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="../_static/llvm.css" type="text/css">
14 <h1>Kaleidoscope: Adding JIT and Optimizer Support</h1>
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 <h2><a name="intro">Chapter 4 Introduction</a></h2>
37 <!-- *********************************************************************** -->
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 <h2><a name="trivialconstfold">Trivial Constant Folding</a></h2>
52 <!-- *********************************************************************** -->
57 Our demonstration for Chapter 3 is elegant and easy to extend. Unfortunately,
58 it does not produce wonderful code. The IRBuilder, however, does give us
59 obvious optimizations when compiling simple code:</p>
61 <div class="doc_code">
63 ready> <b>def test(x) 1+2+x;</b>
64 Read function definition:
65 define double @test(double %x) {
67 %addtmp = fadd double 3.000000e+00, %x
73 <p>This code is not a literal transcription of the AST built by parsing the
76 <div class="doc_code">
78 ready> <b>def test(x) 1+2+x;</b>
79 Read function definition:
80 define double @test(double %x) {
82 %addtmp = fadd double 2.000000e+00, 1.000000e+00
83 %addtmp1 = fadd double %addtmp, %x
89 <p>Constant folding, as seen above, in particular, is a very common and very
90 important optimization: so much so that many language implementors implement
91 constant folding support in their AST representation.</p>
93 <p>With LLVM, you don't need this support in the AST. Since all calls to build
94 LLVM IR go through the LLVM IR builder, the builder itself checked to see if
95 there was a constant folding opportunity when you call it. If so, it just does
96 the constant fold and return the constant instead of creating an instruction.
98 <p>Well, that was easy :). In practice, we recommend always using
99 <tt>IRBuilder</tt> when generating code like this. It has no
100 "syntactic overhead" for its use (you don't have to uglify your compiler with
101 constant checks everywhere) and it can dramatically reduce the amount of
102 LLVM IR that is generated in some cases (particular for languages with a macro
103 preprocessor or that use a lot of constants).</p>
105 <p>On the other hand, the <tt>IRBuilder</tt> is limited by the fact
106 that it does all of its analysis inline with the code as it is built. If you
107 take a slightly more complex example:</p>
109 <div class="doc_code">
111 ready> <b>def test(x) (1+2+x)*(x+(1+2));</b>
112 ready> Read function definition:
113 define double @test(double %x) {
115 %addtmp = fadd double 3.000000e+00, %x
116 %addtmp1 = fadd double %x, 3.000000e+00
117 %multmp = fmul double %addtmp, %addtmp1
123 <p>In this case, the LHS and RHS of the multiplication are the same value. We'd
124 really like to see this generate "<tt>tmp = x+3; result = tmp*tmp;</tt>" instead
125 of computing "<tt>x+3</tt>" twice.</p>
127 <p>Unfortunately, no amount of local analysis will be able to detect and correct
128 this. This requires two transformations: reassociation of expressions (to
129 make the add's lexically identical) and Common Subexpression Elimination (CSE)
130 to delete the redundant add instruction. Fortunately, LLVM provides a broad
131 range of optimizations that you can use, in the form of "passes".</p>
135 <!-- *********************************************************************** -->
136 <h2><a name="optimizerpasses">LLVM Optimization Passes</a></h2>
137 <!-- *********************************************************************** -->
141 <p>LLVM provides many optimization passes, which do many different sorts of
142 things and have different tradeoffs. Unlike other systems, LLVM doesn't hold
143 to the mistaken notion that one set of optimizations is right for all languages
144 and for all situations. LLVM allows a compiler implementor to make complete
145 decisions about what optimizations to use, in which order, and in what
148 <p>As a concrete example, LLVM supports both "whole module" passes, which look
149 across as large of body of code as they can (often a whole file, but if run
150 at link time, this can be a substantial portion of the whole program). It also
151 supports and includes "per-function" passes which just operate on a single
152 function at a time, without looking at other functions. For more information
153 on passes and how they are run, see the <a href="../WritingAnLLVMPass.html">How
154 to Write a Pass</a> document and the <a href="../Passes.html">List of LLVM
157 <p>For Kaleidoscope, we are currently generating functions on the fly, one at
158 a time, as the user types them in. We aren't shooting for the ultimate
159 optimization experience in this setting, but we also want to catch the easy and
160 quick stuff where possible. As such, we will choose to run a few per-function
161 optimizations as the user types the function in. If we wanted to make a "static
162 Kaleidoscope compiler", we would use exactly the code we have now, except that
163 we would defer running the optimizer until the entire file has been parsed.</p>
165 <p>In order to get per-function optimizations going, we need to set up a
166 <a href="../WritingAnLLVMPass.html#passmanager">FunctionPassManager</a> to hold and
167 organize the LLVM optimizations that we want to run. Once we have that, we can
168 add a set of optimizations to run. The code looks like this:</p>
170 <div class="doc_code">
172 FunctionPassManager OurFPM(TheModule);
174 // Set up the optimizer pipeline. Start with registering info about how the
175 // target lays out data structures.
