Added GPU performance metrics.

tools/perf/metrics/gpu_timeline.py

Index: tools/perf/metrics/gpu_timeline.py
diff --git a/tools/perf/metrics/gpu_timeline.py b/tools/perf/metrics/gpu_timeline.py
new file mode 100644
index 0000000000000000000000000000000000000000..78e2dee9a7b4b56b9298a19762329406b87d4661
--- /dev/null
+++ b/tools/perf/metrics/gpu_timeline.py
@@ -0,0 +1,217 @@
+# Copyright 2015 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 math
+import sys
+
+from telemetry.timeline import model as model_module
+from telemetry.timeline import async_slice as async_slice_module
+from telemetry.value import scalar
+from telemetry.value import list_of_scalar_values
+from telemetry.web_perf.metrics import timeline_based_metric
+
+TOPLEVEL_GL_CATEGORY = 'gpu_toplevel'
+TOPLEVEL_SERVICE_CATEGORY = 'disabled-by-default-gpu.service'
+TOPLEVEL_DEVICE_CATEGORY = 'disabled-by-default-gpu.device'
+
+SERVICE_FRAME_END_MARKER = (TOPLEVEL_SERVICE_CATEGORY, 'SwapBuffer')
+DEVICE_FRAME_END_MARKER = (TOPLEVEL_DEVICE_CATEGORY, 'SwapBuffer')
+
+TRACKED_NAMES = { 'RenderCompositor': 'render_compositor',
+                  'BrowserCompositor': 'browser_compositor',
+                  'Compositor': 'browser_compositor' }
+
+GPU_SERVICE_DEVICE_VARIANCE = 5
+
+
+def CalculateFrameTimes(events_per_frame):
+  """Given a list of events per frame, returns a list of frame times."""
+  times_per_frame = []
+  for event_list in events_per_frame:
+    # Prefer to use thread_duration but use duration as fallback.
+    event_times = [(event.thread_duration or event.duration)
+                   for event in event_list]
+    times_per_frame.append(sum(event_times))
+  return times_per_frame
+
+
+def TimelineName(name, source_type, value_type):
+  """Constructs the standard name given in the timeline.
+
+  Args:
+    name: The name of the timeline, for example "total", or "render_compositor".
+    source_type: One of "cpu", "gpu" or None. None is only used for total times.
+    value_type: the type of value. For example "mean", "stddev"...etc.
+  """
+  if source_type:
+    return '%s_%s_%s_time' % (name, value_type, source_type)
+  else:
+    return '%s_%s_time' % (name, value_type)
+
+
+class GPUTimelineMetric(timeline_based_metric.TimelineBasedMetric):
+  """Computes GPU based metrics."""
+
+  def __init__(self):
+    super(GPUTimelineMetric, self).__init__()
+
+  def AddResults(self, model, _, interaction_records, results):

[nednguyen, 2015/01/24 00:32:22]

Does this implementation support overlapped interaction records? If not, you may
want to make the call self.VerifyNonOverlappedRecords

[David Yen, 2015/01/24 00:41:10]

Done.

