1=================================
2LLVM Testing Infrastructure Guide
3=================================
4
5.. contents::
6   :local:
7
8.. toctree::
9   :hidden:
10
11   TestSuiteMakefileGuide
12
13Overview
14========
15
16This document is the reference manual for the LLVM testing
17infrastructure. It documents the structure of the LLVM testing
18infrastructure, the tools needed to use it, and how to add and run
19tests.
20
21Requirements
22============
23
24In order to use the LLVM testing infrastructure, you will need all of the
25software required to build LLVM, as well as `Python <http://python.org>`_ 2.7 or
26later.
27
28LLVM testing infrastructure organization
29========================================
30
31The LLVM testing infrastructure contains two major categories of tests:
32regression tests and whole programs. The regression tests are contained
33inside the LLVM repository itself under ``llvm/test`` and are expected
34to always pass -- they should be run before every commit.
35
36The whole programs tests are referred to as the "LLVM test suite" (or
37"test-suite") and are in the ``test-suite`` module in subversion. For
38historical reasons, these tests are also referred to as the "nightly
39tests" in places, which is less ambiguous than "test-suite" and remains
40in use although we run them much more often than nightly.
41
42Regression tests
43----------------
44
45The regression tests are small pieces of code that test a specific
46feature of LLVM or trigger a specific bug in LLVM. The language they are
47written in depends on the part of LLVM being tested. These tests are driven by
48the :doc:`Lit <CommandGuide/lit>` testing tool (which is part of LLVM), and
49are located in the ``llvm/test`` directory.
50
51Typically when a bug is found in LLVM, a regression test containing just
52enough code to reproduce the problem should be written and placed
53somewhere underneath this directory. For example, it can be a small
54piece of LLVM IR distilled from an actual application or benchmark.
55
56``test-suite``
57--------------
58
59The test suite contains whole programs, which are pieces of code which
60can be compiled and linked into a stand-alone program that can be
61executed. These programs are generally written in high level languages
62such as C or C++.
63
64These programs are compiled using a user specified compiler and set of
65flags, and then executed to capture the program output and timing
66information. The output of these programs is compared to a reference
67output to ensure that the program is being compiled correctly.
68
69In addition to compiling and executing programs, whole program tests
70serve as a way of benchmarking LLVM performance, both in terms of the
71efficiency of the programs generated as well as the speed with which
72LLVM compiles, optimizes, and generates code.
73
74The test-suite is located in the ``test-suite`` Subversion module.
75
76Debugging Information tests
77---------------------------
78
79The test suite contains tests to check quality of debugging information.
80The test are written in C based languages or in LLVM assembly language.
81
82These tests are compiled and run under a debugger. The debugger output
83is checked to validate of debugging information. See README.txt in the
84test suite for more information . This test suite is located in the
85``debuginfo-tests`` Subversion module.
86
87Quick start
88===========
89
90The tests are located in two separate Subversion modules. The
91regressions tests are in the main "llvm" module under the directory
92``llvm/test`` (so you get these tests for free with the main LLVM tree).
93Use ``make check-all`` to run the regression tests after building LLVM.
94
95The more comprehensive test suite that includes whole programs in C and C++
96is in the ``test-suite`` module. See :ref:`test-suite Quickstart
97<test-suite-quickstart>` for more information on running these tests.
98
99Regression tests
100----------------
101
102To run all of the LLVM regression tests use the check-llvm target:
103
104.. code-block:: bash
105
106    % make check-llvm
107
108If you have `Clang <http://clang.llvm.org/>`_ checked out and built, you
109can run the LLVM and Clang tests simultaneously using:
110
111.. code-block:: bash
112
113    % make check-all
114
115To run the tests with Valgrind (Memcheck by default), use the ``LIT_ARGS`` make
116variable to pass the required options to lit. For example, you can use:
117
118.. code-block:: bash
119
120    % make check LIT_ARGS="-v --vg --vg-leak"
121
122to enable testing with valgrind and with leak checking enabled.
123
124To run individual tests or subsets of tests, you can use the ``llvm-lit``
125script which is built as part of LLVM. For example, to run the
126``Integer/BitPacked.ll`` test by itself you can run:
127
128.. code-block:: bash
129
130    % llvm-lit ~/llvm/test/Integer/BitPacked.ll
131
132or to run all of the ARM CodeGen tests:
133
134.. code-block:: bash
135
136    % llvm-lit ~/llvm/test/CodeGen/ARM
137
138For more information on using the :program:`lit` tool, see ``llvm-lit --help``
139or the :doc:`lit man page <CommandGuide/lit>`.
