[PATCH 2/5] math: Remove ldbl-96 fma implementation

[Wilco Dijkstra]

Hi Adhemerval,

> It is worse than the ldbl-64 version on recent x86 hardware.  With
> Zen3 and gcc-15:
>
> ldbl-96 removal
> reciprocal-throughput        master        patched   improvement
> x86_64                    1176.2200       289.4640         4.06x
> i686                      1476.0600       636.8660         2.32x
>
> latency                      master        patched   improvement
> x86_64                    1176.2200        293.7360        4.00x
> i686                      1480.0700        658.4160        2.25x

LGTM - not sure why this was ever accepted given performance is that bad...

Reviewed-by: Wilco Dijkstra  <Wilco.Dijkstra@arm.com>

Cheers,
Wilco