[lowrisc-dev] Open GPU for the first CPU

Theo Markettos theo.markettos at cl.cam.ac.uk
Fri Feb 13 19:12:19 GMT 2015


In article <CAHJRFazdKwkdqBWrhBcQk9a+uqv=OOaFryqJCCbf8+CmC-h=Eg at mail.gmail.com> you wrote:
> I don't have strong religion about the license--early on, this actually
> used Apache 2.0. Since I'm the only author right now, that's pretty easy
> to change
> 
> Perhaps this is a foolish question, but what's the concern with LGPL? :)

Trying to avoid questions of religion, it doesn't map well to hardware
because of the need to provide the ability to relink with a modified version
of the library.  I can slightly see what relinking means in an FPGA context
(partial reconfiguration?) but it doesn't make sense for chip fab - are you
going to saw up the die so you can replace a silicon component?

This was one of the motivations for coming up with the BERI
Hardware/Software Licence, which is essentially 'Apache for hardware':
http://www.beri-open-systems.org/legal/beri-hardware-software-license.txt
It also takes care of some other rights beyond copyrights (eg semiconductor
mask rights) that traditional software licences don't cover.

> This project is kinda experimental, because it has no fixed function
> graphics hardware. While I don't think the performance is horrible, I
> wouldn't expect it to be competitive with a traditional embedded GPU for
> graphics workloads, especially at the low end. However, I haven't
> made apples-to-apples comparisons.
> 
> I've been thinking about a new design that looks more like a traditional
> GPU and would be more graphics-centric.

How much of your infrastructure (test/compiler/etc) do you think might be reusable?

Theo
(not on the lowrisc project)



More information about the lowrisc-dev mailing list