pennylane-qrack 0.24.1__py3-none-win_amd64.whl

pennylane_qrack/qrack_ace_device.py

@@ -0,0 +1,447 @@
+# Copyright 2020 Xanadu Quantum Technologies Inc.
+
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+
+#     http://www.apache.org/licenses/LICENSE-2.0
+
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Base device class for PennyLane-Qrack.
+"""
+from functools import reduce
+import cmath, math
+import os
+import pathlib
+import sys
+import itertools as it
+
+import numpy as np
+
+# PennyLane v0.42 introduced the `exceptions` module and will raise
+# deprecation warnings if they are imported from the top-level module.
+
+# This ensures backwards compatibility with older versions of PennyLane.
+try:
+    from pennylane.exceptions import DeviceError, QuantumFunctionError
+except (ModuleNotFoundError, ImportError) as import_error:
+    from pennylane import DeviceError, QuantumFunctionError
+
+from pennylane.devices import QubitDevice
+from pennylane.ops import (
+    BasisState,
+    Adjoint,
+)
+from pennylane.wires import Wires
+
+from pyqrack import QrackAceBackend, Pauli
+
+from ._version import __version__
+from sys import platform as _platform
+
+# tolerance for numerical errors
+tolerance = 1e-10
+
+
+class QrackAceDevice(QubitDevice):
+    """Qrack Ace device"""
+
+    name = "Qrack Ace device"
+    short_name = "qrack.ace"
+    pennylane_requires = ">=0.11.0"
+    version = __version__
+    author = "Daniel Strano, adapted from Steven Oud and Xanadu"
+
+    _capabilities = {
+        "model": "qubit",
+        "tensor_observables": True,
+        "inverse_operations": True,
+        "returns_state": False,
+    }
+
+    _observable_map = {
+        "PauliX": Pauli.PauliX,
+        "PauliY": Pauli.PauliY,
+        "PauliZ": Pauli.PauliZ,
+        "Identity": Pauli.PauliI,
+        "Hadamard": None,
+        "Hermitian": None,
+        "Prod": None,
+        # "Sum": None,
+        # "SProd": None,
+        # "Exp": None,
+        # "Projector": None,
+        # "Hamiltonian": None,
+        # "SparseHamiltonian": None
+    }
+
+    observables = _observable_map.keys()
+    operations = {
+        "Identity",
+        "C(Identity)",
+        "MultiRZ",
+        "SWAP",
+        "ISWAP",
+        "PSWAP",
+        "CNOT",
+        "CY",
+        "CZ",
+        "S",
+        "T",
+        "RX",
+        "RY",
+        "RZ",
+        "PauliX",
+        "C(PauliX)",
+        "PauliY",
+        "C(PauliY)",
+        "PauliZ",
+        "C(PauliZ)",
+        "Hadamard",
+        "SX",
+        "PhaseShift",
+        "U3",
+        "Rot",
+    }
+
+    config_filepath = pathlib.Path(
+        os.path.dirname(sys.modules[__name__].__file__) + "/QrackAceDeviceConfig.toml"
+    )
+
+    # Use "hybrid" stabilizer optimization? (Default is "true"; non-Clifford circuits will fall back to near-Clifford or universal simulation)
+    isStabilizerHybrid = False
+    # Use "tensor network" optimization? (Default is "false"; prevents dynamic qubit de-allocation; might function sub-optimally with "hybrid" stabilizer enabled)
+    isTensorNetwork = False
+    # Use Schmidt decomposition optimizations? (Default is "true")
+    isSchmidtDecompose = True
+    # Distribute Schmidt-decomposed qubit subsystems to multiple GPUs or accelerators, if available? (Default is "true"; mismatched device capacities might hurt overall performance)
+    isSchmidtDecomposeMulti = True
+    # Use "quantum binary decision diagram" ("QBDD") methods? (Default is "false"; note that QBDD is CPU-only)
+    isBinaryDecisionTree = False
+    # Use GPU acceleration? (Default is "true")
+    isOpenCL = True
+    # Use multi-GPU (or "multi-page") acceleration? (Default is "false")
+    isPaged = True
+    # Use CPU/GPU method hybridization? (Default is "false")
+    isCpuGpuHybrid = True
