xref: /llvm-project/llvm/utils/UpdateTestChecks/common.py (revision 7169ea391a7cc21da171e1c8acb5731e390d6852)
1from __future__ import print_function
2import re
3import string
4import subprocess
5import sys
6import copy
7
8if sys.version_info[0] > 2:
9  class string:
10    expandtabs = str.expandtabs
11else:
12  import string
13
14##### Common utilities for update_*test_checks.py
15
16def should_add_line_to_output(input_line, prefix_set):
17  # Skip any blank comment lines in the IR.
18  if input_line.strip() == ';':
19    return False
20  # Skip any blank lines in the IR.
21  #if input_line.strip() == '':
22  #  return False
23  # And skip any CHECK lines. We're building our own.
24  m = CHECK_RE.match(input_line)
25  if m and m.group(1) in prefix_set:
26    return False
27
28  return True
29
30# Invoke the tool that is being tested.
31def invoke_tool(exe, cmd_args, ir):
32  with open(ir) as ir_file:
33    # TODO Remove the str form which is used by update_test_checks.py and
34    # update_llc_test_checks.py
35    # The safer list form is used by update_cc_test_checks.py
36    if isinstance(cmd_args, list):
37      stdout = subprocess.check_output([exe] + cmd_args, stdin=ir_file)
38    else:
39      stdout = subprocess.check_output(exe + ' ' + cmd_args,
40                                       shell=True, stdin=ir_file)
41    if sys.version_info[0] > 2:
42      stdout = stdout.decode()
43  # Fix line endings to unix CR style.
44  return stdout.replace('\r\n', '\n')
45
46##### LLVM IR parser
47
48RUN_LINE_RE = re.compile('^\s*[;#]\s*RUN:\s*(.*)$')
49CHECK_PREFIX_RE = re.compile('--?check-prefix(?:es)?[= ](\S+)')
50PREFIX_RE = re.compile('^[a-zA-Z0-9_-]+$')
51CHECK_RE = re.compile(r'^\s*[;#]\s*([^:]+?)(?:-NEXT|-NOT|-DAG|-LABEL)?:')
52
53OPT_FUNCTION_RE = re.compile(
54    r'^\s*define\s+(?:internal\s+)?[^@]*@(?P<func>[\w-]+?)\s*\('
55    r'(\s+)?[^)]*[^{]*\{\n(?P<body>.*?)^\}$',
56    flags=(re.M | re.S))
57
58ANALYZE_FUNCTION_RE = re.compile(
59    r'^\s*\'(?P<analysis>[\w\s-]+?)\'\s+for\s+function\s+\'(?P<func>[\w-]+?)\':'
60    r'\s*\n(?P<body>.*)$',
61    flags=(re.X | re.S))
62
63IR_FUNCTION_RE = re.compile('^\s*define\s+(?:internal\s+)?[^@]*@(\w+)\s*\(')
64TRIPLE_IR_RE = re.compile(r'^\s*target\s+triple\s*=\s*"([^"]+)"$')
65TRIPLE_ARG_RE = re.compile(r'-mtriple[= ]([^ ]+)')
66MARCH_ARG_RE = re.compile(r'-march[= ]([^ ]+)')
67
68SCRUB_LEADING_WHITESPACE_RE = re.compile(r'^(\s+)')
69SCRUB_WHITESPACE_RE = re.compile(r'(?!^(|  \w))[ \t]+', flags=re.M)
70SCRUB_TRAILING_WHITESPACE_RE = re.compile(r'[ \t]+$', flags=re.M)
71SCRUB_KILL_COMMENT_RE = re.compile(r'^ *#+ +kill:.*\n')
72SCRUB_LOOP_COMMENT_RE = re.compile(
73    r'# =>This Inner Loop Header:.*|# in Loop:.*', flags=re.M)
74
75
76def error(msg, test_file=None):
77  if test_file:
78    msg = '{}: {}'.format(msg, test_file)
79  print('ERROR: {}'.format(msg), file=sys.stderr)
80
81def warn(msg, test_file=None):
82  if test_file:
83    msg = '{}: {}'.format(msg, test_file)
84  print('WARNING: {}'.format(msg), file=sys.stderr)
85
86def scrub_body(body):
87  # Scrub runs of whitespace out of the assembly, but leave the leading
88  # whitespace in place.
89  body = SCRUB_WHITESPACE_RE.sub(r' ', body)
90  # Expand the tabs used for indentation.
