Significant performance regression of transpilation in 2.1.0 compared to 2.0.2 #14653
Description
Environment
- Qiskit version: 2.1.0
- Python version: 3.12.11
- Operating system: macOS 15.5
What is happening?
Transpilation in Qiskit 2.1.0 requires 9 times more time compared to version 2.0.2.
How can we reproduce the issue?
from timeit import timeit
from qiskit import __version__, generate_preset_pass_manager
from qiskit.circuit.library import zz_feature_map
from qiskit_ibm_runtime.fake_provider import FakeFez
print(f"qiskit_version={__version__}")
qubits = range(4, 30, 4)
circuits = [zz_feature_map(i, reps=4) for i in qubits]
backend = FakeFez()
pm = generate_preset_pass_manager(backend=backend, optimization_level=3)
print(f"{timeit(lambda: pm.run(circuits), number=1)} sec")qiskit_version=2.1.0
34.34797391600023 sec
qiskit_version=2.0.3
3.7458205000002636 sec
What should happen?
Transpiler should be as fast as Qiskit 2.0.2
Any suggestions?
No idea.
Attach a frame graph just FYI