On Thu Mar 14 10:48:59 2024 +0000, Giovanni Mascellani wrote:
So both these two can be used interchangeably for both float and doubles?
Fma is fused multiply-add, with a single rounding, and Fmad is with two roundings, but it's unclear if the latter is only when marked precise. The proton fork uses Fma unless it's marked precise. We have no VSIR instruction for Fmad, but maybe that should be added now.