1from __future__ import print_function
2
3import copy
4import glob
5import os
6import re
7import subprocess
8import sys
9
10if sys.version_info[0] > 2:
11  class string:
12    expandtabs = str.expandtabs
13else:
14  import string
15
16##### Common utilities for update_*test_checks.py
17
18
19_verbose = False
20_prefix_filecheck_ir_name = ''
21
22def parse_commandline_args(parser):
23  parser.add_argument('--include-generated-funcs', action='store_true',
24                      help='Output checks for functions not in source')
25  parser.add_argument('-v', '--verbose', action='store_true',
26                      help='Show verbose output')
27  parser.add_argument('-u', '--update-only', action='store_true',
28                      help='Only update test if it was already autogened')
29  parser.add_argument('--force-update', action='store_true',
30                      help='Update test even if it was autogened by a different script')
31  parser.add_argument('--enable', action='store_true', dest='enabled', default=True,
32                       help='Activate CHECK line generation from this point forward')
33  parser.add_argument('--disable', action='store_false', dest='enabled',
34                      help='Deactivate CHECK line generation from this point forward')
35  parser.add_argument('--replace-value-regex', nargs='+', default=[],
36                      help='List of regular expressions to replace matching value names')
37  parser.add_argument('--prefix-filecheck-ir-name', default='',
38                      help='Add a prefix to FileCheck IR value names to avoid conflicts with scripted names')
39  parser.add_argument('--global-value-regex', nargs='+', default=[],
40                      help='List of regular expressions that a global value declaration must match to generate a check (has no effect if checking globals is not enabled)')
41  parser.add_argument('--global-hex-value-regex', nargs='+', default=[],
42                      help='List of regular expressions such that, for matching global value declarations, literal integer values should be encoded in hex in the associated FileCheck directives')
43  args = parser.parse_args()
44  global _verbose, _global_value_regex, _global_hex_value_regex
45  _verbose = args.verbose
46  _global_value_regex = args.global_value_regex
47  _global_hex_value_regex = args.global_hex_value_regex
48  return args
49
50
51class InputLineInfo(object):
52  def __init__(self, line, line_number, args, argv):
53    self.line = line
54    self.line_number = line_number
55    self.args = args
56    self.argv = argv
57
58
59class TestInfo(object):
60  def __init__(self, test, parser, script_name, input_lines, args, argv,
61               comment_prefix, argparse_callback):
62    self.parser = parser
63    self.argparse_callback = argparse_callback
64    self.path = test
65    self.args = args
66    if args.prefix_filecheck_ir_name:
67      global _prefix_filecheck_ir_name
68      _prefix_filecheck_ir_name = args.prefix_filecheck_ir_name
69    self.argv = argv
70    self.input_lines = input_lines
71    self.run_lines = find_run_lines(test, self.input_lines)
72    self.comment_prefix = comment_prefix
73    if self.comment_prefix is None:
74      if self.path.endswith('.mir'):
75        self.comment_prefix = '#'
76      else:
77        self.comment_prefix = ';'
78    self.autogenerated_note_prefix = self.comment_prefix + ' ' + UTC_ADVERT
79    self.test_autogenerated_note = self.autogenerated_note_prefix + script_name
80    self.test_autogenerated_note += get_autogennote_suffix(parser, self.args)
81
82  def ro_iterlines(self):
83    for line_num, input_line in enumerate(self.input_lines):
84      args, argv = check_for_command(input_line, self.parser,
85                                     self.args, self.argv, self.argparse_callback)
86      yield InputLineInfo(input_line, line_num, args, argv)
87
88  def iterlines(self, output_lines):
89    output_lines.append(self.test_autogenerated_note)
90    for line_info in self.ro_iterlines():
91      input_line = line_info.line
92      # Discard any previous script advertising.
93      if input_line.startswith(self.autogenerated_note_prefix):
94        continue
95      self.args = line_info.args
96      self.argv = line_info.argv
97      if not self.args.enabled:
98        output_lines.append(input_line)
99        continue
100      yield line_info
101
102def itertests(test_patterns, parser, script_name, comment_prefix=None, argparse_callback=None):
103  for pattern in test_patterns:
104    # On Windows we must expand the patterns ourselves.
105    tests_list = glob.glob(pattern)
106    if not tests_list:
107      warn("Test file pattern '%s' was not found. Ignoring it." % (pattern,))
108      continue
109    for test in tests_list:
110      with open(test) as f:
111        input_lines = [l.rstrip() for l in f]
112      args = parser.parse_args()
113      if argparse_callback is not None:
114        argparse_callback(args)
115      argv = sys.argv[:]
116      first_line = input_lines[0] if input_lines else ""
117      if UTC_ADVERT in first_line:
118        if script_name not in first_line and not args.force_update:
119          warn("Skipping test which wasn't autogenerated by " + script_name, test)
120          continue
121        args, argv = check_for_command(first_line, parser, args, argv, argparse_callback)
122      elif args.update_only:
123        assert UTC_ADVERT not in first_line
124        warn("Skipping test which isn't autogenerated: " + test)
125        continue
126      yield TestInfo(test, parser, script_name, input_lines, args, argv,
127                     comment_prefix, argparse_callback)
128
129
130def should_add_line_to_output(input_line, prefix_set, skip_global_checks = False, comment_marker = ';'):
131  # Skip any blank comment lines in the IR.
132  if not skip_global_checks and input_line.strip() == comment_marker:
133    return False
134  # Skip a special double comment line we use as a separator.
135  if input_line.strip() == comment_marker + SEPARATOR:
136    return False
137  # Skip any blank lines in the IR.
138  #if input_line.strip() == '':
139  #  return False
140  # And skip any CHECK lines. We're building our own.
