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 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(TheModule);
175 FunctionPassManager OurFPM(&OurModuleProvider);
177 // Set up the optimizer pipeline. Start with registering info about how the
178 // target lays out data structures.
179 OurFPM.add(new TargetData(*TheExecutionEngine->getTargetData()));
180 // Do simple "peephole" optimizations and bit-twiddling optzns.
181 OurFPM.add(createInstructionCombiningPass());
182 // Reassociate expressions.
183 OurFPM.add(createReassociatePass());
184 // Eliminate Common SubExpressions.
185 OurFPM.add(createGVNPass());
186 // Simplify the control flow graph (deleting unreachable blocks, etc).
187 OurFPM.add(createCFGSimplificationPass());
189 // Set the global so the code gen can use this.
190 TheFPM = &OurFPM;
192 // Run the main "interpreter loop" now.
197 <p>This code defines two objects, an <tt>ExistingModuleProvider</tt> and a
198 <tt>FunctionPassManager</tt>. The former is basically a wrapper around our
199 <tt>Module</tt> that the PassManager requires. It provides certain flexibility
200 that we're not going to take advantage of here, so I won't dive into any details
203 <p>The meat of the matter here, is the definition of "<tt>OurFPM</tt>". It
204 requires a pointer to the <tt>Module</tt> (through the <tt>ModuleProvider</tt>)
205 to construct itself. Once it is set up, we use a series of "add" calls to add
206 a bunch of LLVM passes. The first pass is basically boilerplate, it adds a pass
207 so that later optimizations know how the data structures in the program are
208 layed out. The "<tt>TheExecutionEngine</tt>" variable is related to the JIT,
209 which we will get to in the next section.</p>
211 <p>In this case, we choose to add 4 optimization passes. The passes we chose
212 here are a pretty standard set of "cleanup" optimizations that are useful for
213 a wide variety of code. I won't delve into what they do but, believe me,
214 they are a good starting place :).</p>
216 <p>Once the PassManager is set up, we need to make use of it. We do this by
217 running it after our newly created function is constructed (in
218 <tt>FunctionAST::Codegen</tt>), but before it is returned to the client:</p>
220 <div class="doc_code">
222 if (Value *RetVal = Body->Codegen()) {
223 // Finish off the function.
224 Builder.CreateRet(RetVal);
226 // Validate the generated code, checking for consistency.
227 verifyFunction(*TheFunction);
229 <b>// Optimize the function.
230 TheFPM->run(*TheFunction);</b>
237 <p>As you can see, this is pretty straightforward. The
238 <tt>FunctionPassManager</tt> optimizes and updates the LLVM Function* in place,
239 improving (hopefully) its body. With this in place, we can try our test above
242 <div class="doc_code">
244 ready> <b>def test(x) (1+2+x)*(x+(1+2));</b>
245 ready> Read function definition:
246 define double @test(double %x) {
248 %addtmp = add double %x, 3.000000e+00
249 %multmp = mul double %addtmp, %addtmp
255 <p>As expected, we now get our nicely optimized code, saving a floating point
256 add instruction from every execution of this function.</p>
258 <p>LLVM provides a wide variety of optimizations that can be used in certain
259 circumstances. Some <a href="../Passes.html">documentation about the various
260 passes</a> is available, but it isn't very complete. Another good source of
261 ideas can come from looking at the passes that <tt>llvm-gcc</tt> or
262 <tt>llvm-ld</tt> run to get started. The "<tt>opt</tt>" tool allows you to
263 experiment with passes from the command line, so you can see if they do
266 <p>Now that we have reasonable code coming out of our front-end, lets talk about
271 <!-- *********************************************************************** -->
272 <div class="doc_section"><a name="jit">Adding a JIT Compiler</a></div>
273 <!-- *********************************************************************** -->
275 <div class="doc_text">
277 <p>Code that is available in LLVM IR can have a wide variety of tools
278 applied to it. For example, you can run optimizations on it (as we did above),
279 you can dump it out in textual or binary forms, you can compile the code to an
280 assembly file (.s) for some target, or you can JIT compile it. The nice thing
281 about the LLVM IR representation is that it is the "common currency" between
282 many different parts of the compiler.
