On Mon, 16 Nov 2020, Hans Verkuil wrote:
On 15/11/2020 19:59, Martin Cerveny wrote:
Hello.
On Thu, 5 Nov 2020, Hans Verkuil wrote:
Hi Martin,
On 12/09/2020 16:30, Martin Cerveny wrote:
First patch extends cedrus capability to all decoders
because V3s missing MPEG2 decoder.
Next two patches add system control node (SRAM C1) and
next three patches add support for Cedrus VPU.
Tested on "Lichee Zero" V3s platform with testing LCD patch
( https://github.com/mcerveny/linux/tree/v3s_videocodec_v4 )
and V4L2 raw API testing utility
( https://github.com/mcerveny/v4l2-request-test ):
- enabled LCD (DRM dual VI and sigle UI planes)
- added RGB panel
- enabled PWM
There is low memory on V3s (64MB) and maximum must be available to CMA:
- CONFIG_CMA_SIZE_MBYTES=28
- add swap to swapout other processes
- decrease buffers in v4l2-request-test (.buffers_count from 16 to 6)
Only H.264 decoder working - MPEG and H.265 unsupported by V3s,
JPEG/MJPEG still unimplemented, encoder unimplemented
When I tried to merged these patches I got merge conflicts.
Possibly due to other 5.10 changes, but certainly because of conflicts
with patches from Jernej:
https://patchwork.linuxtv.org/project/linux-media/patch/20200825173523.1289379-4-jernej.skrabec@xxxxxxxx/
https://patchwork.linuxtv.org/project/linux-media/patch/20200825173523.1289379-5-jernej.skrabec@xxxxxxxx/
I've merged Jerne's patches and posted a PR for that:
https://patchwork.linuxtv.org/project/linux-media/patch/f3b8e5e2-5f0e-fb6f-e5b2-7f44f7e365e7@xxxxxxxxx/
Can you rebase your patches on top of my branch that contains Jernej's patches?
https://git.linuxtv.org/hverkuil/media_tree.git/log/?h=for-v5.11e
Once my PR is merged into the media_tree master I can take your rebased
patches.
I updated patches:
https://github.com/mcerveny/linux/tree/media_tree_for-v5.11e
BUT, commit (555 commits) for v5.10-1
https://github.com/torvalds/linux/commit/fd5c32d80884268a381ed0e67cccef0b3d37750b
disrupts usability of Cedrus H.264 (at least for my Allwinner V3s):
1) colors are disrupted
There are missing some initialization now.
If I use "5.9" compatible code
(last bisect good point https://github.com/torvalds/linux/commit/647412daeb454b6dad12a6c6961ab90aac9e5d29 )
then reboot (not power-off!) and use new code
( https://github.com/mcerveny/linux/tree/media_tree_for-v5.11e )
and colors are OK.
Could this or the next issue be related to this bug fix?
https://git.linuxtv.org/media_tree.git/commit/?h=fixes&id=9ac924b98728c3733c91c6c59fc410827d0da49f
That fix isn't yet in our master tree.
Confirmed. It recovers colors !
2) decoding of complex streams fails
( https://github.com/mcerveny/v4l2-request-test/tree/v5.10 )
- bbb-h264-all-i-32 - OK
- bbb-h264-32 - bad from frame 5
- bbb-h264-high-32 - bad from frame 6
I think cedrus devs need to take a look at these issues.
Maybe something wrong in my testing code,
problematic commit swapped some variables between structures :-(
I try to investigate more, regards Martin
Regards,
Hans
best regards,
Martin
Changes since v1:
- patch 0005 rename
- added testing description
Martin Cerveny (6):
media: cedrus: Register all codecs as capability
dt-bindings: sram: allwinner,sun4i-a10-system-control: Add V3s
compatibles
ARM: dts: sun8i: v3s: Add node for system control
media: cedrus: Add support for V3s
dt-bindings: media: cedrus: Add V3s compatible
ARM: dts: sun8i: v3s: Add video engine node
.../allwinner,sun4i-a10-video-engine.yaml | 1 +
.../allwinner,sun4i-a10-system-control.yaml | 6 ++++
arch/arm/boot/dts/sun8i-v3s.dtsi | 33 +++++++++++++++++++
drivers/staging/media/sunxi/cedrus/cedrus.c | 28 +++++++++++++++-
drivers/staging/media/sunxi/cedrus/cedrus.h | 2 ++
.../staging/media/sunxi/cedrus/cedrus_video.c | 2 ++
6 files changed, 71 insertions(+), 1 deletion(-)
