Option for skipping OptimizeSwapBeforeMeasure when not all the wires are measured.

[1ucian0]

Summary

This PR extends OptimizeSwapBeforeMeasure to only remove swaps when both wires are measured. Instead of changing the default behaviour, I added a parameter for backwards compatibility (and in case we want to change it back for some transpilation cases). When OptimizeSwapBeforeMeasure is used in level 3, the option is activated. This fixes #4911

Details and comments

When the SWAP should not be remove

circ = QuantumCircuit(2, 1)
circ.swap(0, 1)
circ.measure(0, 0)

display(circ.draw())
display(OptimizeSwapBeforeMeasure(all_measurement=True)(circ).draw())

        ┌─┐
q_0: ─X─┤M├
      │ └╥┘
q_1: ─X──╫─
         ║ 
c: 1/════╩═
         0 
        ┌─┐
q_0: ─X─┤M├
      │ └╥┘
q_1: ─X──╫─
         ║ 
c: 1/════╩═
         0 

When the SWAP can be removed:

from qiskit import QuantumCircuit
from qiskit.transpiler.passes import OptimizeSwapBeforeMeasure

circ = QuantumCircuit(2, 2)
circ.swap(0, 1)
circ.measure(range(2), range(2))

display(circ.draw())
display(OptimizeSwapBeforeMeasure(all_measurement=True)(circ).draw())

        ┌─┐   
q_0: ─X─┤M├───
      │ └╥┘┌─┐
q_1: ─X──╫─┤M├
         ║ └╥┘
c: 2/════╩══╩═
         0  1 
        ┌─┐
q_0: ───┤M├
     ┌─┐└╥┘
q_1: ┤M├─╫─
     └╥┘ ║ 
c: 2/═╩══╩═
      0  1 

[kdk]

Code changes look good. One question generally. Would it be more inline with the direction of #4608 to have this pass switch the order of the swap and measure instructions, instead of removing the swap? That would preserve the the layout in all cases and still allow the optimization in cases where only some of the qubits are measured.

[1ucian0]

#4608 transpiles a circuit without measurement and then measures. Therefore, this PR already solves #4608 since this pass has no effect if no measurement is found.

As swapping the swap and the measure, I have no strong opinion (maybe @ajavadia?). Probably makes sense when intermediate measurements.

[1ucian0]

For example:

from qiskit import QuantumCircuit
from qiskit.transpiler.passes import OptimizeSwapBeforeMeasure

circ = QuantumCircuit(2, 2)
circ.h(0)
circ.x(0)
circ.swap(0, 1)
circ.measure(range(2), range(2))
circ.h(0)
circ.x(0)
circ.swap(0, 1)
circ.measure(range(2), range(2))

print(circ)
print(OptimizeSwapBeforeMeasure(all_measurement=True, move_swap=True)(circ))

     ┌───┐┌───┐   ┌─┐┌───┐┌───┐   ┌─┐
q_0: ┤ H ├┤ X ├─X─┤M├┤ H ├┤ X ├─X─┤M├───
     └───┘└───┘ │ └╥┘└┬─┬┘└───┘ │ └╥┘┌─┐
q_1: ───────────X──╫──┤M├───────X──╫─┤M├
                   ║  └╥┘          ║ └╥┘
c: 2/══════════════╩═══╩═══════════╩══╩═
                   0   1           0  1
     ┌───┐┌───┐┌───┐┌───┐┌─┐   ┌─┐
q_0: ┤ H ├┤ X ├┤ H ├┤ X ├┤M├─X─┤M├────X─
     └┬─┬┘└───┘└───┘└───┘└╥┘ │ └╥┘┌─┐ │
q_1: ─┤M├─────────────────╫──X──╫─┤M├─X─
      └╥┘                 ║     ║ └╥┘
c: 2/══╩══════════════════╩═════╩══╩════
       0                  1     1  0

Does this make sense?

[kdk]

I had been thinking something even simpler, I think. For the circuit above, I'd expect an output (after one iteration) like

     ┌───┐┌───┐┌─┐   ┌───┐┌───┐┌─┐   
q_0: ┤ H ├┤ X ├┤M├─X─┤ H ├┤ X ├┤M├─X─
     └┬─┬┘└───┘└╥┘ │ └┬─┬┘└───┘└╥┘ │ 
q_1: ─┤M├───────╫──X──┤M├───────╫──X─
      └╥┘       ║     └╥┘       ║    
c: 2/══╩════════╩══════╩════════╩════
       0        1      0        1    

But looking at the behavior from master, I wonder if there isn't already a bug here with mid-circuit measurements.

     ┌───┐┌───┐┌───┐┌───┐┌─┐┌─┐
q_0: ┤ H ├┤ X ├┤ H ├┤ X ├┤M├┤M├
     └┬─┬┘└┬─┬┘└───┘└───┘└╥┘└╥┘
q_1: ─┤M├──┤M├────────────╫──╫─
      └╥┘  └╥┘            ║  ║ 
c: 2/══╩════╩═════════════╩══╩═
       0    0             1  1 

doesn't look right to me. I'd expect (regardless of how the optimization is done) for there to be at least one measurement between the first H-X and second H-X.

[1ucian0]

Good catch. Reported in #5905, fixed in #5906

[ajavadia]

I don't understand what all_measurement does. Can you explain?

[jakelishman]

This appears stale to me now with all its conflicts, and no longer really as in line with how we treat layouts, routings, and the responsibilities of the preset pass managers. I'll close it as staled, but feel free to re-open if there's more to do here.