# This code is part of Qiskit.
#
# (C) Copyright IBM 2021.
#
# 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.
"""
====================================================================
Synthesis Plugins (:mod:`qiskit.transpiler.passes.synthesis.plugin`)
====================================================================
.. currentmodule:: qiskit.transpiler.passes.synthesis.plugin
This module defines the plugin interfaces for the synthesis transpiler passes
in Qiskit. These provide a hook point for external python packages to implement
their own synthesis techniques and have them seamlessly exposed as opt-in
options to users when they run :func:`~qiskit.compiler.transpile`.
The plugin interfaces are built using setuptools
`entry points <https://setuptools.readthedocs.io/en/latest/userguide/entry_point.html>`__
which enable packages external to qiskit to advertise they include a synthesis
plugin.
See :mod:`qiskit.transpiler.preset_passmanagers.plugin` for details on how
to write plugins for transpiler stages.
Writing Plugins
===============
Unitary Synthesis Plugins
-------------------------
To write a unitary synthesis plugin there are 2 main steps. The first step is
to create a subclass of the abstract plugin class:
:class:`~qiskit.transpiler.passes.synthesis.plugin.UnitarySynthesisPlugin`.
The plugin class defines the interface and contract for unitary synthesis
plugins. The primary method is
:meth:`~qiskit.transpiler.passes.synthesis.plugin.UnitarySynthesisPlugin.run`
which takes in a single positional argument, a unitary matrix as a numpy array,
and is expected to return a :class:`~qiskit.dagcircuit.DAGCircuit` object
representing the synthesized circuit from that unitary matrix. Then to inform
the Qiskit transpiler about what information is necessary for the pass there
are several required property methods that need to be implemented such as
``supports_basis_gates`` and ``supports_coupling_map`` depending on whether the
plugin supports and/or requires that input to perform synthesis. For the full
details refer to the
:class:`~qiskit.transpiler.passes.synthesis.plugin.UnitarySynthesisPlugin`
documentation for all the required fields. An example plugin class would look
something like::
from qiskit.transpiler.passes.synthesis import plugin
from qiskit_plugin_pkg.synthesis import generate_dag_circuit_from_matrix
class SpecialUnitarySynthesis(plugin.UnitarySynthesisPlugin):
@property
def supports_basis_gates(self):
return True
@property
def supports_coupling_map(self):
return False
@property
def supports_natural_direction(self):
return False
@property
def supports_pulse_optimize(self):
return False
@property
def supports_gate_lengths(self):
return False
@property
def supports_gate_errors(self):
return False
@property
def supports_gate_lengths_by_qubit(self):
return False
@property
def supports_gate_errors_by_qubit(self):
return False
@property
def min_qubits(self):
return None
@property
def max_qubits(self):
return None
@property
def supported_bases(self):
return None
def run(self, unitary, **options):
basis_gates = options['basis_gates']
dag_circuit = generate_dag_circuit_from_matrix(unitary, basis_gates)
return dag_circuit
If for some reason the available inputs to the
:meth:`~qiskit.transpiler.passes.synthesis.plugin.UnitarySynthesisPlugin.run`
method are insufficient please open an issue and we can discuss expanding the
plugin interface with new opt-in inputs that can be added in a backwards
compatible manner for future releases. Do note though that this plugin interface
is considered stable and guaranteed to not change in a breaking manner. If
changes are needed (for example to expand the available optional input options)
it will be done in a way that will **not** require changes from existing
plugins.
.. note::
All methods prefixed with ``supports_`` are reserved on a
``UnitarySynthesisPlugin`` derived class for part of the interface. You
should not define any custom ``supports_*`` methods on a subclass that
are not defined in the abstract class.
