Gitiles
Code Review Sign In
gerrit-public.fairphone.software / fp2-dev / platform / external / llvm / 9ef76b9985a0c408e126affa049698c413ad8664 / . / docs / tutorial / OCamlLangImpl4.html
blob: 116c618d02507dc4b294c2afca7d5d53789014f4 [file] [log] [blame]
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
3
4<html>
5<head>
6 <title>Kaleidoscope: 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 <meta name="author" content="Erick Tryzelaar">
10 <link rel="stylesheet" href="../llvm.css" type="text/css">
11</head>
12
13<body>
14
15<div class="doc_title">Kaleidoscope: Adding JIT and Optimizer Support</div>
16
17<ul>
18<li><a href="index.html">Up to Tutorial Index</a></li>
19<li>Chapter 4
20 <ol>
21 <li><a href="#intro">Chapter 4 Introduction</a></li>
22 <li><a href="#trivialconstfold">Trivial Constant Folding</a></li>
23 <li><a href="#optimizerpasses">LLVM Optimization Passes</a></li>
24 <li><a href="#jit">Adding a JIT Compiler</a></li>
25 <li><a href="#code">Full Code Listing</a></li>
26 </ol>
27</li>
28<li><a href="OCamlLangImpl5.html">Chapter 5</a>: Extending the Language: Control
29Flow</li>
30</ul>
31
32<div class="doc_author">
33 <p>
34 Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
35 and <a href="mailto:idadesub@users.sourceforge.net">Erick Tryzelaar</a>
36 </p>
37</div>
38
39<!-- *********************************************************************** -->
40<div class="doc_section"><a name="intro">Chapter 4 Introduction</a></div>
41<!-- *********************************************************************** -->
42
43<div class="doc_text">
44
45<p>Welcome to Chapter 4 of the "<a href="index.html">Implementing a language
46with LLVM</a>" tutorial. Chapters 1-3 described the implementation of a simple
47language and added support for generating LLVM IR. This chapter describes
48two new techniques: adding optimizer support to your language, and adding JIT
49compiler support. These additions will demonstrate how to get nice, efficient code
50for the Kaleidoscope language.</p>
51
52</div>
53
54<!-- *********************************************************************** -->
55<div class="doc_section"><a name="trivialconstfold">Trivial Constant
56Folding</a></div>
57<!-- *********************************************************************** -->
58
59<div class="doc_text">
60
Duncan Sands89f6d882008-04-13 06:22:09 +0000[diff] [blame]61<p><b>Note:</b> the default <tt>IRBuilder</tt> now always includes the constant
62folding optimisations below.<p>
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]63
64<p>
65Our demonstration for Chapter 3 is elegant and easy to extend. Unfortunately,
66it does not produce wonderful code. For example, when compiling simple code,
67we don't get obvious optimizations:</p>
68
69<div class="doc_code">
70<pre>
71ready&gt; <b>def test(x) 1+2+x;</b>
72Read function definition:
73define double @test(double %x) {
74entry:
Dan Gohmana9445e12010-03-02 01:11:08 +0000[diff] [blame]75 %addtmp = fadd double 1.000000e+00, 2.000000e+00
76 %addtmp1 = fadd double %addtmp, %x
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]77 ret double %addtmp1
78}
79</pre>
80</div>
81
82<p>This code is a very, very literal transcription of the AST built by parsing
83the input. As such, this transcription lacks optimizations like constant folding
84(we'd like to get "<tt>add x, 3.0</tt>" in the example above) as well as other
85more important optimizations. Constant folding, in particular, is a very common
86and very important optimization: so much so that many language implementors
87implement constant folding support in their AST representation.</p>
88
89<p>With LLVM, you don't need this support in the AST. Since all calls to build
90LLVM IR go through the LLVM builder, it would be nice if the builder itself
91checked to see if there was a constant folding opportunity when you call it.
92If so, it could just do the constant fold and return the constant instead of
93creating an instruction. This is exactly what the <tt>LLVMFoldingBuilder</tt>
94class does.