176 OurFPM.add(new DataLayout(*TheExecutionEngine->getDataLayout()));
177 // Provide basic AliasAnalysis support for GVN.
178 OurFPM.add(createBasicAliasAnalysisPass());
179 // Do simple "peephole" optimizations and bit-twiddling optzns.
180 OurFPM.add(createInstructionCombiningPass());
181 // Reassociate expressions.
182 OurFPM.add(createReassociatePass());
183 // Eliminate Common SubExpressions.
184 OurFPM.add(createGVNPass());
185 // Simplify the control flow graph (deleting unreachable blocks, etc).
186 OurFPM.add(createCFGSimplificationPass());
188 OurFPM.doInitialization();
190 // Set the global so the code gen can use this.
191 TheFPM = &OurFPM;
193 // Run the main "interpreter loop" now.
198 <p>This code defines a <tt>FunctionPassManager</tt>, "<tt>OurFPM</tt>". It
199 requires a pointer to the <tt>Module</tt> to construct itself. Once it is set
200 up, we use a series of "add" calls to add a bunch of LLVM passes. The first
201 pass is basically boilerplate, it adds a pass so that later optimizations know
202 how the data structures in the program are laid out. The
203 "<tt>TheExecutionEngine</tt>" variable is related to the JIT, which we will get
204 to in the next section.</p>
206 <p>In this case, we choose to add 4 optimization passes. The passes we chose
207 here are a pretty standard set of "cleanup" optimizations that are useful for
208 a wide variety of code. I won't delve into what they do but, believe me,
209 they are a good starting place :).</p>
211 <p>Once the PassManager is set up, we need to make use of it. We do this by
212 running it after our newly created function is constructed (in
213 <tt>FunctionAST::Codegen</tt>), but before it is returned to the client:</p>
215 <div class="doc_code">
217 if (Value *RetVal = Body->Codegen()) {
218 // Finish off the function.
219 Builder.CreateRet(RetVal);
221 // Validate the generated code, checking for consistency.
222 verifyFunction(*TheFunction);
224 <b>// Optimize the function.
225 TheFPM->run(*TheFunction);</b>
232 <p>As you can see, this is pretty straightforward. The
233 <tt>FunctionPassManager</tt> optimizes and updates the LLVM Function* in place,
234 improving (hopefully) its body. With this in place, we can try our test above
237 <div class="doc_code">
239 ready> <b>def test(x) (1+2+x)*(x+(1+2));</b>
240 ready> Read function definition:
241 define double @test(double %x) {
243 %addtmp = fadd double %x, 3.000000e+00
244 %multmp = fmul double %addtmp, %addtmp
250 <p>As expected, we now get our nicely optimized code, saving a floating point
251 add instruction from every execution of this function.</p>
253 <p>LLVM provides a wide variety of optimizations that can be used in certain
254 circumstances. Some <a href="../Passes.html">documentation about the various
255 passes</a> is available, but it isn't very complete. Another good source of
256 ideas can come from looking at the passes that <tt>Clang</tt> runs to get
257 started. The "<tt>opt</tt>" tool allows you to experiment with passes from the
258 command line, so you can see if they do anything.</p>
260 <p>Now that we have reasonable code coming out of our front-end, lets talk about
265 <!-- *********************************************************************** -->
266 <h2><a name="jit">Adding a JIT Compiler</a></h2>
267 <!-- *********************************************************************** -->
271 <p>Code that is available in LLVM IR can have a wide variety of tools
272 applied to it. For example, you can run optimizations on it (as we did above),
273 you can dump it out in textual or binary forms, you can compile the code to an
274 assembly file (.s) for some target, or you can JIT compile it. The nice thing
275 about the LLVM IR representation is that it is the "common currency" between
276 many different parts of the compiler.
