[Mesa-dev] [PATCH v3] st/vdpau: change the order in which filters are applied(v3)

[Nayan Deshmukh]

Agreed.

I will file a bug report for the chroma artifacts and submit a patch
which uses temporary surfaces to
while writing. Meanwhile, you can revert the patch.

Regards,
Nayan.

On Thu, Aug 25, 2016 at 3:34 PM, Christian König
<christian.koenig at amd.com> wrote:
> Sorry for not coming back to this earlier, but you know as busy as usual :)
>
>> I thought that a possible reason could be that the Cb and Cr should be
>> first interpolated to the original
>
> Mhm, could be. But in theory that shouldn't make a difference.
>
> Please open up a bug report and attach a picture of the artifacts. I will
> try to take a look when I have time.
>
> In the meantime I suggest that we revert the patch for now and work on
> stopping writing/reading from the same surface first, cause that is the bug
> which is currently a bit problematic for me right now.
>
> Regards,
> Christian.
>
>
> Am 25.08.2016 um 11:58 schrieb Nayan Deshmukh:
>>
>> ping
>>
>> On Mon, Aug 22, 2016 at 7:49 PM, Nayan Deshmukh
>> <nayan26deshmukh at gmail.com> wrote:
>>>
>>> Hi Christian,
>>>
>>> I checked the height/width of the surface. I get a luma channel with
>>> the original size. The Cb and Cr
>>> channel with height/2 and width/2. I had tested videos with 4:2:0
>>> chroma format. But I still get the
>>> artifacts.
>>>
>>> I thought that a possible reason could be that the Cb and Cr should be
>>> first interpolated to the original
>>> and then only we can apply the filter. So, I tried to use a temporary
>>> buffer with original size and
>>> writing on it. But I was still getting the artifacts. But since the
>>> interpolation that would occur in matrix
>>> filter would use the nearest neighbor so I tried to do it using linear
>>> interpolation and the artifacts were
>>> reduced but still, there were minor artifacts.
>>>
>>> Regards,
>>> Nayan.
>>>
>>> On Fri, Aug 19, 2016 at 5:57 PM, Christian König
>>> <christian.koenig at amd.com> wrote:
>>>>
>>>> Hi Nayan,
>>>>
>>>> double check the width/height of the surface and also how many you got
>>>> with
>>>> a debugger.
>>>>
>>>> Actually thinking more about it it could be that top/bottom fields are
>>>> separated and so you got width&height/2 and width/2&height/4 compared to
>>>> the
>>>> video mixer.
>>>>
>>>> Maybe try your idea instead of caching the initialized filters with the
>>>> different sizes.
>>>>
>>>> Regards,
>>>> Christian.
>>>>
>>>>
>>>> Am 18.08.2016 um 18:58 schrieb Nayan Deshmukh:
>>>>>
>>>>> Hi Christian,
>>>>>
>>>>> I tried using 3 instances of the filter with original height & width,
>>>>> height/2 & width and height/2 and width/2.
>>>>> I am applying the 3 filters by checking the width and height of
>>>>> surfaces respectively.
>>>>>
>>>>> The chroma artifacts are still present, different from the initial
>>>>> ones. Any suggestion for the fixup?
>>>>> I tried debugging, but I think I am missing something.
>>>>>
>>>>> Regards,
>>>>> Nayan.
>>>>>
>>>>> On Wed, Aug 17, 2016 at 2:59 PM, Andy Furniss <adf.lists at gmail.com>
>>>>> wrote:
>>>>>>
>>>>>> Christian König wrote:
>>>>>>>
>>>>>>> Top and Bottom field are separated in this representation.
>>>>>>>
>>>>>>> So you got a maximum of 3 planes multiplied by two fields.
>>>>>>
>>>>>>
>>>>>> Ahh, thanks.
>>>>>>
>>>>>>
>>>>>>> Regards,
>>>>>>> Christian.
>>>>>>>
>>>>>>> Am 17.08.2016 um 11:11 schrieb Andy Furniss:
>>>>>>>>
>>>>>>>> Nayan Deshmukh wrote:
>>>>>>>>
>>>>>>>>> Sorry for the misleading language. What I meant was that the filter
>>>>>>>>> should only
>>>>>>>>> be applied to first resource i.e. use only first sampler_view and
>>>>>>>>> surface. As you
>>>>>>>>> replaced i by 0 the filter gets applied multiple times. I tried
>>>>>>>>> doing
>>>>>>>>> that and I am
>>>>>>>>> experiencing same problems i.e. artifacts with positive sharpen.
>>>>>>>>>
>>>>>>>>> So can you please it try it again by replacing VL_MAX_SURFACES with
>>>>>>>>> 1
>>>>>>>>> in the
>>>>>>>>> for loop. It should probably fix the problems.
>>>>>>>>
>>>>>>>>
>>>>>>>> Possibly a stupid question, because I don't know how this stuff
>>>>>>>> works,
>>>>>>>> but why does it loop 6 times per frame rather than 3?
>>>>>>>>
>>>>>>>> I mean as seen if I put a printf as below.
>>>>>>>>
>>>>>>>>>>>>>> +   for(i = 0; i < VL_MAX_SURFACES; ++i) {
>>>>>>>>>>>>>> +      if(sampler_views[i] != NULL && surfaces[i] != NULL) {
>>>>>>>>
>>>>>>>>                    fprintf(stderr,"ADF: i = %d\n", i);
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> +         if (vmixer->noise_reduction.filter)
>>>>>>>>>>>>>> + vl_median_filter_render(vmixer->noise_reduction.filter,
>>>>>>>>>>>>>> +                                    sampler_views[i],
>>>>>>>>>>>>>> surfaces[i]);
>>>>>>>>>>>>>> +
>>>>>>>>>>>>>> +         if (vmixer->sharpness.filter)
>>>>>>>>>>>>>> + vl_matrix_filter_render(vmixer->sharpness.filter,
>>>>>>>>>>>>>> +                                    sampler_views[i],
>>>>>>>>>>>>>> surfaces[i]);
>>>>>>>>
>>>>>>>>
>