Gitiles
Code Review Sign In
gerrit-public.fairphone.software / fp2-dev / platform / external / llvm / efcc3f5c5e12c4558dbde25fba0986f511686d42 / . / docs / tutorial / LangImpl8.html
blob: a92fac832be87ebb5f35608841a5fc78e09de9d6 [file] [log] [blame]
Chris Lattnerb8fc6502007-11-05 01:58:13 +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>
Chris Lattner99005a42007-11-05 19:10:15 +0000[diff] [blame]6 <title>Kaleidoscope: Conclusion and other useful LLVM tidbits</title>
Chris Lattnerb8fc6502007-11-05 01:58:13 +0000[diff] [blame]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">
10</head>
11
12<body>
13
Chris Lattner99005a42007-11-05 19:10:15 +0000[diff] [blame]14<div class="doc_title">Kaleidoscope: Conclusion and other useful LLVM
15 tidbits</div>
16
17<ul>
Chris Lattner0e555b12007-11-05 20:04:56 +0000[diff] [blame]18<li><a href="index.html">Up to Tutorial Index</a></li>
Chris Lattner99005a42007-11-05 19:10:15 +0000[diff] [blame]19<li>Chapter 8
20 <ol>
21 <li><a href="#conclusion">Tutorial Conclusion</a></li>
22 <li><a href="#llvmirproperties">Properties of LLVM IR</a>
23 <ul>
24 <li><a href="#targetindep">Target Independence</a></li>
25 <li><a href="#safety">Safety Guarantees</a></li>
26 <li><a href="#langspecific">Language-Specific Optimizations</a></li>
27 </ul>
28 </li>
29 <li><a href="#tipsandtricks">Tips and Tricks</a>
30 <ul>
31 <li><a href="#offsetofsizeof">Implementing portable
32 offsetof/sizeof</a></li>
33 <li><a href="#gcstack">Garbage Collected Stack Frames</a></li>
34 </ul>
35 </li>
36 </ol>
37</li>
38</ul>
39
Chris Lattnerb8fc6502007-11-05 01:58:13 +0000[diff] [blame]40
41<div class="doc_author">
42 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
43</div>
44
45<!-- *********************************************************************** -->
Chris Lattner99005a42007-11-05 19:10:15 +0000[diff] [blame]46<div class="doc_section"><a name="conclusion">Tutorial Conclusion</a></div>
Chris Lattnerb8fc6502007-11-05 01:58:13 +0000[diff] [blame]47<!-- *********************************************************************** -->
48
49<div class="doc_text">
50
51<p>Welcome to the the final chapter of the "<a href="index.html">Implementing a
52language with LLVM</a>" tutorial. In the course of this tutorial, we have grown
53our little Kaleidoscope language from being a useless toy, to being a
54semi-interesting (but probably still useless) toy. :)</p>
55
56<p>It is interesting to see how far we've come, and how little code it has
57taken. We built the entire lexer, parser, AST, code generator, and an
58interactive run-loop (with a JIT!) by-hand in under 700 lines of
59(non-comment/non-blank) code.</p>
60
61<p>Our little language supports a couple of interesting features: it supports
62user defined binary and unary operators, it uses JIT compilation for immediate
63evaluation, and it supports a few control flow constructs with SSA construction.
64</p>
65
66<p>Part of the idea of this tutorial was to show you how easy and fun it can be
67to define, build, and play with languages. Building a compiler need not be a
68scary or mystical process! Now that you've seen some of the basics, I strongly
69encourage you to take the code and hack on it. For example, try adding:</p>
70
71<ul>
72<li><b>global variables</b> - While global variables have questional value in
73modern software engineering, they are often useful when putting together quick
74little hacks like the Kaleidoscope compiler itself. Fortunately, our current
75setup makes it very easy to add global variables: just have value lookup check
76to see if an unresolved variable is in the global variable symbol table before
77rejecting it. To create a new global variable, make an instance of the LLVM
78<tt>GlobalVariable</tt> class.</li>
79
80<li><b>typed variables</b> - Kaleidoscope currently only supports variables of
81type double. This gives the language a very nice elegance, because only
82supporting one type means that you never have to specify types. Different
83languages have different ways of handling this. The easiest way is to require
84the user to specify types for every variable definition, and record the type
85of the variable in the symbol table along with its Value*.</li>
86
87<li><b>arrays, structs, vectors, etc</b> - Once you add types, you can start
88extending the type system in all sorts of interesting ways. Simple arrays are
89very easy and are quite useful for many different applications. Adding them is
90mostly an exercise in learning how the LLVM <a
91href="../LangRef.html#i_getelementptr">getelementptr</a> instruction works.