285 <p>In this section, we'll add JIT compiler support to our interpreter. The
286 basic idea that we want for Kaleidoscope is to have the user enter function
287 bodies as they do now, but immediately evaluate the top-level expressions they
288 type in. For example, if they type in "1 + 2;", we should evaluate and print
289 out 3. If they define a function, they should be able to call it from the
292 <p>In order to do this, we first declare and initialize the JIT. This is done
293 by adding a global variable and a call in <tt>main</tt>:</p>
295 <div class="doc_code">
297 <b>static ExecutionEngine *TheExecutionEngine;</b>
301 <b>// Create the JIT.
302 TheExecutionEngine = ExecutionEngine::create(TheModule);</b>
308 <p>This creates an abstract "Execution Engine" which can be either a JIT
309 compiler or the LLVM interpreter. LLVM will automatically pick a JIT compiler
310 for you if one is available for your platform, otherwise it will fall back to
313 <p>Once the <tt>ExecutionEngine</tt> is created, the JIT is ready to be used.
314 There are a variety of APIs that are useful, but the simplest one is the
315 "<tt>getPointerToFunction(F)</tt>" method. This method JIT compiles the
316 specified LLVM Function and returns a function pointer to the generated machine
317 code. In our case, this means that we can change the code that parses a
318 top-level expression to look like this:</p>
320 <div class="doc_code">
322 static void HandleTopLevelExpression() {
323 // Evaluate a top level expression into an anonymous function.
324 if (FunctionAST *F = ParseTopLevelExpr()) {
325 if (Function *LF = F->Codegen()) {
326 LF->dump(); // Dump the function for exposition purposes.
328 <b>// JIT the function, returning a function pointer.
329 void *FPtr = TheExecutionEngine->getPointerToFunction(LF);
331 // Cast it to the right type (takes no arguments, returns a double) so we
332 // can call it as a native function.
333 double (*FP)() = (double (*)())FPtr;
334 fprintf(stderr, "Evaluated to %f\n", FP());</b>
339 <p>Recall that we compile top-level expressions into a self-contained LLVM
340 function that takes no arguments and returns the computed double. Because the
341 LLVM JIT compiler matches the native platform ABI, this means that you can just
342 cast the result pointer to a function pointer of that type and call it directly.
343 This means, there is no difference between JIT compiled code and native machine
344 code that is statically linked into your application.</p>
346 <p>With just these two changes, lets see how Kaleidoscope works now!</p>
348 <div class="doc_code">
350 ready> <b>4+5;</b>
351 define double @""() {
353 ret double 9.000000e+00
356 <em>Evaluated to 9.000000</em>
360 <p>Well this looks like it is basically working. The dump of the function
361 shows the "no argument function that always returns double" that we synthesize
362 for each top level expression that is typed in. This demonstrates very basic
363 functionality, but can we do more?</p>
365 <div class="doc_code">
367 ready> <b>def testfunc(x y) x + y*2; </b>
368 Read function definition:
369 define double @testfunc(double %x, double %y) {
371 %multmp = mul double %y, 2.000000e+00
372 %addtmp = add double %multmp, %x
376 ready> <b>testfunc(4, 10);</b>
377 define double @""() {
379 %calltmp = call double @testfunc( double 4.000000e+00, double 1.000000e+01 )
383 <em>Evaluated to 24.000000</em>
387 <p>This illustrates that we can now call user code, but there is something a bit subtle
388 going on here. Note that we only invoke the JIT on the anonymous functions
389 that <em>call testfunc</em>, but we never invoked it on <em>testfunc
392 <p>What actually happened here is that the anonymous function was
393 JIT'd when requested. When the Kaleidoscope app calls through the function
394 pointer that is returned, the anonymous function starts executing. It ends up
395 making the call to the "testfunc" function, and ends up in a stub that invokes
396 the JIT, lazily, on testfunc. Once the JIT finishes lazily compiling testfunc,
397 it returns and the code re-executes the call.</p>
399 <p>In summary, the JIT will lazily JIT code, on the fly, as it is needed. The
400 JIT provides a number of other more advanced interfaces for things like freeing
