Re: [Boost-Users] [uBLAS] binary operation of vector and matrix with different type

From: jhr.walter_at_[hidden]
Date: 2003-05-12 02:07:12

Hi Zhang,

you wrote:

>    the following code generates a compile error raising from
>     a macro in /ublas/traits.hpp:
>    BOOST_STATIC_ASSERT ((boost::is_same<T1, T2>::value));
>
>    Because I'm trying to prod on a vector of size_t and a matrix of
unsigned
>    char.
>
>    Because I want to save memory (the matrix is really huge) and I
>    know the value in m will never exceed 255. The expression is ok to
>    me (of course the result value is size_t).
>    But the ASSERT prevent me from doing this.
>
>    Is there a workaround that permits me doing this kind of case without
>    sacrificing the memory efficiency?

I'd try to add specializations of the form

    template<>
    struct promote_traits<std::size_t, unsigned char> {
        typedef std::size_t promote_type;
    };
    template<>
    struct promote_traits<unsigned char, std::size_t> {
        typedef std::size_t promote_type;
    };

to the namespace boost::numeric::ublas. If there's enough interest we'll
probaby have to extend promote_traits systematically one day.

> Using size_t in the matrix will
>    use three times more memory.

Looks like you're working on an interesting problem :-)

>    ublas::vector<size_t> v1(5);
>    matrix<unsigned char> m(5,5);
>    prod(m,v1);
>
>    the same problem arises when I do
>    prod(m,v2); here v2 is a vector<double>

The same solution should apply here.

HTH,

Joerg