[325] | 1 | #!/usr/bin/python -O
|
---|
[2] | 2 |
|
---|
| 3 | """ A Python Benchmark Suite
|
---|
| 4 |
|
---|
| 5 | """
|
---|
| 6 | #
|
---|
| 7 | # Note: Please keep this module compatible to Python 1.5.2.
|
---|
| 8 | #
|
---|
| 9 | # Tests may include features in later Python versions, but these
|
---|
| 10 | # should then be embedded in try-except clauses in the configuration
|
---|
| 11 | # module Setup.py.
|
---|
| 12 | #
|
---|
| 13 |
|
---|
| 14 | # pybench Copyright
|
---|
| 15 | __copyright__ = """\
|
---|
| 16 | Copyright (c), 1997-2006, Marc-Andre Lemburg (mal@lemburg.com)
|
---|
| 17 | Copyright (c), 2000-2006, eGenix.com Software GmbH (info@egenix.com)
|
---|
| 18 |
|
---|
| 19 | All Rights Reserved.
|
---|
| 20 |
|
---|
| 21 | Permission to use, copy, modify, and distribute this software and its
|
---|
| 22 | documentation for any purpose and without fee or royalty is hereby
|
---|
| 23 | granted, provided that the above copyright notice appear in all copies
|
---|
| 24 | and that both that copyright notice and this permission notice appear
|
---|
| 25 | in supporting documentation or portions thereof, including
|
---|
| 26 | modifications, that you make.
|
---|
| 27 |
|
---|
| 28 | THE AUTHOR MARC-ANDRE LEMBURG DISCLAIMS ALL WARRANTIES WITH REGARD TO
|
---|
| 29 | THIS SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND
|
---|
| 30 | FITNESS, IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL,
|
---|
| 31 | INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING
|
---|
| 32 | FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT,
|
---|
| 33 | NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION
|
---|
| 34 | WITH THE USE OR PERFORMANCE OF THIS SOFTWARE !
|
---|
| 35 | """
|
---|
| 36 |
|
---|
| 37 | import sys, time, operator, string, platform
|
---|
| 38 | from CommandLine import *
|
---|
| 39 |
|
---|
| 40 | try:
|
---|
| 41 | import cPickle
|
---|
| 42 | pickle = cPickle
|
---|
| 43 | except ImportError:
|
---|
| 44 | import pickle
|
---|
| 45 |
|
---|
| 46 | # Version number; version history: see README file !
|
---|
| 47 | __version__ = '2.0'
|
---|
| 48 |
|
---|
| 49 | ### Constants
|
---|
| 50 |
|
---|
| 51 | # Second fractions
|
---|
| 52 | MILLI_SECONDS = 1e3
|
---|
| 53 | MICRO_SECONDS = 1e6
|
---|
| 54 |
|
---|
| 55 | # Percent unit
|
---|
| 56 | PERCENT = 100
|
---|
| 57 |
|
---|
| 58 | # Horizontal line length
|
---|
| 59 | LINE = 79
|
---|
| 60 |
|
---|
| 61 | # Minimum test run-time
|
---|
| 62 | MIN_TEST_RUNTIME = 1e-3
|
---|
| 63 |
|
---|
| 64 | # Number of calibration runs to use for calibrating the tests
|
---|
| 65 | CALIBRATION_RUNS = 20
|
---|
| 66 |
|
---|
| 67 | # Number of calibration loops to run for each calibration run
|
---|
| 68 | CALIBRATION_LOOPS = 20
|
---|
| 69 |
|
---|
| 70 | # Allow skipping calibration ?
|
---|
| 71 | ALLOW_SKIPPING_CALIBRATION = 1
|
---|
| 72 |
|
---|
| 73 | # Timer types
|
---|
| 74 | TIMER_TIME_TIME = 'time.time'
|
---|
| 75 | TIMER_TIME_CLOCK = 'time.clock'
|
---|
| 76 | TIMER_SYSTIMES_PROCESSTIME = 'systimes.processtime'
|
---|
| 77 |
|
---|
| 78 | # Choose platform default timer
|
---|
| 79 | if sys.platform[:3] == 'win':
|
---|
| 80 | # On WinXP this has 2.5ms resolution
|
---|
| 81 | TIMER_PLATFORM_DEFAULT = TIMER_TIME_CLOCK
|
---|
| 82 | else:
|
---|
| 83 | # On Linux this has 1ms resolution
|
---|
| 84 | TIMER_PLATFORM_DEFAULT = TIMER_TIME_TIME
|
---|
| 85 |
|
---|
| 86 | # Print debug information ?
|
---|
| 87 | _debug = 0
|
---|
| 88 |
|
---|
| 89 | ### Helpers
|
---|
| 90 |
|
---|
| 91 | def get_timer(timertype):
|
---|
| 92 |
|
---|
| 93 | if timertype == TIMER_TIME_TIME:
|
---|
| 94 | return time.time
|
---|
| 95 | elif timertype == TIMER_TIME_CLOCK:
|
---|
| 96 | return time.clock
|
---|
| 97 | elif timertype == TIMER_SYSTIMES_PROCESSTIME:
|
---|
| 98 | import systimes
|
---|
| 99 | return systimes.processtime
|
---|
| 100 | else:
|
---|
| 101 | raise TypeError('unknown timer type: %s' % timertype)
|
---|
| 102 |
|
---|
| 103 | def get_machine_details():
|
---|
| 104 |
|
---|
| 105 | if _debug:
|
---|
| 106 | print 'Getting machine details...'
