[PATCH] add inputs to auto-libm-test-in yielding larger errors
Joseph Myers
joseph@codesourcery.com
Fri Nov 27 19:13:04 GMT 2020
On Thu, 26 Nov 2020, Paul Zimmermann wrote:
> thank you Joseph, I will commit that patch.
>
> I also found a case where lgamma() yields more than 10 ulps of error:
>
> Checking lgamma
> x=-0x1.f60c969a239f2p+1 d=1.0410549778851831e+01
>
> This should be considered as a bug and entered into bugzilla?
Any case where a libm function produces errors exceeding the maximum
accepted in ulps files (9 ulps for most formats, 16 for IBM long double),
in any rounding mode, measured in the way the libm testsuite does
(comparing against a correctly rounded result for that rounding mode, not
against an infinite-precision result) should be reported as a bug to
Bugzilla if there isn't already a bug open for inaccuracy of that
function.
--
Joseph S. Myers
joseph@codesourcery.com
More information about the Libc-alpha
mailing list