Switch low bw audio test to histograms.

Also requires a recipe change so the results processor switches to
histogram mode when this CL is landed.

Bug: chromium:1029452
Change-Id: Ic09deefc3f4f9d7a82ffeafeb5209fcfc361aece
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/171683
Reviewed-by: Artem Titov <titovartem@webrtc.org>
Commit-Queue: Patrik Höglund <phoglund@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#30884}

3 files changed, 60 insertions, 32 deletions

diff --git a/audio/BUILD.gn b/audio/BUILD.gn
index 142b140a44..c90292274a 100644
--- a/audio/BUILD.gn
+++ b/audio/BUILD.gn
@@ -206,7 +206,6 @@ if (rtc_include_tests) {
       if (is_android) {
         deps += [ "//testing/android/native_test:native_test_native_code" ]
       }
-
       data = [
         "../resources/voice_engine/audio_tiny16.wav",
         "../resources/voice_engine/audio_tiny48.wav",
@@ -223,6 +222,8 @@ if (rtc_include_tests) {
         "../resources/voice_engine/audio_tiny16.wav",
         "../resources/voice_engine/audio_tiny48.wav",
       ]
+      data_deps =
+          [ "//third_party/catapult/tracing/tracing/proto:histogram_proto" ]
       if (is_win) {
         data += [ "${root_out_dir}/low_bandwidth_audio_test.exe" ]
       } else {
diff --git a/audio/test/low_bandwidth_audio_test.py b/audio/test/low_bandwidth_audio_test.py
index 44ad1a1b44..0744889c64 100755
--- a/audio/test/low_bandwidth_audio_test.py
+++ b/audio/test/low_bandwidth_audio_test.py
@@ -16,7 +16,6 @@ output files will be performed.
 
 import argparse
 import collections
-import json
 import logging
 import os
 import re
@@ -58,7 +57,7 @@ def _ParseArgs():
   parser.add_argument('--num-retries', default='0',
                       help='Number of times to retry the test on Android.')
   parser.add_argument('--isolated_script_test_perf_output', default=None,
-      help='Path to store perf results in chartjson format.')
+      help='Path to store perf results in histogram proto format.')
   parser.add_argument('--extra-test-args', default=[], action='append',
       help='Extra args to path to the test binary.')
 
@@ -170,7 +169,7 @@ def _RunPesq(executable_path, reference_file, degraded_file,
   if match:
     raw_mos, _ = match.groups()
 
-    return {'pesq_mos': (raw_mos, 'score')}
+    return {'pesq_mos': (raw_mos, 'unitless')}
   else:
     logging.error('PESQ: %s', out.splitlines()[-1])
     return {}
@@ -196,41 +195,66 @@ def _RunPolqa(executable_path, reference_file, degraded_file):
     return {}
 
   mos_lqo, = match.groups()
-  return {'polqa_mos_lqo': (mos_lqo, 'score')}
+  return {'polqa_mos_lqo': (mos_lqo, 'unitless')}
 
 
-def _AddChart(charts, metric, test_name, value, units):
-  chart = charts.setdefault(metric, {})
-  chart[test_name] = {
-      "type": "scalar",
-      "value": value,
-      "units": units,
-  }
+def _MergeInPerfResultsFromCcTests(histograms, run_perf_results_file):
+  from tracing.value import histogram_set
 
-
-def _AddRunPerfResults(charts, run_perf_results_file):
+  cc_histograms = histogram_set.HistogramSet()
   with open(run_perf_results_file, 'rb') as f:
-    per_run_perf_results = json.load(f)
-  if 'charts' not in per_run_perf_results:
-    return
-  for metric, cases in per_run_perf_results['charts'].items():
-    chart = charts.setdefault(metric, {})
-    for case_name, case_value in cases.items():
-      if case_name in chart:
-        logging.error('Overriding results for %s/%s', metric, case_name)
-      chart[case_name] = case_value
+    contents = f.read()
+    if not contents:
+      return
+
+    cc_histograms.ImportProto(contents)
+
+  histograms.Merge(cc_histograms)
 
 
 Analyzer = collections.namedtuple('Analyzer', ['name', 'func', 'executable',
                                                'sample_rate_hz'])
 