279 <p>In this section, we'll add JIT compiler support to our interpreter. The
280 basic idea that we want for Kaleidoscope is to have the user enter function
281 bodies as they do now, but immediately evaluate the top-level expressions they
282 type in. For example, if they type in "1 + 2;", we should evaluate and print
283 out 3. If they define a function, they should be able to call it from the
286 <p>In order to do this, we first declare and initialize the JIT. This is done
287 by adding a global variable and a call in <tt>main</tt>:</p>
289 <div class="doc_code">
291 <b>static ExecutionEngine *TheExecutionEngine;</b>
295 <b>// Create the JIT. This takes ownership of the module.
296 TheExecutionEngine = EngineBuilder(TheModule).create();</b>
302 <p>This creates an abstract "Execution Engine" which can be either a JIT
303 compiler or the LLVM interpreter. LLVM will automatically pick a JIT compiler
304 for you if one is available for your platform, otherwise it will fall back to
307 <p>Once the <tt>ExecutionEngine</tt> is created, the JIT is ready to be used.
308 There are a variety of APIs that are useful, but the simplest one is the
309 "<tt>getPointerToFunction(F)</tt>" method. This method JIT compiles the
310 specified LLVM Function and returns a function pointer to the generated machine
311 code. In our case, this means that we can change the code that parses a
312 top-level expression to look like this:</p>
314 <div class="doc_code">
316 static void HandleTopLevelExpression() {
317 // Evaluate a top-level expression into an anonymous function.
318 if (FunctionAST *F = ParseTopLevelExpr()) {
319 if (Function *LF = F->Codegen()) {
320 LF->dump(); // Dump the function for exposition purposes.
322 <b>// JIT the function, returning a function pointer.
323 void *FPtr = TheExecutionEngine->getPointerToFunction(LF);
325 // Cast it to the right type (takes no arguments, returns a double) so we
326 // can call it as a native function.
327 double (*FP)() = (double (*)())(intptr_t)FPtr;
328 fprintf(stderr, "Evaluated to %f\n", FP());</b>
333 <p>Recall that we compile top-level expressions into a self-contained LLVM
334 function that takes no arguments and returns the computed double. Because the
335 LLVM JIT compiler matches the native platform ABI, this means that you can just
336 cast the result pointer to a function pointer of that type and call it directly.
337 This means, there is no difference between JIT compiled code and native machine
338 code that is statically linked into your application.</p>
340 <p>With just these two changes, lets see how Kaleidoscope works now!</p>
342 <div class="doc_code">
344 ready> <b>4+5;</b>
345 Read top-level expression:
348 ret double 9.000000e+00
351 <em>Evaluated to 9.000000</em>
355 <p>Well this looks like it is basically working. The dump of the function
356 shows the "no argument function that always returns double" that we synthesize
357 for each top-level expression that is typed in. This demonstrates very basic
358 functionality, but can we do more?</p>
360 <div class="doc_code">
362 ready> <b>def testfunc(x y) x + y*2; </b>
363 Read function definition:
364 define double @testfunc(double %x, double %y) {
366 %multmp = fmul double %y, 2.000000e+00
367 %addtmp = fadd double %multmp, %x
371 ready> <b>testfunc(4, 10);</b>
372 Read top-level expression:
375 %calltmp = call double @testfunc(double 4.000000e+00, double 1.000000e+01)
379 <em>Evaluated to 24.000000</em>
383 <p>This illustrates that we can now call user code, but there is something a bit
384 subtle going on here. Note that we only invoke the JIT on the anonymous
385 functions that <em>call testfunc</em>, but we never invoked it
386 on <em>testfunc</em> itself. What actually happened here is that the JIT
387 scanned for all non-JIT'd functions transitively called from the anonymous
388 function and compiled all of them before returning
389 from <tt>getPointerToFunction()</tt>.</p>
391 <p>The JIT provides a number of other more advanced interfaces for things like
392 freeing allocated machine code, rejit'ing functions to update them, etc.
