[Mesa-dev] [PATCH 3/4] ac/llvm: set xnack like radeonsi does.

Marek Olšák maraeo at gmail.com
Thu Jul 6 20:50:40 UTC 2017


On Thu, Jul 6, 2017 at 10:14 PM, Matt Arsenault <arsenm2 at gmail.com> wrote:
>
>> On Jul 6, 2017, at 13:08, Dave Airlie <airlied at gmail.com> wrote:
>>
>> On 7 July 2017 at 05:07, Matt Arsenault <arsenm2 at gmail.com> wrote:
>>>
>>>> On Jul 5, 2017, at 19:09, Dave Airlie <airlied at gmail.com> wrote:
>>>>
>>>> From: Dave Airlie <airlied at redhat.com>
>>>>
>>>> Use family, but only set xnack+ for gfx9.
>>>>
>>>
>>> The driver shouldn’t be explicitly setting this. This should be set as part of the subtarget chosen
>>
>> Well I expect Marek knows more (I'm just aligning the drivers).
>>
>> If I had to guess, it's probably because Carrizo sets xnack in the
>> subtarget but we don't want to use it.
>>
>> Dave.
>
> This is the problematic part, if you explicitly disable a subtarget feature when it is set in the subtarget’s feature list, it for some reason disables all of the features in the subtarget

I don't understand that. Wouldn't it break everything?

xnack should be a function attribute, because xnack is a per-VMID
setting that can be changed anytime. For example, you can have a
compute VMID where xnack is disabled and another compute VMID where
xnack is enabled.

Marek


More information about the mesa-dev mailing list