Implement transform for full single-qubit gate fusion #1458

glassnotes · 2021-07-15T14:32:46Z

Context: Addition of circuit optimization / compilation tools to PennyLane.

Description of the Change: Adds a new transform that merges arbitrary-length sequences of adjacent single-qubit gates. Adds related tests, and an additional attribute to express single-qubit gates as three rotation angles.

Benefits: Can significantly reduce the length of circuits by squashing any number of single-qubit gates down to one (technically, three, but a single Rot).

Possible Drawbacks: None

Related GitHub Issues: None

…ndant

…ps://github.com/PennyLaneAI/pennylane into ch7104-implement-transforms-to-cancel-redundant

…ndant

Co-authored-by: Josh Izaac <josh146@gmail.com>

…trol wires case.

* tests/transforms/test_optimization/utils.py:

…ps://github.com/PennyLaneAI/PennyLane into ch7104-implement-transforms-to-cancel-redundant

github-actions · 2021-07-15T14:33:06Z

Hello. You may have forgotten to update the changelog!
Please edit .github/CHANGELOG.md with:

A one-to-two sentence description of the change. You may include a small working example for new features.
A link back to this PR.
Your name (or GitHub username) in the contributors section.

josh146

🚤

pennylane/transforms/optimization/single_qubit_fusion.py

josh146 · 2021-07-16T13:21:49Z

pennylane/transforms/optimization/single_qubit_fusion.py

+            # Apply the Rot equivalent of the gate, for consistency
+            Rot(*cumulative_angles, wires=current_gate.wires)


Oh interesting --- wouldn't it be better to leave the gates as they are? E.g., CNOT rather than Rot(pi/4, 0, 0)?

Because if the circuit is to end up on hardware, something like a CNOT will be easier to implement than Rot(pi/4, 0, 0) (even though they are equivalent)

We could do either! I was just looking at the docs and realized I'd written "one or more gates gets fused to Rot", so to match I turned everything to Rot. But if it's easier to implement just the regular gates, I'll adjust the docstring and revert this back to apply(current_gate).

josh146 · 2021-07-16T13:23:31Z

pennylane/transforms/optimization/single_qubit_fusion.py

@@ -109,7 +130,7 @@ def qfunc(r1, r2):
            next_gate_idx = find_next_gate(current_gate.wires, list_copy[1:])

        # Only apply if the cumulative angle is not close to 0
-        if not allclose(cumulative_angles, zeros(3), atol=tol):
+        if not allclose(cumulative_angles, zeros(3), atol=atol):


Do you want to set rtol=0 here? Otherwise, there remains a relative tolerance check in addition to the absolute tolerance check happening behind the scenes

Alternatively, you could allow the user to pass atol and rtol as keyword arguments for full control 😆

Let's set rtol=0. I'm just thinking intuitively, I would be looking for absolute closeness to 0.

Co-authored-by: Josh Izaac <josh146@gmail.com>

…h7105-implement-transform-for-full-single-qubit

pennylane/transforms/optimization/single_qubit_fusion.py

…-qubit

…r excluded gates.

…tps://github.com/PennyLaneAI/PennyLane into ch7105-implement-transform-for-full-single-qubit

…-qubit

josh146

Looks great @glassnotes! I think the only thing still missing are tests for the single_qubit_rot_angles methods added to all the operations. Is there a clean way to easily test that they have the correct angles hardcoded?

This is a small thing that could easily be hiding bugs

josh146 · 2021-07-19T16:45:13Z

pennylane/transforms/optimization/single_qubit_fusion.py

+# 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.
+"""Transforms for optimizing quantum circuits."""


Should probably make this slightly more descriptive I suppose

Oops, good catch!

josh146 · 2021-07-19T16:46:23Z

pennylane/transforms/optimization/single_qubit_fusion.py

+        if next_gate_idx is None:
+            apply(current_gate)
+            list_copy.pop(0)
+            continue


Looks like this is missing coverage?

josh146 · 2021-07-19T16:46:39Z

pennylane/transforms/optimization/single_qubit_fusion.py

+            if exclude_gates is not None:
+                if next_gate.name in exclude_gates:
+                    break


…tps://github.com/PennyLaneAI/PennyLane into ch7105-implement-transform-for-full-single-qubit

glassnotes · 2021-07-19T22:32:57Z

Looks great @glassnotes! I think the only thing still missing are tests for the single_qubit_rot_angles methods added to all the operations. Is there a clean way to easily test that they have the correct angles hardcoded?

