1 #!/usr/bin/env python2.7 2 3 """A test case update script. 4 5 This script is a utility to update LLVM X86 '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 import argparse 11 import itertools 12 import string 13 import subprocess 14 import sys 15 import tempfile 16 import re 17 18 19 def llc(args, cmd_args, ir): 20 with open(ir) as ir_file: 21 stdout = subprocess.check_output(args.llc_binary + ' ' + cmd_args, 22 shell=True, stdin=ir_file) 23 return stdout 24 25 26 ASM_SCRUB_WHITESPACE_RE = re.compile(r'(?!^(| \w))[ \t]+', flags=re.M) 27 ASM_SCRUB_TRAILING_WHITESPACE_RE = re.compile(r'[ \t]+$', flags=re.M) 28 ASM_SCRUB_SHUFFLES_RE = ( 29 re.compile( 30 r'^(\s*\w+) [^#\n]+#+ ((?:[xyz]mm\d+|mem) = .*)$', 31 flags=re.M)) 32 ASM_SCRUB_SP_RE = re.compile(r'\d+\(%(esp|rsp)\)') 33 ASM_SCRUB_RIP_RE = re.compile(r'[.\w]+\(%rip\)') 34 ASM_SCRUB_KILL_COMMENT_RE = re.compile(r'^ *#+ +kill:.*\n') 35 36 37 def scrub_asm(asm): 38 # Scrub runs of whitespace out of the assembly, but leave the leading 39 # whitespace in place. 40 asm = ASM_SCRUB_WHITESPACE_RE.sub(r' ', asm) 41 # Expand the tabs used for indentation. 42 asm = string.expandtabs(asm, 2) 43 # Detect shuffle asm comments and hide the operands in favor of the comments. 44 asm = ASM_SCRUB_SHUFFLES_RE.sub(r'\1 {{.*#+}} \2', asm) 45 # Generically match the stack offset of a memory operand. 46 asm = ASM_SCRUB_SP_RE.sub(r'{{[0-9]+}}(%\1)', asm) 47 # Generically match a RIP-relative memory operand. 48 asm = ASM_SCRUB_RIP_RE.sub(r'{{.*}}(%rip)', asm) 49 # Strip kill operands inserted into the asm. 50 asm = ASM_SCRUB_KILL_COMMENT_RE.sub('', asm) 51 # Strip trailing whitespace. 52 asm = ASM_SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm) 53 return asm 54 55 56 def main(): 57 parser = argparse.ArgumentParser(description=__doc__) 58 parser.add_argument('-v', '--verbose', action='store_true', 59 help='Show verbose output') 60 parser.add_argument('--llc-binary', default='llc', 61 help='The "llc" binary to use to generate the test case') 62 parser.add_argument( 63 '--function', help='The function in the test file to update') 64 parser.add_argument('tests', nargs='+') 65 args = parser.parse_args() 66 67 run_line_re = re.compile('^\s*;\s*RUN:\s*(.*)$') 68 ir_function_re = re.compile('^\s*define\s+(?:internal\s+)?[^@]*@(\w+)\s*\(') 69 asm_function_re = re.compile( 70 r'^_?(?P<f>[^:]+):[ \t]*#+[ \t]*@(?P=f)\n[^:]*?' 71 r'(?P<body>^##?[ \t]+[^:]+:.*?)\s*' 72 r'^\s*(?:[^:\n]+?:\s*\n\s*\.size|\.cfi_endproc|\.globl|\.comm|\.(?:sub)?section)', 73 flags=(re.M | re.S)) 74 check_prefix_re = re.compile('--check-prefix=(\S+)') 75 check_re = re.compile(r'^\s*;\s*([^:]+?)(?:-NEXT|-NOT|-DAG|-LABEL)?:') 76 autogenerated_note = ('; NOTE: Assertions have been autogenerated by ' 77 'utils/update_llc_test_checks.py') 78 79 for test in args.tests: 80 if args.verbose: 81 print >>sys.stderr, 'Scanning for RUN lines in test file: %s' % (test,) 82 with open(test) as f: 83 test_lines = [l.rstrip() for l in f] 84 85 run_lines = [m.group(1) 86 for m in [run_line_re.match(l) for l in test_lines] if m] 87 if args.verbose: 88 print >>sys.stderr, 'Found %d RUN lines:' % (len(run_lines),) 89 for l in run_lines: 90 print >>sys.stderr, ' RUN: ' + l 91 92 checks = [] 93 for l in run_lines: 94 (llc_cmd, filecheck_cmd) = tuple([cmd.strip() for cmd in l.split('|', 1)]) 95 if not llc_cmd.startswith('llc '): 96 print >>sys.stderr, 'WARNING: Skipping non-llc RUN line: ' + l 97 continue 98 99 if not filecheck_cmd.startswith('FileCheck '): 100 print >>sys.stderr, 'WARNING: Skipping non-FileChecked RUN line: ' + l 101 continue 102 103 llc_cmd_args = llc_cmd[len('llc'):].strip() 104 llc_cmd_args = llc_cmd_args.replace('< %s', '').replace('%s', '').strip() 105 106 check_prefixes = [m.group(1) 107 for m in check_prefix_re.finditer(filecheck_cmd)] 108 if not check_prefixes: 109 check_prefixes = ['CHECK'] 110 111 # FIXME: We should use multiple check prefixes to common check lines. For 112 # now, we just ignore all but the last. 113 checks.append((check_prefixes, llc_cmd_args)) 114 115 asm = {} 116 for prefixes, _ in checks: 117 for prefix in prefixes: 118 asm.update({prefix: dict()}) 119 for prefixes, llc_args in checks: 120 if args.verbose: 121 print >>sys.stderr, 'Extracted LLC cmd: llc ' + llc_args 122 print >>sys.stderr, 'Extracted FileCheck prefixes: ' + str(prefixes) 123 raw_asm = llc(args, llc_args, test) 124 # Build up a dictionary of all the function bodies. 125 for m in asm_function_re.finditer(raw_asm): 126 if not m: 127 continue 128 f = m.group('f') 129 f_asm = scrub_asm(m.group('body')) 130 if f.startswith('stress'): 131 # We only use the last line of the asm for stress tests. 132 f_asm = '\n'.join(f_asm.splitlines()[-1:]) 133 if args.verbose: 134 print >>sys.stderr, 'Processing asm for function: ' + f 135 for l in f_asm.splitlines(): 136 print >>sys.stderr, ' ' + l 137 for prefix in prefixes: 138 if f in asm[prefix] and asm[prefix][f] != f_asm: 139 if prefix == prefixes[-1]: 140 print >>sys.stderr, ('WARNING: Found conflicting asm under the ' 141 'same prefix: %r!' % (prefix,)) 142 else: 143 asm[prefix][f] = None 144 continue 145 146 asm[prefix][f] = f_asm 147 148 is_in_function = False 149 is_in_function_start = False 150 prefix_set = set([prefix for prefixes, _ in checks for prefix in prefixes]) 151 if args.verbose: 152 print >>sys.stderr, 'Rewriting FileCheck prefixes: %s' % (prefix_set,) 153 fixed_lines = [] 154 fixed_lines.append(autogenerated_note) 155 156 for l in test_lines: 157 if is_in_function_start: 158 if l.lstrip().startswith(';'): 159 m = check_re.match(l) 160 if not m or m.group(1) not in prefix_set: 161 fixed_lines.append(l) 162 continue 163 164 # Print out the various check lines here 165 printed_prefixes = [] 166 for prefixes, _ in checks: 167 for prefix in prefixes: 168 if prefix in printed_prefixes: 169 break 170 if not asm[prefix][name]: 171 continue 172 if len(printed_prefixes) != 0: 173 fixed_lines.append(';') 174 printed_prefixes.append(prefix) 175 fixed_lines.append('; %s-LABEL: %s:' % (prefix, name)) 176 asm_lines = asm[prefix][name].splitlines() 177 fixed_lines.append('; %s: %s' % (prefix, asm_lines[0])) 178 for asm_line in asm_lines[1:]: 179 fixed_lines.append('; %s-NEXT: %s' % (prefix, asm_line)) 180 break 181 is_in_function_start = False 182 183 if is_in_function: 184 # Skip any blank comment lines in the IR. 185 if l.strip() == ';': 186 continue 187 # And skip any CHECK lines. We'll build our own. 188 m = check_re.match(l) 189 if m and m.group(1) in prefix_set: 190 continue 191 # Collect the remaining lines in the function body and look for the end 192 # of the function. 193 fixed_lines.append(l) 194 if l.strip() == '}': 195 is_in_function = False 196 continue 197 198 if l == autogenerated_note: 199 continue 200 fixed_lines.append(l) 201 202 m = ir_function_re.match(l) 203 if not m: 204 continue 205 name = m.group(1) 206 if args.function is not None and name != args.function: 207 # When filtering on a specific function, skip all others. 208 continue 209 is_in_function = is_in_function_start = True 210 211 if args.verbose: 212 print>>sys.stderr, 'Writing %d fixed lines to %s...' % ( 213 len(fixed_lines), test) 214 with open(test, 'w') as f: 215 f.writelines([l + '\n' for l in fixed_lines]) 216 217 218 if __name__ == '__main__': 219 main() 220