140
141Debugging Information tests
142---------------------------
143
144To run debugging information tests simply checkout the tests inside
145clang/test directory.
146
147.. code-block:: bash
148
149    % cd clang/test
150    % svn co http://llvm.org/svn/llvm-project/debuginfo-tests/trunk debuginfo-tests
151
152These tests are already set up to run as part of clang regression tests.
153
154Regression test structure
155=========================
156
157The LLVM regression tests are driven by :program:`lit` and are located in the
158``llvm/test`` directory.
159
160This directory contains a large array of small tests that exercise
161various features of LLVM and to ensure that regressions do not occur.
162The directory is broken into several sub-directories, each focused on a
163particular area of LLVM.
164
165Writing new regression tests
166----------------------------
167
168The regression test structure is very simple, but does require some
169information to be set. This information is gathered via ``configure``
170and is written to a file, ``test/lit.site.cfg`` in the build directory.
171The ``llvm/test`` Makefile does this work for you.
172
173In order for the regression tests to work, each directory of tests must
174have a ``lit.local.cfg`` file. :program:`lit` looks for this file to determine
175how to run the tests. This file is just Python code and thus is very
176flexible, but we've standardized it for the LLVM regression tests. If
177you're adding a directory of tests, just copy ``lit.local.cfg`` from
178another directory to get running. The standard ``lit.local.cfg`` simply
179specifies which files to look in for tests. Any directory that contains
180only directories does not need the ``lit.local.cfg`` file. Read the :doc:`Lit
181documentation <CommandGuide/lit>` for more information.
182
183Each test file must contain lines starting with "RUN:" that tell :program:`lit`
184how to run it. If there are no RUN lines, :program:`lit` will issue an error
185while running a test.
186
187RUN lines are specified in the comments of the test program using the
188keyword ``RUN`` followed by a colon, and lastly the command (pipeline)
189to execute. Together, these lines form the "script" that :program:`lit`
190executes to run the test case. The syntax of the RUN lines is similar to a
191shell's syntax for pipelines including I/O redirection and variable
192substitution. However, even though these lines may *look* like a shell
193script, they are not. RUN lines are interpreted by :program:`lit`.
194Consequently, the syntax differs from shell in a few ways. You can specify
195as many RUN lines as needed.
196
197:program:`lit` performs substitution on each RUN line to replace LLVM tool names
198with the full paths to the executable built for each tool (in
199``$(LLVM_OBJ_ROOT)/$(BuildMode)/bin)``. This ensures that :program:`lit` does
200not invoke any stray LLVM tools in the user's path during testing.
201
202Each RUN line is executed on its own, distinct from other lines unless
203its last character is ``\``. This continuation character causes the RUN
204line to be concatenated with the next one. In this way you can build up
205long pipelines of commands without making huge line lengths. The lines
206ending in ``\`` are concatenated until a RUN line that doesn't end in
207``\`` is found. This concatenated set of RUN lines then constitutes one
208execution. :program:`lit` will substitute variables and arrange for the pipeline
209to be executed. If any process in the pipeline fails, the entire line (and
210test case) fails too.
211
212Below is an example of legal RUN lines in a ``.ll`` file:
213
214.. code-block:: llvm
215
216    ; RUN: llvm-as < %s | llvm-dis > %t1
217    ; RUN: llvm-dis < %s.bc-13 > %t2
218    ; RUN: diff %t1 %t2
219
220As with a Unix shell, the RUN lines permit pipelines and I/O
221redirection to be used.
222
223There are some quoting rules that you must pay attention to when writing
224your RUN lines. In general nothing needs to be quoted. :program:`lit` won't
225strip off any quote characters so they will get passed to the invoked program.
226To avoid this use curly braces to tell :program:`lit` that it should treat
227everything enclosed as one value.
228
229In general, you should strive to keep your RUN lines as simple as possible,
230using them only to run tools that generate textual output you can then examine.
