[Telemetry] Add a profiler based loading measurement.

This measurement reports the hottest functions during page load. I plan to run
it over the top million sites.

It works by loading each page twice, the profile is captured for the second load
only. This is because cross-origin navigations cause the renderer to be swapped
out during profiling and to get more stable results.

This required a couple of core modifications:
1. csv_page_measurement_results should work when the page runner creates a
   dummy results object because discard_first_results is enabled.
2. discard_first_results should discard the first result of each page in
   the page set. Not just the first page.

BUG=None

Review URL: https://chromiumcodereview.appspot.com/19857003

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@213279 0039d316-1c4b-4281-b951-d872f2087c98

7 files changed, 106 insertions, 18 deletions

diff --git a/tools/telemetry/telemetry/core/browser.py b/tools/telemetry/telemetry/core/browser.py
index c604cae..c2a7037 100644
--- a/tools/telemetry/telemetry/core/browser.py
+++ b/tools/telemetry/telemetry/core/browser.py
@@ -192,26 +192,32 @@ class Browser(object):
     del result['ProcessCount']
     return result
 
-  def StartProfiling(self, options, base_output_file):
-    """Starts profiling using |options|.profiler_tool. Results are saved to
+  def StartProfiling(self, profiler_name, base_output_file):
+    """Starts profiling using |profiler_name|. Results are saved to
     |base_output_file|.<process_name>."""
     assert not self._active_profilers
 
-    profiler_class = profiler_finder.FindProfiler(options.profiler)
+    profiler_class = profiler_finder.FindProfiler(profiler_name)
 
-    if not profiler_class.is_supported(options):
+    if not profiler_class.is_supported(self._browser_backend.options):
       raise Exception('The %s profiler is not '
-                      'supported on this platform.' % options.profiler_tool)
+                      'supported on this platform.' % profiler_name)
 
     self._active_profilers.append(
         profiler_class(self._browser_backend, self._platform_backend,
             base_output_file))
 
   def StopProfiling(self):
-    """Stops all active profilers and saves their results."""
+    """Stops all active profilers and saves their results.
+
+    Returns:
+      A list of filenames produced by the profiler.
+    """
+    output_files = []
     for profiler in self._active_profilers:
-      profiler.CollectProfile()
+      output_files.extend(profiler.CollectProfile())
     self._active_profilers = []
+    return output_files
 
   def StartTracing(self, custom_categories=None, timeout=10):
     return self._browser_backend.StartTracing(custom_categories, timeout)
diff --git a/tools/telemetry/telemetry/core/platform/profiler/perf_profiler.py b/tools/telemetry/telemetry/core/platform/profiler/perf_profiler.py
index ab4c918..58cea8a 100644
--- a/tools/telemetry/telemetry/core/platform/profiler/perf_profiler.py
+++ b/tools/telemetry/telemetry/core/platform/profiler/perf_profiler.py
@@ -3,6 +3,8 @@
 # found in the LICENSE file.
 
 import logging
+import os
+import re
 import signal
 import subprocess
 import sys
@@ -40,6 +42,7 @@ class _SingleProcessPerfProfiler(object):
       self._tmp_output_file.close()
     print 'To view the profile, run:'
     print '  perf report -i %s' % self._output_file
+    return self._output_file
 
   def _GetStdOut(self):
     self._tmp_output_file.flush()
@@ -79,5 +82,33 @@ class PerfProfiler(profiler.Profiler):
       return False
 
   def CollectProfile(self):
+    output_files = []
     for single_process in self._process_profilers:
-      single_process.CollectProfile()
+      output_files.append(single_process.CollectProfile())
+    return output_files
+
+  @classmethod
+  def GetTopSamples(cls, file_name, number):
+    """Parses the perf generated profile in |file_name| and returns a
+    {function: period} dict of the |number| hottests functions.
+    """
+    assert os.path.exists(file_name)
+    report = subprocess.Popen(
+        ['perf', 'report', '--show-total-period', '-U', '-t', '^', '-i',
+         file_name],
+        stdout=subprocess.PIPE, stderr=open(os.devnull, 'w')).communicate()[0]
+    period_by_function = {}
+    for line in report.split('\n'):
+      if not line or line.startswith('#'):
+        continue
+      fields = line.split('^')
+      if len(fields) != 5:
+        continue
+      period = int(fields[1])
+      function = fields[4].partition(' ')[2]
+      function = re.sub('<.*>', '', function)  # Strip template params.
+      function = re.sub('[(].*[)]', '', function)  # Strip function params.
+      period_by_function[function] = period
+      if len(period_by_function) == number:
+        break
+    return period_by_function
diff --git a/tools/telemetry/telemetry/core/platform/profiler/perf_profiler_unittest.py b/tools/telemetry/telemetry/core/platform/profiler/perf_profiler_unittest.py
new file mode 100644
index 0000000..8ca407f
--- /dev/null
+++ b/tools/telemetry/telemetry/core/platform/profiler/perf_profiler_unittest.py
@@ -0,0 +1,31 @@
+# Copyright 2013 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+import os
+import logging
+import unittest
+
+from telemetry.core.platform.profiler import perf_profiler
+from telemetry.unittest import options_for_unittests
+
+class TestPerfProfiler(unittest.TestCase):
+  def testPerfProfiler(self):
+    options = options_for_unittests.GetCopy()
+    if not perf_profiler.PerfProfiler.is_supported(options):
+      logging.warning('PerfProfiler is not supported. Skipping test')
+      return
+
+    profile_file = os.path.join(os.path.dirname(__file__),
+                                'testdata', 'perf.profile')
+    self.assertEqual(perf_profiler.PerfProfiler.GetTopSamples(profile_file, 10),
+        { 'v8::internal::StaticMarkingVisitor::MarkMapContents': 63615201,
+          'v8::internal::RelocIterator::next': 38271931,
+          'v8::internal::LAllocator::MeetConstraintsBetween': 42913933,
+          'v8::internal::FlexibleBodyVisitor::Visit': 31909537,
+          'v8::internal::LiveRange::CreateAssignedOperand': 42913933,
+          'void v8::internal::RelocInfo::Visit': 96878864,
+          'WebCore::HTMLTokenizer::nextToken': 48240439,
+          'v8::internal::Scanner::ScanIdentifierOrKeyword': 46054550,
+          'sk_memset32_SSE2': 45121317,
+          'v8::internal::HeapObject::Size': 39786862
+          })
diff --git a/tools/telemetry/telemetry/core/platform/profiler/testdata/perf.profile b/tools/telemetry/telemetry/core/platform/profiler/testdata/perf.profile
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/tools/telemetry/telemetry/core/platform/profiler/testdata/perf.profile
diff --git a/tools/telemetry/telemetry/page/csv_page_measurement_results.py b/tools/telemetry/telemetry/page/csv_page_measurement_results.py
index bd2e8d2..2440dd8 100644
--- a/tools/telemetry/telemetry/page/csv_page_measurement_results.py
+++ b/tools/telemetry/telemetry/page/csv_page_measurement_results.py
@@ -7,9 +7,10 @@ from telemetry.page import page_measurement_results
 
 class CsvPageMeasurementResults(
     page_measurement_results.PageMeasurementResults):
-  def __init__(self, output_stream, output_after_every_page):
+  def __init__(self, output_stream=None, output_after_every_page=None):
     super(CsvPageMeasurementResults, self).__init__()
-    self._results_writer = csv.writer(output_stream)
+    if output_stream:
+      self._results_writer = csv.writer(output_stream)
     self._did_output_header = False
     self._header_names_written_to_writer = None
     self._output_after_every_page = output_after_every_page
diff --git a/tools/telemetry/telemetry/page/page_runner.py b/tools/telemetry/telemetry/page/page_runner.py
index a3f1b85..8e20cec 100644
--- a/tools/telemetry/telemetry/page/page_runner.py
+++ b/tools/telemetry/telemetry/page/page_runner.py
@@ -1,6 +1,7 @@
 # Copyright (c) 2012 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
+import collections
 import glob
 import logging
 import os
@@ -27,7 +28,7 @@ class _RunState(object):
     self._append_to_existing_wpr = False
     self._last_archive_path = None
     self._first_browser = True
-    self.first_page = True
+    self.first_page = collections.defaultdict(lambda: True)
     self.profiler_dir = None
 
   def StartBrowser(self, test, page_set, page, possible_browser,
@@ -69,8 +70,8 @@ class _RunState(object):
     if not self.tab:
       self.tab = self.browser.tabs[0]
 
-    if self.first_page:
-      self.first_page = False
+    if self.first_page[page]:
+      self.first_page[page] = False
       test.WillRunPageSet(self.tab)
 
   def StopBrowser(self):
@@ -93,7 +94,7 @@ class _RunState(object):
     output_file = os.path.join(self.profiler_dir, page.url_as_file_safe_name)
     if options.page_repeat != 1 or options.pageset_repeat != 1:
       output_file = _GetSequentialFileName(output_file)
-    self.browser.StartProfiling(options, output_file)
+    self.browser.StartProfiling(options.profiler, output_file)
 
   def StopProfiling(self):
     self.browser.StopProfiling()
@@ -192,7 +193,7 @@ def Run(test, page_set, expectations, options):
         else:
           possible_browser.options.wpr_mode = wpr_modes.WPR_OFF
       results_for_current_run = results
-      if state.first_page and test.discard_first_result:
+      if state.first_page[page] and test.discard_first_result:
         # If discarding results, substitute a dummy object.
         results_for_current_run = type(results)()
       results_for_current_run.StartTest(page)
diff --git a/tools/telemetry/telemetry/page/page_runner_unittest.py b/tools/telemetry/telemetry/page/page_runner_unittest.py
index 8aa37a1..e0ad39d 100644
--- a/tools/telemetry/telemetry/page/page_runner_unittest.py
+++ b/tools/telemetry/telemetry/page/page_runner_unittest.py
@@ -62,11 +62,14 @@ class PageRunnerTests(unittest.TestCase):
   def testDiscardFirstResult(self):
     ps = page_set.PageSet()
     expectations = test_expectations.TestExpectations()
-    page = page_module.Page(
+    ps.pages.append(page_module.Page(
         'file:///' + os.path.join('..', '..', 'unittest_data', 'blank.html'),
         ps,
-        base_dir=os.path.dirname(__file__))
-    ps.pages.append(page)
+        base_dir=os.path.dirname(__file__)))
+    ps.pages.append(page_module.Page(
+        'file:///' + os.path.join('..', '..', 'unittest_data', 'blank.html'),
+        ps,
+        base_dir=os.path.dirname(__file__)))
 
     class Test(page_test.PageTest):
       @property
@@ -77,10 +80,25 @@ class PageRunnerTests(unittest.TestCase):
 
     options = options_for_unittests.GetCopy()
     options.output_format = 'none'
+
+    options.page_repeat = 1
+    options.pageset_repeat = 1
     results = page_runner.Run(Test('RunTest'), ps, expectations, options)
     self.assertEquals(0, len(results.successes))
     self.assertEquals(0, len(results.failures))
 
+    options.page_repeat = 1
+    options.pageset_repeat = 2
+    results = page_runner.Run(Test('RunTest'), ps, expectations, options)
+    self.assertEquals(2, len(results.successes))
+    self.assertEquals(0, len(results.failures))
+
+    options.page_repeat = 2
+    options.pageset_repeat = 1
+    results = page_runner.Run(Test('RunTest'), ps, expectations, options)
+    self.assertEquals(2, len(results.successes))
+    self.assertEquals(0, len(results.failures))
+
   def testCredentialsWhenLoginFails(self):
     credentials_backend = StubCredentialsBackend(login_return_value=False)
     did_run = self.runCredentialsTest(credentials_backend)