Source code for quantify_scheduler.backends.qblox.helpers

# Repository: https://gitlab.com/quantify-os/quantify-scheduler
# Licensed according to the LICENCE file on the main branch
"""Helper functions for Qblox backend."""
from __future__ import annotations

import dataclasses
import math
import warnings
from collections import defaultdict
from copy import deepcopy
from typing import TYPE_CHECKING, Any, Callable

import numpy as np

from quantify_scheduler.backends.qblox import constants
from quantify_scheduler.backends.types.qblox import (
    ComplexChannelDescription,
    DigitalChannelDescription,
    OpInfo,
    RealChannelDescription,
)
from quantify_scheduler.helpers.schedule import _extract_port_clocks_used
from quantify_scheduler.helpers.waveforms import exec_waveform_function
from quantify_scheduler.operations.control_flow_library import (
    ConditionalOperation,
    ControlFlowOperation,
    LoopOperation,
)
from quantify_scheduler.operations.operation import Operation
from quantify_scheduler.operations.pulse_library import WindowOperation
from quantify_scheduler.schedules.schedule import Schedule, ScheduleBase

if TYPE_CHECKING:
    from quantify_scheduler.backends.qblox.instrument_compilers import ClusterCompiler


[docs] def generate_waveform_data( data_dict: dict, sampling_rate: float, duration: float | None = None ) -> np.ndarray: """ Generates an array using the parameters specified in ``data_dict``. Parameters ---------- data_dict : dict The dictionary that contains the values needed to parameterize the waveform. ``data_dict['wf_func']`` is then called to calculate the values. sampling_rate : float The sampling rate used to generate the time axis values. duration : float or None, optional The duration of the waveform in seconds. This parameter can be used if ``data_dict`` does not contain a ``'duration'`` key. By default None. Returns ------- wf_data : np.ndarray The (possibly complex) values of the generated waveform. The number of values is determined by rounding to the nearest integer. Raises ------ TypeError If ``data_dict`` does not contain a ``'duration'`` entry and ``duration is None``. """ if duration is None: try: duration = data_dict["duration"] except KeyError as exc: raise TypeError( "Parameter 'duration' has value None. If 'data_dict' does not contain " "'duration', the function parameter can be used instead." ) from exc num_samples = round(duration * sampling_rate) t = np.arange(start=0, stop=num_samples, step=1) / sampling_rate wf_data = exec_waveform_function( wf_func=data_dict["wf_func"], t=t, pulse_info=data_dict ) return wf_data
[docs] def generate_waveform_names_from_uuid(uuid: Any) -> tuple[str, str]: """ Generates names for the I and Q parts of the complex waveform based on a unique identifier for the pulse/acquisition. Parameters ---------- uuid A unique identifier for a pulse/acquisition. Returns ------- uuid_I: Name for the I waveform. uuid_Q: Name for the Q waveform. """ return f"{str(uuid)}_I", f"{str(uuid)}_Q"
[docs] def generate_uuid_from_wf_data(wf_data: np.ndarray, decimals: int = 12) -> str: """ Creates a unique identifier from the waveform data, using a hash. Identical arrays yield identical strings within the same process. Parameters ---------- wf_data: The data to generate the unique id for. decimals: The number of decimal places to consider. Returns ------- : A unique identifier. """ waveform_hash = hash(wf_data.round(decimals=decimals).tobytes()) return str(waveform_hash)
[docs] def add_to_wf_dict_if_unique(wf_dict: dict[str, Any], waveform: np.ndarray) -> int: """ Adds a waveform to the waveform dictionary if it is not yet in there and returns the uuid and index. If it is already present it simply returns the uuid and index. Parameters ---------- wf_dict: The waveform dict in the format expected by the sequencer. waveform: The waveform to add. Returns ------- Dict[str, Any] The (updated) wf_dict. str The uuid of the waveform. int The index. """ def generate_entry(name: str, data: np.ndarray, idx: int) -> dict[str, Any]: return {name: {"data": data.tolist(), "index": idx}} def find_first_free_wf_index(): index = 0 reserved_indices = [wf_dict[uuid]["index"] for uuid in wf_dict] while index in reserved_indices: index += 1 return index if not np.isrealobj(waveform): raise RuntimeError("This function only accepts real arrays.") uuid = generate_uuid_from_wf_data(waveform) if uuid in wf_dict: index: int = wf_dict[uuid]["index"] else: index: int = find_first_free_wf_index() wf_dict.update(generate_entry(uuid, waveform, index)) return index