231The recommended way to examine output to figure out if the test passes is using
232the :doc:`FileCheck tool <CommandGuide/FileCheck>`. *[The usage of grep in RUN
233lines is deprecated - please do not send or commit patches that use it.]*
234
235Put related tests into a single file rather than having a separate file per
236test. Check if there are files already covering your feature and consider
237adding your code there instead of creating a new file.
238
239Extra files
240-----------
241
242If your test requires extra files besides the file containing the ``RUN:``
243lines, the idiomatic place to put them is in a subdirectory ``Inputs``.
244You can then refer to the extra files as ``%S/Inputs/foo.bar``.
245
246For example, consider ``test/Linker/ident.ll``. The directory structure is
247as follows::
248
249  test/
250    Linker/
251      ident.ll
252      Inputs/
253        ident.a.ll
254        ident.b.ll
255
256For convenience, these are the contents:
257
258.. code-block:: llvm
259
260  ;;;;; ident.ll:
261
262  ; RUN: llvm-link %S/Inputs/ident.a.ll %S/Inputs/ident.b.ll -S | FileCheck %s
263
264  ; Verify that multiple input llvm.ident metadata are linked together.
265
266  ; CHECK-DAG: !llvm.ident = !{!0, !1, !2}
267  ; CHECK-DAG: "Compiler V1"
268  ; CHECK-DAG: "Compiler V2"
269  ; CHECK-DAG: "Compiler V3"
270
271  ;;;;; Inputs/ident.a.ll:
272
273  !llvm.ident = !{!0, !1}
274  !0 = metadata !{metadata !"Compiler V1"}
275  !1 = metadata !{metadata !"Compiler V2"}
276
277  ;;;;; Inputs/ident.b.ll:
278
279  !llvm.ident = !{!0}
280  !0 = metadata !{metadata !"Compiler V3"}
281
282For symmetry reasons, ``ident.ll`` is just a dummy file that doesn't
283actually participate in the test besides holding the ``RUN:`` lines.
284
285.. note::
286
287  Some existing tests use ``RUN: true`` in extra files instead of just
288  putting the extra files in an ``Inputs/`` directory. This pattern is
289  deprecated.
290
291Fragile tests
292-------------
293
294It is easy to write a fragile test that would fail spuriously if the tool being
295tested outputs a full path to the input file.  For example, :program:`opt` by
296default outputs a ``ModuleID``:
297
298.. code-block:: console
299
300  $ cat example.ll
301  define i32 @main() nounwind {
302      ret i32 0
303  }
304
305  $ opt -S /path/to/example.ll
306  ; ModuleID = '/path/to/example.ll'
307
308  define i32 @main() nounwind {
309      ret i32 0
310  }
311
312``ModuleID`` can unexpetedly match against ``CHECK`` lines.  For example:
313
314.. code-block:: llvm
315
316  ; RUN: opt -S %s | FileCheck
317
318  define i32 @main() nounwind {
319      ; CHECK-NOT: load
320      ret i32 0
321  }
322
323This test will fail if placed into a ``download`` directory.
324
325To make your tests robust, always use ``opt ... < %s`` in the RUN line.
326:program:`opt` does not output a ``ModuleID`` when input comes from stdin.
327
328Platform-Specific Tests
329-----------------------
330
331Whenever adding tests that require the knowledge of a specific platform,
332either related to code generated, specific output or back-end features,
333you must make sure to isolate the features, so that buildbots that
334run on different architectures (and don't even compile all back-ends),
335don't fail.
336
337The first problem is to check for target-specific output, for example sizes
338of structures, paths and architecture names, for example:
339
340* Tests containing Windows paths will fail on Linux and vice-versa.
341* Tests that check for ``x86_64`` somewhere in the text will fail anywhere else.
342* Tests where the debug information calculates the size of types and structures.
343
344Also, if the test rely on any behaviour that is coded in any back-end, it must
345go in its own directory. So, for instance, code generator tests for ARM go
346into ``test/CodeGen/ARM`` and so on. Those directories contain a special
347``lit`` configuration file that ensure all tests in that directory will
348only run if a specific back-end is compiled and available.
349
350For instance, on ``test/CodeGen/ARM``, the ``lit.local.cfg`` is:
351
352.. code-block:: python
353
354  config.suffixes = ['.ll', '.c', '.cpp', '.test']
355  if not 'ARM' in config.root.targets:
356    config.unsupported = True
357
358Other platform-specific tests are those that depend on a specific feature
359of a specific sub-architecture, for example only to Intel chips that support ``AVX2``.
