Home | History | Annotate | Download | only in docs
      1 ===================================================================
      2 How To Cross-Compile Clang/LLVM using Clang/LLVM
      3 ===================================================================
      4 
      5 Introduction
      6 ============
      7 
      8 This document contains information about building LLVM and
      9 Clang on host machine, targeting another platform.
     10 
     11 For more information on how to use Clang as a cross-compiler,
     12 please check http://clang.llvm.org/docs/CrossCompilation.html.
     13 
     14 TODO: Add MIPS and other platforms to this document.
     15 
     16 Cross-Compiling from x86_64 to ARM
     17 ==================================
     18 
     19 In this use case, we'll be using CMake and Ninja, on a Debian-based Linux
     20 system, cross-compiling from an x86_64 host (most Intel and AMD chips
     21 nowadays) to a hard-float ARM target (most ARM targets nowadays).
     22 
     23 The packages you'll need are:
     24 
     25  * ``cmake``
     26  * ``ninja-build`` (from backports in Ubuntu)
     27  * ``gcc-4.7-arm-linux-gnueabihf``
     28  * ``gcc-4.7-multilib-arm-linux-gnueabihf``
     29  * ``binutils-arm-linux-gnueabihf``
     30  * ``libgcc1-armhf-cross``
     31  * ``libsfgcc1-armhf-cross``
     32  * ``libstdc++6-armhf-cross``
     33  * ``libstdc++6-4.7-dev-armhf-cross``
     34 
     35 Configuring CMake
     36 -----------------
     37 
     38 For more information on how to configure CMake for LLVM/Clang,
     39 see :doc:`CMake`.
     40 
     41 The CMake options you need to add are:
     42  * ``-DCMAKE_CROSSCOMPILING=True``
     43  * ``-DCMAKE_INSTALL_PREFIX=<install-dir>``
     44  * ``-DLLVM_TABLEGEN=<path-to-host-bin>/llvm-tblgen``
     45  * ``-DCLANG_TABLEGEN=<path-to-host-bin>/clang-tblgen``
     46  * ``-DLLVM_DEFAULT_TARGET_TRIPLE=arm-linux-gnueabihf``
     47  * ``-DLLVM_TARGET_ARCH=ARM``
     48  * ``-DLLVM_TARGETS_TO_BUILD=ARM``
     49  * ``-DCMAKE_CXX_FLAGS='-target armv7a-linux-gnueabihf -mcpu=cortex-a9 -I/usr/arm-linux-gnueabihf/include/c++/4.7.2/arm-linux-gnueabihf/ -I/usr/arm-linux-gnueabihf/include/ -mfloat-abi=hard -ccc-gcc-name arm-linux-gnueabihf-gcc'``
     50 
     51 The TableGen options are required to compile it with the host compiler,
     52 so you'll need to compile LLVM (or at least ``llvm-tblgen``) to your host
     53 platform before you start. The CXX flags define the target, cpu (which
     54 defaults to ``fpu=VFP3`` with NEON), and forcing the hard-float ABI. If you're
     55 using Clang as a cross-compiler, you will *also* have to set ``-ccc-gcc-name``,
     56 to make sure it picks the correct linker.
     57 
     58 Most of the time, what you want is to have a native compiler to the
     59 platform itself, but not others. It might not even be feasible to
     60 produce x86 binaries from ARM targets, so there's no point in compiling
     61 all back-ends. For that reason, you should also set the
     62 ``TARGETS_TO_BUILD`` to only build the ARM back-end.
     63 
     64 You must set the ``CMAKE_INSTALL_PREFIX``, otherwise a ``ninja install``
     65 will copy ARM binaries to your root filesystem, which is not what you
     66 want.
