One way to implement a system capable of parallel computing is with a Kahn process network (KPN). Wikipedia states: “A Kahn process network (KPN, or process network) is a distributed model of computation in which a group of deterministic sequential processes communicate through unbounded first in, first out channels. The model requires that reading from a channel is blocking while writing is non-blocking. Due to these key restrictions, the resulting process network exhibits deterministic behavior that does not depend on the timing of computation nor on communication delays.”
Figure 5 shows an example of a KPN. Note that all the nodes T1 through T4 may run simultaneously as they are separate processes. The key contribution of Kahn’s proposal was defining when a node executes.
Figure 5: Example of a Kahn Process Network (KPN)
As unbounded first-in, first-out (FIFO) channels are physically unrealizable, the AIE array implements a modified KPN with bound channels. A well-designed system with balanced delays is still deterministic as possible stalls (caused by empty input buffers, full output buffers, or resource contention) always consume the same number of cycles.
A modified KPN described in an ADF graph encapsulates how AIE tiles (the “nodes” in the modified KPN) exchange data with the outside world. In an AI engine array, you implement the buffers (also called edges) as streams or shared memories. Note that a sequential process (also called a “node”) stalls or halts execution, when the following happens:
an input stream is empty
an output stream is full
another tile is accessing a shared memory bank
You can minimize stalls by allocating FIFOs of sufficient depth, or dedicating a memory bank to a kernel.
For details, refer to AI Engine Programming: A Kahn Process Network Evolution (WP552).