95
96<p>All we did was switch from <tt>LLVMBuilder</tt> to
97<tt>LLVMFoldingBuilder</tt>. Though we change no other code, we now have all of our
98instructions implicitly constant folded without us having to do anything
99about it. For example, the input above now compiles to:</p>
100
101<div class="doc_code">
102<pre>
103ready&gt; <b>def test(x) 1+2+x;</b>
104Read function definition:
105define double @test(double %x) {
106entry:
Dan Gohmana9445e12010-03-02 01:11:08 +0000[diff] [blame]107 %addtmp = fadd double 3.000000e+00, %x
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]108 ret double %addtmp
109}
110</pre>
111</div>
112
113<p>Well, that was easy :). In practice, we recommend always using
114<tt>LLVMFoldingBuilder</tt> when generating code like this. It has no
115"syntactic overhead" for its use (you don't have to uglify your compiler with
116constant checks everywhere) and it can dramatically reduce the amount of
117LLVM IR that is generated in some cases (particular for languages with a macro
118preprocessor or that use a lot of constants).</p>
119
120<p>On the other hand, the <tt>LLVMFoldingBuilder</tt> is limited by the fact
121that it does all of its analysis inline with the code as it is built. If you
122take a slightly more complex example:</p>
123
124<div class="doc_code">
125<pre>
126ready&gt; <b>def test(x) (1+2+x)*(x+(1+2));</b>
127ready&gt; Read function definition:
128define double @test(double %x) {
129entry:
Dan Gohmana9445e12010-03-02 01:11:08 +0000[diff] [blame]130 %addtmp = fadd double 3.000000e+00, %x
131 %addtmp1 = fadd double %x, 3.000000e+00
132 %multmp = fmul double %addtmp, %addtmp1
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]133 ret double %multmp
134}
135</pre>
136</div>
137
138<p>In this case, the LHS and RHS of the multiplication are the same value. We'd
139really like to see this generate "<tt>tmp = x+3; result = tmp*tmp;</tt>" instead
140of computing "<tt>x*3</tt>" twice.</p>
141
142<p>Unfortunately, no amount of local analysis will be able to detect and correct
143this. This requires two transformations: reassociation of expressions (to
144make the add's lexically identical) and Common Subexpression Elimination (CSE)
145to delete the redundant add instruction. Fortunately, LLVM provides a broad
146range of optimizations that you can use, in the form of "passes".</p>
147
148</div>
149
150<!-- *********************************************************************** -->
151<div class="doc_section"><a name="optimizerpasses">LLVM Optimization
152 Passes</a></div>
153<!-- *********************************************************************** -->
154
155<div class="doc_text">
156
157<p>LLVM provides many optimization passes, which do many different sorts of
158things and have different tradeoffs. Unlike other systems, LLVM doesn't hold
159to the mistaken notion that one set of optimizations is right for all languages
160and for all situations. LLVM allows a compiler implementor to make complete
161decisions about what optimizations to use, in which order, and in what
162situation.</p>
163
164<p>As a concrete example, LLVM supports both "whole module" passes, which look
165across as large of body of code as they can (often a whole file, but if run
166at link time, this can be a substantial portion of the whole program). It also
167supports and includes "per-function" passes which just operate on a single
168function at a time, without looking at other functions. For more information
169on passes and how they are run, see the <a href="../WritingAnLLVMPass.html">How
170to Write a Pass</a> document and the <a href="../Passes.html">List of LLVM
171Passes</a>.</p>
172
173<p>For Kaleidoscope, we are currently generating functions on the fly, one at
174a time, as the user types them in. We aren't shooting for the ultimate
175optimization experience in this setting, but we also want to catch the easy and
176quick stuff where possible. As such, we will choose to run a few per-function
177optimizations as the user types the function in. If we wanted to make a "static
178Kaleidoscope compiler", we would use exactly the code we have now, except that
179we would defer running the optimizer until the entire file has been parsed.</p>
180
181<p>In order to get per-function optimizations going, we need to set up a
182<a href="../WritingAnLLVMPass.html#passmanager">Llvm.PassManager</a> to hold and
183organize the LLVM optimizations that we want to run. Once we have that, we can
184add a set of optimizations to run. The code looks like this:</p>
185
186<div class="doc_code">
187<pre>
188 (* Create the JIT. *)
Erick Tryzelaar9ef76b92010-03-08 19:32:18 +0000[diff] [blame^]189 let the_execution_engine = ExecutionEngine.create Codegen.the_module in
190 let the_fpm = PassManager.create_function Codegen.the_module in
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]191
192 (* Set up the optimizer pipeline. Start with registering info about how the
193 * target lays out data structures. *)
194 TargetData.add (ExecutionEngine.target_data the_execution_engine) the_fpm;
195
196 (* Do simple "peephole" optimizations and bit-twiddling optzn. *)
197 add_instruction_combining the_fpm;
198
199 (* reassociate expressions. *)
200 add_reassociation the_fpm;
201
202 (* Eliminate Common SubExpressions. *)
203 add_gvn the_fpm;
204
205 (* Simplify the control flow graph (deleting unreachable blocks, etc). *)
206 add_cfg_simplification the_fpm;
207
Erick Tryzelaar126d86b2009-09-14 21:54:15 +0000[diff] [blame]208 ignore (PassManager.initialize the_fpm);
209
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]210 (* Run the main "interpreter loop" now. *)
211 Toplevel.main_loop the_fpm the_execution_engine stream;
212</pre>
213</div>
214
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]215<p>The meat of the matter here, is the definition of "<tt>the_fpm</tt>". It
Erick Tryzelaar9ef76b92010-03-08 19:32:18 +0000[diff] [blame^]216requires a pointer to the <tt>the_module</tt> to construct itself. Once it is
217set up, we use a series of "add" calls to add a bunch of LLVM passes. The
218first pass is basically boilerplate, it adds a pass so that later optimizations
219know how the data structures in the program are laid out. The
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]220"<tt>the_execution_engine</tt>" variable is related to the JIT, which we will
221get to in the next section.</p>
222
223<p>In this case, we choose to add 4 optimization passes. The passes we chose
224here are a pretty standard set of "cleanup" optimizations that are useful for
225a wide variety of code. I won't delve into what they do but, believe me,
226they are a good starting place :).</p>
227
228<p>Once the <tt>Llvm.PassManager.</tt> is set up, we need to make use of it.