The second step is to expose the
:class:`~qiskit.transpiler.passes.synthesis.plugin.UnitarySynthesisPlugin` as
a setuptools entry point in the package metadata. This is done by simply adding
an ``entry_points`` entry to the ``setuptools.setup`` call in the ``setup.py``
for the plugin package with the necessary entry points under the
``qiskit.unitary_synthesis`` namespace. For example::
entry_points = {
'qiskit.unitary_synthesis': [
'special = qiskit_plugin_pkg.module.plugin:SpecialUnitarySynthesis',
]
},
(note that the entry point ``name = path`` is a single string not a Python
expression). There isn't a limit to the number of plugins a single package can
include as long as each plugin has a unique name. So a single package can
expose multiple plugins if necessary. The name ``default`` is used by Qiskit
itself and can't be used in a plugin.
Unitary Synthesis Plugin Configuration
''''''''''''''''''''''''''''''''''''''
For some unitary synthesis plugins that expose multiple options and tunables
the plugin interface has an option for users to provide a free form
configuration dictionary. This will be passed through to the ``run()`` method
as the ``config`` kwarg. If your plugin has these configuration options you
should clearly document how a user should specify these configuration options
and how they're used as it's a free form field.
High-Level Synthesis Plugins
----------------------------
Writing a high-level synthesis plugin is conceptually similar to writing a
unitary synthesis plugin. The first step is to create a subclass of the
abstract plugin class:
:class:`~qiskit.transpiler.passes.synthesis.plugin.HighLevelSynthesisPlugin`,
which defines the interface and contract for high-level synthesis plugins.
The primary method is
:meth:`~qiskit.transpiler.passes.synthesis.plugin.HighLevelSynthesisPlugin.run`.
It takes in a single positional argument, a "higher-level-object" to be
synthesized, which is any object of type :class:`~qiskit.circuit.Operation`
(including, for example,
:class:`~qiskit.circuit.library.generalized_gates.linear_function.LinearFunction` or
:class:`~qiskit.quantum_info.operators.symplectic.clifford.Clifford`).
The method
:meth:`~qiskit.transpiler.passes.synthesis.plugin.HighLevelSynthesisPlugin.run`
is expected to return a :class:`~qiskit.circuit.QuantumCircuit` object
representing the synthesized circuit from that higher-level-object.
It is also allowed to return ``None`` representing that the synthesis method is
unable to synthesize the given higher-level-object.
The actual synthesis of higher-level objects is performed by
:class:`~qiskit.transpiler.passes.synthesis.high_level_synthesis.HighLevelSynthesis`
transpiler pass.
In the near future,
:class:`~qiskit.transpiler.passes.synthesis.plugin.HighLevelSynthesisPlugin`
will be extended with additional information necessary to run this transpiler
pass, for instance whether the plugin supports and/or requires ``coupling_map``
to perform synthesis.
For the full details refer to the
:class:`~qiskit.transpiler.passes.synthesis.plugin.HighLevelSynthesisPlugin`
documentation for all the required fields. An example plugin class would look
something like::
from qiskit.transpiler.passes.synthesis.plugin import HighLevelSynthesisPlugin
from qiskit.synthesis.clifford import synth_clifford_bm
class SpecialSynthesisClifford(HighLevelSynthesisPlugin):
def run(self, high_level_object, **options):
if higher_level_object.num_qubits <= 3:
return synth_clifford_bm(high_level_object)
else:
return None
The above example creates a plugin to synthesize objects of type
:class:`~qiskit.quantum_info.operators.symplectic.clifford.Clifford that have
at most 3 qubits, using the method ``synth_clifford_bm``.
The second step is to expose the
:class:`~qiskit.transpiler.passes.synthesis.plugin.HighLevelSynthesisPlugin` as
a setuptools entry point in the package metadata. This is done by adding
an ``entry_points`` entry to the ``setuptools.setup`` call in the ``setup.py``
for the plugin package with the necessary entry points under the
``qiskit.synthesis`` namespace. For example::
entry_points = {
'qiskit.synthesis': [
'clifford.special = qiskit_plugin_pkg.module.plugin:SpecialSynthesisClifford',
]
},
(note that the entry point ``name = path`` is a single string not a Python
expression). The ``name`` consists of two parts separated by dot ".": the
name of the
type of :class:`~qiskit.circuit.Operation` to which the synthesis plugin applies
(``clifford``), and the name of the plugin (``special``).
