>From: Jens M Andreasen <jens.andreasen(a)chello.se>
>
>Here is an early paper:
>
> http://www.cs.washington.edu/homes/oskin/thompson-micro2002.pdf
>
>It is about general-purpose-computing of which audio-processing is a
>subset. (BTW: Check out the aging references on the last page.)
I found nothing on the audio processing. So, the paper may be next to
nothing. But it was said that the GPU can be used as general purpose
computational unit for other applications.
In Europe, e.g., an apparatus, in which analog parts are replaced
by their digital equivalents, is not patentable. (Source: a patent
officer in a magazine.) So, if that big change is not patentable,
then why would a move of the audio application to GPU? Specially after
it is already said it could be used as an alternative to CPU.
We should just ignore the patents and keep up with the audio research.
Prior the patent the research has moved to GPU. We should not wait
20 years before we take first steps in researching the possibilities
of the GPUs.
Juhana
--
http://music.columbia.edu/mailman/listinfo/linux-graphics-dev
for developers of open source graphics software
The following proggi does the job:
#include <xmmintrin.h>
#define _MM_DENORM_ZERO_ON 0x0040
main()
{
// enable flush to zero
_mm_setcsr(_MM_FLUSH_ZERO_ON | _MM_MASK_UNDERFLOW | _mm_getcsr());
// enable denormals are zero
_mm_setcsr(_MM_DENORM_ZERO_ON | _mm_getcsr());
}
Note: you will need compiler option -mfpmath=sse and -march=pentium4
(works for both P4 and Athlon64)
After reboot, both FZ and DAZ flags are reset to zero, so you should run this
program in your /etc/profile. But as said earlier, works only for
applications that use (are compiled to use) the SSE unit instead of the old
style X86 FPU.
Here is one of music-dsp articles (critical date becomes important).
http://aulos.calarts.edu/pipermail/music-dsp/2004-May/027009.html
Perhaps they read that article, patented and implemented during the
summer. I don't know.
Juhana
== cut ==
Bill Baldwin wbaldwin at austin.rr.com
Sat May 1 15:18:11 PDT 2004
It's definitely possible, but far from convenient or optimal with today's
hardware.
The h/w is steadily getting better, but you'd likely run into FP precision
issues on many cards.
The new pixel shader 3.0 h/w is starting to appear and will become more
common over the next year - this extends the instruction length of shaders,
adds better flow control, and lots of other stuff that increases the
potential for doing interesting audio processing.
Getting the data in and out is another issue - you could treat an audio
channel as a 1-dimensional texture, turn off h/w filtering, etc. - but you'd
end up locking resources very frequently to write your input and read back
your output, which tends to stall the graphics pipeline and thus slow things
down.
Still, it's worth exploring given that all that GPU power that your
sequencer is ignoring...
-Bill
-----Original Message-----
On Behalf Of vesa norilo
Hi all,
Has anyone thought about it? I bet someone has.
There's plenty of horsepower sitting idle in PCs with modern video
cards. The in them shaders are completely programmable, but I don't
really know about streaming data to and fro. I just thought that it
would be very neat to write a VSTi synth that runs on, say, Radeon X800
or GF6800. Is it possible at all?
Vesa
http://shoko.calarts.edu/musicdsphttp://ceait.calarts.edu/mailman/listinfo/music-dsp
== end ==
Hello, everybody!
I am developing a small speech server for emacspeak. For audio output I
am using /dev/dsp. But I don't know how to stop playback exactly in
moment I choose. Please tell me how to implement correctly this behavior
or give me the link where I can read information about it.
--
Best wishes. Michael Pozhidaev. E-mail: msp(a)altlinux.ru.
Tomsk state university.
Computer science department. (http://www.inf.tsu.ru)