141  m = CHECK_RE.match(input_line)
142  if m and m.group(1) in prefix_set:
143    if skip_global_checks:
144      global_ir_value_re = re.compile('\[\[', flags=(re.M))
145      return not global_ir_value_re.search(input_line)
146    return False
147
148  return True
149
150# Perform lit-like substitutions
151def getSubstitutions(sourcepath):
152  sourcedir = os.path.dirname(sourcepath)
153  return [('%s', sourcepath),
154          ('%S', sourcedir),
155          ('%p', sourcedir),
156          ('%{pathsep}', os.pathsep)]
157
158def applySubstitutions(s, substitutions):
159  for a,b in substitutions:
160    s = s.replace(a, b)
161  return s
162
163# Invoke the tool that is being tested.
164def invoke_tool(exe, cmd_args, ir, preprocess_cmd=None, verbose=False):
165  with open(ir) as ir_file:
166    substitutions = getSubstitutions(ir)
167
168    # TODO Remove the str form which is used by update_test_checks.py and
169    # update_llc_test_checks.py
170    # The safer list form is used by update_cc_test_checks.py
171    if preprocess_cmd:
172      # Allow pre-processing the IR file (e.g. using sed):
173      assert isinstance(preprocess_cmd, str)  # TODO: use a list instead of using shell
174      preprocess_cmd = applySubstitutions(preprocess_cmd, substitutions).strip()
175      if verbose:
176        print('Pre-processing input file: ', ir, " with command '",
177              preprocess_cmd, "'", sep="", file=sys.stderr)
178      # Python 2.7 doesn't have subprocess.DEVNULL:
179      with open(os.devnull, 'w') as devnull:
180        pp = subprocess.Popen(preprocess_cmd, shell=True, stdin=devnull,
181                              stdout=subprocess.PIPE)
182        ir_file = pp.stdout
183
184    if isinstance(cmd_args, list):
185      args = [applySubstitutions(a, substitutions) for a in cmd_args]
186      stdout = subprocess.check_output([exe] + args, stdin=ir_file)
187    else:
188      stdout = subprocess.check_output(exe + ' ' + applySubstitutions(cmd_args, substitutions),
189                                       shell=True, stdin=ir_file)
190    if sys.version_info[0] > 2:
191      stdout = stdout.decode()
192  # Fix line endings to unix CR style.
193  return stdout.replace('\r\n', '\n')
194
195##### LLVM IR parser
196RUN_LINE_RE = re.compile(r'^\s*(?://|[;#])\s*RUN:\s*(.*)$')
197CHECK_PREFIX_RE = re.compile(r'--?check-prefix(?:es)?[= ](\S+)')
198PREFIX_RE = re.compile('^[a-zA-Z0-9_-]+$')
199CHECK_RE = re.compile(r'^\s*(?://|[;#])\s*([^:]+?)(?:-NEXT|-NOT|-DAG|-LABEL|-SAME|-EMPTY)?:')
200
201UTC_ARGS_KEY = 'UTC_ARGS:'
202UTC_ARGS_CMD = re.compile(r'.*' + UTC_ARGS_KEY + '\s*(?P<cmd>.*)\s*$')
203UTC_ADVERT = 'NOTE: Assertions have been autogenerated by '
204
205OPT_FUNCTION_RE = re.compile(
206    r'^(\s*;\s*Function\sAttrs:\s(?P<attrs>[\w\s]+?))?\s*define\s+(?:internal\s+)?[^@]*@(?P<func>[\w.$-]+?)\s*'
207    r'(?P<args_and_sig>\((\)|(.*?[\w.-]+?)\))[^{]*\{)\n(?P<body>.*?)^\}$',
208    flags=(re.M | re.S))
209
210ANALYZE_FUNCTION_RE = re.compile(
211    r'^\s*\'(?P<analysis>[\w\s-]+?)\'\s+for\s+function\s+\'(?P<func>[\w.$-]+?)\':'
212    r'\s*\n(?P<body>.*)$',
213    flags=(re.X | re.S))
214
215IR_FUNCTION_RE = re.compile(r'^\s*define\s+(?:internal\s+)?[^@]*@"?([\w.$-]+)"?\s*\(')
216TRIPLE_IR_RE = re.compile(r'^\s*target\s+triple\s*=\s*"([^"]+)"$')
217TRIPLE_ARG_RE = re.compile(r'-mtriple[= ]([^ ]+)')
218MARCH_ARG_RE = re.compile(r'-march[= ]([^ ]+)')
219
220SCRUB_LEADING_WHITESPACE_RE = re.compile(r'^(\s+)')
221SCRUB_WHITESPACE_RE = re.compile(r'(?!^(|  \w))[ \t]+', flags=re.M)
222SCRUB_TRAILING_WHITESPACE_RE = re.compile(r'[ \t]+$', flags=re.M)
223SCRUB_TRAILING_WHITESPACE_TEST_RE = SCRUB_TRAILING_WHITESPACE_RE
224SCRUB_TRAILING_WHITESPACE_AND_ATTRIBUTES_RE = re.compile(r'([ \t]|(#[0-9]+))+$', flags=re.M)
225SCRUB_KILL_COMMENT_RE = re.compile(r'^ *#+ +kill:.*\n')
226SCRUB_LOOP_COMMENT_RE = re.compile(
227    r'# =>This Inner Loop Header:.*|# in Loop:.*', flags=re.M)
228SCRUB_TAILING_COMMENT_TOKEN_RE = re.compile(r'(?<=\S)+[ \t]*#$', flags=re.M)
229
230SEPARATOR = '.'