|
---|
| 107 | buildno, builddate = platform.python_build()
|
---|
| 108 | python = platform.python_version()
|
---|
| 109 | try:
|
---|
| 110 | unichr(100000)
|
---|
| 111 | except ValueError:
|
---|
| 112 | # UCS2 build (standard)
|
---|
| 113 | unicode = 'UCS2'
|
---|
| 114 | except NameError:
|
---|
| 115 | unicode = None
|
---|
| 116 | else:
|
---|
| 117 | # UCS4 build (most recent Linux distros)
|
---|
| 118 | unicode = 'UCS4'
|
---|
| 119 | bits, linkage = platform.architecture()
|
---|
| 120 | return {
|
---|
| 121 | 'platform': platform.platform(),
|
---|
| 122 | 'processor': platform.processor(),
|
---|
| 123 | 'executable': sys.executable,
|
---|
| 124 | 'implementation': getattr(platform, 'python_implementation',
|
---|
| 125 | lambda:'n/a')(),
|
---|
| 126 | 'python': platform.python_version(),
|
---|
| 127 | 'compiler': platform.python_compiler(),
|
---|
| 128 | 'buildno': buildno,
|
---|
| 129 | 'builddate': builddate,
|
---|
| 130 | 'unicode': unicode,
|
---|
| 131 | 'bits': bits,
|
---|
| 132 | }
|
---|
| 133 |
|
---|
| 134 | def print_machine_details(d, indent=''):
|
---|
| 135 |
|
---|
| 136 | l = ['Machine Details:',
|
---|
| 137 | ' Platform ID: %s' % d.get('platform', 'n/a'),
|
---|
| 138 | ' Processor: %s' % d.get('processor', 'n/a'),
|
---|
| 139 | '',
|
---|
| 140 | 'Python:',
|
---|
| 141 | ' Implementation: %s' % d.get('implementation', 'n/a'),
|
---|
| 142 | ' Executable: %s' % d.get('executable', 'n/a'),
|
---|
| 143 | ' Version: %s' % d.get('python', 'n/a'),
|
---|
| 144 | ' Compiler: %s' % d.get('compiler', 'n/a'),
|
---|
| 145 | ' Bits: %s' % d.get('bits', 'n/a'),
|
---|
| 146 | ' Build: %s (#%s)' % (d.get('builddate', 'n/a'),
|
---|
| 147 | d.get('buildno', 'n/a')),
|
---|
| 148 | ' Unicode: %s' % d.get('unicode', 'n/a'),
|
---|
| 149 | ]
|
---|
| 150 | print indent + string.join(l, '\n' + indent) + '\n'
|
---|
| 151 |
|
---|
| 152 | ### Test baseclass
|
---|
| 153 |
|
---|
| 154 | class Test:
|
---|
| 155 |
|
---|
| 156 | """ All test must have this class as baseclass. It provides
|
---|
| 157 | the necessary interface to the benchmark machinery.
|
---|
| 158 |
|
---|
| 159 | The tests must set .rounds to a value high enough to let the
|
---|
| 160 | test run between 20-50 seconds. This is needed because
|
---|
| 161 | clock()-timing only gives rather inaccurate values (on Linux,
|
---|
| 162 | for example, it is accurate to a few hundreths of a
|
---|
| 163 | second). If you don't want to wait that long, use a warp
|
---|
| 164 | factor larger than 1.
|
---|
| 165 |
|
---|
| 166 | It is also important to set the .operations variable to a
|
---|
| 167 | value representing the number of "virtual operations" done per
|
---|
| 168 | call of .run().
|
---|
| 169 |
|
---|
| 170 | If you change a test in some way, don't forget to increase
|
---|
| 171 | its version number.
|
---|
| 172 |
|
---|
| 173 | """
|
---|
| 174 |
|
---|
| 175 | ### Instance variables that each test should override
|
---|
| 176 |
|
---|
| 177 | # Version number of the test as float (x.yy); this is important
|
---|
| 178 | # for comparisons of benchmark runs - tests with unequal version
|
---|
| 179 | # number will not get compared.
|
---|
| 180 | version = 2.0
|
---|
| 181 |
|
---|
| 182 | # The number of abstract operations done in each round of the
|
---|
| 183 | # test. An operation is the basic unit of what you want to
|
---|
| 184 | # measure. The benchmark will output the amount of run-time per
|
---|
| 185 | # operation. Note that in order to raise the measured timings
|
---|
| 186 | # significantly above noise level, it is often required to repeat
|
---|
| 187 | # sets of operations more than once per test round. The measured
|
---|
| 188 | # overhead per test round should be less than 1 second.
|
---|
| 189 | operations = 1
|
---|
| 190 |
|
---|
| 191 | # Number of rounds to execute per test run. This should be
|
---|
| 192 | # adjusted to a figure that results in a test run-time of between
|
---|
| 193 | # 1-2 seconds.
|
---|
| 194 | rounds = 100000
|
---|
| 195 |
|
---|
| 196 | ### Internal variables
|
---|
| 197 |
|
---|
| 198 | # Mark this class as implementing a test
|
---|
| 199 | is_a_test = 1
|
---|
| 200 |
|
---|
| 201 | # Last timing: (real, run, overhead)
|
---|
| 202 | last_timing = (0.0, 0.0, 0.0)
|
---|
| 203 |
|
---|
| 204 | # Warp factor to use for this test
|
---|
| 205 | warp = 1
|
---|
| 206 |
|
---|
| 207 | # Number of calibration runs to use
|
---|
| 208 | calibration_runs = CALIBRATION_RUNS
|
---|
| 209 |
|
---|
| 210 | # List of calibration timings
|
---|
| 211 | overhead_times = None
|
---|
| 212 |
|
---|
| 213 | # List of test run timings
|
---|
| 214 | times = []
|
---|
| 215 |
|
---|
| 216 | # Timer used for the benchmark
|
---|
| 217 | timer = TIMER_PLATFORM_DEFAULT
|
---|
| 218 |
|
---|
| 219 | def __init__(self, warp=None, calibration_runs=None, timer=None):
|
---|
| 220 |
|
---|
| 221 | # Set parameters
|
---|
| 222 | if warp is not None:
|
---|
| 223 | self.rounds = int(self.rounds / warp)
|
---|
| 224 | if self.rounds == 0:
|
---|
| 225 | raise ValueError('warp factor set too high')
|
---|
| 226 | self.warp = warp
|
---|
| 227 | if calibration_runs is not None:
|
---|
| 228 | if (not ALLOW_SKIPPING_CALIBRATION and
|
---|
| 229 | calibration_runs < 1):
|
---|
| 230 | raise ValueError('at least one calibration run is required')
|
---|
| 231 | self.calibration_runs = calibration_runs
|
---|
| 232 | if timer is not None:
|
---|
[391] | 233 | self.timer = timer
|
---|
[2] | 234 |
|
---|
| 235 | # Init variables
|
---|
| 236 | self.times = []
|
---|
| 237 | self.overhead_times = []
|
---|
| 238 |
|
---|
| 239 | # We want these to be in the instance dict, so that pickle
|
---|
| 240 | # saves them
|
---|
| 241 | self.version = self.version
|
---|
| 242 | self.operations = self.operations
|
---|
| 243 | self.rounds = self.rounds
|
---|
| 244 |
|
---|
| 245 | def get_timer(self):
|
---|
| 246 |
|
---|
| 247 | """ Return the timer function to use for the test.
