xref: /llvm-project/llvm/docs/CommandGuide/FileCheck.rst (revision 95db1e7fb9292daf70fedf02b5efafeabf3ea33e)
1FileCheck - Flexible pattern matching file verifier
2===================================================
3
4.. program:: FileCheck
5
6SYNOPSIS
7--------
8
9:program:`FileCheck` *match-filename* [*--check-prefix=XXX*] [*--strict-whitespace*]
10
11DESCRIPTION
12-----------
13
14:program:`FileCheck` reads two files (one from standard input, and one
15specified on the command line) and uses one to verify the other.  This
16behavior is particularly useful for the testsuite, which wants to verify that
17the output of some tool (e.g. :program:`llc`) contains the expected information
18(for example, a movsd from esp or whatever is interesting).  This is similar to
19using :program:`grep`, but it is optimized for matching multiple different
20inputs in one file in a specific order.
21
22The ``match-filename`` file specifies the file that contains the patterns to
23match.  The file to verify is read from standard input unless the
24:option:`--input-file` option is used.
25
26OPTIONS
27-------
28
29Options are parsed from the environment variable ``FILECHECK_OPTS``
30and from the command line.
31
32.. option:: -help
33
34 Print a summary of command line options.
35
36.. option:: --check-prefix prefix
37
38 FileCheck searches the contents of ``match-filename`` for patterns to
39 match.  By default, these patterns are prefixed with "``CHECK:``".
40 If you'd like to use a different prefix (e.g. because the same input
41 file is checking multiple different tool or options), the
42 :option:`--check-prefix` argument allows you to specify (without the trailing
43 "``:``") one or more prefixes to match. Multiple prefixes are useful for tests
44 which might change for different run options, but most lines remain the same.
45
46 FileCheck does not permit duplicate prefixes, even if one is a check prefix
47 and one is a comment prefix (see :option:`--comment-prefixes` below).
48
49.. option:: --check-prefixes prefix1,prefix2,...
50
51 An alias of :option:`--check-prefix` that allows multiple prefixes to be
52 specified as a comma separated list.
53
54.. option:: --comment-prefixes prefix1,prefix2,...
55
56 By default, FileCheck ignores any occurrence in ``match-filename`` of any check
57 prefix if it is preceded on the same line by "``COM:``" or "``RUN:``". See the
58 section `The "COM:" directive`_ for usage details.
59
60 These default comment prefixes can be overridden by
61 :option:`--comment-prefixes` if they are not appropriate for your testing
62 environment. However, doing so is not recommended in LLVM's LIT-based test
63 suites, which should be easier to maintain if they all follow a consistent
64 comment style. In that case, consider proposing a change to the default
65 comment prefixes instead.
66
67.. option:: --input-file filename
68
69  File to check (defaults to stdin).
70
71.. option:: --match-full-lines
72
73 By default, FileCheck allows matches of anywhere on a line. This
74 option will require all positive matches to cover an entire
75 line. Leading and trailing whitespace is ignored, unless
76 :option:`--strict-whitespace` is also specified. (Note: negative
77 matches from ``CHECK-NOT`` are not affected by this option!)
78
79 Passing this option is equivalent to inserting ``{{^ *}}`` or
80 ``{{^}}`` before, and ``{{ *$}}`` or ``{{$}}`` after every positive
81 check pattern.
82
83.. option:: --strict-whitespace
84
85 By default, FileCheck canonicalizes input horizontal whitespace (spaces and
86 tabs) which causes it to ignore these differences (a space will match a tab).
87 The :option:`--strict-whitespace` argument disables this behavior. End-of-line
88 sequences are canonicalized to UNIX-style ``\n`` in all modes.
89
90.. option:: --ignore-case
91
92  By default, FileCheck uses case-sensitive matching. This option causes
93  FileCheck to use case-insensitive matching.
94
95.. option:: --implicit-check-not check-pattern
96
97  Adds implicit negative checks for the specified patterns between positive
98  checks. The option allows writing stricter tests without stuffing them with
99  ``CHECK-NOT``\ s.
100
101  For example, "``--implicit-check-not warning:``" can be useful when testing
102  diagnostic messages from tools that don't have an option similar to ``clang
103  -verify``. With this option FileCheck will verify that input does not contain
104  warnings not covered by any ``CHECK:`` patterns.
