Something is embarrassingly slow (with current message passing backend?)

[awintel]

@ashishrao7's quadratic programming process in lava-optim seems to be super slow compared to a pure numpy implementation. Since the only main difference between the two implementation seems to be the event-based message passing in between, we need to figure out why that is the case.

[ashishrao7]

To provide more context the process model for the solver is shown below:

The update equations (top right) running in numpy take ~20ms for 1000 iterations. The process models implementing the same update equations in LAVA when run on the same machine take about 100 seconds for 1000 iterations.

[awintel]

Please also check what is setup time and what is actual execution time.

[ashishrao7]

Adding the profiling results on some small QPs. Most of the time seems to be spent on some acquire method of a thread.lock object
Added scripts for problems of size
3x3 with 2 constraints => ports of size (3,1) and (2,1),
3x3QPprofiling.txt

4x4 with 3 constraints => ports of size (4,1) and (3,1)
4x4QPprofiling.txt

5x5 with 3 constraints => ports of size (5,1) and (3,1)
5x5QPprofiling.txt