[Mesa-dev] [Mesa-stable] [PATCH] nvc0: do not clear surfaces bins in the validate function

[Samuel Pitoiset]

On 06/05/2016 11:50 PM, Emil Velikov wrote:
> On 5 June 2016 at 22:36, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>> On Sun, Jun 5, 2016 at 5:34 PM, Emil Velikov <emil.l.velikov at gmail.com> wrote:
>>> On 5 June 2016 at 22:17, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>>>> On Sun, Jun 5, 2016 at 5:16 PM, Emil Velikov <emil.l.velikov at gmail.com> wrote:
>>>>> On 5 June 2016 at 22:13, Emil Velikov <emil.l.velikov at gmail.com> wrote:
>>>>>> On 5 June 2016 at 17:56, Samuel Pitoiset <samuel.pitoiset at gmail.com> wrote:
>>>>>>> We should not call nouveau_bufctx_reset() inside a validate function.
>>>>>> This seems to contradict the changes introduced in nvc0_compute.c.
>>>>>> Worth explaining a bit better the dos and don'ts ?
>>>>>>
>>>>> As this is already in master, can you please provide a more
>>>>> elaborate/correct summary for -stable ?
>>>>
>>>> I think it's fine as is.
>>>>
>>>> Do: reset bufctx when setting dirty bit
>>>> Don't: reset bufctx in validate logic, since it's "too late" by then.
>>>> (Not strictly wrong, but just should do it earlier.)
>>>
>>> So nvc0_compute_*validate*_surfaces is not validate logic ? Err...
>>> what a confusing name it has ;-)
>>
>> It validates compute. And it invalidates (and clears) the 3d bin.
>>
> So one can reset_bufctx(3d) from the compute validate and vice-versa.
> While doing reset_bufctx(foo) from foo validate is a bad idea ?
> Shouldn't one just say so in the commit message ?

Because the common practice is to clear foo bins at the same place where 
the dirty_3d |= foo is updated, this makes sense. :)

>
> -Emil
>