Many algorithms for scientific computing compute transcendental functions (log, sin, and cos) on arrays of floating point values. These are heavily used operations with strict correctness requirements and thus have been the target of many optimization efforts over the years. Faster versions of these functions can have a huge impact on the performance of many applications in the scientific computing domain.
In this area, the Yeppp! software suite can be considered state-of-the-art. Primarily written at Georgia Institute of Technology by Marat Dukhan, Yeppp! provides optimized implementations of modern processors of these functions, which are much faster compared to the implementations in system libraries.
Julia has a very easy-to-use binding to Yeppp! within a package. It can be installed using the in-built package management mechanism Pkg.add("Yeppp")
. Once installed, the functions are available with the Yeppp
module. There is no simpler way to get a 4x performance boost. With performance...