From: Jonas Karlman <jonas@kwiboo.se>
To: Ezequiel Garcia <ezequiel@collabora.com>,
linux-media@vger.kernel.org, linux-kernel@vger.kernel.org
Cc: Tomasz Figa <tfiga@chromium.org>,
kernel@collabora.com, Hans Verkuil <hverkuil@xs4all.nl>,
Alexandre Courbot <acourbot@chromium.org>,
Jeffrey Kardatzke <jkardatzke@chromium.org>,
Nicolas Dufresne <nicolas.dufresne@collabora.com>,
Philipp Zabel <p.zabel@pengutronix.de>,
Maxime Ripard <mripard@kernel.org>,
Paul Kocialkowski <paul.kocialkowski@bootlin.com>
Subject: Re: [PATCH 06/10] media: uapi: h264: Cleanup DPB entry interface
Date: Mon, 27 Jul 2020 23:39:13 +0000 (UTC) [thread overview]
Message-ID: <e435253e-67ea-fdec-bf18-059eb4c70a0e@kwiboo.se> (raw)
In-Reply-To: <6bac45d9b22deacf8ac7d68f5b51a5a6c30649f3.camel@collabora.com>
Hi,
On 2020-07-24 21:08, Ezequiel Garcia wrote:
> Hello Jonas,
>
> On Wed, 2020-07-22 at 21:52 +0000, Jonas Karlman wrote:
>> On 2020-07-15 22:22, Ezequiel Garcia wrote:
>>> As discussed recently, the current interface for the
>>> Decoded Picture Buffer is not enough to properly
>>> support field coding.
>>>
>>> This commit introduces enough semantics to support
>>> frame and field coding, and to signal how DPB entries
>>> are "used for reference".
>>>
>>> Signed-off-by: Ezequiel Garcia <ezequiel@collabora.com>
>>> ---
>>> .../media/v4l/ext-ctrls-codec.rst | 46 ++++++++++++-------
>>> drivers/media/v4l2-core/v4l2-h264.c | 4 +-
>>> drivers/staging/media/rkvdec/rkvdec-h264.c | 8 ++--
>>> include/media/h264-ctrls.h | 8 +++-
>>> 4 files changed, 42 insertions(+), 24 deletions(-)
>>>
>>> diff --git a/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst b/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst
>>> index dd8e5a2e8986..46d4c8c6ad47 100644
>>> --- a/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst
>>> +++ b/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst
>>> @@ -2058,10 +2058,35 @@ enum v4l2_mpeg_video_h264_hierarchical_coding_type -
>>> * - __s32
>>> - ``bottom_field_order_cnt``
>>> -
>>> + * - enum :c:type:`v4l2_h264_dpb_reference`
>>> + - ``reference``
>>> + - Specifies how the DPB entry is referenced.
>>> * - __u32
>>> - ``flags``
>>> - See :ref:`DPB Entry Flags <h264_dpb_flags>`
>>>
>>> +.. c:type:: v4l2_h264_dpb_reference
>>> +
>>> +.. cssclass:: longtable
>>> +
>>> +.. flat-table::
>>> + :header-rows: 0
>>> + :stub-columns: 0
>>> + :widths: 1 1 2
>>> +
>>> + * - ``V4L2_H264_DPB_TOP_REF``
>>> + - 0x1
>>> + - The top field in field pair is used for
>>> + short-term reference.
>>> + * - ``V4L2_H264_DPB_BOTTOM_REF``
>>> + - 0x2
>>> + - The bottom field in field pair is used for
>>> + short-term reference.
>>> + * - ``V4L2_H264_DPB_FRAME_REF``
>>> + - 0x3
>>> + - The frame (or the top/bottom fields, if it's a field pair)
>>> + is used for short-term reference.
>>> +
>>> .. _h264_dpb_flags:
>>>
>>> ``DPB Entries Flags``
>>> @@ -2075,29 +2100,16 @@ enum v4l2_mpeg_video_h264_hierarchical_coding_type -
>>>
>>> * - ``V4L2_H264_DPB_ENTRY_FLAG_VALID``
>>> - 0x00000001
>>> - - The DPB entry is valid and should be considered
>>> + - The DPB entry is valid (non-empty) and should be considered.
>>> * - ``V4L2_H264_DPB_ENTRY_FLAG_ACTIVE``
>>> - 0x00000002
>>> - - The DPB entry is currently being used as a reference frame
>>> + - The DPB entry is used for reference.
>>> * - ``V4L2_H264_DPB_ENTRY_FLAG_LONG_TERM``
>>> - 0x00000004
>>> - - The DPB entry is a long term reference frame
>>> + - The DPB entry is used for long-term reference.
