[Mesa-dev] [PATCH 1/3] vl/dri3: use external texture as back buffers(v4)

[Andy Furniss]

Nayan Deshmukh wrote:
> Hi Andy,
>
> Can you try this patch? This should help with the tearing.

Patch seems to be good - I get page flipping again so DC, modesetting
and "normal" setup all work OK.

>
> diff --git a/src/gallium/state_trackers/vdpau/output.c
> b/src/gallium/state_trackers/vdpau/output.c
> index 48e3133..98a8011 100644
> --- a/src/gallium/state_trackers/vdpau/output.c
> +++ b/src/gallium/state_trackers/vdpau/output.c
> @@ -82,7 +82,7 @@ vlVdpOutputSurfaceCreate(VdpDevice device,
>      res_tmpl.depth0 = 1;
>      res_tmpl.array_size = 1;
>      res_tmpl.bind = PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET |
> -                   PIPE_BIND_SHARED;
> +                   PIPE_BIND_SHARED | PIPE_BIND_SCANOUT;
>      res_tmpl.usage = PIPE_USAGE_DEFAULT;
>
>      pipe_mutex_lock(dev->mutex);
>
> Regards,
> Nayan
>
> On Wed, Jan 11, 2017 at 5:11 PM, Andy Furniss <adf.lists at gmail.com> wrote:
>> Michel Dänzer wrote:
>>>
>>> On 11/01/17 05:13 PM, Nayan Deshmukh wrote:
>>>>
>>>> On Wed, Jan 11, 2017 at 12:44 PM, Michel Dänzer <michel at daenzer.net>
>>>> wrote:
>>>>>
>>>>> On 10/01/17 06:53 PM, Nayan Deshmukh wrote:
>>>>>>
>>>>>> On Sat, Jan 7, 2017 at 12:42 PM, Michel Dänzer <michel at daenzer.net>
>>>>>> wrote:
>>>>>>>
>>>>>>> On 06/01/17 05:50 AM, Andy Furniss wrote:
>>>>>>>>
>>>>>>>> Christian König wrote:
>>>>>>>>>
>>>>>>>>> Am 04.01.2017 um 18:13 schrieb Nayan Deshmukh:
>>>>>>>>>>
>>>>>>>>>> dri3 allows us to send handle of a texture directly to X
>>>>>>>>>> so this patch allows a state tracker to directly send its
>>>>>>>>>> texture to X to be used as back buffer and avoids extra
>>>>>>>>>> copying
>>>>>>>>>>
>>>>>>>>>> v2: use clip width/height to display a portion of the surface
>>>>>>>>>> v3: remove redundant variables, fix wrapping, rename variables
>>>>>>>>>>        handle vaapi path
>>>>>>>>>> v3.1: we need clip_width/height for every frame so we don't need
>>>>>>>>>>          to maintain it for each buffer instead use a global
>>>>>>>>>> variable
>>>>>>>>>> v4: In case of single gpu we can cache the buffers as applications
>>>>>>>>>>        use constant number of buffer and we can avoid calls to
>>>>>>>>>> present
>>>>>>>>>>        extension for every frame
>>>>>>>>>>
>>>>>>>>>> Suggested-by: Leo Liu <leo.liu at amd.com>
>>>>>>>>>> Signed-off-by: Nayan Deshmukh <nayan26deshmukh at gmail.com>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Acked-by: Christian König <christian.koenig at amd.com>.
>>>>>>>>>
>>>>>>>>> Andy & Leo did you guys already had a chance to test it? To me it
>>>>>>>>> looks
>>>>>>>>> like this should work now.
>>>>>>>>
>>>>>>>>
>>>>>>>> Well there is still the tearing issue from loosing pageflips.
>>>>>>>>
>>>>>>>> Maybe different GPUs don't see this. I can fix by forcing perf but I
>>>>>>>> just tested dal and it's not even fixable running that.
>>>>>>>>
>>>>>>>> I guess that may not count as an issue with these patches as such if
>>>>>>>> xorg/xf86-video-amdgpu can work around, but it's a very noticeable
>>>>>>>> regression until that happens.
>>>>>>>
>>>>>>>
>>>>>>> Somebody should track down why the buffers sent for presentation in
>>>>>>> this
>>>>>>> case don't use the same tiling parameters as buffers used for GL via
>>>>>>> DRI3.
>>>>>>>
>>>>>> I can look into this, but I don't know where to look exactly. Can you
>>>>>> give some
>>>>>> pointers to get started.
>>>>>
>>>>>
>>>>> Looking at src/gallium/auxiliary/vl/vl_winsys_dri3.c and the patches
>>>>> again, my guess is that it's due to PIPE_BIND_SCANOUT not being set when
>>>>> creating the buffers that are now being directly sent to the X server
>>>>> for presentation.
>>>>>
>>>> So the only way to avoid this is to have a PIPE_BIND_SCANOUT for the
>>>> output surfaces of the state tracker. Will introducing
>>>> PIPE_BIND_SCANOUT lead to performance loss for these surfaces?
>>>
>>>
>>> Potentially, but I doubt it'll make a big difference for this use case.
>>> In the future, there might be a feedback mechanism which allows
>>> re-allocating the buffer with/out PIPE_BIND_SCANOUT according to the
>>> current circumstances, but for now it's probably better to set it (at
>>> least in cases where we don't know that the buffer can never be scanned
>>> out directly) to allow for page flipping.
>>
>>
>> Pure luck noticing this because I haven't tested modesetting driver for
>> ages, but -
>>
>> These patches also break full screen vdpau playback when using that.
>>
>> Result is a screen of mostly junk with a hint of the vid - looks like
>> when direct scan out fails on wayland due to tiling mismatch.
>>
>