360
361For instance, ``test/CodeGen/X86/psubus.ll`` tests three sub-architecture
362variants:
363
364.. code-block:: llvm
365
366  ; RUN: llc -mcpu=core2 < %s | FileCheck %s -check-prefix=SSE2
367  ; RUN: llc -mcpu=corei7-avx < %s | FileCheck %s -check-prefix=AVX1
368  ; RUN: llc -mcpu=core-avx2 < %s | FileCheck %s -check-prefix=AVX2
369
370And the checks are different:
371
372.. code-block:: llvm
373
374  ; SSE2: @test1
375  ; SSE2: psubusw LCPI0_0(%rip), %xmm0
376  ; AVX1: @test1
377  ; AVX1: vpsubusw LCPI0_0(%rip), %xmm0, %xmm0
378  ; AVX2: @test1
379  ; AVX2: vpsubusw LCPI0_0(%rip), %xmm0, %xmm0
380
381So, if you're testing for a behaviour that you know is platform-specific or
382depends on special features of sub-architectures, you must add the specific
383triple, test with the specific FileCheck and put it into the specific
384directory that will filter out all other architectures.
385
386
387Substitutions
388-------------
389
390Besides replacing LLVM tool names the following substitutions are performed in
391RUN lines:
392
393``%%``
394   Replaced by a single ``%``. This allows escaping other substitutions.
395
396``%s``
397   File path to the test case's source. This is suitable for passing on the
398   command line as the input to an LLVM tool.
399
400   Example: ``/home/user/llvm/test/MC/ELF/foo_test.s``
401
402``%S``
403   Directory path to the test case's source.
404
405   Example: ``/home/user/llvm/test/MC/ELF``
406
407``%t``
408   File path to a temporary file name that could be used for this test case.
409   The file name won't conflict with other test cases. You can append to it
410   if you need multiple temporaries. This is useful as the destination of
411   some redirected output.
412
413   Example: ``/home/user/llvm.build/test/MC/ELF/Output/foo_test.s.tmp``
414
415``%T``
416   Directory of ``%t``.
417
418   Example: ``/home/user/llvm.build/test/MC/ELF/Output``
419
420``%{pathsep}``
421
422   Expands to the path separator, i.e. ``:`` (or ``;`` on Windows).
423
424
425**LLVM-specific substitutions:**
426
427``%shlibext``
428   The suffix for the host platforms shared library files. This includes the
429   period as the first character.
430
431   Example: ``.so`` (Linux), ``.dylib`` (OS X), ``.dll`` (Windows)
432
433``%exeext``
434   The suffix for the host platforms executable files. This includes the
435   period as the first character.
436
437   Example: ``.exe`` (Windows), empty on Linux.
438
439``%(line)``, ``%(line+<number>)``, ``%(line-<number>)``
440   The number of the line where this substitution is used, with an optional
441   integer offset. This can be used in tests with multiple RUN lines, which
442   reference test file's line numbers.
443
444
445**Clang-specific substitutions:**
446
447``%clang``
448   Invokes the Clang driver.
449
450``%clang_cpp``
451   Invokes the Clang driver for C++.
452
453``%clang_cl``
454   Invokes the CL-compatible Clang driver.
455
456``%clangxx``
457   Invokes the G++-compatible Clang driver.
458
459``%clang_cc1``
460   Invokes the Clang frontend.
461
462``%itanium_abi_triple``, ``%ms_abi_triple``
463   These substitutions can be used to get the current target triple adjusted to
464   the desired ABI. For example, if the test suite is running with the
465   ``i686-pc-win32`` target, ``%itanium_abi_triple`` will expand to
466   ``i686-pc-mingw32``. This allows a test to run with a specific ABI without
467   constraining it to a specific triple.
468
469To add more substituations, look at ``test/lit.cfg`` or ``lit.local.cfg``.
470
471
472Options
473-------
474
475The llvm lit configuration allows to customize some things with user options:
476
477``llc``, ``opt``, ...
478    Substitute the respective llvm tool name with a custom command line. This
479    allows to specify custom paths and default arguments for these tools.