231
232def error(msg, test_file=None):
233  if test_file:
234    msg = '{}: {}'.format(msg, test_file)
235  print('ERROR: {}'.format(msg), file=sys.stderr)
236
237def warn(msg, test_file=None):
238  if test_file:
239    msg = '{}: {}'.format(msg, test_file)
240  print('WARNING: {}'.format(msg), file=sys.stderr)
241
242def debug(*args, **kwargs):
243  # Python2 does not allow def debug(*args, file=sys.stderr, **kwargs):
244  if 'file' not in kwargs:
245    kwargs['file'] = sys.stderr
246  if _verbose:
247    print(*args, **kwargs)
248
249def find_run_lines(test, lines):
250  debug('Scanning for RUN lines in test file:', test)
251  raw_lines = [m.group(1)
252               for m in [RUN_LINE_RE.match(l) for l in lines] if m]
253  run_lines = [raw_lines[0]] if len(raw_lines) > 0 else []
254  for l in raw_lines[1:]:
255    if run_lines[-1].endswith('\\'):
256      run_lines[-1] = run_lines[-1].rstrip('\\') + ' ' + l
257    else:
258      run_lines.append(l)
259  debug('Found {} RUN lines in {}:'.format(len(run_lines), test))
260  for l in run_lines:
261    debug('  RUN: {}'.format(l))
262  return run_lines
263
264def scrub_body(body):
265  # Scrub runs of whitespace out of the assembly, but leave the leading
266  # whitespace in place.
267  body = SCRUB_WHITESPACE_RE.sub(r' ', body)
268  # Expand the tabs used for indentation.
269  body = string.expandtabs(body, 2)
270  # Strip trailing whitespace.
271  body = SCRUB_TRAILING_WHITESPACE_TEST_RE.sub(r'', body)
272  return body
273
274def do_scrub(body, scrubber, scrubber_args, extra):
275  if scrubber_args:
276    local_args = copy.deepcopy(scrubber_args)
277    local_args[0].extra_scrub = extra
278    return scrubber(body, *local_args)
279  return scrubber(body, *scrubber_args)
280
281# Build up a dictionary of all the function bodies.
282class function_body(object):
283  def __init__(self, string, extra, args_and_sig, attrs):
284    self.scrub = string
285    self.extrascrub = extra
286    self.args_and_sig = args_and_sig
287    self.attrs = attrs
288  def is_same_except_arg_names(self, extrascrub, args_and_sig, attrs, is_asm):
289    arg_names = set()
290    def drop_arg_names(match):
291        arg_names.add(match.group(variable_group_in_ir_value_match))
292        if match.group(attribute_group_in_ir_value_match):
293            attr = match.group(attribute_group_in_ir_value_match)
294        else:
295            attr = ''
296        return match.group(1) + attr + match.group(match.lastindex)
297    def repl_arg_names(match):
298        if match.group(variable_group_in_ir_value_match) is not None and match.group(variable_group_in_ir_value_match) in arg_names:
299            return match.group(1) + match.group(match.lastindex)
300        return match.group(1) + match.group(2) + match.group(match.lastindex)
301    if self.attrs != attrs:
302      return False
303    ans0 = IR_VALUE_RE.sub(drop_arg_names, self.args_and_sig)
304    ans1 = IR_VALUE_RE.sub(drop_arg_names, args_and_sig)
305    if ans0 != ans1:
306        return False
307    if is_asm:
308        # Check without replacements, the replacements are not applied to the
309        # body for asm checks.
310        return self.extrascrub == extrascrub
311
312    es0 = IR_VALUE_RE.sub(repl_arg_names, self.extrascrub)
313    es1 = IR_VALUE_RE.sub(repl_arg_names, extrascrub)
314    es0 = SCRUB_IR_COMMENT_RE.sub(r'', es0)
315    es1 = SCRUB_IR_COMMENT_RE.sub(r'', es1)
316    return es0 == es1
317
318  def __str__(self):
319    return self.scrub
320
321class FunctionTestBuilder:
322  def __init__(self, run_list, flags, scrubber_args, path):
323    self._verbose = flags.verbose
324    self._record_args = flags.function_signature
325    self._check_attributes = flags.check_attributes
326    self._scrubber_args = scrubber_args
327    self._path = path
328    # Strip double-quotes if input was read by UTC_ARGS
329    self._replace_value_regex = list(map(lambda x: x.strip('"'), flags.replace_value_regex))
330    self._func_dict = {}
331    self._func_order = {}
332    self._global_var_dict = {}
333    for tuple in run_list:
334      for prefix in tuple[0]:
335        self._func_dict.update({prefix:dict()})
336        self._func_order.update({prefix: []})
337        self._global_var_dict.update({prefix:dict()})
338
339  def finish_and_get_func_dict(self):
340    for prefix in self._get_failed_prefixes():
341      warn('Prefix %s had conflicting output from different RUN lines for all functions in test %s' % (prefix,self._path,))
342    return self._func_dict
343
344  def func_order(self):
345    return self._func_order
346
347  def global_var_dict(self):
348    return self._global_var_dict
349
350  def process_run_line(self, function_re, scrubber, raw_tool_output, prefixes, is_asm):
351    build_global_values_dictionary(self._global_var_dict, raw_tool_output, prefixes)
352    for m in function_re.finditer(raw_tool_output):
353      if not m:
354        continue
355      func = m.group('func')
356      body = m.group('body')
357      attrs = m.group('attrs') if self._check_attributes else ''
358      # Determine if we print arguments, the opening brace, or nothing after the
359      # function name
360      if self._record_args and 'args_and_sig' in m.groupdict():
361          args_and_sig = scrub_body(m.group('args_and_sig').strip())
362      elif 'args_and_sig' in m.groupdict():
363          args_and_sig = '('
364      else:
365          args_and_sig = ''
366      scrubbed_body = do_scrub(body, scrubber, self._scrubber_args,
367                               extra=False)
368      scrubbed_extra = do_scrub(body, scrubber, self._scrubber_args,
369                                extra=True)
370      if 'analysis' in m.groupdict():
371        analysis = m.group('analysis')
372        if analysis.lower() != 'cost model analysis':
373          warn('Unsupported analysis mode: %r!' % (analysis,))
374      if func.startswith('stress'):
375        # We only use the last line of the function body for stress tests.
