Hello Sreerenj!

I have another question, before I asked it in this mailing list, but had no 
responce.
I have used vaapi tools and gstreamer-vaapi for encoding sample FullHD video 
(Big buck bunny) from yuv420 source.
I have used different settings, but I can`t get reasonable bitrate (1Mbit/s) 
for FullHD video with normal quality. 
Even for 10Mbit/s quality was not very good (there were artefacts sometimes).

I have tried same hw encoding on Windows, using Handbrake tool. It works much 
better and I got 1Mbit/s good quality encoding.
My conclusion was that the problem is not in hardware, but with software.

I tried followings CPUs: Haswell 4770K, 4770T, 4790T, Broadwell desktop i5. 
Same result. All my Haswell cpus have HD4600 gpu,
Broadwell has some more improved, but same result.

Maybe you can help me and share some of your expirience. Or encode some sample 
1920x1080 video and check output?


11.11.2015, 14:38, "William Katsak" <william.kat...@ericsson.com>:
> Hello,
>
> Thanks very much for your reply. The way you described it is what I inferred 
> from the source.
>
> What I had expected was that setting the capsfilter would tell the encoder to 
> use whatever tools are available that are part of the specified profile. For 
> example, if my vainfo reports these:
>
>       VAProfileH264ConstrainedBaseline: VAEntrypointEncSlice
>       VAProfileH264Main : VAEntrypointEncSlice
>       VAProfileH264High : VAEntrypointEncSlice
>
> I would have liked to be able to specify profile=high, and have whatever 
> tools are part of VAProfileH264Main selected. Indeed, I think this is how 
> x264enc works (and maybe gstreamer-omx too?).
>
> I will open the bug report shortly, but, I wanted to, ask something else: If 
> I need to approximate the behaviour for some experiments, is it reasonable to 
> just ask for whatever tools are part of each profile? Is there any way to see 
> exactly what would be part of each profile (other than checking the 
> standard)? Is the set of options exported to gstreamer inclusive of 
> everything in the VAProfiles?
>
> Thanks so much again,
> Bill Katsak
> ________________________________________
> From: Libva [libva-boun...@lists.freedesktop.org] on behalf of Sreerenj 
> [sreerenj.balachand...@intel.com]
> Sent: Wednesday, 11 November 2015 5:40 AM
> To: libva@lists.freedesktop.org
> Subject: Re: [Libva] gstreamer-vaapi strange issue
>
> Hi,
>
> Right now gst-vaapi-encoder doesn't have support to negotiate the custom
> profile set via capsfilter. But we are
> selecting an appropriate profile based on other coding tools. We start
> from "constrained-baseline" as the default choice,
> then if user ask for b frame encoding or cabac entropy coding, it will
> switch to "main" profile.. Similarly asking for 8x8 dct
> will switch the profile to "high"..
>
> Basically if you are not choosing any of those coding tools, it could be
> enough to use constrained-baseline..
> Any decoder which can support main profile should be able enough to
> decode the constrained-baseline too.
> Similarly high profile decoder can decode main profile too..
>
> But adding a provision for selecting profiles as per user's choice seems
> fine for me.
> If you can file a bug against gstreamer-vaapi
> (https://bugzilla.gnome.org/enter_bug.cgi?product=gstreamer-vaapi)
> asking for this feature,
> we can discuss it there..
>
> On 10.11.2015 20:23, William Katsak wrote:
>>  I've been looking at the source this afternoon, and my interpretation of it 
>> is that specifying the profile via caps is only serving a restrictive role, 
>> vs. a configuration role. Obviously, I need to study it more, but is this 
>> true? To target a different profile, do we need to specifically specify 
>> which encoding tools that we want to use?
>>
>>  Thanks,
>>  Bill
>>
>>  ________________________________________
>>  From: Libva [libva-boun...@lists.freedesktop.org] on behalf of William 
>> Katsak [william.kat...@ericsson.com]
>>  Sent: Tuesday, 10 November 2015 10:33 AM
>>  To: libva@lists.freedesktop.org
>>  Subject: [Libva] gstreamer-vaapi strange issue
>>
>>  Hello,
>>
>>  I am not sure if this the right list for this, but I am seeing a strange 
>> issue with gstreamer-vaapi on Intel Haswell.
>>
>>  I am doing some experimentation using gstreamer-vaapi on an Intel i3-4370 
>> CPU. I am running libva 1.6.1, and libva Intel driver 1.6.0 on Ubuntu 15.04.
>>
>>  My issue is that I cannot seem to get the thing to do anything except 
>> constrained baseline, despite asking it to do other profiles.
>>
>>  If I do something like this:
>>
>>  gst-launch-1.0 filesrc 
>> location=/ramdisk/bbb_sunflower_1080p_30fps_normal.mp4 ! qtdemux ! 
>> vaapidecode ! vaapiencode_h264 ! video/x-h264,profile=high ! qtmux ! 
>> filesink location=/ramdisk/tmp.mov
>>
>>  It appears to work fine, but if I check with this:
>>
>>  avprobe -show_streams /ramdisk/tmp.mov
>>
>>  the output will indicate:
>>  profile=Constrained Baseline
>>
>>  However, if I change the encode element to:
>>  vaapiencode_h264 dct8x8=1
>>  it does change the output profile.
>>
>>  My question is, is this correct behavior? Do we need dct8x8 to do anything 
>> higher than Constrained Baseline? If so, why? Is this a fundamental detail 
>> of H.264, or is this an implementation detail?
>>
>>  Thanks in advance for any help or pointers that you might able to give me.
>>  -Bill
>>
>>  _______________________________________________
>>  Libva mailing list
>>  Libva@lists.freedesktop.org
>>  http://lists.freedesktop.org/mailman/listinfo/libva
>>  _______________________________________________
>>  Libva mailing list
>>  Libva@lists.freedesktop.org
>>  http://lists.freedesktop.org/mailman/listinfo/libva
>
> --
> Thanks
> Sree
>
> ---------------------------------------------------------------------
> Intel Finland Oy
> Registered Address: PL 281, 00181 Helsinki
> Business Identity Code: 0357606 - 4
> Domiciled in Helsinki
>
> This e-mail and any attachments may contain confidential material for
> the sole use of the intended recipient(s). Any review or distribution
> by others is strictly prohibited. If you are not the intended
> recipient, please contact the sender and delete all copies.
> _______________________________________________
> Libva mailing list
> Libva@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/libva
> _______________________________________________
> Libva mailing list
> Libva@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/libva

-- 
Nikita Orlov
Skype: nik_stet
QQ: 2717846083
_______________________________________________
Libva mailing list
Libva@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/libva

Reply via email to