| blob | 43c414d4c974ef877cabcca5e699fbb5a42fe9ec |
3 <html>
4 <head>
7 </head>
8 <body>
11 LLVM Testing Infrastructure Guide
12 </div>
14 <ol>
18 <ul>
21 </ul>
22 </li>
24 <ul>
27 </ul>
28 </li>
30 <ul>
35 </ul>
36 </li>
39 <ul>
44 </ul>
45 </li>
47 </ol>
52 </div>
54 <!--=========================================================================-->
56 <!--=========================================================================-->
60 <p>This document is the reference manual for the LLVM testing infrastructure. It documents
61 the structure of the LLVM testing infrastructure, the tools needed to use it,
62 and how to add and run tests.</p>
64 </div>
66 <!--=========================================================================-->
68 <!--=========================================================================-->
72 <p>In order to use the LLVM testing infrastructure, you will need all of the software
73 required to build LLVM, plus the following:</p>
75 <dl>
82 </dl>
84 </div>
86 <!--=========================================================================-->
88 <!--=========================================================================-->
92 <p>The LLVM testing infrastructure contains two major categories of tests: code
93 fragments and whole programs. Code fragments are referred to as the "DejaGNU
97 </p>
99 </div>
101 <!-- _______________________________________________________________________ -->
103 <!-- _______________________________________________________________________ -->
107 <p>Code fragments are small pieces of code that test a specific
108 feature of LLVM or trigger a specific bug in LLVM. They are usually
109 written in LLVM assembly language, but can be written in other
110 languages if the test targets a particular language front end (and the
113 are driven by the DejaGNU testing framework, which is hidden behind a
114 few simple makefiles.</p>
116 <p>These code fragments are not complete programs. The code generated
117 from them is never executed to determine correct behavior.</p>
120 directory.</p>
122 <p>Typically when a bug is found in LLVM, a regression test containing
123 just enough code to reproduce the problem should be written and placed
124 somewhere underneath this directory. In most cases, this will be a small
125 piece of LLVM assembly language code, often distilled from an actual
126 application or benchmark.</p>
128 </div>
130 <!-- _______________________________________________________________________ -->
132 <!-- _______________________________________________________________________ -->
136 <p>The test suite contains whole programs, which are pieces of
137 code which can be compiled and linked into a stand-alone program that can be
138 executed. These programs are generally written in high level languages such as
139 C or C++, but sometimes they are written straight in LLVM assembly.</p>
141 <p>These programs are compiled and then executed using several different
142 methods (native compiler, LLVM C backend, LLVM JIT, LLVM native code generation,
143 etc). The output of these programs is compared to ensure that LLVM is compiling
144 the program correctly.</p>
146 <p>In addition to compiling and executing programs, whole program tests serve as
147 a way of benchmarking LLVM performance, both in terms of the efficiency of the
148 programs generated as well as the speed with which LLVM compiles, optimizes, and
149 generates code.</p>
153 </div>
155 <!--=========================================================================-->
157 <!--=========================================================================-->
161 <p>The tests are located in two separate Subversion modules. The
162 DejaGNU tests are in the main "llvm" module under the directory
164 The more comprehensive test suite that includes whole
167 then the default "test-suite", for then the test suite will be run every time
173 <!-- _______________________________________________________________________ -->
175 <!-- _______________________________________________________________________ -->
176 <p>To run all of the simple tests in LLVM using DejaGNU, use the master Makefile
180 <pre>
181 % gmake -C llvm/test
182 </pre>
183 </div>
188 <pre>
189 % gmake check
190 </pre>
191 </div>
194 Transforms), just set the TESTSUITE variable to the path of the
198 <pre>
199 % gmake TESTSUITE=Transforms check
200 </pre>
201 </div>
204 must have run the complete testsuite before you can specify a
205 subdirectory.</b></p>
211 <pre>
212 % gmake TESTONE=Feature/basictest.ll check-one
213 </pre>
214 </div>
216 <p>To run the tests with Valgrind (Memcheck by default), just append
220 <pre>
221 % gmake check VG=1
222 </pre>
223 </div>
225 <!-- _______________________________________________________________________ -->
227 <!-- _______________________________________________________________________ -->
229 <p>To run the comprehensive test suite (tests that compile and execute whole
233 <pre>
234 % cd llvm/projects
235 % svn co http://llvm.org/svn/llvm-project/test-suite/trunk test-suite
236 % cd ..
