1 ===================== 2 LLVM test-suite Guide 3 ===================== 4 5 .. contents:: 6 :local: 7 8 Overview 9 ======== 10 11 This document describes the features of the Makefile-based LLVM 12 test-suite as well as the cmake based replacement. This way of interacting 13 with the test-suite is deprecated in favor of running the test-suite using LNT, 14 but may continue to prove useful for some users. See the Testing 15 Guide's :ref:`test-suite Quickstart <test-suite-quickstart>` section for more 16 information. 17 18 Test suite Structure 19 ==================== 20 21 The ``test-suite`` module contains a number of programs that can be 22 compiled with LLVM and executed. These programs are compiled using the 23 native compiler and various LLVM backends. The output from the program 24 compiled with the native compiler is assumed correct; the results from 25 the other programs are compared to the native program output and pass if 26 they match. 27 28 When executing tests, it is usually a good idea to start out with a 29 subset of the available tests or programs. This makes test run times 30 smaller at first and later on this is useful to investigate individual 31 test failures. To run some test only on a subset of programs, simply 32 change directory to the programs you want tested and run ``gmake`` 33 there. Alternatively, you can run a different test using the ``TEST`` 34 variable to change what tests or run on the selected programs (see below 35 for more info). 36 37 In addition for testing correctness, the ``test-suite`` directory also 38 performs timing tests of various LLVM optimizations. It also records 39 compilation times for the compilers and the JIT. This information can be 40 used to compare the effectiveness of LLVM's optimizations and code 41 generation. 42 43 ``test-suite`` tests are divided into three types of tests: MultiSource, 44 SingleSource, and External. 45 46 - ``test-suite/SingleSource`` 47 48 The SingleSource directory contains test programs that are only a 49 single source file in size. These are usually small benchmark 50 programs or small programs that calculate a particular value. Several 51 such programs are grouped together in each directory. 52 53 - ``test-suite/MultiSource`` 54 55 The MultiSource directory contains subdirectories which contain 56 entire programs with multiple source files. Large benchmarks and 57 whole applications go here. 58 59 - ``test-suite/External`` 60 61 The External directory contains Makefiles for building code that is 62 external to (i.e., not distributed with) LLVM. The most prominent 63 members of this directory are the SPEC 95 and SPEC 2000 benchmark 64 suites. The ``External`` directory does not contain these actual 65 tests, but only the Makefiles that know how to properly compile these 66 programs from somewhere else. The presence and location of these 67 external programs is configured by the test-suite ``configure`` 68 script. 69 70 Each tree is then subdivided into several categories, including 71 applications, benchmarks, regression tests, code that is strange 72 grammatically, etc. These organizations should be relatively self 73 explanatory. 74 75 Some tests are known to fail. Some are bugs that we have not fixed yet; 76 others are features that we haven't added yet (or may never add). In the 77 regression tests, the result for such tests will be XFAIL (eXpected 78 FAILure). In this way, you can tell the difference between an expected 79 and unexpected failure. 80 81 The tests in the test suite have no such feature at this time. If the 82 test passes, only warnings and other miscellaneous output will be 83 generated. If a test fails, a large <program> FAILED message will be 84 displayed. This will help you separate benign warnings from actual test 85 failures. 86 87 Running the test suite via CMake 88 ================================ 89 90 To run the test suite, you need to use the following steps: 91 92 #. The test suite uses the lit test runner to run the test-suite, 93 you need to have lit installed first. Check out LLVM and install lit: 94 95 .. code-block:: bash 96 97 % svn co http://llvm.org/svn/llvm-project/llvm/trunk llvm 98 % cd llvm/utils/lit 99 % sudo python setup.py install # Or without sudo, install in virtual-env. 100 running install 101 running bdist_egg 102 running egg_info 103 writing lit.egg-info/PKG-INFO 104 ... 105 % lit --version 106 lit 0.5.0dev 107 108 #. Check out the ``test-suite`` module with: 109 110 .. code-block:: bash 111 112 % svn co http://llvm.org/svn/llvm-project/test-suite/trunk test-suite 113 114 #. Use CMake to configure the test suite in a new directory. You cannot build 115 the test suite in the source tree. 116 117 .. code-block:: bash 118 119 % mkdir test-suite-build 120 % cd test-suite-build 121 % cmake ../test-suite 122 123 #. Build the benchmarks, using the makefiles CMake generated. 