linux-media.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Adrian Ratiu <adrian.ratiu@collabora.com>
To: Philipp Zabel <p.zabel@pengutronix.de>, linux-media@vger.kernel.org
Cc: Ezequiel Garcia <ezequiel@collabora.com>,
	Mirela Rabulea <mirela.rabulea@nxp.com>,
	kernel@pengutronix.de, Tim Harvey <tharvey@gateworks.com>
Subject: Re: [PATCH v2 0/6] v4l2 JPEG helpers and CODA960 JPEG decoder
Date: Mon, 23 Mar 2020 08:56:32 +0200	[thread overview]
Message-ID: <87a747h84f.fsf@collabora.com> (raw)
In-Reply-To: <20200318183536.15779-1-p.zabel@pengutronix.de>

[-- Attachment #1: Type: text/plain, Size: 3259 bytes --]

On Wed, 18 Mar 2020, Philipp Zabel <p.zabel@pengutronix.de> wrote:
> Hi, 
> 
> the JPEG header parser is updated to accept up to four 
> components, baseline and extended-sequential DCT encoded images, 
> 8-bit and 12-bit precision, as well as 8-bit and 16-bit 
> quantization tables. As a consequence, all drivers will have to 
> check the number of components, precision, and quantization 
> table lengths. 
> 
> I have not yet added support parsing the Adobe APP14 headers to 
> determine the color encoding, as it is unclear to me how it 
> could be used to signal RGBA components - for 4-component images 
> it is defined to disambiguate between CMYK and YCCK 
> encodings. This is implemented in libjpeg.  Patching the header 
> data in place to normalize the component identifiers is not part 
> of the parser and will be added in a separate patch. 
> 
> For now the rcar_jpu, s5p-jpeg and mtk-jpeg conversions are 
> dropped.  Instead, a few CODA fixes were added that should avoid 
> alignment issues with odd-sized JPEG images and that stop 
> tricking GStreamer into negotiating NV12 and then switching to 
> YUV420 instead in S_FMT. 
> 
> regards Philipp 
> 
> Philipp Zabel (6): 
>   media: coda: round up decoded buffer size for all codecs 
>   media: add v4l2 JPEG helpers media: coda: jpeg: add CODA960 
>   JPEG decoder support media: coda: split marking last meta into 
>   helper function media: coda: mark last capture buffer media: 
>   coda: lock capture queue wakeup against decoder stop command 
> 
>  drivers/media/platform/Kconfig            |   1 + 
>  drivers/media/platform/coda/coda-common.c | 188 ++++++- 
>  drivers/media/platform/coda/coda-jpeg.c   | 572 
>  ++++++++++++++++++++ drivers/media/platform/coda/coda.h 
>  |  10 +- drivers/media/v4l2-core/Kconfig           |   4 + 
>  drivers/media/v4l2-core/Makefile          |   2 + 
>  drivers/media/v4l2-core/v4l2-jpeg.c       | 632 
>  ++++++++++++++++++++++ include/media/v4l2-jpeg.h 
>  | 135 +++++ 8 files changed, 1519 insertions(+), 25 
>  deletions(-) create mode 100644 
>  drivers/media/v4l2-core/v4l2-jpeg.c create mode 100644 
>  include/media/v4l2-jpeg.h 
> 
> --  2.20.1 

Hi again,

When testing both v1 and v2 series, a subtle corner-case bug
leading to color degradation was observed on our IMX6DL board.

I'm documenting here so it will be a known issue if someone
else hits the rare corner-case.

In the attached archive there is an example img test420.jpeg.

It can be decoded & displayed correctly using the pipeline:

gst-launch-1.0 -v filesrc location=./test420.jpeg ! jpegparse ! 
v4l2jpegdec ! video/x-raw,format=NV12 ! imagefreeze ! kmssink

If any GIMP created jpeg like gimp_created_borks_colors.jpeg
is decoded in between, then all subsequent raw decoded images
until next reboot have their colors "degraded".

The archive has are two NV12 decoded images which exemplify
this degradation: before_gimp.raw and after_gimp.raw.

The raw files can be viewed with mplayer like this:
mplayer -demuxer rawvideo -rawvideo w=800:h=480:format=NV12 
./before_gimp.raw -loop 0

I am unsure what might be causing this degradation, we only
see it with GIMP-created images, but since our applications
are not creating jpegs with GIMP this is no problem for us.


[-- Attachment #2: test_jpegs.tar.xz --]
[-- Type: application/x-xz, Size: 76236 bytes --]

[-- Attachment #3: Type: text/plain, Size: 22 bytes --]


Thanks again,
Adrian

      parent reply	other threads:[~2020-03-23  6:55 UTC|newest]

Thread overview: 11+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-03-18 18:35 [PATCH v2 0/6] v4l2 JPEG helpers and CODA960 JPEG decoder Philipp Zabel
2020-03-18 18:35 ` [PATCH v2 1/6] media: coda: round up decoded buffer size for all codecs Philipp Zabel
2020-03-18 18:35 ` [PATCH v2 2/6] media: add v4l2 JPEG helpers Philipp Zabel
2020-03-18 18:35 ` [PATCH v2 3/6] media: coda: jpeg: add CODA960 JPEG decoder support Philipp Zabel
2020-03-18 18:35 ` [PATCH v2 4/6] media: coda: split marking last meta into helper function Philipp Zabel
2020-03-18 18:35 ` [PATCH v2 5/6] media: coda: mark last capture buffer Philipp Zabel
2020-03-18 18:35 ` [PATCH v2 6/6] media: coda: lock capture queue wakeup against decoder stop command Philipp Zabel
2020-03-19 15:37 ` [PATCH v2 0/6] v4l2 JPEG helpers and CODA960 JPEG decoder Tim Harvey
2020-03-22 21:41 ` Adrian Ratiu
2020-03-22 22:52   ` Ezequiel Garcia
2020-03-23  6:56 ` Adrian Ratiu [this message]

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=87a747h84f.fsf@collabora.com \
    --to=adrian.ratiu@collabora.com \
    --cc=ezequiel@collabora.com \
    --cc=kernel@pengutronix.de \
    --cc=linux-media@vger.kernel.org \
    --cc=mirela.rabulea@nxp.com \
    --cc=p.zabel@pengutronix.de \
    --cc=tharvey@gateworks.com \
    /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).