|
---|
| 248 |
|
---|
| 249 | """
|
---|
| 250 | return get_timer(self.timer)
|
---|
| 251 |
|
---|
| 252 | def compatible(self, other):
|
---|
| 253 |
|
---|
| 254 | """ Return 1/0 depending on whether the test is compatible
|
---|
| 255 | with the other Test instance or not.
|
---|
| 256 |
|
---|
| 257 | """
|
---|
| 258 | if self.version != other.version:
|
---|
| 259 | return 0
|
---|
| 260 | if self.rounds != other.rounds:
|
---|
| 261 | return 0
|
---|
| 262 | return 1
|
---|
| 263 |
|
---|
| 264 | def calibrate_test(self):
|
---|
| 265 |
|
---|
| 266 | if self.calibration_runs == 0:
|
---|
| 267 | self.overhead_times = [0.0]
|
---|
| 268 | return
|
---|
| 269 |
|
---|
| 270 | calibrate = self.calibrate
|
---|
| 271 | timer = self.get_timer()
|
---|
| 272 | calibration_loops = range(CALIBRATION_LOOPS)
|
---|
| 273 |
|
---|
| 274 | # Time the calibration loop overhead
|
---|
| 275 | prep_times = []
|
---|
| 276 | for i in range(self.calibration_runs):
|
---|
| 277 | t = timer()
|
---|
| 278 | for i in calibration_loops:
|
---|
| 279 | pass
|
---|
| 280 | t = timer() - t
|
---|
[391] | 281 | prep_times.append(t / CALIBRATION_LOOPS)
|
---|
[2] | 282 | min_prep_time = min(prep_times)
|
---|
| 283 | if _debug:
|
---|
| 284 | print
|
---|
| 285 | print 'Calib. prep time = %.6fms' % (
|
---|
| 286 | min_prep_time * MILLI_SECONDS)
|
---|
| 287 |
|
---|
| 288 | # Time the calibration runs (doing CALIBRATION_LOOPS loops of
|
---|
| 289 | # .calibrate() method calls each)
|
---|
| 290 | for i in range(self.calibration_runs):
|
---|
| 291 | t = timer()
|
---|
| 292 | for i in calibration_loops:
|
---|
| 293 | calibrate()
|
---|
| 294 | t = timer() - t
|
---|
| 295 | self.overhead_times.append(t / CALIBRATION_LOOPS
|
---|
| 296 | - min_prep_time)
|
---|
| 297 |
|
---|
| 298 | # Check the measured times
|
---|
| 299 | min_overhead = min(self.overhead_times)
|
---|
| 300 | max_overhead = max(self.overhead_times)
|
---|
| 301 | if _debug:
|
---|
| 302 | print 'Calib. overhead time = %.6fms' % (
|
---|
| 303 | min_overhead * MILLI_SECONDS)
|
---|
| 304 | if min_overhead < 0.0:
|
---|
| 305 | raise ValueError('calibration setup did not work')
|
---|
| 306 | if max_overhead - min_overhead > 0.1:
|
---|
| 307 | raise ValueError(
|
---|
| 308 | 'overhead calibration timing range too inaccurate: '
|
---|
| 309 | '%r - %r' % (min_overhead, max_overhead))
|
---|
| 310 |
|
---|
| 311 | def run(self):
|
---|
| 312 |
|
---|
| 313 | """ Run the test in two phases: first calibrate, then
|
---|
| 314 | do the actual test. Be careful to keep the calibration
|
---|
| 315 | timing low w/r to the test timing.
|
---|
| 316 |
|
---|
| 317 | """
|
---|
| 318 | test = self.test
|
---|
| 319 | timer = self.get_timer()
|
---|
| 320 |
|
---|
| 321 | # Get calibration
|
---|
| 322 | min_overhead = min(self.overhead_times)
|
---|
| 323 |
|
---|
| 324 | # Test run
|
---|
| 325 | t = timer()
|
---|
| 326 | test()
|
---|
| 327 | t = timer() - t
|
---|
| 328 | if t < MIN_TEST_RUNTIME:
|
---|
| 329 | raise ValueError('warp factor too high: '
|
---|
| 330 | 'test times are < 10ms')
|
---|
| 331 | eff_time = t - min_overhead
|
---|
| 332 | if eff_time < 0:
|
---|
| 333 | raise ValueError('wrong calibration')
|
---|
| 334 | self.last_timing = (eff_time, t, min_overhead)
|
---|
| 335 | self.times.append(eff_time)
|
---|
| 336 |
|
---|
| 337 | def calibrate(self):
|
---|
| 338 |
|
---|
| 339 | """ Calibrate the test.
|
---|
| 340 |
|
---|
| 341 | This method should execute everything that is needed to
|
---|
| 342 | setup and run the test - except for the actual operations
|
---|
| 343 | that you intend to measure. pybench uses this method to
|
---|
| 344 | measure the test implementation overhead.
|
---|
| 345 |
|
---|
| 346 | """
|
---|
| 347 | return
|
---|
| 348 |
|
---|
| 349 | def test(self):
|
---|
| 350 |
|
---|
| 351 | """ Run the test.
|
---|
| 352 |
|
---|
| 353 | The test needs to run self.rounds executing
|
---|
| 354 | self.operations number of operations each.