[docs] def generate_waveform_dict(waveforms_complex: dict[str, np.ndarray]) -> dict[str, dict]: """ Takes a dictionary with complex waveforms and generates a new dictionary with real valued waveforms with a unique index, as required by the hardware. Parameters ---------- waveforms_complex Dictionary containing the complex waveforms. Keys correspond to a unique identifier, value is the complex waveform. Returns ------- dict[str, dict] A dictionary with as key the unique name for that waveform, as value another dictionary containing the real-valued data (list) as well as a unique index. Note that the index of the Q waveform is always the index of the I waveform +1. .. admonition:: Examples .. jupyter-execute:: import numpy as np from quantify_scheduler.backends.qblox.helpers import generate_waveform_dict complex_waveforms = {12345: np.array([1, 2])} generate_waveform_dict(complex_waveforms) # {'12345_I': {'data': [1, 2], 'index': 0}, # '12345_Q': {'data': [0, 0], 'index': 1}} """ wf_dict = {} for idx, (uuid, complex_data) in enumerate(waveforms_complex.items()): name_i, name_q = generate_waveform_names_from_uuid(uuid) to_add = { name_i: {"data": complex_data.real.tolist(), "index": 2 * idx}, name_q: {"data": complex_data.imag.tolist(), "index": 2 * idx + 1}, } wf_dict.update(to_add) return wf_dict
[docs] def to_grid_time(time: float, grid_time_ns: int = constants.GRID_TIME) -> int: """ Convert time value in s to time in ns, and verify that it is aligned with grid time. Takes a float value representing a time in seconds as used by the schedule, and returns the integer valued time in nanoseconds that the sequencer uses. The time value needs to be aligned with grid time, i.e., needs to be a multiple of :data:`~.constants.GRID_TIME`, within a tolerance of 1 picosecond. Parameters ---------- time A time value in seconds. grid_time_ns The grid time to use in nanoseconds. Returns ------- : The integer valued nanosecond time. Raises ------ ValueError If ``time`` is not a multiple of :data:`~.constants.GRID_TIME` within the tolerance. """ time_ns_float = time * 1e9 time_ns = int(round(time_ns_float)) tolerance = constants.GRID_TIME_TOLERANCE_TIME if ( not math.isclose( time_ns_float, time_ns, abs_tol=tolerance, rel_tol=0 ) # rel_tol=0 results in: abs(a-b) <= max(0, abs_tol) or time_ns % grid_time_ns != 0 ): raise ValueError( f"Attempting to use a time value of {time_ns_float} ns." f" Please ensure that the durations of operations and wait times between" f" operations are multiples of {grid_time_ns} ns" f" (tolerance: {tolerance:.0e} ns). If you think this is a mistake, try " "increasing the tolerance by setting e.g.:" f" `quantify_scheduler.backends.qblox.constants.GRID_TIME_TOLERANCE_TIME = 1.1e-2` " "at the top of your script." ) return time_ns
[docs] def is_multiple_of_grid_time( time: float, grid_time_ns: int = constants.GRID_TIME ) -> bool: """ Determine whether a time value in seconds is a multiple of the grid time. Within a tolerance as defined by :meth:`~quantify_scheduler.backends.qblox.helpers.to_grid_time`. Parameters ---------- time A time value in seconds. grid_time_ns The grid time to use in nanoseconds. Returns ------- : ``True`` if ``time`` is a multiple of the grid time, ``False`` otherwise. """ try: _ = to_grid_time(time=time, grid_time_ns=grid_time_ns) except ValueError: return False return True
[docs] def get_nco_phase_arguments(phase_deg: float) -> int: """ Converts a phase in degrees to the int arguments the NCO phase instructions expect. We take ``phase_deg`` modulo 360 to account for negative phase and phase larger than 360. Parameters ---------- phase_deg The phase in degrees Returns ------- : The int corresponding to the phase argument. """ phase_deg %= 360 return round(phase_deg * constants.NCO_PHASE_STEPS_PER_DEG)
[docs] def get_nco_set_frequency_arguments(frequency_hz: float) -> int: """ Converts a frequency in Hz to the int argument the NCO set_freq instruction expects. Parameters ---------- frequency_hz The frequency in Hz. Returns ------- : The frequency expressed in steps for the NCO set_freq instruction. Raises ------ ValueError If the frequency_hz is out of range. """ frequency_steps = round(frequency_hz * constants.NCO_FREQ_STEPS_PER_HZ) if ( frequency_steps < -constants.NCO_FREQ_LIMIT_STEPS or frequency_steps > constants.NCO_FREQ_LIMIT_STEPS ): min_max_frequency_in_hz = ( constants.NCO_FREQ_LIMIT_STEPS / constants.NCO_FREQ_STEPS_PER_HZ ) raise ValueError( f"Attempting to set NCO frequency. " f"The frequency must be between and including " f"-{min_max_frequency_in_hz:e} Hz and {min_max_frequency_in_hz:e} Hz. " f"Got {frequency_hz:e} Hz." ) return frequency_steps
@dataclasses.dataclass
[docs] class Frequencies: """Holds and validates frequencies."""
