This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Fix ldbl-128ibm hypotl inaccuracy for arguments with large ratio(bug 14868)
- From: Andreas Jaeger <aj at suse dot com>
- To: "Joseph S. Myers" <joseph at codesourcery dot com>
- Cc: libc-alpha at sourceware dot org
- Date: Thu, 22 Nov 2012 07:38:05 +0100
- Subject: Re: Fix ldbl-128ibm hypotl inaccuracy for arguments with large ratio(bug 14868)
- References: <Pine.LNX.4.64.1211212354050.11924@digraph.polyomino.org.uk>
On 11/22/2012 12:55 AM, Joseph S. Myers wrote:
Bug 14868 is inaccuracy of ldbl128-ibm hypotl when the ratio of the
arguments is above 2**60, but not big enough for a+b to be a correct
calculation of the result. 2**60 is the ratio used in the dbl-64
code; this patch fixes the ldbl128-ibm code by changing the ratio
there to 2**120.
Tested that this fixes the test in question, and the previously seen
16 ulp failures for csqrtl, on powerpc, and tested x86_64 and x86 that
no new ulps are needed for the new test. (On powerpc there's still a
spurious underflow exception from csqrtl in the results, but that
appears to be a separate bug.)
Thanks, this is fine,
Andreas
--
Andreas Jaeger aj@{suse.com,opensuse.org} Twitter/Identica: jaegerandi
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn,Jennifer Guild,Felix Imendörffer,HRB16746 (AG Nürnberg)
GPG fingerprint = 93A3 365E CE47 B889 DF7F FED1 389A 563C C272 A126