MatrixComputation

The project performs a benchmark test for matrices computation in Java, C++, Eigen, OpenCV, RenderScript on Android platform

The purpose of this application is not to compare the performance of algorithms performing matrix multiplication. All computations are implemented in naive way

======

CONTENT

1. [Application Manifest] (#1)
2. [Java] (#2)
3. [C++] (#3)
4. [OpenCV] (#4)
5. [Eigen] (#5)
6. [RenderScript] (#6)
7. [Contacts] (#7)

======

APPLICATION MANIFEST

• Target SDK Version: 19
• Minimum SDK Version: 16
• Uses Permission
  • android.permission.READ_EXTERNAL_STORAGE
  • android.permission.WRITE_EXTERNAL_STORAGE
  • android.permission.READ_LOGS

Java

This is the simplest way to perform matrices computation on Android devices. Yet, the performance is the worst among all approaches demonstrated in this application.

C++

This approach is faster than Java. Note that Java Native Interface is needed in order to run native code in Android applications.

OpenCV

OpenCV is common library for mathematic computations. Details and tutorial about [OpenCV on Android] (http://opencv.org/platforms/android.html) can be found [here] (http://docs.opencv.org/doc/tutorials/introduction/android_binary_package/O4A_SDK.html). NOTE OpenCV is not available in release 1.2

Eigen

[Eigen] (http://eigen.tuxfamily.org/index.php?title=Main_Page) is another choice for computation. Eigen usually outperform OpenCV on Android devices.

RenderScript

[RenderScript] (http://developer.android.com/guide/topics/renderscript/compute.html) is a framework for running computationally intensive tasks at high performance on Android, usually used for image processing on Android devices. It accelerates the computation by running on GPUs. Please note that RenderScript is device-dependent. Version problem limits the compatibility of RenderScript.

======

Contacts

Ching-Lun Lin, clallenlin@sv.cmu.edu, 2014