Re: [Discuss-gnuradio] portaudio sink, partial succes on windows

[Martin Dvh]

Robert McGwier wrote:

> Thanks to tons of grunt work by Eric,  we have used the highly tuned 
> ring buffers in the audio_portaudio_sink code and it functions nicely 
> with a little bit of operator help.    I can now get solid performance 
> on all Linux hosts with the example code
>
> mono_tone_portaudio.py and multi_tone_portaudio.py.
>
> All worked on oss, alsa, and jack.
>
> Eric got similar performance.
>
> Thomas Schmid ran mono_tone_portaudio.py successfully on his mac with 
> coreaudio automatically identified as the host.  We had to modify the 
> buffer size calculation (by forcing it).  This will improve over the 
> next few days as the audio_portaudio_source gets built and these very 
> large portaudio messages  which are delivered in device and stream info 
> structs are parsed more carefully.
>
> It is not clear to me that we want to prefer portaudio talking to jack 
> over directly talking to jack.  This might change if we (Stephane,  me, 
> others)  improve the portaudio jack handler.   For now,  my thinking is 
> we are probably better off and we will be more versatile talking 
> directly to jack without portaudio but we can do this kind of thinking 
> after we get this portaudio stuff fully stabilized.
>
> I am hoping the Windows folks can bring their code up now as well.  
> Portaudio is really solid on Windows as I have been using it there for 
> ASIO, MME,  and DirectX for two years and after I helped Eric Wachsman 
> get a leg up (by leaning on my friends for help) he got WDM-KS running.  
> WDM-KS is the single lowest latency sound host I have seen in ANY 
> personal computer but let me say that I have never used a Mac.  If 
> anyone working on the windows code needs a dll, export library,  
> Microsoft .NET 2003 project for v-19 devel, please let me know.  We 
> should be able to build the dll under mingw but I have not tried yet.  
> That said,  all of this portaudio stuff about latency this and that make 
> very little sense for what we are doing.  What we need is not 
> necessarily very low latency but rock solid performance using buffer 
> sizes that make sense for our applications.
A partial success on windows.
I have portaudio running on windows (just the wmme driver) and for the first 
time I hear sounds without too much crackles here.

I needed only a few tweaks.

Most important, don't use standard cvs HEAD of portaudio but the special cvs 
v19-devel version
I missed the following comment in one of the portaudio mails
>It will be important to stay current:
>
>cvs -d:pserver:address@hidden:/home/cvs co -r v19-devel portaudio

I had done a normal cvs HEAD checkout and got loads of problems first.
You really need the -r v19-devel  options.
Once that was sorted out I did the following.
This is all using mingw:
Update/build/install gnuradio-core
do a configure/build/install for portaudio-v19-devel (only default configure 
options, so it is using the wmme driver)
Then I had to make the following symbolic link:
/usr/local/lib/pkgconfig/portaudio-2.0.pc  -> 
/usr/local/lib/pkgconfig/portaudio.pc

then checkout gr-audio-portaudio
do a bootstrap
Check that the latest libtool version is used
do a configure/build/install

update gnuradio-examples
run:
dial_tone_portaudio.py
mono_tone_portaudio.py
multi_tone_portaudio.py
All run fine ;-)

modify usrp/usrp_wfm_rcv.py to use portaudio
run it
listen to my favourite FM station without too much crackles.
(The audio is not completely clean yet, but probably the ks driver will help 
here)

now run usrp/usrp_wfm_pll_PA.py
runtime error popup immediately

change from non-blocking to blocking portaudio
runtime error popup immediately

disable fft displays
runs ok for 5 to 20 seconds
after that a runtime error popup

enable non-blocking again
runtime error popup immediately


The runtime error  I get is a popup:
(also see attached image runtime_error.gif)
"Microsoft Visual C++ Runtime Library

Runtime Error!
Program: d:\Python24\python.exe
abnormal program termination"

So it only works with blocking enabled and no ffts, and only for a few seconds.
I think it has to do with something not keeping up.
I can force this error by clicking on any other open program window or moving 
another window.

I then tried opening/moving windows while running multi_tone.py and this causes 
multi_tone.py to stop too with the same runtime error.
Running a high computational load in the background doesn't seem to hurt 
multi_tone.py.

Many thanks for making this work, so far.

No it is time for finetuning and getting rid of problems like these.

I people want to try this at home, I have binary installers for windows 
available for this partially working portaudio driver.


Greetings,
Martin