The code looks good to me. However, having looked closer at the code there are some points I particularly don't like. E.g. nint on 0.5, the facpm as you mention.
In general these routines could do with a careful re-coding. I would e.g. be cautious on non-IEEE standard optimizations and this code? Does it actually do the right thing?
The code looks good to me. However, having looked closer at the code there are some points I particularly don't like. E.g. nint on 0.5, the facpm as you mention.
In general these routines could do with a careful re-coding. I would e.g. be cautious on non-IEEE standard optimizations and this code? Does it actually do the right thing?