237 % ./configure --with-llvmgccdir=$LLVM_GCC_DIR
238 </pre>
239 </div>
247 respectively. If this is not the case,
249 executable's location.</p>
252 directory:</p>
255 <pre>
256 % cd projects/test-suite
257 % gmake
258 </pre>
259 </div>
262 let it generate a report by running:</p>
265 <pre>
266 % cd projects/test-suite
267 % gmake TEST=nightly report report.html
268 </pre>
269 </div>
271 <p>Any of the above commands can also be run in a subdirectory of
273 that subdirectory.</p>
275 </div>
277 <!--=========================================================================-->
279 <!--=========================================================================-->
281 <p>The LLVM DejaGNU tests are driven by DejaGNU together with GNU Make and are
284 <p>This directory contains a large array of small tests
285 that exercise various features of LLVM and to ensure that regressions do not
286 occur. The directory is broken into several sub-directories, each focused on
287 a particular area of LLVM. A few of the important ones are:</p>
289 <ul>
298 transforms to ensure they make the right transformations.</li>
300 </ul>
302 </div>
304 <!-- _______________________________________________________________________ -->
306 <!-- _______________________________________________________________________ -->
308 <p>The DejaGNU structure is very simple, but does require some information to
311 Makefile does this work for you.</p>
313 <p>In order for DejaGNU to work, each directory of tests must have a
315 tests. This file is just a Tcl script and it can do anything you want, but
316 we've standardized it for the LLVM regression tests. If you're adding a
320 function defined in that library with a list of file names to run. The names
321 are obtained by using Tcl's glob command. Any directory that contains only
326 that specify how the test is to be run. So, each test script must contain
327 RUN lines if it is to do anything. If there are no RUN lines, the
329 fail.</p>
331 <p>RUN lines are specified in the comments of the test program using the
333 to execute. Together, these lines form the "script" that
335 RUN lines is similar to a shell's syntax for pipelines including I/O
336 redirection and variable substitution. However, even though these lines
339 shell. Consequently the syntax differs from normal shell script syntax in a
340 few ways. You can specify as many RUN lines as needed.</p>
342 <p>Each RUN line is executed on its own, distinct from other lines unless
344 line to be concatenated with the next one. In this way you can build up long
345 pipelines of commands without making huge line lengths. The lines ending in
347 found. This concatenated set of RUN lines then constitutes one execution.
348 Tcl will substitute variables and arrange for the pipeline to be executed. If
349 any process in the pipeline fails, the entire line (and test case) fails too.
350 </p>
355 <pre>
358 ; RUN: diff %t1 %t2
359 </pre>
360 </div>
362 <p>As with a Unix shell, the RUN: lines permit pipelines and I/O redirection
363 to be used. However, the usage is slightly different than for Bash. To check
364 what's legal, see the documentation for the
366 command and the
368 The major differences are:</p>
369 <ul>
374 <li>You can only redirect to a file, not to another descriptor and not from
375 a here document.</li>
376 <li>tcl supports redirecting to open files with the @ syntax but you
377 shouldn't use that here.</li>
378 </ul>
380 <p>There are some quoting rules that you must pay attention to when writing
381 your RUN lines. In general nothing needs to be quoted. Tcl won't strip off any
382 ' or " so they will get passed to the invoked program. For example:</p>
385 <pre>
386 ... | grep 'find this string'
387 </pre>
388 </div>
390 <p>This will fail because the ' characters are passed to grep. This would
391 instruction grep to look for <tt>'find</tt> in the files <tt>this</tt> and
392 <tt>string'</tt>. To avoid this use curly braces to tell Tcl that it should
393 treat everything enclosed as one value. So our example would become:</p>
396 <pre>
397 ... | grep {find this string}
398 </pre>
399 </div>
401 <p>Additionally, the characters <tt>[</tt> and <tt>]</tt> are treated
402 specially by Tcl. They tell Tcl to interpret the content as a command to
403 execute. Since these characters are often used in regular expressions this can
404 have disastrous results and cause the entire test run in a directory to fail.
