3 ================================
4 Frequently Asked Questions (FAQ)
5 ================================
14 Does the University of Illinois Open Source License really qualify as an "open source" license?
15 -----------------------------------------------------------------------------------------------
16 Yes, the license is `certified
17 <http://www.opensource.org/licenses/UoI-NCSA.php>`_ by the Open Source
21 Can I modify LLVM source code and redistribute the modified source?
22 -------------------------------------------------------------------
23 Yes. The modified source distribution must retain the copyright notice and
24 follow the three bulletted conditions listed in the `LLVM license
25 <http://llvm.org/svn/llvm-project/llvm/trunk/LICENSE.TXT>`_.
28 Can I modify the LLVM source code and redistribute binaries or other tools based on it, without redistributing the source?
29 --------------------------------------------------------------------------------------------------------------------------
30 Yes. This is why we distribute LLVM under a less restrictive license than GPL,
31 as explained in the first question above.
37 In what language is LLVM written?
38 ---------------------------------
39 All of the LLVM tools and libraries are written in C++ with extensive use of
43 How portable is the LLVM source code?
44 -------------------------------------
45 The LLVM source code should be portable to most modern Unix-like operating
46 systems. Most of the code is written in standard C++ with operating system
47 services abstracted to a support library. The tools required to build and
48 test LLVM have been ported to a plethora of platforms.
50 Some porting problems may exist in the following areas:
52 * The autoconf/makefile build system relies heavily on UNIX shell tools,
53 like the Bourne Shell and sed. Porting to systems without these tools
54 (MacOS 9, Plan 9) will require more effort.
56 What API do I use to store a value to one of the virtual registers in LLVM IR's SSA representation?
57 ---------------------------------------------------------------------------------------------------
59 In short: you can't. It's actually kind of a silly question once you grok
60 what's going on. Basically, in code like:
64 %result = add i32 %foo, %bar
66 , ``%result`` is just a name given to the ``Value`` of the ``add``
67 instruction. In other words, ``%result`` *is* the add instruction. The
68 "assignment" doesn't explicitly "store" anything to any "virtual register";
69 the "``=``" is more like the mathematical sense of equality.
71 Longer explanation: In order to generate a textual representation of the
72 IR, some kind of name has to be given to each instruction so that other
73 instructions can textually reference it. However, the isomorphic in-memory
74 representation that you manipulate from C++ has no such restriction since
75 instructions can simply keep pointers to any other ``Value``'s that they
76 reference. In fact, the names of dummy numbered temporaries like ``%1`` are
77 not explicitly represented in the in-memory representation at all (see
78 ``Value::getName()``).
83 When I run configure, it finds the wrong C compiler.
84 ----------------------------------------------------
85 The ``configure`` script attempts to locate first ``gcc`` and then ``cc``,
86 unless it finds compiler paths set in ``CC`` and ``CXX`` for the C and C++
87 compiler, respectively.
89 If ``configure`` finds the wrong compiler, either adjust your ``PATH``
90 environment variable or set ``CC`` and ``CXX`` explicitly.
93 The ``configure`` script finds the right C compiler, but it uses the LLVM tools from a previous build. What do I do?
94 ---------------------------------------------------------------------------------------------------------------------
95 The ``configure`` script uses the ``PATH`` to find executables, so if it's
96 grabbing the wrong linker/assembler/etc, there are two ways to fix it:
98 #. Adjust your ``PATH`` environment variable so that the correct program
99 appears first in the ``PATH``. This may work, but may not be convenient
100 when you want them *first* in your path for other work.
102 #. Run ``configure`` with an alternative ``PATH`` that is correct. In a
103 Bourne compatible shell, the syntax would be:
105 .. code-block:: console
107 % PATH=[the path without the bad program] ./configure ...
109 This is still somewhat inconvenient, but it allows ``configure`` to do its
110 work without having to adjust your ``PATH`` permanently.