|
---|
| 355 |
|
---|
| 356 | """
|
---|
| 357 | return
|
---|
| 358 |
|
---|
| 359 | def stat(self):
|
---|
| 360 |
|
---|
| 361 | """ Return test run statistics as tuple:
|
---|
| 362 |
|
---|
| 363 | (minimum run time,
|
---|
| 364 | average run time,
|
---|
| 365 | total run time,
|
---|
| 366 | average time per operation,
|
---|
| 367 | minimum overhead time)
|
---|
| 368 |
|
---|
| 369 | """
|
---|
| 370 | runs = len(self.times)
|
---|
| 371 | if runs == 0:
|
---|
| 372 | return 0.0, 0.0, 0.0, 0.0
|
---|
| 373 | min_time = min(self.times)
|
---|
| 374 | total_time = reduce(operator.add, self.times, 0.0)
|
---|
| 375 | avg_time = total_time / float(runs)
|
---|
| 376 | operation_avg = total_time / float(runs
|
---|
| 377 | * self.rounds
|
---|
| 378 | * self.operations)
|
---|
| 379 | if self.overhead_times:
|
---|
| 380 | min_overhead = min(self.overhead_times)
|
---|
| 381 | else:
|
---|
| 382 | min_overhead = self.last_timing[2]
|
---|
| 383 | return min_time, avg_time, total_time, operation_avg, min_overhead
|
---|
| 384 |
|
---|
| 385 | ### Load Setup
|
---|
| 386 |
|
---|
| 387 | # This has to be done after the definition of the Test class, since
|
---|
| 388 | # the Setup module will import subclasses using this class.
|
---|
| 389 |
|
---|
| 390 | import Setup
|
---|
| 391 |
|
---|
| 392 | ### Benchmark base class
|
---|
| 393 |
|
---|
| 394 | class Benchmark:
|
---|
| 395 |
|
---|
| 396 | # Name of the benchmark
|
---|
| 397 | name = ''
|
---|
| 398 |
|
---|
| 399 | # Number of benchmark rounds to run
|
---|
| 400 | rounds = 1
|
---|
| 401 |
|
---|
| 402 | # Warp factor use to run the tests
|
---|
| 403 | warp = 1 # Warp factor
|
---|
| 404 |
|
---|
| 405 | # Average benchmark round time
|
---|
| 406 | roundtime = 0
|
---|
| 407 |
|
---|
| 408 | # Benchmark version number as float x.yy
|
---|
| 409 | version = 2.0
|
---|
| 410 |
|
---|
| 411 | # Produce verbose output ?
|
---|
| 412 | verbose = 0
|
---|
| 413 |
|
---|
| 414 | # Dictionary with the machine details
|
---|
| 415 | machine_details = None
|
---|
| 416 |
|
---|
| 417 | # Timer used for the benchmark
|
---|
| 418 | timer = TIMER_PLATFORM_DEFAULT
|
---|
| 419 |
|
---|
| 420 | def __init__(self, name, verbose=None, timer=None, warp=None,
|
---|
| 421 | calibration_runs=None):
|
---|
| 422 |
|
---|
| 423 | if name:
|
---|
| 424 | self.name = name
|
---|
| 425 | else:
|
---|
| 426 | self.name = '%04i-%02i-%02i %02i:%02i:%02i' % \
|
---|
| 427 | (time.localtime(time.time())[:6])
|
---|
| 428 | if verbose is not None:
|
---|
| 429 | self.verbose = verbose
|
---|
| 430 | if timer is not None:
|
---|
| 431 | self.timer = timer
|
---|
| 432 | if warp is not None:
|
---|
| 433 | self.warp = warp
|
---|
| 434 | if calibration_runs is not None:
|
---|
| 435 | self.calibration_runs = calibration_runs
|
---|
| 436 |
|
---|
| 437 | # Init vars
|
---|
| 438 | self.tests = {}
|
---|
| 439 | if _debug:
|
---|
| 440 | print 'Getting machine details...'
|
---|
| 441 | self.machine_details = get_machine_details()
|
---|
| 442 |
|
---|
| 443 | # Make .version an instance attribute to have it saved in the
|
---|
| 444 | # Benchmark pickle
|
---|
| 445 | self.version = self.version
|
---|
| 446 |
|
---|
| 447 | def get_timer(self):
|
---|
| 448 |
|
---|
| 449 | """ Return the timer function to use for the test.
|
---|
| 450 |
|
---|
| 451 | """
|
---|
| 452 | return get_timer(self.timer)
|
---|
| 453 |
|
---|
| 454 | def compatible(self, other):
|
---|
| 455 |
|
---|
| 456 | """ Return 1/0 depending on whether the benchmark is
|
---|
| 457 | compatible with the other Benchmark instance or not.
|
---|
| 458 |
|
---|
| 459 | """
|
---|
| 460 | if self.version != other.version:
|
---|
| 461 | return 0
|
---|
| 462 | if (self.machine_details == other.machine_details and
|
---|
| 463 | self.timer != other.timer):
|
---|
| 464 | return 0
|
---|
| 465 | if (self.calibration_runs == 0 and
|
---|
| 466 | other.calibration_runs != 0):
|
---|
| 467 | return 0
|
---|
| 468 | if (self.calibration_runs != 0 and
|
---|
| 469 | other.calibration_runs == 0):
|
---|
| 470 | return 0
|
---|
| 471 | return 1
|
---|
| 472 |
|
---|
| 473 | def load_tests(self, setupmod, limitnames=None):
|
---|
| 474 |
|
---|
| 475 | # Add tests
|
---|
| 476 | if self.verbose:
|
---|
| 477 | print 'Searching for tests ...'