376        scrubbed_body = '\n'.join(scrubbed_body.splitlines()[-1:])
377      if self._verbose:
378        print('Processing function: ' + func, file=sys.stderr)
379        for l in scrubbed_body.splitlines():
380          print('  ' + l, file=sys.stderr)
381      for prefix in prefixes:
382        # Replace function names matching the regex.
383        for regex in self._replace_value_regex:
384          # Pattern that matches capture groups in the regex in leftmost order.
385          group_regex = re.compile('\(.*?\)')
386          # Replace function name with regex.
387          match = re.match(regex, func)
388          if match:
389            func_repl = regex
390            # Replace any capture groups with their matched strings.
391            for g in match.groups():
392              func_repl = group_regex.sub(re.escape(g), func_repl, count=1)
393            func = re.sub(func_repl, '{{' + func_repl + '}}', func)
394
395          # Replace all calls to regex matching functions.
396          matches = re.finditer(regex, scrubbed_body)
397          for match in matches:
398            func_repl = regex
399            # Replace any capture groups with their matched strings.
400            for g in match.groups():
401                func_repl = group_regex.sub(re.escape(g), func_repl, count=1)
402            # Substitute function call names that match the regex with the same
403            # capture groups set.
404            scrubbed_body = re.sub(func_repl, '{{' + func_repl + '}}',
405                                   scrubbed_body)
406
407        if func in self._func_dict[prefix]:
408          if (self._func_dict[prefix][func] is None or
409              str(self._func_dict[prefix][func]) != scrubbed_body or
410              self._func_dict[prefix][func].args_and_sig != args_and_sig or
411                  self._func_dict[prefix][func].attrs != attrs):
412            if (self._func_dict[prefix][func] is not None and
413                self._func_dict[prefix][func].is_same_except_arg_names(
414                scrubbed_extra,
415                args_and_sig,
416                attrs,
417                is_asm)):
418              self._func_dict[prefix][func].scrub = scrubbed_extra
419              self._func_dict[prefix][func].args_and_sig = args_and_sig
420              continue
421            else:
422              # This means a previous RUN line produced a body for this function
423              # that is different from the one produced by this current RUN line,
424              # so the body can't be common accross RUN lines. We use None to
425              # indicate that.
426              self._func_dict[prefix][func] = None
427              continue
428
429        self._func_dict[prefix][func] = function_body(
430            scrubbed_body, scrubbed_extra, args_and_sig, attrs)
431        self._func_order[prefix].append(func)
432
433  def _get_failed_prefixes(self):
434    # This returns the list of those prefixes that failed to match any function,
435    # because there were conflicting bodies produced by different RUN lines, in
436    # all instances of the prefix. Effectively, this prefix is unused and should
437    # be removed.
438    for prefix in self._func_dict:
439      if (self._func_dict[prefix] and
440          (not [fct for fct in self._func_dict[prefix]
441                if self._func_dict[prefix][fct] is not None])):
442        yield prefix
443
444
445##### Generator of LLVM IR CHECK lines
446
447SCRUB_IR_COMMENT_RE = re.compile(r'\s*;.*')
448
449# TODO: We should also derive check lines for global, debug, loop declarations, etc..
450
451class NamelessValue:
452    def __init__(self, check_prefix, check_key, ir_prefix, global_ir_prefix, global_ir_prefix_regexp,
453                 ir_regexp, global_ir_rhs_regexp, is_before_functions):
454        self.check_prefix = check_prefix
455        self.check_key = check_key
456        self.ir_prefix = ir_prefix
457        self.global_ir_prefix = global_ir_prefix
458        self.global_ir_prefix_regexp = global_ir_prefix_regexp
459        self.ir_regexp = ir_regexp
460        self.global_ir_rhs_regexp = global_ir_rhs_regexp
461        self.is_before_functions = is_before_functions
462
463# Description of the different "unnamed" values we match in the IR, e.g.,
464# (local) ssa values, (debug) metadata, etc.
465nameless_values = [
466    NamelessValue(r'TMP'  , '%' , r'%'           , None            , None                   , r'[\w$.-]+?' , None                 , False) ,
467    NamelessValue(r'ATTR' , '#' , r'#'           , None            , None                   , r'[0-9]+'    , None                 , False) ,
468    NamelessValue(r'ATTR' , '#' , None           , r'attributes #' , r'[0-9]+'              , None         , r'{[^}]*}'           , False) ,
469    NamelessValue(r'GLOB' , '@' , r'@'           , None            , None                   , r'[0-9]+'    , None                 , False) ,
470    NamelessValue(r'GLOB' , '@' , None           , r'@'            , r'[a-zA-Z0-9_$"\\.-]+' , None         , r'.+'                , True)  ,
471    NamelessValue(r'DBG'  , '!' , r'!dbg '       , None            , None                   , r'![0-9]+'   , None                 , False) ,
472    NamelessValue(r'PROF' , '!' , r'!prof '      , None            , None                   , r'![0-9]+'   , None                 , False) ,
473    NamelessValue(r'TBAA' , '!' , r'!tbaa '      , None            , None                   , r'![0-9]+'   , None                 , False) ,
474    NamelessValue(r'RNG'  , '!' , r'!range '     , None            , None                   , r'![0-9]+'   , None                 , False) ,
475    NamelessValue(r'LOOP' , '!' , r'!llvm.loop ' , None            , None                   , r'![0-9]+'   , None                 , False) ,
476    NamelessValue(r'META' , '!' , r'metadata '   , None            , None                   , r'![0-9]+'   , None                 , False) ,
477    NamelessValue(r'META' , '!' , None           , r''             , r'![0-9]+'             , None         , r'(?:distinct |)!.*' , False) ,
478]
479
480def createOrRegexp(old, new):
481    if not old:
482        return new
483    if not new:
484        return old
485    return old + '|' + new
486
487def createPrefixMatch(prefix_str, prefix_re):
488    if prefix_str is None or prefix_re is None:
489        return ''
490    return '(?:' + prefix_str + '(' + prefix_re + '))'
491
492# Build the regexp that matches an "IR value". This can be a local variable,
493# argument, global, or metadata, anything that is "named". It is important that
494# the PREFIX and SUFFIX below only contain a single group, if that changes
495# other locations will need adjustment as well.
