On 01/23/2013 09:04 PM, Brandon Surmanski wrote:
> Considering pocl's main compatibility layer is based on LLVM, and LLVM has
> preexisting backends for NVidea PTX and R600, how much work would be required
> to
> create a device for a GPU? Is there any structural incompatibility currently
> limiting such a device?
There shouldn't be any structural incompatibility as we have tested
pocl with a couple of different heterogeneous style device setups
already (tce and cellspu).
> Would the process of creating a new device (assuming it has some LLVM backend)
> simply rely on creating the device file (eg pthread.{c,h})?
Yes, a device layer implementation which communicates with the device is
needed. The Gallium API could be possibly used for this (I'm not an expert
in it): http://dri.freedesktop.org/wiki/GalliumCompute
It might be possible to share code in the pocl device layer so the
implementation could support all GPUs that have Gallium drivers.
On top of that, possibly some device-specific kernel runtime implementations
e.g., for get_global_id() are needed.
I've wanted to take a poke at this, but it seems I'm not going to have time
for such fun in the near future :I
--
--Pekka
------------------------------------------------------------------------------
Master Visual Studio, SharePoint, SQL, ASP.NET, C# 2012, HTML5, CSS,
MVC, Windows 8 Apps, JavaScript and much more. Keep your skills current
with LearnDevNow - 3,200 step-by-step video tutorials by Microsoft
MVPs and experts. ON SALE this month only -- learn more at:
http://p.sf.net/sfu/learnnow-d2d
_______________________________________________
pocl-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/pocl-devel
