[LAD] Mixing audio: Implementing pan and balance

Raphaël BOLLEN raphael.bollen at mobistar.be
Sun Mar 24 13:01:27 UTC 2013


On 03/24/2013 12:56 PM, Fons Adriaensen wrote:
> On Sun, Mar 24, 2013 at 01:57:18AM +0100, Jörn Nettingsmeier wrote:
>
>> On 03/23/2013 06:05 PM, Raphaël BOLLEN wrote:
>>
>>> Awesome, however cpu usage is ~12% on my system when playing silence and
>>> drops to ~2% with some audio going through.
>>
>> smells like a denormals issue.
>
> It sure does, but I can't reproduce it on my P4 based system which is
> normally very picky about denormals - CPU load is around 2.5% all the
> time. Also checked all parts of the code that could produce denormals:
> the HP processing filters, the k-meter and c-meter DSP code, and the
> volume controls. All have denormal protection.
>
> * Is the CPU load you mention the one measured by Jack, or some other ?

measured by jack 0.121.0 as reported by QJackCtl on 64bits system

> * Try to find out which controls affect the CPU load. E.g. what happens
> when you have a signal but the volume is set to off etc.

As I said, without any input connected or only streaming silence to input 1, the cpu % rises by ~10%.
To summarize:

- no input connected: cpu rise 10%
- silence on input 1: cpu rise 10%
- audio (not silence) on input 1: no cpu rise
- silence on 1 ch of input 1, audio on the other channel: cpu rise by ~5%, same if you reverse the 
input channel

other inputs are not affected by this. They can be left un-connected or streamed with silence cpu % 
doesn't change.

With the patch I sent, you can leave input 1 un-connected cpu % stays at ~2% but if you stream 
silence to it cpu % rises again to ~10%

With the modifs suggested by Hermann, problem is gone.

Cheers

--
Raphaël.

>
> Ciao,
>



More information about the Linux-audio-dev mailing list