401 allocated machine code, rejit'ing functions to update them, etc. However, even
402 with this simple code, we get some surprisingly powerful capabilities - check
403 this out (I removed the dump of the anonymous functions, you should get the idea
406 <div class="doc_code">
408 ready> <b>extern sin(x);</b>
410 declare double @sin(double)
412 ready> <b>extern cos(x);</b>
414 declare double @cos(double)
416 ready> <b>sin(1.0);</b>
417 <em>Evaluated to 0.841471</em>
419 ready> <b>def foo(x) sin(x)*sin(x) + cos(x)*cos(x);</b>
420 Read function definition:
421 define double @foo(double %x) {
423 %calltmp = call double @sin( double %x )
424 %multmp = mul double %calltmp, %calltmp
425 %calltmp2 = call double @cos( double %x )
426 %multmp4 = mul double %calltmp2, %calltmp2
427 %addtmp = add double %multmp, %multmp4
431 ready> <b>foo(4.0);</b>
432 <em>Evaluated to 1.000000</em>
436 <p>Whoa, how does the JIT know about sin and cos? The answer is surprisingly
438 example, the JIT started execution of a function and got to a function call. It
439 realized that the function was not yet JIT compiled and invoked the standard set
440 of routines to resolve the function. In this case, there is no body defined
441 for the function, so the JIT ended up calling "<tt>dlsym("sin")</tt>" on the
442 Kaleidoscope process itself.
443 Since "<tt>sin</tt>" is defined within the JIT's address space, it simply
444 patches up calls in the module to call the libm version of <tt>sin</tt>
447 <p>The LLVM JIT provides a number of interfaces (look in the
448 <tt>ExecutionEngine.h</tt> file) for controlling how unknown functions get
449 resolved. It allows you to establish explicit mappings between IR objects and
450 addresses (useful for LLVM global variables that you want to map to static
451 tables, for example), allows you to dynamically decide on the fly based on the
452 function name, and even allows you to have the JIT abort itself if any lazy
453 compilation is attempted.</p>
455 <p>One interesting application of this is that we can now extend the language
456 by writing arbitrary C++ code to implement operations. For example, if we add:
459 <div class="doc_code">
461 /// putchard - putchar that takes a double and returns 0.
463 double putchard(double X) {
470 <p>Now we can produce simple output to the console by using things like:
471 "<tt>extern putchard(x); putchard(120);</tt>", which prints a lowercase 'x' on
472 the console (120 is the ASCII code for 'x'). Similar code could be used to
473 implement file I/O, console input, and many other capabilities in
476 <p>This completes the JIT and optimizer chapter of the Kaleidoscope tutorial. At
477 this point, we can compile a non-Turing-complete programming language, optimize
478 and JIT compile it in a user-driven way. Next up we'll look into <a
479 href="LangImpl5.html">extending the language with control flow constructs</a>,
480 tackling some interesting LLVM IR issues along the way.</p>
484 <!-- *********************************************************************** -->
485 <div class="doc_section"><a name="code">Full Code Listing</a></div>
486 <!-- *********************************************************************** -->
488 <div class="doc_text">
491 Here is the complete code listing for our running example, enhanced with the
492 LLVM JIT and optimizer. To build this example, use:
495 <div class="doc_code">
498 g++ -g toy.cpp `llvm-config --cppflags --ldflags --libs core jit native` -O3 -o toy
504 <p>Here is the code:</p>
506 <div class="doc_code">
508 #include "llvm/DerivedTypes.h"
509 #include "llvm/ExecutionEngine/ExecutionEngine.h"
510 #include "llvm/Module.h"
511 #include "llvm/ModuleProvider.h"
512 #include "llvm/PassManager.h"
513 #include "llvm/Analysis/Verifier.h"
514 #include "llvm/Target/TargetData.h"
515 #include "llvm/Transforms/Scalar.h"
516 #include "llvm/Support/IRBuilder.h"
517 #include <cstdio>
518 #include <string>
520 #include <vector>
521 using namespace llvm;
523 //===----------------------------------------------------------------------===//
525 //===----------------------------------------------------------------------===//
527 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
528 // of these for known things.
