Dear Group,
I’m comparing the speed of the BBB and Raspberry-Pi when running various audio encoders. Normally the BBB wins by a factor of 2 or 3 as expected, but in the case of oggenc (from the vorbis-tools package) the BBB is actually slightly slower. I’ve tried recompiling the vorbis-tools package, but I must be missing some optimisation. Can anybody give me a hint?
The BBB is running Robert C Nelson’s Debian Wheezy image (2013-09-27 I think) at 1000MHz and the RPi is running Raspbian (recompiled Debian Wheezy) at 800MHz.
Recompilation of vorbis-tools:
`
$ sudo apt-get build-dep vorbis-tools
$ sudo apt-get --build source vorbis-tools
$ sudo dpkg -i vorbis-tools_1.4.0-1_armhf.deb
`
Dear Group,
I'm comparing the speed of the BBB and Raspberry-Pi when running various
audio encoders. Normally the BBB wins by a factor of 2 or 3 as expected,
but in the case of oggenc (from the vorbis-tools package) the BBB is
actually slightly slower. I've tried recompiling the vorbis-tools package,
but I must be missing some optimisation. Can anybody give me a hint?
Does it use a crap-load of 'double'?
Can you try the compiler flags "-ffast-math
-fsingle-precision-constant -fshort-double" for both targets?
http://forum.kde.org/viewtopic.php?f=74&t=117429 has some related background.
The double vs. float is just a guess, but there aren't many reasons
you'll see any code run close to as fast on a Pi as it does on a Bone.
Thanks for the suggestion. I’ll look into it and post again if I have any success.