+    # Allocate GPU buffer from general host heap? (Default is "false"; "true" might improve performance or reliability in certain cases, like if using an Intel HD as accelerator)
+    isHostPointer = True if os.environ.get("PYQRACK_HOST_POINTER_DEFAULT_ON") else False
+    # Noise parameter. (Default is "0"; depolarizing noise intensity can also be controlled by "QRACK_GATE_DEPOLARIZATION" environment variable)
+    noise = 0
+    # How many full simulation columns, between border columns
+    long_range_columns = 4
+    # How many full simulation rows, between border rows
+    long_range_rows = 4
+    # Whether to transpose rows and columns
+    is_transpose = False
+
+    def __init__(self, wires=0, shots=None, **kwargs):
+        options = dict(kwargs)
+        if "isStabilizerHybrid" in options:
+            self.isStabilizerHybrid = options["isStabilizerHybrid"]
+        if "isTensorNetwork" in options:
+            self.isTensorNetwork = options["isTensorNetwork"]
+        if "isSchmidtDecompose" in options:
+            self.isSchmidtDecompose = options["isSchmidtDecompose"]
+        if "isBinaryDecisionTree" in options:
+            self.isBinaryDecisionTree = options["isBinaryDecisionTree"]
+        if "isOpenCL" in options:
+            self.isOpenCL = options["isOpenCL"]
+        if "isPaged" in options:
+            self.isPaged = options["isPaged"]
+        if "isCpuGpuHybrid" in options:
+            self.isCpuGpuHybrid = options["isCpuGpuHybrid"]
+        if "isHostPointer" in options:
+            self.isHostPointer = options["isHostPointer"]
+        if "noise" in options:
+            self.noise = options["noise"]
+            if (self.noise != 0) and (shots is None):
+                raise ValueError("Shots must be finite for noisy simulation (not analytical mode).")
+        if "long_range_columns" in options:
+            self.long_range_columns = options["long_range_columns"]
+        if "long_range_rows" in options:
+            self.long_range_rows = options["long_range_rows"]
+        if "is_transpose" in options:
+            self.is_transpose = options["is_transpose"]
+
+        super().__init__(wires=wires, shots=shots)
+        self.shots = shots
+        self._state = QrackAceBackend(
+            self.num_wires,
+            long_range_columns=self.long_range_columns,
+            long_range_rows=self.long_range_rows,
+            is_transpose=self.is_transpose,
+            isStabilizerHybrid=self.isStabilizerHybrid,
+            isTensorNetwork=self.isTensorNetwork,
+            isSchmidtDecompose=self.isSchmidtDecompose,
+            isBinaryDecisionTree=self.isBinaryDecisionTree,
+            isOpenCL=self.isOpenCL,
+            isCpuGpuHybrid=self.isCpuGpuHybrid,
+            isHostPointer=self.isHostPointer,
+            noise=self.noise,
+        )
+        self.device_kwargs = {
+            "long_range_columns": self.long_range_columns,
+            "long_range_rows": self.long_range_rows,
+            "is_transpose": self.is_transpose,
+            "is_hybrid_stabilizer": self.isStabilizerHybrid,
+            "is_tensor_network": self.isTensorNetwork,
+            "is_schmidt_decompose": self.isSchmidtDecompose,
+            "is_schmidt_decompose_parallel": self.isSchmidtDecomposeMulti,
+            "is_qpdd": self.isBinaryDecisionTree,
+            "is_gpu": self.isOpenCL,
+            "is_paged": self.isPaged,
+            "is_hybrid_cpu_gpu": self.isCpuGpuHybrid,
+            "is_host_pointer": self.isHostPointer,
+            "noise": self.noise,
+        }
+        self._circuit = []
+
+    def _reverse_state(self):
+        end = self.num_wires - 1
+        mid = self.num_wires >> 1
+        for i in range(mid):
+            self._state.swap(i, end - i)
+
+    def apply(self, operations, **kwargs):
+        """Apply the circuit operations to the state.
+
+        This method serves as an auxiliary method to :meth:`~.QrackDevice.apply`.
