Fast transpile with MixIn #1455
Conversation
| ... | ||
|
|
||
|
|
||
| class SimpleCircuitExtenderMixin: |
There was a problem hiding this comment.
I wonder if we want to make this a feature of BaseExperiment rather than a mixin. I am not sure what the variable should be called but experiments could set something like layout_and_translate_only = True and if BaseExperiment._transpiled_circuits could just do the code here in that case.
| ): | ||
| # In Qiskit provider model barrier is not included in target. | ||
| # Use standard circuit transpile when circuit is not ISA. | ||
| return transpile( |
There was a problem hiding this comment.
Don't you at least want to set optimization_level? Better than that, could we create a pass manager with just BasisTranslator here?
There was a problem hiding this comment.
Could we quickly check for custom pulse gates as well? I think that is the one other pass that users sometimes want for QE (though in the future that will probably go away).
| p_qregs = QuantumRegister(n_qubits) | ||
| v_p_map = {q: p_qregs[self.physical_qubits[i]] for i, q in enumerate(v_circ.qubits)} | ||
| p_circ = QuantumCircuit(p_qregs, *v_circ.cregs) | ||
| p_circ.metadata = v_circ.metadata | ||
| for inst, v_qubits, clbits in v_circ.data: | ||
| p_qubits = list(map(v_p_map.get, v_qubits)) | ||
| p_circ._append(inst, p_qubits, clbits) | ||
| return p_circ |
There was a problem hiding this comment.
Would the equivalent of this make sense in Qiskit as a SimpleLayout pass? I have not understood why the simplest layout in Qiskit has to be four stages and involve ancilla expansion like it does.
|
close this in favor of #1459 |
Summary
Call to Qiskit transpiler is overkill for simple experiment that directly creates ISA circuits, like T1 and T2 experiment. This PR adds a mixin class that automatically implements
_transpiled_circuits()method that only maps virtual qubit index to physical qubit index without invoking the computationally heavy transpiler.Details and comments