DOI: https://doi.org/10.26089/NumMet.v27r103

High-performance implementation of exponential functions for RISC-V processors

Authors

  • Elena A. Panova
  • Valentin D. Volokitin
  • Evgeny A. Kozinov
  • Iosif B. Meyerov

Keywords:

mathematical functions
exponent function
exp
exp2
expm1
vector calculations
high-performance calculations
RISC-V architecture

Abstract

The development of a high-performance vector library for standard mathematical functions represents a significant objective for the advancement of HPC on platforms based on the RISC-V architecture. This paper presents a detailed implementation of the exp, exp2, and expm1 functions for double-precision, single-precision, and half-precision floating-point types. Special attention is paid to aspects of computational accuracy and performance. A detailed comparison with other existing vector math libraries for the RISC-V architecture is provided.


Downloads

Published

2026-01-29

Issue

Vol. 27 (2026): Issue 1.

Section

Parallel software tools and technologies

Authors

Elena A. Panova

National Research Lobachevsky State University of Nizhny Novgorod

• Assistant

Valentin D. Volokitin

National Research Lobachevsky State University of Nizhny Novgorod

• Senior Lecturer

Evgeny A. Kozinov

National Research Lobachevsky State University of Nizhny Novgorod

• Associate Professor

Iosif B. Meyerov

National Research Lobachevsky State University of Nizhny Novgorod

• Head of Department

References

  1. J.-M. Muller, “Elementary functions and approximate computing,” Proceedings of the IEEE. 108 (12), 2136–2149 (2020).
    doi 10.1109/JPROC.2020.2991885
  2. W. Liu, F. Lombardi, and M. Shulte, “A Retrospective and Prospective View of Approximate Computing [Point of View,” Proceedings of the IEEE. 108 (3), 394–399 (2020).
    doi 10.1109/JPROC.2020.2975695
  3. J. E. Volder, “The CORDIC trigonometric computing technique,” IRE Transactions on electronic computers. EC-8 (3), 330–334 (1959).
    doi 10.1109/TEC.1959.5222693
  4. E. Remes, “Sur un procédé convergent d’approximations successives pour déterminer les polynômes d’approximation,” Comptes rendus de l’Académie des Sciences. 198, 2063–2065 (1934).
  5. N. Brisebarre, S. Chevillard, “Efficient polynomial L^infty-approximations,” in 18th IEEE Symposium on Computer Arithmetic (ARITH ’07), Jun 2007, Montpellier, France(IEEE, 2007), 169–176.
    doi 10.1109/ARITH.2007.17
  6. Jr. Thacher, C. Henry, “Software Manual for the Elementary Functions (W. J. Cody,  Jr. and W. Waite),” SIAM Rev. 24 (1), 91–93 (1982).
    doi 10.1137/1024023
  7. P. Kornerup, J.-M. Muller, “Extending the Range of the Cody and Waite Range Reduction Method,” 2005,
    https://www.academia.edu/62418560/Extending_the_Range_of_the_Cody_and_Waite_Range_Reduction_Method Cited January 21, 2026.
  8. P.-T. P. Tang, “Table-driven implementation of the exponential function in IEEE floating-point arithmetic,” ACM Transactions on Mathematical Software (TOMS). 15 (2), 144–157 (1989).
    doi 10.1145/63522.214389
  9. P.-T. P. Tang, “Table-driven implementation of the logarithm function in IEEE floating-point arithmetic,” ACM Transactions on Mathematical Software (TOMS). 16 (4), 378–400 (1990).
    doi 10.1145/98267.98294
  10. J.-M. Muller, Elementary functions: Algorithms and Implementation. 3rd Edition(Birkhäuser Boston, MA, 2016).
    doi 10.1007/978-1-4899-7983-4
  11. N. Brisebarre, G. Hanrot, J.-M. Muller, et al., “Correctly-Rounded Evaluation of a Function: Why, How, and at What Cost?’’ACM Computing Surveys. 58 (1), 1–34 (2025).
    doi 10.1145/3747840
  12. The GNU MPFR Library.
    https://www.mpfr.org/ Cited January 21, 2026.
  13. C. Daramy, D. Defour, F. de Dinechin, and J.-M. Muller, “CR-LIBM: a correctly rounded elementary function library,” inProc. SPIE 5205, Advanced Signal Processing Algorithms, Architectures, and Implementations XIII(SPIE, 2003), 5205, 458–464.
    doi 10.1117/12.505591
  14. SLEEF Vectorized Math Library.
    https://sleef.org/ Cited January 21, 2026.
  15. GitHub - rvvpl/rvvmf.
    https://github.com/rvvpl/rvvmf Cited January 21, 2026.
  16. P.-T. P. Tang, “An Open-Source RISC-V Vector Math Library,” in 2024 IEEE 31st Symposium on Computer Arithmetic (ARITH), Malaga, Spain, June 10–12, 2024(IEEE, 2024), pp. 60–67.
    doi 10.1109/ARITH61463.2024.00019
  17. GitHub - rivosinc/veclibm: Vector math library using RISC-V vector ISA via C intrinsic.
    https://github.com/rivosinc/veclibm Cited January 22, 2025.
  18. E. A. Panova, V. D. Volokitin, E. A. Kozinov, and I. B. Meyerov, “High-performance implementation of exp and expm1 functions for RISC-V processors,” in Proc. Int. Conf. on Russian Supercomputing Days, Moscow, Russia, September 29–30, 2025, (MAX Press, Moscow, 2025), pp. 67–84. [in Russian].
  19. J.-M. Muller, On the definition of ulp(x): research report / Thème SYM – Systèmes symboliques, Projet Arénaire. Rapport de recherche n° 5504, 2005Institut National de Recherche en Informatique et en Automatique (INRIA), 2005.
    https://www.academia.edu/62418489/On_the_definition_of_ulp_x_ Cited January 22, 2025.
  20. The GNU C Library - GNU Project - Free Software Foundation (FSF).
    https://www.gnu.org/software/libc/manual/html_node/index.html Cited January 22, 2025.
  21. B. Gladman, V. Innocente, J. Mather, and P Zimmermann, Accuracy of Mathematical Functions in Single, Double, Double Extended, and Quadruple Precision, HAL preprint, 2025,
    https://inria.hal.science/hal-03141101v8 Cited January 22, 2025.
  22. Sollya software tool.
    https://www.sollya.org/ Cited January 22, 2025.
  23. G. Estrin, “Organization of computer systems: the fixed plus variable structure computer,” in Papers presented at the May 3–5, 1960, western joint IRE-AIEE-ACM computer conference, May 3–5, 1960, San Francisco, California, USA(ACM, New York, NY, USA, 1960), 33–40.
    doi 10.1145/1460361.1460365
  24. T. J. Dekker, “A floating-point technique for extending the available precision,” Numerische Mathematik. 18 (3), 224–242 (1971).
    doi 10.1007/BF01397083
  25. C. Daramy, D. Defour, F. de Dinechin, and J.-M. Muller, CR-LIBM: The evaluation of theexponential: research report / LIP RR–2003–37 Laboratoire de l’informatique du parallélisme, 2003, HAL preprint, 2+37 p.
    https://hal-lara.archives-ouvertes.fr/hal-02102084/ Cited January 22, 2025.
  26. S. Linnainmaa, “Software for doubled-precision floating-point computations,” ACM Transactions on Mathematical Software (TOMS). 7 (3), 272–283 (1981).
    doi 10.1145/355958.355960
  27. V. Lefevre, J.-M. Muller, and A. Tisserand, “Toward correctly rounded transcendentals,” IEEE Transactions on Computers. 47 (11), 1235–1243 (1998).
    doi 10.1109/12.736435
  28. A. Ziv, “Fast evaluation of elementary mathematical functions with correctly rounded last bit,” ACM Transactions on Mathematical Software (TOMS). 17 (3), 410–423 (1991).
    doi 10.1145/114697.116813
  29. P.-T. P. Tang, “Table-driven implementation of the Expm1 function in IEEE floating-point arithmetic,” ACM Transactions on Mathematical Software (TOMS). 18 (2), 211–222 (1992).
    doi 10.1145/146847.146928
  30. V. D. Volokitin, E. P. Vasiliev, E. A. Kozinov, et al., “Improved Vectorization of OpenCV Algorithms for RISC-V CPUs,” Lobachevskii J. Math. 45 (1), 130–142 (2024).
    doi 10.1134/S1995080224010530
  31. A. Pirova, A. Vodeneeva, K. Kovalev, et al., “Performance optimization of BLAS algorithms with band matrices for RISC-V processors,” Future Generation Computer Systems. 174, Article No. 107936 (2025).
    doi 10.1016/j.future.2025.107936

License

Copyright (c) 2026 Е. А. Панова, В. Д. Волокитин, Е.А. Козинов, И.Б. Мееров

Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 International License.