Skip to main contentIBM Quantum Documentation
This page is from an old version of Qiskit SDK. Go to the latest version.

SabreSwap

class qiskit.transpiler.passes.SabreSwap(*args, **kwargs)

GitHub

Bases: TransformationPass

Map input circuit onto a backend topology via insertion of SWAPs.

Implementation of the SWAP-based heuristic search from the SABRE qubit mapping paper [1] (Algorithm 1). The heuristic aims to minimize the number of lossy SWAPs inserted and the depth of the circuit.

This algorithm starts from an initial layout of virtual qubits onto physical qubits, and iterates over the circuit DAG until all gates are exhausted, inserting SWAPs along the way. It only considers 2-qubit gates as only those are germane for the mapping problem (it is assumed that 3+ qubit gates are already decomposed).

In each iteration, it will first check if there are any gates in the front_layer that can be directly applied. If so, it will apply them and remove them from front_layer, and replenish that layer with new gates if possible. Otherwise, it will try to search for SWAPs, insert the SWAPs, and update the mapping.

The search for SWAPs is restricted, in the sense that we only consider physical qubits in the neighborhood of those qubits involved in front_layer. These give rise to a swap_candidate_list which is scored according to some heuristic cost function. The best SWAP is implemented and current_layout updated.

This transpiler pass adds onto the SABRE algorithm in that it will run multiple trials of the algorithm with different seeds. The best output, determined by the trial with the least amount of SWAPed inserted, will be selected from the random trials.

References:

[1] Li, Gushu, Yufei Ding, and Yuan Xie. “Tackling the qubit mapping problem for NISQ-era quantum devices.” ASPLOS 2019. arXiv:1809.02573

SabreSwap initializer.

Parameters

  • coupling_map (Union[CouplingMap, Target]) – CouplingMap of the target backend.
  • heuristic (str) – The type of heuristic to use when deciding best swap strategy (‘basic’ or ‘lookahead’ or ‘decay’).
  • seed (int) – random seed used to tie-break among candidate swaps.
  • fake_run (bool) – if true, it only pretend to do routing, i.e., no swap is effectively added.
  • trials (int) – The number of seed trials to run sabre with. These will be run in parallel (unless the PassManager is already running in parallel). If not specified this defaults to the number of physical CPUs on the local system. For reproducible results it is recommended that you set this explicitly, as the output will be deterministic for a fixed number of trials.

Raises

TranspilerError – If the specified heuristic is not valid.

Additional Information:

The search space of possible SWAPs on physical qubits is explored by assigning a score to the layout that would result from each SWAP. The goodness of a layout is evaluated based on how viable it makes the remaining virtual gates that must be applied. A few heuristic cost functions are supported

  • ‘basic’:

The sum of distances for corresponding physical qubits of interacting virtual qubits in the front_layer.

Hbasic=gateFD[π(gate.q1)][π(gate.q2)]H_{basic} = \sum_{gate \in F} D[\pi(gate.q_1)][\pi(gate.q2)]
  • ‘lookahead’:

This is the sum of two costs: first is the same as the basic cost. Second is the basic cost but now evaluated for the extended set as well (i.e. E|E| number of upcoming successors to gates in front_layer F). This is weighted by some amount EXTENDED_SET_WEIGHT (W) to signify that upcoming gates are less important that the front_layer.

Hdecay=1FgateFD[π(gate.q1)][π(gate.q2)]+W1EgateED[π(gate.q1)][π(gate.q2)]H_{decay}=\frac{1}{\left|{F}\right|}\sum_{gate \in F} D[\pi(gate.q_1)][\pi(gate.q2)] + W*\frac{1}{\left|{E}\right|} \sum_{gate \in E} D[\pi(gate.q_1)][\pi(gate.q2)]
  • ‘decay’:

This is the same as ‘lookahead’, but the whole cost is multiplied by a decay factor. This increases the cost if the SWAP that generated the trial layout was recently used (i.e. it penalizes increase in depth).

Hdecay=max(decay(SWAP.q1),decay(SWAP.q2))1FgateFD[π(gate.q1)][π(gate.q2)]+W1EgateED[π(gate.q1)][π(gate.q2)]H_{decay} = max(decay(SWAP.q_1), decay(SWAP.q_2)) { \frac{1}{\left|{F}\right|} \sum_{gate \in F} D[\pi(gate.q_1)][\pi(gate.q2)]\\ + W *\frac{1}{\left|{E}\right|} \sum_{gate \in E} D[\pi(gate.q_1)][\pi(gate.q2)] }

Attributes

is_analysis_pass

Check if the pass is an analysis pass.

If the pass is an AnalysisPass, that means that the pass can analyze the DAG and write the results of that analysis in the property set. Modifications on the DAG are not allowed by this kind of pass.

is_transformation_pass

Check if the pass is a transformation pass.

If the pass is a TransformationPass, that means that the pass can manipulate the DAG, but cannot modify the property set (but it can be read).


Methods

execute

execute(passmanager_ir, state, callback=None)

Execute optimization task for input Qiskit IR.

Parameters

  • passmanager_ir (Any) – Qiskit IR to optimize.
  • state (PassManagerState) – State associated with workflow execution by the pass manager itself.
  • callback (Callable | None) – A callback function which is caller per execution of optimization task.

Returns

Optimized Qiskit IR and state of the workflow.

Return type

tuple[Any, qiskit.passmanager.compilation_status.PassManagerState]

name

name()

Name of the pass.

Return type

str

run

run(dag)

Run the SabreSwap pass on dag.

Parameters

dag (DAGCircuit) – the directed acyclic graph to be mapped.

Returns

A dag mapped to be compatible with the coupling_map.

Return type

DAGCircuit

Raises

  • TranspilerError – if the coupling map or the layout are not
  • compatible with the DAG**, or **if the coupling_map=None

update_status

update_status(state, run_state)

Update workflow status.

Parameters

  • state (PassManagerState) – Pass manager state to update.
  • run_state (RunState) – Completion status of current task.

Returns

Updated pass manager state.

Return type

PassManagerState

Was this page helpful?
Report a bug or request content on GitHub.