1. Use latest nightly
2. Go to https://www.techspot.com/news/93931-nvidia-instant-nerf-ai-renders-3d-scenes-2d.html
3. There is an embedded video in the middle of the page. Click on play, drag it till approx half of the video, and then pause the video.
4. Scroll the page up and down with touchpad

ER: Smooth scrolling
AR: Choppy scrolling

Regression range picked up bug 1667303.

I tried setting "gfx.webrender.dcomp-video-vp-scaling-win= FALSE" and restarting, but the stuttering remained. I assume it means that the rest of the code might have some bug.

Profile: https://share.firefox.dev/387oHaJ

If i turn on SW-WR, the stuttering disappears.
Profile : https://share.firefox.dev/3tGNhHA

Set release status flags based on info from the regressing bug 1667303

:sotaro, since you are the author of the regressor, bug 1667303, could you take a look?
For more information, please visit auto_nag documentation.

And now i cant repro anymore, despite the fact that it was 100% reproducible earlier and I could even get a regression range.

OK. reopening this bug as I can repro it.. When I am able to reproduce this, the page will stutter across multiple browser restarts, and even system restarts. If I am not able to reproduce, it just wont repro.
It just might be limited to my setup/machine or AMD drivers.

Another profile on a youtube page : https://share.firefox.dev/36CgfQ2

Just start the video, drag the slider to midway, and pause the video. Then try to scroll the page.

If I disable the compositor, or enable sw-wr, the slowness disappears.

Mayank, what does Edge or Chrome say in about:gpu for: "Supports overlay", "YUY2 overlay support" and "NV12 overlay support"?

Can you look at the GPU section of the "Performance" tab of Task Manager and check if you notice any differences in the numbers when you can reproduce the problem vs when you can't.

I think I might be able to repro this 100% of the time. STR:

1. Go to https://www.youtube.com/watch?v=y6sPHFoZnVk
2. Start the video, and then pause
3. Scroll the page while the video is on the screen.
   If I disable the compositor or enable sw-wr, the stutter disappears.

Compositor on :https://share.firefox.dev/3qId0O0
Compositor off: https://share.firefox.dev/3qHpLs7

With compositor : CPU : 6-10%, GPU 11-16%
Without compositor: CPU: 8-12%, GPU 6-10%
So I wouldnt say there is much of a difference.

Happy to run any debug build or share any logging etc.

Do you have different categories in the GPU pane? I have "3D', "Copy", "Video Decode" and "Video Processing" on Intel?

I tried with all the categories. The only categiry that registers anything was "3d", which i reported.
Profile with surprisingly some extra symbols : https://share.firefox.dev/3NsLgXy

What categories do you have?

3d, copy, High priority compute, compute 0, compute 1, timer 0, Security 1, Video Decode 1, Video Encode 0, Video Codec 0

might be more reproducible on https://economictimes.indiatimes.com/ .. There is an embedded scaled video in the right pane of the page (how far down will depend on your screen and the number of ads)... scrolling past that is stuttery and 100% reproducible.

Since bug 1667303 fix, DCSurfaceVideo::PresentVideo() is called even when it is not necessary. DCSurfaceVideo::PresentVideo() needs to be called only when RenderTextureHost, swapChainSize or VideoSwapChain are updated.

I created Bug 1761906.

This was fixed by Bug 1761906.

this was a bad regression.. Can a test be added ?

Or if a perf profile captured via the gecko profiler can capture the slow video (with appropriate actionable information)

(In reply to Mayank Bansal from comment #20)

this was a bad regression.. Can a test be added ?

It seems possible. I am going to look into it.