Add a startup benchmark and rudimentary scripts for running benchmarks.

benchmarks/startup/run.py

+# Copyright 2014 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 imp
+import os
+import sys
+import subprocess
+import timeit
+
+try:
+  imp.find_module('mopy')
+except ImportError:
+  sys.path.append(os.path.abspath(os.path.join(
+      __file__, os.pardir, os.pardir, os.pardir, 'mojo', 'tools')))
+from mopy.paths import Paths
+
+
+def run(args):
+  if args.release:
+    build_directory = os.path.join('out', 'Release')
+  else:
+    build_directory = os.path.join('out', 'Debug')
+  paths = Paths(build_directory)

[abarth-chromium, 2014/11/14 04:44:13]

Why not just pass in |paths| instead of making the benchmark do this work.  It
seems like every benchmark will need to repeat the boiler plate.

[yzshen1, 2014/11/14 18:00:15]

Done.

+
+  rounds = 1000
+
+  # Because mojo_benchmark_startup terminates the process immediately when its
+  # MojoMain() is called. The overall execution time reflects the startup
+  # performance of the mojo shell.
+  startup_time = timeit.timeit(
+      ("subprocess.call(['%s', 'mojo:mojo_benchmark_startup'])" %
+           paths.mojo_shell_path),
+      "import subprocess", number=rounds)
+
+  # The execution time of a noop executable is also measured, in order to offset
+  # the cost of timeit()/subprocess.call()/etc.
+  noop_time = timeit.timeit(
+      ("subprocess.call(['%s'])" %
+           os.path.join(paths.build_dir, 'mojo_benchmark_startup_noop')),
+      "import subprocess", number=rounds)
+
+  # TODO(yzshen): Consider also testing the startup time when
+  # mojo_benchmark_startup is served by an HTTP server.
+
+  # Convert the execution time to milliseconds and compute the average for
+  # a single run.
+  result = (startup_time - noop_time) * 1000 / rounds
+  return ("Result: rounds tested: %d; average startup time: %f ms" %
+          (rounds, result))