The Etrog pipeline is a much faster reimplementation of the Bergamot example pipeline, featuring a very simple core (see: pipelines/etrog/implementation/simplecore.py) which orchestrates all the operations of each instruction's execution; to wit:
- fetch PC from the register file
- fetch instruction bytes from main memory
- decode instruction from raw bytes
- fetch source operands from register file
- execute instruction
- write back destination register
- update PC
As with the Bergamot implementation, none of these operations overlap (meaning that the Etrog example, arguably, does not model a proper pipeline).
However, unlike Bergamot, which spawns the core, decoder, register file, and execute units, as well as the main memory inside their own individual processes, Etrog only spawns two processes: one for the core, one for main memory. The decoder, register file, and execute units are instantiated as objects inside the core; and messages are passed between these objects entirely inside the core. This significantly reduces the amount of interprocess communication, which significantly improves simulation speed.
On my development laptop, Etrog executed "examples/bin/sieve 20" at a rate of 24,000 simulated cycles per real-world second. Compare this to Bergamot, which executed the same benchmark at a rate of only 3,200 simulated cycles per real-world second.