92The getelementptr instruction is so nifty/unconventional, it <a
Chris Lattner9ca08f32007-11-05 19:28:07 +0000[diff] [blame]93href="../GetElementPtr.html">has its own FAQ</a>!). If you add support
94for recursive types (e.g. linked lists), make sure to read the <a
95href="../ProgrammersManual.html#TypeResolve">section in the LLVM
96Programmer's Manual</a> that describes how to construct them.</li>
Chris Lattnerb8fc6502007-11-05 01:58:13 +0000[diff] [blame]97
98<li><b>standard runtime</b> - Our current language allows the user to access
99arbitrary external functions, and we use it for things like "printd" and
100"putchard". As you extend the language to add higher-level constructs, often
101these constructs make the most amount of sense to be lowered into calls into a
102language-supplied runtime. For example, if you add hash tables to the language,
103it would probably make sense to add the routines to a runtime, instead of
104inlining them all the way.</li>
105
106<li><b>memory management</b> - Currently we can only access the stack in
107Kaleidoscope. It would also be useful to be able to allocate heap memory,
108either with calls to the standard libc malloc/free interface or with a garbage
109collector. If you choose to use garbage collection, note that LLVM fully
110supports <a href="../GarbageCollection.html">Accurate Garbage Collection</a>
111including algorithms that move objects and need to scan/update the stack.</li>
112
113<li><b>debugger support</b> - LLVM supports generation of <a
114href="../SourceLevelDebugging.html">DWARF Debug info</a> which is understood by
115common debuggers like GDB. Adding support for debug info is fairly
116straight-forward. The best way to understand it is to compile some C/C++ code
117with "<tt>llvm-gcc -g -O0</tt>" and taking a look at what it produces.</li>
118
Chris Lattnera3f07ef2007-11-05 07:00:54 +0000[diff] [blame]119<li><b>exception handling support</b> - LLVM supports generation of <a
Chris Lattnerb8fc6502007-11-05 01:58:13 +0000[diff] [blame]120href="../ExceptionHandling.html">zero cost exceptions</a> which interoperate
121with code compiled in other languages. You could also generate code by
122implicitly making every function return an error value and checking it. You
123could also make explicit use of setjmp/longjmp. There are many different ways
124to go here.</li>
125
126<li><b>object orientation, generics, database access, complex numbers,
127geometric programming, ...</b> - Really, there is
128no end of crazy features that you can add to the language.</li>
129
Chris Lattnera3f07ef2007-11-05 07:00:54 +0000[diff] [blame]130<li><b>unusual domains</b> - We've been talking about applying LLVM to a domain
131that many people are interested in: building a compiler for a specific language.
132However, there are many other domains that can use compiler technology that are
133not typically considered. For example, LLVM has been used to implement OpenGL
134graphics acceleration, translate C++ code to ActionScript, and many other
135cute and clever things. Maybe you will be the first to JIT compile a regular
136expression interpreter into native code with LLVM?</li>
137
Chris Lattnerb8fc6502007-11-05 01:58:13 +0000[diff] [blame]138</ul>
139
140<p>
141Have fun - try doing something crazy and unusual. Building a language like
142everyone else always has is much less fun than trying something a little crazy
143and off the wall and seeing how it turns out. If you get stuck or want to talk
144about it, feel free to email the <a
145href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">llvmdev mailing
146list</a>: it has lots of people who are interested in languages and are often
147willing to help out.
