A Case for Correctly Rounded Math Libraries

Explore a comprehensive overview of the RLIBM project in this 43-minute conference talk from Papers We Love Conf 2022. Delve into the innovative approach of approximating the correctly rounded result of elementary functions, rather than their real values, for multiple representations and rounding modes. Discover how this method creates larger intervals of freedom for polynomial approximations, leading to more accurate results. Learn about the structuring of polynomial approximation generation as a linear programming problem to produce correctly rounded results for all inputs. Gain insights into why the RLIBM project's findings support mandating correctly rounded results for representations with 32 bits or fewer. Understand the potential impact of this research on the future of mathematical libraries and computational accuracy.