Report the duration of sampled tasks with histogram.

components/scheduler/renderer/task_duration_reporter.cc

+// Copyright 2016 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.
+
+#include "components/scheduler/renderer/task_duration_reporter.h"
+
+#include "base/metrics/histogram_macros.h"
+
+namespace scheduler {
+namespace {
+  const int kSampleInterval = 100;

[tdresser, 2016/07/27 18:04:25]

This shouldn't be indented, should it? Play around with "git cl format", or
integrating clang format into your editor.

[sunyunjia, 2016/07/27 19:51:04]

Done.

+}
+
+  TaskDurationReporter::TaskDurationReporter() : sample_count_(0) {}

[majidvp, 2016/07/27 18:34:02]

nit: same as above. Should not be indented.

[sunyunjia, 2016/07/27 19:51:04]

Done.

+
+  TaskDurationReporter::~TaskDurationReporter() {}
+
+  void TaskDurationReporter::ReportSampleDuration(base::TimeDelta duration) {
+    ++sample_count_;
+    if (sample_count_ == kSampleInterval) {
+      // [1msec ~ 10sec], 50 bins. All tasks longer than 10sec will be put into
+      // the same bin.
+      UMA_HISTOGRAM_TIMES("RendererScheduler.TaskDuration", duration);

[tdresser, 2016/07/27 18:04:25]

For input latency, we use 

UMA_HISTOGRAM_CUSTOM_COUNTS(                                                \
      name, (end.event_time - start.event_time).InMicroseconds(), 1, 1000000, \
      100);

That would probably be reasonable for this metric as well. I think we do want to
register tasks shorter than 1ms.

Whoever reviews histograms.xml will have feedback on whether or not that's
reasonable.

[sunyunjia, 2016/07/27 19:51:04]

Done.

+      sample_count_ = 0;
+    }
+  }
+}