+
+        Args:
+            operations (List[pennylane.Operation]): operations to be applied
+        """
+
+        self._circuit = self._circuit + operations
+        if self.noise == 0:
+            self._apply()
+            self._circuit = []
+        # else: Defer application until shots or expectation values are requested
+
+    def _apply(self):
+        for op in self._circuit:
+            if isinstance(op, BasisState):
+                self._apply_basis_state(op)
+            else:
+                self._apply_gate(op)
+
+    def _apply_basis_state(self, op):
+        """Initialize a basis state"""
+        wires = self.map_wires(Wires(op.wires))
+        par = op.parameters[0]
+        wire_count = len(wires)
+        n_basis_state = len(par)
+
+        if not set(par).issubset({0, 1}):
+            raise ValueError("BasisState parameter must consist of 0 or 1 integers.")
+        if n_basis_state != wire_count:
+            raise ValueError("BasisState parameter and wires must be of equal length.")
+
+        for i in range(wire_count):
+            index = wires.labels[i]
+            if par[i] != self._state.m(index):
+                self._state.x(index)
+
+    def _apply_gate(self, op):
+        """Apply native qrack gate"""
+
+        opname = op.name
+        if isinstance(op, Adjoint):
+            op = op.base
+            opname = op.name + ".inv"
+
+        par = op.parameters
+
+        if opname == "MultiRZ":
+            device_wires = self.map_wires(op.wires)
+            for q in device_wires:
+                self._state.r(Pauli.PauliZ, par[0], q)
+            return
+
+        # translate op wire labels to consecutive wire labels used by the device
+        device_wires = self.map_wires(
+            (op.control_wires + op.wires) if op.control_wires else op.wires
+        )
+
+        if opname in [
+            "".join(p)
+            for p in it.product(
+                [
+                    "CNOT",
+                    "C(PauliX)",
+                ],
+                ["", ".inv"],
+            )
+        ]:
+            self._state.mcx(device_wires.labels[:-1], device_wires.labels[-1])
+        elif opname in ["C(PauliY)", "C(PauliY).inv"]:
+            self._state.mcy(device_wires.labels[:-1], device_wires.labels[-1])
+        elif opname in ["C(PauliZ)", "C(PauliZ).inv"]:
+            self._state.mcz(device_wires.labels[:-1], device_wires.labels[-1])
+        elif opname in ["SWAP", "SWAP.inv"]:
+            self._state.swap(device_wires.labels[0], device_wires.labels[1])
+        elif opname == "ISWAP":
+            self._state.iswap(device_wires.labels[0], device_wires.labels[1])
+        elif opname == "ISWAP.inv":
+            self._state.adjiswap(device_wires.labels[0], device_wires.labels[1])
+        elif opname in ["CY", "CY.inv", "C(CY)", "C(CY).inv"]:
+            self._state.mcy(device_wires.labels[:-1], device_wires.labels[-1])
+        elif opname in ["CZ", "CZ.inv", "C(CZ)", "C(CZ).inv"]:
+            self._state.mcz(device_wires.labels[:-1], device_wires.labels[-1])
+        elif opname == "S":
+            for label in device_wires.labels:
+                self._state.s(label)
+        elif opname == "S.inv":
+            for label in device_wires.labels:
+                self._state.adjs(label)
+        elif opname == "T":
+            for label in device_wires.labels:
+                self._state.t(label)
+        elif opname == "T.inv":
+            for label in device_wires.labels:
+                self._state.adjt(label)
+        elif opname == "RX":
+            for label in device_wires.labels:
+                self._state.r(Pauli.PauliX, par[0], label)
+        elif opname == "RX.inv":
+            for label in device_wires.labels:
+                self._state.r(Pauli.PauliX, -par[0], label)
+        elif opname == "RY":
+            for label in device_wires.labels:
+                self._state.r(Pauli.PauliY, par[0], label)
+        elif opname == "RY.inv":
+            for label in device_wires.labels:
+                self._state.r(Pauli.PauliY, -par[0], label)
+        elif opname == "RZ":
+            for label in device_wires.labels:
+                self._state.r(Pauli.PauliZ, par[0], label)
+        elif opname == "RZ.inv":
+            for label in device_wires.labels:
+                self._state.r(Pauli.PauliZ, -par[0], label)
+        elif opname in ["PauliX", "PauliX.inv"]:
+            for label in device_wires.labels:
+                self._state.x(label)
+        elif opname in ["PauliY", "PauliY.inv"]:
+            for label in device_wires.labels:
+                self._state.y(label)
+        elif opname in ["PauliZ", "PauliZ.inv"]:
+            for label in device_wires.labels:
