So, I’m doing this thing. My goal is pretty simple. I want to check in once a day, improve or tweak something, however trivial, and take a screenshot of it.
Today I tried some new palettes for a stylized renderer I’m working on:
Hopefully I’ll find some use for it during Ludum Dare.
Continuing work on the 2D soft shape renderer. The plan is to make something akin to mudbun/clayxels, but for 2D.
While the shapes now seem to blend rather smoothly (pun totally intentended), today I worked on getting colors to blend as well.
I spent a lot of time botching it up.
But finally I got something I was happy with. It’s a weighted average color where the weights are the negative distance from the source shape + some offset to the power of 3.19, which is kind of weird. It probably makes sense to make it configurable per shape (right now it’s on the material).
I was also struggling to keep the number of render passes low, because I needed 3 color channels, alpha, distance, and weighted average total. Eventually, I figured out how to properly do multiple render targets in one pass, which means that for the moment I have plenty of room (4 render targets * 4 channels = 16).
In any case, I’m quite optimistic with how it works, and I think I can make something that is both very performant, author-friendly and flexible.
As intended, scaling now produces no artifacts :)
I spent 3 hours trying to get GPU instancing to work for a custom sprite renderer shader with a material property block, only to figure out it’s a known bug that Unity decided not to fix and not to warn about. I mean they could at least have printed some warning in the console that setting a material property block on a sprite renderer will break instancing.
Anyways, once I found out it was their bug, I spent 15 minutes switching over to the mesh renderer and instancing is now working smoothly.
So I still get four draw calls, but at least most sprites with the same shape are now drawn in the same draw call. That’s happening because currently, each of the primitives have their own material. I guess maybe it would make sense to make them all use the same shader and just branch based on an enum value (instead of using #ifdefs like I do now). Anyways, I’m probably going to postpone that until I encounter performance issues to avoid doing too much premature optimization.
