Improve rendering performance for multiple display usecase

[moetayuko]

I'm unfamiliar with OpenGL, so please let me know if I'm wrong.

If I read the code correctly, each window has its own FBO, and the scenes and videos are rendered separately. This is fine for single-screen but inefficient for multi-screen, and I observe increased GPU usage in the latter case.

From what I understand, if the same content is used for multiple screens, it only needs to be rendered once, and then copied to the rest screens, maybe it can be done by glBlitFramebuffer?

[Almamu]

Backgrounds that have interaction (mouse parallax or web backgrounds for example) won't always look the same on all screens as they react to user's movements, so in order to support that, each screen has it's own wallpaper rendered separately. We could improve the situation in a couple different ways tho:

• Add a command-line switch to clone the same background to all screens and override this behavior
• Detect what backgrounds that use parallax or web and only render those separately

It's been some time since I last tried this on Windows, so I'm not 100% sure on how it handles these situations.

[moetayuko]

I'm wondering if the following policy makes sense:

• web backgrounds are rendered with cef, and can continue to render separately
• video backgrounds are rendered with mpv which does not accept interactions, so they should be rendered once and copied to each display
• for scene backgrounds, a cmd line toggle can be added to let the user trade-off between accurate interaction and efficiency. Detecting the presence of interaction can serve as a complement to the toggle.

[Almamu]

That sounds like a reasonable approach, yeah. There's a special case that might require some extra handling too. As it stands right now, one instance of linux-wallpaperengine can run different backgrounds on each screen, so that has to be taken into account too (I guess applying the same logic for screens sharing the same background id should be enough).