[MPlayer-users] Rather high CPU usage on arm (Raspberry Pi) when streaming.

[David Woodfall]

>On Fri, Jun 06, 2014 at 06:53:25AM +0100, David Woodfall wrote:
>> I'm finding CPU usage is between 20-40% on my Pi board when streaming
>> radio URLs:
>>
>> mplayer -playlist  http://www.bbc.co.uk/radio/listen/live/r3.asx
>>
>> is a good example. Local files tend to be ~= the 10% mark.
>
>Too little information I am afraid.
>Did you try capturing with -dumpstream and then playing that file?
>I doubt your local files are in WMA format, in which case the comparison
>doesn't make sense.

Indeed a stream dump plays with the same CPU %.

>Also cpu frequency scaling unless you have disabled it will result in
>CPU usage numbers being completely misleading.
>The commands "perf record mplayer ..." and then after it is finished
>"perf report" will tell where all the CPU time is used.
>Also, which MPlayer version you are using would be helpful information.
>However with the ridiculously outdated CPU of the Pi that kind of CPU
>usage is most likely as expected, as it doesn't even include NEON,
>which causes issues for the data conversions required.
>In addition if whatever distribution you uses uses "softfloat" instead
>of "hardfloat" that is likely to cost you a significant amount of
>performance, as well as if your MPlayer isn't compiled specifically
>for it (runtime CPU detection is a mess and basically does not work
>on ARM, as a result generic distribution binaries a generally useless
>performance-wise).

Yes, I think the softfloat is the problem here. I'm using
armed-slack (Slackware) which looks like it's compiled with soft.

I'm not sure if I can do anything about it for individual
applications, but I'll look into it.