533 tok_def = -2, tok_extern = -3,
536 tok_identifier = -4, tok_number = -5,
539 static std::string IdentifierStr; // Filled in if tok_identifier
540 static double NumVal; // Filled in if tok_number
542 /// gettok - Return the next token from standard input.
543 static int gettok() {
544 static int LastChar = ' ';
546 // Skip any whitespace.
547 while (isspace(LastChar))
548 LastChar = getchar();
550 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
551 IdentifierStr = LastChar;
552 while (isalnum((LastChar = getchar())))
553 IdentifierStr += LastChar;
555 if (IdentifierStr == "def") return tok_def;
556 if (IdentifierStr == "extern") return tok_extern;
557 return tok_identifier;
560 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+
564 LastChar = getchar();
565 } while (isdigit(LastChar) || LastChar == '.');
567 NumVal = strtod(NumStr.c_str(), 0);
571 if (LastChar == '#') {
572 // Comment until end of line.
573 do LastChar = getchar();
574 while (LastChar != EOF && LastChar != '\n' && LastChar != '\r');
580 // Check for end of file. Don't eat the EOF.
584 // Otherwise, just return the character as its ascii value.
585 int ThisChar = LastChar;
586 LastChar = getchar();
590 //===----------------------------------------------------------------------===//
591 // Abstract Syntax Tree (aka Parse Tree)
592 //===----------------------------------------------------------------------===//
594 /// ExprAST - Base class for all expression nodes.
597 virtual ~ExprAST() {}
598 virtual Value *Codegen() = 0;
601 /// NumberExprAST - Expression class for numeric literals like "1.0".
602 class NumberExprAST : public ExprAST {
605 NumberExprAST(double val) : Val(val) {}
606 virtual Value *Codegen();
609 /// VariableExprAST - Expression class for referencing a variable, like "a".
610 class VariableExprAST : public ExprAST {
613 VariableExprAST(const std::string &name) : Name(name) {}
614 virtual Value *Codegen();
617 /// BinaryExprAST - Expression class for a binary operator.
618 class BinaryExprAST : public ExprAST {
622 BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs)
623 : Op(op), LHS(lhs), RHS(rhs) {}
624 virtual Value *Codegen();
627 /// CallExprAST - Expression class for function calls.
628 class CallExprAST : public ExprAST {
630 std::vector<ExprAST*> Args;
632 CallExprAST(const std::string &callee, std::vector<ExprAST*> &args)
633 : Callee(callee), Args(args) {}
634 virtual Value *Codegen();
637 /// PrototypeAST - This class represents the "prototype" for a function,
638 /// which captures its argument names as well as if it is an operator.
641 std::vector<std::string> Args;
643 PrototypeAST(const std::string &name, const std::vector<std::string> &args)
644 : Name(name), Args(args) {}
649 /// FunctionAST - This class represents a function definition itself.
654 FunctionAST(PrototypeAST *proto, ExprAST *body)
655 : Proto(proto), Body(body) {}
660 //===----------------------------------------------------------------------===//
662 //===----------------------------------------------------------------------===//
664 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
665 /// token the parser it looking at. getNextToken reads another token from the
666 /// lexer and updates CurTok with its results.
668 static int getNextToken() {
669 return CurTok = gettok();
672 /// BinopPrecedence - This holds the precedence for each binary operator that is
674 static std::map<char, int> BinopPrecedence;
676 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
677 static int GetTokPrecedence() {
678 if (!isascii(CurTok))
681 // Make sure it's a declared binop.
