Відео

How come? It's just regular 3D graphics :)

@@andreelindgren522 it just gives off old school primitive early 2000's games sort of vibes, which usually had a fixed POV that was from one angle. That and the angles of the landscape changing really confuse my brain

Haha, I agree 😅

Haha, nice :)

Yeah, unfortunately a lot of niche things going on without much explanation really

Me neither :D this is going to be a game that I would like play myself. However, I'm not using any game engine, nor any other third party libs, so there is a lot of work to be done.

Haha, sorry about that. Just added some simple deformation to the terrain, to make sure the new collision stuff worked properly. Previously, I only did one check against the height of the terrain, which was fast but teleported the player up on ledges, and didn't support walls or caves.

taking notes :)

Played Diablo2 a lot, it's one of the big sources of inspiration :)

yes, a couple of human batteries were used just for those, no doubt ;)

I never played that game, but I'm more going for a fantasy kind of game with this one

@@andreelindgren522 wish yeah the most luck.

Thank you! And don't give up :))

Thank you! I'm kind of in a prototype phase now.

Most rendering engines (that don't use ray tracing) only have one basic translucent layer pass. That means, when two translucent objects pass behind one another, the further one usually disappears behind the front one. No color mixing, and no shape details visible through the first level of translucency. Minecraft is a great example of this - If you stack multiple walls of different-colored tinted glass, and you try to look through that structure, you'll only see the nearest color of glass and the rest simply aren't visible through it. This video shows off a technique which allows a ton of shapes to render through one another, in the correct order for color mixing, and with good performance. I don't know if this adequately explains what OP put in to this project, but that's a rough breakdown of the outcome

@@dianjonker Good explanation :) However, most engines tries to render translucency back to front, which usually works quite well, but sometimes breaks.

Thanks! I probably should mention though, that pixels are processed sequentially, meaning I can take advantage of spatial coherence (the triangle order stays mostly the same). So I'm not sure how well it would work on a GPU.

Thanks! Hehe, yes there are quite a lot of objects :D By per-object clipping I meant that there is no scene hierarchy for speeding up the clipping, which would make sense in a real (more static) scene, but the system was built for debug rendering and doesn't know anything about any scene. And no occluders, only view frustum culling. Yes, you are correct about the EPA stuff, except the convex hulls are described using convex components, i.e. not only vertices, but also transformed boxes, spheres etc.. Also, all the object positions are animated using sin/cos, which turned out to be quite CPU-heavy (it's close to 1B/s towards the end). I actually had to cache and reuse the same values over many objects, to not waste too much of the the available CPU, and even though the program is heavily threaded, constructing all the convex hulls every frame used almost all the available CPU. I don't know how something like EPA would run on the GPU, but would be interesting to see.

i agree lol this is insane

Thank you! I've never really implemented SAT myself, but I would say GJK + EPA is quite a lot to implement correctly, especially if you're only going to use it for tetrahedra. SAT can be optimized for temporal coherence, but you might even want to consider some direct solution, specific for tetrahedra-pairs. Like, for example, in the box-box collision scenario, where it's possible to check closest planes/edges/points, and get many contact points at the same time.

my pleasure! I guess the UA-cam algorithm is kind enough to let the cats in ;)

The heavy part is learning the math I suppose. However, you need to be coding in some language at least, if you want to implement these algorithms. What you see in the video is used for collision detection, in my game engine where speed is essential, so it's all coded in c++.

Education helps but a lot of people, me included, are self taught in many ways, and all of the information needed is on the Internet nowadays. I suggest using modern GPU features for better performance than I'm getting. Good luck with your engine!

I suggest you start by bringing in some existing library (eg. Dear ImGui) to your projects. Learn how to use an IMGui first, and see what problems it solves or doesn't. There might even be some Lisp bindings for Dear ImGui, who knows :)

Hmm, good question! Must have been my background music :) Sounds like it is something from a "programming" music playlist, you can check for it in one of these playlists: ua-cam.com/video/pmxYePDPV6M/v-deo.html

Oh! No wonder I felt it was so familiar.