sabre_swap.py

# This code is part of Qiskit.
#
# (C) Copyright IBM 2017, 2020.
#
# This code is licensed under the Apache License, Version 2.0. You may
# obtain a copy of this license in the LICENSE.txt file in the root directory
# of this source tree or at http://www.apache.org/licenses/LICENSE-2.0.
#
# Any modifications or derivative works of this code must retain this
# copyright notice, and modified files need to carry a notice indicating
# that they have been altered from the originals.

"""Routing via SWAP insertion using the SABRE method from Li et al."""

import logging
from collections import defaultdict
from copy import copy, deepcopy
import numpy as np

from qiskit.circuit.library.standard_gates import SwapGate
from qiskit.circuit.quantumregister import Qubit
from qiskit.transpiler.basepasses import TransformationPass
from qiskit.transpiler.exceptions import TranspilerError
from qiskit.transpiler.layout import Layout
from qiskit.dagcircuit import DAGOpNode

logger = logging.getLogger(__name__)

EXTENDED_SET_SIZE = 20  # Size of lookahead window. TODO: set dynamically to len(current_layout)
EXTENDED_SET_WEIGHT = 0.5  # Weight of lookahead window compared to front_layer.

DECAY_RATE = 0.001  # Decay coefficient for penalizing serial swaps.
DECAY_RESET_INTERVAL = 5  # How often to reset all decay rates to 1.


class SabreSwap(TransformationPass):
    r"""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.

    **References:**

    [1] Li, Gushu, Yufei Ding, and Yuan Xie. "Tackling the qubit mapping problem
    for NISQ-era quantum devices." ASPLOS 2019.
    `arXiv:1809.02573 <https://arxiv.org/pdf/1809.02573.pdf>`_
    """

    def __init__(self, coupling_map, heuristic="basic", seed=None, fake_run=False):
        r"""SabreSwap initializer.

        Args:
            coupling_map (CouplingMap): 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.

        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.

            .. math::

                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. :math:`|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.

            .. math::

                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).

            .. math::

                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)]
                    }
        """

        super().__init__()

        # Assume bidirectional couplings, fixing gate direction is easy later.
        if coupling_map.is_symmetric:
            self.coupling_map = coupling_map
        else:
            self.coupling_map = deepcopy(coupling_map)
            self.coupling_map.make_symmetric()

        self.heuristic = heuristic
        self.seed = seed
        self.fake_run = fake_run
        self.applied_predecessors = None
        self.qubits_decay = None
        self._bit_indices = None

    def run(self, dag):
        """Run the SabreSwap pass on `dag`.

        Args:
            dag (DAGCircuit): the directed acyclic graph to be mapped.
        Returns:
            DAGCircuit: A dag mapped to be compatible with the coupling_map.
        Raises:
            TranspilerError: if the coupling map or the layout are not
            compatible with the DAG
        """
        if len(dag.qregs) != 1 or dag.qregs.get("q", None) is None:
            raise TranspilerError("Sabre swap runs on physical circuits only.")

        if len(dag.qubits) > self.coupling_map.size():
            raise TranspilerError("More virtual qubits exist than physical.")

        rng = np.random.default_rng(self.seed)

        # Preserve input DAG's name, regs, wire_map, etc. but replace the graph.
        mapped_dag = None
        if not self.fake_run:
            mapped_dag = dag._copy_circuit_metadata()

        canonical_register = dag.qregs["q"]
        current_layout = Layout.generate_trivial_layout(canonical_register)

        self._bit_indices = {bit: idx for idx, bit in enumerate(canonical_register)}

        # A decay factor for each qubit used to heuristically penalize recently
        # used qubits (to encourage parallelism).
        self.qubits_decay = {qubit: 1 for qubit in dag.qubits}

        # Start algorithm from the front layer and iterate until all gates done.
        num_search_steps = 0
        front_layer = dag.front_layer()
        self.applied_predecessors = defaultdict(int)
        for _, input_node in dag.input_map.items():
            for successor in self._successors(input_node, dag):
                self.applied_predecessors[successor] += 1
        while front_layer:
            execute_gate_list = []

            # Remove as many immediately applicable gates as possible
            for node in front_layer:
                if len(node.qargs) == 2:
                    v0, v1 = node.qargs
                    if self.coupling_map.graph.has_edge(current_layout[v0], current_layout[v1]):
                        execute_gate_list.append(node)
                else:  # Single-qubit gates as well as barriers are free
                    execute_gate_list.append(node)

            if execute_gate_list:
                for node in execute_gate_list:
                    self._apply_gate(mapped_dag, node, current_layout, canonical_register)
                    front_layer.remove(node)
                    for successor in self._successors(node, dag):
                        self.applied_predecessors[successor] += 1
                        if self._is_resolved(successor):
                            front_layer.append(successor)

                    if node.qargs:
                        self._reset_qubits_decay()