405 For example, a common idiom is to look for some basicblock number:</p>
408 <pre>
409 ... | grep bb[2-8]
410 </pre>
411 </div>
413 <p>This, however, will cause Tcl to fail because its going to try to execute
417 <pre>
418 ... | grep {bb\[2-8\]}
419 </pre>
420 </div>
422 <p>Finally, if you need to pass the <tt>\</tt> character down to a program,
423 then it must be doubled. This is another Tcl special character. So, suppose
424 you had:
427 <pre>
428 ... | grep 'i32\*'
429 </pre>
430 </div>
432 <p>This will fail to match what you want (a pointer to i32). First, the
433 <tt>'</tt> do not get stripped off. Second, the <tt>\</tt> gets stripped off
434 by Tcl so what grep sees is: <tt>'i32*'</tt>. That's not likely to match
435 anything. To resolve this you must use <tt>\\</tt> and the <tt>{}</tt>, like
436 this:</p>
439 <pre>
440 ... | grep {i32\\*}
441 </pre>
442 </div>
444 <p>If your system includes GNU <tt>grep</tt>, make sure
445 that <tt>GREP_OPTIONS</tt> is not set in your environment. Otherwise,
446 you may get invalid results (both false positives and false
447 negatives).</p>
449 </div>
451 <!-- _______________________________________________________________________ -->
453 <!-- _______________________________________________________________________ -->
457 <p>A powerful feature of the RUN: lines is that it allows any arbitrary commands
458 to be executed as part of the test harness. While standard (portable) unix
459 tools like 'grep' work fine on run lines, as you see above, there are a lot
460 of caveats due to interaction with Tcl syntax, and we want to make sure the
461 run lines are portable to a wide range of systems. Another major problem is
462 that grep is not very good at checking to verify that the output of a tools
463 contains a series of different output in a specific order. The FileCheck
464 tool was designed to help with these problems.</p>
466 <p>FileCheck (whose basic command line arguments are described in <a
468 designed to read a file to check from standard input, and the set of things
469 to verify from a file specified as a command line argument. A simple example
470 of using FileCheck from a RUN line looks like this:</p>
473 <pre>
474 ; RUN: llvm-as < %s | llc -march=x86-64 | <b>FileCheck %s</b>
475 </pre>
476 </div>
479 llc, then pipe the output of llc into FileCheck. This means that FileCheck will
480 be verifying its standard input (the llc output) against the filename argument
482 lets look at the rest of the .ll file (after the RUN line):</p>
485 <pre>
486 define void @sub1(i32* %p, i32 %v) {
487 entry:
488 ; <b>CHECK: sub1:</b>
489 ; <b>CHECK: subl</b>
490 %0 = tail call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %p, i32 %v)
491 ret void
492 }
494 define void @inc4(i64* %p) {
495 entry:
496 ; <b>CHECK: inc4:</b>
497 ; <b>CHECK: incq</b>
498 %0 = tail call i64 @llvm.atomic.load.add.i64.p0i64(i64* %p, i64 1)
499 ret void
500 }
501 </pre>
502 </div>
505 how the file is piped into llvm-as, then llc, and the machine code output is
506 what we are verifying. FileCheck checks the machine code output to verify that
509 <p>The syntax of the CHECK: lines is very simple: they are fixed strings that
510 must occur in order. FileCheck defaults to ignoring horizontal whitespace
511 differences (e.g. a space is allowed to match a tab) but otherwise, the contents
512 of the CHECK: line is required to match some thing in the test file exactly.</p>
514 <p>One nice thing about FileCheck (compared to grep) is that it allows merging
515 test cases together into logical groups. For example, because the test above
519 file.</p>
521 </div>
523 <!-- _______________________________________________________________________ -->
529 <p>The FileCheck -check-prefix option allows multiple test configurations to be
530 driven from one .ll file. This is useful in many circumstances, for example,
531 testing different architectural variants with llc. Here's a simple example:</p>
534 <pre>
535 ; RUN: llvm-as < %s | llc -mtriple=i686-apple-darwin9 -mattr=sse41 \
536 ; RUN: | <b>FileCheck %s -check-prefix=X32</b>
537 ; RUN: llvm-as < %s | llc -mtriple=x86_64-apple-darwin9 -mattr=sse41 \
538 ; RUN: | <b>FileCheck %s -check-prefix=X64</b>
540 define <4 x i32> @pinsrd_1(i32 %s, <4 x i32> %tmp) nounwind {
541 %tmp1 = insertelement <4 x i32> %tmp, i32 %s, i32 1
542 ret <4 x i32> %tmp1
543 ; <b>X32:</b> pinsrd_1:
544 ; <b>X32:</b> pinsrd $1, 4(%esp), %xmm0
546 ; <b>X64:</b> pinsrd_1:
547 ; <b>X64:</b> pinsrd $1, %edi, %xmm0
548 }
549 </pre>
550 </div>
552 <p>In this case, we're testing that we get the expected code generation with
553 both 32-bit and 64-bit code generation.</p>
555 </div>
557 <!-- _______________________________________________________________________ -->
563 <p>Sometimes you want to match lines and would like to verify that matches
564 happen on exactly consequtive lines with no other lines in between them. In
565 this case, you can use CHECK: and CHECK-NEXT: directives to specify this. If
567 example, something like this works as you'd expect:</p>
570 <pre>
571 define void @t2(<2 x double>* %r, <2 x double>* %A, double %B) {
572 %tmp3 = load <2 x double>* %A, align 16
573 %tmp7 = insertelement <2 x double> undef, double %B, i32 0
574 %tmp9 = shufflevector <2 x double> %tmp3,
575 <2 x double> %tmp7,
576 <2 x i32> < i32 0, i32 2 >
577 store <2 x double> %tmp9, <2 x double>* %r, align 16
578 ret void
580 ; <b>CHECK:</b> t2:
581 ; <b>CHECK:</b> movl 8(%esp), %eax
582 ; <b>CHECK-NEXT:</b> movapd (%eax), %xmm0