229We do this by running it after our newly created function is constructed (in
230<tt>Codegen.codegen_func</tt>), but before it is returned to the client:</p>
231
232<div class="doc_code">
233<pre>
234let codegen_func the_fpm = function
Erick Tryzelaar35295ff2008-03-31 08:44:50 +0000[diff] [blame]235 ...
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]236 try
237 let ret_val = codegen_expr body in
238
239 (* Finish off the function. *)
240 let _ = build_ret ret_val builder in
241
242 (* Validate the generated code, checking for consistency. *)
243 Llvm_analysis.assert_valid_function the_function;
244
245 (* Optimize the function. *)
246 let _ = PassManager.run_function the_function the_fpm in
247
248 the_function
249</pre>
250</div>
251
252<p>As you can see, this is pretty straightforward. The <tt>the_fpm</tt>
253optimizes and updates the LLVM Function* in place, improving (hopefully) its
254body. With this in place, we can try our test above again:</p>
255
256<div class="doc_code">
257<pre>
258ready&gt; <b>def test(x) (1+2+x)*(x+(1+2));</b>
259ready&gt; Read function definition:
260define double @test(double %x) {
261entry:
Dan Gohmana9445e12010-03-02 01:11:08 +0000[diff] [blame]262 %addtmp = fadd double %x, 3.000000e+00
263 %multmp = fmul double %addtmp, %addtmp
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]264 ret double %multmp
265}
266</pre>
267</div>
268
269<p>As expected, we now get our nicely optimized code, saving a floating point
270add instruction from every execution of this function.</p>
271
272<p>LLVM provides a wide variety of optimizations that can be used in certain
273circumstances. Some <a href="../Passes.html">documentation about the various
274passes</a> is available, but it isn't very complete. Another good source of
275ideas can come from looking at the passes that <tt>llvm-gcc</tt> or
276<tt>llvm-ld</tt> run to get started. The "<tt>opt</tt>" tool allows you to
277experiment with passes from the command line, so you can see if they do
278anything.</p>
279
280<p>Now that we have reasonable code coming out of our front-end, lets talk about
281executing it!</p>
282
283</div>
284
285<!-- *********************************************************************** -->
286<div class="doc_section"><a name="jit">Adding a JIT Compiler</a></div>
287<!-- *********************************************************************** -->
288
289<div class="doc_text">
290
291<p>Code that is available in LLVM IR can have a wide variety of tools
292applied to it. For example, you can run optimizations on it (as we did above),
293you can dump it out in textual or binary forms, you can compile the code to an
294assembly file (.s) for some target, or you can JIT compile it. The nice thing
295about the LLVM IR representation is that it is the "common currency" between
296many different parts of the compiler.
297</p>
298
299<p>In this section, we'll add JIT compiler support to our interpreter. The
300basic idea that we want for Kaleidoscope is to have the user enter function
301bodies as they do now, but immediately evaluate the top-level expressions they
302type in. For example, if they type in "1 + 2;", we should evaluate and print
303out 3. If they define a function, they should be able to call it from the
304command line.</p>
305
306<p>In order to do this, we first declare and initialize the JIT. This is done
307by adding a global variable and a call in <tt>main</tt>:</p>
308
309<div class="doc_code">
310<pre>
311...
312let main () =
313 ...
Erick Tryzelaar35295ff2008-03-31 08:44:50 +0000[diff] [blame]314 <b>(* Create the JIT. *)
Erick Tryzelaar9ef76b92010-03-08 19:32:18 +0000[diff] [blame^]315 let the_execution_engine = ExecutionEngine.create Codegen.the_module in</b>
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]316 ...
317</pre>
318</div>
319
320<p>This creates an abstract "Execution Engine" which can be either a JIT
321compiler or the LLVM interpreter. LLVM will automatically pick a JIT compiler
322for you if one is available for your platform, otherwise it will fall back to
323the interpreter.</p>
324
325<p>Once the <tt>Llvm_executionengine.ExecutionEngine.t</tt> is created, the JIT
326is ready to be used. There are a variety of APIs that are useful, but the
327simplest one is the "<tt>Llvm_executionengine.ExecutionEngine.run_function</tt>"
328function. This method JIT compiles the specified LLVM Function and returns a
329function pointer to the generated machine code. In our case, this means that we
330can change the code that parses a top-level expression to look like this:</p>
331
332<div class="doc_code">
333<pre>
334 (* Evaluate a top-level expression into an anonymous function. *)
335 let e = Parser.parse_toplevel stream in
336 print_endline "parsed a top-level expr";
337 let the_function = Codegen.codegen_func the_fpm e in
338 dump_value the_function;
339
340 (* JIT the function, returning a function pointer. *)
341 let result = ExecutionEngine.run_function the_function [||]
342 the_execution_engine in
343
344 print_string "Evaluated to ";
Erick Tryzelaar9ef76b92010-03-08 19:32:18 +0000[diff] [blame^]345 print_float (GenericValue.as_float Codegen.double_type result);
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]346 print_newline ();
347</pre>
348</div>
349
350<p>Recall that we compile top-level expressions into a self-contained LLVM
351function that takes no arguments and returns the computed double. Because the
352LLVM JIT compiler matches the native platform ABI, this means that you can just
353cast the result pointer to a function pointer of that type and call it directly.