     67 
     68 Hacks
     69 -----
     70 
     71 There are some bugs in current LLVM, which require some fiddling before
     72 running CMake:
     73 
     74 #. If you're using Clang as the cross-compiler, there is a problem in
     75    the LLVM ARM back-end that is producing absolute relocations on
     76    position-independent code (``R_ARM_THM_MOVW_ABS_NC``), so for now, you
     77    should disable PIC:
     78 
     79    .. code-block:: bash
     80 
     81       -DLLVM_ENABLE_PIC=False
     82 
     83    This is not a problem, since Clang/LLVM libraries are statically
     84    linked anyway, it shouldn't affect much.
     85 
     86 #. The ARM libraries won't be installed in your system, and possibly
     87    not easily installable anyway, so you'll have to build/download
     88    them separately. But the CMake prepare step, which checks for
     89    dependencies, will check the *host* libraries, not the *target*
     90    ones.
     91 
     92    A quick way of getting the libraries is to download them from
     93    a distribution repository, like Debian (http://packages.debian.org/wheezy/),
     94    and download the missing libraries. Note that the ``libXXX``
     95    will have the shared objects (``.so``) and the ``libXXX-dev`` will
     96    give you the headers and the static (``.a``) library. Just in
     97    case, download both.
     98 
     99    The ones you need for ARM are: ``libtinfo``, ``zlib1g``,
    100    ``libxml2`` and ``liblzma``. In the Debian repository you'll
    101    find downloads for all architectures.
    102 
    103    After you download and unpack all ``.deb`` packages, copy all
    104    ``.so`` and ``.a`` to a directory, make the appropriate
    105    symbolic links (if necessary), and add the relevant ``-L``
    106    and ``-I`` paths to ``-DCMAKE_CXX_FLAGS`` above.
    107 
    108 
    109 Running CMake and Building
    110 --------------------------
    111 
    112 Finally, if you're using your platform compiler, run:
    113 
    114    .. code-block:: bash
    115 
    116      $ cmake -G Ninja <source-dir> <options above>
    117 
    118 If you're using Clang as the cross-compiler, run:
    119 
    120    .. code-block:: bash
    121 
    122      $ CC='clang' CXX='clang++' cmake -G Ninja <source-dir> <options above>
    123 
    124 If you have ``clang``/``clang++`` on the path, it should just work, and special
    125 Ninja files will be created in the build directory. I strongly suggest
    126 you to run ``cmake`` on a separate build directory, *not* inside the
    127 source tree.
    128 
    129 To build, simply type:
    130 
    131    .. code-block:: bash
    132 
    133      $ ninja
    134 
    135 It should automatically find out how many cores you have, what are
    136 the rules that needs building and will build the whole thing.
    137 
    138 You can't run ``ninja check-all`` on this tree because the created
    139 binaries are targeted to ARM, not x86_64.
    140 
    141 Installing and Using
    142 --------------------
    143 
    144 After the LLVM/Clang has built successfully, you should install it
    145 via:
    146 
    147    .. code-block:: bash
    148 
    149      $ ninja install
    150 
    151 which will create a sysroot on the install-dir. You can then tar
    152 that directory into a binary with the full triple name (for easy
    153 identification), like:
    154 
    155    .. code-block:: bash
    156 
    157      $ ln -sf <install-dir> arm-linux-gnueabihf-clang
    158      $ tar zchf arm-linux-gnueabihf-clang.tar.gz arm-linux-gnueabihf-clang
    159 
    160 If you copy that tarball to your target board, you'll be able to use
    161 it for running the test-suite, for example. Follow the guidelines at
    162 http://llvm.org/docs/lnt/quickstart.html, unpack the tarball in the
    163 test directory, and use options:
    164 
    165    .. code-block:: bash
    166 
    167      $ ./sandbox/bin/python sandbox/bin/lnt runtest nt \
    168          --sandbox sandbox \
    169          --test-suite `pwd`/test-suite \
    170          --cc `pwd`/arm-linux-gnueabihf-clang/bin/clang \
    171          --cxx `pwd`/arm-linux-gnueabihf-clang/bin/clang++
    172 
    173 Remember to add the ``-jN`` options to ``lnt`` to the number of CPUs
    174 on your board. Also, the path to your clang has to be absolute, so
    175 you'll need the `pwd` trick above.
    176