393 However, even with this simple code, we get some surprisingly powerful
394 capabilities - check this out (I removed the dump of the anonymous functions,
395 you should get the idea by now :) :</p>
397 <div class="doc_code">
399 ready> <b>extern sin(x);</b>
401 declare double @sin(double)
403 ready> <b>extern cos(x);</b>
405 declare double @cos(double)
407 ready> <b>sin(1.0);</b>
408 Read top-level expression:
411 ret double 0x3FEAED548F090CEE
414 <em>Evaluated to 0.841471</em>
416 ready> <b>def foo(x) sin(x)*sin(x) + cos(x)*cos(x);</b>
417 Read function definition:
418 define double @foo(double %x) {
420 %calltmp = call double @sin(double %x)
421 %multmp = fmul double %calltmp, %calltmp
422 %calltmp2 = call double @cos(double %x)
423 %multmp4 = fmul double %calltmp2, %calltmp2
424 %addtmp = fadd double %multmp, %multmp4
428 ready> <b>foo(4.0);</b>
429 Read top-level expression:
432 %calltmp = call double @foo(double 4.000000e+00)
436 <em>Evaluated to 1.000000</em>
440 <p>Whoa, how does the JIT know about sin and cos? The answer is surprisingly
442 example, the JIT started execution of a function and got to a function call. It
443 realized that the function was not yet JIT compiled and invoked the standard set
444 of routines to resolve the function. In this case, there is no body defined
445 for the function, so the JIT ended up calling "<tt>dlsym("sin")</tt>" on the
446 Kaleidoscope process itself.
447 Since "<tt>sin</tt>" is defined within the JIT's address space, it simply
448 patches up calls in the module to call the libm version of <tt>sin</tt>
451 <p>The LLVM JIT provides a number of interfaces (look in the
452 <tt>ExecutionEngine.h</tt> file) for controlling how unknown functions get
453 resolved. It allows you to establish explicit mappings between IR objects and
454 addresses (useful for LLVM global variables that you want to map to static
455 tables, for example), allows you to dynamically decide on the fly based on the
456 function name, and even allows you to have the JIT compile functions lazily the
457 first time they're called.</p>
459 <p>One interesting application of this is that we can now extend the language
460 by writing arbitrary C++ code to implement operations. For example, if we add:
463 <div class="doc_code">
465 /// putchard - putchar that takes a double and returns 0.
467 double putchard(double X) {
474 <p>Now we can produce simple output to the console by using things like:
475 "<tt>extern putchard(x); putchard(120);</tt>", which prints a lowercase 'x' on
476 the console (120 is the ASCII code for 'x'). Similar code could be used to
477 implement file I/O, console input, and many other capabilities in
480 <p>This completes the JIT and optimizer chapter of the Kaleidoscope tutorial. At
481 this point, we can compile a non-Turing-complete programming language, optimize
482 and JIT compile it in a user-driven way. Next up we'll look into <a
483 href="LangImpl5.html">extending the language with control flow constructs</a>,
484 tackling some interesting LLVM IR issues along the way.</p>
488 <!-- *********************************************************************** -->
489 <h2><a name="code">Full Code Listing</a></h2>
490 <!-- *********************************************************************** -->
495 Here is the complete code listing for our running example, enhanced with the
496 LLVM JIT and optimizer. To build this example, use:
499 <div class="doc_code">
502 clang++ -g toy.cpp `llvm-config --cppflags --ldflags --libs core jit native` -O3 -o toy
509 If you are compiling this on Linux, make sure to add the "-rdynamic" option
510 as well. This makes sure that the external functions are resolved properly
513 <p>Here is the code:</p>
515 <div class="doc_code">
517 #include "llvm/DerivedTypes.h"
518 #include "llvm/ExecutionEngine/ExecutionEngine.h"
519 #include "llvm/ExecutionEngine/JIT.h"
520 #include "llvm/IRBuilder.h"
521 #include "llvm/LLVMContext.h"
522 #include "llvm/Module.h"
523 #include "llvm/PassManager.h"
524 #include "llvm/Analysis/Verifier.h"
525 #include "llvm/Analysis/Passes.h"
526 #include "llvm/DataLayout.h"
527 #include "llvm/Transforms/Scalar.h"
528 #include "llvm/Support/TargetSelect.h"
529 #include <cstdio>
530 #include <string>
532 #include <vector>
533 using namespace llvm;
535 //===----------------------------------------------------------------------===//
537 //===----------------------------------------------------------------------===//
539 // The lexer returns tokens [0-255] if it is an unknown character, otherwise one
540 // of these for known things.
