmp");
919 Function *PrototypeAST::Codegen() {
920 // Make the function type: double(double,double) etc.
921 std::vector
<const Type*
> Doubles(Args.size(), Type::getDoubleTy(getGlobalContext()));
922 FunctionType *FT = FunctionType::get(Type::getDoubleTy(getGlobalContext()), Doubles, false);
924 Function *F = Function::Create(FT, Function::ExternalLinkage, Name, TheModule);
926 // If F conflicted, there was already something named 'Name'. If it has a
927 // body, don't allow redefinition or reextern.
928 if (F-
>getName() != Name) {
929 // Delete the one we just made and get the existing one.
930 F-
>eraseFromParent();
931 F = TheModule-
>getFunction(Name);
933 // If F already has a body, reject this.
934 if (!F-
>empty()) {
935 ErrorF(
"redefinition of function");
939 // If F took a different number of args, reject.
940 if (F-
>arg_size() != Args.size()) {
941 ErrorF(
"redefinition of function with different # args");
946 // Set names for all arguments.
948 for (Function::arg_iterator AI = F-
>arg_begin(); Idx != Args.size();
950 AI-
>setName(Args[Idx]);
952 // Add arguments to variable symbol table.
953 NamedValues[Args[Idx]] = AI;
959 Function *FunctionAST::Codegen() {
962 Function *TheFunction = Proto-
>Codegen();
963 if (TheFunction ==
0)
966 // Create a new basic block to start insertion into.
967 BasicBlock *BB = BasicBlock::Create(getGlobalContext(),
"entry", TheFunction);
968 Builder.SetInsertPoint(BB);
970 if (Value *RetVal = Body-
>Codegen()) {
971 // Finish off the function.
972 Builder.CreateRet(RetVal);
974 // Validate the generated code, checking for consistency.
975 verifyFunction(*TheFunction);
977 // Optimize the function.
978 TheFPM-
>run(*TheFunction);
983 // Error reading body, remove function.
984 TheFunction-
>eraseFromParent();
988 //===----------------------------------------------------------------------===//
989 // Top-Level parsing and JIT Driver
990 //===----------------------------------------------------------------------===//
992 static ExecutionEngine *TheExecutionEngine;
994 static void HandleDefinition() {
995 if (FunctionAST *F = ParseDefinition()) {
996 if (Function *LF = F-
>Codegen()) {
997 fprintf(stderr,
"Read function definition:");
1001 // Skip token for error recovery.
1006 static void HandleExtern() {
1007 if (PrototypeAST *P = ParseExtern()) {
1008 if (Function *F = P-
>Codegen()) {
1009 fprintf(stderr,
"Read extern: ");
1013 // Skip token for error recovery.
1018 static void HandleTopLevelExpression() {
1019 // Evaluate a top level expression into an anonymous function.
1020 if (FunctionAST *F = ParseTopLevelExpr()) {
1021 if (Function *LF = F-
>Codegen()) {
1022 // JIT the function, returning a function pointer.
1023 void *FPtr = TheExecutionEngine-
>getPointerToFunction(LF);
1025 // Cast it to the right type (takes no arguments, returns a double) so we
1026 // can call it as a native function.
1027 double (*FP)() = (double (*)())FPtr;
1028 fprintf(stderr,
"Evaluated to %f\n", FP());
1031 // Skip token for error recovery.
1036 /// top ::= definition | external | expression | ';'
1037 static void MainLoop() {
1039 fprintf(stderr,
"ready> ");
1041 case tok_eof: return;
1042 case ';': getNextToken(); break; // ignore top level semicolons.
1043 case tok_def: HandleDefinition(); break;
1044 case tok_extern: HandleExtern(); break;
1045 default: HandleTopLevelExpression(); break;
1052 //===----------------------------------------------------------------------===//
1053 //
"Library" functions that can be
"extern'd" from user code.
1054 //===----------------------------------------------------------------------===//
1056 /// putchard - putchar that takes a double and returns
0.
1058 double putchard(double X) {
1063 //===----------------------------------------------------------------------===//
1064 // Main driver code.
1065 //===----------------------------------------------------------------------===//
1068 // Install standard binary operators.
1069 //
1 is lowest precedence.
1070 BinopPrecedence['
<'] =
10;
1071 BinopPrecedence['+'] =
20;
1072 BinopPrecedence['-'] =
20;
1073 BinopPrecedence['*'] =
40; // highest.
1075 // Prime the first token.
1076 fprintf(stderr,
"ready> ");
1079 // Make the module, which holds all the code.
1080 TheModule = new Module(
"my cool jit", getGlobalContext());
1082 ExistingModuleProvider *OurModuleProvider =
1083 new ExistingModuleProvider(TheModule);
1085 // Create the JIT. This takes ownership of the module and module provider.
1086 TheExecutionEngine = EngineBuilder(OurModuleProvider).create();
1088 FunctionPassManager OurFPM(OurModuleProvider);
1090 // Set up the optimizer pipeline. Start with registering info about how the
1091 // target lays out data structures.
1092 OurFPM.add(new TargetData(*TheExecutionEngine-
>getTargetData()));
1093 // Do simple
"peephole" optimizations and bit-twiddling optzns.
1094 OurFPM.add(createInstructionCombiningPass());
1095 // Reassociate expressions.
1096 OurFPM.add(createReassociatePass());
1097 // Eliminate Common SubExpressions.
1098 OurFPM.add(createGVNPass());
1099 // Simplify the control flow graph (deleting unreachable blocks, etc).
1100 OurFPM.add(createCFGSimplificationPass());
1102 // Set the global so the code gen can use this.
1103 TheFPM =
&OurFPM;
1105 // Run the main
"interpreter loop" now.
1110 // Print out all of the generated code.
1111 TheModule-
>dump();
1118 <a href=
"LangImpl5.html">Next: Extending the language: control flow
</a>
1121 <!-- *********************************************************************** -->
1124 <a href=
"http://jigsaw.w3.org/css-validator/check/referer"><img
1125 src=
"http://jigsaw.w3.org/css-validator/images/vcss" alt=
"Valid CSS!"></a>
1126 <a href=
"http://validator.w3.org/check/referer"><img
1127 src=
"http://www.w3.org/Icons/valid-html401" alt=
"Valid HTML 4.01!"></a>
1129 <a href=
"mailto:sabre@nondot.org">Chris Lattner
</a><br>
1130 <a href=
"http://llvm.org">The LLVM Compiler Infrastructure
</a><br>
1131 Last modified: $Date:
2007-
10-
17 11:
05:
13 -
0700 (Wed,
17 Oct
2007) $