| 1 |
#!/usr/bin/env python |
| 2 |
# |
| 3 |
# Copyright (C) 2016 The Android Open Source Project |
| 4 |
# |
| 5 |
# Licensed under the Apache License, Version 2.0 (the "License"); |
| 6 |
# you may not use this file except in compliance with the License. |
| 7 |
# You may obtain a copy of the License at |
| 8 |
# |
| 9 |
# http://www.apache.org/licenses/LICENSE-2.0 |
| 10 |
# |
| 11 |
# Unless required by applicable law or agreed to in writing, software |
| 12 |
# distributed under the License is distributed on an "AS IS" BASIS, |
| 13 |
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 |
# See the License for the specific language governing permissions and |
| 15 |
# limitations under the License. |
| 16 |
# |
| 17 |
|
| 18 |
"""annotate.py: annotate source files based on perf.data. |
| 19 |
""" |
| 20 |
|
| 21 |
|
| 22 |
import argparse |
| 23 |
import os |
| 24 |
import os.path |
| 25 |
import shutil |
| 26 |
|
| 27 |
from simpleperf_report_lib import ReportLib |
| 28 |
from utils import log_info, log_warning, log_exit |
| 29 |
from utils import Addr2Nearestline, extant_dir, flatten_arg_list, is_windows, SourceFileSearcher |
| 30 |
|
| 31 |
class SourceLine(object): |
| 32 |
def __init__(self, file_id, function, line): |
| 33 |
self.file = file_id |
| 34 |
self.function = function |
| 35 |
self.line = line |
| 36 |
|
| 37 |
@property |
| 38 |
def file_key(self): |
| 39 |
return self.file |
| 40 |
|
| 41 |
@property |
| 42 |
def function_key(self): |
| 43 |
return (self.file, self.function) |
| 44 |
|
| 45 |
@property |
| 46 |
def line_key(self): |
| 47 |
return (self.file, self.line) |
| 48 |
|
| 49 |
|
| 50 |
class Addr2Line(object): |
| 51 |
"""collect information of how to map [dso_name, vaddr] to [source_file:line]. |
| 52 |
""" |
| 53 |
def __init__(self, ndk_path, binary_cache_path, source_dirs): |
| 54 |
self.addr2line = Addr2Nearestline(ndk_path, binary_cache_path, True) |
| 55 |
self.source_searcher = SourceFileSearcher(source_dirs) |
| 56 |
|
| 57 |
def add_addr(self, dso_path, func_addr, addr): |
| 58 |
self.addr2line.add_addr(dso_path, func_addr, addr) |
| 59 |
|
| 60 |
def convert_addrs_to_lines(self): |
| 61 |
self.addr2line.convert_addrs_to_lines() |
| 62 |
|
| 63 |
def get_sources(self, dso_path, addr): |
| 64 |
dso = self.addr2line.get_dso(dso_path) |
| 65 |
if not dso: |
| 66 |
return [] |
| 67 |
source = self.addr2line.get_addr_source(dso, addr) |
| 68 |
if not source: |
| 69 |
return [] |
| 70 |
result = [] |
| 71 |
for (source_file, source_line, function_name) in source: |
| 72 |
source_file_path = self.source_searcher.get_real_path(source_file) |
| 73 |
if not source_file_path: |
| 74 |
source_file_path = source_file |
| 75 |
result.append(SourceLine(source_file_path, function_name, source_line)) |
| 76 |
return result |
| 77 |
|
| 78 |
|
| 79 |
class Period(object): |
| 80 |
"""event count information. It can be used to represent event count |
| 81 |
of a line, a function, a source file, or a binary. It contains two |
| 82 |
parts: period and acc_period. |
| 83 |
When used for a line, period is the event count occurred when running |
| 84 |
that line, acc_period is the accumulated event count occurred when |
| 85 |
running that line and functions called by that line. Same thing applies |
| 86 |
when it is used for a function, a source file, or a binary. |
