Currently scalar operators expect actual scalars (e.g. Reals). However they should allow for size one (or maybe strictly rank-zero) tensor nodes coming from another computation.
E.g:
NodePtr<> x, y;
auto z = Sum(x) * y;
Here Sum(x) is a size one tensor (possibly rank 0) and it is meaningful to use it as a multiplication operand, but there is no such mechanism.
Currently scalar operators expect actual scalars (e.g.
Reals). However they should allow for size one (or maybe strictly rank-zero) tensor nodes coming from another computation.E.g:
NodePtr<> x, y; auto z = Sum(x) * y;Here
Sum(x)is a size one tensor (possibly rank 0) and it is meaningful to use it as a multiplication operand, but there is no such mechanism.