105
106.. option:: --dump-input <mode>
107
108  Dump input to stderr, adding annotations representing currently enabled
109  diagnostics.  Do this either 'always', on 'fail' (default), or 'never'.
110  Specify 'help' to explain the dump format and quit.
111
112.. option:: --enable-var-scope
113
114  Enables scope for regex variables.
115
116  Variables with names that start with ``$`` are considered global and
117  remain set throughout the file.
118
119  All other variables get undefined after each encountered ``CHECK-LABEL``.
120
121.. option:: -D<VAR=VALUE>
122
123  Sets a filecheck pattern variable ``VAR`` with value ``VALUE`` that can be
124  used in ``CHECK:`` lines.
125
126.. option:: -D#<FMT>,<NUMVAR>=<NUMERIC EXPRESSION>
127
128  Sets a filecheck numeric variable ``NUMVAR`` of matching format ``FMT`` to
129  the result of evaluating ``<NUMERIC EXPRESSION>`` that can be used in
130  ``CHECK:`` lines.  See section
131  ``FileCheck Numeric Variables and Expressions`` for details on supported
132  numeric expressions.
133
134.. option:: -version
135
136 Show the version number of this program.
137
138.. option:: -v
139
140  Print good directive pattern matches.  However, if ``-input-dump=fail`` or
141  ``-input-dump=always``, add those matches as input annotations instead.
142
143.. option:: -vv
144
145  Print information helpful in diagnosing internal FileCheck issues, such as
146  discarded overlapping ``CHECK-DAG:`` matches, implicit EOF pattern matches,
147  and ``CHECK-NOT:`` patterns that do not have matches.  Implies ``-v``.
148  However, if ``-input-dump=fail`` or ``-input-dump=always``, just add that
149  information as input annotations instead.
150
151.. option:: --allow-deprecated-dag-overlap
152
153  Enable overlapping among matches in a group of consecutive ``CHECK-DAG:``
154  directives.  This option is deprecated and is only provided for convenience
155  as old tests are migrated to the new non-overlapping ``CHECK-DAG:``
156  implementation.
157
158.. option:: --color
159
160  Use colors in output (autodetected by default).
161
162EXIT STATUS
163-----------
164
165If :program:`FileCheck` verifies that the file matches the expected contents,
166it exits with 0.  Otherwise, if not, or if an error occurs, it will exit with a
167non-zero value.
168
169TUTORIAL
170--------
171
172FileCheck is typically used from LLVM regression tests, being invoked on the RUN
173line of the test.  A simple example of using FileCheck from a RUN line looks
174like this:
175
176.. code-block:: llvm
177
178   ; RUN: llvm-as < %s | llc -march=x86-64 | FileCheck %s
179
180This syntax says to pipe the current file ("``%s``") into ``llvm-as``, pipe
181that into ``llc``, then pipe the output of ``llc`` into ``FileCheck``.  This
182means that FileCheck will be verifying its standard input (the llc output)
183against the filename argument specified (the original ``.ll`` file specified by
184"``%s``").  To see how this works, let's look at the rest of the ``.ll`` file
185(after the RUN line):
186
187.. code-block:: llvm
188
189   define void @sub1(i32* %p, i32 %v) {
190   entry:
191   ; CHECK: sub1:
192   ; CHECK: subl
193           %0 = tail call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %p, i32 %v)
194           ret void
195   }
196
197   define void @inc4(i64* %p) {
198   entry:
199   ; CHECK: inc4:
200   ; CHECK: incq
201           %0 = tail call i64 @llvm.atomic.load.add.i64.p0i64(i64* %p, i64 1)
202           ret void
203   }
204
205Here you can see some "``CHECK:``" lines specified in comments.  Now you can
206see how the file is piped into ``llvm-as``, then ``llc``, and the machine code
207output is what we are verifying.  FileCheck checks the machine code output to
208verify that it matches what the "``CHECK:``" lines specify.
209
210The syntax of the "``CHECK:``" lines is very simple: they are fixed strings that
211must occur in order.  FileCheck defaults to ignoring horizontal whitespace
212differences (e.g. a space is allowed to match a tab) but otherwise, the contents
213of the "``CHECK:``" line is required to match some thing in the test file exactly.
214
215One nice thing about FileCheck (compared to grep) is that it allows merging
216test cases together into logical groups.  For example, because the test above
217is checking for the "``sub1:``" and "``inc4:``" labels, it will not match
218unless there is a "``subl``" in between those labels.  If it existed somewhere
219else in the file, that would not count: "``grep subl``" matches if "``subl``"
220exists anywhere in the file.
