Re: [Starpu-devel] Staticaly distributing task on one given GPU

[Nathalie Furmento <nathalie.furmento@labri.fr>]

On 28/06/2013 16:31, Xavier Lacoste wrote:
> Hello,
>
> For some experiment I need to force some tasks to run on one given GPU.
>
> I found 2 solutions, and I wonder which one is the best or if there is a 
> better one :
>
> - Using one context for each GPU and submiting tasks to that context.
> - Using the can_execute() field of the starpu_codelet_t structure.
>
> any advice ?
>
> Have a nice day,
>
> XL.
> _______________________________________________
> Starpu-devel mailing list
> Starpu-devel@lists.gforge.inria.fr
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/starpu-devel

Static scheduling is documented here:

http://runtime.bordeaux.inria.fr/StarPU/doc/index.html#Static-scheduling

The function can_execute of the codelet is more suited to enable 
implementation according to the capabilities of the devices.

Cheers,

Nathalie