Fast IAPWS-IF97 Evaluation: Profiling-Guided Loop Tiling and Shared-Power Scaling

The IAPWS-IF97 formulation is the international standard for water and steam thermodynamic properties, but direct formula evaluation without optimization cannot meet the performance demands of compute-intensive applications requiring millions of property calculations. Existing fast direct formula evaluations achieve only limited acceleration through repeated squaring. Alternative methods such as TTSE and SBTL avoid direct evaluation but sacrifice formula-level accuracy. This paper presents two acceleration techniques that directly optimize the IAPWS-IF97 formula evaluation while maintaining full formula accuracy: profiling-guided loop tiling and shared-power scaling. Profiling-guided loop tiling parti tions polynomial summation into cache-friendly tiles with empirically determined boundaries, enabling more effective SIMD vectorization. Shared-power scaling exploits the mathematical relationship between Gibbs/Helmholtz free energy polynomials and their partial derivatives to compute them simultaneously in a single pass, eliminating redundant power calculations. A Rust-based software package, SEUIF97, implements these methods. Benchmark results demonstrate 5x to 36x speedups over baseline implementations and 14x to 30x speedups over CoolProp IF97, the most widely used open-source implementation, with the largest improvements in Regions 1, 2, and 3.