221
222The FileCheck -check-prefix option
223~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
224
225The FileCheck `-check-prefix` option allows multiple test
226configurations to be driven from one `.ll` file.  This is useful in many
227circumstances, for example, testing different architectural variants with
228:program:`llc`.  Here's a simple example:
229
230.. code-block:: llvm
231
232   ; RUN: llvm-as < %s | llc -mtriple=i686-apple-darwin9 -mattr=sse41 \
233   ; RUN:              | FileCheck %s -check-prefix=X32
234   ; RUN: llvm-as < %s | llc -mtriple=x86_64-apple-darwin9 -mattr=sse41 \
235   ; RUN:              | FileCheck %s -check-prefix=X64
236
237   define <4 x i32> @pinsrd_1(i32 %s, <4 x i32> %tmp) nounwind {
238           %tmp1 = insertelement <4 x i32>; %tmp, i32 %s, i32 1
239           ret <4 x i32> %tmp1
240   ; X32: pinsrd_1:
241   ; X32:    pinsrd $1, 4(%esp), %xmm0
242
243   ; X64: pinsrd_1:
244   ; X64:    pinsrd $1, %edi, %xmm0
245   }
246
247In this case, we're testing that we get the expected code generation with
248both 32-bit and 64-bit code generation.
249
250The "COM:" directive
251~~~~~~~~~~~~~~~~~~~~
252
253Sometimes you want to disable a FileCheck directive without removing it
254entirely, or you want to write comments that mention a directive by name. The
255"``COM:``" directive makes it easy to do this. For example, you might have:
256
257.. code-block:: llvm
258
259   ; X32: pinsrd_1:
260   ; X32:    pinsrd $1, 4(%esp), %xmm0
261
262   ; COM: FIXME: X64 isn't working correctly yet for this part of codegen, but
263   ; COM: X64 will have something similar to X32:
264   ; COM:
265   ; COM:   X64: pinsrd_1:
266   ; COM:   X64:    pinsrd $1, %edi, %xmm0
267
268Without "``COM:``", you would need to use some combination of rewording and
269directive syntax mangling to prevent FileCheck from recognizing the commented
270occurrences of "``X32:``" and "``X64:``" above as directives. Moreover,
271FileCheck diagnostics have been proposed that might complain about the above
272occurrences of "``X64``" that don't have the trailing "``:``" because they look
273like directive typos. Dodging all these problems can be tedious for a test
274author, and directive syntax mangling can make the purpose of test code unclear.
275"``COM:``" avoids all these problems.
276
277A few important usage notes:
278
279* "``COM:``" within another directive's pattern does *not* comment out the
280  remainder of the pattern. For example:
281
282  .. code-block:: llvm
283
284     ; X32: pinsrd $1, 4(%esp), %xmm0 COM: This is part of the X32 pattern!
285
286  If you need to temporarily comment out part of a directive's pattern, move it
287  to another line. The reason is that FileCheck parses "``COM:``" in the same
288  manner as any other directive: only the first directive on the line is
289  recognized as a directive.
290
291* For the sake of LIT, FileCheck treats "``RUN:``" just like "``COM:``". If this
292  is not suitable for your test environment, see :option:`--comment-prefixes`.
293
294* FileCheck does not recognize "``COM``", "``RUN``", or any user-defined comment
295  prefix as a comment directive if it's combined with one of the usual check
296  directive suffixes, such as "``-NEXT:``" or "``-NOT:``", discussed below.
297  FileCheck treats such a combination as plain text instead. If it needs to act
298  as a comment directive for your test environment, define it as such with
299  :option:`--comment-prefixes`.
300
301The "CHECK-NEXT:" directive
302~~~~~~~~~~~~~~~~~~~~~~~~~~~
303
304Sometimes you want to match lines and would like to verify that matches
305happen on exactly consecutive lines with no other lines in between them.  In
306this case, you can use "``CHECK:``" and "``CHECK-NEXT:``" directives to specify
307this.  If you specified a custom check prefix, just use "``<PREFIX>-NEXT:``".
