Skeletons that are primarily characterized by their topology.

Simple pipe where the parent process creates all pipe processes. The processes communicate their results via the caller process.

Process pipe where the processes communicate their Remote Data handles via the caller process but fetch the actual data from their predecessor processes

Simple ring skeleton (tutorial version) using remote data for providing direct inter-ring communication without input distribution and output combination

The ring establishes a ring topology, the ring process function transforms the initial input of a ring process and the input stream from the ring into the ring output stream and the ring processes final result. The same function is used by every ring process. Use ringFl if you need different functions in the processes. Use ringAt if explicit placement is desired.

The ringFl establishes a ring topology, the ring process functions transform the initial input of a ring process and the input stream from the ring into the ring output stream and the ring processes' final result. Every ring process applies an individual function which e.g. allows to route individual offline input into the ring processes. Use ringFlAt if explicit placement is desired.

Skeleton ringAt establishes a ring topology, the ring process function transforms the initial input of a ring process and the input stream from the ring into the ring output stream and the ring processes' final result. The same function is used by every ring process. Use ringFlAt if you need different functions in the processes. This version uses explicit placement.

The ringFlAt establishes a ring topology, the ring process functions transform the initial input of a ring process and the input stream from the ring into the ring output stream and the ring processes' final result. Every ring process applies its individual function which e.g. allows to route individual offline input into the ring processes. This version uses explicit placement.

Parallel torus skeleton (tutorial version) with stream rotation in 2 directions: initial inputs for each torus element are given. The node function is used on each torus element to transform the initial input and a stream of inputs from each direction to a stream of outputs to each direction. Each torus input should have the same size in both dimensions, otherwise the smaller input will determine the size of the torus.

The allToAllRD skeleton creates as many processes as elements in the input list (np). The processes are all-to-all connected, each process' input is transformed to np intermediate values by the first parameter function, where the i-th value will be sent to process i. The second transformation function combines the initial input and the np received intermediate values to the final output.

Parallel transposition for matrices which are row-wise round robin distributed among the machines, the transposed result matrix is also row-wise round robin distributed.

Performs an all-reduce with the reduce function using a butterfly scheme. The input list should have length 2^i, where i is an arbitrary natural number. If not, the input list will be truncated to the next smaller power of two. The initial transformation is applied in the processes to obtain the values that will be reduced. The final combine function is used to create a processes result from the initial input and the reduced value.