|
---|
| 478 | print '--------------------------------------'
|
---|
| 479 | for testclass in setupmod.__dict__.values():
|
---|
| 480 | if not hasattr(testclass, 'is_a_test'):
|
---|
| 481 | continue
|
---|
| 482 | name = testclass.__name__
|
---|
| 483 | if name == 'Test':
|
---|
| 484 | continue
|
---|
| 485 | if (limitnames is not None and
|
---|
| 486 | limitnames.search(name) is None):
|
---|
| 487 | continue
|
---|
| 488 | self.tests[name] = testclass(
|
---|
| 489 | warp=self.warp,
|
---|
| 490 | calibration_runs=self.calibration_runs,
|
---|
| 491 | timer=self.timer)
|
---|
| 492 | l = self.tests.keys()
|
---|
| 493 | l.sort()
|
---|
| 494 | if self.verbose:
|
---|
| 495 | for name in l:
|
---|
| 496 | print ' %s' % name
|
---|
| 497 | print '--------------------------------------'
|
---|
| 498 | print ' %i tests found' % len(l)
|
---|
| 499 | print
|
---|
| 500 |
|
---|
| 501 | def calibrate(self):
|
---|
| 502 |
|
---|
| 503 | print 'Calibrating tests. Please wait...',
|
---|
| 504 | sys.stdout.flush()
|
---|
| 505 | if self.verbose:
|
---|
| 506 | print
|
---|
| 507 | print
|
---|
| 508 | print 'Test min max'
|
---|
| 509 | print '-' * LINE
|
---|
| 510 | tests = self.tests.items()
|
---|
| 511 | tests.sort()
|
---|
| 512 | for i in range(len(tests)):
|
---|
| 513 | name, test = tests[i]
|
---|
| 514 | test.calibrate_test()
|
---|
| 515 | if self.verbose:
|
---|
| 516 | print '%30s: %6.3fms %6.3fms' % \
|
---|
| 517 | (name,
|
---|
| 518 | min(test.overhead_times) * MILLI_SECONDS,
|
---|
| 519 | max(test.overhead_times) * MILLI_SECONDS)
|
---|
| 520 | if self.verbose:
|
---|
| 521 | print
|
---|
| 522 | print 'Done with the calibration.'
|
---|
| 523 | else:
|
---|
| 524 | print 'done.'
|
---|
| 525 | print
|
---|
| 526 |
|
---|
| 527 | def run(self):
|
---|
| 528 |
|
---|
| 529 | tests = self.tests.items()
|
---|
| 530 | tests.sort()
|
---|
| 531 | timer = self.get_timer()
|
---|
| 532 | print 'Running %i round(s) of the suite at warp factor %i:' % \
|
---|
| 533 | (self.rounds, self.warp)
|
---|
| 534 | print
|
---|
| 535 | self.roundtimes = []
|
---|
| 536 | for i in range(self.rounds):
|
---|
| 537 | if self.verbose:
|
---|
| 538 | print ' Round %-25i effective absolute overhead' % (i+1)
|
---|
| 539 | total_eff_time = 0.0
|
---|
| 540 | for j in range(len(tests)):
|
---|
| 541 | name, test = tests[j]
|
---|
| 542 | if self.verbose:
|
---|
| 543 | print '%30s:' % name,
|
---|
| 544 | test.run()
|
---|
| 545 | (eff_time, abs_time, min_overhead) = test.last_timing
|
---|
| 546 | total_eff_time = total_eff_time + eff_time
|
---|
| 547 | if self.verbose:
|
---|
| 548 | print ' %5.0fms %5.0fms %7.3fms' % \
|
---|
| 549 | (eff_time * MILLI_SECONDS,
|
---|
| 550 | abs_time * MILLI_SECONDS,
|
---|
| 551 | min_overhead * MILLI_SECONDS)
|
---|
| 552 | self.roundtimes.append(total_eff_time)
|
---|
| 553 | if self.verbose:
|
---|
| 554 | print (' '
|
---|
| 555 | ' ------------------------------')
|
---|
| 556 | print (' '
|
---|
| 557 | ' Totals: %6.0fms' %
|
---|
| 558 | (total_eff_time * MILLI_SECONDS))
|
---|
| 559 | print
|
---|
| 560 | else:
|
---|
| 561 | print '* Round %i done in %.3f seconds.' % (i+1,
|
---|
| 562 | total_eff_time)
|
---|
| 563 | print
|
---|
| 564 |
|
---|
| 565 | def stat(self):
|
---|
| 566 |
|
---|
| 567 | """ Return benchmark run statistics as tuple:
|
---|
| 568 |
|
---|
| 569 | (minimum round time,
|
---|
| 570 | average round time,
|
---|
| 571 | maximum round time)
|
---|
| 572 |
|
---|
| 573 | XXX Currently not used, since the benchmark does test
|
---|
| 574 | statistics across all rounds.
|
---|
| 575 |
|
---|
| 576 | """
|
---|
| 577 | runs = len(self.roundtimes)
|
---|
| 578 | if runs == 0:
|
---|
| 579 | return 0.0, 0.0
|
---|
| 580 | min_time = min(self.roundtimes)
|
---|
| 581 | total_time = reduce(operator.add, self.roundtimes, 0.0)
|
---|
| 582 | avg_time = total_time / float(runs)
|
---|
| 583 | max_time = max(self.roundtimes)
|
---|
| 584 | return (min_time, avg_time, max_time)
|
---|
| 585 |
|
---|
| 586 | def print_header(self, title='Benchmark'):
|
---|
| 587 |
|
---|
| 588 | print '-' * LINE
|
---|
| 589 | print '%s: %s' % (title, self.name)
|
---|
| 590 | print '-' * LINE
|
---|
| 591 | print
|
---|
| 592 | print ' Rounds: %s' % self.rounds
|
---|
| 593 | print ' Warp: %s' % self.warp
|
---|
| 594 | print ' Timer: %s' % self.timer
|
---|
| 595 | print
|
---|
| 596 | if self.machine_details:
|
---|
| 597 | print_machine_details(self.machine_details, indent=' ')
|
---|
| 598 | print
|
---|
| 599 |
|
---|
| 600 | def print_benchmark(self, hidenoise=0, limitnames=None):
|
---|
| 601 |
|
---|
| 602 | print ('Test '
|
---|
| 603 | ' minimum average operation overhead')
|
---|
| 604 | print '-' * LINE
|
---|
| 605 | tests = self.tests.items()
|
---|
| 606 | tests.sort()
|
---|
| 607 | total_min_time = 0.0
|
---|
| 608 | total_avg_time = 0.0
|
---|
| 609 | for name, test in tests:
|
---|