308For example, something like this works as you'd expect:
309
310.. code-block:: llvm
311
312   define void @t2(<2 x double>* %r, <2 x double>* %A, double %B) {
313 	%tmp3 = load <2 x double>* %A, align 16
314 	%tmp7 = insertelement <2 x double> undef, double %B, i32 0
315 	%tmp9 = shufflevector <2 x double> %tmp3,
316                               <2 x double> %tmp7,
317                               <2 x i32> < i32 0, i32 2 >
318 	store <2 x double> %tmp9, <2 x double>* %r, align 16
319 	ret void
320
321   ; CHECK:          t2:
322   ; CHECK: 	        movl	8(%esp), %eax
323   ; CHECK-NEXT: 	movapd	(%eax), %xmm0
324   ; CHECK-NEXT: 	movhpd	12(%esp), %xmm0
325   ; CHECK-NEXT: 	movl	4(%esp), %eax
326   ; CHECK-NEXT: 	movapd	%xmm0, (%eax)
327   ; CHECK-NEXT: 	ret
328   }
329
330"``CHECK-NEXT:``" directives reject the input unless there is exactly one
331newline between it and the previous directive.  A "``CHECK-NEXT:``" cannot be
332the first directive in a file.
333
334The "CHECK-SAME:" directive
335~~~~~~~~~~~~~~~~~~~~~~~~~~~
336
337Sometimes you want to match lines and would like to verify that matches happen
338on the same line as the previous match.  In this case, you can use "``CHECK:``"
339and "``CHECK-SAME:``" directives to specify this.  If you specified a custom
340check prefix, just use "``<PREFIX>-SAME:``".
341
342"``CHECK-SAME:``" is particularly powerful in conjunction with "``CHECK-NOT:``"
343(described below).
344
345For example, the following works like you'd expect:
346
347.. code-block:: llvm
348
349   !0 = !DILocation(line: 5, scope: !1, inlinedAt: !2)
350
351   ; CHECK:       !DILocation(line: 5,
352   ; CHECK-NOT:               column:
353   ; CHECK-SAME:              scope: ![[SCOPE:[0-9]+]]
354
355"``CHECK-SAME:``" directives reject the input if there are any newlines between
356it and the previous directive.  A "``CHECK-SAME:``" cannot be the first
357directive in a file.
358
359The "CHECK-EMPTY:" directive
360~~~~~~~~~~~~~~~~~~~~~~~~~~~~
361
362If you need to check that the next line has nothing on it, not even whitespace,
363you can use the "``CHECK-EMPTY:``" directive.
364
365.. code-block:: llvm
366
367   declare void @foo()
368
369   declare void @bar()
370   ; CHECK: foo
371   ; CHECK-EMPTY:
372   ; CHECK-NEXT: bar
373
374Just like "``CHECK-NEXT:``" the directive will fail if there is more than one
375newline before it finds the next blank line, and it cannot be the first
376directive in a file.
377
378The "CHECK-NOT:" directive
379~~~~~~~~~~~~~~~~~~~~~~~~~~
380
381The "``CHECK-NOT:``" directive is used to verify that a string doesn't occur
382between two matches (or before the first match, or after the last match).  For
383example, to verify that a load is removed by a transformation, a test like this
384can be used:
385
386.. code-block:: llvm
387
388   define i8 @coerce_offset0(i32 %V, i32* %P) {
389     store i32 %V, i32* %P
390
391     %P2 = bitcast i32* %P to i8*
392     %P3 = getelementptr i8* %P2, i32 2
393
394     %A = load i8* %P3
395     ret i8 %A
396   ; CHECK: @coerce_offset0
397   ; CHECK-NOT: load
398   ; CHECK: ret i8
399   }
400
401The "CHECK-COUNT:" directive
402~~~~~~~~~~~~~~~~~~~~~~~~~~~~
403
404If you need to match multiple lines with the same pattern over and over again
405you can repeat a plain ``CHECK:`` as many times as needed. If that looks too
406boring you can instead use a counted check "``CHECK-COUNT-<num>:``", where
407``<num>`` is a positive decimal number. It will match the pattern exactly
408``<num>`` times, no more and no less. If you specified a custom check prefix,
409just use "``<PREFIX>-COUNT-<num>:``" for the same effect.
