Maybe for you it is, do you know that before Gino and Ericson re-wrote JGK algorithm?Erwin Coumans wrote:Wait, there is more then 2 people. I can confirm it is more intuitive
There are over a dozen packages (physics engines and stand alone collision systems) that are doing the same thing, some of them are extremetely efficient, no one claims a rediscovery of Girvert and Johnson algorithm.
Aren?t does your own personal opinions as facts? has you tested with simplexes of 5 or 10000 edges or more? it should be equallt efficient.Erwin Coumans wrote: I did performance and robustness comparisons, and they are not much different. Just run the latest bullet demo if you like. Earlier in this threat I posted the Solid 3.5 integration files to compare it yourself
Code size between bullet simplex solver and solid simplex solver is similar too, I think that bullet simplex solver code is actually smaller, but I have to verify.
So in the end, not much difference:
BTW I run the demos, like you say bullet is not better.
Again these are just personal opinions, nothing there say Bullet has the edge over Solid or any other collision library for that matter, because it is using a better version of GJK. Like I say it requires a stronger proof than just a speculations.Erwin Coumans wrote:(Bullet versus Solid 3.5 simplex solver)
0) the geometric version is more intuitive
1) robustness, code size and performance is similar
1) caching can hurt, keeping data in registers is better
2) between frames: its better to cache the separating vector, and avoid gjk all toghether, instead of caching the determinants
Erwin
About Cache on PS3 obviously you are talking on and area of expertise very few people has knowledge since it is not a released nor it is an open platform. Whoever it seems that from your point of view the specific detail of a particular platform are reason enough for overthrown good practices of software engineering. I am so convince yet.