354This means, there is no difference between JIT compiled code and native machine
355code that is statically linked into your application.</p>
356
357<p>With just these two changes, lets see how Kaleidoscope works now!</p>
358
359<div class="doc_code">
360<pre>
361ready&gt; <b>4+5;</b>
362define double @""() {
363entry:
364 ret double 9.000000e+00
365}
366
367<em>Evaluated to 9.000000</em>
368</pre>
369</div>
370
371<p>Well this looks like it is basically working. The dump of the function
372shows the "no argument function that always returns double" that we synthesize
373for each top level expression that is typed in. This demonstrates very basic
374functionality, but can we do more?</p>
375
376<div class="doc_code">
377<pre>
378ready&gt; <b>def testfunc(x y) x + y*2; </b>
379Read function definition:
380define double @testfunc(double %x, double %y) {
381entry:
Dan Gohmana9445e12010-03-02 01:11:08 +0000[diff] [blame]382 %multmp = fmul double %y, 2.000000e+00
383 %addtmp = fadd double %multmp, %x
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]384 ret double %addtmp
385}
386
387ready&gt; <b>testfunc(4, 10);</b>
388define double @""() {
389entry:
390 %calltmp = call double @testfunc( double 4.000000e+00, double 1.000000e+01 )
391 ret double %calltmp
392}
393
394<em>Evaluated to 24.000000</em>
395</pre>
396</div>
397
398<p>This illustrates that we can now call user code, but there is something a bit
399subtle going on here. Note that we only invoke the JIT on the anonymous
Jeffrey Yasskindc857242009-10-27 20:30:28 +0000[diff] [blame]400functions that <em>call testfunc</em>, but we never invoked it
401on <em>testfunc</em> itself. What actually happened here is that the JIT
402scanned for all non-JIT'd functions transitively called from the anonymous
403function and compiled all of them before returning
404from <tt>run_function</tt>.</p>
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]405
Jeffrey Yasskindc857242009-10-27 20:30:28 +0000[diff] [blame]406<p>The JIT provides a number of other more advanced interfaces for things like
407freeing allocated machine code, rejit'ing functions to update them, etc.
408However, even with this simple code, we get some surprisingly powerful
409capabilities - check this out (I removed the dump of the anonymous functions,
410you should get the idea by now :) :</p>
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]411
412<div class="doc_code">
413<pre>
414ready&gt; <b>extern sin(x);</b>
415Read extern:
416declare double @sin(double)
417
418ready&gt; <b>extern cos(x);</b>
419Read extern:
420declare double @cos(double)
421
422ready&gt; <b>sin(1.0);</b>
423<em>Evaluated to 0.841471</em>
424
425ready&gt; <b>def foo(x) sin(x)*sin(x) + cos(x)*cos(x);</b>
426Read function definition:
427define double @foo(double %x) {
428entry:
429 %calltmp = call double @sin( double %x )
Dan Gohmana9445e12010-03-02 01:11:08 +0000[diff] [blame]430 %multmp = fmul double %calltmp, %calltmp
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]431 %calltmp2 = call double @cos( double %x )
Dan Gohmana9445e12010-03-02 01:11:08 +0000[diff] [blame]432 %multmp4 = fmul double %calltmp2, %calltmp2
433 %addtmp = fadd double %multmp, %multmp4
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]434 ret double %addtmp
435}
436
437ready&gt; <b>foo(4.0);</b>
438<em>Evaluated to 1.000000</em>
439</pre>
440</div>
441
442<p>Whoa, how does the JIT know about sin and cos? The answer is surprisingly
443simple: in this example, the JIT started execution of a function and got to a
444function call. It realized that the function was not yet JIT compiled and
445invoked the standard set of routines to resolve the function. In this case,
446there is no body defined for the function, so the JIT ended up calling
447"<tt>dlsym("sin")</tt>" on the Kaleidoscope process itself. Since
448"<tt>sin</tt>" is defined within the JIT's address space, it simply patches up
449calls in the module to call the libm version of <tt>sin</tt> directly.</p>
450
451<p>The LLVM JIT provides a number of interfaces (look in the
452<tt>llvm_executionengine.mli</tt> file) for controlling how unknown functions
453get resolved. It allows you to establish explicit mappings between IR objects
454and addresses (useful for LLVM global variables that you want to map to static
455tables, for example), allows you to dynamically decide on the fly based on the
Jeffrey Yasskindc857242009-10-27 20:30:28 +0000[diff] [blame]456function name, and even allows you to have the JIT compile functions lazily the
457first time they're called.</p>
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]458
459<p>One interesting application of this is that we can now extend the language
460by writing arbitrary C code to implement operations. For example, if we add:
461</p>
462
463<div class="doc_code">