410Here is a simple example:
411
412.. code-block:: text
413
414   Loop at depth 1
415   Loop at depth 1
416   Loop at depth 1
417   Loop at depth 1
418     Loop at depth 2
419       Loop at depth 3
420
421   ; CHECK-COUNT-6: Loop at depth {{[0-9]+}}
422   ; CHECK-NOT:     Loop at depth {{[0-9]+}}
423
424The "CHECK-DAG:" directive
425~~~~~~~~~~~~~~~~~~~~~~~~~~
426
427If it's necessary to match strings that don't occur in a strictly sequential
428order, "``CHECK-DAG:``" could be used to verify them between two matches (or
429before the first match, or after the last match). For example, clang emits
430vtable globals in reverse order. Using ``CHECK-DAG:``, we can keep the checks
431in the natural order:
432
433.. code-block:: c++
434
435    // RUN: %clang_cc1 %s -emit-llvm -o - | FileCheck %s
436
437    struct Foo { virtual void method(); };
438    Foo f;  // emit vtable
439    // CHECK-DAG: @_ZTV3Foo =
440
441    struct Bar { virtual void method(); };
442    Bar b;
443    // CHECK-DAG: @_ZTV3Bar =
444
445``CHECK-NOT:`` directives could be mixed with ``CHECK-DAG:`` directives to
446exclude strings between the surrounding ``CHECK-DAG:`` directives. As a result,
447the surrounding ``CHECK-DAG:`` directives cannot be reordered, i.e. all
448occurrences matching ``CHECK-DAG:`` before ``CHECK-NOT:`` must not fall behind
449occurrences matching ``CHECK-DAG:`` after ``CHECK-NOT:``. For example,
450
451.. code-block:: llvm
452
453   ; CHECK-DAG: BEFORE
454   ; CHECK-NOT: NOT
455   ; CHECK-DAG: AFTER
456
457This case will reject input strings where ``BEFORE`` occurs after ``AFTER``.
458
459With captured variables, ``CHECK-DAG:`` is able to match valid topological
460orderings of a DAG with edges from the definition of a variable to its use.
461It's useful, e.g., when your test cases need to match different output
462sequences from the instruction scheduler. For example,
463
464.. code-block:: llvm
465
466   ; CHECK-DAG: add [[REG1:r[0-9]+]], r1, r2
467   ; CHECK-DAG: add [[REG2:r[0-9]+]], r3, r4
468   ; CHECK:     mul r5, [[REG1]], [[REG2]]
469
470In this case, any order of that two ``add`` instructions will be allowed.
471
472If you are defining `and` using variables in the same ``CHECK-DAG:`` block,
473be aware that the definition rule can match `after` its use.
474
475So, for instance, the code below will pass:
476
477.. code-block:: text
478
479  ; CHECK-DAG: vmov.32 [[REG2:d[0-9]+]][0]
480  ; CHECK-DAG: vmov.32 [[REG2]][1]
481  vmov.32 d0[1]
482  vmov.32 d0[0]
483
484While this other code, will not:
485
486.. code-block:: text
487
488  ; CHECK-DAG: vmov.32 [[REG2:d[0-9]+]][0]
489  ; CHECK-DAG: vmov.32 [[REG2]][1]
490  vmov.32 d1[1]
491  vmov.32 d0[0]
492
493While this can be very useful, it's also dangerous, because in the case of
494register sequence, you must have a strong order (read before write, copy before
495use, etc). If the definition your test is looking for doesn't match (because
496of a bug in the compiler), it may match further away from the use, and mask
497real bugs away.
498
499In those cases, to enforce the order, use a non-DAG directive between DAG-blocks.
500
501A ``CHECK-DAG:`` directive skips matches that overlap the matches of any
502preceding ``CHECK-DAG:`` directives in the same ``CHECK-DAG:`` block.  Not only
503is this non-overlapping behavior consistent with other directives, but it's
504also necessary to handle sets of non-unique strings or patterns.  For example,
505the following directives look for unordered log entries for two tasks in a
506parallel program, such as the OpenMP runtime:
507
508.. code-block:: text
509
510    // CHECK-DAG: [[THREAD_ID:[0-9]+]]: task_begin
511    // CHECK-DAG: [[THREAD_ID]]: task_end
512    //
513    // CHECK-DAG: [[THREAD_ID:[0-9]+]]: task_begin
514    // CHECK-DAG: [[THREAD_ID]]: task_end
515
516The second pair of directives is guaranteed not to match the same log entries
517as the first pair even though the patterns are identical and even if the text
518of the log entries is identical because the thread ID manages to be reused.
