Transpiler fails after 1000 iterations on simple circuit at optimization_level=3

[nonhermitian]

Information

• Qiskit Terra version: latest
• Python version:
• Operating system:

What is the current behavior?

The following circuit + backend + transpiler (optim level =3) combo:

qc = QuantumCircuit(3, 1) 
qc.reset(range(3))

qc.x(0)
qc.barrier()
qc.ccx(0, 1, 2) 
qc.barrier()
qc.measure(2, 0) 

backend = provider.get_backend('ibmq_athens')
qc_trans = transpile(qc, backend, initial_layout=[2,3,4], optimization_level=3)

fails with the error:

Traceback (most recent call last):
File "", line 10, in
qc_trans = transpile(qc, backend2, initial_layout=[2,3,4], optimization_level=3)
File "/opt/miniconda3/envs/qiskit/lib/python3.7/site-packages/qiskit/compiler/transpile.py", line 243, in transpile
circuits = parallel_map(_transpile_circuit, list(zip(circuits, transpile_args)))
File "/opt/miniconda3/envs/qiskit/lib/python3.7/site-packages/qiskit/tools/parallel.py", line 106, in parallel_map
return [task(values[0], *task_args, **task_kwargs)]
File "/opt/miniconda3/envs/qiskit/lib/python3.7/site-packages/qiskit/compiler/transpile.py", line 327, in _transpile_circuit
output_name=transpile_config['output_name'])
File "/opt/miniconda3/envs/qiskit/lib/python3.7/site-packages/qiskit/transpiler/passmanager.py", line 225, in run
return self._run_single_circuit(circuits, output_name, callback)
File "/opt/miniconda3/envs/qiskit/lib/python3.7/site-packages/qiskit/transpiler/passmanager.py", line 288, in run_single_circuit
result = running_passmanager.run(circuit, output_name=output_name, callback=callback)
File "/opt/miniconda3/envs/qiskit/lib/python3.7/site-packages/qiskit/transpiler/runningpassmanager.py", line 112, in run
for pass in passset:
File "/opt/miniconda3/envs/qiskit/lib/python3.7/site-packages/qiskit/transpiler/runningpassmanager.py", line 311, in iter
raise TranspilerError("Maximum iteration reached. max_iteration=%i" % self.max_iteration)
TranspilerError: 'Maximum iteration reached. max_iteration=1000'

It only fails for level 3.

Steps to reproduce the problem

What is the expected behavior?

Suggested solutions

[kdk]

I'm not currently able to replicate this on master. (A similar issue was recently fixed in #5583 .)

Can you define a callback like:

def callback(pass_, dag, time, property_set, count):
    from qiskit.transpiler import TransformationPass
    if isinstance(pass_, TransformationPass):
        print(pass_.name(), dag.depth(), list(dag.count_ops()))

re-run transpile(..., callback=callback) and post the log?

[nonhermitian]

I should point out it is also stochastic in it failure. More often than not though

callback.txt

[nonhermitian]

If it is fixed in Master than feel free to close. Can re-open if need be.

[kdk]

Thanks, this should fixed by #5583 and will be included in the next patch release. Feel free to re-open if this persists.

[ishapoval]

Hi! After updating qiskit-terra from 0.16.1 to 0.16.2 I started getting

TranspilerError: 'Maximum iteration reached. max_iteration=1000'

when transpiling my own circuit in optimization level 3. Level 2 passes through just fine. Also, hitting the max iteration seems to be device-dependent. E.g., transpiling onto ibmq_manhattan yields the error, but ibmq_16_melbourne does not.

[mtreinish]

@ishapoval can you open a new issue with a recreate scenario. 0.16.2 included a backport of the fix for this original issue (#5583 and it's backport #5585) so I suspect something else might be causing the failure for you, but I'll need to dig into it to be sure.

[ishapoval]

@mtreinish Thanks for the prompt reply! Unfortunately, I can't post my circuit at this point and I didn't manage to reproduce the problem with a simpler scenario. My circuit is parametrized, and while in 0.16.1 it could be transpiled for all my parameters, in 0.16.2 it doesn't transpile for some of them. I narrowed down the problem to the difference in RZZGate angle. Another observation is that if I dump that offending circuit to the QASM string and construct a new circuit back from that string then it can be successfully transpiled. I will open a new issue if I have enough details to reliably reproduce the case in a simple scenario.