Arrays

An Adlib array is a rectangular distributed array of any rank, whose elements are partioned or replicated across some set of processes. The layout of the array is described by an object of type DAD which contains

• The rank, R of the array (ie, its number of dimensions). R is greater than or equal to zero.
• The process group over which the elements of the array are distributed.
• A vector of R dimension map objects, describing the shape of the array, and the mapping of its index space into the dimensions of of the process group, and the memory strides in the locally-held segment of the array.

The definition of the array is completed by a local array, or a pointer to a base address in the memory of each process, where the elements are actually stored. The type of the array elements is determined by the type of the local array segment--the DAD itself is blind to element type.

Note that the Adlib kernel makes no intrinsic assumptions about the ordering of array elements in local memory. For example, when a new distributed array is created, the dimension map vectors can set up to describe the situation where the first or the last dimension is ``most rapidly varying'' in memory. Similarly, no assumption is made that array elements fill a contiguous region of memory. In Fortran 90 terms, a DAD record can describe an arbitrary regular section of some parent array.

• struct Map
  • Constructors and destructors
  • Methods
• struct DAD
  • Public data fields
  • Constructors
  • Methods
• Examples
• Array shape
• Alignment and value-alignment
• Accessibility and value-accessibility