583 ; <b>CHECK-NEXT:</b> movhpd 12(%esp), %xmm0
584 ; <b>CHECK-NEXT:</b> movl 4(%esp), %eax
585 ; <b>CHECK-NEXT:</b> movapd %xmm0, (%eax)
586 ; <b>CHECK-NEXT:</b> ret
587 }
588 </pre>
589 </div>
591 <p>CHECK-NEXT: directives reject the input unless there is exactly one newline
592 between it an the previous directive. A CHECK-NEXT cannot be the first
593 directive in a file.</p>
595 </div>
597 <!-- _______________________________________________________________________ -->
603 <p>The CHECK-NOT: directive is used to verify that a string doesn't occur
604 between two matches (or the first match and the beginning of the file). For
605 example, to verify that a load is removed by a transformation, a test like this
606 can be used:</p>
609 <pre>
610 define i8 @coerce_offset0(i32 %V, i32* %P) {
611 store i32 %V, i32* %P
613 %P2 = bitcast i32* %P to i8*
614 %P3 = getelementptr i8* %P2, i32 2
616 %A = load i8* %P3
617 ret i8 %A
618 ; <b>CHECK:</b> @coerce_offset0
619 ; <b>CHECK-NOT:</b> load
620 ; <b>CHECK:</b> ret i8
621 }
622 </pre>
623 </div>
625 </div>
627 <!-- _______________________________________________________________________ -->
633 <p>The CHECK: and CHECK-NOT: directives both take a pattern to match. For most
634 uses of FileCheck, fixed string matching is perfectly sufficient. For some
635 things, a more flexible form of matching is desired. To support this, FileCheck
636 allows you to specify regular expressions in matching strings, surrounded by
637 double braces: <b>{{yourregex}}</b>. Because we want to use fixed string
638 matching for a majority of what we do, FileCheck has been designed to support
639 mixing and matching fixed string matching with regular expressions. This allows
640 you to write things like this:</p>
643 <pre>
644 ; CHECK: movhpd <b>{{[0-9]+}}</b>(%esp), <b>{{%xmm[0-7]}}</b>
645 </pre>
646 </div>
648 <p>In this case, any offset from the ESP register will be allowed, and any xmm
649 register will be allowed.</p>
651 <p>Because regular expressions are enclosed with double braces, they are
652 visually distinct, and you don't need to use escape characters within the double
653 braces like you would in C. In the rare case that you want to match double
654 braces explicitly from the input, you can use something ugly like
655 <b>{{[{][{]}}</b> as your pattern.</p>
657 </div>
659 <!-- _______________________________________________________________________ -->
665 <p>It is often useful to match a pattern and then verify that it occurs again
666 later in the file. For codegen tests, this can be useful to allow any register,
667 but verify that that register is used consistently later. To do this, FileCheck
668 allows named variables to be defined and substituted into patterns. Here is a
669 simple example:</p>
672 <pre>
673 ; CHECK: test5:
674 ; CHECK: notw <b>[[REGISTER:%[a-z]+]]</b>
675 ; CHECK: andw {{.*}}<b>[[REGISTER]]</b>
676 </pre>
677 </div>
679 <p>The first check line matches a regex (<tt>%[a-z]+</tt>) and captures it into
682 always contained in <tt>[[ ]]</tt> pairs, are named, and their names can be
684 name, then it is a definition of the variable, if not, it is a use.</p>
686 <p>FileCheck variables can be defined multiple times, and uses always get the
688 and are all defined at the end. This means that if you have something like
690 value of the XYZ variable and define a new one after the match is performed. If
691 you need to do something like this you can probably take advantage of the fact
692 that FileCheck is not actually line-oriented when it matches, this allows you to
693 define two separate CHECK lines that match on the same line.
694 </p>
698 </div>
700 <!-- _______________________________________________________________________ -->
702 substitutions</a></div>
703 <!-- _______________________________________________________________________ -->
705 <p>With a RUN line there are a number of substitutions that are permitted. In
706 general, any Tcl variable that is available in the <tt>substitute</tt>
707 function (in <tt>test/lib/llvm.exp</tt>) can be substituted into a RUN line.
708 To make a substitution just write the variable's name preceded by a $.
709 Additionally, for compatibility reasons with previous versions of the test
710 library, certain names can be accessed with an alternate syntax: a % prefix.
711 These alternates are deprecated and may go away in a future version.
712 </p>
713 <p>Here are the available variable names. The alternate syntax is listed in
714 parentheses.</p>
717 <dt><b>$test</b> (%s)</dt>
718 <dd>The full path to the test case's source. This is suitable for passing
719 on the command line as the input to an llvm tool.</dd>
721 <dt><b>$srcdir</b></dt>
724 <dt><b>objdir</b></dt>
725 <dd>The object directory that corresponds to the <tt>$srcdir</tt>.</dd>
727 <dt><b>subdir</b></dt>
728 <dd>A partial path from the <tt>test</tt> directory that contains the
729 sub-directory that contains the test source being executed.</dd>
731 <dt><b>srcroot</b></dt>
732 <dd>The root directory of the LLVM src tree.</dd>
734 <dt><b>objroot</b></dt>
735 <dd>The root directory of the LLVM object tree. This could be the same
736 as the srcroot.</dd>
738 <dt><b>path</b><dt>
739 <dd>The path to the directory that contains the test case source. This is
740 for locating any supporting files that are not generated by the test, but
741 used by the test.</dd>
743 <dt><b>tmp</b></dt>
744 <dd>The path to a temporary file name that could be used for this test case.