| 610 | if (limitnames is not None and
|
---|
| 611 | limitnames.search(name) is None):
|
---|
| 612 | continue
|
---|
| 613 | (min_time,
|
---|
| 614 | avg_time,
|
---|
| 615 | total_time,
|
---|
| 616 | op_avg,
|
---|
| 617 | min_overhead) = test.stat()
|
---|
| 618 | total_min_time = total_min_time + min_time
|
---|
| 619 | total_avg_time = total_avg_time + avg_time
|
---|
| 620 | print '%30s: %5.0fms %5.0fms %6.2fus %7.3fms' % \
|
---|
| 621 | (name,
|
---|
| 622 | min_time * MILLI_SECONDS,
|
---|
| 623 | avg_time * MILLI_SECONDS,
|
---|
| 624 | op_avg * MICRO_SECONDS,
|
---|
| 625 | min_overhead *MILLI_SECONDS)
|
---|
| 626 | print '-' * LINE
|
---|
| 627 | print ('Totals: '
|
---|
| 628 | ' %6.0fms %6.0fms' %
|
---|
| 629 | (total_min_time * MILLI_SECONDS,
|
---|
| 630 | total_avg_time * MILLI_SECONDS,
|
---|
| 631 | ))
|
---|
| 632 | print
|
---|
| 633 |
|
---|
| 634 | def print_comparison(self, compare_to, hidenoise=0, limitnames=None):
|
---|
| 635 |
|
---|
| 636 | # Check benchmark versions
|
---|
| 637 | if compare_to.version != self.version:
|
---|
| 638 | print ('* Benchmark versions differ: '
|
---|
| 639 | 'cannot compare this benchmark to "%s" !' %
|
---|
| 640 | compare_to.name)
|
---|
| 641 | print
|
---|
| 642 | self.print_benchmark(hidenoise=hidenoise,
|
---|
| 643 | limitnames=limitnames)
|
---|
| 644 | return
|
---|
| 645 |
|
---|
| 646 | # Print header
|
---|
| 647 | compare_to.print_header('Comparing with')
|
---|
| 648 | print ('Test '
|
---|
| 649 | ' minimum run-time average run-time')
|
---|
| 650 | print (' '
|
---|
| 651 | ' this other diff this other diff')
|
---|
| 652 | print '-' * LINE
|
---|
| 653 |
|
---|
| 654 | # Print test comparisons
|
---|
| 655 | tests = self.tests.items()
|
---|
| 656 | tests.sort()
|
---|
| 657 | total_min_time = other_total_min_time = 0.0
|
---|
| 658 | total_avg_time = other_total_avg_time = 0.0
|
---|
| 659 | benchmarks_compatible = self.compatible(compare_to)
|
---|
| 660 | tests_compatible = 1
|
---|
| 661 | for name, test in tests:
|
---|
| 662 | if (limitnames is not None and
|
---|
| 663 | limitnames.search(name) is None):
|
---|
| 664 | continue
|
---|
| 665 | (min_time,
|
---|
| 666 | avg_time,
|
---|
| 667 | total_time,
|
---|
| 668 | op_avg,
|
---|
| 669 | min_overhead) = test.stat()
|
---|
| 670 | total_min_time = total_min_time + min_time
|
---|
| 671 | total_avg_time = total_avg_time + avg_time
|
---|
| 672 | try:
|
---|
| 673 | other = compare_to.tests[name]
|
---|
| 674 | except KeyError:
|
---|
| 675 | other = None
|
---|
| 676 | if other is None:
|
---|
| 677 | # Other benchmark doesn't include the given test
|
---|
| 678 | min_diff, avg_diff = 'n/a', 'n/a'
|
---|
| 679 | other_min_time = 0.0
|
---|
| 680 | other_avg_time = 0.0
|
---|
| 681 | tests_compatible = 0
|
---|
| 682 | else:
|
---|
| 683 | (other_min_time,
|
---|
| 684 | other_avg_time,
|
---|
| 685 | other_total_time,
|
---|
| 686 | other_op_avg,
|
---|
| 687 | other_min_overhead) = other.stat()
|
---|
| 688 | other_total_min_time = other_total_min_time + other_min_time
|
---|
| 689 | other_total_avg_time = other_total_avg_time + other_avg_time
|
---|
| 690 | if (benchmarks_compatible and
|
---|
| 691 | test.compatible(other)):
|
---|
[391] | 692 | # Both benchmark and tests are comparable
|
---|
[2] | 693 | min_diff = ((min_time * self.warp) /
|
---|
| 694 | (other_min_time * other.warp) - 1.0)
|
---|
| 695 | avg_diff = ((avg_time * self.warp) /
|
---|
| 696 | (other_avg_time * other.warp) - 1.0)
|
---|
| 697 | if hidenoise and abs(min_diff) < 10.0:
|
---|
| 698 | min_diff = ''
|
---|
| 699 | else:
|
---|
| 700 | min_diff = '%+5.1f%%' % (min_diff * PERCENT)
|
---|
| 701 | if hidenoise and abs(avg_diff) < 10.0:
|
---|
| 702 | avg_diff = ''
|
---|
| 703 | else:
|
---|
| 704 | avg_diff = '%+5.1f%%' % (avg_diff * PERCENT)
|
---|
| 705 | else:
|
---|
[391] | 706 | # Benchmark or tests are not comparable
|
---|
[2] | 707 | min_diff, avg_diff = 'n/a', 'n/a'
|
---|
| 708 | tests_compatible = 0
|
---|
| 709 | print '%30s: %5.0fms %5.0fms %7s %5.0fms %5.0fms %7s' % \
|
---|
| 710 | (name,
|
---|
| 711 | min_time * MILLI_SECONDS,
|
---|
| 712 | other_min_time * MILLI_SECONDS * compare_to.warp / self.warp,
|
---|
| 713 | min_diff,
|
---|
| 714 | avg_time * MILLI_SECONDS,
|
---|
| 715 | other_avg_time * MILLI_SECONDS * compare_to.warp / self.warp,
|
---|
| 716 | avg_diff)
|
---|
| 717 | print '-' * LINE
|
---|
| 718 |
|
---|
| 719 | # Summarise test results
|
---|
| 720 | if not benchmarks_compatible or not tests_compatible:
|
---|
| 721 | min_diff, avg_diff = 'n/a', 'n/a'
|
---|
| 722 | else:
|
---|
| 723 | if other_total_min_time != 0.0:
|
---|
| 724 | min_diff = '%+5.1f%%' % (
|
---|
| 725 | ((total_min_time * self.warp) /
|
---|
| 726 | (other_total_min_time * compare_to.warp) - 1.0) * PERCENT)
|
---|
| 727 | else:
|
---|
| 728 | min_diff = 'n/a'
|
---|
| 729 | if other_total_avg_time != 0.0:
|
---|
