[Blitz-support] shrinking an array by taking the mean of every 2 numbers?

[Patrik J. <pa...@uc...>]

Hi all,

I need to take an array expression and shrink it by replacing every 2
(or 4, 8, etc) numbers with the mean of those numbers. Thus (1,2,3,4) ->
(1.5, 3.5). This happens often and must be as efficient as possible, so
I want to do it without temporaries, but I'm having some trouble
figuring out how to do this.

One way I thought was to make a function:

    template <typename T>
      <return type> shrink(const blitz::_bz_ArrayExpr<T> a) {
      return (a(2*blitz::tensor::i)+a(2*blitz::tensor::i+1)*0.5);
    };

so you can do

shrunken_array =3D shrink(<expr>)

but this doesn't work. It seems you can't use the tensor notation to
index with stride? I thought it would work because the indices are
evaluated on the left-hand side, so if that array is length N and the
expression is length 2N, then it should work out fine.

Another option would be to do something like:

    template <typename T>
      <return type> shrink(const blitz::_bz_ArrayExpr<T> a) {
	Range r(0, a.size()-2, 2);
      return (a(r)+a(r+1)*0.5);
    };

but that doesn't seem to work because _bz_ArrayExpr's don't have a
size() member, and even if they did, you don't seem to be able to apply
the Range to an expression.

Then I thought I would try to use a StridedDomain<1> instead of the
Range, but that fares no better.

Can anyone offer a suggestion how to accomplish this?

Thanks,

/Patrik