We need SDL3.0 support render YUVA format (which include alpha channel).

[huangjinshe]

Could we let SDL3 support YUVA data with alpha channel now? because in SDL2 we need to use ffmpeg format filter convert avframe(original format YUVA420) to RGBA and then render it, I don't know after we use zero-copy(for hardware acceleration) does it support it or not for directly update texture with yuva format. you know such format like vp8(has alpha tunnel video) more common and popular now.
I've already mention it in here before: Added an example program of video decoding with ffmpeg @slouken

Here is the vp8 video file(yuva420p, use libvpx in ffmpeg) example(randomly download from some website):
v.webm

[slouken]

I don't know of any hardware decoder that outputs YUVA, so using swscale to convert to RGBA seems like a reasonable solution here.

It is in theory possible to add shaders to directly render YUVA image formats, but writing, testing, and maintaining those across all the supported renderers is a lot of work. Let's leave it here for now and revisit that once the SDL GPU API is implemented.

If you find that you need that level of performance sooner, feel free to open a PR for your work in progress.

[huangjinshe]

I thought support VP8 format hardware decoding mean support YUVA. Intel support vp8 since 2012 around, more than 10 years.
Last time when I use vp8 decoe by libvpx in cpu, we need to manually set the format type from YUV420P to YUVA420P before use sws_scale, and then convert to RGBA . I don't know how to handle afer hardware decoding, because it will turn to AV_PIX_FMT_NV12.
swscale good, but in SDL2 it only support convert in CPU, I don't know how in SDL3, if don't we still need to lose zero-copy. huge sacrifice of performance.

[slouken]

As far as I can tell VP8 hardware decode outputs NV12, losing any transparency information - as you mentioned, output to AV_PIX_FMT_NV12.

[huangjinshe]

Or maybe provide a callback before render? let us convert pixel format or change the width or height, or something(like create new avframe use GPU power to convert all stuf we want). but I'm not really sure ffmpeg support directly edit avframe by GPU or no.