[pulseaudio-discuss] Bluetooth hfp high cpu load

[Jonas Potgrave (Riedel)]

Hi all,

I am experiencing very high cpu load (100%, just pulseaudio) on an embedded system in hfp mode (call audio). A2dp mode which runs on a much higher sample rate (but just in one direction) in comparison uses only 30% cpu. Playing wav files is also no problem.

My system specs are:
- pulseaudio 9.0
- bluez 5.41
- ofono 1.18
- ARM Cortex-A8 embedded system
- soundcard used is a dummy driver, communicating over bidirectional 8 channel TDM with an FPGA.
- soundcard is running at 48k sample rate and s32le format
- pulseaudio adapts these settings during alsa init (alsa-sink, alsa-source)

I've tried messing with the settings (e.g. tsched, fragments, resampling). Of those only resampling has an effect (different profiling load distribution). But trivial and speex-1 are equally bad (trivial uses multiple divisions that are not optimized for ARM).

Today I even patched out the resampling completely. Still uses 100%. Profiling shows the cpu usage is broadly distributed across multiple pulseaudio fuctions.

Does anyone have an idea why the performance is so bad in hfp only?

Regards,
Jonas