1 ==============================
2 LLVM test-suite Makefile Guide
3 ==============================
5 Written by John T. Criswell, Daniel Dunbar, Reid Spencer, and Tanya
14 This document describes the features of the Makefile-based LLVM
15 test-suite. This way of interacting with the test-suite is deprecated in
16 favor of running the test-suite using LNT, but may continue to prove
17 useful for some users. See the Testing Guide's :ref:`test-suite Quickstart
18 <test-suite-quickstart>` section for more information.
23 The ``test-suite`` module contains a number of programs that can be
24 compiled with LLVM and executed. These programs are compiled using the
25 native compiler and various LLVM backends. The output from the program
26 compiled with the native compiler is assumed correct; the results from
27 the other programs are compared to the native program output and pass if
30 When executing tests, it is usually a good idea to start out with a
31 subset of the available tests or programs. This makes test run times
32 smaller at first and later on this is useful to investigate individual
33 test failures. To run some test only on a subset of programs, simply
34 change directory to the programs you want tested and run ``gmake``
35 there. Alternatively, you can run a different test using the ``TEST``
36 variable to change what tests or run on the selected programs (see below
39 In addition for testing correctness, the ``test-suite`` directory also
40 performs timing tests of various LLVM optimizations. It also records
41 compilation times for the compilers and the JIT. This information can be
42 used to compare the effectiveness of LLVM's optimizations and code
45 ``test-suite`` tests are divided into three types of tests: MultiSource,
46 SingleSource, and External.
48 - ``test-suite/SingleSource``
50 The SingleSource directory contains test programs that are only a
51 single source file in size. These are usually small benchmark
52 programs or small programs that calculate a particular value. Several
53 such programs are grouped together in each directory.
55 - ``test-suite/MultiSource``
57 The MultiSource directory contains subdirectories which contain
58 entire programs with multiple source files. Large benchmarks and
59 whole applications go here.
61 - ``test-suite/External``
63 The External directory contains Makefiles for building code that is
64 external to (i.e., not distributed with) LLVM. The most prominent
65 members of this directory are the SPEC 95 and SPEC 2000 benchmark
66 suites. The ``External`` directory does not contain these actual
67 tests, but only the Makefiles that know how to properly compile these
68 programs from somewhere else. The presence and location of these
69 external programs is configured by the test-suite ``configure``
72 Each tree is then subdivided into several categories, including
73 applications, benchmarks, regression tests, code that is strange
74 grammatically, etc. These organizations should be relatively self
77 Some tests are known to fail. Some are bugs that we have not fixed yet;
78 others are features that we haven't added yet (or may never add). In the
79 regression tests, the result for such tests will be XFAIL (eXpected
80 FAILure). In this way, you can tell the difference between an expected
81 and unexpected failure.
83 The tests in the test suite have no such feature at this time. If the
84 test passes, only warnings and other miscellaneous output will be
85 generated. If a test fails, a large <program> FAILED message will be
86 displayed. This will help you separate benign warnings from actual test
89 Running the test suite
90 ======================
92 First, all tests are executed within the LLVM object directory tree.
93 They *are not* executed inside of the LLVM source tree. This is because
94 the test suite creates temporary files during execution.
96 To run the test suite, you need to use the following steps:
98 #. ``cd`` into the ``llvm/projects`` directory in your source tree.
99 #. Check out the ``test-suite`` module with:
103 % svn co http://llvm.org/svn/llvm-project/test-suite/trunk test-suite
105 This will get the test suite into ``llvm/projects/test-suite``.
107 #. Configure and build ``llvm``.
109 #. Configure and build ``llvm-gcc``.
111 #. Install ``llvm-gcc`` somewhere.
113 #. *Re-configure* ``llvm`` from the top level of each build tree (LLVM
114 object directory tree) in which you want to run the test suite, just
115 as you do before building LLVM.
117 During the *re-configuration*, you must either: (1) have ``llvm-gcc``
118 you just built in your path, or (2) specify the directory where your
119 just-built ``llvm-gcc`` is installed using
120 ``--with-llvmgccdir=$LLVM_GCC_DIR``.
122 You must also tell the configure machinery that the test suite is
123 available so it can be configured for your build tree:
127 % cd $LLVM_OBJ_ROOT ; $LLVM_SRC_ROOT/configure [--with-llvmgccdir=$LLVM_GCC_DIR]
129 [Remember that ``$LLVM_GCC_DIR`` is the directory where you
130 *installed* llvm-gcc, not its src or obj directory.]
132 #. You can now run the test suite from your build tree as follows:
136 % cd $LLVM_OBJ_ROOT/projects/test-suite
139 Note that the second and third steps only need to be done once. After
140 you have the suite checked out and configured, you don't need to do it
141 again (unless the test code or configure script changes).
143 Configuring External Tests
144 --------------------------
146 In order to run the External tests in the ``test-suite`` module, you
147 must specify *--with-externals*. This must be done during the
148 *re-configuration* step (see above), and the ``llvm`` re-configuration
149 must recognize the previously-built ``llvm-gcc``. If any of these is
150 missing or neglected, the External tests won't work.
