[Mesa-dev] Potentially EOL ilo gallium driver

[Ilia Mirkin]

On Tue, Dec 6, 2016 at 3:00 PM, Rob Clark <robdclark at gmail.com> wrote:
> On Tue, Dec 6, 2016 at 2:11 PM, Jason Ekstrand <jason at jlekstrand.net> wrote:
>> On Tue, Dec 6, 2016 at 8:39 AM, Rob Clark <robdclark at gmail.com> wrote:
>>>
>>> On Tue, Dec 6, 2016 at 8:42 AM, Emil Velikov <emil.l.velikov at gmail.com>
>>> wrote:
>>> > On 6 December 2016 at 03:16, Edward O'Callaghan
>>> > <funfunctor at folklore1984.net> wrote:
>>> >> This patch is to potentially remove ourself from the maintaince
>>> >> burden of the ilo driver that appears to now be essentially
>>> >> unmaintained?
>>> >>
>>> >> I am not sure of our policy here or if there are too many
>>> >> users so this patch is really only to gauge a response of
>>> >> how folks feel?
>>> >>
>>> > Surely you want to CC the core/sole developer of the driver when
>>> > considering its removal.
>>> > Maybe mailman was "nice" and hid his email in the header ;-)
>>> >
>>> > Either way adding Chia-I Wu to the list.
>>> >
>>> > -Emil
>>> > P.S. Not sure/sold how much of an actual burden the driver is, yet I
>>> > don't make serious gallium infra changes.
>>>
>>> really hasn't been a problem for me..
>>>
>>> That said, it would be nice if someday someone wired this up to use
>>> glsl_to_nir path in gallium and re-used i965's nir backend.  I think
>>> that would make ilo somewhat more interesting..
>>
>>
>> We had a bit of a chat about this on IRC and what I told Ilia there was that
>> the more interesting thing to do, if someone really wanted to do Intel on
>> gallium, would probably be to build a new driver based on ISL, blorp, the
>> i965 compiler, NIR, and genxml.  We've made a pretty good driver-building
>> toolbox.  Having an almost unmaintained driver that has it's own hand-rolled
>> and inferrior compiler, surface layout, etc. isn't doing much good.
>>
>
> yeah, reusing the other bits would be nice too, and hopefully would be
> the long term goal if someone where to spend time on this.. I guess
> I'd prefer a more incremental approach of converting parts one by one
> if I were doing it myself.  It's kind of a moot point either way until
> someone has time/motivation to spend on it.
>
> But I've no real objection to dropping ilo until then if others feel
> strongly.. it's still there in git history so it can be resurrected if
> someone wants to convert to reuse other i965 bits incrementally rather
> than starting from scratch.

As mentioned on IRC, I think the real use-case that ilo could cover
that i965/anv can't (easily) handle is acting as a gallium-nine
backend. (I know someone's working on DX9 over vulkan, but that's
hardly ready, and will never be available on gen6.)

However at this time, it's not sufficiently functional to handle
gallium-nine, so I don't see any serious downside to dropping it.

  -ilia