[docs] clock: float
[docs] LO: float | None = None
[docs] IF: float | None = None
def __post_init__(self): if self.LO is not None and math.isnan(self.LO): self.LO = None if self.IF is not None and math.isnan(self.IF): self.IF = None
[docs] def validate(self): """Validates frequencies.""" if self.clock is None or math.isnan(self.clock): raise ValueError(f"Clock frequency must be specified ({self.clock=}).") for freq in [self.LO, self.IF]: if freq is not None and math.isnan(freq): raise ValueError( f"Frequencies must be a number or None, not NaN ({self.LO=}, {self.IF=})." )
[docs] def determine_clock_lo_interm_freqs( freqs: Frequencies, downconverter_freq: float | None = None, mix_lo: bool | None = True, ) -> Frequencies: r""" From known frequency for the local oscillator or known intermodulation frequency, determine any missing frequency, after optionally applying ``downconverter_freq`` to the clock frequency. If ``mix_lo`` is ``True``, the following relation is obeyed: :math:`f_{RF} = f_{LO} + f_{IF}`. If ``mix_lo`` is ``False``, :math:`f_{RF} = f_{LO}` is upheld. .. warning:: Using ``downconverter_freq`` requires custom Qblox hardware, do not use otherwise. Parameters ---------- freqs : Frequencies Frequencies object containing clock, local oscillator (LO) and Intermodulation frequency (IF), the frequency of the numerically controlled oscillator (NCO). downconverter_freq : Optional[float] Frequency for downconverting the clock frequency, using: :math:`f_\mathrm{out} = f_\mathrm{downconverter} - f_\mathrm{in}`. mix_lo : bool Flag indicating whether IQ mixing is enabled with the LO. Returns ------- : :class:`.Frequencies` object containing the determined LO and IF frequencies and the optionally downconverted clock frequency. Warns ----- RuntimeWarning In case ``downconverter_freq`` is set equal to 0, warns to unset via ``null``/``None`` instead. RuntimeWarning In case LO is overridden to clock due to ``mix_lo`` being `False` Raises ------ ValueError In case ``downconverter_freq`` is less than 0. ValueError In case ``downconverter_freq`` is less than ``clock_freq``. ValueError In case ``mix_lo`` is ``True`` and neither LO frequency nor IF has been supplied. ValueError In case ``mix_lo`` is ``True`` and both LO frequency and IF have been supplied and do not adhere to :math:`f_{RF} = f_{LO} + f_{IF}`. """ def _downconvert_clock(downconverter_freq: float, clock_freq: float) -> float: if downconverter_freq == 0: warnings.warn( "Downconverter frequency 0 supplied. To unset 'downconverter_freq', " "set to `null` (json) / `None` instead in hardware configuration.", RuntimeWarning, ) if downconverter_freq < 0: raise ValueError( f"Downconverter frequency must be positive ({downconverter_freq=:e})" ) if downconverter_freq < clock_freq: raise ValueError( f"Downconverter frequency must be greater than clock frequency " f"({downconverter_freq=:e}, {clock_freq=:e})" ) return downconverter_freq - clock_freq freqs.validate() if downconverter_freq is not None: freqs.clock = _downconvert_clock( downconverter_freq=downconverter_freq, clock_freq=freqs.clock, ) if not mix_lo: if freqs.LO is not None and not math.isclose(freqs.LO, freqs.clock): warnings.warn( f"Overriding {freqs.LO=} to {freqs.clock=} due to mix_lo=False." ) freqs.LO = freqs.clock else: if freqs.LO is None and freqs.IF is None: raise ValueError( f"Frequency settings underconstrained for {freqs.clock=}." f" Neither LO nor IF supplied ({freqs.LO=}, {freqs.IF=})." ) elif freqs.LO is not None and freqs.IF is not None: if not math.isclose(freqs.LO + freqs.IF, freqs.clock): raise ValueError( f"Frequency settings overconstrained." f" {freqs.clock=} must be equal to {freqs.LO=}+{freqs.IF=} if both are supplied." ) elif freqs.LO is None and freqs.IF is not None: freqs.LO = freqs.clock - freqs.IF elif freqs.LO is not None and freqs.IF is None: freqs.IF = freqs.clock - freqs.LO return freqs
[docs] def generate_port_clock_to_device_map( device_compilers: dict[str, Any] ) -> dict[tuple[str, str], str]: """ Generates a mapping that specifies which port-clock combinations belong to which device. Here, device means a top-level entry in the hardware config, e.g. a Cluster, not which module within the Cluster. Each port-clock combination may only occur once. Parameters ---------- device_compilers: Dictionary containing compiler configs. Returns ------- : A dictionary with as key a tuple representing a port-clock combination, and as value the name of the device. Note that multiple port-clocks may point to the same device. Raises ------ ValueError If a port-clock combination occurs multiple times in the hardware configuration. """ portclock_map = {} for device_name, device_compiler in device_compilers.items(): if hasattr(device_compiler, "portclock_to_path"): for portclock in device_compiler.portclock_to_path.keys(): portclock_map[portclock] = device_name return portclock_map
[docs] class LoopBegin(Operation): """ Operation to indicate the beginning of a loop. Parameters ---------- repetitions : int number of repetitions t0 : float, optional time offset, by default 0 """ def __init__(self, repetitions: int, t0: float = 0) -> None: super().__init__(name="Loop") self.data.update( { "name": "Loop", "control_flow_info": { "t0": t0, "repetitions": repetitions, }, } ) self._update() def __str__(self) -> str: """ Represent the Operation as string. Returns ------- str description """ return self._get_signature(self.data["control_flow_info"])
