ifloat

Another simpler implementation for float only uses integer types, we should use -msoft-float first or we get complex result, so we need this with huge or tiny exponents float.