+    service_times = self._CalculateGPUTimelineData(model)
+    for value_item, durations in service_times.iteritems():
+      count = len(durations)
+      avg = 0.0
+      stddev = 0.0
+      maximum = 0.0
+      if count:
+        avg = sum(durations) / count
+        stddev = math.sqrt(sum((d - avg) ** 2 for d in durations) / count)
+        maximum = max(durations)
+
+      name, src = value_item
+
+      if src:
+        frame_times_name = '%s_%s_frame_times' % (name, src)
+      else:
+        frame_times_name = '%s_frame_times' % (name)
+
+      if durations:
+        results.AddValue(list_of_scalar_values.ListOfScalarValues(
+            results.current_page, frame_times_name, 'ms', durations))
+
+      results.AddValue(scalar.ScalarValue(results.current_page,
+                                          TimelineName(name, src, 'max'),
+                                          'ms', maximum))
+      results.AddValue(scalar.ScalarValue(results.current_page,
+                                          TimelineName(name, src, 'mean'),
+                                          'ms', avg))
+      results.AddValue(scalar.ScalarValue(results.current_page,
+                                          TimelineName(name, src, 'stddev'),
+                                          'ms', stddev))
+
+  def _CalculateGPUTimelineData(self, model):
+    """Uses the model and calculates the times for various values for each
+       frame. The return value will be a dictionary of the following format:
+         {
+           EVENT_NAME1: [FRAME0_TIME, FRAME1_TIME...etc.],
+           EVENT_NAME2: [FRAME0_TIME, FRAME1_TIME...etc.],
+         }
+
+       Event Names:
+         mean_frame - Mean time each frame is calculated to be.
+         mean_gpu_service-cpu: Mean time the GPU service took per frame.
+         mean_gpu_device-gpu: Mean time the GPU device took per frame.
+         TRACKED_NAMES_service-cpu: Using the TRACKED_NAMES dictionary, we
+                                    include service traces per frame for the
+                                    tracked name.
+         TRACKED_NAMES_device-gpu: Using the TRACKED_NAMES dictionary, we
+                                   include device traces per frame for the
+                                   tracked name.
+    """
+    all_service_events = []
+    current_service_frame_end = sys.maxint
+    current_service_events = []
+
+    all_device_events = []
+    current_device_frame_end = sys.maxint
+    current_device_events = []
+
+    tracked_events = {}
+    tracked_events.update(dict([((value, 'cpu'), [])
+                                for value in TRACKED_NAMES.itervalues()]))
+    tracked_events.update(dict([((value, 'gpu'), [])
+                                for value in TRACKED_NAMES.itervalues()]))
+
+    current_tracked_service_events = collections.defaultdict(list)
+    current_tracked_device_events = collections.defaultdict(list)
+
+    event_iter = model.IterAllEvents(
+        event_type_predicate=model_module.IsSliceOrAsyncSlice)
+    for event in event_iter:
+      # Look for frame end markers
+      if (event.category, event.name) == SERVICE_FRAME_END_MARKER:
+        current_service_frame_end = event.end
+      elif (event.category, event.name) == DEVICE_FRAME_END_MARKER:
+        current_device_frame_end = event.end
+
+      # Track all other toplevel gl category markers
+      elif event.args.get('gl_category', None) == TOPLEVEL_GL_CATEGORY:
+        base_name = event.name
+        dash_index = base_name.rfind('-')
+        if dash_index != -1:
+          base_name = base_name[:dash_index]
+        tracked_name = TRACKED_NAMES.get(base_name, None)
+
+        if event.category == TOPLEVEL_SERVICE_CATEGORY:
+          # Check if frame has ended.
+          if event.start >= current_service_frame_end:
+            if current_service_events:
+              all_service_events.append(current_service_events)
+              for value in TRACKED_NAMES.itervalues():
+                tracked_events[(value, 'cpu')].append(
+                    current_tracked_service_events[value])
+            current_service_events = []
+            current_service_frame_end = sys.maxint
+            current_tracked_service_events.clear()
+
+          current_service_events.append(event)
+          if tracked_name:
+            current_tracked_service_events[tracked_name].append(event)
+
+        elif event.category == TOPLEVEL_DEVICE_CATEGORY:
+          # Check if frame has ended.
+          if event.start >= current_device_frame_end:
+            if current_device_events:
+              all_device_events.append(current_device_events)
+              for value in TRACKED_NAMES.itervalues():
+                tracked_events[(value, 'gpu')].append(
+                    current_tracked_device_events[value])
+            current_device_events = []
+            current_device_frame_end = sys.maxint
+            current_tracked_device_events.clear()
+
+          current_device_events.append(event)
+          if tracked_name:
+            current_tracked_device_events[tracked_name].append(event)
+
+    # Append Data for Last Frame.
+    if current_service_events:
+      all_service_events.append(current_service_events)
+      for value in TRACKED_NAMES.itervalues():
+        tracked_events[(value, 'cpu')].append(
+            current_tracked_service_events[value])
+    if current_device_events:
+      all_device_events.append(current_device_events)
+      for value in TRACKED_NAMES.itervalues():
+        tracked_events[(value, 'gpu')].append(
+            current_tracked_device_events[value])
+
+    # Calculate Mean Frame Time for the CPU side.
+    frame_times = []
+    if all_service_events:
+      prev_frame_end = all_service_events[0][0].start
+      for event_list in all_service_events:
+        last_service_event_in_frame = event_list[-1]
+        frame_times.append(last_service_event_in_frame.end - prev_frame_end)
+        prev_frame_end = last_service_event_in_frame.end
+
+    # Create the timeline data dictionary for service side traces.
+    total_frame_value = ('total', None)
+    cpu_frame_value = ('total', 'cpu')
+    gpu_frame_value = ('total', 'gpu')
+    timeline_data = {}
+    timeline_data[total_frame_value] = frame_times
+    timeline_data[cpu_frame_value] = CalculateFrameTimes(all_service_events)
+    for value in TRACKED_NAMES.itervalues():
+      cpu_value = (value, 'cpu')
+      timeline_data[cpu_value] = CalculateFrameTimes(tracked_events[cpu_value])
+
+    # Add in GPU side traces if it was supported (IE. device traces exist).
+    if all_device_events:
+      timeline_data[gpu_frame_value] = CalculateFrameTimes(all_device_events)
+      for value in TRACKED_NAMES.itervalues():
+        gpu_value = (value, 'gpu')
+        tracked_gpu_event = tracked_events[gpu_value]
+        timeline_data[gpu_value] = CalculateFrameTimes(tracked_gpu_event)
+
+    return timeline_data