| 87 |
""" |
| 88 |
def __init__(self, period=0, acc_period=0): |
| 89 |
self.period = period |
| 90 |
self.acc_period = acc_period |
| 91 |
|
| 92 |
|
| 93 |
def __iadd__(self, other): |
| 94 |
self.period += other.period |
| 95 |
self.acc_period += other.acc_period |
| 96 |
return self |
| 97 |
|
| 98 |
|
| 99 |
class DsoPeriod(object): |
| 100 |
"""Period for each shared library""" |
| 101 |
def __init__(self, dso_name): |
| 102 |
self.dso_name = dso_name |
| 103 |
self.period = Period() |
| 104 |
|
| 105 |
|
| 106 |
def add_period(self, period): |
| 107 |
self.period += period |
| 108 |
|
| 109 |
|
| 110 |
class FilePeriod(object): |
| 111 |
"""Period for each source file""" |
| 112 |
def __init__(self, file_id): |
| 113 |
self.file = file_id |
| 114 |
self.period = Period() |
| 115 |
# Period for each line in the file. |
| 116 |
self.line_dict = {} |
| 117 |
# Period for each function in the source file. |
| 118 |
self.function_dict = {} |
| 119 |
|
| 120 |
|
| 121 |
def add_period(self, period): |
| 122 |
self.period += period |
| 123 |
|
| 124 |
|
| 125 |
def add_line_period(self, line, period): |
| 126 |
a = self.line_dict.get(line) |
| 127 |
if a is None: |
| 128 |
self.line_dict[line] = a = Period() |
| 129 |
a += period |
| 130 |
|
| 131 |
|
| 132 |
def add_function_period(self, function_name, function_start_line, period): |
| 133 |
a = self.function_dict.get(function_name) |
| 134 |
if not a: |
| 135 |
if function_start_line is None: |
| 136 |
function_start_line = -1 |
| 137 |
self.function_dict[function_name] = a = [function_start_line, Period()] |
| 138 |
a[1] += period |
| 139 |
|
| 140 |
|
| 141 |
class SourceFileAnnotator(object): |
| 142 |
"""group code for annotating source files""" |
| 143 |
def __init__(self, config): |
| 144 |
# check config variables |
| 145 |
config_names = ['perf_data_list', 'source_dirs', 'comm_filters', |
| 146 |
'pid_filters', 'tid_filters', 'dso_filters', 'ndk_path'] |
| 147 |
for name in config_names: |
| 148 |
if name not in config: |
| 149 |
log_exit('config [%s] is missing' % name) |
| 150 |
symfs_dir = 'binary_cache' |
| 151 |
if not os.path.isdir(symfs_dir): |
| 152 |
symfs_dir = None |
| 153 |
kallsyms = 'binary_cache/kallsyms' |
| 154 |
if not os.path.isfile(kallsyms): |
| 155 |
kallsyms = None |
| 156 |
|
| 157 |
# init member variables |
| 158 |
self.config = config |
| 159 |
self.symfs_dir = symfs_dir |
| 160 |
self.kallsyms = kallsyms |
| 161 |
self.comm_filter = set(config['comm_filters']) if config.get('comm_filters') else None |
| 162 |
if config.get('pid_filters'): |
| 163 |
self.pid_filter = {int(x) for x in config['pid_filters']} |
| 164 |
else: |
| 165 |
self.pid_filter = None |
| 166 |
if config.get('tid_filters'): |
| 167 |
self.tid_filter = {int(x) for x in config['tid_filters']} |
| 168 |
else: |
| 169 |
self.tid_filter = None |
| 170 |
self.dso_filter = set(config['dso_filters']) if config.get('dso_filters') else None |
| 171 |
|
| 172 |
config['annotate_dest_dir'] = 'annotated_files' |
| 173 |
output_dir = config['annotate_dest_dir'] |
| 174 |
if os.path.isdir(output_dir): |
| 175 |
shutil.rmtree(output_dir) |
| 176 |
os.makedirs(output_dir) |
| 177 |
|
| 178 |
|
| 179 |
self.addr2line = Addr2Line(self.config['ndk_path'], symfs_dir, config.get('source_dirs')) |
| 180 |
self.period = 0 |
| 181 |
self.dso_periods = {} |
| 182 |