[docs] class ConditionalBegin(Operation): """ Operation to indicate the beginning of a conditional. Parameters ---------- qubit_name The name of the qubit to condition on. feedback_trigger_address Feedback trigger address t0 Time offset, by default 0 """ def __init__( self, qubit_name: str, feedback_trigger_address: int, t0: float ) -> None:
[docs] class_name = self.__class__.__name__
super().__init__(name=class_name) self.data.update( { "name": class_name, "control_flow_info": { "qubit_name": qubit_name, "t0": t0, "feedback_trigger_address": feedback_trigger_address, }, } ) self._update() def __str__(self) -> str: """ Represent the Operation as string. Returns ------- str The string representation of this operation. """ return self._get_signature(self.data["control_flow_info"])
[docs] def _get_control_flow_begin( control_flow_operation: ControlFlowOperation, ) -> Operation: port_clocks = _extract_port_clocks_used(control_flow_operation) if isinstance(control_flow_operation, LoopOperation): begin_operation: Operation = LoopBegin( control_flow_operation.data["control_flow_info"]["repetitions"], control_flow_operation.data["control_flow_info"]["t0"], ) elif isinstance(control_flow_operation, ConditionalOperation): begin_operation: Operation = ConditionalBegin( control_flow_operation.data["control_flow_info"]["qubit_name"], control_flow_operation.data["control_flow_info"][ "feedback_trigger_address" ], control_flow_operation.data["control_flow_info"]["t0"], ) begin_operation["pulse_info"] = [ { "wf_func": None, "clock": clock, "port": port, "duration": 0, "control_flow_begin": True, **begin_operation["control_flow_info"], } for port, clock in port_clocks ] return begin_operation
[docs] class _ControlFlowReturn(Operation): """ An operation that signals the end of the current control flow statement. Cannot be added to Schedule manually. Parameters ---------- t0 : float, optional time offset, by default 0 """ def __init__(self, t0: float = 0) -> None: super().__init__(name="ControlFlowReturn") self.data.update( { "name": "ControlFlowReturn", "control_flow_info": { "t0": t0, "duration": 0.0, "return_stack": True, }, } ) self._update() def __str__(self) -> str: return self._get_signature(self.data["control_flow_info"])
[docs] def _get_control_flow_end( control_flow_operation: ControlFlowOperation, ) -> Operation: port_clocks = _extract_port_clocks_used(control_flow_operation) if isinstance(control_flow_operation, (LoopOperation, ConditionalOperation)): end_operation: Operation = _ControlFlowReturn() end_operation["pulse_info"] = [ { "wf_func": None, "clock": clock, "port": port, "duration": 0, "control_flow_end": True, **end_operation["control_flow_info"], } for port, clock in port_clocks ] return end_operation
[docs] def _get_list_of_operations_for_op_info_creation( operation: Operation | Schedule, time_offset: float, accumulator: list[tuple[float, Operation]], ) -> None: if isinstance(operation, ScheduleBase): for schedulable in operation.schedulables.values(): abs_time = schedulable["abs_time"] inner_operation = operation.operations[schedulable["operation_id"]] _get_list_of_operations_for_op_info_creation( inner_operation, time_offset + abs_time, accumulator ) elif isinstance(operation, ControlFlowOperation): accumulator.append( (to_grid_time(time_offset) * 1e-9, _get_control_flow_begin(operation)) ) _get_list_of_operations_for_op_info_creation( operation.body, time_offset, accumulator ) accumulator.append( ( to_grid_time(time_offset + operation.body.duration) * 1e-9, _get_control_flow_end(operation), ) ) else: accumulator.append((to_grid_time(time_offset) * 1e-9, operation))
[docs] def assign_pulse_and_acq_info_to_devices( schedule: Schedule, device_compilers: dict[str, ClusterCompiler], ): """ Traverses the schedule and generates `OpInfo` objects for every pulse and acquisition, and assigns it to the correct `ClusterCompiler`. Parameters ---------- schedule The schedule to extract the pulse and acquisition info from. device_compilers Dictionary containing InstrumentCompilers as values and their names as keys. portclock_to_path Dictionary containing the hardware path to connected to each portclock. Raises ------ RuntimeError This exception is raised then the function encountered an operation that has no pulse or acquisition info assigned to it. KeyError This exception is raised when attempting to assign a pulse with a port-clock combination that is not defined in the hardware configuration. KeyError This exception is raised when attempting to assign an acquisition with a port-clock combination that is not defined in the hardware configuration. """ portclock_mapping = generate_port_clock_to_device_map(device_compilers) list_of_operations: list[tuple[float, Operation]] = list() _get_list_of_operations_for_op_info_creation(schedule, 0, list_of_operations) list_of_operations.sort(key=lambda abs_time_and_op: abs_time_and_op[0]) for operation_start_time, op_data in list_of_operations: assert isinstance(op_data, Operation) if isinstance(op_data, WindowOperation): continue if not op_data.valid_pulse and not op_data.valid_acquisition: raise RuntimeError( f"Operation is not a valid pulse or acquisition. Please check" f" whether the device compilation been performed successfully. " f"Operation data: {repr(op_data)}" ) for pulse_data in op_data.data["pulse_info"]: if "t0" in pulse_data: pulse_start_time = operation_start_time + pulse_data["t0"] else: pulse_start_time = operation_start_time # Check whether start time aligns with grid time try: _ = to_grid_time(pulse_start_time) except ValueError as exc: raise ValueError( f"An operation start time of {pulse_start_time * 1e9} ns does not " f"align with a grid time of {constants.GRID_TIME} ns. Please make " f"sure the start time of all operations is a multiple of " f"{constants.GRID_TIME} ns.