496IR_VALUE_REGEXP_PREFIX = r'(\s*)'
497IR_VALUE_REGEXP_STRING = r''
498for nameless_value in nameless_values:
499    lcl_match = createPrefixMatch(nameless_value.ir_prefix, nameless_value.ir_regexp)
500    glb_match = createPrefixMatch(nameless_value.global_ir_prefix, nameless_value.global_ir_prefix_regexp)
501    assert((lcl_match or glb_match) and not (lcl_match and glb_match))
502    if lcl_match:
503        IR_VALUE_REGEXP_STRING = createOrRegexp(IR_VALUE_REGEXP_STRING, lcl_match)
504    elif glb_match:
505        IR_VALUE_REGEXP_STRING = createOrRegexp(IR_VALUE_REGEXP_STRING, '^' + glb_match)
506IR_VALUE_REGEXP_SUFFIX = r'([,\s\(\)]|\Z)'
507IR_VALUE_RE = re.compile(IR_VALUE_REGEXP_PREFIX + r'(' + IR_VALUE_REGEXP_STRING + r')' + IR_VALUE_REGEXP_SUFFIX)
508
509# The entire match is group 0, the prefix has one group (=1), the entire
510# IR_VALUE_REGEXP_STRING is one group (=2), and then the nameless values start.
511first_nameless_group_in_ir_value_match = 3
512
513# constants for the group id of special matches
514variable_group_in_ir_value_match = 3
515attribute_group_in_ir_value_match = 4
516
517# Check a match for IR_VALUE_RE and inspect it to determine if it was a local
518# value, %..., global @..., debug number !dbg !..., etc. See the PREFIXES above.
519def get_idx_from_ir_value_match(match):
520    for i in range(first_nameless_group_in_ir_value_match, match.lastindex):
521        if match.group(i) is not None:
522            return i - first_nameless_group_in_ir_value_match
523    error("Unable to identify the kind of IR value from the match!")
524    return 0
525
526# See get_idx_from_ir_value_match
527def get_name_from_ir_value_match(match):
528    return match.group(get_idx_from_ir_value_match(match) + first_nameless_group_in_ir_value_match)
529
530# Return the nameless prefix we use for this kind or IR value, see also
531# get_idx_from_ir_value_match
532def get_nameless_check_prefix_from_ir_value_match(match):
533    return nameless_values[get_idx_from_ir_value_match(match)].check_prefix
534
535# Return the IR prefix and check prefix we use for this kind or IR value, e.g., (%, TMP) for locals,
536# see also get_idx_from_ir_value_match
537def get_ir_prefix_from_ir_value_match(match):
538    idx = get_idx_from_ir_value_match(match)
539    if nameless_values[idx].ir_prefix and match.group(0).strip().startswith(nameless_values[idx].ir_prefix):
540        return nameless_values[idx].ir_prefix, nameless_values[idx].check_prefix
541    return nameless_values[idx].global_ir_prefix, nameless_values[idx].check_prefix
542
543def get_check_key_from_ir_value_match(match):
544    idx = get_idx_from_ir_value_match(match)
545    return nameless_values[idx].check_key
546
547# Return the IR regexp we use for this kind or IR value, e.g., [\w.-]+? for locals,
548# see also get_idx_from_ir_value_match
549def get_ir_prefix_from_ir_value_re_match(match):
550    # for backwards compatibility we check locals with '.*'
551    if is_local_def_ir_value_match(match):
552        return '.*'
553    idx = get_idx_from_ir_value_match(match)
554    if nameless_values[idx].ir_prefix and match.group(0).strip().startswith(nameless_values[idx].ir_prefix):
555        return nameless_values[idx].ir_regexp
556    return nameless_values[idx].global_ir_prefix_regexp
557
558# Return true if this kind of IR value is "local", basically if it matches '%{{.*}}'.
559def is_local_def_ir_value_match(match):
560    return nameless_values[get_idx_from_ir_value_match(match)].ir_prefix == '%'
561
562# Return true if this kind of IR value is "global", basically if it matches '#{{.*}}'.
563def is_global_scope_ir_value_match(match):
564    return nameless_values[get_idx_from_ir_value_match(match)].global_ir_prefix is not None
565
566# Return true if var clashes with the scripted FileCheck check_prefix.
567def may_clash_with_default_check_prefix_name(check_prefix, var):
568  return check_prefix and re.match(r'^' + check_prefix + r'[0-9]+?$', var, re.IGNORECASE)
569
570# Create a FileCheck variable name based on an IR name.
571def get_value_name(var, check_prefix):
572  var = var.replace('!', '')
573  # This is a nameless value, prepend check_prefix.
574  if var.isdigit():
575    var = check_prefix + var
576  else:
577    # This is a named value that clashes with the check_prefix, prepend with _prefix_filecheck_ir_name,
578    # if it has been defined.
579    if may_clash_with_default_check_prefix_name(check_prefix, var) and _prefix_filecheck_ir_name:
580      var = _prefix_filecheck_ir_name + var
581  var = var.replace('.', '_')
582  var = var.replace('-', '_')
583  return var.upper()
584
585# Create a FileCheck variable from regex.
586def get_value_definition(var, match):
587  # for backwards compatibility we check locals with '.*'
588  if is_local_def_ir_value_match(match):
589    return '[[' + get_value_name(var, get_nameless_check_prefix_from_ir_value_match(match)) + ':' + \
590            get_ir_prefix_from_ir_value_match(match)[0] + get_ir_prefix_from_ir_value_re_match(match) + ']]'
591  prefix = get_ir_prefix_from_ir_value_match(match)[0]
592  return prefix + '[[' + get_value_name(var, get_nameless_check_prefix_from_ir_value_match(match)) + ':' + get_ir_prefix_from_ir_value_re_match(match) + ']]'
593
594# Use a FileCheck variable.