545 tok_def = -2, tok_extern = -3,
548 tok_identifier = -4, tok_number = -5
551 static std::string IdentifierStr; // Filled in if tok_identifier
552 static double NumVal; // Filled in if tok_number
554 /// gettok - Return the next token from standard input.
555 static int gettok() {
556 static int LastChar = ' ';
558 // Skip any whitespace.
559 while (isspace(LastChar))
560 LastChar = getchar();
562 if (isalpha(LastChar)) { // identifier: [a-zA-Z][a-zA-Z0-9]*
563 IdentifierStr = LastChar;
564 while (isalnum((LastChar = getchar())))
565 IdentifierStr += LastChar;
567 if (IdentifierStr == "def") return tok_def;
568 if (IdentifierStr == "extern") return tok_extern;
569 return tok_identifier;
572 if (isdigit(LastChar) || LastChar == '.') { // Number: [0-9.]+
576 LastChar = getchar();
577 } while (isdigit(LastChar) || LastChar == '.');
579 NumVal = strtod(NumStr.c_str(), 0);
583 if (LastChar == '#') {
584 // Comment until end of line.
585 do LastChar = getchar();
586 while (LastChar != EOF && LastChar != '\n' && LastChar != '\r');
592 // Check for end of file. Don't eat the EOF.
596 // Otherwise, just return the character as its ascii value.
597 int ThisChar = LastChar;
598 LastChar = getchar();
602 //===----------------------------------------------------------------------===//
603 // Abstract Syntax Tree (aka Parse Tree)
604 //===----------------------------------------------------------------------===//
606 /// ExprAST - Base class for all expression nodes.
609 virtual ~ExprAST() {}
610 virtual Value *Codegen() = 0;
613 /// NumberExprAST - Expression class for numeric literals like "1.0".
614 class NumberExprAST : public ExprAST {
617 NumberExprAST(double val) : Val(val) {}
618 virtual Value *Codegen();
621 /// VariableExprAST - Expression class for referencing a variable, like "a".
622 class VariableExprAST : public ExprAST {
625 VariableExprAST(const std::string &name) : Name(name) {}
626 virtual Value *Codegen();
629 /// BinaryExprAST - Expression class for a binary operator.
630 class BinaryExprAST : public ExprAST {
634 BinaryExprAST(char op, ExprAST *lhs, ExprAST *rhs)
635 : Op(op), LHS(lhs), RHS(rhs) {}
636 virtual Value *Codegen();
639 /// CallExprAST - Expression class for function calls.
640 class CallExprAST : public ExprAST {
642 std::vector<ExprAST*> Args;
644 CallExprAST(const std::string &callee, std::vector<ExprAST*> &args)
645 : Callee(callee), Args(args) {}
646 virtual Value *Codegen();
649 /// PrototypeAST - This class represents the "prototype" for a function,
650 /// which captures its name, and its argument names (thus implicitly the number
651 /// of arguments the function takes).
654 std::vector<std::string> Args;
656 PrototypeAST(const std::string &name, const std::vector<std::string> &args)
657 : Name(name), Args(args) {}
662 /// FunctionAST - This class represents a function definition itself.
667 FunctionAST(PrototypeAST *proto, ExprAST *body)
668 : Proto(proto), Body(body) {}
673 //===----------------------------------------------------------------------===//
675 //===----------------------------------------------------------------------===//
677 /// CurTok/getNextToken - Provide a simple token buffer. CurTok is the current
678 /// token the parser is looking at. getNextToken reads another token from the
679 /// lexer and updates CurTok with its results.
681 static int getNextToken() {
682 return CurTok = gettok();
685 /// BinopPrecedence - This holds the precedence for each binary operator that is
687 static std::map<char, int> BinopPrecedence;
689 /// GetTokPrecedence - Get the precedence of the pending binary operator token.
690 static int GetTokPrecedence() {
691 if (!isascii(CurTok))
694 // Make sure it's a declared binop.
