Fangrui Song | 3823fc4 | 2018-02-02 16:41:07 +0000 | [diff] [blame] | 1 | from __future__ import print_function |
Fangrui Song | ee4e2e7 | 2018-01-30 00:40:05 +0000 | [diff] [blame] | 2 | import re |
Fangrui Song | 4f0f426 | 2018-02-10 05:01:33 +0000 | [diff] [blame^] | 3 | import string |
Fangrui Song | ee4e2e7 | 2018-01-30 00:40:05 +0000 | [diff] [blame] | 4 | import subprocess |
Fangrui Song | 3823fc4 | 2018-02-02 16:41:07 +0000 | [diff] [blame] | 5 | import sys |
Fangrui Song | ee4e2e7 | 2018-01-30 00:40:05 +0000 | [diff] [blame] | 6 | |
Fangrui Song | 4f0f426 | 2018-02-10 05:01:33 +0000 | [diff] [blame^] | 7 | if sys.version_info[0] > 2: |
| 8 | class string: |
| 9 | expandtabs = str.expandtabs |
| 10 | else: |
| 11 | import string |
Fangrui Song | ee4e2e7 | 2018-01-30 00:40:05 +0000 | [diff] [blame] | 12 | |
Fangrui Song | 4f0f426 | 2018-02-10 05:01:33 +0000 | [diff] [blame^] | 13 | ##### Common utilities for update_*test_checks.py |
Fangrui Song | ee4e2e7 | 2018-01-30 00:40:05 +0000 | [diff] [blame] | 14 | |
| 15 | def 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. |
| 30 | def invoke_tool(exe, cmd_args, ir): |
| 31 | with open(ir) as ir_file: |
| 32 | stdout = subprocess.check_output(exe + ' ' + cmd_args, |
| 33 | shell=True, stdin=ir_file) |
Fangrui Song | 3823fc4 | 2018-02-02 16:41:07 +0000 | [diff] [blame] | 34 | if sys.version_info[0] > 2: |
| 35 | stdout = stdout.decode() |
Fangrui Song | ee4e2e7 | 2018-01-30 00:40:05 +0000 | [diff] [blame] | 36 | # Fix line endings to unix CR style. |
Fangrui Song | 3823fc4 | 2018-02-02 16:41:07 +0000 | [diff] [blame] | 37 | return stdout.replace('\r\n', '\n') |
Fangrui Song | ee4e2e7 | 2018-01-30 00:40:05 +0000 | [diff] [blame] | 38 | |
Fangrui Song | 4f0f426 | 2018-02-10 05:01:33 +0000 | [diff] [blame^] | 39 | ##### LLVM IR parser |
| 40 | |
| 41 | RUN_LINE_RE = re.compile('^\s*;\s*RUN:\s*(.*)$') |
| 42 | CHECK_PREFIX_RE = re.compile('--?check-prefix(?:es)?=(\S+)') |
| 43 | CHECK_RE = re.compile(r'^\s*;\s*([^:]+?)(?:-NEXT|-NOT|-DAG|-LABEL)?:') |
| 44 | |
| 45 | OPT_FUNCTION_RE = re.compile( |
| 46 | r'^\s*define\s+(?:internal\s+)?[^@]*@(?P<func>[\w-]+?)\s*\(' |
| 47 | r'(\s+)?[^)]*[^{]*\{\n(?P<body>.*?)^\}$', |
| 48 | flags=(re.M | re.S)) |
| 49 | |
| 50 | IR_FUNCTION_RE = re.compile('^\s*define\s+(?:internal\s+)?[^@]*@(\w+)\s*\(') |
| 51 | TRIPLE_IR_RE = re.compile(r'^target\s+triple\s*=\s*"([^"]+)"$') |
| 52 | TRIPLE_ARG_RE = re.compile(r'-mtriple=([^ ]+)') |
| 53 | |
| 54 | SCRUB_LEADING_WHITESPACE_RE = re.compile(r'^(\s+)') |
| 55 | SCRUB_WHITESPACE_RE = re.compile(r'(?!^(| \w))[ \t]+', flags=re.M) |
| 56 | SCRUB_TRAILING_WHITESPACE_RE = re.compile(r'[ \t]+$', flags=re.M) |
| 57 | SCRUB_KILL_COMMENT_RE = re.compile(r'^ *#+ +kill:.*\n') |
| 58 | SCRUB_LOOP_COMMENT_RE = re.compile( |
| 59 | r'# =>This Inner Loop Header:.*|# in Loop:.*', flags=re.M) |
| 60 | |
| 61 | def scrub_body(body): |
| 62 | # Scrub runs of whitespace out of the assembly, but leave the leading |
| 63 | # whitespace in place. |
| 64 | body = SCRUB_WHITESPACE_RE.sub(r' ', body) |
| 65 | # Expand the tabs used for indentation. |
| 66 | body = string.expandtabs(body, 2) |
| 67 | # Strip trailing whitespace. |
