Excessive memory usage while transpiling at optimization level 2

[ArfatSalman]

Environment

• Qiskit Terra version: {'qiskit-terra': '0.23.2', 'qiskit-aer': '0.11.2', 'qiskit-ignis': None, 'qiskit-ibmq-provider': '0.20.0', 'qiskit': '0.41.0', 'qiskit-nature': None, 'qiskit-finance': None, 'qiskit-optimization': None, 'qiskit-machine-learning': None}
• Python version: Python 3.9.12
• Operating system: MacOS Ventura 13.1 (22C65)

What is happening?

While transpiling this circuit, the python process tried to allocate more than 70 GB of memory (as far as I could see in the Activity Monitor of MacOS). The process is eventually killed.

At any other optimization level (0, 1, or 3), the circuit behaves normally.

Not sure what the issue is. It seems like it is related to this issue #9197 which leads to a large memory allocation.

How can we reproduce the issue?

from qiskit import QuantumCircuit, ClassicalRegister, QuantumRegister
from qiskit.circuit.library.standard_gates import *


qr = QuantumRegister(7, name="qr")
cr = ClassicalRegister(7, name="cr")
qc = QuantumCircuit(qr, cr, name="qc")
qc.append(
    UGate(2.2050953212701585, 3.3615540885941595, 4.353999488160196),
    qargs=[qr[4]],
    cargs=[],
)
qc.append(C3XGate(), qargs=[qr[4], qr[1], qr[3], qr[5]], cargs=[])
qc.append(
    CUGate(
        1.8216381126900292, 1.1397633674319172, 5.883270079132554, 6.151280357960994
    ),
    qargs=[qr[2], qr[6]],
    cargs=[],
)
qc.append(RZGate(0.8251147604978816), qargs=[qr[1]], cargs=[])
qc.append(XGate(), qargs=[qr[1]], cargs=[])
qc.append(RZZGate(3.4573000463617367), qargs=[qr[5], qr[0]], cargs=[])
qc.append(CXGate(), qargs=[qr[6], qr[4]], cargs=[])
qc.append(RCCXGate(), qargs=[qr[1], qr[3], qr[4]], cargs=[])
qc.append(U2Gate(2.647545737192122, 3.1045763954315304), qargs=[qr[6]], cargs=[])
qc.append(ZGate(), qargs=[qr[1]], cargs=[])
qc.append(
    CU3Gate(1.3925947961767187, 3.8341186215245604, 3.461675537985181),
    qargs=[qr[5], qr[1]],
    cargs=[],
)
qc.append(ZGate(), qargs=[qr[6]], cargs=[])
qc.append(C3XGate(), qargs=[qr[3], qr[0], qr[4], qr[2]], cargs=[])
qc.append(RZXGate(6.045230636776243), qargs=[qr[4], qr[6]], cargs=[])
qc.append(TdgGate(), qargs=[qr[1]], cargs=[])
qc.append(RZZGate(1.627140919622475), qargs=[qr[4], qr[2]], cargs=[])
qc.append(C3XGate(), qargs=[qr[3], qr[5], qr[1], qr[2]], cargs=[])
qc.append(
    CU3Gate(1.4873061282104487, 6.193271645272907, 0.08920760414730318),
    qargs=[qr[5], qr[1]],
    cargs=[],
)
qc.append(IGate(), qargs=[qr[3]], cargs=[])
qc.append(HGate(), qargs=[qr[4]], cargs=[])
qc.append(C4XGate(), qargs=[qr[0], qr[2], qr[6], qr[5], qr[3]], cargs=[])
qc.append(ECRGate(), qargs=[qr[1], qr[0]], cargs=[])
qc.append(CXGate(), qargs=[qr[6], qr[2]], cargs=[])
qc.append(CXGate(), qargs=[qr[6], qr[1]], cargs=[])

qc.measure(qr, cr)

from qiskit import transpile

qc = transpile(qc, basis_gates=None, optimization_level=2, coupling_map=None)

What should happen?

The circuit should transpile correctly.

Any suggestions?

No response

[mtreinish]

Yeah I'm pretty sure this is a duplicate of #9197, it's only effecting you on optimization level 2 because at optimization level 3 the other passes run decompose the large multiqubit gates to 1 and 2 qubit gates (in order to do additional optimization passes). For lower optimization levels commutative cancellation isn't run. #9201 should fix this