1================================
2Frequently Asked Questions (FAQ)
3================================
4
5.. contents::
6   :local:
7
8
9License
10=======
11
12Does the University of Illinois Open Source License really qualify as an "open source" license?
13-----------------------------------------------------------------------------------------------
14Yes, the license is `certified
15<http://www.opensource.org/licenses/UoI-NCSA.php>`_ by the Open Source
16Initiative (OSI).
17
18
19Can I modify LLVM source code and redistribute the modified source?
20-------------------------------------------------------------------
21Yes.  The modified source distribution must retain the copyright notice and
22follow the three bulleted conditions listed in the `LLVM license
23<http://llvm.org/svn/llvm-project/llvm/trunk/LICENSE.TXT>`_.
24
25
26Can I modify the LLVM source code and redistribute binaries or other tools based on it, without redistributing the source?
27--------------------------------------------------------------------------------------------------------------------------
28Yes. This is why we distribute LLVM under a less restrictive license than GPL,
29as explained in the first question above.
30
31
32Source Code
33===========
34
35In what language is LLVM written?
36---------------------------------
37All of the LLVM tools and libraries are written in C++ with extensive use of
38the STL.
39
40
41How portable is the LLVM source code?
42-------------------------------------
43The LLVM source code should be portable to most modern Unix-like operating
44systems.  Most of the code is written in standard C++ with operating system
45services abstracted to a support library.  The tools required to build and
46test LLVM have been ported to a plethora of platforms.
47
48Some porting problems may exist in the following areas:
49
50* The autoconf/makefile build system relies heavily on UNIX shell tools,
51  like the Bourne Shell and sed.  Porting to systems without these tools
52  (MacOS 9, Plan 9) will require more effort.
53
54What API do I use to store a value to one of the virtual registers in LLVM IR's SSA representation?
55---------------------------------------------------------------------------------------------------
56
57In short: you can't. It's actually kind of a silly question once you grok
58what's going on. Basically, in code like:
59
60.. code-block:: llvm
61
62    %result = add i32 %foo, %bar
63
64, ``%result`` is just a name given to the ``Value`` of the ``add``
65instruction. In other words, ``%result`` *is* the add instruction. The
66"assignment" doesn't explicitly "store" anything to any "virtual register";
67the "``=``" is more like the mathematical sense of equality.
68
69Longer explanation: In order to generate a textual representation of the
70IR, some kind of name has to be given to each instruction so that other
71instructions can textually reference it. However, the isomorphic in-memory
72representation that you manipulate from C++ has no such restriction since
73instructions can simply keep pointers to any other ``Value``'s that they
74reference. In fact, the names of dummy numbered temporaries like ``%1`` are
75not explicitly represented in the in-memory representation at all (see
76``Value::getName()``).
77
78
79Source Languages
80================
81
82What source languages are supported?
83------------------------------------
84
85LLVM currently has full support for C and C++ source languages through
86`Clang <http://clang.llvm.org/>`_. Many other language frontends have
87been written using LLVM, and an incomplete list is available at
88`projects with LLVM <http://llvm.org/ProjectsWithLLVM/>`_.
89
90
91I'd like to write a self-hosting LLVM compiler. How should I interface with the LLVM middle-end optimizers and back-end code generators?
92----------------------------------------------------------------------------------------------------------------------------------------
93Your compiler front-end will communicate with LLVM by creating a module in the
94LLVM intermediate representation (IR) format. Assuming you want to write your
95language's compiler in the language itself (rather than C++), there are 3
96major ways to tackle generating LLVM IR from a front-end:
97
981. **Call into the LLVM libraries code using your language's FFI (foreign
99   function interface).**
100
101  * *for:* best tracks changes to the LLVM IR, .ll syntax, and .bc format
102
103  * *for:* enables running LLVM optimization passes without a emit/parse
104    overhead
105
106  * *for:* adapts well to a JIT context
107
108  * *against:* lots of ugly glue code to write
109
1102. **Emit LLVM assembly from your compiler's native language.**
111
112  * *for:* very straightforward to get started
113
114  * *against:* the .ll parser is slower than the bitcode reader when
115    interfacing to the middle end
116
117  * *against:* it may be harder to track changes to the IR
118
1193. **Emit LLVM bitcode from your compiler's native language.**
120
121  * *for:* can use the more-efficient bitcode reader when interfacing to the
122    middle end
123
124  * *against:* you'll have to re-engineer the LLVM IR object model and bitcode
125    writer in your language
126
127  * *against:* it may be harder to track changes to the IR
128
129If you go with the first option, the C bindings in include/llvm-c should help
130a lot, since most languages have strong support for interfacing with C. The
131most common hurdle with calling C from managed code is interfacing with the
132garbage collector. The C interface was designed to require very little memory
133management, and so is straightforward in this regard.
