run_perf_test.py 19.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12
#!/usr/bin/env python
# Copyright 2014 the V8 project authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.

from collections import namedtuple
import coverage
import json
from mock import DEFAULT
from mock import MagicMock
import os
from os import path, sys
13
import platform
14
import shutil
15
import subprocess
16 17 18 19 20 21
import tempfile
import unittest

# Requires python-coverage and python-mock. Native python coverage
# version >= 3.7.1 should be installed to get the best speed.

22 23 24 25
BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
RUN_PERF = os.path.join(BASE_DIR, 'run_perf.py')
TEST_DATA = os.path.join(BASE_DIR, 'unittests', 'testdata')

26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85
TEST_WORKSPACE = path.join(tempfile.gettempdir(), "test-v8-run-perf")

V8_JSON = {
  "path": ["."],
  "binary": "d7",
  "flags": ["--flag"],
  "main": "run.js",
  "run_count": 1,
  "results_regexp": "^%s: (.+)$",
  "tests": [
    {"name": "Richards"},
    {"name": "DeltaBlue"},
  ]
}

V8_NESTED_SUITES_JSON = {
  "path": ["."],
  "flags": ["--flag"],
  "run_count": 1,
  "units": "score",
  "tests": [
    {"name": "Richards",
     "path": ["richards"],
     "binary": "d7",
     "main": "run.js",
     "resources": ["file1.js", "file2.js"],
     "run_count": 2,
     "results_regexp": "^Richards: (.+)$"},
    {"name": "Sub",
     "path": ["sub"],
     "tests": [
       {"name": "Leaf",
        "path": ["leaf"],
        "run_count_x64": 3,
        "units": "ms",
        "main": "run.js",
        "results_regexp": "^Simple: (.+) ms.$"},
     ]
    },
    {"name": "DeltaBlue",
     "path": ["delta_blue"],
     "main": "run.js",
     "flags": ["--flag2"],
     "results_regexp": "^DeltaBlue: (.+)$"},
    {"name": "ShouldntRun",
     "path": ["."],
     "archs": ["arm"],
     "main": "run.js"},
  ]
}

V8_GENERIC_JSON = {
  "path": ["."],
  "binary": "cc",
  "flags": ["--flag"],
  "generic": True,
  "run_count": 1,
  "units": "ms",
}

86
Output = namedtuple("Output", "stdout, stderr, timed_out")
87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121

class PerfTest(unittest.TestCase):
  @classmethod
  def setUpClass(cls):
    cls.base = path.dirname(path.dirname(path.abspath(__file__)))
    sys.path.append(cls.base)
    cls._cov = coverage.coverage(
        include=([os.path.join(cls.base, "run_perf.py")]))
    cls._cov.start()
    import run_perf
    from testrunner.local import commands
    global commands
    global run_perf

  @classmethod
  def tearDownClass(cls):
    cls._cov.stop()
    print ""
    print cls._cov.report()

  def setUp(self):
    self.maxDiff = None
    if path.exists(TEST_WORKSPACE):
      shutil.rmtree(TEST_WORKSPACE)
    os.makedirs(TEST_WORKSPACE)

  def tearDown(self):
    if path.exists(TEST_WORKSPACE):
      shutil.rmtree(TEST_WORKSPACE)

  def _WriteTestInput(self, json_content):
    self._test_input = path.join(TEST_WORKSPACE, "test.json")
    with open(self._test_input, "w") as f:
      f.write(json.dumps(json_content))

122
  def _MockCommand(self, *args, **kwargs):
123
    # Fake output for each test run.
124 125 126 127
    test_outputs = [Output(stdout=arg,
                           stderr=None,
                           timed_out=kwargs.get("timed_out", False))
                    for arg in args[1]]
128 129 130 131 132 133 134 135 136 137
    def execute(*args, **kwargs):
      return test_outputs.pop()
    commands.Execute = MagicMock(side_effect=execute)

