Accuracy of Mathematical Functions in Single Precision for glibc 2.31
paul zimmermann
Paul.Zimmermann@inria.fr
Mon Feb 3 06:33:00 GMT 2020
Hi,
I have checked by exhaustive search the accuracy of single precision
math functions on x86_64 (for rounding to nearest) in the just released
2.31 version:
https://homepages.loria.fr/PZimmermann/glibc/glibc231.pdf
Results are quite interesting, in particular:
- exp10 is close from being correctly rounded (only one exception)
- j0, j1, y0, y1 are quite bad (this was already known, but the "Known
Maximum Errors in Math Functions" in the reference manual do not reflect it)
- some functions have only very few cases with error >= 2 ulps
Paul Zimmermann
More information about the Libc-alpha
mailing list