519
520The "CHECK-LABEL:" directive
521~~~~~~~~~~~~~~~~~~~~~~~~~~~~
522
523Sometimes in a file containing multiple tests divided into logical blocks, one
524or more ``CHECK:`` directives may inadvertently succeed by matching lines in a
525later block. While an error will usually eventually be generated, the check
526flagged as causing the error may not actually bear any relationship to the
527actual source of the problem.
528
529In order to produce better error messages in these cases, the "``CHECK-LABEL:``"
530directive can be used. It is treated identically to a normal ``CHECK``
531directive except that FileCheck makes an additional assumption that a line
532matched by the directive cannot also be matched by any other check present in
533``match-filename``; this is intended to be used for lines containing labels or
534other unique identifiers. Conceptually, the presence of ``CHECK-LABEL`` divides
535the input stream into separate blocks, each of which is processed independently,
536preventing a ``CHECK:`` directive in one block matching a line in another block.
537If ``--enable-var-scope`` is in effect, all local variables are cleared at the
538beginning of the block.
539
540For example,
541
542.. code-block:: llvm
543
544  define %struct.C* @C_ctor_base(%struct.C* %this, i32 %x) {
545  entry:
546  ; CHECK-LABEL: C_ctor_base:
547  ; CHECK: mov [[SAVETHIS:r[0-9]+]], r0
548  ; CHECK: bl A_ctor_base
549  ; CHECK: mov r0, [[SAVETHIS]]
550    %0 = bitcast %struct.C* %this to %struct.A*
551    %call = tail call %struct.A* @A_ctor_base(%struct.A* %0)
552    %1 = bitcast %struct.C* %this to %struct.B*
553    %call2 = tail call %struct.B* @B_ctor_base(%struct.B* %1, i32 %x)
554    ret %struct.C* %this
555  }
556
557  define %struct.D* @D_ctor_base(%struct.D* %this, i32 %x) {
558  entry:
559  ; CHECK-LABEL: D_ctor_base:
560
561The use of ``CHECK-LABEL:`` directives in this case ensures that the three
562``CHECK:`` directives only accept lines corresponding to the body of the
563``@C_ctor_base`` function, even if the patterns match lines found later in
564the file. Furthermore, if one of these three ``CHECK:`` directives fail,
565FileCheck will recover by continuing to the next block, allowing multiple test
566failures to be detected in a single invocation.
567
568There is no requirement that ``CHECK-LABEL:`` directives contain strings that
569correspond to actual syntactic labels in a source or output language: they must
570simply uniquely match a single line in the file being verified.
571
572``CHECK-LABEL:`` directives cannot contain variable definitions or uses.
573
574FileCheck Regex Matching Syntax
575~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
576
577All FileCheck directives take a pattern to match.
578For most uses of FileCheck, fixed string matching is perfectly sufficient.  For
579some things, a more flexible form of matching is desired.  To support this,
580FileCheck allows you to specify regular expressions in matching strings,
581surrounded by double braces: ``{{yourregex}}``. FileCheck implements a POSIX
582regular expression matcher; it supports Extended POSIX regular expressions
583(ERE). Because we want to use fixed string matching for a majority of what we
584do, FileCheck has been designed to support mixing and matching fixed string
585matching with regular expressions.  This allows you to write things like this:
586
587.. code-block:: llvm
588
589   ; CHECK: movhpd	{{[0-9]+}}(%esp), {{%xmm[0-7]}}
590
591In this case, any offset from the ESP register will be allowed, and any xmm
592register will be allowed.
593
594Because regular expressions are enclosed with double braces, they are
595visually distinct, and you don't need to use escape characters within the double
596braces like you would in C.  In the rare case that you want to match double
597braces explicitly from the input, you can use something ugly like
598``{{[}][}]}}`` as your pattern.  Or if you are using the repetition count
599syntax, for example ``[[:xdigit:]]{8}`` to match exactly 8 hex digits, you
600would need to add parentheses like this ``{{([[:xdigit:]]{8})}}`` to avoid
601confusion with FileCheck's closing double-brace.