91  body = string.expandtabs(body, 2)
92  # Strip trailing whitespace.
93  body = SCRUB_TRAILING_WHITESPACE_RE.sub(r'', body)
94  return body
95
96def do_scrub(body, scrubber, scrubber_args, extra):
97  if scrubber_args:
98    local_args = copy.deepcopy(scrubber_args)
99    local_args[0].extra_scrub = extra
100    return scrubber(body, *local_args)
101  return scrubber(body, *scrubber_args)
102
103# Build up a dictionary of all the function bodies.
104class function_body(object):
105  def __init__(self, string, extra):
106    self.scrub = string
107    self.extrascrub = extra
108  def __str__(self):
109    return self.scrub
110
111def build_function_body_dictionary(function_re, scrubber, scrubber_args, raw_tool_output, prefixes, func_dict, verbose):
112  for m in function_re.finditer(raw_tool_output):
113    if not m:
114      continue
115    func = m.group('func')
116    body = m.group('body')
117    scrubbed_body = do_scrub(body, scrubber, scrubber_args, extra = False)
118    scrubbed_extra = do_scrub(body, scrubber, scrubber_args, extra = True)
119    if 'analysis' in m.groupdict():
120      analysis = m.group('analysis')
121      if analysis.lower() != 'cost model analysis':
122        warn('Unsupported analysis mode: %r!' % (analysis,))
123    if func.startswith('stress'):
124      # We only use the last line of the function body for stress tests.
125      scrubbed_body = '\n'.join(scrubbed_body.splitlines()[-1:])
126    if verbose:
127      print('Processing function: ' + func, file=sys.stderr)
128      for l in scrubbed_body.splitlines():
129        print('  ' + l, file=sys.stderr)
130    for prefix in prefixes:
131      if func in func_dict[prefix] and str(func_dict[prefix][func]) != scrubbed_body:
132        if func_dict[prefix][func] and func_dict[prefix][func].extrascrub == scrubbed_extra:
133          func_dict[prefix][func].scrub = scrubbed_extra
134          continue
135        else:
136          if prefix == prefixes[-1]:
137            warn('Found conflicting asm under the same prefix: %r!' % (prefix,))
138          else:
139            func_dict[prefix][func] = None
140            continue
141
142      func_dict[prefix][func] = function_body(scrubbed_body, scrubbed_extra)
143
144##### Generator of LLVM IR CHECK lines
145
146SCRUB_IR_COMMENT_RE = re.compile(r'\s*;.*')
147
148# Match things that look at identifiers, but only if they are followed by
149# spaces, commas, paren, or end of the string
150IR_VALUE_RE = re.compile(r'(\s+)%([\w\.\-]+?)([,\s\(\)]|\Z)')
151
152# Create a FileCheck variable name based on an IR name.
153def get_value_name(var):
154  if var.isdigit():
155    var = 'TMP' + var
156  var = var.replace('.', '_')
157  var = var.replace('-', '_')
158  return var.upper()
159
160
161# Create a FileCheck variable from regex.
162def get_value_definition(var):
163  return '[[' + get_value_name(var) + ':%.*]]'
164
165
166# Use a FileCheck variable.
167def get_value_use(var):
168  return '[[' + get_value_name(var) + ']]'
169
170# Replace IR value defs and uses with FileCheck variables.
171def genericize_check_lines(lines, is_analyze):
172  # This gets called for each match that occurs in
173  # a line. We transform variables we haven't seen
174  # into defs, and variables we have seen into uses.
175  def transform_line_vars(match):
176    var = match.group(2)
177    if var in vars_seen:
178      rv = get_value_use(var)
179    else:
180      vars_seen.add(var)
181      rv = get_value_definition(var)
182    # re.sub replaces the entire regex match
183    # with whatever you return, so we have
184    # to make sure to hand it back everything
185    # including the commas and spaces.
186    return match.group(1) + rv + match.group(3)
187
188  vars_seen = set()
189  lines_with_def = []
190
191  for i, line in enumerate(lines):
192    # An IR variable named '%.' matches the FileCheck regex string.
193    line = line.replace('%.', '%dot')
194    # Ignore any comments, since the check lines will too.
