Information

• Qiskit Terra version: Master since: 9040bb9c
• Python version: Any
• Operating system: Any

What is the current behavior?

Since #5486 merged there has been a performance regression across the board with everything that operates on bits and/or registers in qiskit. For example:

https://qiskit.github.io/qiskit/#converters.ConverterBenchmarks.time_dag_to_circuit?machine=dedicated-benchmarking-softlayer-baremetal&os=Linux%204.15.0-46-generic&ram=16GB&p-n_qubits=8&p-depth=128&commits=9040bb9c

or

https://qiskit.github.io/qiskit/#transpiler_levels.TranspilerLevelBenchmarks.time_quantum_volume_transpile_50_x_20?machine=dedicated-benchmarking-softlayer-baremetal&os=Linux%204.15.0-46-generic&ram=16GB&p-transpiler%20optimization%20level=0&commits=9040bb9c

My guess is that this is around the changes made in Register.__eq__ in #5486 but haven't profiled it to confirm.

The hope is that the remaining patches in the optional registers patch series will address the performance impact: https://github.com/Qiskit/qiskit-terra/pulls?q=is%3Apr+is%3Aopen+label%3Aoptional-registers This issue is so we continue to track the regression as a release blocker to ensure we confirm the performance regression is addressed before we release with the code.