| 730 | avg_diff = '%+5.1f%%' % (
|
---|
| 731 | ((total_avg_time * self.warp) /
|
---|
| 732 | (other_total_avg_time * compare_to.warp) - 1.0) * PERCENT)
|
---|
| 733 | else:
|
---|
| 734 | avg_diff = 'n/a'
|
---|
| 735 | print ('Totals: '
|
---|
| 736 | ' %5.0fms %5.0fms %7s %5.0fms %5.0fms %7s' %
|
---|
| 737 | (total_min_time * MILLI_SECONDS,
|
---|
| 738 | (other_total_min_time * compare_to.warp/self.warp
|
---|
| 739 | * MILLI_SECONDS),
|
---|
| 740 | min_diff,
|
---|
| 741 | total_avg_time * MILLI_SECONDS,
|
---|
| 742 | (other_total_avg_time * compare_to.warp/self.warp
|
---|
| 743 | * MILLI_SECONDS),
|
---|
| 744 | avg_diff
|
---|
| 745 | ))
|
---|
| 746 | print
|
---|
| 747 | print '(this=%s, other=%s)' % (self.name,
|
---|
| 748 | compare_to.name)
|
---|
| 749 | print
|
---|
| 750 |
|
---|
| 751 | class PyBenchCmdline(Application):
|
---|
| 752 |
|
---|
| 753 | header = ("PYBENCH - a benchmark test suite for Python "
|
---|
| 754 | "interpreters/compilers.")
|
---|
| 755 |
|
---|
| 756 | version = __version__
|
---|
| 757 |
|
---|
| 758 | debug = _debug
|
---|
| 759 |
|
---|
| 760 | options = [ArgumentOption('-n',
|
---|
| 761 | 'number of rounds',
|
---|
| 762 | Setup.Number_of_rounds),
|
---|
| 763 | ArgumentOption('-f',
|
---|
| 764 | 'save benchmark to file arg',
|
---|
| 765 | ''),
|
---|
| 766 | ArgumentOption('-c',
|
---|
| 767 | 'compare benchmark with the one in file arg',
|
---|
| 768 | ''),
|
---|
| 769 | ArgumentOption('-s',
|
---|
| 770 | 'show benchmark in file arg, then exit',
|
---|
| 771 | ''),
|
---|
| 772 | ArgumentOption('-w',
|
---|
| 773 | 'set warp factor to arg',
|
---|
| 774 | Setup.Warp_factor),
|
---|
| 775 | ArgumentOption('-t',
|
---|
| 776 | 'run only tests with names matching arg',
|
---|
| 777 | ''),
|
---|
| 778 | ArgumentOption('-C',
|
---|
| 779 | 'set the number of calibration runs to arg',
|
---|
| 780 | CALIBRATION_RUNS),
|
---|
| 781 | SwitchOption('-d',
|
---|
| 782 | 'hide noise in comparisons',
|
---|
| 783 | 0),
|
---|
| 784 | SwitchOption('-v',
|
---|
| 785 | 'verbose output (not recommended)',
|
---|
| 786 | 0),
|
---|
| 787 | SwitchOption('--with-gc',
|
---|
| 788 | 'enable garbage collection',
|
---|
| 789 | 0),
|
---|
| 790 | SwitchOption('--with-syscheck',
|
---|
| 791 | 'use default sys check interval',
|
---|
| 792 | 0),
|
---|
| 793 | ArgumentOption('--timer',
|
---|
| 794 | 'use given timer',
|
---|
| 795 | TIMER_PLATFORM_DEFAULT),
|
---|
| 796 | ]
|
---|
| 797 |
|
---|
| 798 | about = """\
|
---|
| 799 | The normal operation is to run the suite and display the
|
---|
| 800 | results. Use -f to save them for later reuse or comparisons.
|
---|
| 801 |
|
---|
| 802 | Available timers:
|
---|
| 803 |
|
---|
| 804 | time.time
|
---|
| 805 | time.clock
|
---|
| 806 | systimes.processtime
|
---|
| 807 |
|
---|
| 808 | Examples:
|
---|
| 809 |
|
---|
| 810 | python2.1 pybench.py -f p21.pybench
|
---|
| 811 | python2.5 pybench.py -f p25.pybench
|
---|
| 812 | python pybench.py -s p25.pybench -c p21.pybench
|
---|
| 813 | """
|
---|
| 814 | copyright = __copyright__
|
---|
| 815 |
|
---|
| 816 | def main(self):
|
---|
| 817 |
|
---|
| 818 | rounds = self.values['-n']
|
---|
| 819 | reportfile = self.values['-f']
|
---|
| 820 | show_bench = self.values['-s']
|
---|
| 821 | compare_to = self.values['-c']
|
---|
| 822 | hidenoise = self.values['-d']
|
---|
| 823 | warp = int(self.values['-w'])
|
---|
| 824 | withgc = self.values['--with-gc']
|
---|
| 825 | limitnames = self.values['-t']
|
---|
| 826 | if limitnames:
|
---|
| 827 | if _debug:
|
---|
| 828 | print '* limiting test names to one with substring "%s"' % \
|
---|
| 829 | limitnames
|
---|
| 830 | limitnames = re.compile(limitnames, re.I)
|
---|
| 831 | else:
|
---|
| 832 | limitnames = None
|
---|
| 833 | verbose = self.verbose
|
---|
| 834 | withsyscheck = self.values['--with-syscheck']
|
---|
| 835 | calibration_runs = self.values['-C']
|
---|
| 836 | timer = self.values['--timer']
|
---|
| 837 |
|
---|
| 838 | print '-' * LINE
|
---|
| 839 | print 'PYBENCH %s' % __version__
|
---|
| 840 | print '-' * LINE
|
---|
| 841 | print '* using %s %s' % (
|
---|
| 842 | getattr(platform, 'python_implementation', lambda:'Python')(),
|
---|
| 843 | string.join(string.split(sys.version), ' '))
|
---|
| 844 |
|
---|
| 845 | # Switch off garbage collection
|
---|
| 846 | if not withgc:
|
---|
| 847 | try:
|
---|
| 848 | import gc
|
---|
| 849 | except ImportError:
|
---|
| 850 | print '* Python version doesn\'t support garbage collection'
|
---|
| 851 | else:
|
---|
| 852 | try:
|
---|
| 853 | gc.disable()
|
---|
| 854 | except NotImplementedError:
|
---|
| 855 | print '* Python version doesn\'t support gc.disable'
|
---|
| 856 | else:
|
---|
| 857 | print '* disabled garbage collection'
|
---|
| 858 |
|
---|
| 859 | # "Disable" sys check interval
|
---|
| 860 | if not withsyscheck:
|
---|
| 861 | # Too bad the check interval uses an int instead of a long...
