Re: [PATCH v2 1/7] media: rkvdec: Add HEVC backend

[Detlev Casanova <detlev.casanova@collabora.com>]

Hi Jonas,

On Sunday, 17 August 2025 12:39:01 EDT Jonas Karlman wrote:
> Hi Detlev,
> 
> On 8/12/2025 9:54 PM, Detlev Casanova wrote:
> > On Tuesday, 12 August 2025 09:07:33 EDT Nicolas Dufresne wrote:
> >> Le mardi 12 août 2025 à 02:58 +0200, Jonas Karlman a écrit :
> >>> Hi Nicolas,
> >>> 
> >>> On 8/11/2025 11:07 PM, Nicolas Dufresne wrote:
> >>>> First pass of review,
> >>>> 
> >>>> There is obvious conflict between Detlev RK3388/3576 series, please
> >>>> coordinate. By picking the same direction, the common code can be split
> >>>> from platform code, and then both sets can share that common code and
> >>>> avoid conflicts.>
> >>> 
> >>> Yes, it is very unfortunate that Detlev did not base the work on v1 of
> >>> this series, this has been sitting idle on the list for almost 2 years
> >>> only waiting on its dependent H264 High10/4:2:2 series to be merged.
> >>> 
> >>> Most feedback for v1 was related to unstaging and the series could be
> >>> applied clean until iommu detach/attach was introduced very recently.
> >>> 
> >>> This series has also been referenced in multiple threads related to
> >>> rkvdec2 work, most recent at [1]. Unfortunately, I have missed both v1
> >>> and v2 of Detlev's new rkvdec2 series, and was not able to send a v2 of
> >>> this until now due to family tragedy.
> >>> 
> >>> I will take a closer look at Detlev's new rkvdec2 series and see what I
> >>> can do in a v3 to ease any merge conflicts.
> >> 
> >> All in all, I'd like to have both of these series ready for 6.18. To
> >> avoid
> >> the need for dependant series, I can pick top patches if the preparation
> >> is
> >> properly split from the platform specific code. Please make you
> >> recommendation.
> >> 
> >> My main concern at the moment is the back and forth from register to
> >> structure, Detlev already ported everything else to structures, so
> >> porting
> >> a single codec for a single SoC seems like the least amount of work. I'd
> >> like to ear if any of you would be willing to convert it. This way I can
> >> land the conversion early, and we remove one massive clash between the
> >> two
> >> sets.
> > 
> > The conversion won't be too difficult to manage, I can do it as I did for
> > h264 and add that to my serie over this one.
> 
> Thanks, I very much appreciate that!, I won't have time to look into a
> conversion until at earliest next weekend.

I pushed a branch with my series rebased on yours[1]. It ports the hevc code 
to using structs (for registers).
Also move the cabac table and common hevc functions.

Reports indicate that all works well :)

> > As my series is already doing a conversion, I'm not against keeping this
> > one without structs and simply do the conversion later, even if that
> > means merging code we know will change later.
> > 
> > I'd also like to mention that currently, my serie changes the writel to
> > structs only for the registers, not for PPS and SPS.
> > So I don't mind keeping that part as is for rkvdec-hevc.
> > 
> > For the variants support, I can also merge in my changes: my rkvdec_config
> > would just become an extention your rkvdec_variant.
> > 
> > Finally, this series has been here for longer than mine, so it's only fair
> > to have me rebase mine, especially because it means doing a bit more of
> > what I already did.
> 
> Great, I will try to get a v3 out in next few days after addressing some
> of the feedback.

I will rebase on your v3 and send my v3 when this one is accepted.


[1]: https://gitlab.collabora.com/detlev/linux/-/commits/add-vdpu381-and-383-to-rkvdec-v3
Regards,
Detlev.