464<pre>
465/* putchard - putchar that takes a double and returns 0. */
466extern "C"
467double putchard(double X) {
468 putchar((char)X);
469 return 0;
470}
471</pre>
472</div>
473
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
476the console (120 is the ASCII code for 'x'). Similar code could be used to
477implement file I/O, console input, and many other capabilities in
478Kaleidoscope.</p>
479
480<p>This completes the JIT and optimizer chapter of the Kaleidoscope tutorial. At
481this point, we can compile a non-Turing-complete programming language, optimize
482and JIT compile it in a user-driven way. Next up we'll look into <a
483href="OCamlLangImpl5.html">extending the language with control flow
484constructs</a>, tackling some interesting LLVM IR issues along the way.</p>
485
486</div>
487
488<!-- *********************************************************************** -->
489<div class="doc_section"><a name="code">Full Code Listing</a></div>
490<!-- *********************************************************************** -->
491
492<div class="doc_text">
493
494<p>
495Here is the complete code listing for our running example, enhanced with the
496LLVM JIT and optimizer. To build this example, use:
497</p>
498
Erick Tryzelaar35295ff2008-03-31 08:44:50 +0000[diff] [blame]499<div class="doc_code">
500<pre>
501# Compile
502ocamlbuild toy.byte
503# Run
504./toy.byte
505</pre>
506</div>
507
508<p>Here is the code:</p>
509
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]510<dl>
511<dt>_tags:</dt>
512<dd class="doc_code">
513<pre>
514&lt;{lexer,parser}.ml&gt;: use_camlp4, pp(camlp4of)
515&lt;*.{byte,native}&gt;: g++, use_llvm, use_llvm_analysis
516&lt;*.{byte,native}&gt;: use_llvm_executionengine, use_llvm_target
517&lt;*.{byte,native}&gt;: use_llvm_scalar_opts, use_bindings
518</pre>
519</dd>
520
521<dt>myocamlbuild.ml:</dt>
522<dd class="doc_code">
523<pre>
524open Ocamlbuild_plugin;;
525
526ocaml_lib ~extern:true "llvm";;
527ocaml_lib ~extern:true "llvm_analysis";;
528ocaml_lib ~extern:true "llvm_executionengine";;
529ocaml_lib ~extern:true "llvm_target";;
530ocaml_lib ~extern:true "llvm_scalar_opts";;
531
532flag ["link"; "ocaml"; "g++"] (S[A"-cc"; A"g++"]);;
533dep ["link"; "ocaml"; "use_bindings"] ["bindings.o"];;
534</pre>
535</dd>
536
537<dt>token.ml:</dt>
538<dd class="doc_code">
539<pre>
540(*===----------------------------------------------------------------------===
541 * Lexer Tokens
542 *===----------------------------------------------------------------------===*)
543
544(* The lexer returns these 'Kwd' if it is an unknown character, otherwise one of
545 * these others for known things. *)
546type token =
547 (* commands *)
548 | Def | Extern
549
550 (* primary *)
551 | Ident of string | Number of float
552
553 (* unknown *)
554 | Kwd of char
555</pre>
556</dd>
557
558<dt>lexer.ml:</dt>
559<dd class="doc_code">
560<pre>
561(*===----------------------------------------------------------------------===
562 * Lexer
563 *===----------------------------------------------------------------------===*)
564
565let rec lex = parser
566 (* Skip any whitespace. *)
567 | [&lt; ' (' ' | '\n' | '\r' | '\t'); stream &gt;] -&gt; lex stream
568
569 (* identifier: [a-zA-Z][a-zA-Z0-9] *)
570 | [&lt; ' ('A' .. 'Z' | 'a' .. 'z' as c); stream &gt;] -&gt;
571 let buffer = Buffer.create 1 in
572 Buffer.add_char buffer c;
573 lex_ident buffer stream
574
575 (* number: [0-9.]+ *)
576 | [&lt; ' ('0' .. '9' as c); stream &gt;] -&gt;
577 let buffer = Buffer.create 1 in
578 Buffer.add_char buffer c;
579 lex_number buffer stream
580
581 (* Comment until end of line. *)
582 | [&lt; ' ('#'); stream &gt;] -&gt;
583 lex_comment stream
584
585 (* Otherwise, just return the character as its ascii value. *)
586 | [&lt; 'c; stream &gt;] -&gt;
587 [&lt; 'Token.Kwd c; lex stream &gt;]
588
589 (* end of stream. *)
590 | [&lt; &gt;] -&gt; [&lt; &gt;]
591
592and lex_number buffer = parser
593 | [&lt; ' ('0' .. '9' | '.' as c); stream &gt;] -&gt;
594 Buffer.add_char buffer c;
595 lex_number buffer stream
596 | [&lt; stream=lex &gt;] -&gt;
597 [&lt; 'Token.Number (float_of_string (Buffer.contents buffer)); stream &gt;]
598
599and lex_ident buffer = parser
600 | [&lt; ' ('A' .. 'Z' | 'a' .. 'z' | '0' .. '9' as c); stream &gt;] -&gt;
601 Buffer.add_char buffer c;
602 lex_ident buffer stream
603 | [&lt; stream=lex &gt;] -&gt;
604 match Buffer.contents buffer with
605 | "def" -&gt; [&lt; 'Token.Def; stream &gt;]
606 | "extern" -&gt; [&lt; 'Token.Extern; stream &gt;]
607 | id -&gt; [&lt; 'Token.Ident id; stream &gt;]
608
609and lex_comment = parser
610 | [&lt; ' ('\n'); stream=lex &gt;] -&gt; stream
611 | [&lt; 'c; e=lex_comment &gt;] -&gt; e
612 | [&lt; &gt;] -&gt; [&lt; &gt;]
613</pre>
614</dd>
615
616<dt>ast.ml:</dt>
617<dd class="doc_code">
618<pre>
