Re: [Discuss-gnuradio] build gnuradio on windows with cygwin mingw msvc

[Martin Dvh]

Thanks for the info,

> > The cygwin wiki seems out of date (talks about gnuradio-0.9)
> Which cygwin wiki? The afore-mentioned? I've updated it 2 days ago.
Yes, I missed the "Starting with" in this line:
   Starting with GnuRadio version 0.9, there's no need for the ipc-daemon 
service.

> cppunit and fftw3 patches have been submitted upstream.
> Hopefully they will be in next release.
> I have patched versions on my site though: http://f8cfe.free.fr/ham/gnuradio/
This is very usefull, thank you very much

> I'm taking the opportunity to call for volunteers for testing
> the low-level USRP under Windows. 
Sorry, I don't have an usrp(yet). Still working with my hacked bttv TVcapture 
card as frontend and ADC.


> > As a sidenote, I was able to compile a  part of gnuradio with microsoft 
> > visual C++.net 2003
> > Just made a project and added all relevant cpp and .h files
>
> This is promising!
> We should be very close to it, as gnuradio-core compiles with MinGW
> (until the link stage which fails for obscure reason). 
> If you plan to use GNU Radio with Visual C++, you have to know that you
> won't be able to link against MinGW or Cygwin generated libraries,
> because the C++ ABI differs.
Yes, I know.
mingw and visual c++ can sometimes be combined but this is very hackish, cygwin 
and visual c++ will never work.
I will settle for any solution (all cygwin, all mingw or all visual c++)
As long as it is able to combine with opengl (I am working on a fir 
implementation on the GPU)
I suspect that combining cygwin with opengl might be troublesome.

> Chances are the only doprod available was the generic one.
Yes, but the dotprod performance is why I am trying to build under windows.
I am working on a fir (dotprod and more) implementation on the GPU (using 
several Cg fragment shaders).
To do this I need render-to-texture which at the moment is only available with 
the windows drivers of ATI and nvidia.
(This is going to change as soon as FBO (Framebuffer objects) is implemented in 
the linux drivers which is expected to happen in the next months)
I already have the most part of the fir implementation ready but now I want to 
integrate it with the rest of gnuradio to start testing.
The expected theoretical maximum performance of this GPU implementation is 
about 800 Mtaps/sec for a fcc fir on my Geforce FX 5700 LE.
This is higher then my cpu (amd duron 800) can do and the cpu can be used for 
other things.
Other parts of gnuradio could be implemented on the GPU too possibly leading to 
a complete SDR on the GPU.
It might even be possible to hack the driver of any frontend to do direct DMA 
to the videocards memory without using the cpu.
(inspired by the linux bttv driver which can capture directly to the video 
framebuffer)
In a sense, the videocards processing power could replace the fpga.
The only thing needed then is a dumb pci or pci-express ADC card with 
appropriate frontend.

The other way around (TX)is also possible.
Make an interesting signal with the cpu.
Send it to the GPU for upconverting using a fragment shader.
Send it to the world using the (secondary) VGA output.
My videocards ramdac can do 400Ms/s and it has Three channels(colors) per vga 
output.
So I could do three channels at once or one complex (0-400 Mhz) an one real 
(0-200 Mhz) channel.

Best Regards,

Martin Dvh