602
603FileCheck String Substitution Blocks
604~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
605
606It is often useful to match a pattern and then verify that it occurs again
607later in the file.  For codegen tests, this can be useful to allow any
608register, but verify that that register is used consistently later.  To do
609this, :program:`FileCheck` supports string substitution blocks that allow
610string variables to be defined and substituted into patterns.  Here is a simple
611example:
612
613.. code-block:: llvm
614
615   ; CHECK: test5:
616   ; CHECK:    notw	[[REGISTER:%[a-z]+]]
617   ; CHECK:    andw	{{.*}}[[REGISTER]]
618
619The first check line matches a regex ``%[a-z]+`` and captures it into the
620string variable ``REGISTER``.  The second line verifies that whatever is in
621``REGISTER`` occurs later in the file after an "``andw``". :program:`FileCheck`
622string substitution blocks are always contained in ``[[ ]]`` pairs, and string
623variable names can be formed with the regex ``[a-zA-Z_][a-zA-Z0-9_]*``.  If a
624colon follows the name, then it is a definition of the variable; otherwise, it
625is a substitution.
626
627:program:`FileCheck` variables can be defined multiple times, and substitutions
628always get the latest value.  Variables can also be substituted later on the
629same line they were defined on. For example:
630
631.. code-block:: llvm
632
633    ; CHECK: op [[REG:r[0-9]+]], [[REG]]
634
635Can be useful if you want the operands of ``op`` to be the same register,
636and don't care exactly which register it is.
637
638If ``--enable-var-scope`` is in effect, variables with names that
639start with ``$`` are considered to be global. All others variables are
640local.  All local variables get undefined at the beginning of each
641CHECK-LABEL block. Global variables are not affected by CHECK-LABEL.
642This makes it easier to ensure that individual tests are not affected
643by variables set in preceding tests.
644
645FileCheck Numeric Substitution Blocks
646~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
647
648:program:`FileCheck` also supports numeric substitution blocks that allow
649defining numeric variables and checking for numeric values that satisfy a
650numeric expression constraint based on those variables via a numeric
651substitution. This allows ``CHECK:`` directives to verify a numeric relation
652between two numbers, such as the need for consecutive registers to be used.
653
654The syntax to define a numeric variable is ``[[#%<fmtspec>,<NUMVAR>:]]`` where:
655
656* ``%<fmtspec>`` is an optional scanf-style matching format specifier to
657  indicate what number format to match (e.g. hex number).  Currently accepted
658  format specifiers are ``%u``, ``%d``, ``%x`` and ``%X``.  If absent, the
659  format specifier defaults to ``%u``.
660
661* ``<NUMVAR>`` is the name of the numeric variable to define to the matching
662  value.
663
664For example:
665
666.. code-block:: llvm
667
668    ; CHECK: mov r[[#REG:]], 0x[[#%X,IMM:]]
669
670would match ``mov r5, 0xF0F0`` and set ``REG`` to the value ``5`` and ``IMM``
671to the value ``0xF0F0``.
672
673The syntax of a numeric substitution is
674``[[#%<fmtspec>: <constraint> <expr>]]`` where:
675
676* ``%<fmtspec>`` is the same matching format specifier as for defining numeric
677  variables but acting as a printf-style format to indicate how a numeric
678  expression value should be matched against.  If absent, the format specifier
679  is inferred from the matching format of the numeric variable(s) used by the
680  expression constraint if any, and defaults to ``%u`` if no numeric variable
681  is used.  In case of conflict between matching formats of several numeric
682  variables the format specifier is mandatory.
683
684* ``<constraint>`` is the constraint describing how the value to match must
685  relate to the value of the numeric expression. The only currently accepted
686  constraint is ``==`` for an exact match and is the default if
687  ``<constraint>`` is not provided. No matching constraint must be specified
688  when the ``<expr>`` is empty.
689
690* ``<expr>`` is an expression. An expression is in turn recursively defined
691  as:
692
693  * a numeric operand, or
694  * an expression followed by an operator and a numeric operand.
695
696  A numeric operand is a previously defined numeric variable, an integer
697  literal, or a function. Spaces are accepted before, after and between any of
698  these elements. Numeric operands have 64-bit precision. Overflow and underflow
699  are rejected. There is no support for operator precendence, but parentheses
700  can be used to change the evaluation order.
701
702The supported operators are:
703
704  * ``+`` - Returns the sum of its two operands.
705  * ``-`` - Returns the difference of its two operands.
706
707The syntax of a function call is ``<name>(<arguments>)`` where:
708
709* ``name`` is a predefined string literal. Accepted values are:
710
711  * add - Returns the sum of its two operands.
712  * div - Returns the quotient of its two operands.
713  * max - Returns the largest of its two operands.
714  * min - Returns the smallest of its two operands.
715  * mul - Returns the product of its two operands.
716  * sub - Returns the difference of its two operands.
