double precision FP and operations

[dhardy]

As I understand it, epiphany only supports single-precision (32-bit) FP with a limited number of operations (correct?). So are there plans to support double-precision along with division, log, exp (and thus pow) in a future generation? Can they be efficiently implemented anyway? These are key to some categories of scientific computing (and probably a few other things).

[ysapir]

[noryn]

Hello
We have several libraries for TI fixed and floating point DSPs. You can see the details here: http://www.sundancedsp.com/sundance-libraries

One of these libraries peforms IEEE-754 floating point format numbers on a fixed point DSPs. We are planning to port several of these libraies including the GDD600 to this platform if we see reasonable interest. We appreciate your feedbacks to help us make the rght decision.

Rgds

Nory
tel: 775 8273103

[piotr5]

decision-making is simple: if you believe in the success of this project, then prepare something for people to buy. otherwise you can still jump on the brandwagon lateron too. however, with all details laying open, you will have a hard time competing with open-source. maybe you could contribute something open-source for promotion, to show the quality of your products? guess if you could write a 32bit division-algorithm, one that is faster than the one currently used, and maybe make it even smaller too...

[noryn]

Thanks. It is not a matter of faith and believing. The platofrm may be very successful but users may not have any need/requirement for the libraries we are considering. That is why it would be encouraging to us if we hear that people are actually asking for usch dev tools for this processor.

Cheers

Nory