[PATCH] math: Add optimization barrier to ensure a1 + u.d is not reused [BZ #30664]

[Joseph Myers]

On Wed, 26 Feb 2025, Adhemerval Zanella Netto wrote:

> Last time I checked a integer based implementation (musl), it was faster
> than glibc one even on recent hardware.  But I don't think this is pressing
> issue since now most hardware does have fma instructions.

We have an implementation based on soft-fp (in sysdeps/ieee754/soft-fp/).  
That does require a suitable sfp-machine.h, including the parts 
integrating with hardware exceptions and rounding modes in the case of 
architectures supporting those.

-- 
Joseph S. Myers
josmyers@redhat.com