619(*===----------------------------------------------------------------------===
620 * Abstract Syntax Tree (aka Parse Tree)
621 *===----------------------------------------------------------------------===*)
622
623(* expr - Base type for all expression nodes. *)
624type expr =
625 (* variant for numeric literals like "1.0". *)
626 | Number of float
627
628 (* variant for referencing a variable, like "a". *)
629 | Variable of string
630
631 (* variant for a binary operator. *)
632 | Binary of char * expr * expr
633
634 (* variant for function calls. *)
635 | Call of string * expr array
636
637(* proto - This type represents the "prototype" for a function, which captures
638 * its name, and its argument names (thus implicitly the number of arguments the
639 * function takes). *)
640type proto = Prototype of string * string array
641
642(* func - This type represents a function definition itself. *)
643type func = Function of proto * expr
644</pre>
645</dd>
646
647<dt>parser.ml:</dt>
648<dd class="doc_code">
649<pre>
650(*===---------------------------------------------------------------------===
651 * Parser
652 *===---------------------------------------------------------------------===*)
653
654(* binop_precedence - This holds the precedence for each binary operator that is
655 * defined *)
656let binop_precedence:(char, int) Hashtbl.t = Hashtbl.create 10
657
658(* precedence - Get the precedence of the pending binary operator token. *)
659let precedence c = try Hashtbl.find binop_precedence c with Not_found -&gt; -1
660
661(* primary
662 * ::= identifier
663 * ::= numberexpr
664 * ::= parenexpr *)
665let rec parse_primary = parser
666 (* numberexpr ::= number *)
667 | [&lt; 'Token.Number n &gt;] -&gt; Ast.Number n
668
669 (* parenexpr ::= '(' expression ')' *)
670 | [&lt; 'Token.Kwd '('; e=parse_expr; 'Token.Kwd ')' ?? "expected ')'" &gt;] -&gt; e
671
672 (* identifierexpr
673 * ::= identifier
674 * ::= identifier '(' argumentexpr ')' *)
675 | [&lt; 'Token.Ident id; stream &gt;] -&gt;
676 let rec parse_args accumulator = parser
677 | [&lt; e=parse_expr; stream &gt;] -&gt;
678 begin parser
679 | [&lt; 'Token.Kwd ','; e=parse_args (e :: accumulator) &gt;] -&gt; e
680 | [&lt; &gt;] -&gt; e :: accumulator
681 end stream
682 | [&lt; &gt;] -&gt; accumulator
683 in
684 let rec parse_ident id = parser
685 (* Call. *)
686 | [&lt; 'Token.Kwd '(';
687 args=parse_args [];
688 'Token.Kwd ')' ?? "expected ')'"&gt;] -&gt;
689 Ast.Call (id, Array.of_list (List.rev args))
690
691 (* Simple variable ref. *)
692 | [&lt; &gt;] -&gt; Ast.Variable id
693 in
694 parse_ident id stream
695
696 | [&lt; &gt;] -&gt; raise (Stream.Error "unknown token when expecting an expression.")
697
698(* binoprhs
699 * ::= ('+' primary)* *)
700and parse_bin_rhs expr_prec lhs stream =
701 match Stream.peek stream with
702 (* If this is a binop, find its precedence. *)
703 | Some (Token.Kwd c) when Hashtbl.mem binop_precedence c -&gt;
704 let token_prec = precedence c in
705
706 (* If this is a binop that binds at least as tightly as the current binop,
707 * consume it, otherwise we are done. *)
708 if token_prec &lt; expr_prec then lhs else begin
709 (* Eat the binop. *)
710 Stream.junk stream;
711
712 (* Parse the primary expression after the binary operator. *)
713 let rhs = parse_primary stream in
714
715 (* Okay, we know this is a binop. *)
716 let rhs =
717 match Stream.peek stream with
718 | Some (Token.Kwd c2) -&gt;
719 (* If BinOp binds less tightly with rhs than the operator after
720 * rhs, let the pending operator take rhs as its lhs. *)
721 let next_prec = precedence c2 in
722 if token_prec &lt; next_prec
723 then parse_bin_rhs (token_prec + 1) rhs stream
724 else rhs
725 | _ -&gt; rhs
726 in
727
728 (* Merge lhs/rhs. *)
729 let lhs = Ast.Binary (c, lhs, rhs) in
730 parse_bin_rhs expr_prec lhs stream
731 end
732 | _ -&gt; lhs
733
734(* expression
735 * ::= primary binoprhs *)
736and parse_expr = parser
737 | [&lt; lhs=parse_primary; stream &gt;] -&gt; parse_bin_rhs 0 lhs stream
738
739(* prototype
740 * ::= id '(' id* ')' *)
741let parse_prototype =
742 let rec parse_args accumulator = parser
743 | [&lt; 'Token.Ident id; e=parse_args (id::accumulator) &gt;] -&gt; e
744 | [&lt; &gt;] -&gt; accumulator
745 in
746
747 parser
748 | [&lt; 'Token.Ident id;
749 'Token.Kwd '(' ?? "expected '(' in prototype";
750 args=parse_args [];
751 'Token.Kwd ')' ?? "expected ')' in prototype" &gt;] -&gt;
752 (* success. *)
753 Ast.Prototype (id, Array.of_list (List.rev args))
754
755 | [&lt; &gt;] -&gt;
756 raise (Stream.Error "expected function name in prototype")
757
758(* definition ::= 'def' prototype expression *)
759let parse_definition = parser
760 | [&lt; 'Token.Def; p=parse_prototype; e=parse_expr &gt;] -&gt;
761 Ast.Function (p, e)
762
763(* toplevelexpr ::= expression *)