195    scrubbed_line = SCRUB_IR_COMMENT_RE.sub(r'', line)
196    if is_analyze:
197      lines[i] = scrubbed_line
198    else:
199      lines[i] = IR_VALUE_RE.sub(transform_line_vars, scrubbed_line)
200  return lines
201
202
203def add_checks(output_lines, comment_marker, prefix_list, func_dict, func_name, check_label_format, is_asm, is_analyze):
204  printed_prefixes = []
205  for p in prefix_list:
206    checkprefixes = p[0]
207    for checkprefix in checkprefixes:
208      if checkprefix in printed_prefixes:
209        break
210      # TODO func_dict[checkprefix] may be None, '' or not exist.
211      # Fix the call sites.
212      if func_name not in func_dict[checkprefix] or not func_dict[checkprefix][func_name]:
213        continue
214
215      # Add some space between different check prefixes, but not after the last
216      # check line (before the test code).
217      if is_asm:
218        if len(printed_prefixes) != 0:
219          output_lines.append(comment_marker)
220
221      printed_prefixes.append(checkprefix)
222      output_lines.append(check_label_format % (checkprefix, func_name))
223      func_body = str(func_dict[checkprefix][func_name]).splitlines()
224
225      # For ASM output, just emit the check lines.
226      if is_asm:
227        output_lines.append('%s %s:       %s' % (comment_marker, checkprefix, func_body[0]))
228        for func_line in func_body[1:]:
229          output_lines.append('%s %s-NEXT:  %s' % (comment_marker, checkprefix, func_line))
230        break
231
232      # For IR output, change all defs to FileCheck variables, so we're immune
233      # to variable naming fashions.
234      func_body = genericize_check_lines(func_body, is_analyze)
235
236      # This could be selectively enabled with an optional invocation argument.
237      # Disabled for now: better to check everything. Be safe rather than sorry.
238
239      # Handle the first line of the function body as a special case because
240      # it's often just noise (a useless asm comment or entry label).
241      #if func_body[0].startswith("#") or func_body[0].startswith("entry:"):
242      #  is_blank_line = True
243      #else:
244      #  output_lines.append('%s %s:       %s' % (comment_marker, checkprefix, func_body[0]))
245      #  is_blank_line = False
246
247      is_blank_line = False
248
249      for func_line in func_body:
250        if func_line.strip() == '':
251          is_blank_line = True
252          continue
253        # Do not waste time checking IR comments.
254        func_line = SCRUB_IR_COMMENT_RE.sub(r'', func_line)
255
256        # Skip blank lines instead of checking them.
257        if is_blank_line:
258          output_lines.append('{} {}:       {}'.format(
259              comment_marker, checkprefix, func_line))
260        else:
261          output_lines.append('{} {}-NEXT:  {}'.format(
262              comment_marker, checkprefix, func_line))
263        is_blank_line = False
264
265      # Add space between different check prefixes and also before the first
266      # line of code in the test function.
267      output_lines.append(comment_marker)
268      break
269
270def add_ir_checks(output_lines, comment_marker, prefix_list, func_dict, func_name):
271  # Label format is based on IR string.
272  check_label_format = '{} %s-LABEL: @%s('.format(comment_marker)
273  add_checks(output_lines, comment_marker, prefix_list, func_dict, func_name, check_label_format, False, False)
274
275def add_analyze_checks(output_lines, comment_marker, prefix_list, func_dict, func_name):
276  check_label_format = '{} %s-LABEL: \'%s\''.format(comment_marker)
277  add_checks(output_lines, comment_marker, prefix_list, func_dict, func_name, check_label_format, False, True)
278
279
280def check_prefix(prefix):
281  if not PREFIX_RE.match(prefix):
282        hint = ""
283        if ',' in prefix:
284          hint = " Did you mean '--check-prefixes=" + prefix + "'?"
285        warn(("Supplied prefix '%s' is invalid. Prefix must contain only alphanumeric characters, hyphens and underscores." + hint) %
286             (prefix))
287
288
289def verify_filecheck_prefixes(fc_cmd):
290  fc_cmd_parts = fc_cmd.split()
291  for part in fc_cmd_parts:
292    if "check-prefix=" in part:
293      prefix = part.split('=', 1)[1]
294      check_prefix(prefix)
295    elif "check-prefixes=" in part:
296      prefixes = part.split('=', 1)[1].split(',')
297      for prefix in prefixes:
298        check_prefix(prefix)
299        if prefixes.count(prefix) > 1:
300          warn("Supplied prefix '%s' is not unique in the prefix list." % (prefix,))
301