148</p>
149
150<p>Before we end, I want to talk about some "tips and tricks" for generating
151LLVM IR. These are some of the more subtle things that may not be obvious, but
152are very useful if you want to take advantage of LLVM's capabilities.</p>
153
154</div>
155
156<!-- *********************************************************************** -->
Chris Lattnera3f07ef2007-11-05 07:00:54 +0000[diff] [blame]157<div class="doc_section"><a name="llvmirproperties">Properties of LLVM
158IR</a></div>
159<!-- *********************************************************************** -->
160
161<div class="doc_text">
162
163<p>We have a couple common questions about code in the LLVM IR form, lets just
164get these out of the way right now shall we?</p>
165
166</div>
167
168<!-- ======================================================================= -->
169<div class="doc_subsubsection"><a name="targetindep">Target
170Independence</a></div>
171<!-- ======================================================================= -->
172
173<div class="doc_text">
174
175<p>Kaleidoscope is an example of a "portable language": any program written in
176Kaleidoscope will work the same way on any target that it runs on. Many other
177languages have this property, e.g. lisp, java, haskell, javascript, python, etc
178(note that while these languages are portable, not all their libraries are).</p>
179
180<p>One nice aspect of LLVM is that it is often capable of preserving language
181independence in the IR: you can take the LLVM IR for a Kaleidoscope-compiled
182program and run it on any target that LLVM supports, even emitting C code and
183compiling that on targets that LLVM doesn't support natively. You can trivially
184tell that the Kaleidoscope compiler generates target-independent code because it
185never queries for any target-specific information when generating code.</p>
186
187<p>The fact that LLVM provides a compact target-independent representation for
188code gets a lot of people excited. Unfortunately, these people are usually
189thinking about C or a language from the C family when they are asking questions
190about language portability. I say "unfortunately", because there is really no
191way to make (fully general) C code portable, other than shipping the source code
192around (and of course, C source code is not actually portable in general
193either - ever port a really old application from 32- to 64-bits?).</p>
194
195<p>The problem with C (again, in its full generality) is that it is heavily
196laden with target specific assumptions. As one simple example, the preprocessor
197often destructively removes target-independence from the code when it processes
198the input text:</p>
199
200<div class="doc_code">
201<pre>
202#ifdef __i386__
203 int X = 1;
204#else
205 int X = 42;
206#endif
207</pre>
208</div>
209
210<p>While it is possible to engineer more and more complex solutions to problems
211like this, it cannot be solved in full generality in a way better than shipping
212the actual source code.</p>
213
214<p>That said, there are interesting subsets of C that can be made portable. If
215you are willing to fix primitive types to a fixed size (say int = 32-bits,
216and long = 64-bits), don't care about ABI compatibility with existing binaries,
217and are willing to give up some other minor features, you can have portable
218code. This can even make real sense for specialized domains such as an
219in-kernel language.</p>
220
221</div>
222
223<!-- ======================================================================= -->
224<div class="doc_subsubsection"><a name="safety">Safety Guarantees</a></div>
225<!-- ======================================================================= -->
226
227<div class="doc_text">
228
229<p>Many of the languages above are also "safe" languages: it is impossible for
230a program written in Java to corrupt its address space and crash the process.
231Safety is an interesting property that requires a combination of language
232design, runtime support, and often operating system support.</p>
233
234<p>It is certainly possible to implement a safe language in LLVM, but LLVM IR
235does not itself guarantee safety. The LLVM IR allows unsafe pointer casts,
236use after free bugs, buffer over-runs, and a variety of other problems. Safety
237needs to be implemented as a layer on top of LLVM and, conveniently, several
238groups have investigated this. Ask on the <a
239href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">llvmdev mailing
240list</a> if you are interested in more details.</p>
241
242</div>
243
244<!-- ======================================================================= -->
245<div class="doc_subsubsection"><a name="langspecific">Language-Specific
246Optimizations</a></div>
247<!-- ======================================================================= -->
248
249<div class="doc_text">
250
251<p>One thing about LLVM that turns off many people is that it does not solve all
252the world's problems in one system (sorry 'world hunger', someone else will have
253to solve you some other day). One specific complaint is that people perceive
254LLVM as being incapable of performing high-level language-specific optimization:
255LLVM "loses too much information".</p>
256
257<p>Unfortunately, this is really not the place to give you a full and unified
258version of "Chris Lattner's theory of compiler design". Instead, I'll make a
259few observations:</p>
260
261<p>First, you're right that LLVM does lose information. For example, as of this
262writing, there is no way to distinguish in the LLVM IR whether an SSA-value came
263from a C "int" or a C "long" on an ILP32 machine (other than debug info). Both
264get compiled down to an 'i32' value and the information about what it came from
265is lost. The more general issue here is that the LLVM type system uses
266"structural equivalence" instead of "name equivalence". Another place this
267surprises people is if you have two types in a high-level language that have the
268same structure (e.g. two different structs that have a single int field): these
269types will compile down into a single LLVM type and it will be impossible to
270tell what it came from.</p>
271
272<p>Second, while LLVM does lose information, LLVM is not a fixed target: we
273continue to enhance and improve it in many different ways. In addition to
274adding new features (LLVM did not always support exceptions or debug info), we
275also extend the IR to capture important information for optimization (e.g.
