Pipeline and Sequence Modes

The two main modes of this simulation are pipeline and sequence, following the terminology of AFF3CT.

Sequence

A sequence consists of chaining together several AFF3CT tasks. A task is an operation performed on data (e.g., modulation) and belongs to a module, which can be seen as a class. For example, the modulation task belongs to the modulator module, which may also include other tasks such as demodulation. In a sequence, tasks are executed one after another.

Pipeline

A pipeline can be seen as a sequence split into multiple stages, where each stage contains several tasks. Once stage 0 finishes processing the first batch of data, it can immediately start processing the second batch while stage 1 handles the first one. This overlap enables parallel execution across stages. AFF3CT pipelines allow each stages to be executed in parallel by multiple workers.

How This Simulation Uses Both

In this simulation, we primarily use sequences as described above. However, pipelines are used in a more specific way.

Only the decoder task is parallelised (which represents itself a pipeline stage), as it is by far the time-consuming (80% to 99% of the total runtime, depending on the simulation parameters). By doing so, multiple batches of data can be processed simultaneously within the decoder, instead of being blocked behind it in a purely sequential execution.