695 int TokPrec = BinopPrecedence[CurTok];
696 if (TokPrec <= 0) return -1;
700 /// Error* - These are little helper functions for error handling.
701 ExprAST *Error(const char *Str) { fprintf(stderr, "Error: %s\n", Str);return 0;}
702 PrototypeAST *ErrorP(const char *Str) { Error(Str); return 0; }
703 FunctionAST *ErrorF(const char *Str) { Error(Str); return 0; }
705 static ExprAST *ParseExpression();
709 /// ::= identifier '(' expression* ')'
710 static ExprAST *ParseIdentifierExpr() {
711 std::string IdName = IdentifierStr;
713 getNextToken(); // eat identifier.
715 if (CurTok != '(') // Simple variable ref.
716 return new VariableExprAST(IdName);
719 getNextToken(); // eat (
720 std::vector<ExprAST*> Args;
723 ExprAST *Arg = ParseExpression();
727 if (CurTok == ')') break;
730 return Error("Expected ')' or ',' in argument list");
738 return new CallExprAST(IdName, Args);
741 /// numberexpr ::= number
742 static ExprAST *ParseNumberExpr() {
743 ExprAST *Result = new NumberExprAST(NumVal);
744 getNextToken(); // consume the number
748 /// parenexpr ::= '(' expression ')'
749 static ExprAST *ParseParenExpr() {
750 getNextToken(); // eat (.
751 ExprAST *V = ParseExpression();
755 return Error("expected ')'");
756 getNextToken(); // eat ).
761 /// ::= identifierexpr
764 static ExprAST *ParsePrimary() {
766 default: return Error("unknown token when expecting an expression");
767 case tok_identifier: return ParseIdentifierExpr();
768 case tok_number: return ParseNumberExpr();
769 case '(': return ParseParenExpr();
774 /// ::= ('+' primary)*
775 static ExprAST *ParseBinOpRHS(int ExprPrec, ExprAST *LHS) {
776 // If this is a binop, find its precedence.
778 int TokPrec = GetTokPrecedence();
780 // If this is a binop that binds at least as tightly as the current binop,
781 // consume it, otherwise we are done.
782 if (TokPrec < ExprPrec)
785 // Okay, we know this is a binop.
787 getNextToken(); // eat binop
789 // Parse the primary expression after the binary operator.
790 ExprAST *RHS = ParsePrimary();
793 // If BinOp binds less tightly with RHS than the operator after RHS, let
794 // the pending operator take RHS as its LHS.
795 int NextPrec = GetTokPrecedence();
796 if (TokPrec < NextPrec) {
797 RHS = ParseBinOpRHS(TokPrec+1, RHS);
798 if (RHS == 0) return 0;
802 LHS = new BinaryExprAST(BinOp, LHS, RHS);
807 /// ::= primary binoprhs
809 static ExprAST *ParseExpression() {
810 ExprAST *LHS = ParsePrimary();
813 return ParseBinOpRHS(0, LHS);
817 /// ::= id '(' id* ')'
818 static PrototypeAST *ParsePrototype() {
819 if (CurTok != tok_identifier)
820 return ErrorP("Expected function name in prototype");
822 std::string FnName = IdentifierStr;
826 return ErrorP("Expected '(' in prototype");
828 std::vector<std::string> ArgNames;
829 while (getNextToken() == tok_identifier)
830 ArgNames.push_back(IdentifierStr);
832 return ErrorP("Expected ')' in prototype");
835 getNextToken(); // eat ')'.
837 return new PrototypeAST(FnName, ArgNames);
840 /// definition ::= 'def' prototype expression
841 static FunctionAST *ParseDefinition() {
842 getNextToken(); // eat def.
843 PrototypeAST *Proto = ParsePrototype();
844 if (Proto == 0) return 0;
846 if (ExprAST *E = ParseExpression())
847 return new FunctionAST(Proto, E);
851 /// toplevelexpr ::= expression
852 static FunctionAST *ParseTopLevelExpr() {
853 if (ExprAST *E = ParseExpression()) {
854 // Make an anonymous proto.
855 PrototypeAST *Proto = new PrototypeAST("", std::vector<std::string>());
856 return new FunctionAST(Proto, E);
861 /// external ::= 'extern' prototype
862 static PrototypeAST *ParseExtern() {
863 getNextToken(); // eat extern.