134
135What support is there for a higher level source language constructs for building a compiler?
136--------------------------------------------------------------------------------------------
137Currently, there isn't much. LLVM supports an intermediate representation
138which is useful for code representation but will not support the high level
139(abstract syntax tree) representation needed by most compilers. There are no
140facilities for lexical nor semantic analysis.
141
142
143I don't understand the ``GetElementPtr`` instruction. Help!
144-----------------------------------------------------------
145See `The Often Misunderstood GEP Instruction <GetElementPtr.html>`_.
146
147
148Using the C and C++ Front Ends
149==============================
150
151Can I compile C or C++ code to platform-independent LLVM bitcode?
152-----------------------------------------------------------------
153No. C and C++ are inherently platform-dependent languages. The most obvious
154example of this is the preprocessor. A very common way that C code is made
155portable is by using the preprocessor to include platform-specific code. In
156practice, information about other platforms is lost after preprocessing, so
157the result is inherently dependent on the platform that the preprocessing was
158targeting.
159
160Another example is ``sizeof``. It's common for ``sizeof(long)`` to vary
161between platforms. In most C front-ends, ``sizeof`` is expanded to a
162constant immediately, thus hard-wiring a platform-specific detail.
163
164Also, since many platforms define their ABIs in terms of C, and since LLVM is
165lower-level than C, front-ends currently must emit platform-specific IR in
166order to have the result conform to the platform ABI.
167
168
169Questions about code generated by the demo page
170===============================================
171
172What is this ``llvm.global_ctors`` and ``_GLOBAL__I_a...`` stuff that happens when I ``#include <iostream>``?
173-------------------------------------------------------------------------------------------------------------
174If you ``#include`` the ``<iostream>`` header into a C++ translation unit,
175the file will probably use the ``std::cin``/``std::cout``/... global objects.
176However, C++ does not guarantee an order of initialization between static
177objects in different translation units, so if a static ctor/dtor in your .cpp
178file used ``std::cout``, for example, the object would not necessarily be
179automatically initialized before your use.
180
181To make ``std::cout`` and friends work correctly in these scenarios, the STL
182that we use declares a static object that gets created in every translation
183unit that includes ``<iostream>``.  This object has a static constructor
184and destructor that initializes and destroys the global iostream objects
185before they could possibly be used in the file.  The code that you see in the
186``.ll`` file corresponds to the constructor and destructor registration code.
187
188If you would like to make it easier to *understand* the LLVM code generated
189by the compiler in the demo page, consider using ``printf()`` instead of
190``iostream``\s to print values.
191
192
193Where did all of my code go??
194-----------------------------
195If you are using the LLVM demo page, you may often wonder what happened to
196all of the code that you typed in.  Remember that the demo script is running
197the code through the LLVM optimizers, so if your code doesn't actually do
198anything useful, it might all be deleted.
199
200To prevent this, make sure that the code is actually needed.  For example, if
201you are computing some expression, return the value from the function instead
202of leaving it in a local variable.  If you really want to constrain the
203optimizer, you can read from and assign to ``volatile`` global variables.
204
205
206What is this "``undef``" thing that shows up in my code?
