Posts by tkalvas@mastodontti.fi
 (DIR) Post #AyAExLOiPhcSkK9vrE by tkalvas@mastodontti.fi
       2025-09-08T12:37:54Z
       
       0 likes, 0 repeats
       
       @blog asin and acos are often smells of poor numerical accuracy and/or singularity problems. Often the right way is to do your intermediate calculations with a bunch of vector operations (possibly including said dot products) and come back to angles with atan2 at the end.It might have been partially a performance point originally, but these days the truth is in numerical usefulness concerns.
       
 (DIR) Post #B1e66Z2dVhayDuV5rk by tkalvas@mastodontti.fi
       2025-12-26T07:10:36Z
       
       0 likes, 0 repeats
       
       @mcc This was (maybe partially) solved in many Common Lisp implementations decades ago. We took it as far as we could in SBCL, which compiled everything even if it didn't type well. Of course this has runtime data representation implications, everything has to be directly inspectable.But it was insanely fast when you got the typing right. And the compilation warnings when you didn't helped you to get there.In some ways CL is completely stuck in the 80s, but in others it's still revolutionary