[Pkg-opencl-devel] opencl-icd selection

[Giuseppe Bilotta]

On Mon, Nov 17, 2014 at 7:47 PM, Rebecca N. Palmer
<rebecca_palmer at zoho.com> wrote:
> As the discrete GPU is likely to be faster than the integrated GPU on
> systems with both, I'd suggest sorting nvidia, mesa, beignet if we find more
> than one platform with a GPU device.

While I understand the motivations, I'm honestly a little wary of
adding this policy straight into ocl-icd. For example, one might argue
that 1.2 platforms are better than 1.1 platforms, or that devices
which are not in use by the display are preferred to ones in use by
the display --unless of course you also want to do GL sharing. And what
about amd (fglrx) vs nvidia? Which one should be preferred and why?
(Yes, I have a couple of machines where I have discrete GPUs from both
vendors installed).

Maybe this could be implemented at the distribution level via a
default OCL_ICD_VENDORS= env variable in /etc/profile.d instead?
(After I've actually implemented sorting via the OCL_ICD_VENDORS env
variable, of course.)


-- 
Giuseppe "Oblomov" Bilotta