1 ======================================================
2 Kaleidoscope: Conclusion and other useful LLVM tidbits
3 ======================================================
8 Written by `Chris Lattner <mailto:sabre@nondot.org>`_
13 Welcome to the final chapter of the "`Implementing a language with
14 LLVM <index.html>`_" tutorial. In the course of this tutorial, we have
15 grown our little Kaleidoscope language from being a useless toy, to
16 being a semi-interesting (but probably still useless) toy. :)
18 It is interesting to see how far we've come, and how little code it has
19 taken. We built the entire lexer, parser, AST, code generator, and an
20 interactive run-loop (with a JIT!) by-hand in under 700 lines of
21 (non-comment/non-blank) code.
23 Our little language supports a couple of interesting features: it
24 supports user defined binary and unary operators, it uses JIT
25 compilation for immediate evaluation, and it supports a few control flow
26 constructs with SSA construction.
28 Part of the idea of this tutorial was to show you how easy and fun it
29 can be to define, build, and play with languages. Building a compiler
30 need not be a scary or mystical process! Now that you've seen some of
31 the basics, I strongly encourage you to take the code and hack on it.
32 For example, try adding:
34 - **global variables** - While global variables have questional value
35 in modern software engineering, they are often useful when putting
36 together quick little hacks like the Kaleidoscope compiler itself.
37 Fortunately, our current setup makes it very easy to add global
38 variables: just have value lookup check to see if an unresolved
39 variable is in the global variable symbol table before rejecting it.
40 To create a new global variable, make an instance of the LLVM
41 ``GlobalVariable`` class.
42 - **typed variables** - Kaleidoscope currently only supports variables
43 of type double. This gives the language a very nice elegance, because
44 only supporting one type means that you never have to specify types.
45 Different languages have different ways of handling this. The easiest
46 way is to require the user to specify types for every variable
47 definition, and record the type of the variable in the symbol table
48 along with its Value\*.
49 - **arrays, structs, vectors, etc** - Once you add types, you can start
50 extending the type system in all sorts of interesting ways. Simple
51 arrays are very easy and are quite useful for many different
52 applications. Adding them is mostly an exercise in learning how the
53 LLVM `getelementptr <../LangRef.html#i_getelementptr>`_ instruction
54 works: it is so nifty/unconventional, it `has its own
55 FAQ <../GetElementPtr.html>`_! If you add support for recursive types
56 (e.g. linked lists), make sure to read the `section in the LLVM
57 Programmer's Manual <../ProgrammersManual.html#TypeResolve>`_ that
58 describes how to construct them.
59 - **standard runtime** - Our current language allows the user to access
60 arbitrary external functions, and we use it for things like "printd"
61 and "putchard". As you extend the language to add higher-level
62 constructs, often these constructs make the most sense if they are
63 lowered to calls into a language-supplied runtime. For example, if
64 you add hash tables to the language, it would probably make sense to
65 add the routines to a runtime, instead of inlining them all the way.
66 - **memory management** - Currently we can only access the stack in
67 Kaleidoscope. It would also be useful to be able to allocate heap
68 memory, either with calls to the standard libc malloc/free interface
69 or with a garbage collector. If you would like to use garbage
70 collection, note that LLVM fully supports `Accurate Garbage
71 Collection <../GarbageCollection.html>`_ including algorithms that
72 move objects and need to scan/update the stack.
73 - **debugger support** - LLVM supports generation of `DWARF Debug
74 info <../SourceLevelDebugging.html>`_ which is understood by common
75 debuggers like GDB. Adding support for debug info is fairly
76 straightforward. The best way to understand it is to compile some
77 C/C++ code with "``llvm-gcc -g -O0``" and taking a look at what it
79 - **exception handling support** - LLVM supports generation of `zero
80 cost exceptions <../ExceptionHandling.html>`_ which interoperate with
81 code compiled in other languages. You could also generate code by
82 implicitly making every function return an error value and checking
83 it. You could also make explicit use of setjmp/longjmp. There are
84 many different ways to go here.
85 - **object orientation, generics, database access, complex numbers,
86 geometric programming, ...** - Really, there is no end of crazy
87 features that you can add to the language.
88 - **unusual domains** - We've been talking about applying LLVM to a
89 domain that many people are interested in: building a compiler for a
90 specific language. However, there are many other domains that can use
91 compiler technology that are not typically considered. For example,
92 LLVM has been used to implement OpenGL graphics acceleration,
93 translate C++ code to ActionScript, and many other cute and clever
94 things. Maybe you will be the first to JIT compile a regular
95 expression interpreter into native code with LLVM?
97 Have fun - try doing something crazy and unusual. Building a language
98 like everyone else always has, is much less fun than trying something a
99 little crazy or off the wall and seeing how it turns out. If you get
100 stuck or want to talk about it, feel free to email the `llvmdev mailing
101 list <http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev>`_: it has lots
102 of people who are interested in languages and are often willing to help
105 Before we end this tutorial, I want to talk about some "tips and tricks"
106 for generating LLVM IR. These are some of the more subtle things that
107 may not be obvious, but are very useful if you want to take advantage of
110 Properties of the LLVM IR
111 =========================
113 We have a couple common questions about code in the LLVM IR form - lets
114 just get these out of the way right now, shall we?
119 Kaleidoscope is an example of a "portable language": any program written
120 in Kaleidoscope will work the same way on any target that it runs on.
121 Many other languages have this property, e.g. lisp, java, haskell,
122 javascript, python, etc (note that while these languages are portable,
123 not all their libraries are).
125 One nice aspect of LLVM is that it is often capable of preserving target
126 independence in the IR: you can take the LLVM IR for a
127 Kaleidoscope-compiled program and run it on any target that LLVM
128 supports, even emitting C code and compiling that on targets that LLVM
129 doesn't support natively. You can trivially tell that the Kaleidoscope
130 compiler generates target-independent code because it never queries for
131 any target-specific information when generating code.