|
---|
| 862 | value = 2147483647
|
---|
| 863 | try:
|
---|
| 864 | sys.setcheckinterval(value)
|
---|
| 865 | except (AttributeError, NotImplementedError):
|
---|
| 866 | print '* Python version doesn\'t support sys.setcheckinterval'
|
---|
| 867 | else:
|
---|
| 868 | print '* system check interval set to maximum: %s' % value
|
---|
| 869 |
|
---|
| 870 | if timer == TIMER_SYSTIMES_PROCESSTIME:
|
---|
| 871 | import systimes
|
---|
| 872 | print '* using timer: systimes.processtime (%s)' % \
|
---|
| 873 | systimes.SYSTIMES_IMPLEMENTATION
|
---|
| 874 | else:
|
---|
| 875 | print '* using timer: %s' % timer
|
---|
| 876 |
|
---|
| 877 | print
|
---|
| 878 |
|
---|
| 879 | if compare_to:
|
---|
| 880 | try:
|
---|
| 881 | f = open(compare_to,'rb')
|
---|
| 882 | bench = pickle.load(f)
|
---|
| 883 | bench.name = compare_to
|
---|
| 884 | f.close()
|
---|
| 885 | compare_to = bench
|
---|
| 886 | except IOError, reason:
|
---|
| 887 | print '* Error opening/reading file %s: %s' % (
|
---|
| 888 | repr(compare_to),
|
---|
| 889 | reason)
|
---|
| 890 | compare_to = None
|
---|
| 891 |
|
---|
| 892 | if show_bench:
|
---|
| 893 | try:
|
---|
| 894 | f = open(show_bench,'rb')
|
---|
| 895 | bench = pickle.load(f)
|
---|
| 896 | bench.name = show_bench
|
---|
| 897 | f.close()
|
---|
| 898 | bench.print_header()
|
---|
| 899 | if compare_to:
|
---|
| 900 | bench.print_comparison(compare_to,
|
---|
| 901 | hidenoise=hidenoise,
|
---|
| 902 | limitnames=limitnames)
|
---|
| 903 | else:
|
---|
| 904 | bench.print_benchmark(hidenoise=hidenoise,
|
---|
| 905 | limitnames=limitnames)
|
---|
| 906 | except IOError, reason:
|
---|
| 907 | print '* Error opening/reading file %s: %s' % (
|
---|
| 908 | repr(show_bench),
|
---|
| 909 | reason)
|
---|
| 910 | print
|
---|
| 911 | return
|
---|
| 912 |
|
---|
| 913 | if reportfile:
|
---|
| 914 | print 'Creating benchmark: %s (rounds=%i, warp=%i)' % \
|
---|
| 915 | (reportfile, rounds, warp)
|
---|
| 916 | print
|
---|
| 917 |
|
---|
| 918 | # Create benchmark object
|
---|
| 919 | bench = Benchmark(reportfile,
|
---|
| 920 | verbose=verbose,
|
---|
| 921 | timer=timer,
|
---|
| 922 | warp=warp,
|
---|
| 923 | calibration_runs=calibration_runs)
|
---|
| 924 | bench.rounds = rounds
|
---|
| 925 | bench.load_tests(Setup, limitnames=limitnames)
|
---|
| 926 | try:
|
---|
| 927 | bench.calibrate()
|
---|
| 928 | bench.run()
|
---|
| 929 | except KeyboardInterrupt:
|
---|
| 930 | print
|
---|
| 931 | print '*** KeyboardInterrupt -- Aborting'
|
---|
| 932 | print
|
---|
| 933 | return
|
---|
| 934 | bench.print_header()
|
---|
| 935 | if compare_to:
|
---|
| 936 | bench.print_comparison(compare_to,
|
---|
| 937 | hidenoise=hidenoise,
|
---|
| 938 | limitnames=limitnames)
|
---|
| 939 | else:
|
---|
| 940 | bench.print_benchmark(hidenoise=hidenoise,
|
---|
| 941 | limitnames=limitnames)
|
---|
| 942 |
|
---|
| 943 | # Ring bell
|
---|
| 944 | sys.stderr.write('\007')
|
---|
| 945 |
|
---|
| 946 | if reportfile:
|
---|
| 947 | try:
|
---|
| 948 | f = open(reportfile,'wb')
|
---|
| 949 | bench.name = reportfile
|
---|
| 950 | pickle.dump(bench,f)
|
---|
| 951 | f.close()
|
---|
| 952 | except IOError, reason:
|
---|
| 953 | print '* Error opening/writing reportfile'
|
---|
| 954 | except IOError, reason:
|
---|
| 955 | print '* Error opening/writing reportfile %s: %s' % (
|
---|
| 956 | reportfile,
|
---|
| 957 | reason)
|
---|
| 958 | print
|
---|
| 959 |
|
---|
| 960 | if __name__ == '__main__':
|
---|
| 961 | PyBenchCmdline()
|
---|