1from __future__ import print_function
2import re
3import string
4import subprocess
5import sys
6
7if sys.version_info[0] > 2:
8  class string:
9    expandtabs = str.expandtabs
10else:
11  import string
12
13##### Common utilities for update_*test_checks.py
14
15def should_add_line_to_output(input_line, prefix_set):
16  # Skip any blank comment lines in the IR.
17  if input_line.strip() == ';':
18    return False
19  # Skip any blank lines in the IR.
20  #if input_line.strip() == '':
21  #  return False
22  # And skip any CHECK lines. We're building our own.
23  m = CHECK_RE.match(input_line)
24  if m and m.group(1) in prefix_set:
25    return False
26
27  return True
28
29# Invoke the tool that is being tested.
30def invoke_tool(exe, cmd_args, ir):
31  with open(ir) as ir_file:
32    # TODO Remove the str form which is used by update_test_checks.py and
33    # update_llc_test_checks.py
34    # The safer list form is used by update_cc_test_checks.py
35    if isinstance(cmd_args, list):
36      stdout = subprocess.check_output([exe] + cmd_args, stdin=ir_file)
37    else:
38      stdout = subprocess.check_output(exe + ' ' + cmd_args,
39                                       shell=True, stdin=ir_file)
40    if sys.version_info[0] > 2:
41      stdout = stdout.decode()
42  # Fix line endings to unix CR style.
43  return stdout.replace('\r\n', '\n')
44
45##### LLVM IR parser
46
47RUN_LINE_RE = re.compile('^\s*[;#]\s*RUN:\s*(.*)$')
48CHECK_PREFIX_RE = re.compile('--?check-prefix(?:es)?[= ](\S+)')
49CHECK_RE = re.compile(r'^\s*[;#]\s*([^:]+?)(?:-NEXT|-NOT|-DAG|-LABEL)?:')
50
51OPT_FUNCTION_RE = re.compile(
52    r'^\s*define\s+(?:internal\s+)?[^@]*@(?P<func>[\w-]+?)\s*\('
53    r'(\s+)?[^)]*[^{]*\{\n(?P<body>.*?)^\}$',
54    flags=(re.M | re.S))
55
56IR_FUNCTION_RE = re.compile('^\s*define\s+(?:internal\s+)?[^@]*@(\w+)\s*\(')
57TRIPLE_IR_RE = re.compile(r'^\s*target\s+triple\s*=\s*"([^"]+)"$')
58TRIPLE_ARG_RE = re.compile(r'-mtriple[= ]([^ ]+)')
59MARCH_ARG_RE = re.compile(r'-march[= ]([^ ]+)')
60
61SCRUB_LEADING_WHITESPACE_RE = re.compile(r'^(\s+)')
62SCRUB_WHITESPACE_RE = re.compile(r'(?!^(|  \w))[ \t]+', flags=re.M)
63SCRUB_TRAILING_WHITESPACE_RE = re.compile(r'[ \t]+$', flags=re.M)
64SCRUB_KILL_COMMENT_RE = re.compile(r'^ *#+ +kill:.*\n')
65SCRUB_LOOP_COMMENT_RE = re.compile(
66    r'# =>This Inner Loop Header:.*|# in Loop:.*', flags=re.M)
67
68def scrub_body(body):
69  # Scrub runs of whitespace out of the assembly, but leave the leading
70  # whitespace in place.
71  body = SCRUB_WHITESPACE_RE.sub(r' ', body)
72  # Expand the tabs used for indentation.
73  body = string.expandtabs(body, 2)
74  # Strip trailing whitespace.
75  body = SCRUB_TRAILING_WHITESPACE_RE.sub(r'', body)
76  return body
77
78# Build up a dictionary of all the function bodies.
79def build_function_body_dictionary(function_re, scrubber, scrubber_args, raw_tool_output, prefixes, func_dict, verbose):
80  for m in function_re.finditer(raw_tool_output):
81    if not m:
82      continue
83    func = m.group('func')
84    scrubbed_body = scrubber(m.group('body'), *scrubber_args)
85    if func.startswith('stress'):
86      # We only use the last line of the function body for stress tests.
