It seems like that won't get the "fast": that code does 4 trigonometric functions and a sqrt while the OPs approximation seems to do one cached trig call and th

It seems like that won't get the "fast": that code does 4 trigonometric functions and a sqrt while the OPs approximation seems to do one cached trig call and then one sqrt for each point.