764let parse_toplevel = parser
765 | [&lt; e=parse_expr &gt;] -&gt;
766 (* Make an anonymous proto. *)
767 Ast.Function (Ast.Prototype ("", [||]), e)
768
769(* external ::= 'extern' prototype *)
770let parse_extern = parser
771 | [&lt; 'Token.Extern; e=parse_prototype &gt;] -&gt; e
772</pre>
773</dd>
774
775<dt>codegen.ml:</dt>
776<dd class="doc_code">
777<pre>
778(*===----------------------------------------------------------------------===
779 * Code Generation
780 *===----------------------------------------------------------------------===*)
781
782open Llvm
783
784exception Error of string
785
Erick Tryzelaar1f3d2762009-08-19 17:32:38 +0000[diff] [blame]786let context = global_context ()
787let the_module = create_module context "my cool jit"
788let builder = builder context
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]789let named_values:(string, llvalue) Hashtbl.t = Hashtbl.create 10
Erick Tryzelaar9ef76b92010-03-08 19:32:18 +0000[diff] [blame^]790let double_type = double_type context
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]791
792let rec codegen_expr = function
793 | Ast.Number n -&gt; const_float double_type n
794 | Ast.Variable name -&gt;
795 (try Hashtbl.find named_values name with
796 | Not_found -&gt; raise (Error "unknown variable name"))
797 | Ast.Binary (op, lhs, rhs) -&gt;
798 let lhs_val = codegen_expr lhs in
799 let rhs_val = codegen_expr rhs in
800 begin
801 match op with
802 | '+' -&gt; build_add lhs_val rhs_val "addtmp" builder
803 | '-' -&gt; build_sub lhs_val rhs_val "subtmp" builder
804 | '*' -&gt; build_mul lhs_val rhs_val "multmp" builder
805 | '&lt;' -&gt;
806 (* Convert bool 0/1 to double 0.0 or 1.0 *)
807 let i = build_fcmp Fcmp.Ult lhs_val rhs_val "cmptmp" builder in
808 build_uitofp i double_type "booltmp" builder
809 | _ -&gt; raise (Error "invalid binary operator")
810 end
811 | Ast.Call (callee, args) -&gt;
812 (* Look up the name in the module table. *)
813 let callee =
814 match lookup_function callee the_module with
815 | Some callee -&gt; callee
816 | None -&gt; raise (Error "unknown function referenced")
817 in
818 let params = params callee in
819
820 (* If argument mismatch error. *)
821 if Array.length params == Array.length args then () else
822 raise (Error "incorrect # arguments passed");
823 let args = Array.map codegen_expr args in
824 build_call callee args "calltmp" builder
825
826let codegen_proto = function
827 | Ast.Prototype (name, args) -&gt;
828 (* Make the function type: double(double,double) etc. *)
829 let doubles = Array.make (Array.length args) double_type in
830 let ft = function_type double_type doubles in
831 let f =
832 match lookup_function name the_module with
833 | None -&gt; declare_function name ft the_module
834
835 (* If 'f' conflicted, there was already something named 'name'. If it
836 * has a body, don't allow redefinition or reextern. *)
837 | Some f -&gt;
838 (* If 'f' already has a body, reject this. *)
839 if block_begin f &lt;&gt; At_end f then
840 raise (Error "redefinition of function");
841
842 (* If 'f' took a different number of arguments, reject. *)
843 if element_type (type_of f) &lt;&gt; ft then
844 raise (Error "redefinition of function with different # args");
845 f
846 in
847
848 (* Set names for all arguments. *)
849 Array.iteri (fun i a -&gt;
850 let n = args.(i) in
851 set_value_name n a;
852 Hashtbl.add named_values n a;
853 ) (params f);
854 f
855
856let codegen_func the_fpm = function
857 | Ast.Function (proto, body) -&gt;
858 Hashtbl.clear named_values;
859 let the_function = codegen_proto proto in
860
861 (* Create a new basic block to start insertion into. *)
Erick Tryzelaar9ef76b92010-03-08 19:32:18 +0000[diff] [blame^]862 let bb = append_block context "entry" the_function in
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]863 position_at_end bb builder;
864
865 try
866 let ret_val = codegen_expr body in
867
868 (* Finish off the function. *)
869 let _ = build_ret ret_val builder in
870
871 (* Validate the generated code, checking for consistency. *)
872 Llvm_analysis.assert_valid_function the_function;
873
874 (* Optimize the function. *)
875 let _ = PassManager.run_function the_function the_fpm in
876
877 the_function
878 with e -&gt;
879 delete_function the_function;
880 raise e
881</pre>
882</dd>
883
884<dt>toplevel.ml:</dt>
885<dd class="doc_code">
886<pre>
887(*===----------------------------------------------------------------------===
888 * Top-Level parsing and JIT Driver
889 *===----------------------------------------------------------------------===*)
890
891open Llvm
892open Llvm_executionengine
893
894(* top ::= definition | external | expression | ';' *)
895let rec main_loop the_fpm the_execution_engine stream =
896 match Stream.peek stream with
897 | None -&gt; ()
898
899 (* ignore top-level semicolons. *)
