Double-precision sqrt was removed as the Newton iteration can be completely derived from rsqrt. The alternative would be to use sqrt table values and a sqrt specific Newton iteration, but that itself requires one division per iteration which does not make sense to implement.
Other changes:
- More comments added
- Removed all
clangwarnings