Sabre swap mapper dominated by data copy #5197
Description
Information
- Qiskit Terra version: master
- Python version:
- Operating system:
What is the current behavior?
Running Sabre on a QV circuit over 20 qubits shows that data copying is the current bottleneck:
Abbreviated profile:
9631133 function calls (8921864 primitive calls) in 4.184 seconds
Ordered by: internal time
ncalls tottime percall cumtime percall filename:lineno(function)
497066/663 0.492 0.000 1.516 0.002 copy.py:132(deepcopy)
44375/860 0.312 0.000 1.499 0.002 copy.py:220(_deepcopy_tuple)
389939 0.232 0.000 0.326 0.000 layout.py:97(__getitem__)
191092 0.179 0.000 0.511 0.000 sabre_swap.py:326(<genexpr>)
55118/10740 0.160 0.000 1.537 0.000 copy.py:269(_reconstruct)
1312882 0.112 0.000 0.112 0.000 bit.py:73(__hash__)
1032203 0.107 0.000 0.108 0.000 {method 'get' of 'dict' objects}
44177/660 0.090 0.000 1.497 0.002 copy.py:237(_deepcopy_dict)