This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH COMMITTED] Update Sparc ULPs.
- From: "Joseph S. Myers" <joseph at codesourcery dot com>
- To: David Miller <davem at davemloft dot net>
- Cc: <libc-alpha at sourceware dot org>
- Date: Tue, 14 May 2013 23:28:35 +0000
- Subject: Re: [PATCH COMMITTED] Update Sparc ULPs.
- References: <20130514 dot 154654 dot 1688995032370251284 dot davem at davemloft dot net>
On Tue, 14 May 2013, David Miller wrote:
> Joseph, was the decrease in tgamma precision (end of this patch)
> expected?
The worst-case precision is massively improved - as I noted in the
original patch submission, the testcases added by the patch showed (on
x86_64) errors of up to 203 ulp for float, 1713 ulp for double and 12714
ulp for long double. It only appears as a decrease in precision because
the testsuite coverage for tgamma was previously so limited.
> Also, I'm now getting a bunch of missing underflow errors for casinh
> with "long double".
If those trace back to missing underflows from atanl / atan2l (if the part
of the result of casinh / casin that should underflow is the one computed
by atan2l), then that's another instance of bug 15319, in which case you
could change them to UNDERFLOW_EXCEPTION_OK and add a comment like the
existing instances of
/* Bug 15319: underflow exception may be missing. */
(if they trace to some other cause, a separate bug should be filed).
--
Joseph S. Myers
joseph@codesourcery.com