595def get_value_use(var, match, check_prefix):
596  if is_local_def_ir_value_match(match):
597    return '[[' + get_value_name(var, check_prefix) + ']]'
598  prefix = get_ir_prefix_from_ir_value_match(match)[0]
599  return prefix + '[[' + get_value_name(var, check_prefix) + ']]'
600
601# Replace IR value defs and uses with FileCheck variables.
602def generalize_check_lines(lines, is_analyze, vars_seen, global_vars_seen):
603  # This gets called for each match that occurs in
604  # a line. We transform variables we haven't seen
605  # into defs, and variables we have seen into uses.
606  def transform_line_vars(match):
607    pre, check = get_ir_prefix_from_ir_value_match(match)
608    var = get_name_from_ir_value_match(match)
609    for nameless_value in nameless_values:
610        if may_clash_with_default_check_prefix_name(nameless_value.check_prefix, var):
611          warn("Change IR value name '%s' or use -prefix-ir-filecheck-name to prevent possible conflict"
612            " with scripted FileCheck name." % (var,))
613    key = (var, get_check_key_from_ir_value_match(match))
614    is_local_def = is_local_def_ir_value_match(match)
615    if is_local_def and key in vars_seen:
616      rv = get_value_use(var, match, get_nameless_check_prefix_from_ir_value_match(match))
617    elif not is_local_def and key in global_vars_seen:
618      rv = get_value_use(var, match, global_vars_seen[key])
619    else:
620      if is_local_def:
621         vars_seen.add(key)
622      else:
623         global_vars_seen[key] = get_nameless_check_prefix_from_ir_value_match(match)
624      rv = get_value_definition(var, match)
625    # re.sub replaces the entire regex match
626    # with whatever you return, so we have
627    # to make sure to hand it back everything
628    # including the commas and spaces.
629    return match.group(1) + rv + match.group(match.lastindex)
630
631  lines_with_def = []
632
633  for i, line in enumerate(lines):
634    # An IR variable named '%.' matches the FileCheck regex string.
635    line = line.replace('%.', '%dot')
636    for regex in _global_hex_value_regex:
637      if re.match('^@' + regex + ' = ', line):
638        line = re.sub(r'\bi([0-9]+) ([0-9]+)',
639            lambda m : 'i' + m.group(1) + ' [[#' + hex(int(m.group(2))) + ']]',
640            line)
641        break
642    # Ignore any comments, since the check lines will too.
643    scrubbed_line = SCRUB_IR_COMMENT_RE.sub(r'', line)
644    lines[i] = scrubbed_line
645    if not is_analyze:
646      # It can happen that two matches are back-to-back and for some reason sub
647      # will not replace both of them. For now we work around this by
648      # substituting until there is no more match.
649      changed = True
650      while changed:
651          (lines[i], changed) = IR_VALUE_RE.subn(transform_line_vars, lines[i], count=1)
652  return lines
653
654
655def add_checks(output_lines, comment_marker, prefix_list, func_dict, func_name, check_label_format, is_asm, is_analyze, global_vars_seen_dict):
656  # prefix_exclusions are prefixes we cannot use to print the function because it doesn't exist in run lines that use these prefixes as well.
657  prefix_exclusions = set()
658  printed_prefixes = []
659  for p in prefix_list:
660    checkprefixes = p[0]
661    # If not all checkprefixes of this run line produced the function we cannot check for it as it does not
662    # exist for this run line. A subset of the check prefixes might know about the function but only because
663    # other run lines created it.
664    if any(map(lambda checkprefix: func_name not in func_dict[checkprefix], checkprefixes)):
665        prefix_exclusions |= set(checkprefixes)
666        continue
667
668  # prefix_exclusions is constructed, we can now emit the output
669  for p in prefix_list:
670    global_vars_seen = {}
671    checkprefixes = p[0]
672    for checkprefix in checkprefixes:
673      if checkprefix in global_vars_seen_dict:
674        global_vars_seen.update(global_vars_seen_dict[checkprefix])
675      else:
676        global_vars_seen_dict[checkprefix] = {}
677      if checkprefix in printed_prefixes:
678        break
679
680      # Check if the prefix is excluded.
681      if checkprefix in prefix_exclusions:
682        continue
683
684      # If we do not have output for this prefix we skip it.
685      if not func_dict[checkprefix][func_name]:
686        continue
687
688      # Add some space between different check prefixes, but not after the last
689      # check line (before the test code).
690      if is_asm:
691        if len(printed_prefixes) != 0:
692          output_lines.append(comment_marker)
693
694      if checkprefix not in global_vars_seen_dict:
695          global_vars_seen_dict[checkprefix] = {}
696
697      global_vars_seen_before = [key for key in global_vars_seen.keys()]
698
699      vars_seen = set()
700      printed_prefixes.append(checkprefix)
701      attrs = str(func_dict[checkprefix][func_name].attrs)
702      attrs = '' if attrs == 'None' else attrs
703      if attrs:
704        output_lines.append('%s %s: Function Attrs: %s' % (comment_marker, checkprefix, attrs))
705      args_and_sig = str(func_dict[checkprefix][func_name].args_and_sig)
706      args_and_sig = generalize_check_lines([args_and_sig], is_analyze, vars_seen, global_vars_seen)[0]
707      if '[[' in args_and_sig:
708        output_lines.append(check_label_format % (checkprefix, func_name, ''))
709        output_lines.append('%s %s-SAME: %s' % (comment_marker, checkprefix, args_and_sig))
710      else:
711        output_lines.append(check_label_format % (checkprefix, func_name, args_and_sig))
712      func_body = str(func_dict[checkprefix][func_name]).splitlines()
713
714      # For ASM output, just emit the check lines.
