Calls to qml.ctrl(qml.QubitUnitary)) produces two QubitUnitaries

[erick-xanadu]

Code:

import pennylane as qml

import catalyst

dev = qml.device("lightning.qubit", wires=2)

matrix = qml.matrix(qml.Hadamard(wires=[0]))

@catalyst.qjit(keep_intermediate=True)
@qml.qnode(dev)
def foo():
    catalyst.ctrl(qml.QubitUnitary(matrix, wires=[0]), control=[1])
    return catalyst.measure(0)

We produce the following MLIR, notice that we have the Hadamard, and two QubitUnitaries.

      %out_qubits = quantum.unitary(%2 : tensor<2x2xcomplex<f64>>) %1 : !quantum.bit
      // ... snip ...
      %out_qubits_7, %out_ctrl_qubits = quantum.unitary(%4 : tensor<2x2xcomplex<f64>>) %out_qubits ctrls(%3) ctrlvals(%extracted_6) : !quantum.bit ctrls !quantum.bit

possibly related to #1483 and PennyLaneAI/pennylane#6839 . When running a different commit of Catalyst with PennyLane==0.40 we only have one single qubit unitary. Regarding the Hadamard, it may be that we have never had support for qml.matrix(gate) inside the qjit context. Having a gate would ensure it is traced even if it is used only for obtaining a matrix.