\n\nOffending operation:" f"\n{repr(op_data)}." ) from exc if pulse_data.get("reference_magnitude", None) is not None: warnings.warn( "reference_magnitude parameter not implemented. This parameter will be ignored.", RuntimeWarning, ) port = pulse_data["port"] clock = pulse_data["clock"] portclock = f"{port}-{clock}" combined_data = OpInfo( name=op_data.data["name"], data=pulse_data, timing=pulse_start_time, ) if port is None: # Distribute clock operations to all sequencers utilizing that clock for map_portclock, device_name in portclock_mapping.items(): map_port, map_clock = map_portclock.split("-") if (combined_data.name == "LatchReset") or map_clock == clock: device_compilers[device_name].add_op_info( port=map_port, clock=map_clock, op_info=combined_data ) else: if portclock not in portclock_mapping: raise KeyError( f"Could not assign pulse data to device. The combination " f"of port {port} and clock {clock} could not be found " f"in hardware configuration.\n\nAre both the port and clock " f"specified in the hardware configuration?\n\n" f"Relevant operation:\n{combined_data}." ) device_name = portclock_mapping[portclock] device_compilers[device_name].add_op_info( port=port, clock=clock, op_info=combined_data ) for acq_data in op_data.data["acquisition_info"]: if "t0" in acq_data: acq_start_time = operation_start_time + acq_data["t0"] else: acq_start_time = operation_start_time port = acq_data["port"] clock = acq_data["clock"] portclock = f"{port}-{clock}" if port is None: continue combined_data = OpInfo( name=op_data.data["name"], data=acq_data, timing=acq_start_time, ) if portclock not in portclock_mapping: raise KeyError( f"Could not assign acquisition data to device. The combination " f"of port {port} and clock {clock} could not be found " f"in hardware configuration.\n\nAre both the port and clock " f"specified in the hardware configuration?\n\n" f"Relevant operation:\n{combined_data}." ) device_name = portclock_mapping[portclock] device_compilers[device_name].add_op_info( port=port, clock=clock, op_info=combined_data )
[docs] def calc_from_units_volt( voltage_range, name: str, param_name: str, offset: float | None ) -> float | None: """ Helper method to calculate the offset from mV or V. Then compares to given voltage range, and throws a ValueError if out of bounds. Parameters ---------- voltage_range The range of the voltage levels of the device used. name The name of the device used. param_name The name of the offset parameter this method is using. offset The value of the offset parameter this method is using. Returns ------- : The normalized offsets. Raises ------ RuntimeError When a unit range is given that is not supported, or a value is given that falls outside the allowed range. """ offset_in_arg = offset # Always in volts if offset_in_arg is None: return None conversion_factor = 1 if voltage_range.units == "mV": conversion_factor = 1e3 elif voltage_range.units != "V": raise RuntimeError( f"Parameter {param_name} of {name} specifies " f"the units {voltage_range.units}, but the Qblox " f"backend only supports mV and V." ) calculated_offset = offset_in_arg * conversion_factor if ( calculated_offset < voltage_range.min_val or calculated_offset > voltage_range.max_val ): raise ValueError( f"Attempting to set {param_name} of {name} to " f"{offset_in_arg} V. {param_name} has to be between " f"{voltage_range.min_val / conversion_factor} and " f"{voltage_range.max_val / conversion_factor} V!" ) return calculated_offset
[docs] def single_scope_mode_acquisition_raise(sequencer_0, sequencer_1, module_name): """ Raises an error stating that only one scope mode acquisition can be used per module. Parameters ---------- sequencer_0 First sequencer which attempts to use the scope mode acquisition. sequencer_1 Second sequencer which attempts to use the scope mode acquisition. module_name Name of the module. Raises ------ ValueError Always raises the error message. """ raise ValueError( f"Both sequencer '{sequencer_0}' and '{sequencer_1}' " f"of '{module_name}' attempts to perform scope mode acquisitions. " f"Only one sequencer per device can " f"trigger raw trace capture.\n\nPlease ensure that " f"only one port-clock combination performs " f"raw trace acquisition per instrument." )
