Forum: GNU Radio GNU Radio/Parallella project for GSoC 2013

A35252079e3891e673d08c79e36d95c4?d=identicon&s=25 Roman Redkovich (Guest)
on 2013-04-18 09:28
(Received via mailing list)
Hello, I'm starting this thread to discuss ideas about
support for many-core floating-points accelerators.
As it was said in wiki it could be for example:
  -further development of Performance Counters
   and Block Core Affinity
  -implementing message passing using on-chip networks

http://forums.parallella.org/viewtopic.php?f=20&t=268
0817fa933c74eec9b3fcf3a04e16f418?d=identicon&s=25 Sylvain Munaut (Guest)
on 2013-04-18 09:57
(Received via mailing list)
Hi,

> Hello, I'm starting this thread to discuss ideas about
> support for many-core floating-points accelerators.
> As it was said in wiki it could be for example:
>         -further development of Performance Counters
>          and Block Core Affinity

Shouldn't we work on actually _running_ any of the blocks on the
accelerator before trying to benchmark them ?

Cheers,

   Sylvain
7be2b94c643faaf06871527bc98daa9b?d=identicon&s=25 Andrew Back (Guest)
on 2013-04-18 10:18
(Received via mailing list)
Hi Sylvain,

On 18 April 2013 08:56, Sylvain Munaut <246tnt@gmail.com> wrote:
> Hi,
>
>> Hello, I'm starting this thread to discuss ideas about
>> support for many-core floating-points accelerators.
>> As it was said in wiki it could be for example:
>>         -further development of Performance Counters
>>          and Block Core Affinity
>
> Shouldn't we work on actually _running_ any of the blocks on the
> accelerator before trying to benchmark them ?

Of course, but the suggestion was that a GSoC project involving
Parallella should do more than just port blocks  this could be part
of the project but not all. I believe Tommy Tracy II has started
working on porting, but I am not sure how far he will get before a
student would need to start their project.

Cheers,

Andrew

--
Andrew Back
http://carrierdetect.com
C539637020fd56193dd6daec746c4a84?d=identicon&s=25 Tom Rondeau (Guest)
on 2013-04-18 15:16
(Received via mailing list)
On Thu, Apr 18, 2013 at 4:16 AM, Andrew Back <andrew@carrierdetect.com>
wrote:
>>
>
> Andrew


Yes. We certainly need to get blocks ported over to the chip to do
anything. From a GNU Radio standpoint as a GSoC project, that's not
particularly interesting to us, just work that has to be done. What
we're interested in as a project is improving our understanding of
both many-core processors as well as the use of coprocessors for
signal processing/SDR work.

Tom
E16be4811324adf8f26be26d77e9d29d?d=identicon&s=25 Robert McGwier (Guest)
on 2013-04-19 03:59
(Received via mailing list)
Or possibly more interesting and significant for the future,
heterogeneous
processors in a system.  We're seeing lots of this with ARM's on FPGA's,
GPU on ARM based SoC, .......  In addition to homogeneous multicore
processors with interesting network fabrics........
34ca4305e0456e8510a1d4d7009a12d0?d=identicon&s=25 Tommy Tracy II (Guest)
on 2013-04-19 04:04
(Received via mailing list)
Ideally, an architecture-aware scheduler could capitalize on all
available accelerators / heterogenous processors, and could maximize
performance for that system.

Unfortunately right now we're limited to porting a subset of blocks to
these accelerators.

Tommy James Tracy II
Ph.D Student
High Performance Low Power Lab
University of Virginia
Phone: 913-775-2241
Please log in before posting. Registration is free and takes only a minute.
Existing account

NEW: Do you have a Google/GoogleMail, Yahoo or Facebook account? No registration required!
Log in with Google account | Log in with Yahoo account | Log in with Facebook account
No account? Register here.