self.file_periods = {} |
| 183 |
|
| 184 |
|
| 185 |
def annotate(self): |
| 186 |
self._collect_addrs() |
| 187 |
self._convert_addrs_to_lines() |
| 188 |
self._generate_periods() |
| 189 |
self._write_summary() |
| 190 |
self._annotate_files() |
| 191 |
|
| 192 |
|
| 193 |
def _collect_addrs(self): |
| 194 |
"""Read perf.data, collect all addresses we need to convert to |
| 195 |
source file:line. |
| 196 |
""" |
| 197 |
for perf_data in self.config['perf_data_list']: |
| 198 |
lib = ReportLib() |
| 199 |
lib.SetRecordFile(perf_data) |
| 200 |
if self.symfs_dir: |
| 201 |
lib.SetSymfs(self.symfs_dir) |
| 202 |
if self.kallsyms: |
| 203 |
lib.SetKallsymsFile(self.kallsyms) |
| 204 |
while True: |
| 205 |
sample = lib.GetNextSample() |
| 206 |
if sample is None: |
| 207 |
lib.Close() |
| 208 |
break |
| 209 |
if not self._filter_sample(sample): |
| 210 |
continue |
| 211 |
symbols = [] |
| 212 |
symbols.append(lib.GetSymbolOfCurrentSample()) |
| 213 |
callchain = lib.GetCallChainOfCurrentSample() |
| 214 |
for i in range(callchain.nr): |
| 215 |
symbols.append(callchain.entries[i].symbol) |
| 216 |
for symbol in symbols: |
| 217 |
if self._filter_symbol(symbol): |
| 218 |
self.addr2line.add_addr(symbol.dso_name, symbol.symbol_addr, |
| 219 |
symbol.vaddr_in_file) |
| 220 |
self.addr2line.add_addr(symbol.dso_name, symbol.symbol_addr, |
| 221 |
symbol.symbol_addr) |
| 222 |
|
| 223 |
|
| 224 |
def _filter_sample(self, sample): |
| 225 |
"""Return true if the sample can be used.""" |
| 226 |
if self.comm_filter: |
| 227 |
if sample.thread_comm not in self.comm_filter: |
| 228 |
return False |
| 229 |
if self.pid_filter: |
| 230 |
if sample.pid not in self.pid_filter: |
| 231 |
return False |
| 232 |
if self.tid_filter: |
| 233 |
if sample.tid not in self.tid_filter: |
| 234 |
return False |
| 235 |
return True |
| 236 |
|
| 237 |
|
| 238 |
def _filter_symbol(self, symbol): |
| 239 |
if not self.dso_filter or symbol.dso_name in self.dso_filter: |
| 240 |
return True |
| 241 |
return False |
| 242 |
|
| 243 |
|
| 244 |
def _convert_addrs_to_lines(self): |
| 245 |
self.addr2line.convert_addrs_to_lines() |
| 246 |
|
| 247 |
|
| 248 |
def _generate_periods(self): |
| 249 |
"""read perf.data, collect Period for all types: |
| 250 |
binaries, source files, functions, lines. |
| 251 |
""" |
| 252 |
for perf_data in self.config['perf_data_list']: |
| 253 |
lib = ReportLib() |
| 254 |
lib.SetRecordFile(perf_data) |
| 255 |
if self.symfs_dir: |
| 256 |
lib.SetSymfs(self.symfs_dir) |
| 257 |
if self.kallsyms: |
| 258 |
lib.SetKallsymsFile(self.kallsyms) |
| 259 |
while True: |
| 260 |
sample = lib.GetNextSample() |
| 261 |
if sample is None: |
| 262 |
lib.Close() |
| 263 |
break |
| 264 |
if not self._filter_sample(sample): |
| 265 |
continue |
| 266 |
self._generate_periods_for_sample(lib, sample) |
| 267 |
|
| 268 |
|
| 269 |
def _generate_periods_for_sample(self, lib, sample): |
| 270 |
symbols = [] |
| 271 |
symbols.append(lib.GetSymbolOfCurrentSample()) |
| 272 |
callchain = lib.GetCallChainOfCurrentSample() |
| 273 |
for i in range(callchain.nr): |
| 274 |
symbols.append(callchain.entries[i].symbol) |
| 275 |
# Each sample has a callchain, but its period is only used once |
| 276 |
# to add period for each function/source_line/source_file/binary. |
| 277 |
# For example, if more than one entry in the callchain hits a |
| 278 |
