1#!/usr/bin/env python2.7 2 3"""A test case update script. 4 5This script is a utility to update LLVM X86 'llc' based test cases with new 6FileCheck patterns. It can either update all of the tests in the file or 7a single test function. 8""" 9 10import argparse 11import itertools 12import os # Used to advertise this file's name ("autogenerated_note"). 13import string 14import subprocess 15import sys 16import tempfile 17import re 18 19# Invoke the tool that is being tested. 20def llc(args, cmd_args, ir): 21 with open(ir) as ir_file: 22 stdout = subprocess.check_output(args.llc_binary + ' ' + cmd_args, 23 shell=True, stdin=ir_file) 24 # Fix line endings to unix CR style. 25 stdout = stdout.replace('\r\n', '\n') 26 return stdout 27 28 29# RegEx: this is where the magic happens. 30 31SCRUB_WHITESPACE_RE = re.compile(r'(?!^(| \w))[ \t]+', flags=re.M) 32SCRUB_TRAILING_WHITESPACE_RE = re.compile(r'[ \t]+$', flags=re.M) 33SCRUB_X86_SHUFFLES_RE = ( 34 re.compile( 35 r'^(\s*\w+) [^#\n]+#+ ((?:[xyz]mm\d+|mem)( \{%k\d+\}( \{z\})?)? = .*)$', 36 flags=re.M)) 37SCRUB_X86_SP_RE = re.compile(r'\d+\(%(esp|rsp)\)') 38SCRUB_X86_RIP_RE = re.compile(r'[.\w]+\(%rip\)') 39SCRUB_X86_LCP_RE = re.compile(r'\.LCPI[0-9]+_[0-9]+') 40SCRUB_KILL_COMMENT_RE = re.compile(r'^ *#+ +kill:.*\n') 41 42RUN_LINE_RE = re.compile('^\s*;\s*RUN:\s*(.*)$') 43IR_FUNCTION_RE = re.compile('^\s*define\s+(?:internal\s+)?[^@]*@(\w+)\s*\(') 44ASM_FUNCTION_RE = re.compile( 45 r'^_?(?P<func>[^:]+):[ \t]*#+[ \t]*@(?P=func)\n[^:]*?' 46 r'(?P<body>^##?[ \t]+[^:]+:.*?)\s*' 47 r'^\s*(?:[^:\n]+?:\s*\n\s*\.size|\.cfi_endproc|\.globl|\.comm|\.(?:sub)?section)', 48 flags=(re.M | re.S)) 49CHECK_PREFIX_RE = re.compile('--check-prefix=(\S+)') 50CHECK_RE = re.compile(r'^\s*;\s*([^:]+?)(?:-NEXT|-NOT|-DAG|-LABEL)?:') 51 52 53def scrub_asm(asm): 54 # Scrub runs of whitespace out of the assembly, but leave the leading 55 # whitespace in place. 56 asm = SCRUB_WHITESPACE_RE.sub(r' ', asm) 57 # Expand the tabs used for indentation. 58 asm = string.expandtabs(asm, 2) 59 # Detect shuffle asm comments and hide the operands in favor of the comments. 60 asm = SCRUB_X86_SHUFFLES_RE.sub(r'\1 {{.*#+}} \2', asm) 61 # Generically match the stack offset of a memory operand. 62 asm = SCRUB_X86_SP_RE.sub(r'{{[0-9]+}}(%\1)', asm) 63 # Generically match a RIP-relative memory operand. 64 asm = SCRUB_X86_RIP_RE.sub(r'{{.*}}(%rip)', asm) 65 # Generically match a LCP symbol. 66 asm = SCRUB_X86_LCP_RE.sub(r'{{\.LCPI.*}}', asm) 67 # Strip kill operands inserted into the asm. 68 asm = SCRUB_KILL_COMMENT_RE.sub('', asm) 69 # Strip trailing whitespace. 70 asm = SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm) 71 return asm 72 73 74# Build up a dictionary of all the function bodies. 75def build_function_body_dictionary(raw_tool_output, prefixes, func_dict, verbose): 76 for m in ASM_FUNCTION_RE.finditer(raw_tool_output): 77 if not m: 78 continue 79 func = m.group('func') 80 scrubbed_body = scrub_asm(m.group('body')) 81 if func.startswith('stress'): 82 # We only use the last line of the function body for stress tests. 83 scrubbed_body = '\n'.join(scrubbed_body.splitlines()[-1:]) 84 if verbose: 85 print >>sys.stderr, 'Processing function: ' + func 86 for l in scrubbed_body.splitlines(): 87 print >>sys.stderr, ' ' + l 88 for prefix in prefixes: 89 if func in func_dict[prefix] and func_dict[prefix][func] != scrubbed_body: 90 if prefix == prefixes[-1]: 91 print >>sys.stderr, ('WARNING: Found conflicting asm under the ' 92 'same prefix: %r!' % (prefix,)) 93 else: 94 func_dict[prefix][func] = None 95 continue 96 97 func_dict[prefix][func] = scrubbed_body 98 99 100def add_checks(output_lines, prefix_list, func_dict, func_name): 101 printed_prefixes = [] 102 for checkprefixes, _ in prefix_list: 103 for checkprefix in checkprefixes: 104 if checkprefix in printed_prefixes: 105 break 106 if not func_dict[checkprefix][func_name]: 107 continue 108 # Add some space between different check prefixes. 