There isn't a limit to the number of plugins a single package can
include as long as each plugin has a unique name.
Using Plugins
=============
To use a plugin all you need to do is install the package that includes a
synthesis plugin. Then Qiskit will automatically discover the installed
plugins and expose them as valid options for the appropriate
:func:`~qiskit.compiler.transpile` kwargs and pass constructors. If there are
any installed plugins which can't be loaded/imported this will be logged to
Python logging.
To get the installed list of installed unitary synthesis plugins you can use the
:func:`qiskit.transpiler.passes.synthesis.plugin.unitary_synthesis_plugin_names`
function.
Plugin API
==========
Unitary Synthesis Plugins
-------------------------
.. autosummary::
:toctree: ../stubs/
UnitarySynthesisPlugin
UnitarySynthesisPluginManager
unitary_synthesis_plugin_names
High-Level Synthesis Plugins
----------------------------
.. autosummary::
:toctree: ../stubs/
HighLevelSynthesisPlugin
HighLevelSynthesisPluginManager
"""
import abc
import stevedore
[ドキュメント]class UnitarySynthesisPlugin(abc.ABC):
"""Abstract unitary synthesis plugin class
This abstract class defines the interface for unitary synthesis plugins.
"""
@property
@abc.abstractmethod
def max_qubits(self):
"""Return the maximum number of qubits the unitary synthesis plugin supports.
If the size of the unitary to be synthesized exceeds this value the
``default`` plugin will be used. If there is no upper bound return
``None`` and all unitaries (``>= min_qubits`` if it's defined) will be
passed to this plugin when it's enabled.
"""
pass
@property
@abc.abstractmethod
def min_qubits(self):
"""Return the minimum number of qubits the unitary synthesis plugin supports.
If the size of the unitary to be synthesized is below this value the
``default`` plugin will be used. If there is no lower bound return
``None`` and all unitaries (``<= max_qubits`` if it's defined) will be
passed to this plugin when it's enabled.
"""
pass
@property
@abc.abstractmethod
def supports_basis_gates(self):
"""Return whether the plugin supports taking ``basis_gates``
If this returns ``True`` the plugin's ``run()`` method will be
passed a ``basis_gates`` kwarg with a list of gate names the target
backend supports. For example, ``['sx', 'x', 'cx', 'id', 'rz']``."""
pass
@property
@abc.abstractmethod
def supports_coupling_map(self):
"""Return whether the plugin supports taking ``coupling_map``
If this returns ``True`` the plugin's ``run()`` method will receive
one kwarg ``coupling_map``. The ``coupling_map`` kwarg will be set to a
tuple with the first element being a
:class:`~qiskit.transpiler.CouplingMap` object representing the qubit
connectivity of the target backend, the second element will be a list
of integers that represent the qubit indices in the coupling map that
unitary is on. Note that if the target backend doesn't have a coupling
map set, the ``coupling_map`` kwarg's value will be ``(None, qubit_indices)``.
"""
pass
@property
@abc.abstractmethod
def supports_natural_direction(self):
"""Return whether the plugin supports a toggle for considering
directionality of 2-qubit gates as ``natural_direction``.
Refer to the documentation for :class:`~qiskit.transpiler.passes.UnitarySynthesis`
for the possible values and meaning of these values.
"""
pass
@property
@abc.abstractmethod
def supports_pulse_optimize(self):
"""Return whether the plugin supports a toggle to optimize pulses
during synthesis as ``pulse_optimize``.
Refer to the documentation for :class:`~qiskit.transpiler.passes.UnitarySynthesis`
for the possible values and meaning of these values.
