... ad++¹

ad++ is currently implemented as a set of header files defining distributed arrays as type-secure container class templates, function templates for collective array operations, and macros for distributed control constructs.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... ranges²

Two ranges are considered orthogonal if they are distributed over different dimensions of the same processor arrangement. As a special case a collapsed, on-processor array range is orthogonal to any other.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... array³

Triplet subscripting of ranges works in the same was as triplet subscripting of arrays in Fortran. If lower or upper bounds are omitted they default to the bounds of the subscripted object, so Y(::2) is equivalent to Y(1:N:2)--a range including every 2nd element of Y.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... other⁴

As a rule operations like DA_CSHIFT and DA_REMAP cannot perform in-place updates. Their arguments should be distinct and non-overlapping.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.