[Mesa-dev] [PATCH] st/va: disable cabac for h264 baseline profile

Christian König deathsimple at vodafone.de
Tue Oct 18 15:37:39 UTC 2016

Am 18.10.2016 um 15:42 schrieb Andy Furniss:
> Andy Furniss wrote:
>> Christian König wrote:
>>> Am 18.10.2016 um 11:19 schrieb Andy Furniss:
>>>> boyuan.zhang at amd.com wrote:
>>>>> From: Boyuan Zhang <boyuan.zhang at amd.com>
>>>>> cabac is only supported in the h264 main and higher profiles
>>>> So shouldn't there be code allows it if the user space doesn't
>>>> set baseline?
>>>> I don't know how in gstreamer as it seems to try to use b-frames
>>>> if you use other than baseline which doesn't work.
>>>> With avconv it is possible to call main/high and set b-frames to 0.
>>>> I know it's technically correct spec wise, but seems a shame as it
>>>> costs a fair bit in "free" efficiency.
>>>> On Windows the raptor game recording app produces files flagged as
>>>> high with cabac - but without b-frames.
>>> The problem is that it can easily break decoders. CABAC is simply not
>>> allowed in a stream flagged as baseline compliant.
>> But with ffmpeg/avconv I can make a stream flagged as main/high even
>> if it's really baseline + CABAC. I guess Windows may vary but the test
>> I did seems to take this pragmatic approach, as it seems do other h/w
>> encoders eg. smartphone output.
>>> It's a pity that we don't support B-frames any more.
>> Anymore? Now I am curious, seems to work with omx (cqp single instance)
>>> With that in place
>>> we could easily advertise support for mainline profile.
> Sorry if that came over as being pedantic, silly as I think pragmatism
> is the way to go and I know intel advertise main/high, but doubt they
> do interlaced.

Exactly, I mean we are talking about features to support encoding into 
interlaced format. Is anybody still actively doing that?

But even then, it's not so much of a problem advertising mainline 
profile and then not using MBAFF/PAFF.

But when you advertise B-frames and then can't encode it you got a 
serious problem because your frames are not in the right order any more :)

> In fact vce vaapi is currently advertising them as well (I did mention
> it in some thread). Good for letting ffmpeg flag as such while not using
> b-frames, not so good for gstreamer as they have changed the default to
> high so old command lines will not explicitly fail, but will produce 
> junk.
> I see va.h has a cabac switch and gstreamer exposes it - though it's not
> read by the driver. Maybe if that were hooked up then users could turn
> it on and profit :-).

Yeah, but again turning it on while the SPS/PPS only advertise the 
stream to be baseline compliant is a clear violation of the codec 
standard. (Is that actually encodeable in the stream? or does the 
encoder switch to some higher level automatically if you use it?).


>>> Christian.
>>>>> Signed-off-by: Boyuan Zhang <boyuan.zhang at amd.com>
>>>>> ---
>>>>>   src/gallium/state_trackers/va/picture.c | 1 -
>>>>>   1 file changed, 1 deletion(-)
>>>>> diff --git a/src/gallium/state_trackers/va/picture.c
>>>>> b/src/gallium/state_trackers/va/picture.c
>>>>> index eae5dc4..db08a3c 100644
>>>>> --- a/src/gallium/state_trackers/va/picture.c
>>>>> +++ b/src/gallium/state_trackers/va/picture.c
>>>>> @@ -110,7 +110,6 @@ getEncParamPreset(vlVaContext *context)
>>>>> context->desc.h264enc.motion_est.enc_ime2_search_range_y =
>>>>> 0x00000004;
>>>>>      //pic control preset
>>>>> -   context->desc.h264enc.pic_ctrl.enc_cabac_enable = 0x00000001;
>>>>> context->desc.h264enc.pic_ctrl.enc_constraint_set_flags =
>>>>> 0x00000040;
>>>>>      //rate control
>>>> _______________________________________________
>>>> mesa-dev mailing list
>>>> mesa-dev at lists.freedesktop.org
>>>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev

More information about the mesa-dev mailing list