• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1======================================================
2Kaleidoscope: Conclusion and other useful LLVM tidbits
3======================================================
4
5.. contents::
6   :local:
7
8Tutorial Conclusion
9===================
10
11Welcome to the final chapter of the "`Implementing a language with
12LLVM <index.html>`_" tutorial. In the course of this tutorial, we have
13grown our little Kaleidoscope language from being a useless toy, to
14being a semi-interesting (but probably still useless) toy. :)
15
16It is interesting to see how far we've come, and how little code it has
17taken. We built the entire lexer, parser, AST, code generator, and an
18interactive run-loop (with a JIT!) by-hand in under 700 lines of
19(non-comment/non-blank) code.
20
21Our little language supports a couple of interesting features: it
22supports user defined binary and unary operators, it uses JIT
23compilation for immediate evaluation, and it supports a few control flow
24constructs with SSA construction.
25
26Part of the idea of this tutorial was to show you how easy and fun it
27can be to define, build, and play with languages. Building a compiler
28need not be a scary or mystical process! Now that you've seen some of
29the basics, I strongly encourage you to take the code and hack on it.
30For example, try adding:
31
32-  **global variables** - While global variables have questional value
33   in modern software engineering, they are often useful when putting
34   together quick little hacks like the Kaleidoscope compiler itself.
35   Fortunately, our current setup makes it very easy to add global
36   variables: just have value lookup check to see if an unresolved
37   variable is in the global variable symbol table before rejecting it.
38   To create a new global variable, make an instance of the LLVM
39   ``GlobalVariable`` class.
40-  **typed variables** - Kaleidoscope currently only supports variables
41   of type double. This gives the language a very nice elegance, because
42   only supporting one type means that you never have to specify types.
43   Different languages have different ways of handling this. The easiest
44   way is to require the user to specify types for every variable
45   definition, and record the type of the variable in the symbol table
46   along with its Value\*.
47-  **arrays, structs, vectors, etc** - Once you add types, you can start
48   extending the type system in all sorts of interesting ways. Simple
49   arrays are very easy and are quite useful for many different
50   applications. Adding them is mostly an exercise in learning how the
51   LLVM `getelementptr <../LangRef.html#i_getelementptr>`_ instruction
52   works: it is so nifty/unconventional, it `has its own
53   FAQ <../GetElementPtr.html>`_! If you add support for recursive types
54   (e.g. linked lists), make sure to read the `section in the LLVM
55   Programmer's Manual <../ProgrammersManual.html#TypeResolve>`_ that
56   describes how to construct them.
57-  **standard runtime** - Our current language allows the user to access
58   arbitrary external functions, and we use it for things like "printd"
59   and "putchard". As you extend the language to add higher-level
60   constructs, often these constructs make the most sense if they are
61   lowered to calls into a language-supplied runtime. For example, if
62   you add hash tables to the language, it would probably make sense to
63   add the routines to a runtime, instead of inlining them all the way.
64-  **memory management** - Currently we can only access the stack in
65   Kaleidoscope. It would also be useful to be able to allocate heap
66   memory, either with calls to the standard libc malloc/free interface
67   or with a garbage collector. If you would like to use garbage
68   collection, note that LLVM fully supports `Accurate Garbage
69   Collection <../GarbageCollection.html>`_ including algorithms that
70   move objects and need to scan/update the stack.
71-  **debugger support** - LLVM supports generation of `DWARF Debug
72   info <../SourceLevelDebugging.html>`_ which is understood by common
73   debuggers like GDB. Adding support for debug info is fairly
74   straightforward. The best way to understand it is to compile some
75   C/C++ code with "``clang -g -O0``" and taking a look at what it
76   produces.
77-  **exception handling support** - LLVM supports generation of `zero
78   cost exceptions <../ExceptionHandling.html>`_ which interoperate with
79   code compiled in other languages. You could also generate code by
80   implicitly making every function return an error value and checking
81   it. You could also make explicit use of setjmp/longjmp. There are
82   many different ways to go here.