This is a small thing that could easily be hiding bugs

@josh146 thanks for pointing that out, have added a small test for this in the test_qubit_ops file! Think everything else should be covered now with the most recent updates to tests.

josh146

Thanks @glassnotes!

josh146 · 2021-07-20T07:11:33Z

tests/ops/test_qubit_ops.py

+    @pytest.mark.parametrize(
+        "op",
+        [
+            (qml.Hadamard(wires=0)),
+            (qml.PauliX(wires=0)),
+            (qml.PauliY(wires=0)),
+            (qml.PauliZ(wires=0)),
+            (qml.S(wires=0)),
+            (qml.T(wires=0)),
+            (qml.SX(wires=0)),
+            (qml.RX(0.3, wires=0)),
+            (qml.RY(0.3, wires=0)),
+            (qml.RZ(0.3, wires=0)),
+            (qml.PhaseShift(0.3, wires=0)),
+            (qml.Rot(0.3, 0.4, 0.5, wires=0)),
+        ],
+    )
+    def test_single_qubit_rot_angles(self, op):
+        """Tests that the Rot gates yielded by single_qubit_rot_angles
+        are equivalent to the true operations up to a global phase."""
+        angles = op.single_qubit_rot_angles()
+        obtained_mat = qml.Rot(*angles, wires=0).matrix
+
+        # Check whether the two matrices are each others conjugate transposes
+        mat_product = qml.math.dot(op.matrix, qml.math.conj(obtained_mat.T))
+        mat_product /= mat_product[0, 0]
+
+        assert qml.math.allclose(mat_product, I)


…-qubit

Olivia Di Matteo and others added 29 commits July 8, 2021 12:51

Copy over some files from 1384.

0406efc

Add optimization utils file.

8560c1d

Add updated qubit ops file.

68f9e79

Update init file.

09a3ae1

Streamline utils file.

eba5f51

Update operation.py.

5781743

Add arctan, and add JAX 64-bit support to test.

7a21d41

Remove as_rot_angles.

7118152

Update docstrings.

b156b70

Merge branch 'master' into ch7104-implement-transforms-to-cancel-redu…

0a9abe9

…ndant

Merge branch 'ch7104-implement-transforms-to-cancel-redundant' of htt…

c99f3cd

…ps://github.com/PennyLaneAI/pennylane into ch7104-implement-transforms-to-cancel-redundant

Shorten variable names for better readability.

76947be

Update default values of compilation tags.

2b04ec2

Update CHANGELOG.

95a61a9

Add new transforms to docs.

af55d17

Merge branch 'master' into ch7104-implement-transforms-to-cancel-redu…

86f4f5e

…ndant

Apply suggestions from code review

2f78671

Co-authored-by: Josh Izaac <josh146@gmail.com>

Add support for cancelling adjacent adjoint gates.

0f20fa9

Improve docstrings for compilation attributes. Add symmetric over con…

c26d14b

…trol wires case.

Adjust Toffoli test case.

2ecb449

Convert utility functions to non-private.

f7fb6c7

* tests/transforms/test_optimization/utils.py:

Add tests for utility functions.

dd73732

Add optional parameters for rotation merging.

a783181

Streamline angle fusion for 0 cases.

2b29cf0

Disable warning for too many return statements.

82688ac

Apply suggestions from code review

be50d8d

Add tests for optimization utils.

d960e87

Merge branch 'ch7104-implement-transforms-to-cancel-redundant' of htt…

7ce969f

…ps://github.com/PennyLaneAI/PennyLane into ch7104-implement-transforms-to-cancel-redundant

Add transform for full single-qubit fusion.

20198e7

Rename specify_ops to include_gates.

