Telemetry: adds CPU frequency stats.

tools/perf/metrics/cpu.py

 # 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.
 
 from metrics import Metric
 
 class CpuMetric(Metric):
   """Calulates CPU load over a span of time."""
 
   def __init__(self, browser):
@@ skipping 12 matching lines @@
   def Stop(self, page, tab):
     assert self._start_cpu, 'Must call Start() first'
     self._results = _SubtractCpuStats(self._browser.cpu_stats, self._start_cpu)
 
   # Optional argument trace_name is not in base class Metric.
   # pylint: disable=W0221
   def AddResults(self, tab, results, trace_name='cpu_utilization'):
     assert self._results, 'Must call Stop() first'
     # Add a result for each process type.
     for process_type in self._results:
+      if process_type == 'Global':
+        continue
       trace_name_for_process = '%s_%s' % (trace_name, process_type.lower())
       cpu_percent = 100 * self._results[process_type]
       results.Add(trace_name_for_process, '%', cpu_percent,
                   chart_name='cpu_utilization', data_type='unimportant')
+    # Add a result for Global CPU stats.
+    for k, v in self._results['Global']['GlobalCpuFrequencyStats'].iteritems():
+      results.Add(k, 'ms', v,

[tonyg, 2014/04/18 03:58:56]

I think this needs to either be a histogram result type or else be a single
average_frequency_hz number. In its current form, the data wouldn't visualize
well on the dashboard and html results.

[bulach, 2014/04/22 12:43:10]

ouch, indeed.... averaging it out as average_frequency_hz

+                  chart_name='cpu_freq', data_type='unimportant')
 
 
 def _SubtractCpuStats(cpu_stats, start_cpu_stats):
   """Computes average cpu usage over a time period for different process types.
 
   Each of the two cpu_stats arguments is a dict with the following format:
       {'Browser': {'CpuProcessTime': ..., 'TotalTime': ...},
        'Renderer': {'CpuProcessTime': ..., 'TotalTime': ...}
        'Gpu': {'CpuProcessTime': ..., 'TotalTime': ...}}
 
   The 'CpuProcessTime' fields represent the number of seconds of CPU time
   spent in each process, and total time is the number of real seconds
   that have passed (this may be a Unix timestamp).
 
   Returns:
     A dict of process type names (Browser, Renderer, etc.) to ratios of cpu
     time used to total time elapsed.
   """
   cpu_usage = {}
   for process_type in cpu_stats:
     assert process_type in start_cpu_stats, 'Mismatching process types'
     # Skip any process_types that are empty.
-    if (not cpu_stats[process_type]) or (not start_cpu_stats[process_type]):
+    if ((not cpu_stats[process_type]) or
+        (not start_cpu_stats[process_type]) or
+        process_type == 'Global'):
       continue
     cpu_process_time = (cpu_stats[process_type]['CpuProcessTime'] -
                         start_cpu_stats[process_type]['CpuProcessTime'])
     total_time = (cpu_stats[process_type]['TotalTime'] -
                   start_cpu_stats[process_type]['TotalTime'])
     assert total_time > 0, 'Expected total_time > 0, was: %d' % total_time
     cpu_usage[process_type] = float(cpu_process_time) / total_time
+
+  if 'Global' not in cpu_stats:
+    return cpu_usage
+
+  frequency_stats = cpu_stats['Global']['GlobalCpuFrequencyStats']
+  start_frequency_stats = start_cpu_stats['Global']['GlobalCpuFrequencyStats']
+  total_frequency_stats = {}
+  for k in frequency_stats.iterkeys():
+    total_frequency_stats[k] = frequency_stats[k] - start_frequency_stats[k]
+
+  cpu_usage['Global'] = {}
+  cpu_usage['Global']['GlobalCpuFrequencyStats'] = total_frequency_stats
   return cpu_usage
-