On 02/02/13 04:07, Kelly Hirai wrote:
> fpga seems a natural way to express in silicon, data flow languages like
The Stretch seems to be focussed on video, the examples given to explain their
compiler process show it picking reasonably complex mathematical expressions
inside loops which it then converts to logic with registers in and out, I guess
probably duplicated logic if possible. That way the critical inner loops are
optimised. With video and image recognition stuff it could work very well, there
is lots of iterating over rows of pixels with convolutions and stuff in those
The Xilinx software claims to include a large library of elements to use, not
sure what areas they would cover, and there seems to be some DSP stuff too,
perhaps at extra cost. The example application on the website is building
automotive sound systems ... which would involve filters, eqs, codecs, radio
decoders and so forth.
But the open source stuff seems pretty serious as well, with NASA and the
European Space Agency featuring as contributors. They probably use a lot of DSP,
in many frequency ranges, so maybe they have contributed more than just their
For serious audio the large supply of GPIO pins connected to the FPGA all on the
chip seems to fit a physical control surface with lots of inputs and extremely
low latency very well. And as you say the dataflow languages seem to fit the
FPGA type of machine very well.
But all of this is serious, time consuming, work of course.
Linux-audio-user mailing list