                # Diagnostics
                logger.debug(
                    "free! %s",
                    [
                        (n.name if isinstance(n, DAGOpNode) else None, n.qargs)
                        for n in execute_gate_list
                    ],
                )
                logger.debug(
                    "front_layer: %s",
                    [(n.name if isinstance(n, DAGOpNode) else None, n.qargs) for n in front_layer],
                )

                continue

            # After all free gates are exhausted, heuristically find
            # the best swap and insert it. When two or more swaps tie
            # for best score, pick one randomly.
            extended_set = self._obtain_extended_set(dag, front_layer)
            swap_candidates = self._obtain_swaps(front_layer, current_layout)
            swap_scores = dict.fromkeys(swap_candidates, 0)
            for swap_qubits in swap_scores:
                trial_layout = current_layout.copy()
                trial_layout.swap(*swap_qubits)
                score = self._score_heuristic(
                    self.heuristic, front_layer, extended_set, trial_layout, swap_qubits
                )
                swap_scores[swap_qubits] = score
            min_score = min(swap_scores.values())
            best_swaps = [k for k, v in swap_scores.items() if v == min_score]
            best_swaps.sort(key=lambda x: (self._bit_indices[x[0]], self._bit_indices[x[1]]))
            best_swap = rng.choice(best_swaps)
            swap_node = DAGOpNode(op=SwapGate(), qargs=best_swap)
            self._apply_gate(mapped_dag, swap_node, current_layout, canonical_register)
            current_layout.swap(*best_swap)

            num_search_steps += 1
            if num_search_steps % DECAY_RESET_INTERVAL == 0:
                self._reset_qubits_decay()
            else:
                self.qubits_decay[best_swap[0]] += DECAY_RATE
                self.qubits_decay[best_swap[1]] += DECAY_RATE

            # Diagnostics
            logger.debug("SWAP Selection...")
            logger.debug("extended_set: %s", [(n.name, n.qargs) for n in extended_set])
            logger.debug("swap scores: %s", swap_scores)
            logger.debug("best swap: %s", best_swap)
            logger.debug("qubits decay: %s", self.qubits_decay)

        self.property_set["final_layout"] = current_layout

        if not self.fake_run:
            return mapped_dag
        return dag

    def _apply_gate(self, mapped_dag, node, current_layout, canonical_register):
        if self.fake_run:
            return
        new_node = _transform_gate_for_layout(node, current_layout, canonical_register)
        mapped_dag.apply_operation_back(new_node.op, new_node.qargs, new_node.cargs)

    def _reset_qubits_decay(self):
        """Reset all qubit decay factors to 1 upon request (to forget about
        past penalizations).
        """
        self.qubits_decay = {k: 1 for k in self.qubits_decay.keys()}

    def _successors(self, node, dag):
        for _, successor, edge_data in dag.edges(node):
            if not isinstance(successor, DAGOpNode):
                continue
            if isinstance(edge_data, Qubit):
                yield successor

    def _is_resolved(self, node):
        """Return True if all of a node's predecessors in dag are applied."""
        return self.applied_predecessors[node] == len(node.qargs)

    def _obtain_extended_set(self, dag, front_layer):
        """Populate extended_set by looking ahead a fixed number of gates.
        For each existing element add a successor until reaching limit.
        """
        extended_set = list()
        incremented = list()
        tmp_front_layer = front_layer
        done = False
        while tmp_front_layer and not done:
            new_tmp_front_layer = list()
            for node in tmp_front_layer:
                for successor in self._successors(node, dag):
                    incremented.append(successor)
                    self.applied_predecessors[successor] += 1
                    if self._is_resolved(successor):
                        new_tmp_front_layer.append(successor)
                        if len(successor.qargs) == 2:
                            extended_set.append(successor)
                if len(extended_set) >= EXTENDED_SET_SIZE:
                    done = True
                    break
            tmp_front_layer = new_tmp_front_layer
        for node in incremented:
            self.applied_predecessors[node] -= 1
        return extended_set

    def _obtain_swaps(self, front_layer, current_layout):
        """Return a set of candidate swaps that affect qubits in front_layer.

        For each virtual qubit in front_layer, find its current location
        on hardware and the physical qubits in that neighborhood. Every SWAP
        on virtual qubits that corresponds to one of those physical couplings
        is a candidate SWAP.

        Candidate swaps are sorted so SWAP(i,j) and SWAP(j,i) are not duplicated.
        """
        candidate_swaps = set()
        for node in front_layer:
            for virtual in node.qargs:
                physical = current_layout[virtual]
                for neighbor in self.coupling_map.neighbors(physical):
                    virtual_neighbor = current_layout[neighbor]
                    swap = sorted([virtual, virtual_neighbor], key=lambda q: self._bit_indices[q])
                    candidate_swaps.add(tuple(swap))

        return candidate_swaps

    def _compute_cost(self, layer, layout):
        cost = 0
        for node in layer:
            cost += self.coupling_map.distance(layout[node.qargs[0]], layout[node.qargs[1]])
        return cost

    def _score_heuristic(self, heuristic, front_layer, extended_set, layout, swap_qubits=None):
        """Return a heuristic score for a trial layout.

        Assuming a trial layout has resulted from a SWAP, we now assign a cost
        to it. The goodness of a layout is evaluated based on how viable it makes
        the remaining virtual gates that must be applied.
        """
        first_cost = self._compute_cost(front_layer, layout)
        if heuristic == "basic":
            return first_cost

        first_cost /= len(front_layer)
        second_cost = 0
        if extended_set:
            second_cost = self._compute_cost(extended_set, layout) / len(extended_set)
        total_cost = first_cost + EXTENDED_SET_WEIGHT * second_cost
        if heuristic == "lookahead":
            return total_cost

        if heuristic == "decay":
            return (
                max(self.qubits_decay[swap_qubits[0]], self.qubits_decay[swap_qubits[1]])
                * total_cost
            )

        raise TranspilerError("Heuristic %s not recognized." % heuristic)


def _transform_gate_for_layout(op_node, layout, device_qreg):
    """Return node implementing a virtual op on given layout."""
    mapped_op_node = copy(op_node)

    premap_qargs = op_node.qargs
    mapped_qargs = map(lambda x: device_qreg[layout[x]], premap_qargs)
    mapped_op_node.qargs = list(mapped_qargs)

    return mapped_op_node