This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH]: Update auto-libm-test-out for catan / catanh.
Joseph Myers <joseph@codesourcery.com> writes:
> On Thu, 9 Mar 2017, Tulio Magno Quites Machado Filho wrote:
>
>> > I've cc'ed Tulio so he can regenerate them for power.
>>
>> That isn't necessary powerpc.
>
> To be clear, the point is to *remove* existing catan / catanh ulps before
> regenerating (with the fixed auto-libm-test-out-*), so that the
> regeneration can reduce ulps values there (existing ulps values based on
> buggy expectations) rather than just increasing them as would be the case
> when regen-ulps is used without existing ulps being removed first.
I removed all the *ldouble entries and run the tests on ppc, ppc64 and ppc64le.
There are no ULP changes.
However, binary128 ULPs from branch tuliom/float128 indeed decreased.
I'm updating them.
>> For the record, after applying this patch, I still see 64 errors which aren't
>> appearing in glibc 2.25.
>
> It seems plausible the few auto-libm-test-in entries generating those
> errors should be xfail-rounding:ibm128-libgcc.
Ack. I'll prepare a patch and take a look if the logb failures are also
fixed by the libgcc patch.
Thanks!
--
Tulio Magno