682 int TokPrec = BinopPrecedence[CurTok];
683 if (TokPrec <= 0) return -1;
687 /// Error* - These are little helper functions for error handling.
688 ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;}
689 PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; }
690 FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; }
692 static ExprAST *ParseExpression();
696 /// ::= identifier '(' expression* ')'
697 static ExprAST *ParseIdentifierExpr() {
698 std::string IdName = IdentifierStr;
700 getNextToken(); // eat identifier.
702 if (CurTok != '(') // Simple variable ref.
703 return new VariableExprAST(IdName);
706 getNextToken(); // eat (
707 std::vector<ExprAST*> Args;
710 ExprAST *Arg = ParseExpression();
714 if (CurTok == ')') break;
717 return Error("Expected ')' or ',' in argument list");
725 return new CallExprAST(IdName, Args);
728 /// numberexpr ::= number
729 static ExprAST *ParseNumberExpr() {
730 ExprAST *Result = new NumberExprAST(NumVal);
731 getNextToken(); // consume the number
735 /// parenexpr ::= '(' expression ')'
736 static ExprAST *ParseParenExpr() {
737 getNextToken(); // eat (.
738 ExprAST *V = ParseExpression();
742 return Error("expected ')'");
743 getNextToken(); // eat ).
748 /// ::= identifierexpr
751 static ExprAST *ParsePrimary() {
753 default: return Error("unknown token when expecting an expression");
754 case tok_identifier: return ParseIdentifierExpr();
755 case tok_number: return ParseNumberExpr();
756 case '(': return ParseParenExpr();
761 /// ::= ('+' primary)*
762 static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) {
763 // If this is a binop, find its precedence.
765 int TokPrec = GetTokPrecedence();
767 // If this is a binop that binds at least as tightly as the current binop,
768 // consume it, otherwise we are done.
769 if (TokPrec < ExprPrec)
772 // Okay, we know this is a binop.
774 getNextToken(); // eat binop
776 // Parse the primary expression after the binary operator.
777 ExprAST *RHS = ParsePrimary();
780 // If BinOp binds less tightly with RHS than the operator after RHS, let
781 // the pending operator take RHS as its LHS.
782 int NextPrec = GetTokPrecedence();
783 if (TokPrec < NextPrec) {
784 RHS = ParseBinOpRHS(TokPrec+1, RHS);
785 if (RHS == 0) return 0;
789 LHS = new BinaryExprAST(BinOp, LHS, RHS);
794 /// ::= primary binoprhs
796 static ExprAST *ParseExpression() {
797 ExprAST *LHS = ParsePrimary();
800 return ParseBinOpRHS(0, LHS);
804 /// ::= id '(' id* ')'
805 static PrototypeAST *ParsePrototype() {
806 if (CurTok != tok_identifier)
807 return ErrorP("Expected function name in prototype");
809 std::string FnName = IdentifierStr;
813 return ErrorP("Expected '(' in prototype");
815 std::vector<std::string> ArgNames;
816 while (getNextToken() == tok_identifier)
817 ArgNames.push_back(IdentifierStr);
819 return ErrorP("Expected ')' in prototype");
822 getNextToken(); // eat ')'.
824 return new PrototypeAST(FnName, ArgNames);
827 /// definition ::= 'def' prototype expression
828 static FunctionAST *ParseDefinition() {
829 getNextToken(); // eat def.
830 PrototypeAST *Proto = ParsePrototype();
831 if (Proto == 0) return 0;
833 if (ExprAST *E = ParseExpression())
834 return new FunctionAST(Proto, E);
838 /// toplevelexpr ::= expression
839 static FunctionAST *ParseTopLevelExpr() {
840 if (ExprAST *E = ParseExpression()) {
841 // Make an anonymous proto.
