mbox series

[v4,0/2] media: HEVC: RPS clean up

Message ID 20220104073842.1791639-1-benjamin.gaignard@collabora.com
Headers show
Series media: HEVC: RPS clean up | expand

Message

Benjamin Gaignard Jan. 4, 2022, 7:38 a.m. UTC 
This series aims to clean up Reference Picture Set usage and flags.

Long term flag was named with RPS prefix while it is not used for RPS
but for mark long term references in DBP. Remane it and remove the two
other useless RPS flags.

Clarify documentation about RPS lists content and make sure that Hantro
driver use them correctly (i.e without look up in DBP).

version 4:
- check flags with & and not ==
- remove the 2 last patches which becomes useless since VPU ctrl-blk
  driver introduction.

version 3:
- rebased on top of v5.16-rc1

version 2:
- change DPB field name from rps to flags

GStreamer HEVC plugin merge request can be found here:
https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1079

With those piece of code fluster score is 77/147.

Benjamin Gaignard (2):
  media: hevc: Remove RPS named flags
  media: hevc: Embedded indexes in RPS

 .../media/v4l/ext-ctrls-codec.rst             | 14 +++++-----
 .../staging/media/hantro/hantro_g2_hevc_dec.c | 27 +++++--------------
 .../staging/media/sunxi/cedrus/cedrus_h265.c  |  2 +-
 include/media/hevc-ctrls.h                    |  6 ++---
 4 files changed, 16 insertions(+), 33 deletions(-)

Comments

Ezequiel Garcia Jan. 5, 2022, 12:26 p.m. UTC | #1 
Hi Benjamin,

On Tue, Jan 04, 2022 at 08:38:41AM +0100, Benjamin Gaignard wrote:
> Marking a picture as long-term reference is valid for DPB but not for RPS.
> Change flag name to match with it description in HEVC spec chapiter
> "8.3.2 Decoding process for reference picture set".
> Remove the other unused RPS flags.
> 

A change like this, which is affecting a kernel interface and has impact
on userspace and drivers, requires a lot more attention in the commit description.

If I understand correctly, this change is related to how HEVC was first
introduced and how the DPB was originally represented in V4L2.

Originally, a DPB entry struct v4l2_hevc_dpb_entry had an rps field
which can be:

  V4L2_HEVC_DPB_ENTRY_RPS_ST_CURR_BEFORE
  V4L2_HEVC_DPB_ENTRY_RPS_ST_CURR_AFTER
  V4L2_HEVC_DPB_ENTRY_RPS_LT_CURR

Perhaps this idea followed libva, where a VAPictureHEVC has flags field
which can be:

  VA_PICTURE_HEVC_RPS_ST_CURR_BEFORE, 
  VA_PICTURE_HEVC_RPS_ST_CURR_AFTER,
  VA_PICTURE_HEVC_RPS_LT_CURR,
  VA_PICTURE_HEVC_LONG_TERM_REFERENCE

In this representation, the sets PocStCurrBefore, PocStCurrAfter,
PocLtCurr are implicit, and must be built by the kernel from the DPB
entries struct v4l2_hevc_dpb_entry, using the information in the rps field.

Last year, we changed this with your commit:

commit d395a78db9eabd12633b39e05c80e803543b6590
Author: Benjamin Gaignard <benjamin.gaignard@collabora.com>
Date:   Thu Jun 3 13:49:57 2021 +0200

    media: hevc: Add decode params control

Which added the control v4l2_ctrl_hevc_decode_params, which includes
the sets PocStCurrBefore, PocStCurrAfter, PocLtCurr explicitly and therefore
moved the responsability of creating and maintaining the sets to userspace.

This effectively made the rps field in the struct v4l2_hevc_dpb_entry
useless. The longterm flag is still needed though for a DPB entry.

With this in mind, you could even say this commit is doing actually two
things:

1. Removes the unused rps field.
2. Adds a flag field, for the longterm DPB entry boolean.

Do you think a single byte of flags will be OK for a DPB entry?
I think so, but I'd love yours/others input here.

If the above is more or less accurate then, and provided you
submit a new version with a more detailed commit description:

Reviewed-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>

Also, a small question:

> Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>
> ---
> version 4:
> - check flags with & and not ==
> 
>  Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst | 8 +++-----
>  drivers/staging/media/hantro/hantro_g2_hevc_dec.c         | 2 +-
>  drivers/staging/media/sunxi/cedrus/cedrus_h265.c          | 2 +-
>  include/media/hevc-ctrls.h                                | 6 ++----
>  4 files changed, 7 insertions(+), 11 deletions(-)
> 
> diff --git a/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst b/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst
> index e141f0e4eec9..38da33e61c3d 100644
> --- a/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst
> +++ b/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst
> @@ -3166,11 +3166,9 @@ enum v4l2_mpeg_video_hevc_size_of_length_field -
>  	:c:func:`v4l2_timeval_to_ns()` function to convert the struct
>  	:c:type:`timeval` in struct :c:type:`v4l2_buffer` to a __u64.
>      * - __u8
> -      - ``rps``
> -      - The reference set for the reference frame
> -        (V4L2_HEVC_DPB_ENTRY_RPS_ST_CURR_BEFORE,
> -        V4L2_HEVC_DPB_ENTRY_RPS_ST_CURR_AFTER or
> -        V4L2_HEVC_DPB_ENTRY_RPS_LT_CURR)
> +      - ``flags``
> +      - Long term flag for the reference frame
> +        (V4L2_HEVC_DPB_ENTRY_LONG_TERM_REFERENCE)

Is this longterm flag associated in any way to a syntax element
or an HEVC process? If so, please document that.