83-  **object orientation, generics, database access, complex numbers,
84   geometric programming, ...** - Really, there is no end of crazy
85   features that you can add to the language.
86-  **unusual domains** - We've been talking about applying LLVM to a
87   domain that many people are interested in: building a compiler for a
88   specific language. However, there are many other domains that can use
89   compiler technology that are not typically considered. For example,
90   LLVM has been used to implement OpenGL graphics acceleration,
91   translate C++ code to ActionScript, and many other cute and clever
92   things. Maybe you will be the first to JIT compile a regular
93   expression interpreter into native code with LLVM?
94
95Have fun - try doing something crazy and unusual. Building a language
96like everyone else always has, is much less fun than trying something a
97little crazy or off the wall and seeing how it turns out. If you get
98stuck or want to talk about it, feel free to email the `llvmdev mailing
99list <http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev>`_: it has lots
100of people who are interested in languages and are often willing to help
101out.
102
103Before we end this tutorial, I want to talk about some "tips and tricks"
104for generating LLVM IR. These are some of the more subtle things that
105may not be obvious, but are very useful if you want to take advantage of
106LLVM's capabilities.
107
108Properties of the LLVM IR
109=========================
110
111We have a couple common questions about code in the LLVM IR form - lets
112just get these out of the way right now, shall we?
113
114Target Independence
115-------------------
116
117Kaleidoscope is an example of a "portable language": any program written
118in Kaleidoscope will work the same way on any target that it runs on.
119Many other languages have this property, e.g. lisp, java, haskell,
120javascript, python, etc (note that while these languages are portable,
121not all their libraries are).
122
123One nice aspect of LLVM is that it is often capable of preserving target
124independence in the IR: you can take the LLVM IR for a
125Kaleidoscope-compiled program and run it on any target that LLVM
126supports, even emitting C code and compiling that on targets that LLVM
127doesn't support natively. You can trivially tell that the Kaleidoscope
128compiler generates target-independent code because it never queries for
129any target-specific information when generating code.
130
131The fact that LLVM provides a compact, target-independent,
132representation for code gets a lot of people excited. Unfortunately,
133these people are usually thinking about C or a language from the C
134family when they are asking questions about language portability. I say
135"unfortunately", because there is really no way to make (fully general)
136C code portable, other than shipping the source code around (and of
137course, C source code is not actually portable in general either - ever
138port a really old application from 32- to 64-bits?).
139
140The problem with C (again, in its full generality) is that it is heavily
141laden with target specific assumptions. As one simple example, the
142preprocessor often destructively removes target-independence from the
143code when it processes the input text:
144
145.. code-block:: c
146
147    #ifdef __i386__
148      int X = 1;
149    #else
150      int X = 42;
151    #endif
152
153While it is possible to engineer more and more complex solutions to
154problems like this, it cannot be solved in full generality in a way that
155is better than shipping the actual source code.
156
157That said, there are interesting subsets of C that can be made portable.
158If you are willing to fix primitive types to a fixed size (say int =
15932-bits, and long = 64-bits), don't care about ABI compatibility with
160existing binaries, and are willing to give up some other minor features,
161you can have portable code. This can make sense for specialized domains
162such as an in-kernel language.
163
164Safety Guarantees
165-----------------
166
167Many of the languages above are also "safe" languages: it is impossible
168for a program written in Java to corrupt its address space and crash the
169process (assuming the JVM has no bugs). Safety is an interesting
170property that requires a combination of language design, runtime
171support, and often operating system support.
172
173It is certainly possible to implement a safe language in LLVM, but LLVM
174IR does not itself guarantee safety. The LLVM IR allows unsafe pointer
175casts, use after free bugs, buffer over-runs, and a variety of other
176problems. Safety needs to be implemented as a layer on top of LLVM and,
177conveniently, several groups have investigated this. Ask on the `llvmdev
178mailing list <http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev>`_ if
179you are interested in more details.
