Re: [fluid-dev] problems with fluidsynth 1.1.6 on a raspberry pi

[Pedro Lopez-Cabanillas]

El Domingo, 18 de noviembre de 2012 12:51:58 David Henningsson escribió:
> I'll do some experiments with it when I have some more time.

My bet is floating point exceptions. We have already discussed this theory 
before [1] without fixing too much, and users continue reporting similar 
problems, so this is my advice if you are looking for clues. 

If you are running Fluidsynth on some Intel CPU, you may try to build 
FluidSynth using the following options:

$ cmake -DCMAKE_BUILD_TYPE=Debug \
 -Denable-debug=on \
 -Denable-floats=on \
 -Denable-fpe-check=on \
 -DCMAKE_C_FLAGS=-mno-sse2

The FPE_CHECK option enables debug code on FluidSynth checking floating point  
exceptions into the log, but it is only for Intel processors. Sorry, we don't  
have anything like this debug option for the ARM architecture, but hopefully 
fixing denormals for one CPU would help others as well.

The last option (no SSE2) disables the FP optimizations on modern Intel CPUs, 
because SSE2 enabled CPUs are inmune to denormals. If you have older Intels, 
you don't need this option in order to get the exceptions. That means: using 
the default compiler settings, or your distro's Fluid package, you are likely 
going to suffer the FPE troubles out of the box.

Playing some MIDI songs on the command line with FluidSynth compiled in this 
way, you will get on the console messages like these:

fluidsynth: warning: FPE exception (before or in voice_filter): Denormal 
number
fluidsynth: warning: FPE exception (before or in ??? Remainder of 
synth_one_block ???): Denormal number 

This can be verified with the song "Nights of white satin" referenced by 
David, or even better with this Salsa:
http://www.help4web.net/webmaster/sound/Salsa.mid

Now, once you have got your stream of FPE Exceptions on your log, please post 
it here on the mailing list.

[1] http://lists.nongnu.org/archive/html/fluid-dev/2012-05/msg00016.html

Regards,
Pedro