Implementation of Rlibm-OMP Algorithm Based on Sunway

The performance bottleneck of math library functions has long been a common issue among major manufacturers. To overcome these bottlenecks, this paper proposes the Rlibm-OMP method, which combines the RLibm fast polynomial method with the OpenMP parallel programming model based on Sunway architecture. This approach is aimed at generating polynomial replacements for foundational library functions. The RLibm fast polynomial method aims to produce polynomial replacements that yield correct results for foundational functions across all inputs and various rounding modes. However, due to several adaptation issues when porting external platform algorithms to the Sunway platform, performance often degrades. Thus, an optimization was carried out using the OpenMP parallel programming model based on Sunway to optimize polynomial evaluations. By integrating the RLibm fast polynomial method with the OpenMP parallel programming model, it is possible not only to generate correctly rounded polynomials but also to achieve efficient parallel computations on the Sunway platform. Experimental results indicate that the polynomial results for 32-bit floating-point functions are not only accurate but also, on average, 12% faster than those achieved using the Rlibm fast polynomial method alone.