University of Copenhagen
People doing numerical computing are greedy. They want results now and accurately, and we have been ready to accept loss of readability and extensibility to get it. As a result, the elementary functions we take for granted in paper math are often coded in low-level languages. We can do better. The talk will go through the motivation, the work done, what remains to be done, and what further possibilities there are to improve performance and accuracy of elementary function evaluations in Julia. Collectively, these functions are often provided in so-called “libm”’s (library of mathematical functions), and many languages oriented towards numerical computing tend to call libraries written in C or Fortran. Julia has depended on a collection of functions bundled in the C-based openlibm to provide this functionality. As we will see, there has been a push to free Julia of openlibm and similar C- or Fortran-libraries without loss of speed, accuracy or readability.
Ph.D. student in economics, JuliaNLSolvers owner and developer, Julia nerd.
