k8vavoom: no good thing ever dies! (2023, Dec 22 build)

@vigilantdoomer, there is no way to properly render voxel objects without converting them to 3d models (at least without radically bumping GPU requirements, and writing a brand new shader-based voxel engine). so voxels are converted to 3d models, hence they treated like any other 3d models, and described as such. i see it as a nice thing, because there is no need to learn different syntaxes to add different kind of 3d objects to the game.

the only bad decision was choosing XML for model description format. but that's what Janis did, and i don't feel like introducing another syntax will help (i have to keep XML format anyway, to not break compatibility with the existing Vavoom projects).

Did you ever get around to defining a range of frames instead of having to do them one by one? For md2 models I mean.

Excellent!

Need to experiment with it one of these days.

fun fact: most KVX objects has some invisible voxel faces marked as visible. Boss Cube from reikall, for example, is empty inside, but its contour voxels marked as "visible from inside", and the engine has to create triangles for the faces the player could never see anyway. i implemented visible face recalculation algorithm, and the result is:

before: 11065 unique vertices, 5556 triangles

after: 9587 unique vertices, 4820 triangles

it may not look like a big improvement, but using this algo i managed to get almost +2 FPS on the test scene.

Nice work. Glad to see more projects carrying the metaphorical torch.

Yeah, sure.  I'm pretty lazy and never seem to get around to updating the site of my own volition.  But when someone gives me a nudge to do something then I usually do it.  Keep me posted.     

just for fun: standalone converter (my testbed), in "ultra" mode, with t-junction fixer.

the famous Voxlap test scene, fully converted:


some stats:

thank you! just in case: this isn't any "voxel engine", this is the development testbed for voxel model -> polygon model converter for k8vavoom. of course, nobody would use such huge voxel models in Doom, but hey, if i can make my conversion code fast on THIS… i love to test my algos on data nobody in their sane mind will feed to them. ;-)

fun fact: my old GT 720 GPU can easily render this whole scene (all 1188394 triangles) with 35 FPS. ;-)

but the most important thing here is that i have the tested (i hope ;-) code to perform near-optimal voxel conversion. i believe that my code produces the smallest models (in triangle count) among all "retro FPS engines" that support KVX/KV6 voxels, and it can fix visual "cracks" too. and it does this for DVP/reikall voxels in less then a millisecond per model.

at least i inspected GZDoom code, and it can't come even close to my results. ;-) (and it doesn't even try to fix t-junctions, but can introduce them.)

yes, exactly this.


nope, that's the easiest part. ;-) actually, KVX/KV6 already contains this info. but… but it is sometimes wrong, and has some invisible faces marked as visible. so i implemented my own "hollow fill" code, which leaves only exterior faces, properly marked. as a consequence, my code can load any voxel format, in any order: it will convert that into the internal voxel representation.

the actual challenge is finding the best rectangle decomposition. doing that with brute force is something around O((X*Y*Z*6)*(N*M)). in other words, it is FUCKIN' SLOW. my first iterations converted that voxlap thingy in ~40 minutes. and the current code does it in less than 2 minutes. it can optionally fix t-junctions too.

tbh, i can do it even faster, but that will require adding at least the same amout of code i already wrote, and won't cut the time to seconds anyway. so i decided to stop here. for normal voxel models it is almost instant, and i'm ok with that.

also, the code is basically a self-contained library, that can be used in any engine with minimal modifications. so once i put it into the repo, other engines will be able to take and use it too. the final result of the library is vertex/indicies data and color texture, ready to be uploaded to GPU.

thank you! no, i didn't not really looked for voxel packs, just took what i knew about… i know about DVP, and that's basically all. and i was told about reikall. now about this too, will take a look, thank you again! ;-)

btw, beta of voxel conversion code is in the repo. you may consider stealing it for your great EDGE. it is in "source/render/modelparse/voxelib.*". i really want people to start using this code for voxel conversion, because it hurts me to see various sub-par methods. #trianglecountmatters!

p.s.: of course, i may improve it in the future, and make it completely independent even of corelib. for now, the code is using `TArray` and `TMap` (basically, `std::vector` and `std::map`) from k8vavoom core library. but that's mostly all, and i think i can implement very simple hash table and dynarrays right in the lib code, because why not… ;-)

still, i made "voxelib" usable as a standalone library now. it includes KVX/KV6 loaders too. and you don't need neither other k8vavoom libraries, nor even STL to use it! ;-)

as for voxel support, it is quite easy to implement them as Yet Another Kind Of 3D Models, as i did. so you only need to get my library, and you will get a skin texture, and a triangle soup, which you could simply copy to your data structures. and BOOM! EDGE has voxel support, and you can write a HUGE ANNOUNCEMENT! ;-)

that would be a golden topic title! ;-)


sure. anyway, the thing is still red hot, so it would be wise to wait a little until i iron out all the bugs i'll spot. ;-)