feat: handle IfElseOp in qiskit_to_tk
#437
Conversation
| ) -> Circuit: | ||
| if_box, else_box = _pytket_boxes_from_IfElseOp(if_else_op, qregs, cregs) | ||
| circ_builder = CircuitBuilder(qregs, cregs) | ||
| circ = circ_builder.circuit() |
There was a problem hiding this comment.
print(circ.qubits, circ.bits)
print(if_box.get_circuit().qubits, if_box.get_circuit().bits)
print(qubits)
print(bits)
prints
[q105[0], q105[1]] [c3[0], c3[1]]
[q105[0], q105[1]] [c3[0], c3[1]]
[q105[1]]
[c3[0]]
So the box has 2 qubits and 2 bits, but in add_circbox, args=qubits + bits only provides 1 qubit and 1 bit.
There was a problem hiding this comment.
Thanks.
I figured out what I was doing wrong. The thing that was confusing me was that the two circuits which define the IfElseOp do not have any registers. Their qubits are tied to the register of the circuit containing the IfElseOp. This meant that the subcircuits in my CircBoxes had too many qubits/bits as there were some additional blank wires.
Cleaning up the blank wires with Circuit.remove_blank_wires as in d65a347 seems to fix this.
|
The test case I added is now working after 52458ab. I'll need to add more diverse test cases before I'm confident in merging this as well as negate the "else" condition properly and see what the story is for more complex conditions. |
CalMacCQ
changed the title
IfElseOp in qiskit_to_tkIfElseOp in qiskit_to_tk
CalMacCQ
added
enhancement
| # Examples -> https://docs.quantum.ibm.com/guides/classical-feedforward-and-control-flow | ||
| def _pytket_boxes_from_ifelseop( | ||
| if_else_op: IfElseOp, qregs: list[QuantumRegister], cregs: list[ClassicalRegister] | ||
| ) -> tuple[CircBox, Optional[CircBox]]: |
There was a problem hiding this comment.
I'm not so pleased with this Optional return type in the function signature. This is to deal with the case when the IfElseOp only has a true_body rather than a true_body and a false_body.
There's probably a cleaner way to deal with this.
| circuit.measure(q1, c1) | ||
|
|
||
| tkc = qiskit_to_tk(circuit) | ||
| assert tkc.n_gates_of_type(OpType.Conditional) == 2 |
There was a problem hiding this comment.
Should test this more thoroughly. I ran into this bug -> CQCL/tket#1726 when trying to validate the contents of the two CircBoxes.
| args=qubits, | ||
| condition_bits=bits, | ||
| # TODO: handle conditions over multiple bits/registers? | ||
| condition_value=not bool(if_else_op.condition[1]), |
There was a problem hiding this comment.
Still need to decide if its worth handling more complex conditionals here or just restrict to conditions over a single bit for now.
Description
Converting a qiskit IfElseOp as two conditional pytket
CircBoxes. This sort of control flow is probably better expressed in the HUGR data structure than as a pytketCircuit/CircBox.Needs more diverse test cases before merging and we also need to decide if we want to handle more complex conditions.
Looking at the qiskit docs they also have switch-case as well as for and while loops which we may want to support in the future.
https://docs.quantum.ibm.com/guides/classical-feedforward-and-control-flow
Example
Now the corresponding pytket circuit looks like
The$H$ and $X$ gate respectively.
IfandElseboxes contain a singleI thought about putting string representing the conditon inside the "If" label however this could be awkward. Hopefully its clear from the diagram whats going on.
This conversion ended up being slighly tricker than I expected because the qiskit circuits which define the
true_bodyandfalse_bodydo not have registers. Managed to get around this by using the register from the circuit containing theIfElseOpand cleaning up empty wires withCircuit.remove_blank_wiresto ensure theCircBoxis of the correct size.Related issues
closes #415, #366
Checklist