| 68 | body = SCRUB_TRAILING_WHITESPACE_RE.sub(r'', body) |
| 69 | return body |
| 70 | |
Fangrui Song | ee4e2e7 | 2018-01-30 00:40:05 +0000 | [diff] [blame] | 71 | # Build up a dictionary of all the function bodies. |
| 72 | def build_function_body_dictionary(function_re, scrubber, scrubber_args, raw_tool_output, prefixes, func_dict, verbose): |
| 73 | for m in function_re.finditer(raw_tool_output): |
| 74 | if not m: |
| 75 | continue |
| 76 | func = m.group('func') |
| 77 | scrubbed_body = scrubber(m.group('body'), *scrubber_args) |
| 78 | if func.startswith('stress'): |
| 79 | # We only use the last line of the function body for stress tests. |
| 80 | scrubbed_body = '\n'.join(scrubbed_body.splitlines()[-1:]) |
| 81 | if verbose: |
Fangrui Song | 3823fc4 | 2018-02-02 16:41:07 +0000 | [diff] [blame] | 82 | print('Processing function: ' + func, file=sys.stderr) |
Fangrui Song | ee4e2e7 | 2018-01-30 00:40:05 +0000 | [diff] [blame] | 83 | for l in scrubbed_body.splitlines(): |
Fangrui Song | 3823fc4 | 2018-02-02 16:41:07 +0000 | [diff] [blame] | 84 | print(' ' + l, file=sys.stderr) |
Fangrui Song | ee4e2e7 | 2018-01-30 00:40:05 +0000 | [diff] [blame] | 85 | for prefix in prefixes: |
| 86 | if func in func_dict[prefix] and func_dict[prefix][func] != scrubbed_body: |
| 87 | if prefix == prefixes[-1]: |
Fangrui Song | 3823fc4 | 2018-02-02 16:41:07 +0000 | [diff] [blame] | 88 | print('WARNING: Found conflicting asm under the ' |
| 89 | 'same prefix: %r!' % (prefix,), file=sys.stderr) |
Fangrui Song | ee4e2e7 | 2018-01-30 00:40:05 +0000 | [diff] [blame] | 90 | else: |
| 91 | func_dict[prefix][func] = None |
| 92 | continue |
| 93 | |
| 94 | func_dict[prefix][func] = scrubbed_body |
Fangrui Song | 4f0f426 | 2018-02-10 05:01:33 +0000 | [diff] [blame^] | 95 | |
| 96 | ##### Generator of LLVM IR CHECK lines |
| 97 | |
| 98 | SCRUB_IR_COMMENT_RE = re.compile(r'\s*;.*') |
| 99 | |
| 100 | # Match things that look at identifiers, but only if they are followed by |
| 101 | # spaces, commas, paren, or end of the string |
| 102 | IR_VALUE_RE = re.compile(r'(\s+)%([\w\.]+?)([,\s\(\)]|\Z)') |
| 103 | |
| 104 | # Create a FileCheck variable name based on an IR name. |
| 105 | def get_value_name(var): |
| 106 | if var.isdigit(): |
| 107 | var = 'TMP' + var |
| 108 | var = var.replace('.', '_') |
| 109 | return var.upper() |
| 110 | |
| 111 | |
| 112 | # Create a FileCheck variable from regex. |
| 113 | def get_value_definition(var): |
| 114 | return '[[' + get_value_name(var) + ':%.*]]' |
| 115 | |
| 116 | |
| 117 | # Use a FileCheck variable. |
| 118 | def get_value_use(var): |
| 119 | return '[[' + get_value_name(var) + ']]' |
| 120 | |
| 121 | # Replace IR value defs and uses with FileCheck variables. |
| 122 | def genericize_check_lines(lines): |
| 123 | # This gets called for each match that occurs in |
| 124 | # a line. We transform variables we haven't seen |
| 125 | # into defs, and variables we have seen into uses. |
| 126 | def transform_line_vars(match): |
| 127 | var = match.group(2) |
| 128 | if var in vars_seen: |
| 129 | rv = get_value_use(var) |
| 130 | else: |
| 131 | vars_seen.add(var) |
| 132 | rv = get_value_definition(var) |
