Support for Autoqasm -> QIR

pyproject.toml

@@ -37,7 +37,7 @@ Discord = "https://discord.gg/TPBU2sa8Et"
 [project.optional-dependencies]
 cirq = ["cirq-core>=1.3.0,<1.5.0"]
 qasm3 = ["openqasm3[parser]>=0.4.0,<1.1.0"]
-test = ["qbraid>=0.7.1,<0.8.0", "pytest", "pytest-cov"]
+test = ["qbraid>=0.7.1,<0.8.0", "pytest", "pytest-cov", "autoqasm>=0.1.0"]
 lint = ["black[jupyter]", "isort", "pylint", "qbraid-cli>=0.8.3"]
 docs = ["sphinx>=7.3.7,<7.5.0", "sphinx-autodoc-typehints>=1.24,<2.3", "sphinx-rtd-theme~=2.0.0", "docutils<0.22", "sphinx-copybutton"]
 

qbraid_qir/__init__.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/_version.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/cirq/__init__.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/cirq/convert.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/cirq/elements.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/cirq/exceptions.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/cirq/opsets.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/cirq/passes.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/cirq/visitor.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/exceptions.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/qasm3/__init__.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/qasm3/convert.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/qasm3/elements.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/qasm3/exceptions.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2023 qBraid
+# Copyright (C) 2024 qBraid
 #
 # This file is part of the qBraid-SDK
 #

