1================================ 2How to submit an LLVM bug report 3================================ 4 5Introduction - Got bugs? 6======================== 7 8 9If you're working with LLVM and run into a bug, we definitely want to know 10about it. This document describes what you can do to increase the odds of 11getting it fixed quickly. 12 13�� If you believe that the bug is security related, please follow :ref:`report-security-issue`. �� 14 15Basically you have to do two things at a minimum. First, decide whether the 16bug `crashes the compiler`_ or if the compiler is `miscompiling`_ the program 17(i.e., the compiler successfully produces an executable, but it doesn't run 18right). Based on what type of bug it is, follow the instructions in the 19linked section to narrow down the bug so that the person who fixes it will be 20able to find the problem more easily. 21 22Once you have a reduced test-case, go to `the LLVM Bug Tracking System 23<https://github.com/llvm/llvm-project/issues>`_ and fill out the form with the 24necessary details (note that you don't need to pick a label, just use if you're 25not sure). The bug description should contain the following information: 26 27* All information necessary to reproduce the problem. 28* The reduced test-case that triggers the bug. 29* The location where you obtained LLVM (if not from our Git 30 repository). 31 32Thanks for helping us make LLVM better! 33 34.. _crashes the compiler: 35 36Crashing Bugs 37============= 38 39More often than not, bugs in the compiler cause it to crash---often due to 40an assertion failure of some sort. The most important piece of the puzzle 41is to figure out if it is crashing in the Clang front-end or if it is one of 42the LLVM libraries (e.g. the optimizer or code generator) that has 43problems. 44 45To figure out which component is crashing (the front-end, middle-end 46optimizer, or backend code generator), run the ``clang`` command line as you 47were when the crash occurred, but with the following extra command line 48options: 49 50* ``-emit-llvm -Xclang -disable-llvm-passes``: If ``clang`` still crashes when 51 passed these options (which disable the optimizer and code generator), then 52 the crash is in the front-end. Jump ahead to :ref:`front-end bugs 53 <frontend-crash>`. 54 55* ``-emit-llvm``: If ``clang`` crashes with this option (which disables 56 the code generator), you found a middle-end optimizer bug. Jump ahead to 57 :ref:`middle-end bugs <middleend-crash>`. 58 59* Otherwise, you have a backend code generator crash. Jump ahead to :ref:`code 60 generator bugs <backend-crash>`. 61 62.. _frontend-crash: 63 64Front-end bugs 65-------------- 66 67On a ``clang`` crash, the compiler will dump a preprocessed file and a script 68to replay the ``clang`` command. For example, you should see something like 69 70.. code-block:: text 71 72 PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT: 73 Preprocessed source(s) and associated run script(s) are located at: 74 clang: note: diagnostic msg: /tmp/foo-xxxxxx.c 75 clang: note: diagnostic msg: /tmp/foo-xxxxxx.sh 76 77The `creduce <https://github.com/csmith-project/creduce>`_ tool helps to 78reduce the preprocessed file down to the smallest amount of code that still 79replicates the problem. You're encouraged to use creduce to reduce the code 80to make the developers' lives easier. The 81``clang/utils/creduce-clang-crash.py`` script can be used on the files 82that clang dumps to help with automating creating a test to check for the 83compiler crash. 84 85`cvise <https://github.com/marxin/cvise>`_ is an alternative to ``creduce``. 86 87.. _middleend-crash: 88 89Middle-end optimization bugs 90---------------------------- 91 92If you find that a bug crashes in the optimizer, compile your test-case to a 93``.bc`` file by passing "``-emit-llvm -O1 -Xclang -disable-llvm-passes -c -o 94foo.bc``". The ``-O1`` is important because ``-O0`` adds the ``optnone`` 95function attribute to all functions and many passes don't run on ``optnone`` 96functions. Then run: 97 98.. code-block:: bash 99 100 opt -O3 foo.bc -disable-output 101 102If this doesn't crash, please follow the instructions for a :ref:`front-end 103bug <frontend-crash>`. 104 105If this does crash, then you should be able to debug this with the following 106:doc:`bugpoint <Bugpoint>` command: 107 108.. code-block:: bash 109 110 bugpoint foo.bc -O3 111 112Run this, then file a bug with the instructions and reduced .bc 113files that bugpoint emits. 114 115If bugpoint doesn't reproduce the crash, ``llvm-reduce`` is an alternative 116way to reduce LLVM IR. Create a script that repros the crash and run: 117 118.. code-block:: bash 119 120 llvm-reduce --test=path/to/script foo.bc 121 122which should produce reduced IR that reproduces the crash. Be warned the 123``llvm-reduce`` is still fairly immature and may crash. 124 125If none of the above work, you can get the IR before a crash by running the 126``opt`` command with the ``--print-before-all --print-module-scope`` flags to 127dump the IR before every pass. Be warned that this is very verbose. 