+                self._state.z(label)
+        elif opname in ["Hadamard", "Hadamard.inv"]:
+            for label in device_wires.labels:
+                self._state.h(label)
+        elif opname == "SX":
+            half_pi = math.pi / 2
+            for label in device_wires.labels:
+                self._state.u(label, half_pi, -half_pi, half.pi)
+        elif opname == "SX.inv":
+            half_pi = math.pi / 2
+            for label in device_wires.labels:
+                self._state.u(label, -half_pi, -half.pi, half_pi)
+        elif opname == "PhaseShift":
+            half_par = par[0] / 2
+            for label in device_wires.labels:
+                self._state.u(label, 0, half_par, half_par)
+        elif opname == "PhaseShift.inv":
+            half_par = par[0] / 2
+            for label in device_wires.labels:
+                self._state.u(label, 0, -half_par, -half_par)
+        elif opname == "U3":
+            for label in device_wires.labels:
+                self._state.u(label, par[0], par[1], par[2])
+        elif opname == "U3.inv":
+            for label in device_wires.labels:
+                self._state.u(label, -par[0], -par[2], -par[1])
+        elif opname == "Rot":
+            for label in device_wires.labels:
+                self._state.r(Pauli.PauliZ, par[0], label)
+                self._state.r(Pauli.PauliY, par[1], label)
+                self._state.r(Pauli.PauliZ, par[2], label)
+        elif opname == "Rot.inv":
+            for label in device_wires.labels:
+                self._state.r(Pauli.PauliZ, -par[2], label)
+                self._state.r(Pauli.PauliY, -par[1], label)
+                self._state.r(Pauli.PauliZ, -par[0], label)
+        elif opname not in [
+            "Identity",
+            "Identity.inv",
+            "C(Identity)",
+            "C(Identity).inv",
+        ]:
+            raise DeviceError(f"Operation {opname} is not supported on a {self.short_name} device.")
+
+    def analytic_probability(self, wires=None):
+        raise DeviceError(
+            f"analytic_probability is not supported on a {self.short_name} device. (Specify a finite number of shots, instead.)"
+        )
+
+    def expval(self, observable, **kwargs):
+        if self.shots is None:
+            if isinstance(observable.name, list):
+                b = [self._observable_map[obs] for obs in observable.name]
+            elif observable.name == "Prod":
+                b = [self._observable_map[obs.name] for obs in observable.operands]
+            else:
+                b = [self._observable_map[observable.name]]
+
+            # exact expectation value
+            if callable(observable.eigvals):
+                eigvals = self._asarray(observable.eigvals(), dtype=self.R_DTYPE)
+            else:  # older version of pennylane
+                eigvals = self._asarray(observable.eigvals, dtype=self.R_DTYPE)
+            prob = self.probability(wires=observable.wires)
+            return self._dot(eigvals, prob)
+
+        # estimate the ev
+        return np.mean(self.sample(observable))
+
+    def _generate_sample(self):
+        rev_sample = self._state.m_all()
+        sample = 0
+        for i in range(self.num_wires):
+            if (rev_sample & (1 << i)) > 0:
+                sample |= 1 << (self.num_wires - (i + 1))
+        return sample
+
+    def generate_samples(self):
+        if self.shots is None:
+            raise QuantumFunctionError(
+                "The number of shots has to be explicitly set on the device "
+                "when using sample-based measurements."
+            )
+
+        if self.noise != 0:
+            samples = []
+            for _ in range(self.shots):
+                self._state.reset_all()
+                self._apply()
+                samples.append(self._generate_sample())
+            self._samples = QubitDevice.states_to_binary(np.array(samples), self.num_wires)
+            self._circuit = []
+
+            return self._samples
+
+        if self.shots == 1:
+            self._samples = QubitDevice.states_to_binary(
+                np.array([self._generate_sample()]), self.num_wires
+            )
+
+            return self._samples
+
+        # QubitDevice.states_to_binary() doesn't work for >64qb. (Fix by Elara, the custom OpenAI GPT)
+        samples = self._state.measure_shots(list(range(self.num_wires - 1, -1, -1)), self.shots)
+        self._samples = np.array(
+            [list(format(b, f"0{self.num_wires}b")) for b in samples], dtype=np.int8
+        )
+
+        return self._samples
+
+    def reset(self):
+        for i in range(self.num_wires):
+            if self._state.m(i):
+                self._state.x(i)