[PATCH] dri2proto: add prime protocol support. (v2.1)

[Dave Airlie]

On Fri, Jul 6, 2012 at 4:02 PM, Kristian Høgsberg <krh at bitplanet.net> wrote:
> On Fri, Jul 6, 2012 at 9:41 AM, Dave Airlie <airlied at gmail.com> wrote:
>> On Fri, Jul 6, 2012 at 2:32 PM, Kristian Høgsberg <krh at bitplanet.net> wrote:
>>> On Fri, Jul 6, 2012 at 6:27 AM, Dave Airlie <airlied at gmail.com> wrote:
>>>> From: Dave Airlie <airlied at redhat.com>
>>>>
>>>> So we reserve bits 16->19 for offload device ids, this means we can
>>>> have 6 offload devices, which is plenty for now, and we can bump
>>>> this further later without fear.
>>>>
>>>> v2: I suck at maths, that is all. Fixed up the maths to match reality.
>>>> v2.1: fix typo
>>>
>>> Looks fine to me.  Is there a version bump that goes along with this
>>> so drivers can know they can ask for prime devices or should they just
>>> try and expect a NULL reply if it's not available?
>>
>> Drivers don't ever know about prime, its all hidden in the
>> server/libGL/dri2 layers.
>
> Ah, I meant client side / aiglx there, ie how does the caller know
> that there are prime device types available.  But just asking and
> getting NULL sounds fine.

It out of band via randr it can enumerate the list of offload slaves.

Really the client side only knows by being told to explicitly try via
the env var DRI_PRIME, as apps aren't expected to know/care about what
they are talking to, its really up to the user/DE to pick which GPU to
launch the app on.

maybe if we add support for GLX_AMD_gpu_association one day we'll need
some more enumeration options.

Dave.
Dave.
>
> Kristian