>>> * - ``V4L2_H264_DPB_ENTRY_FLAG_FIELD``
>>> - 0x00000008
>>> - - The DPB entry is a field reference, which means only one of the field
>>> - will be used when decoding the new frame/field. When not set the DPB
>>> - entry is a frame reference (both fields will be used). Note that this
>>> - flag does not say anything about the number of fields contained in the
>>> - reference frame, it just describes the one used to decode the new
>>> - field/frame
>>> - * - ``V4L2_H264_DPB_ENTRY_FLAG_BOTTOM_FIELD``
>>> - - 0x00000010
>>> - - The DPB entry is a bottom field reference (only the bottom field of the
>>> - reference frame is needed to decode the new frame/field). Only valid if
>>> - V4L2_H264_DPB_ENTRY_FLAG_FIELD is set. When
>>> - V4L2_H264_DPB_ENTRY_FLAG_FIELD is set but
>>> - V4L2_H264_DPB_ENTRY_FLAG_BOTTOM_FIELD is not, that means the
>>> - DPB entry is a top field reference
>>> + - The DPB entry is a single field or a complementary field pair.
>>>
>>> ``V4L2_CID_MPEG_VIDEO_H264_DECODE_MODE (enum)``
>>> Specifies the decoding mode to use. Currently exposes slice-based and
>>> diff --git a/drivers/media/v4l2-core/v4l2-h264.c b/drivers/media/v4l2-core/v4l2-h264.c
>>> index edf6225f0522..306a51683606 100644
>>> --- a/drivers/media/v4l2-core/v4l2-h264.c
>>> +++ b/drivers/media/v4l2-core/v4l2-h264.c
>>> @@ -66,10 +66,10 @@ v4l2_h264_init_reflist_builder(struct v4l2_h264_reflist_builder *b,
>>> else
>>> b->refs[i].frame_num = dpb[i].frame_num;
>>>
>>> - if (!(dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_FIELD))
>>> + if (dpb[i].reference & V4L2_H264_DPB_FRAME_REF)
>>
>> This looks wrong, should probably use ==,
>>
>> dpb[i].reference == V4L2_H264_DPB_FRAME_REF
>>
>> else this would match any reference value.
>>
>>> pic_order_count = min(dpb[i].top_field_order_cnt,
>>> dpb[i].bottom_field_order_cnt);
>>> - else if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_BOTTOM_FIELD)
>>> + else if (dpb[i].reference & V4L2_H264_DPB_BOTTOM_REF)
>>> pic_order_count = dpb[i].bottom_field_order_cnt;
>>> else
>>> pic_order_count = dpb[i].top_field_order_cnt;
>>> diff --git a/drivers/staging/media/rkvdec/rkvdec-h264.c b/drivers/staging/media/rkvdec/rkvdec-h264.c
>>> index 7b66e2743a4f..57539c630422 100644
>>> --- a/drivers/staging/media/rkvdec/rkvdec-h264.c
>>> +++ b/drivers/staging/media/rkvdec/rkvdec-h264.c
>>> @@ -953,11 +953,11 @@ static void config_registers(struct rkvdec_ctx *ctx,
>>> RKVDEC_COLMV_USED_FLAG_REF;
>>>
>>> if (!(dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_FIELD))
>>> - refer_addr |= RKVDEC_TOPFIELD_USED_REF |
>>> - RKVDEC_BOTFIELD_USED_REF;
>>> - else if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_BOTTOM_FIELD)
>>> + refer_addr |= RKVDEC_FIELD_REF;
>>> +
>>> + if (dpb[i].reference & V4L2_H264_DPB_TOP_REF)
>>> refer_addr |= RKVDEC_BOTFIELD_USED_REF;
>>> - else
>>> + else if (dpb[i].reference & V4L2_H264_DPB_BOTTOM_REF)
>>
>> This should probably be if and not else if, and BOTFIELD/TOPFIELD_USED_REF
>> seems to be mixed up.
>>
>> I have only taken a quick look so far, I will update ffmpeg and runtime test
>> later this weekend, will get back with result and full review on Sunday evening.
>>
>
> Thanks that would be useful.
>
> However, keep in mind this series is specifically concerned
> with the uAPI review.
>
> This is not supposed to fix the field coded support, or anything
> else in any driver.
>
> IMO, at this stage, fixing drivers is somewhat lower priority
> than discussing and stabilizing the uAPI.
I have now tested rkvdec on a RK3328 device and needed to do 3 fixups, see [1].
Initial ffmpeg update using update h264 uapi is located at [2], the ffmpeg
update still needs to be tested with cedrus and hantro.
So far I have not seen any issue with the uapi changes.
