Fast Distance Approximation
I've implemented ropes in my simulation, but it gets really slow with many ropes and many particles.
Does anyone know a way to quickly approximate the distance between two points, or maybe a fast square root approximation?
Thanks in advance,
Mr. Big
Read this for a fast (and pretty exact) square root appriximation:
http://www.math.purdue.edu/~clomont/Math/Papers/2003/InvSqrt.pdf
If you want code look at the end.
http://www.math.purdue.edu/~clomont/Math/Papers/2003/InvSqrt.pdf
If you want code look at the end.
I found exactly what I needed on this page: http://en.wikipedia.org/wiki/Euclidean_distance
Your link is very useful too!
Thanks!
Your link is very useful too!
Thanks!
You might want to look into SSE. It has special operations for fast square root and fast inverse square root. Since you're simulating many ropes and particles you're probably applying the same algorithms to them. If this is the case using SSE could result in up to a 4x improvement in speed (assuming a good implementation) since it performs 4 floating point operations in parallel.
Quote:Original post by mrbig
Does anyone know a way to quickly approximate the distance between two points, or maybe a fast square root approximation?
d = (A.a - B.a)*(A.a - B.a) + (A.b - B.b) * ...
or
d = (A - B) dot (A - B)
Re SSE
If it will not be killed in throughput and latency.
Quote:Original post by Raghar
d = (A.a - B.a)*(A.a - B.a) + (A.b - B.b) * ...
or
d = (A - B) dot (A - B)
This will compute |A-B|², the square of the distance.
If you can tolerate a maximum error of 6.5%, you should use an octagonal approximation: Approximate the unit circle with an octagon, giving a case-analysis on eight linear functions. The hexadecagon approximation is better yet, but (although still considerably faster than a sqrt) is a fair bit slower than the octagonal function, and is a pain to code.
More accurate approximations yet can be produced using Richardson Extrapolation (daniel_i_l's article seems like a good implementation) and if you are prepared to do the work yourself, you can tailor the speed/error tradeoff to suit your needs.
Regards
Admiral
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement