Ensure translation stage always outputs ISA circuits

[jakelishman]

Summary

This shifts the responsibility of the translation stage slightly to always output ISA circuits. It previously was not 100% required that they did this, at least implicitly because Qiskit's built-in plugins didn't always respect 2q direction.

The builtin translation plugins now always respect 2q direction, which removes the need for the pre_optimization implicit stage, freeing it up for better user customisation.

This (in theory) shouldn't have runtime impacts because the optimisation loop was already having to do this afterwards anyway. For potential plugins that do respect gate direction (like the upcoming BasisConstructor), it can be a speedup in the default pipeline, since they won't need to run the gate-direction check any more.

Details and comments

Built on top of #13620, since it will need to change documentation in that PR.

Fix #13787.

[qiskit-bot]

One or more of the following people are relevant to this code:

• @Qiskit/terra-core

[coveralls]

Pull Request Test Coverage Report for Build 13168271554

Warning: This coverage report may be inaccurate.

This pull request's base commit is no longer the HEAD commit of its target branch. This means it includes changes from outside the original pull request, including, potentially, unrelated coverage changes.

• For more information on this, see Tracking coverage changes with pull request builds.
• To avoid this issue with future PRs, see these Recommended CI Configurations.
• For a quick fix, rebase this PR at GitHub. Your next report should be accurate.

Details

• 10 of 10 (100.0%) changed or added relevant lines in 1 file are covered.
• 15 unchanged lines in 3 files lost coverage.
• Overall coverage decreased (-0.005%) to 88.646%

---

Files with Coverage Reduction	New Missed Lines	%
crates/accelerate/src/euler_one_qubit_decomposer.rs	1	91.5%
crates/qasm2/src/lex.rs	6	92.23%
qiskit/transpiler/preset_passmanagers/common.py	8	85.94%

Totals
Change from base Build 13165103262:	-0.005%
Covered Lines:	79222
Relevant Lines:	89369

---

💛 - Coveralls

[jakelishman]

Now rebased over main. I'll run the benchmark suite overnight.

[jakelishman]

I ran the full benchmark suite, and there (as mostly expected) doesn't seem to be much change from this. There's a small positive improvement in a couple of places of runtime, though I'm not convinced it wasn't just fluctuations on my machine. There's depth improvements in QUEKO, perhaps because in cases where the optimisation loop is entered twice, there's a bit more information available to it? Or maybe it's just because there's the knock-on effect of some changes to the randomisation.

| Change   | Before [4d2265f6] <dag-module^2>   | After [92f30cdf] <translation-stage-direction>   |   Ratio | Benchmark (Parameter)                                                    |
|----------|------------------------------------|--------------------------------------------------|---------|--------------------------------------------------------------------------|
| -        | 1.24±0.04ms                        | 1.07±0.04ms                                      |    0.86 | quantum_info.CliffordDecomposeBench.time_decompose('5,10')               |
| -        | 4.53±0.4ms                         | 3.33±0.07ms                                      |    0.73 | quantum_info.SparsePauliOpBench.time_tensor(150, 100)                    |
| -        | 18.0±0.4ms                         | 13.5±0.3ms                                       |    0.75 | queko.QUEKOTranspilerBench.time_transpile_bigd(3, None)                  |
| -        | 81.9±1ms                           | 61.9±1ms                                         |    0.76 | queko.QUEKOTranspilerBench.time_transpile_bntf(0, 'sabre')               |
| -        | 138±6ms                            | 88.2±1ms                                         |    0.64 | queko.QUEKOTranspilerBench.time_transpile_bntf(0, None)                  |
| -        | 114±2ms                            | 99.8±1ms                                         |    0.87 | queko.QUEKOTranspilerBench.time_transpile_bntf(2, 'sabre')               |
| -        | 165±0.7ms                          | 123±2ms                                          |    0.74 | queko.QUEKOTranspilerBench.time_transpile_bntf(3, 'sabre')               |
| -        | 74                                 | 43                                               |    0.58 | queko.QUEKOTranspilerBench.track_depth_bigd_optimal_depth_45(2, 'sabre') |
| -        | 74                                 | 45                                               |    0.61 | queko.QUEKOTranspilerBench.track_depth_bigd_optimal_depth_45(2, None)    |
| -        | 74                                 | 43                                               |    0.58 | queko.QUEKOTranspilerBench.track_depth_bigd_optimal_depth_45(3, 'sabre') |
| -        | 74                                 | 45                                               |    0.61 | queko.QUEKOTranspilerBench.track_depth_bigd_optimal_depth_45(3, None)    |
| -        | 255                                | 216                                              |    0.85 | queko.QUEKOTranspilerBench.track_depth_bntf_optimal_depth_25(2, 'sabre') |
| -        | 45                                 | 30                                               |    0.67 | queko.QUEKOTranspilerBench.track_depth_bntf_optimal_depth_25(2, None)    |
| -        | 45                                 | 30                                               |    0.67 | queko.QUEKOTranspilerBench.track_depth_bntf_optimal_depth_25(3, None)    |
| -        | 28.7±1ms                           | 22.5±0.7ms                                       |    0.79 | random_circuit_hex.BenchRandomCircuitHex.time_ibmq_backend_transpile(10) |
| -        | 39.3±2ms                           | 28.7±1ms                                         |    0.73 | random_circuit_hex.BenchRandomCircuitHex.time_ibmq_backend_transpile(12) |
| -        | 20.4±0.8ms                         | 17.3±0.6ms                                       |    0.85 | random_circuit_hex.BenchRandomCircuitHex.time_ibmq_backend_transpile(8)  |

[ElePT]

I am absolutely in favor of this change, I think it's much cleaner and flexible, and the changes look good! I just left a question for my own understanding of the passes.

[ElePT]

Just to make sure I understood this sentence, is it GateDirection or CheckGateDirection that doesn't guarantee that 1q gates are ISA?

[jakelishman]

It's GateDirection - that's the one that actually inserts gates. CheckGateDirection just says whether or not we need to run GateDirection and the subsequent fixer.

[ElePT]

LGTM!