745 The file name won't conflict with other test cases. You can append to it if
746 you need multiple temporaries. This is useful as the destination of some
747 redirected output.</dd>
749 <dt><b>llvmlibsdir</b> (%llvmlibsdir)</dt>
750 <dd>The directory where the LLVM libraries are located.</dd>
752 <dt><b>target_triplet</b> (%target_triplet)</dt>
753 <dd>The target triplet that corresponds to the current host machine (the one
756 <dt><b>llvmgcc</b> (%llvmgcc)</dt>
757 <dd>The full path to the <tt>llvm-gcc</tt> executable as specified in the
758 configured LLVM environment</dd>
760 <dt><b>llvmgxx</b> (%llvmgxx)</dt>
761 <dd>The full path to the <tt>llvm-gxx</tt> executable as specified in the
762 configured LLVM environment</dd>
764 <dt><b>llvmgcc_version</b> (%llvmgcc_version)</dt>
765 <dd>The full version number of the <tt>llvm-gcc</tt> executable.</dd>
767 <dt><b>llvmgccmajvers</b> (%llvmgccmajvers)</dt>
768 <dd>The major version number of the <tt>llvm-gcc</tt> executable.</dd>
770 <dt><b>gccpath</b></dt>
771 <dd>The full path to the C compiler used to <i>build </i> LLVM. Note that
772 this might not be gcc.</dd>
774 <dt><b>gxxpath</b></dt>
775 <dd>The full path to the C++ compiler used to <i>build </i> LLVM. Note that
776 this might not be g++.</dd>
778 <dt><b>compile_c</b> (%compile_c)</dt>
779 <dd>The full command line used to compile LLVM C source code. This has all
780 the configured -I, -D and optimization options.</dd>
782 <dt><b>compile_cxx</b> (%compile_cxx)</dt>
783 <dd>The full command used to compile LLVM C++ source code. This has
784 all the configured -I, -D and optimization options.</dd>
786 <dt><b>link</b> (%link)</dt>
787 <dd>This full link command used to link LLVM executables. This has all the
788 configured -I, -L and -l options.</dd>
790 <dt><b>shlibext</b> (%shlibext)</dt>
791 <dd>The suffix for the host platforms share library (dll) files. This
792 includes the period as the first character.</dd>
793 </dl>
794 <p>To add more variables, two things need to be changed. First, add a line in
795 the <tt>test/Makefile</tt> that creates the <tt>site.exp</tt> file. This will
797 <tt>test/lib/llvm.exp</tt> file, in the substitute proc, add the variable name
799 the variable can then be used in test scripts.</p>
800 </div>
802 <!-- _______________________________________________________________________ -->
804 <!-- _______________________________________________________________________ -->
806 <p>To make RUN line writing easier, there are several shell scripts located
807 in the <tt>llvm/test/Scripts</tt> directory. This directory is in the PATH
808 when running tests, so you can just call these scripts using their name. For
809 example:</p>
810 <dl>
811 <dt><b>ignore</b></dt>
812 <dd>This script runs its arguments and then always returns 0. This is useful
813 in cases where the test needs to cause a tool to generate an error (e.g. to
814 check the error output). However, any program in a pipeline that returns a
815 non-zero result will cause the test to fail. This script overcomes that
816 issue and nicely documents that the test case is purposefully ignoring the
817 result code of the tool</dd>
819 <dt><b>not</b></dt>
820 <dd>This script runs its arguments and then inverts the result code from
821 it. Zero result codes become 1. Non-zero result codes become 0. This is
823 succeed only if you don't find X in the input.</dd>
824 </dl>
827 You can easily mark a test as XFAIL just by including <tt>XFAIL: </tt> on a
828 line near the top of the file. This signals that the test case should succeed
829 if the test fails. Such test cases are counted separately by DejaGnu. To
830 specify an expected fail, use the XFAIL keyword in the comments of the test
831 program followed by a colon and one or more regular expressions (separated by
832 a comma). The regular expressions allow you to XFAIL the test conditionally
833 by host platform. The regular expressions following the : are matched against
834 the target triplet or llvmgcc version number for the host machine. If there is
835 a match, the test is expected to fail. If not, the test is expected to
836 succeed. To XFAIL everywhere just specify <tt>XFAIL: *</tt>. When matching
837 the llvm-gcc version, you can specify the major (e.g. 3) or full version
838 (i.e. 3.4) number. Here is an example of an <tt>XFAIL</tt> line:</p>
841 <pre>
842 ; XFAIL: darwin,sun,llvmgcc4
843 </pre>
844 </div>
846 <p>To make the output more useful, the <tt>llvm_runtest</tt> function wil
847 scan the lines of the test case for ones that contain a pattern that matches
848 PR[0-9]+. This is the syntax for specifying a PR (Problem Report) number that