# function, the event count of that function is only increased once. |
| 279 |
# Otherwise, we may get periods > 100%. |
| 280 |
is_sample_used = False |
| 281 |
used_dso_dict = {} |
| 282 |
used_file_dict = {} |
| 283 |
used_function_dict = {} |
| 284 |
used_line_dict = {} |
| 285 |
period = Period(sample.period, sample.period) |
| 286 |
for j, symbol in enumerate(symbols): |
| 287 |
if j == 1: |
| 288 |
period = Period(0, sample.period) |
| 289 |
if not self._filter_symbol(symbol): |
| 290 |
continue |
| 291 |
is_sample_used = True |
| 292 |
# Add period to dso. |
| 293 |
self._add_dso_period(symbol.dso_name, period, used_dso_dict) |
| 294 |
# Add period to source file. |
| 295 |
sources = self.addr2line.get_sources(symbol.dso_name, symbol.vaddr_in_file) |
| 296 |
for source in sources: |
| 297 |
if source.file: |
| 298 |
self._add_file_period(source, period, used_file_dict) |
| 299 |
# Add period to line. |
| 300 |
if source.line: |
| 301 |
self._add_line_period(source, period, used_line_dict) |
| 302 |
# Add period to function. |
| 303 |
sources = self.addr2line.get_sources(symbol.dso_name, symbol.symbol_addr) |
| 304 |
for source in sources: |
| 305 |
if source.file: |
| 306 |
self._add_file_period(source, period, used_file_dict) |
| 307 |
if source.function: |
| 308 |
self._add_function_period(source, period, used_function_dict) |
| 309 |
|
| 310 |
if is_sample_used: |
| 311 |
self.period += sample.period |
| 312 |
|
| 313 |
|
| 314 |
def _add_dso_period(self, dso_name, period, used_dso_dict): |
| 315 |
if dso_name not in used_dso_dict: |
| 316 |
used_dso_dict[dso_name] = True |
| 317 |
dso_period = self.dso_periods.get(dso_name) |
| 318 |
if dso_period is None: |
| 319 |
dso_period = self.dso_periods[dso_name] = DsoPeriod(dso_name) |
| 320 |
dso_period.add_period(period) |
| 321 |
|
| 322 |
|
| 323 |
def _add_file_period(self, source, period, used_file_dict): |
| 324 |
if source.file_key not in used_file_dict: |
| 325 |
used_file_dict[source.file_key] = True |
| 326 |
file_period = self.file_periods.get(source.file) |
| 327 |
if file_period is None: |
| 328 |
file_period = self.file_periods[source.file] = FilePeriod(source.file) |
| 329 |
file_period.add_period(period) |
| 330 |
|
| 331 |
|
| 332 |
def _add_line_period(self, source, period, used_line_dict): |
| 333 |
if source.line_key not in used_line_dict: |
| 334 |
used_line_dict[source.line_key] = True |
| 335 |
file_period = self.file_periods[source.file] |
| 336 |
file_period.add_line_period(source.line, period) |
| 337 |
|
| 338 |
|
| 339 |
def _add_function_period(self, source, period, used_function_dict): |
| 340 |
if source.function_key not in used_function_dict: |
| 341 |
used_function_dict[source.function_key] = True |
| 342 |
file_period = self.file_periods[source.file] |
| 343 |
file_period.add_function_period(source.function, source.line, period) |
| 344 |
|
| 345 |
|
| 346 |
def _write_summary(self): |
| 347 |
summary = os.path.join(self.config['annotate_dest_dir'], 'summary') |
| 348 |
with open(summary, 'w') as f: |
| 349 |
f.write('total period: %d\n\n' % self.period) |
| 350 |
dso_periods = sorted(self.dso_periods.values(), |
| 351 |
key=lambda x: x.period.acc_period, reverse=True) |
| 352 |
for dso_period in dso_periods: |
| 353 |
f.write('dso %s: %s\n' % (dso_period.dso_name, |