[docs] def _preprocess_legacy_hardware_config( hardware_config: dict[str, Any] ) -> dict[str, Any]: """Modify a legacy hardware config into a form that is compatible with the current backend.""" def _modify_inner_dicts( config: dict[str, Any], target_key: str, value_modifier: Callable ) -> dict[str, Any]: for key, value in config.items(): if key == target_key: config[key] = value_modifier(value) elif isinstance(config[key], dict): config[key] = _modify_inner_dicts( config=value, target_key=target_key, value_modifier=value_modifier ) return config def _replace_deprecated_portclock_keys( portclock_configs: list[dict], ) -> list[dict]: for portclock_config in portclock_configs: for deprecated_key, updated_key in { "init_offset_awg_path_0": "init_offset_awg_path_I", "init_offset_awg_path_1": "init_offset_awg_path_Q", "init_gain_awg_path_0": "init_gain_awg_path_I", "init_gain_awg_path_1": "init_gain_awg_path_Q", }.items(): if deprecated_key in portclock_config: warnings.warn( f"'{deprecated_key}' is deprecated and will be removed from the public " f"interface in quantify-scheduler >= 0.20.0. Please use " f"'{updated_key}' instead.", FutureWarning, ) portclock_config[updated_key] = portclock_config[deprecated_key] del portclock_config[deprecated_key] return portclock_configs # Preprocessed config is a deepcopy of original config return _modify_inner_dicts( config=deepcopy(hardware_config), target_key="portclock_configs", value_modifier=_replace_deprecated_portclock_keys, )
[docs] def _generate_new_style_hardware_compilation_config( old_style_config: dict, ) -> dict: """ Generate a new-style QbloxHardwareCompilationConfig from an old-style hardware config. Parameters ---------- old_style_config Old-style hardware config. Returns ------- dict New-style hardware compilation config dictionary. """ def _convert_complex_channel_config( cluster_name: str, module_slot_idx: int, channel_name: str, old_channel_config: dict, new_style_config: dict, ) -> None: """Add information from old-style complex channel config to new-style config.""" new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ][channel_name] = {} port_name = f"{cluster_name}.module{module_slot_idx}.{channel_name}" for ( channel_cfg_key, channel_cfg_value, ) in old_channel_config.items(): # Find attached port-clock combinations: channel_port_clocks = [ f"{pc_cfg['port']}-{pc_cfg['clock']}" for pc_cfg in old_channel_config["portclock_configs"] ] if channel_cfg_key in [ "marker_debug_mode_enable", "mix_lo", "downconverter_freq", ]: new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ][channel_name][channel_cfg_key] = channel_cfg_value elif channel_cfg_key == "lo_name": # Add IQ mixer to the hardware_description: new_style_config["hardware_description"][ f"iq_mixer_{channel_cfg_value}" ] = {"instrument_type": "IQMixer"} # Add LO and IQ mixer to connectivity graph: new_style_config["connectivity"]["graph"].extend( [ ( port_name, f"iq_mixer_{channel_cfg_value}.if", ), ( f"{channel_cfg_value}.output", f"iq_mixer_{channel_cfg_value}.lo", ), ] ) # Overwrite port_name to IQ mixer RF output: port_name = f"iq_mixer_{channel_cfg_value}.rf" if "frequency" in old_style_config[channel_cfg_value]: # Set lo_freq for all port-clock combinations (external LO) for port_clock in channel_port_clocks: new_style_config["hardware_options"]["modulation_frequencies"][ port_clock ]["lo_freq"] = old_style_config[channel_cfg_value]["frequency"] elif channel_cfg_key == "lo_freq": # Set lo_freq for all port-clock combinations (RF modules) for port_clock in channel_port_clocks: new_style_config["hardware_options"]["modulation_frequencies"][ port_clock ]["lo_freq"] = channel_cfg_value elif channel_cfg_key == "auto_lo_cal": # Set auto_lo_cal for all port-clock combinations (RF modules) for port_clock in channel_port_clocks: new_style_config["hardware_options"]["mixer_corrections"][ port_clock ]["auto_lo_cal"] = channel_cfg_value elif channel_cfg_key == "dc_mixer_offset_I": # Set mixer offsets for all port-clock combinations (RF modules) for port_clock in channel_port_clocks: new_style_config["hardware_options"]["mixer_corrections"][ port_clock ]["dc_offset_i"] = channel_cfg_value elif channel_cfg_key == "dc_mixer_offset_Q": # Set mixer offsets for all port-clock combinations (RF modules) for port_clock in channel_port_clocks: new_style_config["hardware_options"]["mixer_corrections"][ port_clock ]["dc_offset_q"] = channel_cfg_value elif channel_cfg_key == "input_gain_I": # Set input gains for all port-clock combinations (RF modules) for port_clock in channel_port_clocks: new_style_config["hardware_options"]["input_gain"][port_clock][ "gain_I" ] = channel_cfg_value elif channel_cfg_key == "input_gain_Q": # Set input gains for all port-clock combinations (RF modules) for port_clock in channel_port_clocks: new_style_config["hardware_options"]["input_gain"][port_clock][ "gain_Q" ] = channel_cfg_value elif channel_cfg_key == "output_att": # Set output attenuation for all port-clock combinations (RF modules) for port_clock in channel_port_clocks: new_style_config["hardware_options"]["output_att"][ port_clock ] = channel_cfg_value elif channel_cfg_key == "input_att": # Set input attenuation for all port-clock combinations (RF modules) for port_clock in channel_port_clocks: new_style_config["hardware_options"]["input_att"][ port_clock ] = channel_cfg_value elif channel_cfg_key == "portclock_configs": # Add connectivity information to connectivity graph: for portclock_cfg in channel_cfg_value: new_style_config["connectivity"]["graph"].append( ( port_name, f"{portclock_cfg['port']}", ) ) port_clock = ( f"{portclock_cfg.pop('port')}-{portclock_cfg.pop('clock')}" ) if "interm_freq" in portclock_cfg: # Set intermodulation freqs from portclock config: new_style_config["hardware_options"]["modulation_frequencies"][ port_clock ]["interm_freq"] = portclock_cfg.pop("interm_freq") if "mixer_amp_ratio" in