850 number. When a PR number is specified, it will be used in the pass/fail
851 reporting. This is useful to quickly get some context when a test fails.</p>
854 interpretation of lines to terminate. This is generally done right after the
855 last RUN: line. This has two side effects: (a) it prevents special
856 interpretation of lines that are part of the test program, not the
857 instructions to the test case, and (b) it speeds things up for really big test
858 cases by avoiding interpretation of the remainder of the file.</p>
860 </div>
862 <!--=========================================================================-->
864 Structure</a></div>
865 <!--=========================================================================-->
869 <p>The <tt>test-suite</tt> module contains a number of programs that can be compiled
870 with LLVM and executed. These programs are compiled using the native compiler
871 and various LLVM backends. The output from the program compiled with the
872 native compiler is assumed correct; the results from the other programs are
873 compared to the native program output and pass if they match.</p>
875 <p>When executing tests, it is usually a good idea to start out with a subset of
876 the available tests or programs. This makes test run times smaller at first and
877 later on this is useful to investigate individual test failures. To run some
878 test only on a subset of programs, simply change directory to the programs you
879 want tested and run <tt>gmake</tt> there. Alternatively, you can run a different
880 test using the <tt>TEST</tt> variable to change what tests or run on the
881 selected programs (see below for more info).</p>
883 <p>In addition for testing correctness, the <tt>llvm-test</tt> directory also
884 performs timing tests of various LLVM optimizations. It also records
885 compilation times for the compilers and the JIT. This information can be
886 used to compare the effectiveness of LLVM's optimizations and code
887 generation.</p>
889 <p><tt>llvm-test</tt> tests are divided into three types of tests: MultiSource,
890 SingleSource, and External.</p>
892 <ul>
893 <li><tt>llvm-test/SingleSource</tt>
894 <p>The SingleSource directory contains test programs that are only a single
895 source file in size. These are usually small benchmark programs or small
896 programs that calculate a particular value. Several such programs are grouped
897 together in each directory.</p></li>
899 <li><tt>llvm-test/MultiSource</tt>
900 <p>The MultiSource directory contains subdirectories which contain entire
901 programs with multiple source files. Large benchmarks and whole applications
902 go here.</p></li>
904 <li><tt>llvm-test/External</tt>
905 <p>The External directory contains Makefiles for building code that is external
906 to (i.e., not distributed with) LLVM. The most prominent members of this
907 directory are the SPEC 95 and SPEC 2000 benchmark suites. The <tt>External</tt>
908 directory does not contain these actual tests, but only the Makefiles that know
909 how to properly compile these programs from somewhere else. The presence and
910 location of these external programs is configured by the llvm-test
911 <tt>configure</tt> script.</p></li>
912 </ul>
914 <p>Each tree is then subdivided into several categories, including applications,
915 benchmarks, regression tests, code that is strange grammatically, etc. These
916 organizations should be relatively self explanatory.</p>
918 <p>Some tests are known to fail. Some are bugs that we have not fixed yet;
919 others are features that we haven't added yet (or may never add). In DejaGNU,
920 the result for such tests will be XFAIL (eXpected FAILure). In this way, you
921 can tell the difference between an expected and unexpected failure.</p>
923 <p>The tests in the test suite have no such feature at this time. If the
924 test passes, only warnings and other miscellaneous output will be generated. If
925 a test fails, a large <program> FAILED message will be displayed. This
926 will help you separate benign warnings from actual test failures.</p>
928 </div>
930 <!--=========================================================================-->
932 <!--=========================================================================-->
936 <p>First, all tests are executed within the LLVM object directory tree. They
937 <i>are not</i> executed inside of the LLVM source tree. This is because the
938 test suite creates temporary files during execution.</p>
940 <p>To run the test suite, you need to use the following steps:</p>
942 <ol>
943 <li><tt>cd</tt> into the <tt>llvm/projects</tt> directory in your source tree.