154 * *--with-externals=<directory>*
156 This tells LLVM where to find any external tests. They are expected to
157 be in specifically named subdirectories of <``directory``>. If
158 ``directory`` is left unspecified, ``configure`` uses the default value
159 ``/home/vadve/shared/benchmarks/speccpu2000/benchspec``. Subdirectory
160 names known to LLVM include:
170 Others are added from time to time, and can be determined from
173 Running different tests
174 -----------------------
176 In addition to the regular "whole program" tests, the ``test-suite``
177 module also provides a mechanism for compiling the programs in different
178 ways. If the variable TEST is defined on the ``gmake`` command line, the
179 test system will include a Makefile named
180 ``TEST.<value of TEST variable>.Makefile``. This Makefile can modify
181 build rules to yield different results.
183 For example, the LLVM nightly tester uses ``TEST.nightly.Makefile`` to
184 create the nightly test reports. To run the nightly tests, run
185 ``gmake TEST=nightly``.
187 There are several TEST Makefiles available in the tree. Some of them are
188 designed for internal LLVM research and will not work outside of the
189 LLVM research group. They may still be valuable, however, as a guide to
190 writing your own TEST Makefile for any optimization or analysis passes
191 that you develop with LLVM.
193 Generating test output
194 ----------------------
196 There are a number of ways to run the tests and generate output. The
197 most simple one is simply running ``gmake`` with no arguments. This will
198 compile and run all programs in the tree using a number of different
199 methods and compare results. Any failures are reported in the output,
200 but are likely drowned in the other output. Passes are not reported
203 Somewhat better is running ``gmake TEST=sometest test``, which runs the
204 specified test and usually adds per-program summaries to the output
205 (depending on which sometest you use). For example, the ``nightly`` test
206 explicitly outputs TEST-PASS or TEST-FAIL for every test after each
207 program. Though these lines are still drowned in the output, it's easy
208 to grep the output logs in the Output directories.
210 Even better are the ``report`` and ``report.format`` targets (where
211 ``format`` is one of ``html``, ``csv``, ``text`` or ``graphs``). The
212 exact contents of the report are dependent on which ``TEST`` you are
213 running, but the text results are always shown at the end of the run and
214 the results are always stored in the ``report.<type>.format`` file (when
215 running with ``TEST=<type>``). The ``report`` also generate a file
216 called ``report.<type>.raw.out`` containing the output of the entire
219 Writing custom tests for the test suite
220 ---------------------------------------
222 Assuming you can run the test suite, (e.g.
223 "``gmake TEST=nightly report``" should work), it is really easy to run
224 optimizations or code generator components against every program in the
225 tree, collecting statistics or running custom checks for correctness. At
226 base, this is how the nightly tester works, it's just one example of a
229 Lets say that you have an LLVM optimization pass, and you want to see
230 how many times it triggers. First thing you should do is add an LLVM
231 `statistic <ProgrammersManual.html#Statistic>`_ to your pass, which will
232 tally counts of things you care about.
234 Following this, you can set up a test and a report that collects these
235 and formats them for easy viewing. This consists of two files, a
236 "``test-suite/TEST.XXX.Makefile``" fragment (where XXX is the name of
237 your test) and a "``test-suite/TEST.XXX.report``" file that indicates
238 how to format the output into a table. There are many example reports of
239 various levels of sophistication included with the test suite, and the
240 framework is very general.
242 If you are interested in testing an optimization pass, check out the
243 "libcalls" test as an example. It can be run like this:
247 % cd llvm/projects/test-suite/MultiSource/Benchmarks # or some other level
248 % make TEST=libcalls report
250 This will do a bunch of stuff, then eventually print a table like this:
254 Name | total | #exit |
256 FreeBench/analyzer/analyzer | 51 | 6 |
257 FreeBench/fourinarow/fourinarow | 1 | 1 |
258 FreeBench/neural/neural | 19 | 9 |
259 FreeBench/pifft/pifft | 5 | 3 |
260 MallocBench/cfrac/cfrac | 1 | * |
261 MallocBench/espresso/espresso | 52 | 12 |
262 MallocBench/gs/gs | 4 | * |
263 Prolangs-C/TimberWolfMC/timberwolfmc | 302 | * |
264 Prolangs-C/agrep/agrep | 33 | 12 |
265 Prolangs-C/allroots/allroots | * | * |
266 Prolangs-C/assembler/assembler | 47 | * |
267 Prolangs-C/bison/mybison | 74 | * |
270 This basically is grepping the -stats output and displaying it in a
271 table. You can also use the "TEST=libcalls report.html" target to get
272 the table in HTML form, similarly for report.csv and report.tex.
274 The source for this is in ``test-suite/TEST.libcalls.*``. The format is
275 pretty simple: the Makefile indicates how to run the test (in this case,
276 "``opt -simplify-libcalls -stats``"), and the report contains one line
277 for each column of the output. The first value is the header for the
278 column and the second is the regex to grep the output of the command
279 for. There are lots of example reports that can do fancy stuff.