87      scrubbed_body = '\n'.join(scrubbed_body.splitlines()[-1:])
88    if verbose:
89      print('Processing function: ' + func, file=sys.stderr)
90      for l in scrubbed_body.splitlines():
91        print('  ' + l, file=sys.stderr)
92    for prefix in prefixes:
93      if func in func_dict[prefix] and func_dict[prefix][func] != scrubbed_body:
94        if prefix == prefixes[-1]:
95          print('WARNING: Found conflicting asm under the '
96                               'same prefix: %r!' % (prefix,), file=sys.stderr)
97        else:
98          func_dict[prefix][func] = None
99          continue
100
101      func_dict[prefix][func] = scrubbed_body
102
103##### Generator of LLVM IR CHECK lines
104
105SCRUB_IR_COMMENT_RE = re.compile(r'\s*;.*')
106
107# Match things that look at identifiers, but only if they are followed by
108# spaces, commas, paren, or end of the string
109IR_VALUE_RE = re.compile(r'(\s+)%([\w\.\-]+?)([,\s\(\)]|\Z)')
110
111# Create a FileCheck variable name based on an IR name.
112def get_value_name(var):
113  if var.isdigit():
114    var = 'TMP' + var
115  var = var.replace('.', '_')
116  var = var.replace('-', '_')
117  return var.upper()
118
119
120# Create a FileCheck variable from regex.
121def get_value_definition(var):
122  return '[[' + get_value_name(var) + ':%.*]]'
123
124
125# Use a FileCheck variable.
126def get_value_use(var):
127  return '[[' + get_value_name(var) + ']]'
128
129# Replace IR value defs and uses with FileCheck variables.
130def genericize_check_lines(lines):
131  # This gets called for each match that occurs in
132  # a line. We transform variables we haven't seen
133  # into defs, and variables we have seen into uses.
134  def transform_line_vars(match):
135    var = match.group(2)
136    if var in vars_seen:
137      rv = get_value_use(var)
138    else:
139      vars_seen.add(var)
140      rv = get_value_definition(var)
141    # re.sub replaces the entire regex match
142    # with whatever you return, so we have
143    # to make sure to hand it back everything
144    # including the commas and spaces.
145    return match.group(1) + rv + match.group(3)
146
147  vars_seen = set()
148  lines_with_def = []
149
150  for i, line in enumerate(lines):
151    # An IR variable named '%.' matches the FileCheck regex string.
152    line = line.replace('%.', '%dot')
153    # Ignore any comments, since the check lines will too.
154    scrubbed_line = SCRUB_IR_COMMENT_RE.sub(r'', line)
155    lines[i] =  IR_VALUE_RE.sub(transform_line_vars, scrubbed_line)
156  return lines
157
158
159def add_ir_checks(output_lines, comment_marker, prefix_list, func_dict, func_name):
160  # Label format is based on IR string.
161  check_label_format = '{} %s-LABEL: @%s('.format(comment_marker)
162
163  printed_prefixes = []
164  for p in prefix_list:
165    checkprefixes = p[0]
166    for checkprefix in checkprefixes:
167      if checkprefix in printed_prefixes:
168        break
169      if not func_dict[checkprefix][func_name]:
170        continue
171      # Add some space between different check prefixes, but not after the last
172      # check line (before the test code).
173      #if len(printed_prefixes) != 0:
174      #  output_lines.append(';')
175      printed_prefixes.append(checkprefix)
176      output_lines.append(check_label_format % (checkprefix, func_name))
177      func_body = func_dict[checkprefix][func_name].splitlines()
178
179      # For IR output, change all defs to FileCheck variables, so we're immune
180      # to variable naming fashions.
181      func_body = genericize_check_lines(func_body)
182
183      # This could be selectively enabled with an optional invocation argument.
184      # Disabled for now: better to check everything. Be safe rather than sorry.
185
186      # Handle the first line of the function body as a special case because
187      # it's often just noise (a useless asm comment or entry label).
188      #if func_body[0].startswith("#") or func_body[0].startswith("entry:"):
189      #  is_blank_line = True
190      #else:
191      #  output_lines.append('; %s:       %s' % (checkprefix, func_body[0]))
192      #  is_blank_line = False
193
194      is_blank_line = False
195
196      for func_line in func_body:
197        if func_line.strip() == '':
198          is_blank_line = True
199          continue
200        # Do not waste time checking IR comments.
201        func_line = SCRUB_IR_COMMENT_RE.sub(r'', func_line)
202
203        # Skip blank lines instead of checking them.
204        if is_blank_line == True:
205          output_lines.append('{} {}:       {}'.format(
206              comment_marker, checkprefix, func_line))
207        else:
208          output_lines.append('{} {}-NEXT:  {}'.format(
209              comment_marker, checkprefix, func_line))
210        is_blank_line = False
211
212      # Add space between different check prefixes and also before the first
213      # line of code in the test function.
214      output_lines.append(comment_marker)
215      break
216  return output_lines
217