portclock_cfg: # Set intermodulation freqs from portclock config: new_style_config["hardware_options"]["mixer_corrections"][ port_clock ]["amp_ratio"] = portclock_cfg.pop("mixer_amp_ratio") if "auto_sideband_cal" in portclock_cfg: # Set auto_sideband_cal from portclock config: new_style_config["hardware_options"]["mixer_corrections"][ port_clock ]["auto_sideband_cal"] = portclock_cfg.pop("auto_sideband_cal") if "mixer_phase_error_deg" in portclock_cfg: # Set intermodulation freqs from portclock config: new_style_config["hardware_options"]["mixer_corrections"][ port_clock ]["phase_error"] = portclock_cfg.pop("mixer_phase_error_deg") if portclock_cfg != {}: # Set remaining portclock config parameters to sequencer options: new_style_config["hardware_options"]["sequencer_options"][ port_clock ] = portclock_cfg def _convert_real_channel_config( cluster_name: str, module_slot_idx: int, channel_name: str, old_channel_config: dict, new_style_config: dict, ) -> None: """Add information from old-style real channel config to new-style config.""" new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ][channel_name] = {} port_name = f"{cluster_name}.module{module_slot_idx}.{channel_name}" for ( channel_cfg_key, channel_cfg_value, ) in old_channel_config.items(): # Find attached port-clock combinations: channel_port_clocks = [ f"{pc_cfg['port']}-{pc_cfg['clock']}" for pc_cfg in old_channel_config["portclock_configs"] ] if channel_cfg_key in ["marker_debug_mode_enable", "mix_lo"]: new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ][channel_name][channel_cfg_key] = channel_cfg_value elif channel_cfg_key in ("input_gain_0", "input_gain_1"): # Set input gains for all port-clock combinations for port_clock in channel_port_clocks: new_style_config["hardware_options"]["input_gain"][ port_clock ] = channel_cfg_value elif channel_cfg_key == "portclock_configs": # Add connectivity information to connectivity graph: for portclock_cfg in channel_cfg_value: new_style_config["connectivity"]["graph"].append( ( port_name, f"{portclock_cfg['port']}", ) ) port_clock = ( f"{portclock_cfg.pop('port')}-{portclock_cfg.pop('clock')}" ) if "interm_freq" in portclock_cfg: # Set intermodulation freqs from portclock config: new_style_config["hardware_options"]["modulation_frequencies"][ port_clock ]["interm_freq"] = portclock_cfg.pop("interm_freq") if "init_gain_awg_path_I" in portclock_cfg: # Set init gain from portclock config: new_style_config["hardware_options"]["sequencer_options"][ port_clock ]["init_gain_awg_path_I"] = portclock_cfg.pop( "init_gain_awg_path_I" ) if "init_gain_awg_path_Q" in portclock_cfg: # Set init gain from portclock config: new_style_config["hardware_options"]["sequencer_options"][ port_clock ]["init_gain_awg_path_Q"] = portclock_cfg.pop( "init_gain_awg_path_Q" ) if any( option in portclock_cfg for option in [ "init_gain_awg_path_I", "init_gain_awg_path_Q", "init_offset_awg_path_I", "init_offset_awg_path_Q", "qasm_hook_func", "ttl_acq_threshold", ] ): # Set remaining portclock config parameters to sequencer options: new_style_config["hardware_options"]["sequencer_options"][ port_clock ] = portclock_cfg if any("optical_control" in pc for pc in channel_port_clocks): channel_mixer = "OpticalModulator" mixer_tag = "optical_mod" mixer_output_tag = "out" else: channel_mixer = "IQMixer" mixer_tag = "iq_mixer" mixer_output_tag = "rf" if channel_cfg_key == "lo_name": # Add optical/iq mixer to the hardware_description new_style_config["hardware_description"][ f"{mixer_tag}_{channel_cfg_value}" ] = {"instrument_type": channel_mixer} # Add LO and mixer to connectivity graph: new_style_config["connectivity"]["graph"].extend( [ ( port_name, f"{mixer_tag}_{channel_cfg_value}.if", ), ( f"{channel_cfg_value}.output", f"{mixer_tag}_{channel_cfg_value}.lo", ), ] ) # Overwrite port_name to mixer output: port_name = f"{mixer_tag}_{channel_cfg_value}.{mixer_output_tag}" if "frequency" in old_style_config[channel_cfg_value]: # Set lo_freq for all port-clock combinations (external LO) for port_clock in channel_port_clocks: new_style_config["hardware_options"]["modulation_frequencies"][ port_clock ]["lo_freq"] = old_style_config[channel_cfg_value]["frequency"] def _convert_digital_channel_config( cluster_name: str, module_slot_idx: int, channel_name: str, old_channel_config: dict, new_style_config: dict, ) -> None: new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ][channel_name] = {} for ( channel_cfg_key, channel_cfg_value, ) in old_channel_config.items(): if channel_cfg_key == "portclock_configs": # Add connectivity information to connectivity graph: for portclock_cfg in channel_cfg_value: new_style_config["connectivity"]["graph"].append( ( f"{cluster_name}.module{module_slot_idx}.