717
718* ``<arguments>`` is a comma seperated list of expressions.
719
720For example:
721
722.. code-block:: llvm
723
724    ; CHECK: load r[[#REG:]], [r0]
725    ; CHECK: load r[[#REG+1]], [r1]
726    ; CHECK: Loading from 0x[[#%x,ADDR:]]
727    ; CHECK-SAME: to 0x[[#ADDR + 7]]
728
729The above example would match the text:
730
731.. code-block:: gas
732
733    load r5, [r0]
734    load r6, [r1]
735    Loading from 0xa0463440 to 0xa0463447
736
737but would not match the text:
738
739.. code-block:: gas
740
741    load r5, [r0]
742    load r7, [r1]
743    Loading from 0xa0463440 to 0xa0463443
744
745Due to ``7`` being unequal to ``5 + 1`` and ``a0463443`` being unequal to
746``a0463440 + 7``.
747
748The syntax also supports an empty expression, equivalent to writing {{[0-9]+}},
749for cases where the input must contain a numeric value but the value itself
750does not matter:
751
752.. code-block:: gas
753
754    ; CHECK-NOT: mov r0, r[[#]]
755
756to check that a value is synthesized rather than moved around.
757
758A numeric variable can also be defined to the result of a numeric expression,
759in which case the numeric expression constraint is checked and if verified the
760variable is assigned to the value. The unified syntax for both defining numeric
761variables and checking a numeric expression is thus
762``[[#%<fmtspec>,<NUMVAR>: <constraint> <expr>]]`` with each element as
763described previously. One can use this syntax to make a testcase more
764self-describing by using variables instead of values:
765
766.. code-block:: gas
767
768    ; CHECK: mov r[[#REG_OFFSET:]], 0x[[#%X,FIELD_OFFSET:12]]
769    ; CHECK-NEXT: load r[[#]], [r[[#REG_BASE:]], r[[#REG_OFFSET]]]
770
771which would match:
772
773.. code-block:: gas
774
775    mov r4, 0xC
776    load r6, [r5, r4]
777
778The ``--enable-var-scope`` option has the same effect on numeric variables as
779on string variables.
780
781Important note: In its current implementation, an expression cannot use a
782numeric variable defined earlier in the same CHECK directive.
783
784FileCheck Pseudo Numeric Variables
785~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
786
787Sometimes there's a need to verify output that contains line numbers of the
788match file, e.g. when testing compiler diagnostics.  This introduces a certain
789fragility of the match file structure, as "``CHECK:``" lines contain absolute
790line numbers in the same file, which have to be updated whenever line numbers
791change due to text addition or deletion.
792
793To support this case, FileCheck expressions understand the ``@LINE`` pseudo
794numeric variable which evaluates to the line number of the CHECK pattern where
795it is found.
796
797This way match patterns can be put near the relevant test lines and include
798relative line number references, for example:
799
800.. code-block:: c++
801
802   // CHECK: test.cpp:[[# @LINE + 4]]:6: error: expected ';' after top level declarator
803   // CHECK-NEXT: {{^int a}}
804   // CHECK-NEXT: {{^     \^}}
805   // CHECK-NEXT: {{^     ;}}
806   int a
807
808To support legacy uses of ``@LINE`` as a special string variable,
809:program:`FileCheck` also accepts the following uses of ``@LINE`` with string
810substitution block syntax: ``[[@LINE]]``, ``[[@LINE+<offset>]]`` and
811``[[@LINE-<offset>]]`` without any spaces inside the brackets and where
812``offset`` is an integer.
813
814Matching Newline Characters
815~~~~~~~~~~~~~~~~~~~~~~~~~~~
816
817To match newline characters in regular expressions the character class
818``[[:space:]]`` can be used. For example, the following pattern:
819
820.. code-block:: c++
821
822   // CHECK: DW_AT_location [DW_FORM_sec_offset] ([[DLOC:0x[0-9a-f]+]]){{[[:space:]].*}}"intd"
823
824matches output of the form (from llvm-dwarfdump):
825
826.. code-block:: text
827
828       DW_AT_location [DW_FORM_sec_offset]   (0x00000233)
829       DW_AT_name [DW_FORM_strp]  ( .debug_str[0x000000c9] = "intd")
830
831letting us set the :program:`FileCheck` variable ``DLOC`` to the desired value
832``0x00000233``, extracted from the line immediately preceding "``intd``".
833