[Phantom Hoover]

CryEngine doesn't give a **** about polycount.

From what I can see, neither does FSO.

I thought FSO used the base model for collision modelling, which grinds to a halt if you have too many polys?

[Shivan Hunter]

Collision between two models whose bounding boxes intersect is O(log n) w.r.t. polies, IIRC. Huge polycounts can be an issue, and it's good to have a simple collision mesh for a variety of reasons, but the real constraint is number of ships in the mission (because sorting the ships and finding out which ones do intersect is a handshake probem AKA O(n^2) {and I do remember something about using sorting algorithms like quicksort in x, z, and y to whittle them down as opposed to comparing each and every one, which could give us results near O(n*log(n)), but idk what the progress on that is if any, it might even have been in that collision update a while back}).

[General Battuta]

Collision between two models whose bounding boxes intersect is O(log n) w.r.t. polies, IIRC. Huge polycounts can be an issue, and it's good to have a simple collision mesh for a variety of reasons, but the real constraint is number of ships in the mission (because sorting the ships and finding out which ones do intersect is a handshake probem AKA O(n^2) {and I do remember something about using sorting algorithms like quicksort in x, z, and y to whittle them down as opposed to comparing each and every one, which could give us results near O(n*log(n)), but idk what the progress on that is if any, it might even have been in that collision update a while back}).

[MatthTheGeek]

I thought FSO used the base model for collision modelling, which grinds to a halt if you have too many polys?

FSO uses all subobjects in LOD0 that aren't marked as $no_collide for collision detection. Which is why everything you see is marked as $no_collide and I have a lower-poly collision mesh set as invisible and $collide_invisible (I don't remember if that's the exact syntax, I don't have the pof in front of me).

Note: even without collision mesh, firing at 18 constellations flying around barely brings me down to solid 20fps. Swifty did a good job.

[The E]

Collision detection in FSO was never that bad.

We perform quite a few first-pass optimizations in order to cut down the number of collisions we evaluate (like, for example, primary weapon shots will never collide. Or things like a weapon moving away from another object. Stuff like that.), and once we get into actually having to evaluate collisions against another object, we go through a BSP tree to quickly evaluate them (this is the reason why the pof format is as omgwtfhuge as it is, BSP trees are efficient to evaluate but not efficient to store).

Basically, our collision detection is usually much better than O(n log n). The issue with large models such as the ones under discussion is that evaluating the BSP tree takes ages, and once we go into it, it is almost guaranteed that we'll be going through it all the way.

That's how collision meshes work: By not evaluating collisions against the main mesh, everything speeds up by a lot.

[Dragon]

Also, IIRC, FSO doesn't do mesh-to-mesh collisions at all. In a, for example, fighter-to-capship collision, the smaller of the ships is approximated by it's radius. I first noticed it when landing a particularly flat fighter on a carrier deck. I'm not sure how collisions between objects of similar size are handled, but IIRC, it's also an approximation (either radius or bounding box is used).

[MatthTheGeek]

So yeah, I was fully expecting that firing on a whole 500k-poly mesh while on a debug build would simply melt FSO, turns out it just shrugged it off and kept solid 60fps like it was nothing. And it'll go even better once I put collision meshes on everything.

So you can stop dreaming about supercomputers. FSO's pretty cool guy, eat polygons for breakfast and doesn't afraid of anything.

[Col. Fishguts]

How many constellations can you get on screen simultaneously until you get a slide show?

[Shivan Hunter]

yay for dicking around with GIMP instead of fredding

[0rph3u5]

I see a hole in space ... not sure if I like that

[X3N0-Life-Form]

You guys gotta stop dividing by zero, it's bad for your skin.

[Luis Dias]

i think it's a not-so-incompetent rendition of a black hole. Of course, the problem is that it's assymetrical. It shouldn't be. Nevertheless.

[Shivan Hunter]

Well, it has two mathematical errors, one of which I can correct by rotating part of it 180 degrees. Luis: It's perfectly circular in the skybox, it only looks asymmetrical in that pic because it's off to the side of the screen. FOV stuff. Same discussion as was had about the sun graphics I did a few pages back.

[Luis Dias]

I suspected as much, but it seemed too skewed.

[esarai]

YEAAAH GRAVITIC LENSING.  FLY INTO THE BLACK HOLE, THE FUTURE IS NOW!

[Kolgena]

oh man there needs to be caustics on that ships lighting and environment map because gravitic lensing

fso most make update

[Shivan Hunter]

?

The hole is GIMP'd into the skybox, and the envmap uses the skybox, so...

Dunno about caustics, either whether it'd be noticeable on ships from a physics perspective or feasible in the engine (or doable with ~shaders~). Could just position a low-intensity sun with a black bitmap and $NoGlare: at the hole so a tiny amount of light comes from its direction.

[General Battuta]

Thank you for ~tilde posting~, the best posting.

[Phantom Hoover]

pretension tildes top all other forms of textual inflection

[Kolgena]

?

The hole is GIMP'd into the skybox, and the envmap uses the skybox, so...

Dunno about caustics, either whether it'd be noticeable on ships from a physics perspective or feasible in the engine (or doable with ~shaders~). Could just position a low-intensity sun with a black bitmap and $NoGlare: at the hole so a tiny amount of light comes from its direction.

oh i thought that it was a warp hole the fighter was entering but i guess that is wrong.

fso most make update anyways