Fresh benchmarks on multiple layouts

[Misty-W]

Fixes #83.
Supersedes #120 .

• Includes recent updates to Qiskit, UCC, and the UCC benchmarks
• Uses Qiskit SabreLayout in UCC instead of the UCC custom pass of the same name, due to changes in Qiskit
• Fixes the order of mapping, routing, and basis translation in the PyTket benchmarking function

Result

Follow-up:
Expand testing of existing Qiskit mapping and routing passes from optimization level 3 and non-default passes

[Misty-W]

Best layout benchmarks to date, using Qiskit level 3 mapping and routing passes along with UCC defaults- gate counts are nearly on par.

For reference, recent benchmarks on all-to-all connectivity:

Also, latest data on compile time to show Qiskit runtime improvement:

[jordandsullivan]

@Misty-W Is this still a draft or ready to merge?

[Misty-W]

@Misty-W Is this still a draft or ready to merge?

I would wait to merge until we can show UCC is better on average wrt gate count across different layouts. I'm currently setting up a test of the mapomatic library to see if it provides a worthwhile improvement.

[jordandsullivan]

Okay cool. If that does not yield fruit, is there a reason we can't just use Qiskit's default mapping pass?

[Misty-W]

With Qiskit's default mapping pass, UCC is almost but not quite as good as Qiskit when averaging across the benchmark layouts (see plots in above comment), so we need a better mapping pass.

[jordandsullivan]

Since this is intended to fix #83 (i.e. create mapping benchmarks), I think it's okay if we merge in the code to generate these mapping benchmarks without merging the plots or adding them to our automated benchmark suite.

@Misty-W Since you are now moving on to work on improving these mapping passes, let's complete this one and begin track your efforts toward mapping improvements in this issue: #85

[jordandsullivan]

LGTM