| 354 |
self._get_percentage_str(dso_period.period))) |
| 355 |
f.write('\n') |
| 356 |
|
| 357 |
file_periods = sorted(self.file_periods.values(), |
| 358 |
key=lambda x: x.period.acc_period, reverse=True) |
| 359 |
for file_period in file_periods: |
| 360 |
f.write('file %s: %s\n' % (file_period.file, |
| 361 |
self._get_percentage_str(file_period.period))) |
| 362 |
for file_period in file_periods: |
| 363 |
f.write('\n\n%s: %s\n' % (file_period.file, |
| 364 |
self._get_percentage_str(file_period.period))) |
| 365 |
values = [] |
| 366 |
for func_name in file_period.function_dict.keys(): |
| 367 |
func_start_line, period = file_period.function_dict[func_name] |
| 368 |
values.append((func_name, func_start_line, period)) |
| 369 |
values = sorted(values, key=lambda x: x[2].acc_period, reverse=True) |
| 370 |
for value in values: |
| 371 |
f.write('\tfunction (%s): line %d, %s\n' % ( |
| 372 |
value[0], value[1], self._get_percentage_str(value[2]))) |
| 373 |
f.write('\n') |
| 374 |
for line in sorted(file_period.line_dict.keys()): |
| 375 |
f.write('\tline %d: %s\n' % ( |
| 376 |
line, self._get_percentage_str(file_period.line_dict[line]))) |
| 377 |
|
| 378 |
|
| 379 |
def _get_percentage_str(self, period, short=False): |
| 380 |
s = 'acc_p: %f%%, p: %f%%' if short else 'accumulated_period: %f%%, period: %f%%' |
| 381 |
return s % self._get_percentage(period) |
| 382 |
|
| 383 |
|
| 384 |
def _get_percentage(self, period): |
| 385 |
if self.period == 0: |
| 386 |
return (0, 0) |
| 387 |
acc_p = 100.0 * period.acc_period / self.period |
| 388 |
p = 100.0 * period.period / self.period |
| 389 |
return (acc_p, p) |
| 390 |
|
| 391 |
|
| 392 |
def _annotate_files(self): |
| 393 |
"""Annotate Source files: add acc_period/period for each source file. |
| 394 |
1. Annotate java source files, which have $JAVA_SRC_ROOT prefix. |
| 395 |
2. Annotate c++ source files. |
| 396 |
""" |
| 397 |
dest_dir = self.config['annotate_dest_dir'] |
| 398 |
for key in self.file_periods: |
| 399 |
from_path = key |
| 400 |
if not os.path.isfile(from_path): |
| 401 |
log_warning("can't find source file for path %s" % from_path) |
| 402 |
continue |
| 403 |
if from_path.startswith('/'): |
| 404 |
to_path = os.path.join(dest_dir, from_path[1:]) |
| 405 |
elif is_windows() and ':\\' in from_path: |
| 406 |
to_path = os.path.join(dest_dir, from_path.replace(':\\', os.sep)) |
| 407 |
else: |
| 408 |
to_path = os.path.join(dest_dir, from_path) |
| 409 |
is_java = from_path.endswith('.java') |
| 410 |
self._annotate_file(from_path, to_path, self.file_periods[key], is_java) |
| 411 |
|
| 412 |
|
| 413 |
def _annotate_file(self, from_path, to_path, file_period, is_java): |
| 414 |
"""Annotate a source file. |
| 415 |
|
| 416 |
Annotate a source file in three steps: |
| 417 |
1. In the first line, show periods of this file. |
| 418 |
2. For each function, show periods of this function. |
| 419 |
3. For each line not hitting the same line as functions, show |
| 420 |
line periods. |
| 421 |
""" |
| 422 |
log_info('annotate file %s' % from_path) |
| 423 |
with open(from_path, 'r') as rf: |
| 424 |
lines = rf.readlines() |
| 425 |
|
| 426 |
annotates = {} |
| 427 |
for line in file_period.line_dict.keys(): |
| 428 |
annotates[line] = self._get_percentage_str(file_period.line_dict[line], True) |
| 429 |
for func_name in file_period.function_dict.keys(): |
