Re: multi-array comments/questions

From: rwgk (rwgk_at_[hidden])
Date: 2002-01-03 20:04:29

>   After a quick look at the code, I'm not seeing the full suite of
> operators, that libraries like MTL and others have. (Expression 
template
> stuff.) Is this because multi-array is designed for a different set 
of
> problems?

How wise would it be to define the full set of operators?
E.g., do you want them to work element-wise or as matrix
operations (a * b), or element-wise (a < b -> multiarray
of bool) or as reductions (a < b -> one bool)?

A few weeks ago I posted a proposal for a scheme that
gives the user the choice of selecting the desired
set of operators at run-time with a using directive.
The basic idea:

void foo()
{
  using namespace matrix_algebra;
  somearray a,b,c;
  c = a * b; // matrix multiplication
}

void bar()
{
  using namespace element_wise_algebra;
  somearray a,b,c;
  c = a * b; // element-wise multiplication
}

This works well with EDG based compilers, gcc, and
Metrowerks.

In this scheme it actually is a disadvantage if
operators are predefined in the class scope.

I believe that a tight integration of operators in multi_array
would be a limitation. A technology that delivers array management
as one building block and operations on these as other, separate
building blocks seems more reusable and therefore superior.
Otherwise we will need two or more versions of multi_array,
with different sets of operators.

An alternative to my proposed scheme that I could think of
is to use inheritance, e.g.:
    class matrix : public multi_array {
      // define the matrix-algebra operators here
    };
    class marray : public multi_array {
      // define the element-wise operators here
    };
But this could easily lead to a proliferation of derived
types (matrix algebra with < working as reduction,
matrix algebra with < working element-wise, etc.).

Ralf