    # Check that d8 is called from the correct cwd for each test run.
    dirs = [path.join(TEST_WORKSPACE, arg) for arg in args[0]]
    def chdir(*args, **kwargs):
      self.assertEquals(dirs.pop(), args[0])
    os.chdir = MagicMock(side_effect=chdir)

138 139 140
    subprocess.check_call = MagicMock()
    platform.system = MagicMock(return_value='Linux')

141 142 143 144 145 146 147 148 149 150
  def _CallMain(self, *args):
    self._test_output = path.join(TEST_WORKSPACE, "results.json")
    all_args=[
      "--json-test-results",
      self._test_output,
      self._test_input,
    ]
    all_args += args
    return run_perf.Main(all_args)

151 152
  def _LoadResults(self, file_name=None):
    with open(file_name or self._test_output) as f:
153 154
      return json.load(f)

155
  def _VerifyResults(self, suite, units, traces, file_name=None):
156 157 158 159 160
    self.assertEquals([
      {"units": units,
       "graphs": [suite, trace["name"]],
       "results": trace["results"],
       "stddev": trace["stddev"]} for trace in traces],
161
      self._LoadResults(file_name)["traces"])
162 163 164 165

  def _VerifyErrors(self, errors):
    self.assertEquals(errors, self._LoadResults()["errors"])

166
  def _VerifyMock(self, binary, *args, **kwargs):
167 168
    arg = [path.join(path.dirname(self.base), binary)]
    arg += args
169 170
    commands.Execute.assert_called_with(
        arg, timeout=kwargs.get("timeout", 60))
171

172
  def _VerifyMockMultiple(self, *args, **kwargs):
173 174 175 176
    expected = []
    for arg in args:
      a = [path.join(path.dirname(self.base), arg[0])]
      a += arg[1:]
177
      expected.append(((a,), {"timeout": kwargs.get("timeout", 60)}))
178 179 180 181 182 183 184 185
    self.assertEquals(expected, commands.Execute.call_args_list)

  def testOneRun(self):
    self._WriteTestInput(V8_JSON)
    self._MockCommand(["."], ["x\nRichards: 1.234\nDeltaBlue: 10657567\ny\n"])
    self.assertEquals(0, self._CallMain())
    self._VerifyResults("test", "score", [
      {"name": "Richards", "results": ["1.234"], "stddev": ""},
186
      {"name": "DeltaBlue", "results": ["10657567.0"], "stddev": ""},
187 188 189 190
    ])
    self._VerifyErrors([])
    self._VerifyMock(path.join("out", "x64.release", "d7"), "--flag", "run.js")

191 192 193 194 195 196 197 198
  def testOneRunWithTestFlags(self):
    test_input = dict(V8_JSON)
    test_input["test_flags"] = ["2", "test_name"]
    self._WriteTestInput(test_input)
    self._MockCommand(["."], ["Richards: 1.234\nDeltaBlue: 10657567"])
    self.assertEquals(0, self._CallMain())
    self._VerifyResults("test", "score", [
      {"name": "Richards", "results": ["1.234"], "stddev": ""},
199
      {"name": "DeltaBlue", "results": ["10657567.0"], "stddev": ""},
200 201 202 203 204
    ])
    self._VerifyErrors([])
    self._VerifyMock(path.join("out", "x64.release", "d7"), "--flag", "run.js",
                     "--", "2", "test_name")

205 206 207 208 209 210 211 212 213 214 215
  def testTwoRuns_Units_SuiteName(self):
    test_input = dict(V8_JSON)
    test_input["run_count"] = 2
    test_input["name"] = "v8"
    test_input["units"] = "ms"
    self._WriteTestInput(test_input)
    self._MockCommand([".", "."],
                      ["Richards: 100\nDeltaBlue: 200\n",
                       "Richards: 50\nDeltaBlue: 300\n"])
    self.assertEquals(0, self._CallMain())
    self._VerifyResults("v8", "ms", [
216 217
      {"name": "Richards", "results": ["50.0", "100.0"], "stddev": ""},
      {"name": "DeltaBlue", "results": ["300.0", "200.0"], "stddev": ""},
218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233
    ])
    self._VerifyErrors([])
    self._VerifyMock(path.join("out", "x64.release", "d7"), "--flag", "run.js")

  def testTwoRuns_SubRegexp(self):
    test_input = dict(V8_JSON)
    test_input["run_count"] = 2
    del test_input["results_regexp"]
    test_input["tests"][0]["results_regexp"] = "^Richards: (.+)$"
    test_input["tests"][1]["results_regexp"] = "^DeltaBlue: (.+)$"
    self._WriteTestInput(test_input)
    self._MockCommand([".", "."],
                      ["Richards: 100\nDeltaBlue: 200\n",
                       "Richards: 50\nDeltaBlue: 300\n"])
    self.assertEquals(0, self._CallMain())
    self._VerifyResults("test", "score", [
234 235
      {"name": "Richards", "results": ["50.0", "100.0"], "stddev": ""},
      {"name": "DeltaBlue", "results": ["300.0", "200.0"], "stddev": ""},
236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252
    ])
    self._VerifyErrors([])
    self._VerifyMock(path.join("out", "x64.release", "d7"), "--flag", "run.js")

  def testNestedSuite(self):
    self._WriteTestInput(V8_NESTED_SUITES_JSON)
    self._MockCommand(["delta_blue", "sub/leaf", "richards"],
                      ["DeltaBlue: 200\n",
                       "Simple: 1 ms.\n",
                       "Simple: 2 ms.\n",
                       "Simple: 3 ms.\n",
                       "Richards: 100\n",
                       "Richards: 50\n"])
    self.assertEquals(0, self._CallMain())
    self.assertEquals([
      {"units": "score",
       "graphs": ["test", "Richards"],
253
       "results": ["50.0", "100.0"],
254 255 256
       "stddev": ""},
      {"units": "ms",
       "graphs": ["test", "Sub", "Leaf"],
257
       "results": ["3.0", "2.0", "1.0"],
258 259 260
       "stddev": ""},
      {"units": "score",
       "graphs": ["test", "DeltaBlue"],
261
       "results": ["200.0"],
262 263 264 265
       "stddev": ""},
      ], self._LoadResults()["traces"])
    self._VerifyErrors([])
    self._VerifyMockMultiple(
266 267
        (path.join("out", "x64.release", "d7"), "--flag", "run.js"),
        (path.join("out", "x64.release", "d7"), "--flag", "run.js"),
268 269 270 271 272 273 274 275 276 277 278 279 280 281
        (path.join("out", "x64.release", "d8"), "--flag", "run.js"),
        (path.join("out", "x64.release", "d8"), "--flag", "run.js"),
        (path.join("out", "x64.release", "d8"), "--flag", "run.js"),
        (path.join("out", "x64.release", "d8"), "--flag", "--flag2", "run.js"))

  def testOneRunStdDevRegExp(self):
    test_input = dict(V8_JSON)
    test_input["stddev_regexp"] = "^%s\-stddev: (.+)$"
    self._WriteTestInput(test_input)
    self._MockCommand(["."], ["Richards: 1.234\nRichards-stddev: 0.23\n"
                              "DeltaBlue: 10657567\nDeltaBlue-stddev: 106\n"])
    self.assertEquals(0, self._CallMain())
    self._VerifyResults("test", "score", [
      {"name": "Richards", "results": ["1.234"], "stddev": "0.23"},
282
      {"name": "DeltaBlue", "results": ["10657567.0"], "stddev": "106"},
283 284 285 286 287 288 289 290 291 292 293 294 295 296 297
    ])
    self._VerifyErrors([])
    self._VerifyMock(path.join("out", "x64.release", "d7"), "--flag", "run.js")