276whether an argument is sign or zero extended, information about pointers
277aliasing, etc. Many of the enhancements are user-driven: people want LLVM to
278do some specific feature, so they go ahead and extend it to do so.</p>
279
280<p>Third, it <em>is certainly possible</em> to add language-specific
281optimizations, and you have a number of choices in how to do it. As one trivial
282example, it is possible to add language-specific optimization passes that
Chris Lattnerefcc3f52007-11-06 01:40:41 +0000[diff] [blame^]283"know" things about code compiled for a language. In the case of the C family,
Chris Lattnera3f07ef2007-11-05 07:00:54 +0000[diff] [blame]284there is an optimziation pass that "knows" about the standard C library
285functions. If you call "exit(0)" in main(), it knows that it is safe to
286optimize that into "return 0;" for example, because C specifies what the 'exit'
287function does.</p>
288
289<p>In addition to simple library knowledge, it is possible to embed a variety of
290other language-specific information into the LLVM IR. If you have a specific
291need and run into a wall, please bring the topic up on the llvmdev list. At the
292very worst, you can always treat LLVM as if it were a "dumb code generator" and
293implement the high-level optimizations you desire in your front-end on the
294language-specific AST.
295</p>
296
297</div>
298
299<!-- *********************************************************************** -->
Chris Lattnerb8fc6502007-11-05 01:58:13 +0000[diff] [blame]300<div class="doc_section"><a name="tipsandtricks">Tips and Tricks</a></div>
301<!-- *********************************************************************** -->
302
303<div class="doc_text">
304
Chris Lattnera3f07ef2007-11-05 07:00:54 +0000[diff] [blame]305<p>There is a variety of useful tips and tricks that you come to know after
306working on/with LLVM that aren't obvious at first glance. Instead of letting
307everyone rediscover them, this section talks about some of these issues.</p>
308
309</div>
310
311<!-- ======================================================================= -->
312<div class="doc_subsubsection"><a name="offsetofsizeof">Implementing portable
313offsetof/sizeof</a></div>
314<!-- ======================================================================= -->
315
316<div class="doc_text">
317
318<p>One interesting thing that comes up if you are trying to keep the code
319generated by your compiler "target independent" is that you often need to know
320the size of some LLVM type or the offset of some field in an llvm structure.
321For example, you might need to pass the size of a type into a function that
322allocates memory.</p>
323
324<p>Unfortunately, this can vary widely across targets: for example the width of
325a pointer is trivially target-specific. However, there is a <a
326href="http://nondot.org/sabre/LLVMNotes/SizeOf-OffsetOf-VariableSizedStructs.txt">clever
327way to use the getelementptr instruction</a> that allows you to compute this
328in a portable way.</p>
329
330</div>
331
332<!-- ======================================================================= -->
333<div class="doc_subsubsection"><a name="gcstack">Garbage Collected
334Stack Frames</a></div>
335<!-- ======================================================================= -->
336
337<div class="doc_text">
338
339<p>Some languages want to explicitly manage their stack frames, often so that
340they are garbage collected or to allow easy implementation of closures. There
341are often better ways to implement these features than explicit stack frames,
342but <a
343href="http://nondot.org/sabre/LLVMNotes/ExplicitlyManagedStackFrames.txt">LLVM
344does support them if you want</a>. It requires your front-end to convert the
345code into <a
346href="http://en.wikipedia.org/wiki/Continuation-passing_style">Continuation
347Passing Style</a> and use of tail calls (which LLVM also supports).</p>
Chris Lattnerb8fc6502007-11-05 01:58:13 +0000[diff] [blame]348
349</div>
350
351<!-- *********************************************************************** -->
352<hr>
353<address>
354 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
355 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
356 <a href="http://validator.w3.org/check/referer"><img
357 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
358
359 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
360 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
361 Last modified: $Date: 2007-10-17 11:05:13 -0700 (Wed, 17 Oct 2007) $
362</address>
363</body>
364</html>