113 When creating a dynamic library, I get a strange GLIBC error.
114 -------------------------------------------------------------
115 Under some operating systems (i.e. Linux), libtool does not work correctly if
116 GCC was compiled with the ``--disable-shared option``. To work around this,
117 install your own version of GCC that has shared libraries enabled by default.
120 I've updated my source tree from Subversion, and now my build is trying to use a file/directory that doesn't exist.
121 -------------------------------------------------------------------------------------------------------------------
122 You need to re-run configure in your object directory. When new Makefiles
123 are added to the source tree, they have to be copied over to the object tree
124 in order to be used by the build.
127 I've modified a Makefile in my source tree, but my build tree keeps using the old version. What do I do?
128 ---------------------------------------------------------------------------------------------------------
129 If the Makefile already exists in your object tree, you can just run the
130 following command in the top level directory of your object tree:
132 .. code-block:: console
134 % ./config.status <relative path to Makefile>;
136 If the Makefile is new, you will have to modify the configure script to copy
140 I've upgraded to a new version of LLVM, and I get strange build errors.
141 -----------------------------------------------------------------------
142 Sometimes, changes to the LLVM source code alters how the build system works.
143 Changes in ``libtool``, ``autoconf``, or header file dependencies are
144 especially prone to this sort of problem.
146 The best thing to try is to remove the old files and re-build. In most cases,
147 this takes care of the problem. To do this, just type ``make clean`` and then
148 ``make`` in the directory that fails to build.
151 I've built LLVM and am testing it, but the tests freeze.
152 --------------------------------------------------------
153 This is most likely occurring because you built a profile or release
154 (optimized) build of LLVM and have not specified the same information on the
155 ``gmake`` command line.
157 For example, if you built LLVM with the command:
159 .. code-block:: console
161 % gmake ENABLE_PROFILING=1
163 ...then you must run the tests with the following commands:
165 .. code-block:: console
168 % gmake ENABLE_PROFILING=1
170 Why do test results differ when I perform different types of builds?
171 --------------------------------------------------------------------
172 The LLVM test suite is dependent upon several features of the LLVM tools and
175 First, the debugging assertions in code are not enabled in optimized or
176 profiling builds. Hence, tests that used to fail may pass.
178 Second, some tests may rely upon debugging options or behavior that is only
179 available in the debug build. These tests will fail in an optimized or
183 Compiling LLVM with GCC 3.3.2 fails, what should I do?
184 ------------------------------------------------------
185 This is `a bug in GCC <http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13392>`_,
186 and affects projects other than LLVM. Try upgrading or downgrading your GCC.
189 Compiling LLVM with GCC succeeds, but the resulting tools do not work, what can be wrong?
190 -----------------------------------------------------------------------------------------
191 Several versions of GCC have shown a weakness in miscompiling the LLVM
192 codebase. Please consult your compiler version (``gcc --version``) to find
193 out whether it is `broken <GettingStarted.html#brokengcc>`_. If so, your only
194 option is to upgrade GCC to a known good version.
197 After Subversion update, rebuilding gives the error "No rule to make target".
198 -----------------------------------------------------------------------------
199 If the error is of the form:
201 .. code-block:: console
203 gmake[2]: *** No rule to make target `/path/to/somefile',
204 needed by `/path/to/another/file.d'.
207 This may occur anytime files are moved within the Subversion repository or
208 removed entirely. In this case, the best solution is to erase all ``.d``
209 files, which list dependencies for source files, and rebuild:
211 .. code-block:: console
214 % rm -f `find . -name \*\.d`
217 In other cases, it may be necessary to run ``make clean`` before rebuilding.
223 What source languages are supported?
224 ------------------------------------
225 LLVM currently has full support for C and C++ source languages. These are
226 available through both `Clang <http://clang.llvm.org/>`_ and `DragonEgg
227 <http://dragonegg.llvm.org/>`_.
229 The PyPy developers are working on integrating LLVM into the PyPy backend so
230 that PyPy language can translate to LLVM.
