...gathering

The MPI call for gather is MPI_Gather.

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

...scatter

The MPI call for scatter is MPI_Scatter.

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

...collect

The MPI call for collect is MPI_Allgather.

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

...broadcast

The MPI call for broadcast is MPI_Bcast.

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

...template

Notice, our use of the word template should not be confused with the use of templates in C++.

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

...global_length

To keep from being confused: once projected, the (multi)vector should be viewed as a special case of the matrix, which exists entirely within the row(s) or column(s) that own the projected vector. Thus, global_length and global_width reflect the length and width of the (multi)vector viewed as a matrix.

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

...rank-k

Notice that the generic name for this operation is ``rank-k'' update. Since k is one of the dimensions of the matrices, we are really performing rank-s updates.

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

...processor

For the Cray T3D and Convex Exemplar, the term node is somewhat confusing. A node board of a Cray T3D contains two processors (processing elements or PEs). A Convex Exemplar node consists of sixteen processors in a shared memory configuration. Notice that in our terminology, one node equals one Cray T3D PE or one Convex Exemplar processor.

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