This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Adjust thresholds in Bessel function implementations (bug 14469)
- From: paul zimmermann <Paul dot Zimmermann at inria dot fr>
- To: Joseph Myers <joseph at codesourcery dot com>
- Cc: libc-alpha at sourceware dot org
- Date: Fri, 14 Feb 2020 13:55:46 +0100
- Subject: Re: Adjust thresholds in Bessel function implementations (bug 14469)
- References: <alpine.DEB.2.21.2002132154260.22316@digraph.polyomino.org.uk>
Dear Joseph,
> Paul, could you run your exhaustive tests of j0f, j1f, y0f,
> y1f with this patch and see what effect it has on the maximum ulp
> error and the number of inputs with large errors?
here are the results on x86_64 for the binary32 functions:
Maximal error in ulps:
glibc-2.31 glibc-2.31 + patch
j0f 4760682496 6177902
j1f 714456064 2246675
y0f 15117099008 4837658
y1f 464033280 6177902
Number of incorrectly rounded results:
glibc-2.31 glibc-2.31 + patch
j0f 1353797232 1334176546
j1f 1369557306 1340594104
y0f 1314115311 1304302535
y1f 1213975420 1199498354
The maximal errors have thus been greatly reduced.
Best regards,
Paul