qbraid_qir/qasm3/linalg.py

@@ -0,0 +1,302 @@
+# Copyright (C) 2024 qBraid
+#
+# This file is part of the qBraid-SDK
+#
+# The qBraid-SDK is free software released under the GNU General Public License v3
+# or later. You can redistribute and/or modify it under the terms of the GPL v3.
+# See the LICENSE file in the project root or <https://www.gnu.org/licenses/gpl-3.0.html>.
+#
+# THERE IS NO WARRANTY for the qBraid-SDK, as per Section 15 of the GPL v3.
+
+# pylint: disable=too-many-locals
+
+"""
+Module for linear algebra functions necessary for gate decomposition.
+
+"""
+
+import cmath
+import functools
+import math
+
+import numpy as np
+
+MAGIC = np.array([[1, 0, 0, 1j], [0, 1j, 1, 0], [0, 1j, -1, 0], [1, 0, 0, -1j]]) * np.sqrt(0.5)
+
+MAGIC_CONJ_T = np.conj(MAGIC.T)
+
+KAK_MAGIC = np.array([[1, 0, 0, 1j], [0, 1j, 1, 0], [0, 1j, -1, 0], [1, 0, 0, -1j]]) * np.sqrt(0.5)
+
+KAK_MAGIC_DAG = np.conjugate(np.transpose(KAK_MAGIC))
+
+KAK_GAMMA = np.array([[1, 1, 1, 1], [1, 1, -1, -1], [-1, 1, -1, 1], [1, -1, -1, 1]]) * 0.25
+
+
+def _helper_svd(mat):
+    """
+    Helper function to perform SVD on a matrix.
+    """
+    if mat.size == 0:
+        return np.zeros((0, 0), dtype=mat.dtype), np.array([]), np.zeros((0, 0), dtype=mat.dtype)
+    return np.linalg.svd(mat)
+
+
+def _merge_dtypes(dtype1, dtype2):
+    return (np.zeros(0, dtype1) + np.zeros(0, dtype2)).dtype
+
+
+def _block_diag(*blocks):
+    """
+    Helper function to perform block diagonalization.
+    """
+    n = sum(b.shape[0] for b in blocks)
+    dtype = functools.reduce(_merge_dtypes, (b.dtype for b in blocks))
+
+    result = np.zeros((n, n), dtype=dtype)
+    i = 0
+    for b in blocks:
+        j = i + b.shape[0]
+        result[i:j, i:j] = b
+        i = j
+
+    return result
+
+
+def _orthogonal_diagonalize(symmetric_matrix, diagonal_matrix):
+    """
+    Find orthogonal matrix that diagonalize mat1 and mat2.
+    """
+
+    def similar_singular(i, j):
+        return np.allclose(diagonal_matrix[i, i], diagonal_matrix[j, j])
+
+    ranges = []
+    start = 0
+    while start < diagonal_matrix.shape[0]:
+        past = start + 1
+        while past < diagonal_matrix.shape[0] and similar_singular(start, past):
+            past += 1
+        ranges.append((start, past))
+        start = past
+
+    p = np.zeros(symmetric_matrix.shape, dtype=np.float64)
+    for start, end in ranges:
+        block = symmetric_matrix[start:end, start:end]
+        _, res = np.linalg.eigh(block)
+        p[start:end, start:end] = res
+
+    return p
+
+
+def orthogonal_bidiagonalize(mat1, mat2):
+    """
+    Find orthogonal matrices that diagonalize mat1 and mat2.
+    """
+    atol = 1e-9
+    base_left, base_diag, base_right = _helper_svd(mat1)
+    base_diag = np.diag(base_diag)
+
+    dim = base_diag.shape[0]
+    rank = dim
+    while rank > 0 and np.all(np.less_equal(np.abs(base_diag[rank - 1, rank - 1]), atol)):
+        rank -= 1
+    base_diag = base_diag[:rank, :rank]
+
+    semi_corrected = np.dot(np.dot(base_left.T, np.real(mat2)), base_right.T)
+
+    overlap = semi_corrected[:rank, :rank]
+    overlap_adjust = _orthogonal_diagonalize(overlap, base_diag)
+
+    extra = semi_corrected[rank:, rank:]
+    extra_left_adjust, _, extra_right_adjust = _helper_svd(extra)
+
+    left_adjust = _block_diag(overlap_adjust, extra_left_adjust)
+    right_adjust = _block_diag(overlap_adjust.T, extra_right_adjust)
+    left = np.dot(left_adjust.T, base_left.T)
+    right = np.dot(base_right.T, right_adjust.T)
+    return left, right
+
+
+def _kronecker_fator(mat):
+    """
+    Split U = kron(A, B) to A and B.
+    """
+    a, b = max(((i, j) for i in range(4) for j in range(4)), key=lambda t: abs(mat[t]))
+
+    f1 = np.zeros((2, 2), dtype=mat.dtype)
+    f2 = np.zeros((2, 2), dtype=mat.dtype)
+    for i in range(2):
+        for j in range(2):
+            f1[(a >> 1) ^ i, (b >> 1) ^ j] = mat[a ^ (i << 1), b ^ (j << 1)]
+            f2[(a & 1) ^ i, (b & 1) ^ j] = mat[a ^ i, b ^ j]
+
+    with np.errstate(divide="ignore", invalid="ignore"):
+        f1 /= np.sqrt(np.linalg.det(f1)) or 1
+        f2 /= np.sqrt(np.linalg.det(f2)) or 1
+
+    g = mat[a, b] / (f1[a >> 1, b >> 1] * f2[a & 1, b & 1])
+    if np.real(g) < 0:
+        f1 *= -1
+        g *= -1
+
+    return g, f1, f2
+
+
+def _so4_to_su2(mat):
+    """
+    Decompose SO(4) matrix to SU(2) matrices.
+    """
+    ab = np.dot(np.dot(MAGIC, mat), MAGIC_CONJ_T)
+    _, a, b = _kronecker_fator(ab)
+
+    return a, b
+
+