"""
pass
@property
def supports_gate_lengths_by_qubit(self):
"""Return whether the plugin supports taking ``gate_lengths_by_qubit``
This differs from ``supports_gate_lengths``/``gate_lengths`` by using a different
view of the same data. Instead of being keyed by gate name this is keyed by qubit
and uses :class:`~.Gate` instances to represent gates (instead of gate names)
``gate_lengths_by_qubit`` will be a dictionary in the form of
``{(qubits,): [Gate, length]}``. For example::
{
(0,): [SXGate(): 0.0006149355812506126, RZGate(): 0.0],
(0, 1): [CXGate(): 0.012012477900732316]
}
where the ``length`` value is in units of seconds.
Do note that this dictionary might not be complete or could be empty
as it depends on the target backend reporting gate lengths on every
gate for each qubit.
This defaults to False
"""
return False
@property
def supports_gate_errors_by_qubit(self):
"""Return whether the plugin supports taking ``gate_errors_by_qubit``
This differs from ``supports_gate_errors``/``gate_errors`` by using a different
view of the same data. Instead of being keyed by gate name this is keyed by qubit
and uses :class:`~.Gate` instances to represent gates (instead of gate names).
``gate_errors_by_qubit`` will be a dictionary in the form of
``{(qubits,): [Gate, error]}``. For example::
{
(0,): [SXGate(): 0.0006149355812506126, RZGate(): 0.0],
(0, 1): [CXGate(): 0.012012477900732316]
}
Do note that this dictionary might not be complete or could be empty
as it depends on the target backend reporting gate errors on every
gate for each qubit. The gate error rates reported in ``gate_errors``
are provided by the target device ``Backend`` object and the exact
meaning might be different depending on the backend.
This defaults to False
"""
return False
@property
@abc.abstractmethod
def supports_gate_lengths(self):
"""Return whether the plugin supports taking ``gate_lengths``
``gate_lengths`` will be a dictionary in the form of
``{gate_name: {(qubit_1, qubit_2): length}}``. For example::
{
'sx': {(0,): 0.0006149355812506126, (1,): 0.0006149355812506126},
'cx': {(0, 1): 0.012012477900732316, (1, 0): 5.191111111111111e-07}
}
where the ``length`` value is in units of seconds.
Do note that this dictionary might not be complete or could be empty
as it depends on the target backend reporting gate lengths on every
gate for each qubit.
"""
pass
@property
@abc.abstractmethod
def supports_gate_errors(self):
"""Return whether the plugin supports taking ``gate_errors``
``gate_errors`` will be a dictionary in the form of
``{gate_name: {(qubit_1, qubit_2): error}}``. For example::
{
'sx': {(0,): 0.0006149355812506126, (1,): 0.0006149355812506126},
'cx': {(0, 1): 0.012012477900732316, (1, 0): 5.191111111111111e-07}
}
Do note that this dictionary might not be complete or could be empty
as it depends on the target backend reporting gate errors on every
gate for each qubit. The gate error rates reported in ``gate_errors``
are provided by the target device ``Backend`` object and the exact
meaning might be different depending on the backend.
"""
pass
@property
@abc.abstractmethod
def supported_bases(self):
"""Returns a dictionary of supported bases for synthesis
This is expected to return a dictionary where the key is a string
basis and the value is a list of gate names that the basis works in.
If the synthesis method doesn't support multiple bases this should
return ``None``. For example::
{
"XZX": ["rz", "rx"],
"XYX": ["rx", "ry"],
}
If a dictionary is returned by this method the run kwargs will be
passed a parameter ``matched_basis`` which contains a list of the
basis strings (i.e. keys in the dictionary) which match the target basis
gate set for the transpilation. If no entry in the dictionary matches
the target basis gate set then the ``matched_basis`` kwarg will be set
to an empty list, and a plugin can choose how to deal with the target
basis gate set not matching the plugin's capabilities.
"""
pass
@property
def supports_target(self):
"""Whether the plugin supports taking ``target`` as an option
``target`` will be a :class:`~.Target` object representing the target
device for the output of the synthesis pass.
