xref: /llvm-project/llvm/utils/UpdateTestChecks/common.py (revision 45be5e477e9216363191a8ac9123bea4585cf14f)
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
75def scrub_body(body):
76  # Scrub runs of whitespace out of the assembly, but leave the leading
77  # whitespace in place.
78  body = SCRUB_WHITESPACE_RE.sub(r' ', body)
79  # Expand the tabs used for indentation.
80  body = string.expandtabs(body, 2)
81  # Strip trailing whitespace.
82  body = SCRUB_TRAILING_WHITESPACE_RE.sub(r'', body)
83  return body
84
85def do_scrub(body, scrubber, scrubber_args, extra):
86  if scrubber_args:
87    local_args = copy.deepcopy(scrubber_args)
88    local_args[0].extra_scrub = extra
89    return scrubber(body, *local_args)
90  return scrubber(body, *scrubber_args)
91
92# Build up a dictionary of all the function bodies.
93class function_body(object):
94  def __init__(self, string, extra):
95    self.scrub = string
96    self.extrascrub = extra
97  def __str__(self):
98    return self.scrub
99
100def build_function_body_dictionary(function_re, scrubber, scrubber_args, raw_tool_output, prefixes, func_dict, verbose):
101  for m in function_re.finditer(raw_tool_output):
102    if not m:
103      continue
104    func = m.group('func')
105    body = m.group('body')
106    scrubbed_body = do_scrub(body, scrubber, scrubber_args, extra = False)
107    scrubbed_extra = do_scrub(body, scrubber, scrubber_args, extra = True)
108    if 'analysis' in m.groupdict():
109      analysis = m.group('analysis')
110      if analysis.lower() != 'cost model analysis':
111        print('WARNING: Unsupported analysis mode: %r!' % (analysis,), file=sys.stderr)
112    if func.startswith('stress'):
113      # We only use the last line of the function body for stress tests.
114      scrubbed_body = '\n'.join(scrubbed_body.splitlines()[-1:])
115    if verbose:
116      print('Processing function: ' + func, file=sys.stderr)
117      for l in scrubbed_body.splitlines():
118        print('  ' + l, file=sys.stderr)
119    for prefix in prefixes:
120      if func in func_dict[prefix] and str(func_dict[prefix][func]) != scrubbed_body:
121        if func_dict[prefix][func] and func_dict[prefix][func].extrascrub == scrubbed_extra:
122          func_dict[prefix][func].scrub = scrubbed_extra
123          continue
124        else:
125          if prefix == prefixes[-1]:
126            print('WARNING: Found conflicting asm under the '
127                                 'same prefix: %r!' % (prefix,), file=sys.stderr)
128          else:
129            func_dict[prefix][func] = None
130            continue
131
132      func_dict[prefix][func] = function_body(scrubbed_body, scrubbed_extra)
133
134##### Generator of LLVM IR CHECK lines
135
136SCRUB_IR_COMMENT_RE = re.compile(r'\s*;.*')
137
138# Match things that look at identifiers, but only if they are followed by
139# spaces, commas, paren, or end of the string
140IR_VALUE_RE = re.compile(r'(\s+)%([\w\.\-]+?)([,\s\(\)]|\Z)')
141
142# Create a FileCheck variable name based on an IR name.
143def get_value_name(var):
144  if var.isdigit():
145    var = 'TMP' + var
146  var = var.replace('.', '_')
147  var = var.replace('-', '_')
148  return var.upper()
149
150
151# Create a FileCheck variable from regex.
152def get_value_definition(var):
153  return '[[' + get_value_name(var) + ':%.*]]'
154
155
156# Use a FileCheck variable.
157def get_value_use(var):
158  return '[[' + get_value_name(var) + ']]'
159
160# Replace IR value defs and uses with FileCheck variables.
161def genericize_check_lines(lines, is_analyze):
162  # This gets called for each match that occurs in
163  # a line. We transform variables we haven't seen
164  # into defs, and variables we have seen into uses.
165  def transform_line_vars(match):
166    var = match.group(2)
167    if var in vars_seen:
168      rv = get_value_use(var)
169    else:
170      vars_seen.add(var)
171      rv = get_value_definition(var)
172    # re.sub replaces the entire regex match
173    # with whatever you return, so we have
174    # to make sure to hand it back everything
175    # including the commas and spaces.
176    return match.group(1) + rv + match.group(3)
177
178  vars_seen = set()
179  lines_with_def = []
180
181  for i, line in enumerate(lines):
182    # An IR variable named '%.' matches the FileCheck regex string.
183    line = line.replace('%.', '%dot')
184    # Ignore any comments, since the check lines will too.
