nir: handle double-precision in fabs, frsq and fsqrt

Reviewed-by: Matt Turner <mattst88@gmail.com>
1 file changed