842 PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>());
843 return new FunctionAST(Proto, E);
848 /// external ::= 'extern' prototype
849 static PrototypeAST *ParseExtern() {
850 getNextToken(); // eat extern.
851 return ParsePrototype();
854 //===----------------------------------------------------------------------===//
856 //===----------------------------------------------------------------------===//
858 static Module *TheModule;
859 static IRBuilder Builder;
860 static std::map<std::string, Value*> NamedValues;
861 static FunctionPassManager *TheFPM;
863 Value *ErrorV(const char *Str) { Error(Str); return 0; }
865 Value *NumberExprAST::Codegen() {
866 return ConstantFP::get(APFloat(Val));
869 Value *VariableExprAST::Codegen() {
870 // Look this variable up in the function.
871 Value *V = NamedValues[Name];
872 return V ? V : ErrorV("Unknown variable name");
875 Value *BinaryExprAST::Codegen() {
876 Value *L = LHS->Codegen();
877 Value *R = RHS->Codegen();
878 if (L == 0 || R == 0) return 0;
881 case '+': return Builder.CreateAdd(L, R, "addtmp");
882 case '-': return Builder.CreateSub(L, R, "subtmp");
883 case '*': return Builder.CreateMul(L, R, "multmp");
885 L = Builder.CreateFCmpULT(L, R, "cmptmp");
886 // Convert bool 0/1 to double 0.0 or 1.0
887 return Builder.CreateUIToFP(L, Type::DoubleTy, "booltmp");
888 default: return ErrorV("invalid binary operator");
892 Value *CallExprAST::Codegen() {
893 // Look up the name in the global module table.
894 Function *CalleeF = TheModule->getFunction(Callee);
896 return ErrorV("Unknown function referenced");
898 // If argument mismatch error.
899 if (CalleeF->arg_size() != Args.size())
900 return ErrorV("Incorrect # arguments passed");
902 std::vector<Value*> ArgsV;
903 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
904 ArgsV.push_back(Args[i]->Codegen());
905 if (ArgsV.back() == 0) return 0;
908 return Builder.CreateCall(CalleeF, ArgsV.begin(), ArgsV.end(), "calltmp");
911 Function *PrototypeAST::Codegen() {
912 // Make the function type: double(double,double) etc.
913 std::vector<const Type*> Doubles(Args.size(), Type::DoubleTy);
914 FunctionType *FT = FunctionType::get(Type::DoubleTy, Doubles, false);
916 Function *F = Function::Create(FT, Function::ExternalLinkage, Name, TheModule);
918 // If F conflicted, there was already something named 'Name'. If it has a
919 // body, don't allow redefinition or reextern.
920 if (F->getName() != Name) {
921 // Delete the one we just made and get the existing one.
922 F->eraseFromParent();
923 F = TheModule->getFunction(Name);
925 // If F already has a body, reject this.
926 if (!F->empty()) {
927 ErrorF("redefinition of function");
931 // If F took a different number of args, reject.
932 if (F->arg_size() != Args.size()) {
933 ErrorF("redefinition of function with different # args");
938 // Set names for all arguments.
940 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
942 AI->setName(Args[Idx]);
944 // Add arguments to variable symbol table.
945 NamedValues[Args[Idx]] = AI;
951 Function *FunctionAST::Codegen() {
954 Function *TheFunction = Proto->Codegen();
955 if (TheFunction == 0)
958 // Create a new basic block to start insertion into.
959 BasicBlock *BB = BasicBlock::Create("entry", TheFunction);
960 Builder.SetInsertPoint(BB);
962 if (Value *RetVal = Body->Codegen()) {
963 // Finish off the function.
964 Builder.CreateRet(RetVal);
966 // Validate the generated code, checking for consistency.
967 verifyFunction(*TheFunction);
969 // Optimize the function.
970 TheFPM->run(*TheFunction);
975 // Error reading body, remove function.
