Simulator can run indefinately when running circuits with conditionals

[IlanIwumbwe]

Benny and I found this bug.

Informations

• Qiskit Aer version: 0.15.0
• Python version: 3.11.6
• Operating system: Ubuntu 23.01 LTS

What is the current behavior?

Runs indefinately

Steps to reproduce the problem

Run this code:

from qiskit import QuantumCircuit, ClassicalRegister
from helpers.qiskit_helpers import compare_statevectors, run_on_simulator, run_routing_simulation

subcirc1 = QuantumCircuit(0)
# subcirc1.x(0)

main_circ = QuantumCircuit(1)
# Adding creg resources 
creg_0 = ClassicalRegister(1)
main_circ.add_register(creg_0)


main_circ.measure(0, creg_0[0])
main_circ.x(0)
with main_circ.if_test((creg_0[0],0)) as else_1:
    main_circ.append(subcirc1)
with else_1:
    main_circ.append(subcirc1)

main_circ.measure_active()
print(main_circ)
run_on_simulator(main_circ, 1)

Making the subroutine non-empty makes it work

from qiskit import QuantumCircuit, ClassicalRegister
from helpers.qiskit_helpers import compare_statevectors, run_on_simulator, run_routing_simulation

subcirc1 = QuantumCircuit(1)
subcirc1.x(0)

main_circ = QuantumCircuit(1)
# Adding creg resources 
creg_0 = ClassicalRegister(1)
main_circ.add_register(creg_0)


main_circ.measure(0, creg_0[0])
main_circ.x(0)
with main_circ.if_test((creg_0[0],0)) as else_1:
    main_circ.append(subcirc1, [0])
with else_1:
    main_circ.append(subcirc1, [0])

main_circ.measure_active()
print(main_circ)
run_on_simulator(main_circ, 1)

Moving the x gate before the measure makes it work

from qiskit import QuantumCircuit, ClassicalRegister
from helpers.qiskit_helpers import compare_statevectors, run_on_simulator, run_routing_simulation

subcirc1 = QuantumCircuit(0)

main_circ = QuantumCircuit(1)
# Adding creg resources 
creg_0 = ClassicalRegister(1)
main_circ.add_register(creg_0)

main_circ.x(0)
main_circ.measure(0, creg_0[0])

with main_circ.if_test((creg_0[0],0)) as else_1:
    main_circ.append(subcirc1)
with else_1:
    main_circ.append(subcirc1)

main_circ.measure_active()
print(main_circ)
run_on_simulator(main_circ, 1)

What is the expected behavior?

Should return results, in the case of empty subroutine, should always treat it as identity

Suggested solutions

[gadial]

I was unable to run your code (I'm missing your qiskit helpers and getting qiskit_aer.aererror.AerError: 'unknown instruction: circuit-166' when trying to replicate on my own) but I did manage to reproduce the infinite loop with

from qiskit import ClassicalRegister
backend = self.backend(method="statevector", device="CPU")
backend.set_options(max_parallel_experiments=0)
main_circ = QuantumCircuit(1)
creg_0 = ClassicalRegister(1)
main_circ.add_register(creg_0)
main_circ.measure(0, creg_0[0])
main_circ.x(0)
with main_circ.if_test((creg_0[0],0)) as else_1:
    pass
with else_1:
    pass

main_circ.measure_active()
result = backend.run(main_circ, shots=1).result()
print(result)

It seems to me that it's not an Aer problem but a transpiler problem. Aer attempts to run the transpiler, and it messes up the circuit, moving jump instructions to the beginning. This results from the conversion to DAG and back, not anything Aer-specific. I've forwarded the issue to the Qiskit crew.

[jakelishman]

Gadi: the "jump" instructions are inserted by Aer itself as part of how it makes the tree-like structure of control-flow into a sequence of instructions. If the jumps and marks are getting set wrong, that'll be happening within Aer.

[gadial]

Thanks Jake. It's now clearer to me what's going on: Aer generates the jump command without assigning it to any qubit (which does not happen in the case of non-empty branches). This leads to the reordering of the instructions when converting back from DAG to circuit. So it's indeed a problem with the way Aer uses the transpiler.

[IlanIwumbwe]

I was unable to run your code

Sorry about that, that was my mistake. I forgot to add the helper functions we were using into the issue.