109 if len(printed_prefixes) != 0: 110 output_lines.append(';') 111 printed_prefixes.append(checkprefix) 112 output_lines.append('; %s-LABEL: %s:' % (checkprefix, func_name)) 113 func_body = func_dict[checkprefix][func_name].splitlines() 114 output_lines.append('; %s: %s' % (checkprefix, func_body[0])) 115 for func_line in func_body[1:]: 116 output_lines.append('; %s-NEXT: %s' % (checkprefix, func_line)) 117 # Add space between different check prefixes and the first line of code. 118 # output_lines.append(';') 119 break 120 return output_lines 121 122 123def should_add_line_to_output(input_line, prefix_set): 124 # Skip any blank comment lines in the IR. 125 if input_line.strip() == ';': 126 return False 127 # Skip any blank lines in the IR. 128 #if input_line.strip() == '': 129 # return False 130 # And skip any CHECK lines. We're building our own. 131 m = CHECK_RE.match(input_line) 132 if m and m.group(1) in prefix_set: 133 return False 134 135 return True 136 137 138def main(): 139 parser = argparse.ArgumentParser(description=__doc__) 140 parser.add_argument('-v', '--verbose', action='store_true', 141 help='Show verbose output') 142 parser.add_argument('--llc-binary', default='llc', 143 help='The "llc" binary to use to generate the test case') 144 parser.add_argument( 145 '--function', help='The function in the test file to update') 146 parser.add_argument('tests', nargs='+') 147 args = parser.parse_args() 148 149 autogenerated_note = ('; NOTE: Assertions have been autogenerated by ' 150 'utils/' + os.path.basename(__file__)) 151 152 for test in args.tests: 153 if args.verbose: 154 print >>sys.stderr, 'Scanning for RUN lines in test file: %s' % (test,) 155 with open(test) as f: 156 input_lines = [l.rstrip() for l in f] 157 158 run_lines = [m.group(1) 159 for m in [RUN_LINE_RE.match(l) for l in input_lines] if m] 160 if args.verbose: 161 print >>sys.stderr, 'Found %d RUN lines:' % (len(run_lines),) 162 for l in run_lines: 163 print >>sys.stderr, ' RUN: ' + l 164 165 prefix_list = [] 166 for l in run_lines: 167 (llc_cmd, filecheck_cmd) = tuple([cmd.strip() for cmd in l.split('|', 1)]) 168 if not llc_cmd.startswith('llc '): 169 print >>sys.stderr, 'WARNING: Skipping non-llc RUN line: ' + l 170 continue 171 172 if not filecheck_cmd.startswith('FileCheck '): 173 print >>sys.stderr, 'WARNING: Skipping non-FileChecked RUN line: ' + l 174 continue 175 176 llc_cmd_args = llc_cmd[len('llc'):].strip() 177 llc_cmd_args = llc_cmd_args.replace('< %s', '').replace('%s', '').strip() 178 179 check_prefixes = [m.group(1) 180 for m in CHECK_PREFIX_RE.finditer(filecheck_cmd)] 181 if not check_prefixes: 182 check_prefixes = ['CHECK'] 183 184 # FIXME: We should use multiple check prefixes to common check lines. For 185 # now, we just ignore all but the last. 186 prefix_list.append((check_prefixes, llc_cmd_args)) 187 188 func_dict = {} 189 for prefixes, _ in prefix_list: 190 for prefix in prefixes: 191 func_dict.update({prefix: dict()}) 192 for prefixes, llc_args in prefix_list: 193 if args.verbose: 194 print >>sys.stderr, 'Extracted LLC cmd: llc ' + llc_args 195 print >>sys.stderr, 'Extracted FileCheck prefixes: ' + str(prefixes) 196 197 raw_tool_output = llc(args, llc_args, test) 198 build_function_body_dictionary(raw_tool_output, prefixes, func_dict, args.verbose) 199 200 is_in_function = False 201 is_in_function_start = False 202 prefix_set = set([prefix for prefixes, _ in prefix_list for prefix in prefixes]) 203 if args.verbose: 204 print >>sys.stderr, 'Rewriting FileCheck prefixes: %s' % (prefix_set,) 205 output_lines = [] 206 output_lines.append(autogenerated_note) 207 208 for input_line in input_lines: 209 if is_in_function_start: 210 if input_line == '': 211 continue 212 if input_line.lstrip().startswith(';'): 213 m = CHECK_RE.match(input_line) 214 if not m or m.group(1) not in prefix_set: 215 output_lines.append(input_line) 216 continue 217 218 # Print out the various check lines here. 219 output_lines = add_checks(output_lines, prefix_list, func_dict, name) 220 is_in_function_start = False 221 222 if is_in_function: 223 if should_add_line_to_output(input_line, prefix_set) == True: 224 # This input line of the function body will go as-is into the output. 225 output_lines.append(input_line) 226 else: 227 continue 228 if input_line.strip() == '}': 229 is_in_function = False 230 continue 231 232 if input_line == autogenerated_note: 233 continue 234 235 # If it's outside a function, it just gets copied to the output. 236 output_lines.append(input_line) 237 238 m = IR_FUNCTION_RE.match(input_line) 239 if not m: 240 continue 241 name = m.group(1) 242 if args.function is not None and name != args.function: 243 # When filtering on a specific function, skip all others. 244 continue 245 is_in_function = is_in_function_start = True 246 247 if args.verbose: 248 print>>sys.stderr, 'Writing %d lines to %s...' % (len(output_lines), test) 249 250 with open(test, 'wb') as f: 251 f.writelines([l + '\n' for l in output_lines]) 252 253 254if __name__ == '__main__': 255 main() 256