Sat Dec 7 12:24:42 EST 2019

Transformation of parallel to serial?

Can it be that simple?

Start with actual "large" objects, then transform the data connections
into sequential ones, and refactor the processing in the same way.

It seems that the important bit is to decide where to make those
time/space tradeoffs.

It also seems simpler to go from parallel data to serial data.
I.e. you "add" stuff.

Going from parallel compute to serial compute is also straightforward,
but it seems backwards in some way because we all get so used to
sequential programs.  However the intuitive sequential programs might
hide some other sequentialization.

It seems better to make resource allocation very explicit: start with
pure data flow, then braid it into sequential machines.

EDIT: Actually this goes back to things I remember from very early on:
splitting a filter into a sequential program using a simpler datapath.