715      if is_asm:
716        output_lines.append('%s %s:       %s' % (comment_marker, checkprefix, func_body[0]))
717        for func_line in func_body[1:]:
718          if func_line.strip() == '':
719            output_lines.append('%s %s-EMPTY:' % (comment_marker, checkprefix))
720          else:
721            output_lines.append('%s %s-NEXT:  %s' % (comment_marker, checkprefix, func_line))
722        break
723
724      # For IR output, change all defs to FileCheck variables, so we're immune
725      # to variable naming fashions.
726      func_body = generalize_check_lines(func_body, is_analyze, vars_seen, global_vars_seen)
727
728      # This could be selectively enabled with an optional invocation argument.
729      # Disabled for now: better to check everything. Be safe rather than sorry.
730
731      # Handle the first line of the function body as a special case because
732      # it's often just noise (a useless asm comment or entry label).
733      #if func_body[0].startswith("#") or func_body[0].startswith("entry:"):
734      #  is_blank_line = True
735      #else:
736      #  output_lines.append('%s %s:       %s' % (comment_marker, checkprefix, func_body[0]))
737      #  is_blank_line = False
738
739      is_blank_line = False
740
741      for func_line in func_body:
742        if func_line.strip() == '':
743          is_blank_line = True
744          continue
745        # Do not waste time checking IR comments.
746        func_line = SCRUB_IR_COMMENT_RE.sub(r'', func_line)
747
748        # Skip blank lines instead of checking them.
749        if is_blank_line:
750          output_lines.append('{} {}:       {}'.format(
751              comment_marker, checkprefix, func_line))
752        else:
753          output_lines.append('{} {}-NEXT:  {}'.format(
754              comment_marker, checkprefix, func_line))
755        is_blank_line = False
756
757      # Add space between different check prefixes and also before the first
758      # line of code in the test function.
759      output_lines.append(comment_marker)
760
761      # Remembe new global variables we have not seen before
762      for key in global_vars_seen:
763          if key not in global_vars_seen_before:
764              global_vars_seen_dict[checkprefix][key] = global_vars_seen[key]
765      break
766
767def add_ir_checks(output_lines, comment_marker, prefix_list, func_dict,
768                  func_name, preserve_names, function_sig, global_vars_seen_dict):
769  # Label format is based on IR string.
770  function_def_regex = 'define {{[^@]+}}' if function_sig else ''
771  check_label_format = '{} %s-LABEL: {}@%s%s'.format(comment_marker, function_def_regex)
772  add_checks(output_lines, comment_marker, prefix_list, func_dict, func_name,
773             check_label_format, False, preserve_names, global_vars_seen_dict)
774
775def add_analyze_checks(output_lines, comment_marker, prefix_list, func_dict, func_name):
776  check_label_format = '{} %s-LABEL: \'%s%s\''.format(comment_marker)
777  global_vars_seen_dict = {}
778  add_checks(output_lines, comment_marker, prefix_list, func_dict, func_name,
779             check_label_format, False, True, global_vars_seen_dict)
780
781def build_global_values_dictionary(glob_val_dict, raw_tool_output, prefixes):
782  for nameless_value in nameless_values:
783    if nameless_value.global_ir_prefix is None:
784      continue
785
786    lhs_re_str = nameless_value.global_ir_prefix + nameless_value.global_ir_prefix_regexp
787    rhs_re_str = nameless_value.global_ir_rhs_regexp
788
789    global_ir_value_re_str = r'^' + lhs_re_str + r'\s=\s' + rhs_re_str + r'$'
790    global_ir_value_re = re.compile(global_ir_value_re_str, flags=(re.M))
791    lines = []
792    for m in global_ir_value_re.finditer(raw_tool_output):
793        lines.append(m.group(0))
794
795    for prefix in prefixes:
796      if glob_val_dict[prefix] is None:
797        continue
798      if nameless_value.check_prefix in glob_val_dict[prefix]:
799        if lines == glob_val_dict[prefix][nameless_value.check_prefix]:
800          continue
801        if prefix == prefixes[-1]:
802          warn('Found conflicting asm under the same prefix: %r!' % (prefix,))
803        else:
804          glob_val_dict[prefix][nameless_value.check_prefix] = None
805          continue
806      glob_val_dict[prefix][nameless_value.check_prefix] = lines
807
808def add_global_checks(glob_val_dict, comment_marker, prefix_list, output_lines, global_vars_seen_dict, is_analyze, is_before_functions):
809  printed_prefixes = set()
810  for nameless_value in nameless_values:
811    if nameless_value.global_ir_prefix is None:
812        continue
813    if nameless_value.is_before_functions != is_before_functions:
814        continue
815    for p in prefix_list:
816      global_vars_seen = {}
817      checkprefixes = p[0]
818      if checkprefixes is None:
819        continue
820      for checkprefix in checkprefixes:
821        if checkprefix in global_vars_seen_dict:
822            global_vars_seen.update(global_vars_seen_dict[checkprefix])
823        else:
824            global_vars_seen_dict[checkprefix] = {}
825        if (checkprefix, nameless_value.check_prefix) in printed_prefixes:
826          break
827        if not glob_val_dict[checkprefix]:
828          continue
829        if nameless_value.check_prefix not in glob_val_dict[checkprefix]:
830          continue
831        if not glob_val_dict[checkprefix][nameless_value.check_prefix]:
832          continue
833
834        check_lines = []
835        global_vars_seen_before = [key for key in global_vars_seen.keys()]
836        for line in glob_val_dict[checkprefix][nameless_value.check_prefix]:
837          if _global_value_regex:
838            matched = False
839            for regex in _global_value_regex:
840              if re.match('^@' + regex + ' = ', line):
841                matched = True
842                break
843            if not matched:
844              continue
845          tmp = generalize_check_lines([line], is_analyze, set(), global_vars_seen)
846          check_line = '%s %s: %s' % (comment_marker, checkprefix, tmp[0])
847          check_lines.append(check_line)
848        if not check_lines:
849          continue
850
851        output_lines.append(comment_marker + SEPARATOR)
852        for check_line in check_lines:
853          output_lines.append(check_line)
854
855        printed_prefixes.add((checkprefix, nameless_value.check_prefix))
856
857        # Remembe new global variables we have not seen before
858        for key in global_vars_seen:
859            if key not in global_vars_seen_before:
860                global_vars_seen_dict[checkprefix][key] = global_vars_seen[key]
861        break
862
863  if printed_prefixes:
864      output_lines.append(comment_marker + SEPARATOR)
865
866
867def check_prefix(prefix):
868  if not PREFIX_RE.match(prefix):
869        hint = ""
870        if ',' in prefix:
871          hint = " Did you mean '--check-prefixes=" + prefix + "'?"
