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