133 The fact that LLVM provides a compact, target-independent,
134 representation for code gets a lot of people excited. Unfortunately,
135 these people are usually thinking about C or a language from the C
136 family when they are asking questions about language portability. I say
137 "unfortunately", because there is really no way to make (fully general)
138 C code portable, other than shipping the source code around (and of
139 course, C source code is not actually portable in general either - ever
140 port a really old application from 32- to 64-bits?).
142 The problem with C (again, in its full generality) is that it is heavily
143 laden with target specific assumptions. As one simple example, the
144 preprocessor often destructively removes target-independence from the
145 code when it processes the input text:
155 While it is possible to engineer more and more complex solutions to
156 problems like this, it cannot be solved in full generality in a way that
157 is better than shipping the actual source code.
159 That said, there are interesting subsets of C that can be made portable.
160 If you are willing to fix primitive types to a fixed size (say int =
161 32-bits, and long = 64-bits), don't care about ABI compatibility with
162 existing binaries, and are willing to give up some other minor features,
163 you can have portable code. This can make sense for specialized domains
164 such as an in-kernel language.
169 Many of the languages above are also "safe" languages: it is impossible
170 for a program written in Java to corrupt its address space and crash the
171 process (assuming the JVM has no bugs). Safety is an interesting
172 property that requires a combination of language design, runtime
173 support, and often operating system support.
175 It is certainly possible to implement a safe language in LLVM, but LLVM
176 IR does not itself guarantee safety. The LLVM IR allows unsafe pointer
177 casts, use after free bugs, buffer over-runs, and a variety of other
178 problems. Safety needs to be implemented as a layer on top of LLVM and,
179 conveniently, several groups have investigated this. Ask on the `llvmdev
180 mailing list <http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev>`_ if
181 you are interested in more details.
183 Language-Specific Optimizations
184 -------------------------------
186 One thing about LLVM that turns off many people is that it does not
187 solve all the world's problems in one system (sorry 'world hunger',
188 someone else will have to solve you some other day). One specific
189 complaint is that people perceive LLVM as being incapable of performing
190 high-level language-specific optimization: LLVM "loses too much
193 Unfortunately, this is really not the place to give you a full and
194 unified version of "Chris Lattner's theory of compiler design". Instead,
195 I'll make a few observations:
197 First, you're right that LLVM does lose information. For example, as of
198 this writing, there is no way to distinguish in the LLVM IR whether an
199 SSA-value came from a C "int" or a C "long" on an ILP32 machine (other
200 than debug info). Both get compiled down to an 'i32' value and the
201 information about what it came from is lost. The more general issue
202 here, is that the LLVM type system uses "structural equivalence" instead
203 of "name equivalence". Another place this surprises people is if you
204 have two types in a high-level language that have the same structure
205 (e.g. two different structs that have a single int field): these types
206 will compile down into a single LLVM type and it will be impossible to
207 tell what it came from.
209 Second, while LLVM does lose information, LLVM is not a fixed target: we
210 continue to enhance and improve it in many different ways. In addition
211 to adding new features (LLVM did not always support exceptions or debug
212 info), we also extend the IR to capture important information for
213 optimization (e.g. whether an argument is sign or zero extended,
214 information about pointers aliasing, etc). Many of the enhancements are
215 user-driven: people want LLVM to include some specific feature, so they
216 go ahead and extend it.
218 Third, it is *possible and easy* to add language-specific optimizations,
219 and you have a number of choices in how to do it. As one trivial
220 example, it is easy to add language-specific optimization passes that
221 "know" things about code compiled for a language. In the case of the C
222 family, there is an optimization pass that "knows" about the standard C
223 library functions. If you call "exit(0)" in main(), it knows that it is
224 safe to optimize that into "return 0;" because C specifies what the
225 'exit' function does.
227 In addition to simple library knowledge, it is possible to embed a
228 variety of other language-specific information into the LLVM IR. If you
229 have a specific need and run into a wall, please bring the topic up on
230 the llvmdev list. At the very worst, you can always treat LLVM as if it
231 were a "dumb code generator" and implement the high-level optimizations
232 you desire in your front-end, on the language-specific AST.
237 There is a variety of useful tips and tricks that you come to know after
238 working on/with LLVM that aren't obvious at first glance. Instead of
239 letting everyone rediscover them, this section talks about some of these
242 Implementing portable offsetof/sizeof
243 -------------------------------------
245 One interesting thing that comes up, if you are trying to keep the code
246 generated by your compiler "target independent", is that you often need
247 to know the size of some LLVM type or the offset of some field in an
248 llvm structure. For example, you might need to pass the size of a type
249 into a function that allocates memory.
251 Unfortunately, this can vary widely across targets: for example the
252 width of a pointer is trivially target-specific. However, there is a
253 `clever way to use the getelementptr
254 instruction <http://nondot.org/sabre/LLVMNotes/SizeOf-OffsetOf-VariableSizedStructs.txt>`_
255 that allows you to compute this in a portable way.
257 Garbage Collected Stack Frames
258 ------------------------------
260 Some languages want to explicitly manage their stack frames, often so
261 that they are garbage collected or to allow easy implementation of
262 closures. There are often better ways to implement these features than
263 explicit stack frames, but `LLVM does support
264 them, <http://nondot.org/sabre/LLVMNotes/ExplicitlyManagedStackFrames.txt>`_
265 if you want. It requires your front-end to convert the code into
266 `Continuation Passing
267 Style <http://en.wikipedia.org/wiki/Continuation-passing_style>`_ and
268 the use of tail calls (which LLVM also supports).