Accuracy of Mathematical Functions
- Reply: Steve Kargl : "Re: Accuracy of Mathematical Functions"
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Tue, 25 Aug 2026 07:29:16 UTC
Dear all, a new update of our comparison is available: https://members.loria.fr/PZimmermann/papers/accuracy.pdf Main remarks about this update: * more correctly-rounded functions in GNU libc and LLVM libc * the Intel Math Library provides some cr_xxx functions: the user can choose between the default function (log) and the correctly rounded one (cr_log) * about two years after the adoption of C23, very few libraries implement the new C23 functions Specific to FreeBSD (we are testing 15.1); * We use the generic x86_64 binary from the FreeBSD distribution. Some slightly different results might be obtained if you recompile FreeBSD using fused-multiply-add * pow still has a large error of 636 ulps for x=0x1.000002c5e2e99p+0 and y=0x1.c9eee35374af6p+31 Looking forward for improvements in the next release! Brian Gladman Vincenzo Innocente John Mather Katsuhisa Ozaki Paul Zimmermann