Hi everybody,I’m a student who is completing his master degree of
computeer engineering (now I’m at UPC-Barcelona, but before I’ve studied
at polytechnic of Turin,Italy).For my thesis, one of my possible idea is
to bring part of GNURADIO over CUDA(I’ve see that there are already
different contributions/papers about it). In this way, my work won’t be
completely useless, as lot of other thesis.In order to do this, one
possible schedule could be:
-do the porting of some existing blocks (or set of modules over CUDA),
which don’t require an enviroment too much CUDA-friendly. In this way,I
hope to get more confidence and start to have problems.-after one or two
month, start to work above schedule(and/or similar stuffs) in order to
avoid host-device memory transfer when there are CUDA blocks in
cascade, and here I will get tons of problems…(for istance,tagging
single packets could be not good for warp divergence)…there could be
different architectural choice,like a kernel for each block,or marging
them togheter(taking advantages from warp) and use a “dynamic”
compilation of device code(CUDA).-other stuffs,optimizations…
I know that some of you can tell me about OpenCL, but I didn’t have any
experience with it, so I prefer CUDA. I want also to minimize problems
which will arise, because surely I will gett lot of them…
I know also that some of you will say that VOLK gives good
performance,but why don’t allow users to get power from both?(maybe when
you want to elaborate different signals at same time).For the moment,
this is not my final choice about my thesis(I’ve also another proposal
about another field, but it wil be useless after the thesis).In any
case, I want to ask you if this could be interesting for your
community.Also because for the moment I have only an EZCAP
tuner(RTL2838),which is fine for tv,audio signals,etc… But probably I
will ask to some of you if you can sniff some frequency(like 802.11) and
give me files, in order to test the correct works also for these
modules.I will have also different choice to do, and it could be
interesting to hear diferent point of view.
For the moment, thanks for your time,marco ribero
Hey Marco,
good luck for your project!
FYI, there have been quite a few projects using GPPs in the past. I
recommend going through the slides from all of the GRCons, you’ll find
something there (at the very least from the first GRCon).
Also, there’s gr-theano, which also uses CUDA, and for inspiration,
having a look at gr-fosphor might be a good idea.
For high-performance GR flow graphs, the overhead for ingress/egress has
always been the problem. If you have ideas for modifying GR itself, it
might be worth contacting the “Accelerators” working group, which
discusses these matters.
Cheers,
M