Perf test to measure loading CDM library.

chrome/browser/media/load_cdm_benchmark_test.cc

Index: chrome/browser/media/load_cdm_benchmark_test.cc
diff --git a/chrome/browser/media/load_cdm_benchmark_test.cc b/chrome/browser/media/load_cdm_benchmark_test.cc
new file mode 100644
index 0000000000000000000000000000000000000000..af5be06ece7517dd0e9e0a02e2a5de973fd3290a
--- /dev/null
+++ b/chrome/browser/media/load_cdm_benchmark_test.cc
@@ -0,0 +1,39 @@
+// Copyright (c) 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.
+
+#include "base/files/file_path.h"
+#include "base/path_service.h"
+#include "base/scoped_native_library.h"
+#include "base/strings/utf_string_conversions.h"
+#include "base/time/time.h"
+#include "testing/gmock/include/gmock/gmock.h"
+#include "testing/perf/perf_test.h"
+
+
+const char kWidevineAdapterName[] = "widevinecdm";

[ddorwin, 2014/02/06 19:57:33]

This file is specific to WV, but the filename is not, and it's really only line
20 that is.

Maybe we can extract a test function and call it with the name of the library we
want to load.

That would also make it easy to run this on ECK, which could serve as a baseline
of the difference between the actual and simplest CDMs.

[shadi1, 2014/02/07 23:37:17]

Done.

+
+TEST(CDMBinLoadPerfTest, Basic) {

[ddorwin, 2014/02/06 19:57:33]

Do we need "Bin"? It seems redundant, but if so, spell out Binary.

[shadi1, 2014/02/07 23:37:17]

Done.

+  base::NativeLibrary native_library;
+  base::FilePath adapter_full_name;
+  adapter_full_name = base::FilePath::FromUTF16Unsafe(
+      base::GetNativeLibraryName(base::ASCIIToUTF16(kWidevineAdapterName)));

[shadi1, 2014/02/01 01:50:07]

This does the magic of appending "lib", "dll", "plugin", etc. magic based on
platform. 

I can replace this with kWidevineCdmFileName but then this will be dependent on
widevine_cdm_common.h as well.

[ddorwin, 2014/02/06 19:57:33]

Coo, maybe we should use this instead of having a different constant for each
platform. I guess that's slightly worse performance for production, though.
Shouldn't this file be dependent on widevine_cdm_version.h, which includes
widevine_cdm_common.h anyway? This test should not exist if
!WIDEVINE_CDM_AVAILABLE, right?

We also need to exclude platforms that don't ship the CDM as a library - IOW,
run it only when ENABLE_PEPPER_CDMS.

+  base::FilePath output_dir;
+  EXPECT_TRUE(PathService::Get(base::DIR_MODULE, &output_dir));
+  base::FilePath adapter_path = output_dir.Append(adapter_full_name);
+  std::string error;
+  base::TimeTicks start = base::TimeTicks::HighResNow();
+  native_library = base::LoadNativeLibrary(adapter_path, &error);
+  double delta = (base::TimeTicks::HighResNow() - start).InMillisecondsF();
+  EXPECT_TRUE(native_library);
+  if (!native_library) {

[ddorwin, 2014/02/06 19:57:33]

How about combining these lines?
  EXPECT_TRUE(native_library) << "Error loading adapter:\n" << error;

Note: adapter is misspelled in the original CL.

[shadi1, 2014/02/07 23:37:17]

Done.

+    std::cout << "Error loading apadter:\n" << error;
+  } else {
+    perf_test::PrintResult("time_to_load_adapter",

[ddorwin, 2014/02/06 19:57:33]

I'm not sure how unique this needs to be, but we should probably name it
time_to_load_widevine_cdm. (adapter is not really relevant to the metric and
there could be other cdm adapters.)

Or does l34 make this unique...? In that case: ...s/_adapter/_cdm/.

[shadi1, 2014/02/07 23:37:17]

True about the other parameter that makes it unique. I dropped the _cdm from the
metric. With current CL the output is something like this:
*RESULT time_to_load: widevinecdm= 525.772 ms
*RESULT time_to_load: clearkeycdm= .848 ms

+                           "",
+                           kWidevineAdapterName,
+                           delta,
+                           "ms",
+                           true);
+  }
+}