Re: [fluid-dev] High DSP load

[josh]

Quoting Pedro Lopez-Cabanillas <address@hidden>:
> What about adding a call to jack_cpu_load() in fluid_jack.c, and logging the
> results?

Good idea.  Just did that, ran some tests, saved the output numbers to  
files and graphed the results from both versions using gnuplot:
plot "FluidSynth-1.1.0.txt" with lines, "FluidSynth-1.0.9.txt" with lines

http://resonance.org/~josh/FluidSynth-AutumnLeaves-graph.png

Very little difference in my case.  In fact 1.0.9 spikes higher than  
1.1.0 for that test on my system.


> > Could you try this?
> ...
> > #define fluid_synth_is_synth_thread(_s)     TRUE
> > #define fluid_synth_should_queue(_s)   FALSE
>
> Very little difference, if any at all.

Good to hear.  That confirms that the event queuing has very little overhead.

> I've also tested 1.1.0 in another system right now, with very   
> similar results.
> The first test machine is an AMD Athlon XP 2800+ Speed: 2 GHz
> The second test machine is an Intel Core2Duo T7200 Speed: 2 GHz

I should probably try running FluidSynth on some other machines to see  
if I can replicate what you are seeing.

> Regards,
> Pedro


Here is the code fragment I added at the top of  
fluid_jack_audio_driver_process() in fluid_jack.c:

  static int stats_frame_counter = 0;
  float cpuload;

  stats_frame_counter += nframes;

  if (stats_frame_counter > 48000)
  {
    cpuload = jack_cpu_load (dev->client);
    printf ("%0.3f\n", cpuload);
    stats_frame_counter = 0;
  }


Regards,
Josh