Re: [boost] [MPL.Math] compile-time double macro and printing

From: Peder Holt (peder.holt_at_[hidden])
Date: 2005-12-21 17:38:07

On 12/19/05, Chris Weed <chrisweed_at_[hidden]> wrote:
> Hi,
> I am trying to simply print a compile-time double using gcc 4.0.0 on Sun.
> The following prints out : 1.9036e+185
>
> #include <boost/metamath/double.hpp>
> #include <boost/metamath/double_macros.hpp>
> #include <iostream>
>
> int main()
> {
>   typedef BOOST_DOUBLE(3.3) D;
>   D d;
>   std::cout << (double)d << std::endl;
>   return 0;
> }
>
> Am I doing something wrong?
> Chris

This has probably got something to do with the way I translate a
mpl::double_ to a double. I use a union to represent the mapping
between a double and the mantissa,exponent and sign (represented as
integers) This mapping is most likely different for different systems,
so we need to either device a better scheme for converting from
compile-time to runtime double, or use different mappings for
different systems.

The relevant code is found in:
boost\mpl\math\double_\aux_\get_value.hpp

    struct encoding
    {
        unsigned long mantissa2;
        unsigned long mantissa1:20;
        unsigned long exp:11;
        unsigned long sign:1;
    };

    union converter_
    {
        double value;
        encoding encode;
    };

    template <typename T>
    double get_value(const T& arg)
    {
        converter_ converter;
        converter.encode.sign = T::sign;
        converter.encode.exp = T::exponent + 1023;
        converter.encode.mantissa1 = T::mantissa::part1 >> 10;
        converter.encode.mantissa2 =
            (T::mantissa::part1 << 22)
          | (T::mantissa::part2 >> 9);
        return converter.value;
    }

Try fiddling with the order of the elements in the encoding struct, e.g.
    struct encoding
    {
        unsigned long mantissa1:20;
        unsigned long exp:11;
        unsigned long sign:1;
        //Move this to the end
        unsigned long mantissa2;
    };

On the other hand, this encoding is probably documented somewhere...

Regards,
Peder

>
> _______________________________________________
> Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
>