[FFmpeg-devel] [PATCH 00/72] Implement support for Vulkan multiplane images and video decoding

[Lynne]

Feb 19, 2023, 16:40 by michael at niedermayer.cc:

> Hi
>
>
> On Sun, Feb 19, 2023 at 01:08:02AM +0100, Lynne wrote:
>
>>
>> Feb 18, 2023, 20:03 by michael at niedermayer.cc:
>>
>> > On Fri, Feb 17, 2023 at 04:43:50AM +0100, Lynne wrote:
>> >
>> >> This small patchset mostly rewrites Vulkan to enable using multiplane images,
>> >> and implements video decode support. Also, many numerous bugs and issues
>> >> were fixed, as well as having quite a lot of performance improvements.
>> >>
>> >> The patchset can be viewed here as well:
>> >> https://github.com/cyanreg/FFmpeg/tree/vulkan_staging
>> >>
>> >> Patches attached.
>> >>
>> >
>> > [...]
>> > 
>> >
>> >> av1dec.c  |    3 +++
>> >>  avcodec.h |    5 +++++
>> >>  h264dec.c |    3 +++
>> >>  hevcdec.c |    3 +++
>> >>  vp8.c     |    3 +++
>> >>  vp9.c     |    3 +++
>> >>  6 files changed, 20 insertions(+)
>> >> 122f9df511e4680d0027afae5d4f9f2f1880874e  0065-avcodec-add-AVHWAccel.flush-callback.patch
>> >> From 93223fa95389c60c015cfcee22784a1bf0fdb05b Mon Sep 17 00:00:00 2001
>> >> From: Lynne <dev at lynne.ee>
>> >> Date: Fri, 6 Jan 2023 03:32:56 +0100
>> >> Subject: [PATCH 65/72] avcodec: add AVHWAccel.flush callback
>> >>
>> >
>> > this patch seems to break fate-vp8-size-change
>> >
>>
>> Thanks, philipl also reported this, fixed in my branch
>>
>> https://github.com/cyanreg/FFmpeg/tree/vulkan_staging
>>
>
> ok that works, that said
> is there consensus that i should create the release branch "now"?
> It seems no review is going on in public of these patches and we should do
> the release "soon", i am asking as i dont want to just surprise anyone with
> making the branch before giving a final call
>

We need a few more days, folks are reviewing the patches
mainly on IRC and on github, since they're large.
Functionality is on-par with the current code, and it fixes
so much, apart from adding new features, I really don't want
anyone to use the old code.
The old code is so bad, even if this patchset is broken in
some ways, it would still be a big improvement over the old code.

Most of the code is code I maintain, and has been tested and
partially reviewed already by two developers who know Vulkan
and GPU code in general. I'd still like to have this in 6.0, so
I think what I should ask is if there are any objections to merging
this as-is, and fixing any issues during the testing period.

Otherwise, I'd like for 6.1 to be released no later than April.