[PATCH 00/23] Add remaining CORE-MATH binary32 implementations to libm

[Florian Weimer]

* DJ Delorie:

> I'm currently[*] seeing ULP failures for atan, atan2, asin, and acos
> when passed PI-related constants, for example:
>
> FAIL: math/test-double-acospi
> FAIL: math/test-double-asinpi
> FAIL: math/test-double-atan2pi
> FAIL: math/test-double-atanpi
> FAIL: math/test-float-acospi
> FAIL: math/test-float-asinpi
> FAIL: math/test-float-atan2pi
> FAIL: math/test-float-atanpi
> . . .
>
> ... on s390, ppc64le, and i686.
>
> The results are all "saw 1 expected 0 ULPs".
>
> Could these be related to this patch set?
>
> [*] while syncing glibc to Fedora Rawhide, at least.  The i686 ci/cd
> trybot is not seeing ULP problems.

Yes, these functions are not expected to be correctly rounded, and as
they are newly added, many architectures do not have ulps bounds for
them yet.

Thanks,
Florian