A performance enhancement of the Payne-Hanek range reduction algorithm (Q7360674)

This paper analyzes the Payne--Hanek range reduction algorithm used for evaluating trigonometric functions with large floating-point inputs and presents a branch-free variation using only floating-point arithmetic to reduce latency and improve SIMD vectorization. This new method operates directly over double-precision inputs ($|x| \ge 2^{16}$) and is well-suited for hardware with FMA instructions, improving both latency and throughput compared to existing implementations.