185    scrubbed_line = SCRUB_IR_COMMENT_RE.sub(r'', line)
186    if is_analyze:
187      lines[i] = scrubbed_line
188    else:
189      lines[i] = IR_VALUE_RE.sub(transform_line_vars, scrubbed_line)
190  return lines
191
192
193def add_checks(output_lines, comment_marker, prefix_list, func_dict, func_name, check_label_format, is_asm, is_analyze):
194  printed_prefixes = []
195  for p in prefix_list:
196    checkprefixes = p[0]
197    for checkprefix in checkprefixes:
198      if checkprefix in printed_prefixes:
199        break
200      # TODO func_dict[checkprefix] may be None, '' or not exist.
201      # Fix the call sites.
202      if func_name not in func_dict[checkprefix] or not func_dict[checkprefix][func_name]:
203        continue
204
205      # Add some space between different check prefixes, but not after the last
206      # check line (before the test code).
207      if is_asm:
208        if len(printed_prefixes) != 0:
209          output_lines.append(comment_marker)
210
211      printed_prefixes.append(checkprefix)
212      output_lines.append(check_label_format % (checkprefix, func_name))
213      func_body = str(func_dict[checkprefix][func_name]).splitlines()
214
215      # For ASM output, just emit the check lines.
216      if is_asm:
217        output_lines.append('%s %s:       %s' % (comment_marker, checkprefix, func_body[0]))
218        for func_line in func_body[1:]:
219          output_lines.append('%s %s-NEXT:  %s' % (comment_marker, checkprefix, func_line))
220        break
221
222      # For IR output, change all defs to FileCheck variables, so we're immune
223      # to variable naming fashions.
224      func_body = genericize_check_lines(func_body, is_analyze)
225
226      # This could be selectively enabled with an optional invocation argument.
227      # Disabled for now: better to check everything. Be safe rather than sorry.
228
229      # Handle the first line of the function body as a special case because
230      # it's often just noise (a useless asm comment or entry label).
231      #if func_body[0].startswith("#") or func_body[0].startswith("entry:"):
232      #  is_blank_line = True
233      #else:
234      #  output_lines.append('%s %s:       %s' % (comment_marker, checkprefix, func_body[0]))
235      #  is_blank_line = False
236
237      is_blank_line = False
238
239      for func_line in func_body:
240        if func_line.strip() == '':
241          is_blank_line = True
242          continue
243        # Do not waste time checking IR comments.
244        func_line = SCRUB_IR_COMMENT_RE.sub(r'', func_line)
245
246        # Skip blank lines instead of checking them.
247        if is_blank_line:
248          output_lines.append('{} {}:       {}'.format(
249              comment_marker, checkprefix, func_line))
250        else:
251          output_lines.append('{} {}-NEXT:  {}'.format(
252              comment_marker, checkprefix, func_line))
253        is_blank_line = False
254
255      # Add space between different check prefixes and also before the first
256      # line of code in the test function.
257      output_lines.append(comment_marker)
258      break
259
260def add_ir_checks(output_lines, comment_marker, prefix_list, func_dict, func_name):
261  # Label format is based on IR string.
262  check_label_format = '{} %s-LABEL: @%s('.format(comment_marker)
263  add_checks(output_lines, comment_marker, prefix_list, func_dict, func_name, check_label_format, False, False)
264
265def add_analyze_checks(output_lines, comment_marker, prefix_list, func_dict, func_name):
266  check_label_format = '{} %s-LABEL: \'%s\''.format(comment_marker)
267  add_checks(output_lines, comment_marker, prefix_list, func_dict, func_name, check_label_format, False, True)
268
269
270def check_prefix(prefix):
271  if not PREFIX_RE.match(prefix):
272        hint = ""
273        if ',' in prefix:
274          hint = " Did you mean '--check-prefixes=" + prefix + "'?"
275        print(("WARNING: Supplied prefix '%s' is invalid. Prefix must contain only alphanumeric characters, hyphens and underscores." + hint) %
276              (prefix), file=sys.stderr)
277
278
279def verify_filecheck_prefixes(fc_cmd):
280  fc_cmd_parts = fc_cmd.split()
281  for part in fc_cmd_parts:
282    if "check-prefix=" in part:
283      prefix = part.split('=', 1)[1]
284      check_prefix(prefix)
285    elif "check-prefixes=" in part:
286      prefixes = part.split('=', 1)[1].split(',')
287      for prefix in prefixes:
288        check_prefix(prefix)
289        if prefixes.count(prefix) > 1:
290          print("WARNING: Supplied prefix '%s' is not unique in the prefix list." %
291                (prefix,), file=sys.stderr)
292