OpenCV for Tegra 2
I would like to build OpenCV for the Tegra 2 platform, whether running WCE7 or Linux. What is the best way to ensure obtaining the highest performance ?
I would like to build OpenCV for the Tegra 2 platform, whether running WCE7 or Linux. What is the best way to ensure obtaining the highest performance ?
The major drawback of the Tegra 2 platform, is that it doesn't have NEON unit. That means that you can't use SIMD instructions to optimize the code. GLSL optimizations are available on Android, Tegra 3, but there just a few of them (see this article).
So, the only available option is to enable multi-threading optimizations. There are not so many of them in OpenCV right now, but they are added from time to time, and upcoming 2.4.3 release will have some new parallel implementations.
And finally, check that you're using proper compiler and compilation flags, because they may affect your performance. For instance, use RELEASE build, enable optimization flags, etc. You can try to build OpenCV4Android from sources and check in CMake output which gcc options are used.
BTW, to control your performance, I would suggest you to run OpenCV performance tests. Look for "Performance Tests" section on this page.
Asked: 2012-10-24 04:16:56 -0600
Seen: 340 times
Last updated: Oct 25 '12
CARMA DevKit Tegra 3 NEON optimizations
OpenCV optimizations on Android Tegra3 platform
Why does my Android program fire "Tegra compatible GPU found" like crazy?
Cannot run OpenCV samples on Tegra T4i device
Can't run sample on Tegra K1 / Nexus 9