976 TheFunction->eraseFromParent();
980 //===----------------------------------------------------------------------===//
981 // Top-Level parsing and JIT Driver
982 //===----------------------------------------------------------------------===//
984 static ExecutionEngine *TheExecutionEngine;
986 static void HandleDefinition() {
987 if (FunctionAST *F = ParseDefinition()) {
988 if (Function *LF = F->Codegen()) {
989 fprintf(stderr, "Read function definition:");
993 // Skip token for error recovery.
998 static void HandleExtern() {
999 if (PrototypeAST *P = ParseExtern()) {
1000 if (Function *F = P->Codegen()) {
1001 fprintf(stderr, "Read extern: ");
1005 // Skip token for error recovery.
1010 static void HandleTopLevelExpression() {
1011 // Evaluate a top level expression into an anonymous function.
1012 if (FunctionAST *F = ParseTopLevelExpr()) {
1013 if (Function *LF = F->Codegen()) {
1014 // JIT the function, returning a function pointer.
1015 void *FPtr = TheExecutionEngine->getPointerToFunction(LF);
1017 // Cast it to the right type (takes no arguments, returns a double) so we
1018 // can call it as a native function.
1019 double (*FP)() = (double (*)())FPtr;
1020 fprintf(stderr, "Evaluated to %f\n", FP());
1023 // Skip token for error recovery.
1028 /// top ::= definition | external | expression | ';'
1029 static void MainLoop() {
1031 fprintf(stderr, "ready> ");
1033 case tok_eof: return;
1034 case ';': getNextToken(); break; // ignore top level semicolons.
1035 case tok_def: HandleDefinition(); break;
1036 case tok_extern: HandleExtern(); break;
1037 default: HandleTopLevelExpression(); break;
1044 //===----------------------------------------------------------------------===//
1045 // "Library" functions that can be "extern'd" from user code.
1046 //===----------------------------------------------------------------------===//
1048 /// putchard - putchar that takes a double and returns 0.
1050 double putchard(double X) {
1055 //===----------------------------------------------------------------------===//
1056 // Main driver code.
1057 //===----------------------------------------------------------------------===//
1060 // Install standard binary operators.
1061 // 1 is lowest precedence.
1062 BinopPrecedence['<'] = 10;
1063 BinopPrecedence['+'] = 20;
1064 BinopPrecedence['-'] = 20;
1065 BinopPrecedence['*'] = 40; // highest.
1067 // Prime the first token.
1068 fprintf(stderr, "ready> ");
1071 // Make the module, which holds all the code.
1072 TheModule = new Module("my cool jit");
1075 TheExecutionEngine = ExecutionEngine::create(TheModule);
1078 ExistingModuleProvider OurModuleProvider(TheModule);
1079 FunctionPassManager OurFPM(&OurModuleProvider);
1081 // Set up the optimizer pipeline. Start with registering info about how the
1082 // target lays out data structures.
1083 OurFPM.add(new TargetData(*TheExecutionEngine->getTargetData()));
1084 // Do simple "peephole" optimizations and bit-twiddling optzns.
1085 OurFPM.add(createInstructionCombiningPass());
1086 // Reassociate expressions.
1087 OurFPM.add(createReassociatePass());
1088 // Eliminate Common SubExpressions.
1089 OurFPM.add(createGVNPass());
1090 // Simplify the control flow graph (deleting unreachable blocks, etc).
1091 OurFPM.add(createCFGSimplificationPass());
1093 // Set the global so the code gen can use this.
1094 TheFPM = &OurFPM;
1096 // Run the main "interpreter loop" now.
1101 // Print out all of the generated code.
1102 TheModule->dump();
1103 } // Free module provider (and thus the module) and pass manager.
1110 <a href="LangImpl5.html">Next: Extending the language: control flow</a>
1113 <!-- *********************************************************************** -->
1116 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1117 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1118 <a href="http://validator.w3.org/check/referer"><img
1119 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
1121 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1122 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
1123 Last modified: $Date: 2007-10-17 11:05:13 -0700 (Wed, 17 Oct 2007) $