+def _kak_canonicalize_vector(x, y, z):
+    """
+    Canonicalize vector for KAK decomposition.
+    """
+    phase = [complex(1)]
+    left = [np.eye(2)] * 2
+    right = [np.eye(2)] * 2
+    v = [x, y, z]
+
+    flippers = [
+        np.array([[0, 1], [1, 0]]) * 1j,
+        np.array([[0, -1j], [1j, 0]]) * 1j,
+        np.array([[1, 0], [0, -1]]) * 1j,
+    ]
+
+    swappers = [
+        np.array([[1, -1j], [1j, -1]]) * 1j * np.sqrt(0.5),
+        np.array([[1, 1], [1, -1]]) * 1j * np.sqrt(0.5),
+        np.array([[0, 1 - 1j], [1 + 1j, 0]]) * 1j * np.sqrt(0.5),
+    ]
+
+    def shift(k, step):
+        v[k] += step * np.pi / 2
+        phase[0] *= 1j**step
+        right[0] = np.dot(flippers[k] ** (step % 4), right[0])
+        right[1] = np.dot(flippers[k] ** (step % 4), right[1])
+
+    def negate(k1, k2):
+        v[k1] *= -1
+        v[k2] *= -1
+        phase[0] *= -1
+        s = flippers[3 - k1 - k2]
+        left[1] = np.dot(left[1], s)
+        right[1] = np.dot(s, right[1])
+
+    def swap(k1, k2):
+        v[k1], v[k2] = v[k2], v[k1]
+        s = swappers[3 - k1 - k2]
+        left[0] = np.dot(left[0], s)
+        left[1] = np.dot(left[1], s)
+        right[0] = np.dot(s, right[0])
+        right[1] = np.dot(s, right[1])
+
+    def canonical_shift(k):
+        while v[k] <= -np.pi / 4:
+            shift(k, +1)
+        while v[k] > np.pi / 4:
+            shift(k, -1)
+
+    def sort():
+        if abs(v[0]) < abs(v[1]):
+            swap(0, 1)
+        if abs(v[1]) < abs(v[2]):
+            swap(1, 2)
+        if abs(v[0]) < abs(v[1]):
+            swap(0, 1)
+
+    canonical_shift(0)
+    canonical_shift(1)
+    canonical_shift(2)
+    sort()
+
+    if v[0] < 0:
+        negate(0, 2)
+    if v[1] < 0:
+        negate(1, 2)
+    canonical_shift(2)
+
+    atol = 1e-9
+    if v[0] > np.pi / 4 - atol and v[2] < 0:
+        shift(0, -1)
+        negate(0, 2)
+
+    return {
+        "single_qubit_operations_after": (left[1], left[0]),
+        "single_qubit_operations_before": (right[1], right[0]),
+    }
+
+
+def _deconstruct_matrix_to_angles(mat):
+    """
+    Decompose matrix into angles.
+    """
+
+    def _phase_matrix(angle):
+        return np.diag([1, np.exp(1j * angle)])
+
+    def _rotation_matrix(angle):
+        c, s = np.cos(angle), np.sin(angle)
+        return np.array([[c, -s], [s, c]])
+
+    right_phase = cmath.phase(mat[0, 1] * np.conj(mat[0, 0])) + np.pi
+    mat = np.dot(mat, _phase_matrix(-right_phase))
+
+    bottom_phase = cmath.phase(mat[1, 0] * np.conj(mat[0, 0]))
+    mat = np.dot(_phase_matrix(-bottom_phase), mat)
+
+    rotation = math.atan2(abs(mat[1, 0]), abs(mat[0, 0]))
+    mat = np.dot(_rotation_matrix(-rotation), mat)
+
+    diagonal_phase = cmath.phase(mat[1, 1] * np.conj(mat[0, 0]))
+
+    return right_phase + diagonal_phase, rotation * 2, bottom_phase
+
+
+def so_bidiagonalize(mat):
+    """
+    Find special orthogonal L and R so that L @ mat @ R is diagonal.
+    """
+    left, right = orthogonal_bidiagonalize(np.real(mat), np.imag(mat))
+    with np.errstate(divide="ignore", invalid="ignore"):
+        if np.linalg.det(left) < 0:
+            left[0, :] *= -1
+        if np.linalg.det(right) < 0:
+            right[:, 0] *= -1
+
+    diag = np.dot(np.dot(left, mat), right)
+
+    return left, np.diag(diag), right
+
+
+def kak_decomposition_angles(mat):
+    """
+    Decompose matrix into KAK decomposition, return all angles.
+    """
+    left, d, right = so_bidiagonalize(KAK_MAGIC_DAG @ mat @ KAK_MAGIC)
+
+    a1, a0 = _so4_to_su2(left.T)
+    b1, b0 = _so4_to_su2(right.T)
+    _, x, y, z = (KAK_GAMMA @ np.angle(d).reshape(-1, 1)).flatten()
+
+    inner_cannon = _kak_canonicalize_vector(x, y, z)
+    b1 = np.dot(inner_cannon["single_qubit_operations_before"][0], b1)
+    b0 = np.dot(inner_cannon["single_qubit_operations_before"][1], b0)
+    a1 = np.dot(a1, inner_cannon["single_qubit_operations_after"][0])
+    a0 = np.dot(a0, inner_cannon["single_qubit_operations_after"][1])
+
+    pre_phase00, rotation00, post_phase00 = _deconstruct_matrix_to_angles(b1)
+    pre_phase01, rotation01, post_phase01 = _deconstruct_matrix_to_angles(b0)
+    pre_phase10, rotation10, post_phase10 = _deconstruct_matrix_to_angles(a1)
+    pre_phase11, rotation11, post_phase11 = _deconstruct_matrix_to_angles(a0)
+
+    return [
+        [rotation00, post_phase00, pre_phase00],
+        [rotation01, post_phase01, pre_phase01],
+        [rotation10, post_phase10, pre_phase10],
+        [rotation11, post_phase11, pre_phase11],
+    ]