124 125 .. code-block:: bash 126 127 % make 128 Scanning dependencies of target timeit-target 129 [ 0%] Building C object tools/CMakeFiles/timeit-target.dir/timeit.c.o 130 [ 0%] Linking C executable timeit-target 131 [ 0%] Built target timeit-target 132 Scanning dependencies of target fpcmp-host 133 [ 0%] [TEST_SUITE_HOST_CC] Building host executable fpcmp 134 [ 0%] Built target fpcmp-host 135 Scanning dependencies of target timeit-host 136 [ 0%] [TEST_SUITE_HOST_CC] Building host executable timeit 137 [ 0%] Built target timeit-host 138 139 140 #. Run the tests with lit: 141 142 .. code-block:: bash 143 144 % lit -v -j 1 . -o results.json 145 -- Testing: 474 tests, 1 threads -- 146 PASS: test-suite :: MultiSource/Applications/ALAC/decode/alacconvert-decode.test (1 of 474) 147 ********** TEST 'test-suite :: MultiSource/Applications/ALAC/decode/alacconvert-decode.test' RESULTS ********** 148 compile_time: 0.2192 149 exec_time: 0.0462 150 hash: "59620e187c6ac38b36382685ccd2b63b" 151 size: 83348 152 ********** 153 PASS: test-suite :: MultiSource/Applications/ALAC/encode/alacconvert-encode.test (2 of 474) 154 155 156 Running the test suite via Makefiles (deprecated) 157 ================================================= 158 159 First, all tests are executed within the LLVM object directory tree. 160 They *are not* executed inside of the LLVM source tree. This is because 161 the test suite creates temporary files during execution. 162 163 To run the test suite, you need to use the following steps: 164 165 #. ``cd`` into the ``llvm/projects`` directory in your source tree. 166 #. Check out the ``test-suite`` module with: 167 168 .. code-block:: bash 169 170 % svn co http://llvm.org/svn/llvm-project/test-suite/trunk test-suite 171 172 This will get the test suite into ``llvm/projects/test-suite``. 173 174 #. Configure and build ``llvm``. 175 176 #. Configure and build ``llvm-gcc``. 177 178 #. Install ``llvm-gcc`` somewhere. 179 180 #. *Re-configure* ``llvm`` from the top level of each build tree (LLVM 181 object directory tree) in which you want to run the test suite, just 182 as you do before building LLVM. 183 184 During the *re-configuration*, you must either: (1) have ``llvm-gcc`` 185 you just built in your path, or (2) specify the directory where your 186 just-built ``llvm-gcc`` is installed using 187 ``--with-llvmgccdir=$LLVM_GCC_DIR``. 188 189 You must also tell the configure machinery that the test suite is 190 available so it can be configured for your build tree: 191 192 .. code-block:: bash 193 194 % cd $LLVM_OBJ_ROOT ; $LLVM_SRC_ROOT/configure [--with-llvmgccdir=$LLVM_GCC_DIR] 195 196 [Remember that ``$LLVM_GCC_DIR`` is the directory where you 197 *installed* llvm-gcc, not its src or obj directory.] 198 199 #. You can now run the test suite from your build tree as follows: 200 201 .. code-block:: bash 202 203 % cd $LLVM_OBJ_ROOT/projects/test-suite 204 % make 205 206 Note that the second and third steps only need to be done once. After 207 you have the suite checked out and configured, you don't need to do it 208 again (unless the test code or configure script changes). 209 210 Configuring External Tests 211 -------------------------- 212 213 In order to run the External tests in the ``test-suite`` module, you 214 must specify *--with-externals*. This must be done during the 215 *re-configuration* step (see above), and the ``llvm`` re-configuration 216 must recognize the previously-built ``llvm-gcc``. If any of these is 217 missing or neglected, the External tests won't work. 218 219 * *--with-externals* 220 221 * *--with-externals=<directory>* 222 223 This tells LLVM where to find any external tests. They are expected to 224 be in specifically named subdirectories of <``directory``>. If 225 ``directory`` is left unspecified, ``configure`` uses the default value 226 ``/home/vadve/shared/benchmarks/speccpu2000/benchspec``. Subdirectory 227 names known to LLVM include: 228 229 * spec95 230 231 * speccpu2000 232 233 * speccpu2006 234 235 * povray31 236 237 Others are added from time to time, and can be determined from 238 ``configure``. 239 240 Running different tests 241 ----------------------- 242 243 In addition to the regular "whole program" tests, the ``test-suite`` 244 module also provides a mechanism for compiling the programs in different 245 ways. If the variable TEST is defined on the ``gmake`` command line, the 246 test system will include a Makefile named 247 ``TEST.<value of TEST variable>.Makefile``. This Makefile can modify 248 build rules to yield different results. 