{channel_name}", f"{portclock_cfg['port']}", ) ) port_clock = ( f"{portclock_cfg.pop('port')}-{portclock_cfg.pop('clock')}" ) if "in_threshold_primary" in portclock_cfg: # Set init gain from portclock config: new_style_config["hardware_options"]["digitization_thresholds"][ port_clock ]["in_threshold_primary"] = portclock_cfg.pop( "in_threshold_primary" ) def _convert_cluster_module_config( cluster_name: str, module_slot_idx: int, old_module_config: dict, new_style_config: dict, ) -> None: """Add information from old-style Cluster module config to new-style config.""" new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ] = {} for module_cfg_key, module_cfg_value in old_module_config.items(): if module_cfg_key in ["instrument_type", "sequence_to_file"]: new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ][module_cfg_key] = module_cfg_value elif module_cfg_key.startswith("complex_"): # Portclock configs dict must be last item in dict for correct conversion old_channel_config = { k: v for k, v in module_cfg_value.items() if k != "portclock_configs" } old_channel_config["portclock_configs"] = module_cfg_value[ "portclock_configs" ] _convert_complex_channel_config( cluster_name=cluster_name, module_slot_idx=module_slot_idx, channel_name=module_cfg_key, old_channel_config=old_channel_config, new_style_config=new_style_config, ) # Remove channel description if only default values are set parsed_channel_description = ComplexChannelDescription.model_validate( new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ][module_cfg_key] ) if ( parsed_channel_description.model_dump() == ComplexChannelDescription().model_dump() ): new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ].pop(module_cfg_key) elif module_cfg_key.startswith("real_"): # Portclock configs dict must be last item in dict for correct conversion old_channel_config = { k: v for k, v in module_cfg_value.items() if k != "portclock_configs" } old_channel_config["portclock_configs"] = module_cfg_value[ "portclock_configs" ] _convert_real_channel_config( cluster_name=cluster_name, module_slot_idx=module_slot_idx, channel_name=module_cfg_key, old_channel_config=module_cfg_value, new_style_config=new_style_config, ) # Remove channel description if only default values are set parsed_channel_description = RealChannelDescription.model_validate( new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ][module_cfg_key] ) if ( parsed_channel_description.model_dump() == RealChannelDescription().model_dump() ): new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ].pop(module_cfg_key) elif module_cfg_key.startswith("digital_"): _convert_digital_channel_config( cluster_name=cluster_name, module_slot_idx=module_slot_idx, channel_name=module_cfg_key, old_channel_config=module_cfg_value, new_style_config=new_style_config, ) # Remove channel description if only default values are set parsed_channel_description = DigitalChannelDescription.model_validate( new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ][module_cfg_key] ) if ( parsed_channel_description.model_dump() == DigitalChannelDescription().model_dump() ): new_style_config["hardware_description"][cluster_name]["modules"][ module_slot_idx ].pop(module_cfg_key) def _convert_cluster_config( cluster_name: str, old_cluster_config: dict, new_style_config: dict ) -> None: """Add information from old-style Cluster config to new-style config.""" new_style_config["hardware_description"][cluster_name] = { "instrument_type": "Cluster", "modules": {}, } for cluster_cfg_key, cluster_cfg_value in old_cluster_config.items(): if cluster_cfg_key in ["ref", "sequence_to_file"]: new_style_config["hardware_description"][cluster_name][ cluster_cfg_key ] = cluster_cfg_value elif "module" in cluster_cfg_key: _convert_cluster_module_config( cluster_name=cluster_name, module_slot_idx=int(cluster_cfg_key.split(sep="module")[1]), old_module_config=cluster_cfg_value, new_style_config=new_style_config, ) warnings.warn( "The hardware configuration dictionary is deprecated and will not be supported in " "quantify-scheduler >= 1.0.0. Please use a `HardwareCompilationConfig` instead. For " "more information on how to migrate from old- to new-style hardware specification, " "please visit https://quantify-os.org/docs/quantify-scheduler/dev/examples/hardware_config_migration.html " "in the documentation.", FutureWarning, ) # Update deprecated keys and avoid modifying the original config old_style_config = _preprocess_legacy_hardware_config(old_style_config) # Initialize new-style hardware compilation config dictionary new_style_config = { "config_type": "quantify_scheduler.backends.qblox_backend.QbloxHardwareCompilationConfig", "hardware_description": {}, "hardware_options": defaultdict(lambda: defaultdict(dict)), "connectivity": {"graph": []}, } # Loop over old-style config and populate new-style input dicts for hw_cfg_key, hw_cfg_value in old_style_config.items(): if hw_cfg_key == "backend": pass elif hw_cfg_key in ["latency_corrections", "distortion_corrections"]: new_style_config["hardware_options"][hw_cfg_key] = hw_cfg_value elif "instrument_type" not in hw_cfg_value: warnings.warn( f"Skipping hardware config entry '{hw_cfg_key}' because it does not specify an instrument type." ) elif hw_cfg_value["instrument_type"] == "Cluster": _convert_cluster_config( cluster_name=hw_cfg_key, old_cluster_config=hw_cfg_value, new_style_config=new_style_config, ) elif hw_cfg_value["instrument_type"] == "LocalOscillator": new_style_config["hardware_description"][hw_cfg_key] = {} for lo_cfg_key, lo_cfg_value in hw_cfg_value.items(): if lo_cfg_key in ["instrument_type", "power"]: new_style_config["hardware_description"][hw_cfg_key][ lo_cfg_key ] = lo_cfg_value elif lo_cfg_key == "frequency": pass else: raise KeyError( f"Unexpected key {lo_cfg_key} in LocalOscillator config." ) else: raise ValueError( f"Unexpected instrument_type {hw_cfg_value['instrument_type']} in old-style hardware config." ) return new_style_config