944 </li>
946 <li><p>Check out the <tt>test-suite</tt> module with:</p>
949 <pre>
950 % svn co http://llvm.org/svn/llvm-project/test-suite/trunk test-suite
951 </pre>
952 </div>
953 <p>This will get the test suite into <tt>llvm/projects/test-suite</tt>.</p>
954 </li>
955 <li><p>Configure and build <tt>llvm</tt>.</p></li>
956 <li><p>Configure and build <tt>llvm-gcc</tt>.</p></li>
957 <li><p>Install <tt>llvm-gcc</tt> somewhere.</p></li>
958 <li><p><em>Re-configure</em> <tt>llvm</tt> from the top level of
959 each build tree (LLVM object directory tree) in which you want
960 to run the test suite, just as you do before building LLVM.</p>
961 <p>During the <em>re-configuration</em>, you must either: (1)
962 have <tt>llvm-gcc</tt> you just built in your path, or (2)
963 specify the directory where your just-built <tt>llvm-gcc</tt> is
964 installed using <tt>--with-llvmgccdir=$LLVM_GCC_DIR</tt>.</p>
965 <p>You must also tell the configure machinery that the test suite
966 is available so it can be configured for your build tree:</p>
968 <pre>
969 % cd $LLVM_OBJ_ROOT ; $LLVM_SRC_ROOT/configure [--with-llvmgccdir=$LLVM_GCC_DIR]
970 </pre>
971 </div>
972 <p>[Remember that <tt>$LLVM_GCC_DIR</tt> is the directory where you
973 <em>installed</em> llvm-gcc, not its src or obj directory.]</p>
974 </li>
976 <li><p>You can now run the test suite from your build tree as follows:</p>
978 <pre>
979 % cd $LLVM_OBJ_ROOT/projects/test-suite
980 % make
981 </pre>
982 </div>
983 </li>
984 </ol>
985 <p>Note that the second and third steps only need to be done once. After you
986 have the suite checked out and configured, you don't need to do it again (unless
987 the test code or configure script changes).</p>
989 </div>
991 <!-- _______________________________________________________________________ -->
994 <!-- _______________________________________________________________________ -->
997 <p>In order to run the External tests in the <tt>test-suite</tt>
998 module, you must specify <i>--with-externals</i>. This
999 must be done during the <em>re-configuration</em> step (see above),
1000 and the <tt>llvm</tt> re-configuration must recognize the
1001 previously-built <tt>llvm-gcc</tt>. If any of these is missing or
1002 neglected, the External tests won't work.</p>
1003 <dl>
1004 <dt><i>--with-externals</i></dt>
1005 <dt><i>--with-externals=<<tt>directory</tt>></i></dt>
1006 </dl>
1007 This tells LLVM where to find any external tests. They are expected to be
1008 in specifically named subdirectories of <<tt>directory</tt>>.
1009 If <tt>directory</tt> is left unspecified,
1010 <tt>configure</tt> uses the default value
1011 <tt>/home/vadve/shared/benchmarks/speccpu2000/benchspec</tt>.
1012 Subdirectory names known to LLVM include:
1013 <dl>
1014 <dt>spec95</dt>
1015 <dt>speccpu2000</dt>
1016 <dt>speccpu2006</dt>
1017 <dt>povray31</dt>
1018 </dl>
1019 Others are added from time to time, and can be determined from
1020 <tt>configure</tt>.
1021 </div>
1023 <!-- _______________________________________________________________________ -->
1026 <!-- _______________________________________________________________________ -->
1029 module also provides a mechanism for compiling the programs in different ways.
1030 If the variable TEST is defined on the <tt>gmake</tt> command line, the test system will
1031 include a Makefile named <tt>TEST.<value of TEST variable>.Makefile</tt>.
1032 This Makefile can modify build rules to yield different results.</p>
1034 <p>For example, the LLVM nightly tester uses <tt>TEST.nightly.Makefile</tt> to
1035 create the nightly test reports. To run the nightly tests, run <tt>gmake
1036 TEST=nightly</tt>.</p>
1038 <p>There are several TEST Makefiles available in the tree. Some of them are
1039 designed for internal LLVM research and will not work outside of the LLVM
1040 research group. They may still be valuable, however, as a guide to writing your
1041 own TEST Makefile for any optimization or analysis passes that you develop with
1042 LLVM.</p>
1044 </div>
1046 <!-- _______________________________________________________________________ -->
1049 <!-- _______________________________________________________________________ -->
1051 <p>There are a number of ways to run the tests and generate output. The most
1052 simple one is simply running <tt>gmake</tt> with no arguments. This will
1053 compile and run all programs in the tree using a number of different methods
1054 and compare results. Any failures are reported in the output, but are likely
1055 drowned in the other output. Passes are not reported explicitely.</p>
1057 <p>Somewhat better is running <tt>gmake TEST=sometest test</tt>, which runs
1058 the specified test and usually adds per-program summaries to the output
1059 (depending on which sometest you use). For example, the <tt>nightly</tt> test
1060 explicitely outputs TEST-PASS or TEST-FAIL for every test after each program.
1061 Though these lines are still drowned in the output, it's easy to grep the
1062 output logs in the Output directories.</p>
1064 <p>Even better are the <tt>report</tt> and <tt>report.format</tt> targets
1065 (where <tt>format</tt> is one of <tt>html</tt>, <tt>csv</tt>, <tt>text</tt> or
1066 <tt>graphs</tt>). The exact contents of the report are dependent on which
1067 <tt>TEST</tt> you are running, but the text results are always shown at the
1068 end of the run and the results are always stored in the
1069 <tt>report.<type>.format</tt> file (when running with
1070 <tt>TEST=<type></tt>).
1072 The <tt>report</tt> also generate a file called
1073 <tt>report.<type>.raw.out</tt> containing the output of the entire test
1074 run.