233 I'd like to write a self-hosting LLVM compiler. How should I interface with the LLVM middle-end optimizers and back-end code generators?
234 ----------------------------------------------------------------------------------------------------------------------------------------
235 Your compiler front-end will communicate with LLVM by creating a module in the
236 LLVM intermediate representation (IR) format. Assuming you want to write your
237 language's compiler in the language itself (rather than C++), there are 3
238 major ways to tackle generating LLVM IR from a front-end:
240 1. **Call into the LLVM libraries code using your language's FFI (foreign
241 function interface).**
243 * *for:* best tracks changes to the LLVM IR, .ll syntax, and .bc format
245 * *for:* enables running LLVM optimization passes without a emit/parse
248 * *for:* adapts well to a JIT context
250 * *against:* lots of ugly glue code to write
252 2. **Emit LLVM assembly from your compiler's native language.**
254 * *for:* very straightforward to get started
256 * *against:* the .ll parser is slower than the bitcode reader when
257 interfacing to the middle end
259 * *against:* it may be harder to track changes to the IR
261 3. **Emit LLVM bitcode from your compiler's native language.**
263 * *for:* can use the more-efficient bitcode reader when interfacing to the
266 * *against:* you'll have to re-engineer the LLVM IR object model and bitcode
267 writer in your language
269 * *against:* it may be harder to track changes to the IR
271 If you go with the first option, the C bindings in include/llvm-c should help
272 a lot, since most languages have strong support for interfacing with C. The
273 most common hurdle with calling C from managed code is interfacing with the
274 garbage collector. The C interface was designed to require very little memory
275 management, and so is straightforward in this regard.
277 What support is there for a higher level source language constructs for building a compiler?
278 --------------------------------------------------------------------------------------------
279 Currently, there isn't much. LLVM supports an intermediate representation
280 which is useful for code representation but will not support the high level
281 (abstract syntax tree) representation needed by most compilers. There are no
282 facilities for lexical nor semantic analysis.
285 I don't understand the ``GetElementPtr`` instruction. Help!
286 -----------------------------------------------------------
287 See `The Often Misunderstood GEP Instruction <GetElementPtr.html>`_.
290 Using the C and C++ Front Ends
291 ==============================
293 Can I compile C or C++ code to platform-independent LLVM bitcode?
294 -----------------------------------------------------------------
295 No. C and C++ are inherently platform-dependent languages. The most obvious
296 example of this is the preprocessor. A very common way that C code is made
297 portable is by using the preprocessor to include platform-specific code. In
298 practice, information about other platforms is lost after preprocessing, so
299 the result is inherently dependent on the platform that the preprocessing was
302 Another example is ``sizeof``. It's common for ``sizeof(long)`` to vary
303 between platforms. In most C front-ends, ``sizeof`` is expanded to a
304 constant immediately, thus hard-wiring a platform-specific detail.
306 Also, since many platforms define their ABIs in terms of C, and since LLVM is
307 lower-level than C, front-ends currently must emit platform-specific IR in
308 order to have the result conform to the platform ABI.
311 Questions about code generated by the demo page
312 ===============================================
314 What is this ``llvm.global_ctors`` and ``_GLOBAL__I_a...`` stuff that happens when I ``#include <iostream>``?
315 -------------------------------------------------------------------------------------------------------------
316 If you ``#include`` the ``<iostream>`` header into a C++ translation unit,
317 the file will probably use the ``std::cin``/``std::cout``/... global objects.
318 However, C++ does not guarantee an order of initialization between static
319 objects in different translation units, so if a static ctor/dtor in your .cpp
320 file used ``std::cout``, for example, the object would not necessarily be
321 automatically initialized before your use.
323 To make ``std::cout`` and friends work correctly in these scenarios, the STL
324 that we use declares a static object that gets created in every translation
325 unit that includes ``<iostream>``. This object has a static constructor
326 and destructor that initializes and destroys the global iostream objects
327 before they could possibly be used in the file. The code that you see in the
328 ``.ll`` file corresponds to the constructor and destructor registration code.