864 return ParsePrototype();
867 //===----------------------------------------------------------------------===//
869 //===----------------------------------------------------------------------===//
871 static Module *TheModule;
872 static IRBuilder<> Builder(getGlobalContext());
873 static std::map<std::string, Value*> NamedValues;
874 static FunctionPassManager *TheFPM;
876 Value *ErrorV(const char *Str) { Error(Str); return 0; }
878 Value *NumberExprAST::Codegen() {
879 return ConstantFP::get(getGlobalContext(), APFloat(Val));
882 Value *VariableExprAST::Codegen() {
883 // Look this variable up in the function.
884 Value *V = NamedValues[Name];
885 return V ? V : ErrorV("Unknown variable name");
888 Value *BinaryExprAST::Codegen() {
889 Value *L = LHS->Codegen();
890 Value *R = RHS->Codegen();
891 if (L == 0 || R == 0) return 0;
894 case '+': return Builder.CreateFAdd(L, R, "addtmp");
895 case '-': return Builder.CreateFSub(L, R, "subtmp");
896 case '*': return Builder.CreateFMul(L, R, "multmp");
898 L = Builder.CreateFCmpULT(L, R, "cmptmp");
899 // Convert bool 0/1 to double 0.0 or 1.0
900 return Builder.CreateUIToFP(L, Type::getDoubleTy(getGlobalContext()),
902 default: return ErrorV("invalid binary operator");
906 Value *CallExprAST::Codegen() {
907 // Look up the name in the global module table.
908 Function *CalleeF = TheModule->getFunction(Callee);
910 return ErrorV("Unknown function referenced");
912 // If argument mismatch error.
913 if (CalleeF->arg_size() != Args.size())
914 return ErrorV("Incorrect # arguments passed");
916 std::vector<Value*> ArgsV;
917 for (unsigned i = 0, e = Args.size(); i != e; ++i) {
918 ArgsV.push_back(Args[i]->Codegen());
919 if (ArgsV.back() == 0) return 0;
922 return Builder.CreateCall(CalleeF, ArgsV, "calltmp");
925 Function *PrototypeAST::Codegen() {
926 // Make the function type: double(double,double) etc.
927 std::vector<Type*> Doubles(Args.size(),
928 Type::getDoubleTy(getGlobalContext()));
929 FunctionType *FT = FunctionType::get(Type::getDoubleTy(getGlobalContext()),
932 Function *F = Function::Create(FT, Function::ExternalLinkage, Name, TheModule);
934 // If F conflicted, there was already something named 'Name'. If it has a
935 // body, don't allow redefinition or reextern.
936 if (F->getName() != Name) {
937 // Delete the one we just made and get the existing one.
938 F->eraseFromParent();
939 F = TheModule->getFunction(Name);
941 // If F already has a body, reject this.
942 if (!F->empty()) {
943 ErrorF("redefinition of function");
947 // If F took a different number of args, reject.
948 if (F->arg_size() != Args.size()) {
949 ErrorF("redefinition of function with different # args");
954 // Set names for all arguments.
956 for (Function::arg_iterator AI = F->arg_begin(); Idx != Args.size();
958 AI->setName(Args[Idx]);
960 // Add arguments to variable symbol table.
961 NamedValues[Args[Idx]] = AI;
967 Function *FunctionAST::Codegen() {
970 Function *TheFunction = Proto->Codegen();
971 if (TheFunction == 0)
974 // Create a new basic block to start insertion into.
975 BasicBlock *BB = BasicBlock::Create(getGlobalContext(), "entry", TheFunction);
976 Builder.SetInsertPoint(BB);
978 if (Value *RetVal = Body->Codegen()) {
979 // Finish off the function.
980 Builder.CreateRet(RetVal);
982 // Validate the generated code, checking for consistency.
983 verifyFunction(*TheFunction);
985 // Optimize the function.
986 TheFPM->run(*TheFunction);
991 // Error reading body, remove function.