  def testTwoRunsStdDevRegExp(self):
    test_input = dict(V8_JSON)
    test_input["stddev_regexp"] = "^%s\-stddev: (.+)$"
    test_input["run_count"] = 2
    self._WriteTestInput(test_input)
    self._MockCommand(["."], ["Richards: 3\nRichards-stddev: 0.7\n"
                              "DeltaBlue: 6\nDeltaBlue-boom: 0.9\n",
                              "Richards: 2\nRichards-stddev: 0.5\n"
                              "DeltaBlue: 5\nDeltaBlue-stddev: 0.8\n"])
    self.assertEquals(1, self._CallMain())
    self._VerifyResults("test", "score", [
298 299
      {"name": "Richards", "results": ["2.0", "3.0"], "stddev": "0.7"},
      {"name": "DeltaBlue", "results": ["5.0", "6.0"], "stddev": "0.8"},
300 301
    ])
    self._VerifyErrors(
302
        ["Test test/Richards should only run once since a stddev is provided "
303
         "by the test.",
304
         "Test test/DeltaBlue should only run once since a stddev is provided "
305 306
         "by the test.",
         "Regexp \"^DeltaBlue\-stddev: (.+)$\" didn't match for test "
307
         "test/DeltaBlue."])
308 309 310 311 312 313 314 315
    self._VerifyMock(path.join("out", "x64.release", "d7"), "--flag", "run.js")

  def testBuildbot(self):
    self._WriteTestInput(V8_JSON)
    self._MockCommand(["."], ["Richards: 1.234\nDeltaBlue: 10657567\n"])
    self.assertEquals(0, self._CallMain("--buildbot"))
    self._VerifyResults("test", "score", [
      {"name": "Richards", "results": ["1.234"], "stddev": ""},
316
      {"name": "DeltaBlue", "results": ["10657567.0"], "stddev": ""},
317 318 319 320 321 322 323 324 325 326 327 328
    ])
    self._VerifyErrors([])
    self._VerifyMock(path.join("out", "Release", "d7"), "--flag", "run.js")

  def testBuildbotWithTotal(self):
    test_input = dict(V8_JSON)
    test_input["total"] = True
    self._WriteTestInput(test_input)
    self._MockCommand(["."], ["Richards: 1.234\nDeltaBlue: 10657567\n"])
    self.assertEquals(0, self._CallMain("--buildbot"))
    self._VerifyResults("test", "score", [
      {"name": "Richards", "results": ["1.234"], "stddev": ""},
329
      {"name": "DeltaBlue", "results": ["10657567.0"], "stddev": ""},
330 331 332 333 334 335 336 337 338
      {"name": "Total", "results": ["3626.49109719"], "stddev": ""},
    ])
    self._VerifyErrors([])
    self._VerifyMock(path.join("out", "Release", "d7"), "--flag", "run.js")

  def testBuildbotWithTotalAndErrors(self):
    test_input = dict(V8_JSON)
    test_input["total"] = True
    self._WriteTestInput(test_input)
339
    self._MockCommand(["."], ["x\nRichards: bla\nDeltaBlue: 10657567\ny\n"])
340 341 342
    self.assertEquals(1, self._CallMain("--buildbot"))
    self._VerifyResults("test", "score", [
      {"name": "Richards", "results": [], "stddev": ""},
343
      {"name": "DeltaBlue", "results": ["10657567.0"], "stddev": ""},
344 345
    ])
    self._VerifyErrors(
346
        ["Regexp \"^Richards: (.+)$\" "
347
         "returned a non-numeric for test test/Richards.",
348 349 350 351 352 353 354 355 356
         "Not all traces have the same number of results."])
    self._VerifyMock(path.join("out", "Release", "d7"), "--flag", "run.js")

  def testRegexpNoMatch(self):
    self._WriteTestInput(V8_JSON)
    self._MockCommand(["."], ["x\nRichaards: 1.234\nDeltaBlue: 10657567\ny\n"])
    self.assertEquals(1, self._CallMain())
    self._VerifyResults("test", "score", [
      {"name": "Richards", "results": [], "stddev": ""},
357
      {"name": "DeltaBlue", "results": ["10657567.0"], "stddev": ""},
358 359
    ])
    self._VerifyErrors(
360
        ["Regexp \"^Richards: (.+)$\" didn't match for test test/Richards."])
361 362 363 364 365 366
    self._VerifyMock(path.join("out", "x64.release", "d7"), "--flag", "run.js")

  def testOneRunGeneric(self):
    test_input = dict(V8_GENERIC_JSON)
    self._WriteTestInput(test_input)
    self._MockCommand(["."], [
367 368
      "RESULT Infra: Constant1= 11 count\n"
      "RESULT Infra: Constant2= [10,5,10,15] count\n"
369 370 371
      "RESULT Infra: Constant3= {12,1.2} count\n"
      "RESULT Infra: Constant4= [10,5,error,15] count\n"])
    self.assertEquals(1, self._CallMain())
372 373 374
    self.assertEquals([
      {"units": "count",
       "graphs": ["test", "Infra", "Constant1"],
375
       "results": ["11.0"],
376 377 378
       "stddev": ""},
      {"units": "count",
       "graphs": ["test", "Infra", "Constant2"],
379
       "results": ["10.0", "5.0", "10.0", "15.0"],
380 381 382
       "stddev": ""},
      {"units": "count",
       "graphs": ["test", "Infra", "Constant3"],
383
       "results": ["12.0"],
384
       "stddev": "1.2"},
385 386 387 388
      {"units": "count",
       "graphs": ["test", "Infra", "Constant4"],
       "results": [],
       "stddev": ""},
389
      ], self._LoadResults()["traces"])
390
    self._VerifyErrors(["Found non-numeric in test/Infra/Constant4"])
391
    self._VerifyMock(path.join("out", "x64.release", "cc"), "--flag", "")
392 393 394 395 396 397 398 399 400 401 402 403

  def testOneRunTimingOut(self):
    test_input = dict(V8_JSON)
    test_input["timeout"] = 70
    self._WriteTestInput(test_input)
    self._MockCommand(["."], [""], timed_out=True)
    self.assertEquals(1, self._CallMain())
    self._VerifyResults("test", "score", [
      {"name": "Richards", "results": [], "stddev": ""},
      {"name": "DeltaBlue", "results": [], "stddev": ""},
    ])
    self._VerifyErrors([
404 405
      "Regexp \"^Richards: (.+)$\" didn't match for test test/Richards.",
      "Regexp \"^DeltaBlue: (.+)$\" didn't match for test test/DeltaBlue.",
406 407 408
    ])
    self._VerifyMock(
        path.join("out", "x64.release", "d7"), "--flag", "run.js", timeout=70)
409 410 411 412 413

  # Simple test that mocks out the android platform. Testing the platform would
  # require lots of complicated mocks for the android tools.
  def testAndroid(self):
    self._WriteTestInput(V8_JSON)
414 415
    # FIXME(machenbach): This is not test-local!
    platform = run_perf.AndroidPlatform
416 417 418 419
    platform.PreExecution = MagicMock(return_value=None)
    platform.PostExecution = MagicMock(return_value=None)
    platform.PreTests = MagicMock(return_value=None)
    platform.Run = MagicMock(
420
        return_value=("Richards: 1.234\nDeltaBlue: 10657567\n", None))
421 422 423
    run_perf.AndroidPlatform = MagicMock(return_value=platform)
    self.assertEquals(
        0, self._CallMain("--android-build-tools", "/some/dir",
424
                          "--arch", "arm"))
425 426
    self._VerifyResults("test", "score", [
      {"name": "Richards", "results": ["1.234"], "stddev": ""},
427
      {"name": "DeltaBlue", "results": ["10657567.0"], "stddev": ""},
428
    ])
429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459

