Starting in 10 minutes!! We might see a models rendering today if I get lucky!!! 🚀

So I'm looking at what a "production" Live2D renderer would look like, and I'm torn because I think I want to use wgpu, but I need some features it doesn't support to do some things efficientlly.

To do 5.3 blend modes efficiently, I need either VK_EXT_blend_operation_advanced or VK_EXT_rasterization_order_attachment_access or subpass self-dependency, but WebGPU has none of those.

For efficient masking, I think it's basically going to have to be subpasses (i.e. something that lets me render a mask to an attachment, then sample it in another subpass, that stays in tile memory/cache).

(Metal has equivalents for all this too)

Another option for both is VK_EXT_fragment_shader_interlock, but that will probably suck on tilers...

How hard is it to get stuff like this added to wgpu as an extension?

I said I wanted to work on a MOC3 optimizer, but ngl first I want to work on a high performance renderer.

Because I'm preeeeeetty sure the Cubism SDK and VTS are doing lots of really sub-optimal stuff.

Maybe there's no need to optimize models if I just make it fast ^^

STREAM: Reverse engineering Live2D MOC3 (part 3)

We have the fields! ✨
Now how do we render stuff?
How do deformers work??
Let's figure all that out!

✨ Important FAQ✨
https://lina.yt/moc3faq

🕐 Jun 10 13:00 JST / 04:00 UTC
▶️ https://youtube.com/live/5AiZTgUx_WM
#VTuber #ENVTuber #Live2D