480    Example:
481
482    % llvm-lit "-Dllc=llc -verify-machineinstrs"
483
484``run_long_tests``
485    Enable the execution of long running tests.
486
487``llvm_site_config``
488    Load the specified lit configuration instead of the default one.
489
490
491Other Features
492--------------
493
494To make RUN line writing easier, there are several helper programs. These
495helpers are in the PATH when running tests, so you can just call them using
496their name. For example:
497
498``not``
499   This program runs its arguments and then inverts the result code from it.
500   Zero result codes become 1. Non-zero result codes become 0.
501
502Sometimes it is necessary to mark a test case as "expected fail" or
503XFAIL. You can easily mark a test as XFAIL just by including ``XFAIL:``
504on a line near the top of the file. This signals that the test case
505should succeed if the test fails. Such test cases are counted separately
506by the testing tool. To specify an expected fail, use the XFAIL keyword
507in the comments of the test program followed by a colon and one or more
508failure patterns. Each failure pattern can be either ``*`` (to specify
509fail everywhere), or a part of a target triple (indicating the test
510should fail on that platform), or the name of a configurable feature
511(for example, ``loadable_module``). If there is a match, the test is
512expected to fail. If not, the test is expected to succeed. To XFAIL
513everywhere just specify ``XFAIL: *``. Here is an example of an ``XFAIL``
514line:
515
516.. code-block:: llvm
517
518    ; XFAIL: darwin,sun
519
520To make the output more useful, :program:`lit` will scan
521the lines of the test case for ones that contain a pattern that matches
522``PR[0-9]+``. This is the syntax for specifying a PR (Problem Report) number
523that is related to the test case. The number after "PR" specifies the
524LLVM bugzilla number. When a PR number is specified, it will be used in
525the pass/fail reporting. This is useful to quickly get some context when
526a test fails.
527
528Finally, any line that contains "END." will cause the special
529interpretation of lines to terminate. This is generally done right after
530the last RUN: line. This has two side effects:
531
532(a) it prevents special interpretation of lines that are part of the test
533    program, not the instructions to the test case, and
534
535(b) it speeds things up for really big test cases by avoiding
536    interpretation of the remainder of the file.
537
538``test-suite`` Overview
539=======================
540
541The ``test-suite`` module contains a number of programs that can be
542compiled and executed. The ``test-suite`` includes reference outputs for
543all of the programs, so that the output of the executed program can be
544checked for correctness.
545
546``test-suite`` tests are divided into three types of tests: MultiSource,
547SingleSource, and External.
548
549-  ``test-suite/SingleSource``
550
551   The SingleSource directory contains test programs that are only a
552   single source file in size. These are usually small benchmark
553   programs or small programs that calculate a particular value. Several
554   such programs are grouped together in each directory.
555
556-  ``test-suite/MultiSource``
557
558   The MultiSource directory contains subdirectories which contain
559   entire programs with multiple source files. Large benchmarks and
560   whole applications go here.
561
562-  ``test-suite/External``
563
564   The External directory contains Makefiles for building code that is
565   external to (i.e., not distributed with) LLVM. The most prominent
566   members of this directory are the SPEC 95 and SPEC 2000 benchmark
567   suites. The ``External`` directory does not contain these actual
568   tests, but only the Makefiles that know how to properly compile these
569   programs from somewhere else. When using ``LNT``, use the
570   ``--test-externals`` option to include these tests in the results.
571
572.. _test-suite-quickstart:
573
574``test-suite`` Quickstart
575-------------------------
576
577The modern way of running the ``test-suite`` is focused on testing and
578benchmarking complete compilers using the
579`LNT <http://llvm.org/docs/lnt>`_ testing infrastructure.
580
581For more information on using LNT to execute the ``test-suite``, please
582see the `LNT Quickstart <http://llvm.org/docs/lnt/quickstart.html>`_
583documentation.
584
585``test-suite`` Makefiles
586------------------------
587
588Historically, the ``test-suite`` was executed using a complicated setup
589of Makefiles. The LNT based approach above is recommended for most
590users, but there are some testing scenarios which are not supported by
591the LNT approach. In addition, LNT currently uses the Makefile setup
592under the covers and so developers who are interested in how LNT works
593under the hood may want to understand the Makefile based setup.
594
595For more information on the ``test-suite`` Makefile setup, please see
596the :doc:`Test Suite Makefile Guide <TestSuiteMakefileGuide>`.
597