5fbd0e4

glassnotes requested a review from josh146 July 16, 2021 13:13

Remove spurious print statement.

2f741fd

josh146 requested changes Jul 16, 2021

View reviewed changes

glassnotes and others added 5 commits July 16, 2021 09:30

Apply suggestions from code review

b472282

Co-authored-by: Josh Izaac <josh146@gmail.com>

Set rtol and clarify fusion is for two or more gates in a row.

056350e

Fix test.

df6b181

Update handling of tolerance.

c39c36a

Merge branch 'ch7104-implement-transforms-to-cancel-redundant' into c…

b282724

…h7105-implement-transform-for-full-single-qubit

glassnotes commented Jul 16, 2021

View reviewed changes

pennylane/transforms/optimization/single_qubit_fusion.py Outdated Show resolved Hide resolved

Update pennylane/transforms/optimization/single_qubit_fusion.py

4e748dd

glassnotes commented Jul 16, 2021

View reviewed changes

pennylane/transforms/optimization/single_qubit_fusion.py Outdated Show resolved Hide resolved

Update pennylane/transforms/optimization/single_qubit_fusion.py

5c9b4db

Base automatically changed from ch7104-implement-transforms-to-cancel-redundant to master July 19, 2021 04:09

josh146 and others added 5 commits July 19, 2021 15:02

Merge branch 'master' into ch7105-implement-transform-for-full-single…

ac052ee

…-qubit

Fix codefactor issues. Simplify some logic and add final test case fo…

7993658

…r excluded gates.

Merge branch 'ch7105-implement-transform-for-full-single-qubit' of ht…

6ce3f82

…tps://github.com/PennyLaneAI/PennyLane into ch7105-implement-transform-for-full-single-qubit

Disable too-many-branches.

449b343

Merge branch 'master' into ch7105-implement-transform-for-full-single…

0fc444b

…-qubit

glassnotes requested a review from josh146 July 19, 2021 14:12

glassnotes mentioned this pull request Jul 19, 2021

Add a transform get_unitary_matrix to obtain matrix representations of circuits #1462

Closed

josh146 requested changes Jul 19, 2021

View reviewed changes

glassnotes added 3 commits July 19, 2021 18:08

Update docstring. Add tests for single_qubit_rot_angles.

ce19bdd

Merge branch 'ch7105-implement-transform-for-full-single-qubit' of ht…

fe64b63

…tps://github.com/PennyLaneAI/PennyLane into ch7105-implement-transform-for-full-single-qubit

Add missing tests to complete coverage.

f94ea21

josh146 approved these changes Jul 20, 2021

View reviewed changes

Merge branch 'master' into ch7105-implement-transform-for-full-single…

142ab05

…-qubit

glassnotes merged commit 2b451de into master Jul 20, 2021

glassnotes deleted the ch7105-implement-transform-for-full-single-qubit branch July 20, 2021 13:22

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implement transform for full single-qubit gate fusion #1458

Implement transform for full single-qubit gate fusion #1458

glassnotes commented Jul 15, 2021 •

edited

Loading

github-actions bot commented Jul 15, 2021

josh146 left a comment

josh146 Jul 16, 2021

josh146 Jul 16, 2021

glassnotes Jul 16, 2021

josh146 Jul 16, 2021

josh146 Jul 16, 2021

glassnotes Jul 16, 2021

josh146 left a comment

josh146 Jul 19, 2021

glassnotes Jul 19, 2021

josh146 Jul 19, 2021

josh146 Jul 19, 2021

glassnotes commented Jul 19, 2021

josh146 left a comment

josh146 Jul 20, 2021

		# Apply the Rot equivalent of the gate, for consistency
		Rot(*cumulative_angles, wires=current_gate.wires)

Implement transform for full single-qubit gate fusion #1458

Implement transform for full single-qubit gate fusion #1458

Conversation

glassnotes commented Jul 15, 2021 • edited Loading

github-actions bot commented Jul 15, 2021

josh146 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

josh146 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

glassnotes commented Jul 19, 2021

josh146 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

glassnotes commented Jul 15, 2021 •

edited

Loading