Sat Aug 17 11:36:27 EDT 2019
Try out ABC: give it a LUT
So given RTL, represent the update functions as LUTs, push it into ABC
and see if that initial LUT form has an effect on what comes out.
What is the primitive here? An N->1 boolean function. Describing
such a thing requires 2^N bits.
But that isn't really the core representation, because it does not
allow for sharing. ABC likely needs to be presented with N->M boolean
functions, represented in some kind of logic gate format that does not
cause table explosion.
So I probably still want to hold on to some shared DAG just to not
explode the representation. Processing of such graphs is likely
iterative, i.e. it needs a seed of _some_ structured/shared
representation as AIG before it can do optimizations in AIG space.
Full LUT description seems too unstructured. Cfr Karnaugh maps: the
whole idea is to identify these islands == term pruning.