Q: ffmpeg will not try to set SLICE_PARAMS or PRED_WEIGHT ctrls for
DECODE_MODE_SLICE_BASED, should userspace check if ctrl exists or is using
DECODE_MODE value okay?
I have also pushed an updated WIP branch at [3] containing high 10,
field encoded and hevc work.
[1] https://github.com/Kwiboo/linux-rockchip/compare/b6b91f27c0cb33520e954e7bb2550e0e07ed4d85...b82b6e93feb9ca44d2c677f25416cf6345f0114d
[2] https://github.com/Kwiboo/FFmpeg/commits/v4l2-request-hwaccel-4.3.1
[3] https://github.com/Kwiboo/linux-rockchip/commits/linuxtv-rkvdec-work-in-progress
Best regards,
Jonas
>
> Thanks,
> Ezequiel
>
>
next prev parent reply other threads:[~2020-07-27 23:39 UTC|newest]
Thread overview: 40+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-07-15 20:22 [PATCH 0/7] media: Clean H264 stateless uAPI Ezequiel Garcia
2020-07-15 20:22 ` [PATCH 01/10] media: uapi: h264: Update reference lists Ezequiel Garcia
2020-07-22 21:43 ` Jonas Karlman
2020-07-15 20:22 ` [PATCH 02/10] media: uapi: h264: Further clarify scaling lists order Ezequiel Garcia
2020-07-16 7:23 ` Hans Verkuil
2020-07-16 11:43 ` Ezequiel Garcia
2020-07-16 11:44 ` Hans Verkuil
2020-07-15 20:22 ` [PATCH 03/10] media: uapi: h264: Split prediction weight parameters Ezequiel Garcia
2020-07-16 7:26 ` Hans Verkuil
2020-07-16 11:14 ` Ezequiel Garcia
2020-07-22 16:03 ` Jernej Škrabec
2020-07-25 13:30 ` Alexandre Courbot
2020-07-30 13:48 ` Nicolas Dufresne
2020-07-15 20:22 ` [PATCH 04/10] media: uapi: h264: Clarify pic_order_cnt_bit_size field Ezequiel Garcia
2020-07-15 20:22 ` [PATCH 05/10] media: uapi: h264: Increase size of 'first_mb_in_slice' field Ezequiel Garcia
2020-07-15 20:22 ` [PATCH 06/10] media: uapi: h264: Cleanup DPB entry interface Ezequiel Garcia
2020-07-22 16:09 ` Jernej Škrabec
2020-07-22 17:11 ` Ezequiel Garcia
2020-07-22 21:52 ` Jonas Karlman
2020-07-24 19:08 ` Ezequiel Garcia
2020-07-27 23:39 ` Jonas Karlman [this message]
2020-07-31 12:49 ` Ezequiel Garcia
2020-07-15 20:22 ` [PATCH 07/10] media: uapi: h264: Increase size of DPB entry pic_num Ezequiel Garcia
2020-07-15 20:22 ` [PATCH 08/10] media: uapi: h264: Clean slice invariants syntax elements Ezequiel Garcia
2020-07-25 14:34 ` Alexandre Courbot
2020-07-27 14:39 ` Ezequiel Garcia
2020-07-27 14:52 ` Tomasz Figa
2020-07-27 16:18 ` Ezequiel Garcia
2020-07-27 18:10 ` Tomasz Figa
2020-07-27 19:43 ` Nicolas Dufresne
2020-08-04 13:35 ` Tomasz Figa
2020-08-05 16:41 ` Ezequiel Garcia
2020-07-28 12:44 ` Maxime Ripard
2020-07-28 21:09 ` Nicolas Dufresne
2020-07-15 20:22 ` [PATCH 09/10] media: hantro: Don't require unneeded H264_SLICE_PARAMS Ezequiel Garcia
2020-07-25 14:45 ` Alexandre Courbot
2020-07-26 13:34 ` Alexandre Courbot
2020-07-27 14:44 ` Ezequiel Garcia
2020-07-15 20:22 ` [PATCH 10/10] media: rkvdec: " Ezequiel Garcia
2020-07-27 22:03 ` Jonas Karlman
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=e435253e-67ea-fdec-bf18-059eb4c70a0e@kwiboo.se \
--to=jonas@kwiboo.se \
--cc=acourbot@chromium.org \
--cc=ezequiel@collabora.com \
--cc=hverkuil@xs4all.nl \
--cc=jkardatzke@chromium.org \
--cc=kernel@collabora.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-media@vger.kernel.org \
--cc=mripard@kernel.org \
--cc=nicolas.dufresne@collabora.com \
--cc=p.zabel@pengutronix.de \
--cc=paul.kocialkowski@bootlin.com \
--cc=tfiga@chromium.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).