| 430 |
func_start_line, period = file_period.function_dict[func_name] |
| 431 |
if func_start_line == -1: |
| 432 |
continue |
| 433 |
line = func_start_line - 1 if is_java else func_start_line |
| 434 |
annotates[line] = '[func] ' + self._get_percentage_str(period, True) |
| 435 |
annotates[1] = '[file] ' + self._get_percentage_str(file_period.period, True) |
| 436 |
|
| 437 |
max_annotate_cols = 0 |
| 438 |
for key in annotates: |
| 439 |
max_annotate_cols = max(max_annotate_cols, len(annotates[key])) |
| 440 |
|
| 441 |
empty_annotate = ' ' * (max_annotate_cols + 6) |
| 442 |
|
| 443 |
dirname = os.path.dirname(to_path) |
| 444 |
if not os.path.isdir(dirname): |
| 445 |
os.makedirs(dirname) |
| 446 |
with open(to_path, 'w') as wf: |
| 447 |
for line in range(1, len(lines) + 1): |
| 448 |
annotate = annotates.get(line) |
| 449 |
if annotate is None: |
| 450 |
if not lines[line-1].strip(): |
| 451 |
annotate = '' |
| 452 |
else: |
| 453 |
annotate = empty_annotate |
| 454 |
else: |
| 455 |
annotate = '/* ' + annotate + ( |
| 456 |
' ' * (max_annotate_cols - len(annotate))) + ' */' |
| 457 |
wf.write(annotate) |
| 458 |
wf.write(lines[line-1]) |
| 459 |
|
| 460 |
def main(): |
| 461 |
parser = argparse.ArgumentParser(description=""" |
| 462 |
Annotate source files based on profiling data. It reads line information from binary_cache |
| 463 |
generated by app_profiler.py or binary_cache_builder.py, and generate annotated source |
| 464 |
files in annotated_files directory.""") |
| 465 |
parser.add_argument('-i', '--perf_data_list', nargs='+', action='append', help=""" |
| 466 |
The paths of profiling data. Default is perf.data.""") |
| 467 |
parser.add_argument('-s', '--source_dirs', type=extant_dir, nargs='+', action='append', help=""" |
| 468 |
Directories to find source files.""") |
| 469 |
parser.add_argument('--comm', nargs='+', action='append', help=""" |
| 470 |
Use samples only in threads with selected names.""") |
| 471 |
parser.add_argument('--pid', nargs='+', action='append', help=""" |
| 472 |
Use samples only in processes with selected process ids.""") |
| 473 |
parser.add_argument('--tid', nargs='+', action='append', help=""" |
| 474 |
Use samples only in threads with selected thread ids.""") |
| 475 |
parser.add_argument('--dso', nargs='+', action='append', help=""" |
| 476 |
Use samples only in selected binaries.""") |
| 477 |
parser.add_argument('--ndk_path', type=extant_dir, help='Set the path of a ndk release.') |
| 478 |
|
| 479 |
args = parser.parse_args() |
| 480 |
config = {} |
| 481 |
config['perf_data_list'] = flatten_arg_list(args.perf_data_list) |
| 482 |
if not config['perf_data_list']: |
| 483 |
config['perf_data_list'].append('perf.data') |
| 484 |
config['source_dirs'] = flatten_arg_list(args.source_dirs) |
| 485 |
config['comm_filters'] = flatten_arg_list(args.comm) |
| 486 |
config['pid_filters'] = flatten_arg_list(args.pid) |
| 487 |
config['tid_filters'] = flatten_arg_list(args.tid) |
| 488 |
config['dso_filters'] = flatten_arg_list(args.dso) |
| 489 |
config['ndk_path'] = args.ndk_path |
| 490 |
|
| 491 |
annotator = SourceFileAnnotator(config) |
| 492 |
annotator.annotate() |
| 493 |
log_info('annotate finish successfully, please check result in annotated_files/.') |
| 494 |
|
| 495 |
if __name__ == '__main__': |
| 496 |
main() |