Description
Catalyst and PennyLane has always interpreted wire labels differently.
For example, when a new label
qml.gate_or_measurement(wires=1000)
is encountered, core pennylane considers "1000" as a pure wire label, and interprets that as one new wire, with the label "1000".
However in catalyst we do not associate wires with arbitrary labels and require wires to be continuous integers from zero, and we would interpret this as "allocate new wires until we have 1001 wires, and act on wire[1000]".
There is a qml.map_wires transform that essentially allow users to map wire indices from one set of labels to another set of labels. This can be used as a workaround for users to still use arbitrary labels in the circuit itself, but have the "arbitrary label to [0,1,2,...]" map tell catalyst what to do.
However, this transform does not currently work with catalyst.
Although we indicated that this is one of the PL transforms without a catalyst pass counterpart (https://github.com/PennyLaneAI/catalyst/blob/main/frontend/catalyst/from_plxpr.py#L205), other implementations are certainly welcome.