  def testTwoRuns_Trybot(self):
    test_input = dict(V8_JSON)
    test_input["run_count"] = 2
    self._WriteTestInput(test_input)
    self._MockCommand([".", ".", ".", "."],
                      ["Richards: 100\nDeltaBlue: 200\n",
                       "Richards: 200\nDeltaBlue: 20\n",
                       "Richards: 50\nDeltaBlue: 200\n",
                       "Richards: 100\nDeltaBlue: 20\n"])
    test_output_no_patch = path.join(TEST_WORKSPACE, "results_no_patch.json")
    self.assertEquals(0, self._CallMain(
        "--outdir-no-patch", "out-no-patch",
        "--json-test-results-no-patch", test_output_no_patch,
    ))
    self._VerifyResults("test", "score", [
      {"name": "Richards", "results": ["100.0", "200.0"], "stddev": ""},
      {"name": "DeltaBlue", "results": ["20.0", "20.0"], "stddev": ""},
    ])
    self._VerifyResults("test", "score", [
      {"name": "Richards", "results": ["50.0", "100.0"], "stddev": ""},
      {"name": "DeltaBlue", "results": ["200.0", "200.0"], "stddev": ""},
    ], test_output_no_patch)
    self._VerifyErrors([])
    self._VerifyMockMultiple(
        (path.join("out", "x64.release", "d7"), "--flag", "run.js"),
        (path.join("out-no-patch", "x64.release", "d7"), "--flag", "run.js"),
        (path.join("out", "x64.release", "d7"), "--flag", "run.js"),
        (path.join("out-no-patch", "x64.release", "d7"), "--flag", "run.js"),
    )

460 461 462 463 464 465 466 467 468 469 470 471 472
  def testWrongBinaryWithProf(self):
    test_input = dict(V8_JSON)
    self._WriteTestInput(test_input)
    self._MockCommand(["."], ["x\nRichards: 1.234\nDeltaBlue: 10657567\ny\n"])
    self.assertEquals(0, self._CallMain("--extra-flags=--prof"))
    self._VerifyResults("test", "score", [
      {"name": "Richards", "results": ["1.234"], "stddev": ""},
      {"name": "DeltaBlue", "results": ["10657567.0"], "stddev": ""},
    ])
    self._VerifyErrors([])
    self._VerifyMock(path.join("out", "x64.release", "d7"),
                     "--flag", "--prof", "run.js")

473 474 475 476 477 478 479
  def testUnzip(self):
    def Gen():
      for i in [1, 2, 3]:
        yield i, i + 1
    l, r = run_perf.Unzip(Gen())
    self.assertEquals([1, 2, 3], list(l()))
    self.assertEquals([2, 3, 4], list(r()))
480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547

  #############################################################################
  ### System tests

  def _RunPerf(self, mocked_d8, test_json):
    output_json = path.join(TEST_WORKSPACE, "output.json")
    args = [
      sys.executable, RUN_PERF,
      "--binary-override-path", os.path.join(TEST_DATA, mocked_d8),
      "--json-test-results", output_json,
      os.path.join(TEST_DATA, test_json),
    ]
    subprocess.check_output(args)
    return self._LoadResults(output_json)

  def testNormal(self):
    results = self._RunPerf("d8_mocked1.py", "test1.json")
    self.assertEquals([], results['errors'])
    self.assertEquals([
      {
        'units': 'score',
        'graphs': ['test1', 'Richards'],
        'results': [u'1.2', u'1.2'],
        'stddev': '',
      },
      {
        'units': 'score',
        'graphs': ['test1', 'DeltaBlue'],
        'results': [u'2.1', u'2.1'],
        'stddev': '',
      },
    ], results['traces'])

  def testResultsProcessor(self):
    results = self._RunPerf("d8_mocked2.py", "test2.json")
    self.assertEquals([], results['errors'])
    self.assertEquals([
      {
        'units': 'score',
        'graphs': ['test2', 'Richards'],
        'results': [u'1.2', u'1.2'],
        'stddev': '',
      },
      {
        'units': 'score',
        'graphs': ['test2', 'DeltaBlue'],
        'results': [u'2.1', u'2.1'],
        'stddev': '',
      },
    ], results['traces'])

  def testResultsProcessorNested(self):
    results = self._RunPerf("d8_mocked2.py", "test3.json")
    self.assertEquals([], results['errors'])
    self.assertEquals([
      {
        'units': 'score',
        'graphs': ['test3', 'Octane', 'Richards'],
        'results': [u'1.2'],
        'stddev': '',
      },
      {
        'units': 'score',
        'graphs': ['test3', 'Octane', 'DeltaBlue'],
        'results': [u'2.1'],
        'stddev': '',
      },
    ], results['traces'])