Fri Feb 22 22:17:38 CET 2013

Conclusions: transfer functions

- Generic gaussian elimination is a nice idea to keep in mind for
  later, but is probably not necessary if we take a shortcut for 1 and
  2 element feedback functions.

- It is probably possible to perform GE in a value-orgiented way, just
  by performing the algorithm at the "macro level".  The result is a
  computational network that still has parallelism (for low-level
  opti) and can be pruned by 1 and 0 ops.

- A mechanism is necessary to write "macro functions" in scheme.
  I.e. "DAG generators".  Things that operate on compile-time data