Re: [PATCH] math: Add math-use-builtins-fabs

[Adhemerval Zanella <adhemerval.zanella@linaro.org>]

On 04/04/2022 18:30, Joseph Myers wrote:
> On Mon, 4 Apr 2022, Adhemerval Zanella via Libc-alpha wrote:
> 
>> Both float and double are assumed to be supported (since generic
>> implementation uses it without a fallback), only long double and
>> float128 are defined.
> 
> The only case where built-in fabs (or copysign) may not work properly is 
> IBM long double in the soft-float case (see 
> sysdeps/powerpc/nofpu/Makefile).  I'd expect the default to be that it 
> works with an override for long double fabs in that one soft-float case.

I am adding a way to override float128 builtin support because I plan
to support building glibc with clang, which does not support inline
__builtin_fabsf128.  I change the patch to assume _Float128 support,
but set it on this patch makes the required further changes simpler.

> 
>> On ia64 it also fixes:
>>
>>   math/test-float64x-fabs
>>   math/test-ldouble-fabs
> 
> If this is fixing a bug that was user-visible in a release, there should 
> be a corresponding report in Bugzilla.
> 

Ack.