By default this will be ``False`` since the plugin interface predates
the :class:`~.Target` class. If a plugin returns ``True`` for this
attribute, it is expected that the plugin will use the
:class:`~.Target` instead of the values passed if any of
``supports_gate_lengths``, ``supports_gate_errors``,
``supports_coupling_map``, and ``supports_basis_gates`` are set
(although ideally all those parameters should contain duplicate
information).
"""
return False
[ドキュメント] @abc.abstractmethod
def run(self, unitary, **options):
"""Run synthesis for the given unitary matrix
Args:
unitary (numpy.ndarray): The unitary matrix to synthesize to a
:class:`~qiskit.dagcircuit.DAGCircuit` object
options: The optional kwargs that are passed based on the output
the ``support_*`` methods on the class. Refer to the
documentation for these methods on
:class:`~qiskit.transpiler.passes.synthesis.plugin.UnitarySynthesisPlugin`
to see what the keys and values are.
Returns:
DAGCircuit: The dag circuit representation of the unitary. Alternatively, you can return
a tuple of the form ``(dag, wires)`` where ``dag`` is the dag circuit representation of
the circuit representation of the unitary and ``wires`` is the mapping wires to use for
:meth:`qiskit.dagcircuit.DAGCircuit.substitute_node_with_dag`. If you return a tuple
and ``wires`` is ``None`` this will behave just as if only a
:class:`~qiskit.dagcircuit.DAGCircuit` was returned. Additionally if this returns
``None`` no substitution will be made.
"""
pass
[ドキュメント]class UnitarySynthesisPluginManager:
"""Unitary Synthesis plugin manager class
This class tracks the installed plugins, it has a single property,
``ext_plugins`` which contains a list of stevedore plugin objects.
"""
def __init__(self):
self.ext_plugins = stevedore.ExtensionManager(
"qiskit.unitary_synthesis", invoke_on_load=True, propagate_map_exceptions=True
)
[ドキュメント]def unitary_synthesis_plugin_names():
"""Return a list of installed unitary synthesis plugin names
Returns:
list: A list of the installed unitary synthesis plugin names. The plugin names are valid
values for the :func:`~qiskit.compiler.transpile` kwarg ``unitary_synthesis_method``.
"""
# NOTE: This is not a shared global instance to avoid an import cycle
# at load time for the default plugin.
plugins = UnitarySynthesisPluginManager()
return plugins.ext_plugins.names()
[ドキュメント]class HighLevelSynthesisPlugin(abc.ABC):
"""Abstract high-level synthesis plugin class.
This abstract class defines the interface for high-level synthesis plugins.
"""
[ドキュメント] @abc.abstractmethod
def run(self, high_level_object, **options):
"""Run synthesis for the given Operation.
Args:
high_level_object (Operation): The Operation to synthesize to a
:class:`~qiskit.dagcircuit.DAGCircuit` object
options: The optional kwargs.
Returns:
QuantumCircuit: The quantum circuit representation of the Operation
when successful, and ``None`` otherwise.
"""
pass
[ドキュメント]class HighLevelSynthesisPluginManager:
"""Class tracking the installed high-level-synthesis plugins."""
def __init__(self):
self.plugins = stevedore.ExtensionManager(
"qiskit.synthesis", invoke_on_load=True, propagate_map_exceptions=True
)
# The registered plugin names should be of the form <OperationName.SynthesisMethodName>.
# Create a dict, mapping <OperationName> to the list of its <SynthesisMethodName>s.
self.plugins_by_op = {}
for plugin_name in self.plugins.names():
op_name, method_name = plugin_name.split(".")
if op_name not in self.plugins_by_op.keys():
self.plugins_by_op[op_name] = []
self.plugins_by_op[op_name].append(method_name)
[ドキュメント] def method_names(self, op_name):
"""Returns plugin methods for op_name."""
if op_name in self.plugins_by_op.keys():
return self.plugins_by_op[op_name]
else:
return []
[ドキュメント] def method(self, op_name, method_name):
"""Returns the plugin for ``op_name`` and ``method_name``."""
plugin_name = op_name + "." + method_name
return self.plugins[plugin_name].obj