Increase heuristic effort for optimization level 2

[mtreinish]

Summary

This commit tweaks the heuristic effort in optimization level 2 to be more of a middle ground between level 1 and 3; with a better balance between output quality and runtime. This places it to be a better default for a pass manager we use if one isn't specified. The tradeoff here is that the vf2layout and vf2postlayout search space is reduced to be the same as level 1. There are diminishing margins of return on the vf2 layout search especially for cases when there are a large number of qubit permutations for the mapping found. Then the number of sabre trials is brought up to the same level as optimization level 3. As this can have a significant impact on output and the extra runtime cost is minimal. The larger change is that the optimization passes from level 3. This ends up mainly being 2q peephole optimization. With the performance improvements from #12010 and #11946 and all the follow-on PRs this is now fast enough to rely on in optimization level 2.

Details and comments

Related to: #7112

[qiskit-bot]

One or more of the the following people are requested to review this:

• @Qiskit/terra-core

[coveralls]

Pull Request Test Coverage Report for Build 8693815185

Details

• 6 of 6 (100.0%) changed or added relevant lines in 2 files are covered.
• 7 unchanged lines in 2 files lost coverage.
• Overall coverage decreased (-0.005%) to 89.354%

---

Files with Coverage Reduction	New Missed Lines	%
crates/qasm2/src/expr.rs	1	94.03%
crates/qasm2/src/lex.rs	6	92.11%

Totals
Change from base Build 8689749566:	-0.005%
Covered Lines:	60163
Relevant Lines:	67331

---

💛 - Coveralls

[mtreinish]

I ran the "utility scale" asv benchmarks with this PR and got the following results:

Benchmarks that have improved:

       before           after         ratio
     [e0be97c1]       [53dec8bd]
     <main>       <level-2-v2>
-      1.37±0.03s       1.22±0.01s     0.89  utility_scale.UtilityScaleBenchmarks.time_square_heisenberg('cx')
-            1896             1618     0.85  utility_scale.UtilityScaleBenchmarks.track_qaoa_depth('cz')
-            1908             1607     0.84  utility_scale.UtilityScaleBenchmarks.track_qaoa_depth('cx')
-            1936             1622     0.84  utility_scale.UtilityScaleBenchmarks.track_qaoa_depth('ecr')
-             972              444     0.46  utility_scale.UtilityScaleBenchmarks.track_square_heisenberg_depth('cx')
-             972              444     0.46  utility_scale.UtilityScaleBenchmarks.track_square_heisenberg_depth('cz')
-             972              444     0.46  utility_scale.UtilityScaleBenchmarks.track_square_heisenberg_depth('ecr')

Benchmarks that have stayed the same:

       before           after         ratio
     [e0be97c1]       [53dec8bd]
     <main>       <level-2-v2>
       21.1±0.06s       25.3±0.02s    ~1.20  utility_scale.UtilityScaleBenchmarks.time_qft('ecr')
          2.18±0s       2.32±0.01s     1.06  utility_scale.UtilityScaleBenchmarks.time_square_heisenberg('ecr')
       24.3±0.06s       25.7±0.02s     1.06  utility_scale.UtilityScaleBenchmarks.time_qft('cz')
       29.8±0.2ms       30.7±0.2ms     1.03  utility_scale.UtilityScaleBenchmarks.time_parse_square_heisenberg_n100('ecr')
       18.6±0.06s       19.1±0.06s     1.03  utility_scale.UtilityScaleBenchmarks.time_qft('cx')
       92.3±0.8ms       94.5±0.4ms     1.02  utility_scale.UtilityScaleBenchmarks.time_parse_qft_n100('cz')
      29.7±0.08ms       30.2±0.1ms     1.02  utility_scale.UtilityScaleBenchmarks.time_parse_square_heisenberg_n100('cx')
         92.9±2ms         94.2±1ms     1.01  utility_scale.UtilityScaleBenchmarks.time_parse_qft_n100('cx')
       92.7±0.7ms       94.0±0.3ms     1.01  utility_scale.UtilityScaleBenchmarks.time_parse_qft_n100('ecr')
      30.0±0.04ms       30.2±0.4ms     1.01  utility_scale.UtilityScaleBenchmarks.time_parse_square_heisenberg_n100('cz')
      8.56±0.04ms      8.60±0.04ms     1.00  utility_scale.UtilityScaleBenchmarks.time_parse_qaoa_n100('cz')
      8.64±0.08ms      8.65±0.02ms     1.00  utility_scale.UtilityScaleBenchmarks.time_parse_qaoa_n100('cx')
       8.59±0.1ms      8.60±0.04ms     1.00  utility_scale.UtilityScaleBenchmarks.time_parse_qaoa_n100('ecr')
             2598             2582     0.99  utility_scale.UtilityScaleBenchmarks.track_qft_depth('cx')
             2598             2582     0.99  utility_scale.UtilityScaleBenchmarks.track_qft_depth('cz')
             2598             2496     0.96  utility_scale.UtilityScaleBenchmarks.track_qft_depth('ecr')
       3.11±0.01s       2.69±0.03s    ~0.87  utility_scale.UtilityScaleBenchmarks.time_square_heisenberg('cz')

Benchmarks that have got worse:

       before           after         ratio
     [e0be97c1]       [53dec8bd]
     <main>       <level-2-v2>
+      2.78±0.04s          5.32±0s     1.92  utility_scale.UtilityScaleBenchmarks.time_qaoa('ecr')
+         852±8ms       1.08±0.01s     1.27  utility_scale.UtilityScaleBenchmarks.time_qaoa('cx')
+      5.12±0.01s          6.45±0s     1.26  utility_scale.UtilityScaleBenchmarks.time_qaoa('cz')

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.
PERFORMANCE DECREASED.

It's about what I expected. Improvements in quality since we're ramping up heurstic effort in a bunch of places but also at the cost of runtime. The ecr qaoa runtime benchark is a bit more severe than I expected, but the ~25% slower is about what I was expecting.

The only other change I really want to do is building off of #12171 I'd like to change the default sabre heuristic we use in level 2 to lookahead instead of decay. Not to save any runtime, I just expect it to produce better output.

I'm also curious to try dropping the sabre trial counts back down to 10 and see where the tradeoffs are there. I'll rerun the benchmarks and see what they look like with that change. The only place the runtime change can realistically come from is in sabre or by running 2q peephole. It didn't make any difference, if anything it was slower (because of more swaps inserted).

[levbishop]

I think there'll be more tweaks to O2 as we speed up and streamline other passes, add sabre heuristics, etc, but this is a solid starting point