992 TheFunction->eraseFromParent();
996 //===----------------------------------------------------------------------===//
997 // Top-Level parsing and JIT Driver
998 //===----------------------------------------------------------------------===//
1000 static ExecutionEngine *TheExecutionEngine;
1002 static void HandleDefinition() {
1003 if (FunctionAST *F = ParseDefinition()) {
1004 if (Function *LF = F->Codegen()) {
1005 fprintf(stderr, "Read function definition:");
1009 // Skip token for error recovery.
1014 static void HandleExtern() {
1015 if (PrototypeAST *P = ParseExtern()) {
1016 if (Function *F = P->Codegen()) {
1017 fprintf(stderr, "Read extern: ");
1021 // Skip token for error recovery.
1026 static void HandleTopLevelExpression() {
1027 // Evaluate a top-level expression into an anonymous function.
1028 if (FunctionAST *F = ParseTopLevelExpr()) {
1029 if (Function *LF = F->Codegen()) {
1030 fprintf(stderr, "Read top-level expression:");
1033 // JIT the function, returning a function pointer.
1034 void *FPtr = TheExecutionEngine->getPointerToFunction(LF);
1036 // Cast it to the right type (takes no arguments, returns a double) so we
1037 // can call it as a native function.
1038 double (*FP)() = (double (*)())(intptr_t)FPtr;
1039 fprintf(stderr, "Evaluated to %f\n", FP());
1042 // Skip token for error recovery.
1047 /// top ::= definition | external | expression | ';'
1048 static void MainLoop() {
1050 fprintf(stderr, "ready> ");
1052 case tok_eof: return;
1053 case ';': getNextToken(); break; // ignore top-level semicolons.
1054 case tok_def: HandleDefinition(); break;
1055 case tok_extern: HandleExtern(); break;
1056 default: HandleTopLevelExpression(); break;
1061 //===----------------------------------------------------------------------===//
1062 // "Library" functions that can be "extern'd" from user code.
1063 //===----------------------------------------------------------------------===//
1065 /// putchard - putchar that takes a double and returns 0.
1067 double putchard(double X) {
1072 //===----------------------------------------------------------------------===//
1073 // Main driver code.
1074 //===----------------------------------------------------------------------===//
1077 InitializeNativeTarget();
1078 LLVMContext &Context = getGlobalContext();
1080 // Install standard binary operators.
1081 // 1 is lowest precedence.
1082 BinopPrecedence['<'] = 10;
1083 BinopPrecedence['+'] = 20;
1084 BinopPrecedence['-'] = 20;
1085 BinopPrecedence['*'] = 40; // highest.
1087 // Prime the first token.
1088 fprintf(stderr, "ready> ");
1091 // Make the module, which holds all the code.
1092 TheModule = new Module("my cool jit", Context);
1094 // Create the JIT. This takes ownership of the module.
1096 TheExecutionEngine = EngineBuilder(TheModule).setErrorStr(&ErrStr).create();
1097 if (!TheExecutionEngine) {
1098 fprintf(stderr, "Could not create ExecutionEngine: %s\n", ErrStr.c_str());
1102 FunctionPassManager OurFPM(TheModule);
1104 // Set up the optimizer pipeline. Start with registering info about how the
1105 // target lays out data structures.
1106 OurFPM.add(new DataLayout(*TheExecutionEngine->getDataLayout()));
1107 // Provide basic AliasAnalysis support for GVN.
1108 OurFPM.add(createBasicAliasAnalysisPass());
1109 // Do simple "peephole" optimizations and bit-twiddling optzns.
1110 OurFPM.add(createInstructionCombiningPass());
1111 // Reassociate expressions.
1112 OurFPM.add(createReassociatePass());
1113 // Eliminate Common SubExpressions.
1114 OurFPM.add(createGVNPass());
1115 // Simplify the control flow graph (deleting unreachable blocks, etc).
1116 OurFPM.add(createCFGSimplificationPass());
1118 OurFPM.doInitialization();
1120 // Set the global so the code gen can use this.
1121 TheFPM = &OurFPM;
1123 // Run the main "interpreter loop" now.
1128 // Print out all of the generated code.
1129 TheModule->dump();
1136 <a href="LangImpl5.html">Next: Extending the language: control flow</a>
1139 <!-- *********************************************************************** -->
1142 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1143 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1144 <a href="http://validator.w3.org/check/referer"><img
1145 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
1147 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1148 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
1149 Last modified: $Date$