900 | Some (Token.Kwd ';') -&gt;
901 Stream.junk stream;
902 main_loop the_fpm the_execution_engine stream
903
904 | Some token -&gt;
905 begin
906 try match token with
907 | Token.Def -&gt;
908 let e = Parser.parse_definition stream in
909 print_endline "parsed a function definition.";
910 dump_value (Codegen.codegen_func the_fpm e);
911 | Token.Extern -&gt;
912 let e = Parser.parse_extern stream in
913 print_endline "parsed an extern.";
914 dump_value (Codegen.codegen_proto e);
915 | _ -&gt;
916 (* Evaluate a top-level expression into an anonymous function. *)
917 let e = Parser.parse_toplevel stream in
918 print_endline "parsed a top-level expr";
919 let the_function = Codegen.codegen_func the_fpm e in
920 dump_value the_function;
921
922 (* JIT the function, returning a function pointer. *)
923 let result = ExecutionEngine.run_function the_function [||]
924 the_execution_engine in
925
926 print_string "Evaluated to ";
Erick Tryzelaar9ef76b92010-03-08 19:32:18 +0000[diff] [blame^]927 print_float (GenericValue.as_float Codegen.double_type result);
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]928 print_newline ();
929 with Stream.Error s | Codegen.Error s -&gt;
930 (* Skip token for error recovery. *)
931 Stream.junk stream;
932 print_endline s;
933 end;
934 print_string "ready&gt; "; flush stdout;
935 main_loop the_fpm the_execution_engine stream
936</pre>
937</dd>
938
939<dt>toy.ml:</dt>
940<dd class="doc_code">
941<pre>
942(*===----------------------------------------------------------------------===
943 * Main driver code.
944 *===----------------------------------------------------------------------===*)
945
946open Llvm
947open Llvm_executionengine
948open Llvm_target
949open Llvm_scalar_opts
950
951let main () =
Erick Tryzelaar46262682009-09-14 21:54:32 +0000[diff] [blame]952 ignore (initialize_native_target ());
953
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]954 (* Install standard binary operators.
955 * 1 is the lowest precedence. *)
956 Hashtbl.add Parser.binop_precedence '&lt;' 10;
957 Hashtbl.add Parser.binop_precedence '+' 20;
958 Hashtbl.add Parser.binop_precedence '-' 20;
959 Hashtbl.add Parser.binop_precedence '*' 40; (* highest. *)
960
961 (* Prime the first token. *)
962 print_string "ready&gt; "; flush stdout;
963 let stream = Lexer.lex (Stream.of_channel stdin) in
964
965 (* Create the JIT. *)
Erick Tryzelaar9ef76b92010-03-08 19:32:18 +0000[diff] [blame^]966 let the_execution_engine = ExecutionEngine.create Codegen.the_module in
967 let the_fpm = PassManager.create_function Codegen.the_module in
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]968
969 (* Set up the optimizer pipeline. Start with registering info about how the
970 * target lays out data structures. *)
971 TargetData.add (ExecutionEngine.target_data the_execution_engine) the_fpm;
972
973 (* Do simple "peephole" optimizations and bit-twiddling optzn. *)
Erick Tryzelaar9ef76b92010-03-08 19:32:18 +0000[diff] [blame^]974 add_instruction_combination the_fpm;
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]975
976 (* reassociate expressions. *)
977 add_reassociation the_fpm;
978
979 (* Eliminate Common SubExpressions. *)
980 add_gvn the_fpm;
981
982 (* Simplify the control flow graph (deleting unreachable blocks, etc). *)
983 add_cfg_simplification the_fpm;
984
Erick Tryzelaar126d86b2009-09-14 21:54:15 +0000[diff] [blame]985 ignore (PassManager.initialize the_fpm);
986
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]987 (* Run the main "interpreter loop" now. *)
988 Toplevel.main_loop the_fpm the_execution_engine stream;
989
990 (* Print out all the generated code. *)
991 dump_module Codegen.the_module
992;;
993
994main ()
995</pre>
996</dd>
997
998<dt>bindings.c</dt>
999<dd class="doc_code">
1000<pre>
1001#include &lt;stdio.h&gt;
1002
1003/* putchard - putchar that takes a double and returns 0. */
1004extern double putchard(double X) {
1005 putchar((char)X);
1006 return 0;
1007}
1008</pre>
1009</dd>
1010</dl>
1011
1012<a href="OCamlLangImpl5.html">Next: Extending the language: control flow</a>
1013</div>
1014
1015<!-- *********************************************************************** -->
1016<hr>
1017<address>
1018 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1019 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1020 <a href="http://validator.w3.org/check/referer"><img
1021 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
1022
1023 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1024 <a href="mailto:idadesub@users.sourceforge.net">Erick Tryzelaar</a><br>
1025 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
Dan Gohman523e3922010-02-03 17:27:31 +0000[diff] [blame]1026 Last modified: $Date$
Erick Tryzelaar37c076b2008-03-30 09:57:12 +0000[diff] [blame]1027</address>
1028</body>
1029</html>