180
181Language-Specific Optimizations
182-------------------------------
183
184One thing about LLVM that turns off many people is that it does not
185solve all the world's problems in one system (sorry 'world hunger',
186someone else will have to solve you some other day). One specific
187complaint is that people perceive LLVM as being incapable of performing
188high-level language-specific optimization: LLVM "loses too much
189information".
190
191Unfortunately, this is really not the place to give you a full and
192unified version of "Chris Lattner's theory of compiler design". Instead,
193I'll make a few observations:
194
195First, you're right that LLVM does lose information. For example, as of
196this writing, there is no way to distinguish in the LLVM IR whether an
197SSA-value came from a C "int" or a C "long" on an ILP32 machine (other
198than debug info). Both get compiled down to an 'i32' value and the
199information about what it came from is lost. The more general issue
200here, is that the LLVM type system uses "structural equivalence" instead
201of "name equivalence". Another place this surprises people is if you
202have two types in a high-level language that have the same structure
203(e.g. two different structs that have a single int field): these types
204will compile down into a single LLVM type and it will be impossible to
205tell what it came from.
206
207Second, while LLVM does lose information, LLVM is not a fixed target: we
208continue to enhance and improve it in many different ways. In addition
209to adding new features (LLVM did not always support exceptions or debug
210info), we also extend the IR to capture important information for
211optimization (e.g. whether an argument is sign or zero extended,
212information about pointers aliasing, etc). Many of the enhancements are
213user-driven: people want LLVM to include some specific feature, so they
214go ahead and extend it.
215
216Third, it is *possible and easy* to add language-specific optimizations,
217and you have a number of choices in how to do it. As one trivial
218example, it is easy to add language-specific optimization passes that
219"know" things about code compiled for a language. In the case of the C
220family, there is an optimization pass that "knows" about the standard C
221library functions. If you call "exit(0)" in main(), it knows that it is
222safe to optimize that into "return 0;" because C specifies what the
223'exit' function does.
224
225In addition to simple library knowledge, it is possible to embed a
226variety of other language-specific information into the LLVM IR. If you
227have a specific need and run into a wall, please bring the topic up on
228the llvmdev list. At the very worst, you can always treat LLVM as if it
229were a "dumb code generator" and implement the high-level optimizations
230you desire in your front-end, on the language-specific AST.
231
232Tips and Tricks
233===============
234
235There is a variety of useful tips and tricks that you come to know after
236working on/with LLVM that aren't obvious at first glance. Instead of
237letting everyone rediscover them, this section talks about some of these
238issues.
239
240Implementing portable offsetof/sizeof
241-------------------------------------
242
243One interesting thing that comes up, if you are trying to keep the code
244generated by your compiler "target independent", is that you often need
245to know the size of some LLVM type or the offset of some field in an
246llvm structure. For example, you might need to pass the size of a type
247into a function that allocates memory.
248
249Unfortunately, this can vary widely across targets: for example the
250width of a pointer is trivially target-specific. However, there is a
251`clever way to use the getelementptr
252instruction <http://nondot.org/sabre/LLVMNotes/SizeOf-OffsetOf-VariableSizedStructs.txt>`_
253that allows you to compute this in a portable way.
254
255Garbage Collected Stack Frames
256------------------------------
257
258Some languages want to explicitly manage their stack frames, often so
259that they are garbage collected or to allow easy implementation of
260closures. There are often better ways to implement these features than
261explicit stack frames, but `LLVM does support
262them, <http://nondot.org/sabre/LLVMNotes/ExplicitlyManagedStackFrames.txt>`_
263if you want. It requires your front-end to convert the code into
264`Continuation Passing
265Style <http://en.wikipedia.org/wiki/Continuation-passing_style>`_ and
266the use of tail calls (which LLVM also supports).
267
268