128 129.. _backend-crash: 130 131Backend code generator bugs 132--------------------------- 133 134If you find a bug that crashes clang in the code generator, compile your 135source file to a .bc file by passing "``-emit-llvm -c -o foo.bc``" to 136clang (in addition to the options you already pass). Once your have 137foo.bc, one of the following commands should fail: 138 139#. ``llc foo.bc`` 140#. ``llc foo.bc -relocation-model=pic`` 141#. ``llc foo.bc -relocation-model=static`` 142 143If none of these crash, please follow the instructions for a :ref:`front-end 144bug<frontend-crash>`. If one of these do crash, you should be able to reduce 145this with one of the following :doc:`bugpoint <Bugpoint>` command lines (use 146the one corresponding to the command above that failed): 147 148#. ``bugpoint -run-llc foo.bc`` 149#. ``bugpoint -run-llc foo.bc --tool-args -relocation-model=pic`` 150#. ``bugpoint -run-llc foo.bc --tool-args -relocation-model=static`` 151 152Please run this, then file a bug with the instructions and reduced .bc file 153that bugpoint emits. If something goes wrong with bugpoint, please submit 154the "foo.bc" file and the option that llc crashes with. 155 156.. _miscompiling: 157 158Miscompilations 159=============== 160 161If clang successfully produces an executable, but that executable doesn't run 162right, this is either a bug in the code or a bug in the compiler. The first 163thing to check is to make sure it is not using undefined behavior (e.g. 164reading a variable before it is defined). In particular, check to see if the 165program is clean under various `sanitizers 166<https://github.com/google/sanitizers>`_ (e.g. ``clang 167-fsanitize=undefined,address``) and `valgrind <http://valgrind.org/>`_. Many 168"LLVM bugs" that we have chased down ended up being bugs in the program being 169compiled, not LLVM. 170 171Once you determine that the program itself is not buggy, you should choose 172which code generator you wish to compile the program with (e.g. LLC or the JIT) 173and optionally a series of LLVM passes to run. For example: 174 175.. code-block:: bash 176 177 bugpoint -run-llc [... optzn passes ...] file-to-test.bc --args -- [program arguments] 178 179bugpoint will try to narrow down your list of passes to the one pass that 180causes an error, and simplify the bitcode file as much as it can to assist 181you. It will print a message letting you know how to reproduce the 182resulting error. 183 184The :doc:`OptBisect <OptBisect>` page shows an alternative method for finding 185incorrect optimization passes. 186 187Incorrect code generation 188========================= 189 190Similarly to debugging incorrect compilation by mis-behaving passes, you 191can debug incorrect code generation by either LLC or the JIT, using 192``bugpoint``. The process ``bugpoint`` follows in this case is to try to 193narrow the code down to a function that is miscompiled by one or the other 194method, but since for correctness, the entire program must be run, 195``bugpoint`` will compile the code it deems to not be affected with the C 196Backend, and then link in the shared object it generates. 197 198To debug the JIT: 199 200.. code-block:: bash 201 202 bugpoint -run-jit -output=[correct output file] [bitcode file] \ 203 --tool-args -- [arguments to pass to lli] \ 204 --args -- [program arguments] 205 206Similarly, to debug the LLC, one would run: 207 208.. code-block:: bash 209 210 bugpoint -run-llc -output=[correct output file] [bitcode file] \ 211 --tool-args -- [arguments to pass to llc] \ 212 --args -- [program arguments] 213 214**Special note:** if you are debugging MultiSource or SPEC tests that 215already exist in the ``llvm/test`` hierarchy, there is an easier way to 216debug the JIT, LLC, and CBE, using the pre-written Makefile targets, which 217will pass the program options specified in the Makefiles: 218 219.. code-block:: bash 220 221 cd llvm/test/../../program 222 make bugpoint-jit 223 224At the end of a successful ``bugpoint`` run, you will be presented 225with two bitcode files: a *safe* file which can be compiled with the C 226backend and the *test* file which either LLC or the JIT 227mis-codegenerates, and thus causes the error. 228 229To reproduce the error that ``bugpoint`` found, it is sufficient to do 230the following: 231 232#. Regenerate the shared object from the safe bitcode file: 233 234 .. code-block:: bash 235 236 llc -march=c safe.bc -o safe.c 237 gcc -shared safe.c -o safe.so 238 239#. If debugging LLC, compile test bitcode native and link with the shared 240 object: 241 242 .. code-block:: bash 243 244 llc test.bc -o test.s 245 gcc test.s safe.so -o test.llc 246 ./test.llc [program options] 247 248#. If debugging the JIT, load the shared object and supply the test 249 bitcode: 250 251 .. code-block:: bash 252 253 lli -load=safe.so test.bc [program options] 254