Qiskit / qiskit / 6672838606

87%

main: 88%

push

github

Reuse VF2 scoring views for all scoring (#11115) (#11135)

* Reuse VF2 scoring views for all scoring

As part of the VF2Layout and VF2PostLayout passes when there are a large
number of matches found we're spending an inordinate amount of time in
scoring rebuilding the same views over and over again of the interaction
graph for each scoring call. For example, in one test cProfile showed
that with #11112 when running transpile() on a 65 Bernstein Vazirani
circuit with a secret of all 1s for FakeSherbrooke with
optimization_level=3 we were calling vf2_utils.score_layout() 161,761
times which took a culmulative time of 14.33 secs. Of that time though
we spent 5.865 secs building the edge list view.

These views are fixed for a given interaction graph which doesn't change
during the duration of the run() method on these passes. To remove this
inefficiency this commit moves the construction of the views to the
beginning of the passes and just reuses them by reference for each
scoring call, avoiding the reconstruction overhead.

* Add EdgeList Rust pyclass to avoid repeated conversion

This commit adds a new pyclass written in rust that wraps a rust
Vec. Previously the scoring function also used an dict->IndexMap
conversion, but the mapping structure wasn't necessary and added
additional overhead, so it was converted to a list/Vec to speed up the
execution even further. By using this new pyclass as the input to the
rust scoring function we avoid converting the edge list from a list to
an Vec on each call which will reduce the overhead even further.

(cherry picked from commit a67fe87b4)

Co-authored-by: Matthew Treinish <mtreinish@kortar.org>

37 of 37 new or added lines in 4 files covered. (100.0%)

73853 of 85027 relevant lines covered (86.86%)

91326.68 hits per line