 
+def _ConfigurePythonPath(args):
+  script_dir = os.path.dirname(os.path.realpath(__file__))
+  checkout_root = os.path.abspath(
+      os.path.join(script_dir, os.pardir, os.pardir))
+
+  sys.path.insert(0, os.path.join(checkout_root, 'third_party', 'catapult',
+                                  'tracing'))
+  sys.path.insert(0, os.path.join(checkout_root, 'third_party', 'protobuf',
+                                  'python'))
+
+  # The low_bandwidth_audio_perf_test gn rule will build the protobuf stub for
+  # python, so put it in the path for this script before we attempt to import
+  # it.
+  histogram_proto_path = os.path.join(
+      args.build_dir, 'pyproto', 'tracing', 'tracing', 'proto')
+  sys.path.insert(0, histogram_proto_path)
+
+  # Fail early in case the proto hasn't been built.
+  from tracing.proto import histogram_proto
+  if not histogram_proto.HAS_PROTO:
+    raise ImportError('Could not find histogram_pb2. You need to build the '
+                      'low_bandwidth_audio_perf_test target before invoking '
+                      'this script. Expected to find '
+                      'histogram_pb2.py in %s.' % histogram_proto_path)
+
+
 def main():
   # pylint: disable=W0101
   logging.basicConfig(level=logging.INFO)
 
   args = _ParseArgs()
 
+  _ConfigurePythonPath(args)
+
+  # Import catapult modules here after configuring the pythonpath.
+  from tracing.value import histogram_set
+  from tracing.value.diagnostics import reserved_infos
+  from tracing.value.diagnostics import generic_set
+
   pesq_path, polqa_path = _GetPathToTools()
   if pesq_path is None:
     return 1
@@ -250,14 +274,14 @@ def main():
   if polqa_path and _RunPolqa(polqa_path, example_path, example_path):
     analyzers.append(Analyzer('polqa', _RunPolqa, polqa_path, 48000))
 
-  charts = {}
-
+  histograms = histogram_set.HistogramSet()
   for analyzer in analyzers:
     # Start the test executable that produces audio files.
     test_process = subprocess.Popen(
         _LogCommand(test_command + [
             '--sample_rate_hz=%d' % analyzer.sample_rate_hz,
-            '--test_case_prefix=%s' % analyzer.name
+            '--test_case_prefix=%s' % analyzer.name,
+            '--write_histogram_proto_json'
           ] + args.extra_test_args),
         stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
     perf_results_file = None
@@ -279,9 +303,12 @@ def main():
         analyzer_results = analyzer.func(analyzer.executable,
                                          reference_file, degraded_file)
         for metric, (value, units) in analyzer_results.items():
-          # Output a result for the perf dashboard.
+          hist = histograms.CreateHistogram(metric, units, [value])
+          user_story = generic_set.GenericSet([test_name])
+          hist.diagnostics[reserved_infos.STORIES.name] = user_story
+
+          # Output human readable results.
           print 'RESULT %s: %s= %s %s' % (metric, test_name, value, units)
-          _AddChart(charts, metric, test_name, value, units)
 
         if args.remove:
           os.remove(reference_file)
@@ -291,13 +318,13 @@ def main():
     if perf_results_file:
       perf_results_file = _GetFile(perf_results_file, out_dir, move=True,
                            android=args.android, adb_prefix=adb_prefix)
-      _AddRunPerfResults(charts, perf_results_file)
+      _MergeInPerfResultsFromCcTests(histograms, perf_results_file)
       if args.remove:
         os.remove(perf_results_file)
 
   if args.isolated_script_test_perf_output:
-    with open(args.isolated_script_test_perf_output, 'w') as f:
-      json.dump({"format_version": "1.0", "charts": charts}, f)
+    with open(args.isolated_script_test_perf_output, 'wb') as f:
+      f.write(histograms.AsProto().SerializeToString())
 
   return test_process.wait()
 
diff --git a/audio/test/pc_low_bandwidth_audio_test.cc b/audio/test/pc_low_bandwidth_audio_test.cc
index 37c80860ff..aafb65f15d 100644
--- a/audio/test/pc_low_bandwidth_audio_test.cc
+++ b/audio/test/pc_low_bandwidth_audio_test.cc
@@ -105,7 +105,7 @@ std::string AudioOutputFile() {
 
 std::string PerfResultsOutputFile() {
   return webrtc::test::OutputPath() + "PCLowBandwidth_perf_" +
-         FileSampleRateSuffix() + ".json";
+         FileSampleRateSuffix() + ".pb";
 }
 
 void LogTestResults() {