Home → Geekbench →

Wrong ARM version detection

• Edit

Cyril Bianconi

26 Jan, 2012 06:32 PM

Dear Geekbench developpers,

Thanks for your great SW.

I've run your benchmark on an ARM-based device (cannot provide more info as the device has not yet been officially announced) running Android and I reach a very poor score.

I did some reverse engineering on the apk and I found that GeekBench is using a lib called libgeekbench-jni.so
In fact, there are 2 libs with the same name. One for armeabi and another one for armeabi-v7a.
I copied my armeabi-v7a into the armeabi one and my perfs improved drastically.

It looks like that GeekBench is doing an arm type detection and it does not select the right lib based on the data that I'm reporting.
Could you please provide me more data on how this detection is done and which value do you expect to use the armeabi-v7a lib as this has never been an issue with the other benchmarks.

Thanks for your support.

1. Support Staff 1 Posted by John on 27 Jan, 2012 08:01 AM

   

   Hi Cyril,

   Thanks for your message!

   Unfortunately Geekbench doesn't pick which native library to load when running the benchmarks; that decision is made by Android itself. I'd recommend getting in touch with the device vendor and seeing what they have to say about this problem, as it might affect other applications that use JNI libraries.

   Best,
   John

   • Edit

2. 2 Posted by sirsolano on 12 Mar, 2012 05:13 PM

   

   Hi all,

   The issue as been reported to Google as a generic issue for ICS:
   http://code.google.com/p/android/issues/detail?id=25321

   Here is the resolution:
   From: [email blocked]
   "I confirm that this affects all ICS platform releases (4.0.1 to 4.0.3). The problem is not directly related to the NDK.
   A fix has been submitted internally, and will appear in the next minor platform release (no ETA for now).
   See the android-ndk thread for discussion and possible work-arounds.
   Sorry about that."

   Workaround:
   https://groups.google.com/group/android-ndk/browse_thread/thread/37...

   The workaround should take place in the app; let us know if Primate Labs will be considering patching it.

   Thanks,
   Jorge

   • Edit

3. Support Staff 3 Posted by John on 12 Mar, 2012 06:16 PM

   

   Hi Jorge,

   Interesting! Thanks for letting us know about the problem, and I'll look into adding the workaround to Geekbench.

   Thanks,
   John

   • Edit

4. John closed this discussion on 09 Jul, 2012 12:24 AM.