207--------------------------------------------------------
208``undef`` is the LLVM way of representing a value that is not defined.  You
209can get these if you do not initialize a variable before you use it.  For
210example, the C function:
211
212.. code-block:: c
213
214   int X() { int i; return i; }
215
216Is compiled to "``ret i32 undef``" because "``i``" never has a value specified
217for it.
218
219
220Why does instcombine + simplifycfg turn a call to a function with a mismatched calling convention into "unreachable"? Why not make the verifier reject it?
221----------------------------------------------------------------------------------------------------------------------------------------------------------
222This is a common problem run into by authors of front-ends that are using
223custom calling conventions: you need to make sure to set the right calling
224convention on both the function and on each call to the function.  For
225example, this code:
226
227.. code-block:: llvm
228
229   define fastcc void @foo() {
230       ret void
231   }
232   define void @bar() {
233       call void @foo()
234       ret void
235   }
236
237Is optimized to:
238
239.. code-block:: llvm
240
241   define fastcc void @foo() {
242       ret void
243   }
244   define void @bar() {
245       unreachable
246   }
247
248... with "``opt -instcombine -simplifycfg``".  This often bites people because
249"all their code disappears".  Setting the calling convention on the caller and
250callee is required for indirect calls to work, so people often ask why not
251make the verifier reject this sort of thing.
252
253The answer is that this code has undefined behavior, but it is not illegal.
254If we made it illegal, then every transformation that could potentially create
255this would have to ensure that it doesn't, and there is valid code that can
256create this sort of construct (in dead code).  The sorts of things that can
257cause this to happen are fairly contrived, but we still need to accept them.
258Here's an example:
259
260.. code-block:: llvm
261
262   define fastcc void @foo() {
263       ret void
264   }
265   define internal void @bar(void()* %FP, i1 %cond) {
266       br i1 %cond, label %T, label %F
267   T:
268       call void %FP()
269       ret void
270   F:
271       call fastcc void %FP()
272       ret void
273   }
274   define void @test() {
275       %X = or i1 false, false
276       call void @bar(void()* @foo, i1 %X)
277       ret void
278   }
279
280In this example, "test" always passes ``@foo``/``false`` into ``bar``, which
281ensures that it is dynamically called with the right calling conv (thus, the
282code is perfectly well defined).  If you run this through the inliner, you
283get this (the explicit "or" is there so that the inliner doesn't dead code
284eliminate a bunch of stuff):
285
286.. code-block:: llvm
287
288   define fastcc void @foo() {
289       ret void
290   }
291   define void @test() {
292       %X = or i1 false, false
293       br i1 %X, label %T.i, label %F.i
294   T.i:
295       call void @foo()
296       br label %bar.exit
297   F.i:
298       call fastcc void @foo()
299       br label %bar.exit
300   bar.exit:
301       ret void
302   }
303
304Here you can see that the inlining pass made an undefined call to ``@foo``
305with the wrong calling convention.  We really don't want to make the inliner
306have to know about this sort of thing, so it needs to be valid code.  In this
307case, dead code elimination can trivially remove the undefined code.  However,
308if ``%X`` was an input argument to ``@test``, the inliner would produce this:
309
310.. code-block:: llvm
311
312   define fastcc void @foo() {
313       ret void
314   }
315
316   define void @test(i1 %X) {
317       br i1 %X, label %T.i, label %F.i
318   T.i:
319       call void @foo()
320       br label %bar.exit
321   F.i:
322       call fastcc void @foo()
323       br label %bar.exit
324   bar.exit:
325       ret void
326   }
327
328The interesting thing about this is that ``%X`` *must* be false for the
329code to be well-defined, but no amount of dead code elimination will be able
330to delete the broken call as unreachable.  However, since
331``instcombine``/``simplifycfg`` turns the undefined call into unreachable, we
332end up with a branch on a condition that goes to unreachable: a branch to
333unreachable can never happen, so "``-inline -instcombine -simplifycfg``" is
334able to produce:
335
336.. code-block:: llvm
337
338   define fastcc void @foo() {
339      ret void
340   }
341   define void @test(i1 %X) {
342   F.i:
343      call fastcc void @foo()
344      ret void
345   }
346