1075 </div>
1077 <!-- _______________________________________________________________________ -->
1080 <!-- _______________________________________________________________________ -->
1085 should work), it is really easy to run optimizations or code generator
1086 components against every program in the tree, collecting statistics or running
1087 custom checks for correctness. At base, this is how the nightly tester works,
1088 it's just one example of a general framework.</p>
1090 <p>Lets say that you have an LLVM optimization pass, and you want to see how
1091 many times it triggers. First thing you should do is add an LLVM
1093 will tally counts of things you care about.</p>
1095 <p>Following this, you can set up a test and a report that collects these and
1096 formats them for easy viewing. This consists of two files, an
1099 format the output into a table. There are many example reports of various
1100 levels of sophistication included with the test suite, and the framework is very
1101 general.</p>
1103 <p>If you are interested in testing an optimization pass, check out the
1107 <pre>
1108 % cd llvm/projects/test-suite/MultiSource/Benchmarks # or some other level
1109 % make TEST=libcalls report
1110 </pre>
1111 </div>
1113 <p>This will do a bunch of stuff, then eventually print a table like this:</p>
1116 <pre>
1117 Name | total | #exit |
1118 ...
1119 FreeBench/analyzer/analyzer | 51 | 6 |
1120 FreeBench/fourinarow/fourinarow | 1 | 1 |
1121 FreeBench/neural/neural | 19 | 9 |
1122 FreeBench/pifft/pifft | 5 | 3 |
1123 MallocBench/cfrac/cfrac | 1 | * |
1124 MallocBench/espresso/espresso | 52 | 12 |
1125 MallocBench/gs/gs | 4 | * |
1126 Prolangs-C/TimberWolfMC/timberwolfmc | 302 | * |
1127 Prolangs-C/agrep/agrep | 33 | 12 |
1128 Prolangs-C/allroots/allroots | * | * |
1129 Prolangs-C/assembler/assembler | 47 | * |
1130 Prolangs-C/bison/mybison | 74 | * |
1131 ...
1132 </pre>
1133 </div>
1135 <p>This basically is grepping the -stats output and displaying it in a table.
1137 form, similarly for report.csv and report.tex.</p>
1139 <p>The source for this is in test-suite/TEST.libcalls.*. The format is pretty
1140 simple: the Makefile indicates how to run the test (in this case,
1142 each column of the output. The first value is the header for the column and the
1143 second is the regex to grep the output of the command for. There are lots of
1144 example reports that can do fancy stuff.</p>
1146 </div>
1149 <!--=========================================================================-->
1151 <!--=========================================================================-->
1155 <p>
1158 program test (described above), run all of the DejaGNU tests,
1159 delete the checked out tree, and then submit the results to
1161 After test results are submitted to
1163 they are processed and displayed on the tests page. An email to
1165 llvm-testresults@cs.uiuc.edu</a> summarizing the results is also generated.
1166 This testing scheme is designed to ensure that programs don't break as well
1167 as keep track of LLVM's progress over time.</p>
1169 <p>If you'd like to set up an instance of the nightly tester to run on your
1170 machine, take a look at the comments at the top of the
1171 <tt>utils/NewNightlyTest.pl</tt> file. If you decide to set up a nightly tester
1172 please choose a unique nickname and invoke <tt>utils/NewNightlyTest.pl</tt>
1175 <p>You can create a shell script to encapsulate the running of the script.
1176 The optimized x86 Linux nightly test is run from just such a script:</p>
1179 <pre>
1180 #!/bin/bash
1181 BASE=/proj/work/llvm/nightlytest
1182 export BUILDDIR=$BASE/build
1183 export WEBDIR=$BASE/testresults
1184 export LLVMGCCDIR=/proj/work/llvm/cfrontend/install
1185 export PATH=/proj/install/bin:$LLVMGCCDIR/bin:$PATH
1186 export LD_LIBRARY_PATH=/proj/install/lib
1187 cd $BASE
1188 cp /proj/work/llvm/llvm/utils/NewNightlyTest.pl .
1189 nice ./NewNightlyTest.pl -nice -release -verbose -parallel -enable-linscan \
1190 -nickname NightlyTester -noexternals > output.log 2>&1
1191 </pre>
1192 </div>
1194 <p>It is also possible to specify the the location your nightly test results
1195 are submitted. You can do this by passing the command line option
1197 <tt>utils/NewNightlyTest.pl</tt>. For example, to submit to the llvm.org
1198 nightly test results page, you would invoke the nightly test script with
1200 If these options are not specified, the nightly test script sends the results
1201 to the llvm.org nightly test results page.</p>
1203 <p>Take a look at the <tt>NewNightlyTest.pl</tt> file to see what all of the
1204 flags and strings do. If you start running the nightly tests, please let us
1205 know. Thanks!</p>
1207 </div>
1209 <!-- *********************************************************************** -->
1211 <hr>
1212 <address>
1218 John T. Criswell, Reid Spencer, and Tanya Lattner<br>
1220 Last modified: $Date$
1221 </address>
1222 </body>
1223 </html>