Add multi-architecture support in telemetry.

tools/telemetry/telemetry/core/platform/profiler/android_profiling_helper.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 glob
 import hashlib
 import logging
 import os
 import platform
 import re
@@ skipping 207 matching lines @@
 
 def PrepareDeviceForPerf(device):
   """Set up a device for running perf.
 
   Args:
     device: DeviceUtils instance identifying the target device.
 
   Returns:
     The path to the installed perf binary on the device.
   """
-  android_prebuilt_profiler_helper.InstallOnDevice(device, 'perf')
+  perf_binary = 'perf_' + device.GetProp('ro.product.cpu.abi')

[tonyg, 2014/10/10 16:51:14]

This is a general problem, not specific to the perf binary or to the android
platform.

Take a look at support_binaries.py and how it chooses the folder automatically
based on the platform. I had pictured that we'd add another level of nesting for
the architecture -- here:

def _GetBinPath(binary_name, platform_name):
  # TODO(tonyg): Add another nesting level for architecture_name.
  return os.path.join(path.GetTelemetryDir(), 'bin', platform_name, binary_name)

So then instead of renaming perf.sha1, we'd move it to
bin/android/armeabi-v7a/perf.sha1.

Would you be willing to rework this CL to generalize this?

[pasko, 2014/10/10 17:04:07]

Oops, I also recently added perfhost_precise and perfhost_trusty :)

I am slightly biased towards perf_armeabi-v7a because in the future we might
need to support multiple ABIs on the single device, which probably means
multiple perf binaries installed side by side (not 100% sure about that). We
would probably never need to run two different perf binaries concurrently
anyway, but maybe it would be easier to tell which binary is used on the device
by looking at the filename itself or the logs on the bots.

Other applications do similar things. For example, the GCC binary I use is named
as "arm-linux-androideabi-gcc".

But I am not strongly opposed to the alternative you proposed, it would work
either way.

+  android_prebuilt_profiler_helper.InstallOnDevice(device, perf_binary)
   # Make sure kernel pointers are not hidden.
   device.WriteFile('/proc/sys/kernel/kptr_restrict', '0', as_root=True)
-  return android_prebuilt_profiler_helper.GetDevicePath('perf')
+  return android_prebuilt_profiler_helper.GetDevicePath(perf_binary)
 
 
 def GetToolchainBinaryPath(library_file, binary_name):
   """Return the path to an Android toolchain binary on the host.
 
   Args:
     library_file: ELF library which is used to identify the used ABI,
         architecture and toolchain.
     binary_name: Binary to search for, e.g., 'objdump'
   Returns:
@@ skipping 18 matching lines @@
     return None
   toolchain_version = toolchain_version.group(1)
 
   path = os.path.join(util.GetChromiumSrcDir(), 'third_party', 'android_tools',
                       'ndk', 'toolchains',
                       '%s-%s' % (toolchain_config, toolchain_version),
                       'prebuilt', '%s-%s' % (host_os, host_machine), 'bin',
                       '%s-%s' % (toolchain_config, binary_name))
   path = os.path.abspath(path)
   return path if os.path.exists(path) else None