Post #760561

@djnavarro Whenever floating point reproducibility comes up there are always two types of comment: “IEEE 754 floating point is deterministic and this is a solved problem” and “floating point is inherently inexact and consistency is impossible”. And the statements are made with such strong belief from both sides! Of course with differing BLAS, compiler and libraries the truth is somewhere in the middle, but the subtleties are hidden until you dive into the weeds and actually try it.