utils/update_llc_test_checks.py

   1 #!/usr/bin/env python
   2
   3 """A test case update script.
   4
   5 This script is a utility to update LLVM 'llc' based test cases with new
   6 FileCheck patterns. It can either update all of the tests in the file or
   7 a single test function.
   8 """
   9
  10 from __future__ import print_function
  11
  12 import argparse
  13 import glob
  14 import os         # Used to advertise this file's name ("autogenerated_note").
  15 import string
  16 import subprocess
  17 import sys
  18 import re
  19
  20 from UpdateTestChecks import asm, common
  21
  22 ADVERT = ' NOTE: Assertions have been autogenerated by '
  23
  24
  25 def main():
  26   parser = argparse.ArgumentParser(description=__doc__)
  27   parser.add_argument('-v', '--verbose', action='store_true',
  28                       help='Show verbose output')
  29   parser.add_argument('--llc-binary', default='llc',
  30                       help='The "llc" binary to use to generate the test case')
  31   parser.add_argument(
  32       '--function', help='The function in the test file to update')
  33   parser.add_argument(
  34       '--extra_scrub', action='store_true',
  35       help='Always use additional regex to further reduce diffs between various subtargets')
  36   parser.add_argument(
  37       '--x86_scrub_rip', action='store_true', default=True,
  38       help='Use more regex for x86 matching to reduce diffs between various subtargets')
  39   parser.add_argument(
  40       '--no_x86_scrub_rip', action='store_false', dest='x86_scrub_rip')
  41   parser.add_argument('-u', '--update-only', action='store_true',
  42                       help='Only update test if it was already autogened')
  43   parser.add_argument('tests', nargs='+')
  44   args = parser.parse_args()
  45
  46   script_name = os.path.basename(__file__)
  47
  48   test_paths = [test for pattern in args.tests for test in glob.glob(pattern)]
  49   for test in test_paths:
  50     if args.verbose:
  51       print('Scanning for RUN lines in test file: %s' % (test,), file=sys.stderr)
  52     with open(test) as f:
  53       input_lines = [l.rstrip() for l in f]
  54
  55     first_line = input_lines[0] if input_lines else ""
  56     if 'autogenerated' in first_line and script_name not in first_line:
  57       common.warn("Skipping test which wasn't autogenerated by " + script_name, test)
  58       continue
  59
  60     if args.update_only:
  61       if not first_line or 'autogenerated' not in first_line:
  62         common.warn("Skipping test which isn't autogenerated: " + test)
  63         continue
  64
  65     triple_in_ir = None
  66     for l in input_lines:
  67       m = common.TRIPLE_IR_RE.match(l)
  68       if m:
  69         triple_in_ir = m.groups()[0]
  70         break
  71
  72     raw_lines = [m.group(1)
  73                  for m in [common.RUN_LINE_RE.match(l) for l in input_lines] if m]
  74     run_lines = [raw_lines[0]] if len(raw_lines) > 0 else []
  75     for l in raw_lines[1:]:
  76       if run_lines[-1].endswith("\\"):
  77         run_lines[-1] = run_lines[-1].rstrip("\\") + " " + l
  78       else:
  79         run_lines.append(l)
  80
  81     if args.verbose:
  82       print('Found %d RUN lines:' % (len(run_lines),), file=sys.stderr)
  83       for l in run_lines:
  84         print('  RUN: ' + l, file=sys.stderr)
  85
  86     run_list = []
  87     for l in run_lines:
  88       if '|' not in l:
  89         common.warn('Skipping unparseable RUN line: ' + l)
  90         continue
  91
  92       commands = [cmd.strip() for cmd in l.split('|', 1)]
  93       llc_cmd = commands[0]
  94       llc_tool = llc_cmd.split(' ')[0]
  95
  96       triple_in_cmd = None
  97       m = common.TRIPLE_ARG_RE.search(llc_cmd)
  98       if m:
  99         triple_in_cmd = m.groups()[0]
 100
 101       march_in_cmd = None
 102       m = common.MARCH_ARG_RE.search(llc_cmd)
 103       if m:
 104         march_in_cmd = m.groups()[0]
 105
 106       filecheck_cmd = ''
 107       if len(commands) > 1:
 108         filecheck_cmd = commands[1]
 109       common.verify_filecheck_prefixes(filecheck_cmd)
 110       if llc_tool != 'llc':
 111         common.warn('Skipping non-llc RUN line: ' + l)
 112         continue
 113
 114       if not filecheck_cmd.startswith('FileCheck '):
 115         common.warn('Skipping non-FileChecked RUN line: ' + l)
 116         continue
 117
 118       llc_cmd_args = llc_cmd[len(llc_tool):].strip()
 119       llc_cmd_args = llc_cmd_args.replace('< %s', '').replace('%s', '').strip()
 120       if test.endswith('.mir'):
 121         llc_cmd_args += ' -x mir'
 122         comment_sym = '#'
 123         check_indent = '  '
 124       else:
 125         comment_sym = ';'
 126         check_indent = ''
 127
 128       check_prefixes = [item for m in common.CHECK_PREFIX_RE.finditer(filecheck_cmd)
 129                                for item in m.group(1).split(',')]
 130       if not check_prefixes:
 131         check_prefixes = ['CHECK']
 132
 133       # FIXME: We should use multiple check prefixes to common check lines. For
 134       # now, we just ignore all but the last.
 135       run_list.append((check_prefixes, llc_cmd_args, triple_in_cmd, march_in_cmd))
 136
 137     autogenerated_note = (comment_sym + ADVERT + 'utils/' + script_name)
 138
 139     func_dict = {}
 140     for p in run_list:
 141       prefixes = p[0]
 142       for prefix in prefixes:
 143         func_dict.update({prefix: dict()})
 144     for prefixes, llc_args, triple_in_cmd, march_in_cmd in run_list:
 145       if args.verbose:
 146         print('Extracted LLC cmd: ' + llc_tool + ' ' + llc_args, file=sys.stderr)
 147         print('Extracted FileCheck prefixes: ' + str(prefixes), file=sys.stderr)
 148
 149       raw_tool_output = common.invoke_tool(args.llc_binary, llc_args, test)
 150       triple = triple_in_cmd or triple_in_ir
 151       if not triple:
 152         triple = asm.get_triple_from_march(march_in_cmd)
 153
 154       asm.build_function_body_dictionary_for_triple(args, raw_tool_output,
 155           triple, prefixes, func_dict)
 156
 157     is_in_function = False
 158     is_in_function_start = False
 159     func_name = None
 160     prefix_set = set([prefix for p in run_list for prefix in p[0]])
 161     if args.verbose:
 162       print('Rewriting FileCheck prefixes: %s' % (prefix_set,), file=sys.stderr)
 163     output_lines = []
 164     output_lines.append(autogenerated_note)
 165
 166     for input_line in input_lines:
 167       if is_in_function_start:
 168         if input_line == '':
 169           continue
 170         if input_line.lstrip().startswith(';'):
 171           m = common.CHECK_RE.match(input_line)
 172           if not m or m.group(1) not in prefix_set:
 173             output_lines.append(input_line)
 174             continue
 175
 176         # Print out the various check lines here.
 177         asm.add_asm_checks(output_lines, check_indent + ';', run_list, func_dict, func_name)
 178         is_in_function_start = False
 179
 180       if is_in_function:
 181         if common.should_add_line_to_output(input_line, prefix_set):
 182           # This input line of the function body will go as-is into the output.
 183           output_lines.append(input_line)
 184         else:
 185           continue
 186         if input_line.strip() == '}':
 187           is_in_function = False
 188         continue
 189
 190       # Discard any previous script advertising.
 191       if input_line.startswith(comment_sym + ADVERT):
 192         continue
 193
 194       # If it's outside a function, it just gets copied to the output.
 195       output_lines.append(input_line)
 196
 197       m = common.IR_FUNCTION_RE.match(input_line)
 198       if not m:
 199         continue
 200       func_name = m.group(1)
 201       if args.function is not None and func_name != args.function:
 202         # When filtering on a specific function, skip all others.
 203         continue
 204       is_in_function = is_in_function_start = True
 205
 206     if args.verbose:
 207       print('Writing %d lines to %s...' % (len(output_lines), test), file=sys.stderr)
 208
 209     with open(test, 'wb') as f:
 210       f.writelines(['{}\n'.format(l).encode('utf-8') for l in output_lines])
 211
 212
 213 if __name__ == '__main__':
 214   main()