[VP] Use forward_references as the reference for ADI #1162
Conversation
VAAPI define forward_references as past referene frame, but vphal regart forward reference as future reference frame. This change will only correct the behavior in DDI level to map forward_references to VPHAL_SURFACE.pBwdRef, and backward_references to VPHAL_SURFACE.pFwdRef
|
Will this be merged at some point? It looks like I can't use hardware deinterlacing on Tiger Lake until it is. |
|
Hope this will get merged some day. |
|
@FurongZhang , please take a look, if no other concern , we should start the merging process. suppose validation reference also need be updated. |
|
Almost 2 years now since the issue has been reported, with no fix in sight. |
|
ping |
|
The original author will not work on this. I picked up this and will follow up. Will run full test and check if there is any regression. |
|
Any plans to merge this? I would love to stop having to compile the driver myself |
|
@FurongZhang - as fyi - this has been part of the LibreELEC code base for 12 months now: LibreELEC/LibreELEC.tv@7fc1462 Hope to see it in the official release in the near future. |
|
Still no merge? |
|
gentle ping as a reminder for this PR ;) |
|
I don't understand either why this PR is ignored (for years now). ADI is an important feature which is completely broken at this point. |
|
monthly ping as a reminder for this PR ;) |
|
monthly ping as a reminder for this PR ;) |
|
It is in our known issues list. https://github.com/intel/media-driver/wiki/Existing-Known-Issue-List#video-processing-forward-and-backward-reference. |
|
@LhGu by "close this pr" do you mean you are going to merge it? If not, can you please explain why? |
|
monthly ping as a reminder for this PR ;) |
1 similar comment
|
monthly ping as a reminder for this PR ;) |
|
ping :-) this ignorance is really wondering me |
FurongZhang
requested review from
FurongZhang
and removed request for
FurongZhang
|
Hi @Simon566 , @bkuhls , @chcore , @smp79 , Best regards, |
|
Hi, any user thats using a intel x86 system with IGPU for watching satellite TV will have issues with interlacing. Also, there is quite alot older video content out there which is interlaced ! in my case its a Kodi /XBMC box on a N5105 CPU. Others are using Libreelec which is basically also the same (XBMC/Kodi) on X86, but they patch the intel driver by default in their repos. Im already happy with BOB deinterlacing , but advanced deinterlace would even be better regards, |
N5105 (Jasper Lake) does not support advanced deinterlacing anyway (hardware limitation), so you are not affected by this issue. The issue affects Tiger Lake/Alder Lake and newer platforms. |
|
possible , but without the mentioned patch , which LibreElec uses , VAAPI BOB doesnt work as well .... |
|
BOB looks much worse than MCDI or any other high quality DI method. In your case I recommend disabling VAAPI hw decoding and use software (ff-h264) decoding with BWDIF deinterlace method. BWDIF looks just as good as hardware MCDI. |
|
BWDIF hasnt arrived in kodi yet. and decoding h265 in SW is too slow |
|
BWDIF is the default sw deinterlace method since Kodi 21. VAAPI hw acceleration can be disabled only for h264/mpeg-2. |
|
oh ok , i got kodi 21 , but i only see bob / deinterlace / vaapi bob |
|
Disable VAAPI acceleration for h264 and mpeg-2, it would default to bwdif. |
|
this doesnt work , 4k is too much in h264 |
@FurongZhang I refer you to my comment on the issue. This effects everyone who is trying to deinterlace HD or SD broadcast TV content using modern Intel iGPU's and this media driver. Adaptive deinterlacing doesn't work properly and results in a noticeably jittery picture. This worked correctly on the old drivers, and on this driver with this PR. Naturally the open source media server projects and their users notice this more as they are playing and watching live and recorded broadcast TV channels. Those projects have been able to work around this by providing distributions that include this driver with this PR. But that doesn't help users who are using standard distributions (e.g. Ubuntu) with standard media player or transcoding applications. As you'll see from my comment this affects both home users and broadcasters (i.e. businesses and customers of Intel server products). Given this Intel really need to provide a fix for this, otherwise playback or transcoding of interlaced video will remain broken for the majority of users, regardless of software used. |
|
@orryverducci , I fully understood your request. |
|
Ping from me. Broadcast TV is still very common media. Without ADI, Intel GPU cannot be used for transcoding with high quality. |
|
ping pong im used to a different kind of quality from intel, wether its Raptor Lake CPUs or their drivers . |
|
Adding @jiafengy1 |
|
afaics the code of this PR was committed to 24.3.1: 8017d9c |
|
Yes, at last. |
|
🎉🎉🎉
…
On 6 Aug 2024 at 08:14, smp79 ***@***.***> wrote:
Yes, at last.
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>
|
VAAPI define forward_references as past referene frame, but vphal regard forward reference as future reference frame. This change will only correct the behavior in DDI level to map forward_references to VPHAL_SURFACE.pBwdRef, and backward_references to VPHAL_SURFACE.pFwdRef