| 133 | # re.sub replaces the entire regex match |
| 134 | # with whatever you return, so we have |
| 135 | # to make sure to hand it back everything |
| 136 | # including the commas and spaces. |
| 137 | return match.group(1) + rv + match.group(3) |
| 138 | |
| 139 | vars_seen = set() |
| 140 | lines_with_def = [] |
| 141 | |
| 142 | for i, line in enumerate(lines): |
| 143 | # An IR variable named '%.' matches the FileCheck regex string. |
| 144 | line = line.replace('%.', '%dot') |
| 145 | # Ignore any comments, since the check lines will too. |
| 146 | scrubbed_line = SCRUB_IR_COMMENT_RE.sub(r'', line) |
| 147 | lines[i] = IR_VALUE_RE.sub(transform_line_vars, scrubbed_line) |
| 148 | return lines |
| 149 | |
| 150 | |
| 151 | def add_ir_checks(output_lines, prefix_list, func_dict, func_name, opt_basename): |
| 152 | # Label format is based on IR string. |
| 153 | check_label_format = "; %s-LABEL: @%s(" |
| 154 | |
| 155 | printed_prefixes = [] |
| 156 | for checkprefixes, _ in prefix_list: |
| 157 | for checkprefix in checkprefixes: |
| 158 | if checkprefix in printed_prefixes: |
| 159 | break |
| 160 | if not func_dict[checkprefix][func_name]: |
| 161 | continue |
| 162 | # Add some space between different check prefixes, but not after the last |
| 163 | # check line (before the test code). |
| 164 | #if len(printed_prefixes) != 0: |
| 165 | # output_lines.append(';') |
| 166 | printed_prefixes.append(checkprefix) |
| 167 | output_lines.append(check_label_format % (checkprefix, func_name)) |
| 168 | func_body = func_dict[checkprefix][func_name].splitlines() |
| 169 | |
| 170 | # For IR output, change all defs to FileCheck variables, so we're immune |
| 171 | # to variable naming fashions. |
| 172 | func_body = genericize_check_lines(func_body) |
| 173 | |
| 174 | # This could be selectively enabled with an optional invocation argument. |
| 175 | # Disabled for now: better to check everything. Be safe rather than sorry. |
| 176 | |
| 177 | # Handle the first line of the function body as a special case because |
| 178 | # it's often just noise (a useless asm comment or entry label). |
| 179 | #if func_body[0].startswith("#") or func_body[0].startswith("entry:"): |
| 180 | # is_blank_line = True |
| 181 | #else: |
| 182 | # output_lines.append('; %s: %s' % (checkprefix, func_body[0])) |
| 183 | # is_blank_line = False |
| 184 | |
| 185 | is_blank_line = False |
| 186 | |
| 187 | for func_line in func_body: |
| 188 | if func_line.strip() == '': |
| 189 | is_blank_line = True |
| 190 | continue |
| 191 | # Do not waste time checking IR comments. |
| 192 | func_line = SCRUB_IR_COMMENT_RE.sub(r'', func_line) |
| 193 | |
| 194 | # Skip blank lines instead of checking them. |
| 195 | if is_blank_line == True: |
| 196 | output_lines.append('; %s: %s' % (checkprefix, func_line)) |
| 197 | else: |
| 198 | output_lines.append('; %s-NEXT: %s' % (checkprefix, func_line)) |
| 199 | is_blank_line = False |
| 200 | |
| 201 | # Add space between different check prefixes and also before the first |
| 202 | # line of code in the test function. |
| 203 | output_lines.append(';') |
| 204 | break |
| 205 | return output_lines |