Thanks,
Ezequiel
Benjamin Gaignard Jan. 7, 2022, 4:02 p.m. UTC | #2 
Le 05/01/2022 à 13:26, Ezequiel Garcia a écrit :
> Hi Benjamin,
>
> On Tue, Jan 04, 2022 at 08:38:41AM +0100, Benjamin Gaignard wrote:
>> Marking a picture as long-term reference is valid for DPB but not for RPS.
>> Change flag name to match with it description in HEVC spec chapiter
>> "8.3.2 Decoding process for reference picture set".
>> Remove the other unused RPS flags.
>>
> A change like this, which is affecting a kernel interface and has impact
> on userspace and drivers, requires a lot more attention in the commit description.
>
> If I understand correctly, this change is related to how HEVC was first
> introduced and how the DPB was originally represented in V4L2.
>
> Originally, a DPB entry struct v4l2_hevc_dpb_entry had an rps field
> which can be:
>
>    V4L2_HEVC_DPB_ENTRY_RPS_ST_CURR_BEFORE
>    V4L2_HEVC_DPB_ENTRY_RPS_ST_CURR_AFTER
>    V4L2_HEVC_DPB_ENTRY_RPS_LT_CURR
>
> Perhaps this idea followed libva, where a VAPictureHEVC has flags field
> which can be:
>
>    VA_PICTURE_HEVC_RPS_ST_CURR_BEFORE,
>    VA_PICTURE_HEVC_RPS_ST_CURR_AFTER,
>    VA_PICTURE_HEVC_RPS_LT_CURR,
>    VA_PICTURE_HEVC_LONG_TERM_REFERENCE
>
> In this representation, the sets PocStCurrBefore, PocStCurrAfter,
> PocLtCurr are implicit, and must be built by the kernel from the DPB
> entries struct v4l2_hevc_dpb_entry, using the information in the rps field.
>
> Last year, we changed this with your commit:
>
> commit d395a78db9eabd12633b39e05c80e803543b6590
> Author: Benjamin Gaignard <benjamin.gaignard@collabora.com>
> Date:   Thu Jun 3 13:49:57 2021 +0200
>
>      media: hevc: Add decode params control
>
> Which added the control v4l2_ctrl_hevc_decode_params, which includes
> the sets PocStCurrBefore, PocStCurrAfter, PocLtCurr explicitly and therefore
> moved the responsability of creating and maintaining the sets to userspace.
>
> This effectively made the rps field in the struct v4l2_hevc_dpb_entry
> useless. The longterm flag is still needed though for a DPB entry.
>
> With this in mind, you could even say this commit is doing actually two
> things:
>
> 1. Removes the unused rps field.
> 2. Adds a flag field, for the longterm DPB entry boolean.
>
> Do you think a single byte of flags will be OK for a DPB entry?
> I think so, but I'd love yours/others input here.

In ITU HEVC spec it is said that:
"When a picture is referred to as being marked as "used for reference",
this collectively refers to the picture being marked as "used
  for
short-term reference" or "used for long-term reference" (but not both)."

So I believe that a single if enough for DPB entry.

>
> If the above is more or less accurate then, and provided you
> submit a new version with a more detailed commit description:

I will follow your suggestions and improve it in v5

>
> Reviewed-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
>
> Also, a small question:
>
>> Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>
>> ---
>> version 4:
>> - check flags with & and not ==
>>
>>   Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst | 8 +++-----
>>   drivers/staging/media/hantro/hantro_g2_hevc_dec.c         | 2 +-
>>   drivers/staging/media/sunxi/cedrus/cedrus_h265.c          | 2 +-
>>   include/media/hevc-ctrls.h                                | 6 ++----
>>   4 files changed, 7 insertions(+), 11 deletions(-)
>>
>> diff --git a/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst b/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst
>> index e141f0e4eec9..38da33e61c3d 100644
>> --- a/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst
>> +++ b/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst
>> @@ -3166,11 +3166,9 @@ enum v4l2_mpeg_video_hevc_size_of_length_field -
>>   	:c:func:`v4l2_timeval_to_ns()` function to convert the struct
>>   	:c:type:`timeval` in struct :c:type:`v4l2_buffer` to a __u64.
>>       * - __u8
>> -      - ``rps``
>> -      - The reference set for the reference frame
>> -        (V4L2_HEVC_DPB_ENTRY_RPS_ST_CURR_BEFORE,
>> -        V4L2_HEVC_DPB_ENTRY_RPS_ST_CURR_AFTER or
>> -        V4L2_HEVC_DPB_ENTRY_RPS_LT_CURR)
>> +      - ``flags``
>> +      - Long term flag for the reference frame
>> +        (V4L2_HEVC_DPB_ENTRY_LONG_TERM_REFERENCE)
> Is this longterm flag associated in any way to a syntax element
> or an HEVC process? If so, please document that.

done in v5.

Thanks for your review.

Benjamin

>
> Thanks,
> Ezequiel