330 If you would like to make it easier to *understand* the LLVM code generated
331 by the compiler in the demo page, consider using ``printf()`` instead of
332 ``iostream``\s to print values.
335 Where did all of my code go??
336 -----------------------------
337 If you are using the LLVM demo page, you may often wonder what happened to
338 all of the code that you typed in. Remember that the demo script is running
339 the code through the LLVM optimizers, so if your code doesn't actually do
340 anything useful, it might all be deleted.
342 To prevent this, make sure that the code is actually needed. For example, if
343 you are computing some expression, return the value from the function instead
344 of leaving it in a local variable. If you really want to constrain the
345 optimizer, you can read from and assign to ``volatile`` global variables.
348 What is this "``undef``" thing that shows up in my code?
349 --------------------------------------------------------
350 ``undef`` is the LLVM way of representing a value that is not defined. You
351 can get these if you do not initialize a variable before you use it. For
352 example, the C function:
356 int X() { int i; return i; }
358 Is compiled to "``ret i32 undef``" because "``i``" never has a value specified
362 Why does instcombine + simplifycfg turn a call to a function with a mismatched calling convention into "unreachable"? Why not make the verifier reject it?
363 ----------------------------------------------------------------------------------------------------------------------------------------------------------
364 This is a common problem run into by authors of front-ends that are using
365 custom calling conventions: you need to make sure to set the right calling
366 convention on both the function and on each call to the function. For
371 define fastcc void @foo() {
383 define fastcc void @foo() {
390 ... with "``opt -instcombine -simplifycfg``". This often bites people because
391 "all their code disappears". Setting the calling convention on the caller and
392 callee is required for indirect calls to work, so people often ask why not
393 make the verifier reject this sort of thing.
395 The answer is that this code has undefined behavior, but it is not illegal.
396 If we made it illegal, then every transformation that could potentially create
397 this would have to ensure that it doesn't, and there is valid code that can
398 create this sort of construct (in dead code). The sorts of things that can
399 cause this to happen are fairly contrived, but we still need to accept them.
404 define fastcc void @foo() {
407 define internal void @bar(void()* %FP, i1 %cond) {
408 br i1 %cond, label %T, label %F
413 call fastcc void %FP()
416 define void @test() {
417 %X = or i1 false, false
418 call void @bar(void()* @foo, i1 %X)
422 In this example, "test" always passes ``@foo``/``false`` into ``bar``, which
423 ensures that it is dynamically called with the right calling conv (thus, the
424 code is perfectly well defined). If you run this through the inliner, you
425 get this (the explicit "or" is there so that the inliner doesn't dead code
426 eliminate a bunch of stuff):
430 define fastcc void @foo() {
433 define void @test() {
434 %X = or i1 false, false
435 br i1 %X, label %T.i, label %F.i
440 call fastcc void @foo()
446 Here you can see that the inlining pass made an undefined call to ``@foo``
447 with the wrong calling convention. We really don't want to make the inliner
448 have to know about this sort of thing, so it needs to be valid code. In this
449 case, dead code elimination can trivially remove the undefined code. However,
450 if ``%X`` was an input argument to ``@test``, the inliner would produce this:
454 define fastcc void @foo() {
458 define void @test(i1 %X) {
459 br i1 %X, label %T.i, label %F.i
464 call fastcc void @foo()
470 The interesting thing about this is that ``%X`` *must* be false for the
471 code to be well-defined, but no amount of dead code elimination will be able
472 to delete the broken call as unreachable. However, since
473 ``instcombine``/``simplifycfg`` turns the undefined call into unreachable, we
474 end up with a branch on a condition that goes to unreachable: a branch to
475 unreachable can never happen, so "``-inline -instcombine -simplifycfg``" is
480 define fastcc void @foo() {
483 define void @test(i1 %X) {
485 call fastcc void @foo()