Convert single-qubit gates to single-gate PhasedXZ for sycamore #2796
Conversation
| mat: np.ndarray, | ||
| atol: float = 0, | ||
| ) -> Optional[ops.SingleQubitGate]: | ||
| """Implements a single-qubit operation with a PhasedXZ gate. |
There was a problem hiding this comment.
Should we add information about the way this is computed?
cirq/optimizers/decompositions.py
Outdated
| return None | ||
|
|
||
| # Special case: XY half-turns can absorb Z rotations. | ||
| if abs(xy_turn) >= 0.5 - atol: |
There was a problem hiding this comment.
Is this correct? Do you mean xy_turn more than 0.5 or xy_turn within atol of 0.5?
(or does xy_turn max out at 0.5?)
There was a problem hiding this comment.
xy_turn is in the range [-0.5, 0.5] thanks to the _signed_mod_1 function, so this is equivalent to checking if abs(xy_turn) is close to 0.5 (it's also done this way in the existing function single_qubit_matrix_to_phased_x_z above. But I agree this is a bit confusing; will change to use math.isclose.
| ) -> None: | ||
| """Canonicalizes runs of single-qubit rotations in a circuit. | ||
|
|
||
| Specifically, any run of non-parameterized circuits will be replaced by an |
There was a problem hiding this comment.
Would "any run of non-parameterized single-qubit operations" be more precise?
There was a problem hiding this comment.
That's definitely better; fixed.
| gate = decompositions.single_qubit_matrix_to_phxz(matrix, atol) | ||
| return [gate(qubit)] if gate else [] | ||
|
|
||
| MergeSingleQubitGates(synthesizer=synth).optimize_circuit(circuit) |
There was a problem hiding this comment.
Is this tested anywhere? If so, I missed it.
There was a problem hiding this comment.
It was being tested indirectly since it gets used in optimized_for_sycamore; will add an explicit test for just this optimizer.
I've added a separate decomposition and merge optimizer for the
PhasedXZcase. I think ideally this would just replace the existing code which emits separatePhasedXPowandZgates, but that is currently used in, for example, the ion trap decompositions where it is not clear that these devices and/or APIs supportPhasedXZ. Once this is in I have a separate PR to add support forPhasedXZin theEjectZoptimizer.