New Algorithms for Vector Elementary Functions

Elementary functions, such as exp, log, and sin, are among the most studied mathematical objects, and they are widely used in mathematical models. Simulating such models and fitting them to data often involves computing elementary functions on large arrays of inputs. While computing exp on large array is often perceived as expensive computational operation, in this talk will suggest algorithms which make such computations memory bandwidth-bound, and discuss why existing algorithms can not achieve the same level of performance.