872        warn(("Supplied prefix '%s' is invalid. Prefix must contain only alphanumeric characters, hyphens and underscores." + hint) %
873             (prefix))
874
875
876def verify_filecheck_prefixes(fc_cmd):
877  fc_cmd_parts = fc_cmd.split()
878  for part in fc_cmd_parts:
879    if "check-prefix=" in part:
880      prefix = part.split('=', 1)[1]
881      check_prefix(prefix)
882    elif "check-prefixes=" in part:
883      prefixes = part.split('=', 1)[1].split(',')
884      for prefix in prefixes:
885        check_prefix(prefix)
886        if prefixes.count(prefix) > 1:
887          warn("Supplied prefix '%s' is not unique in the prefix list." % (prefix,))
888
889
890def get_autogennote_suffix(parser, args):
891  autogenerated_note_args = ''
892  for action in parser._actions:
893    if not hasattr(args, action.dest):
894      continue  # Ignore options such as --help that aren't included in args
895    # Ignore parameters such as paths to the binary or the list of tests
896    if action.dest in ('tests', 'update_only', 'opt_binary', 'llc_binary',
897                       'clang', 'opt', 'llvm_bin', 'verbose'):
898      continue
899    value = getattr(args, action.dest)
900    if action.const is not None:  # action stores a constant (usually True/False)
901      # Skip actions with different constant values (this happens with boolean
902      # --foo/--no-foo options)
903      if value != action.const:
904        continue
905    if parser.get_default(action.dest) == value:
906      continue  # Don't add default values
907    autogenerated_note_args += action.option_strings[0] + ' '
908    if action.const is None:  # action takes a parameter
909      if action.nargs == '+':
910        value = ' '.join(map(lambda v: '"' + v.strip('"') + '"', value))
911      autogenerated_note_args += '%s ' % value
912  if autogenerated_note_args:
913    autogenerated_note_args = ' %s %s' % (UTC_ARGS_KEY, autogenerated_note_args[:-1])
914  return autogenerated_note_args
915
916
917def check_for_command(line, parser, args, argv, argparse_callback):
918    cmd_m = UTC_ARGS_CMD.match(line)
919    if cmd_m:
920        for option in cmd_m.group('cmd').strip().split(' '):
921            if option:
922                argv.append(option)
923        args = parser.parse_args(filter(lambda arg: arg not in args.tests, argv))
924        if argparse_callback is not None:
925          argparse_callback(args)
926    return args, argv
927
928def find_arg_in_test(test_info, get_arg_to_check, arg_string, is_global):
929  result = get_arg_to_check(test_info.args)
930  if not result and is_global:
931    # See if this has been specified via UTC_ARGS.  This is a "global" option
932    # that affects the entire generation of test checks.  If it exists anywhere
933    # in the test, apply it to everything.
934    saw_line = False
935    for line_info in test_info.ro_iterlines():
936      line = line_info.line
937      if not line.startswith(';') and line.strip() != '':
938        saw_line = True
939      result = get_arg_to_check(line_info.args)
940      if result:
941        if warn and saw_line:
942          # We saw the option after already reading some test input lines.
943          # Warn about it.
944          print('WARNING: Found {} in line following test start: '.format(arg_string)
945                + line, file=sys.stderr)
946          print('WARNING: Consider moving {} to top of file'.format(arg_string),
947                file=sys.stderr)
948        break
949  return result
950
951def dump_input_lines(output_lines, test_info, prefix_set, comment_string):
952  for input_line_info in test_info.iterlines(output_lines):
953    line = input_line_info.line
954    args = input_line_info.args
955    if line.strip() == comment_string:
956      continue
957    if line.strip() == comment_string + SEPARATOR:
958      continue
959    if line.lstrip().startswith(comment_string):
960      m = CHECK_RE.match(line)
961      if m and m.group(1) in prefix_set:
962        continue
963    output_lines.append(line.rstrip('\n'))
964
965def add_checks_at_end(output_lines, prefix_list, func_order,
966                      comment_string, check_generator):
967  added = set()
968  for prefix in prefix_list:
969    prefixes = prefix[0]
970    tool_args = prefix[1]
971    for prefix in prefixes:
972      for func in func_order[prefix]:
973        if added:
974          output_lines.append(comment_string)
975        added.add(func)
976
977        # The add_*_checks routines expect a run list whose items are
978        # tuples that have a list of prefixes as their first element and
979        # tool command args string as their second element.  They output
980        # checks for each prefix in the list of prefixes.  By doing so, it
981        # implicitly assumes that for each function every run line will
982        # generate something for that function.  That is not the case for
983        # generated functions as some run lines might not generate them
984        # (e.g. -fopenmp vs. no -fopenmp).
985        #
986        # Therefore, pass just the prefix we're interested in.  This has
987        # the effect of generating all of the checks for functions of a
988        # single prefix before moving on to the next prefix.  So checks
989        # are ordered by prefix instead of by function as in "normal"
990        # mode.
991        check_generator(output_lines,
992                        [([prefix], tool_args)],
993                        func)
994