249 250 For example, the LLVM nightly tester uses ``TEST.nightly.Makefile`` to 251 create the nightly test reports. To run the nightly tests, run 252 ``gmake TEST=nightly``. 253 254 There are several TEST Makefiles available in the tree. Some of them are 255 designed for internal LLVM research and will not work outside of the 256 LLVM research group. They may still be valuable, however, as a guide to 257 writing your own TEST Makefile for any optimization or analysis passes 258 that you develop with LLVM. 259 260 Generating test output 261 ---------------------- 262 263 There are a number of ways to run the tests and generate output. The 264 most simple one is simply running ``gmake`` with no arguments. This will 265 compile and run all programs in the tree using a number of different 266 methods and compare results. Any failures are reported in the output, 267 but are likely drowned in the other output. Passes are not reported 268 explicitly. 269 270 Somewhat better is running ``gmake TEST=sometest test``, which runs the 271 specified test and usually adds per-program summaries to the output 272 (depending on which sometest you use). For example, the ``nightly`` test 273 explicitly outputs TEST-PASS or TEST-FAIL for every test after each 274 program. Though these lines are still drowned in the output, it's easy 275 to grep the output logs in the Output directories. 276 277 Even better are the ``report`` and ``report.format`` targets (where 278 ``format`` is one of ``html``, ``csv``, ``text`` or ``graphs``). The 279 exact contents of the report are dependent on which ``TEST`` you are 280 running, but the text results are always shown at the end of the run and 281 the results are always stored in the ``report.<type>.format`` file (when 282 running with ``TEST=<type>``). The ``report`` also generate a file 283 called ``report.<type>.raw.out`` containing the output of the entire 284 test run. 285 286 Writing custom tests for the test suite 287 --------------------------------------- 288 289 Assuming you can run the test suite, (e.g. 290 "``gmake TEST=nightly report``" should work), it is really easy to run 291 optimizations or code generator components against every program in the 292 tree, collecting statistics or running custom checks for correctness. At 293 base, this is how the nightly tester works, it's just one example of a 294 general framework. 295 296 Lets say that you have an LLVM optimization pass, and you want to see 297 how many times it triggers. First thing you should do is add an LLVM 298 `statistic <ProgrammersManual.html#Statistic>`_ to your pass, which will 299 tally counts of things you care about. 300 301 Following this, you can set up a test and a report that collects these 302 and formats them for easy viewing. This consists of two files, a 303 "``test-suite/TEST.XXX.Makefile``" fragment (where XXX is the name of 304 your test) and a "``test-suite/TEST.XXX.report``" file that indicates 305 how to format the output into a table. There are many example reports of 306 various levels of sophistication included with the test suite, and the 307 framework is very general. 308 309 If you are interested in testing an optimization pass, check out the 310 "libcalls" test as an example. It can be run like this: 311 312 .. code-block:: bash 313 314 % cd llvm/projects/test-suite/MultiSource/Benchmarks # or some other level 315 % make TEST=libcalls report 316 317 This will do a bunch of stuff, then eventually print a table like this: 318 319 :: 320 321 Name | total | #exit | 322 ... 323 FreeBench/analyzer/analyzer | 51 | 6 | 324 FreeBench/fourinarow/fourinarow | 1 | 1 | 325 FreeBench/neural/neural | 19 | 9 | 326 FreeBench/pifft/pifft | 5 | 3 | 327 MallocBench/cfrac/cfrac | 1 | * | 328 MallocBench/espresso/espresso | 52 | 12 | 329 MallocBench/gs/gs | 4 | * | 330 Prolangs-C/TimberWolfMC/timberwolfmc | 302 | * | 331 Prolangs-C/agrep/agrep | 33 | 12 | 332 Prolangs-C/allroots/allroots | * | * | 333 Prolangs-C/assembler/assembler | 47 | * | 334 Prolangs-C/bison/mybison | 74 | * | 335 ... 336 337 This basically is grepping the -stats output and displaying it in a 338 table. You can also use the "TEST=libcalls report.html" target to get 339 the table in HTML form, similarly for report.csv and report.tex. 340 341 The source for this is in ``test-suite/TEST.libcalls.*``. The format is 342 pretty simple: the Makefile indicates how to run the test (in this case, 343 "``opt -simplify-libcalls -stats``"), and the report contains one line 344 for each column of the output. The first value is the header for the 345 column and the second is the regex to grep the output of the command 346 for. There are lots of example reports that can do fancy stuff. 347
