🎨 Update of IonQ and Rigetti Devices #570

nquetschlich · 2025-05-22T14:56:13Z

This PR...

updates the IonQ device calibration data
introduces new devices: IonQ Forte 1 and Rigetti Ankaa 3
introduces a more correct gatesets for these devices

resolves #289

burgholzer

Thanks for the first draft here.
I just rather quickly went over everything here and added a couple of inline comments.
This is going into the right direction but, as expected, will take a little more work to get right.

burgholzer · 2025-05-22T15:19:39Z

src/mqt/bench/targets/devices/__init__.py

+        get_ionq_target("ionq_forte_1"),
+        get_ionq_target("ionq_aria_1"),


I believe the naming of these should be adjusted so that the last number indicates the number of qubits similar to IBM and IQM. The 1 here does not add any value

These are the specific device names and there is also an Aria 2. Therefore, I suggest to keep the name, but this is independent of the qubit number at the end - if you prefer that, I am happy to add it to all devices.

But aria 2 still uses the same layout and the same gateset if I am not mistaken.
Given how the device data is averaged anyway, I do not think there is too much value to this number.

If the gateset should really be different between these, then I would actually add the number to the gateset name and still append an underscore with the number of qubits for the device here.

Same holds for the Rigetti device.

src/mqt/bench/targets/devices/ionq.py

codecov · 2025-05-23T07:17:52Z

Codecov Report

Attention: Patch coverage is 96.96970% with 8 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/mqt/bench/targets/gatesets/rigetti.py	92.8%	4 Missing ⚠️
src/mqt/bench/targets/devices/ionq.py	91.3%	2 Missing ⚠️
src/mqt/bench/targets/gatesets/__init__.py	96.2%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!

burgholzer

Thanks for all the changes. This is definitely moving in the right direction. I hope the new comments help to push this even closer to the finish line.

burgholzer · 2025-05-23T07:35:50Z

src/mqt/bench/targets/devices/ibm.py

@@ -18,7 +18,7 @@

 from qiskit.providers.fake_provider import GenericBackendV2

-from ..gatesets import get_ibm_eagle_gateset, get_ibm_falcon_gateset, get_ibm_heron_gateset
+from mqt.bench.targets.gatesets.ibm import get_ibm_eagle_gateset, get_ibm_falcon_gateset, get_ibm_heron_gateset


This change is unrelated to the PR and I actually would not make that change.

Adjusted it accordingly.

burgholzer · 2025-05-23T07:38:16Z

src/mqt/bench/targets/devices/ionq.py

 from qiskit.transpiler import InstructionProperties, Target

+from mqt.bench.targets.gatesets.ionq import GPI2Gate, GPIGate, MSGate, ZZGate, add_ionq_equivalences


I would have a rather strong preference for relative imports. Tends to work better with editable installs and IDEs.

Adjusted it accordingly.

burgholzer · 2025-05-23T07:44:39Z

src/mqt/bench/targets/devices/ionq.py

-        return get_ionq_aria1_target()
-    if device_name == "ionq_harmony":
-        return get_ionq_harmony_target()
+    add_ionq_equivalences()


I am not yet quite sure this is the best place to put this. As there are multiple IonQ devices, this will inevitably get called multiple times.
I could imagine that this literally duplicates all the rules, which doesn't sound good.

I think it would be good to ensure that this is only called once. And preferably, it would only be called when the target is used, not when it's constructed. We only want to add these right before the transpile call when the target is used.

I'd also like to understand a little bit more about the lifecycle of the SessionEquivalenceLibrary. How long does it persist? Where does it enter the equation in the transpile call? Would moving to the StagedPassManager allow us to use a custom EL?

burgholzer · 2025-05-23T07:52:22Z

src/mqt/bench/targets/gatesets/__init__.py

+        elif gate == "fixed_angle_rx_pi_by_2":
+            target.add_instruction(RXPiOver2Gate())
+        elif gate == "fixed_angle_rx_pi_by_minus_2":
+            target.add_instruction(RXMinusPiOver2Gate())


Slightly more consistent naming could be useful here:

Suggested change

elif gate == "fixed_angle_rx_pi_by_2":

target.add_instruction(RXPiOver2Gate())

elif gate == "fixed_angle_rx_pi_by_minus_2":

target.add_instruction(RXMinusPiOver2Gate())

elif gate == "rxpi2":

target.add_instruction(RXPI2Gate())

elif gate == "rxpi2dg":

target.add_instruction(RXPI2DgGate())

Or something along those lines.
The PI and PI2 in "gpi" and "gpi2" stand for exactly that.

I implemented that now.

burgholzer · 2025-05-23T07:56:12Z

tests/devices/test_ionq_device_support.py

@@ -13,33 +13,32 @@
 import pytest
 from qiskit.transpiler import Target

-from mqt.bench.targets.devices.ionq import get_ionq_target
+from mqt.bench.targets.devices.ionq import _build_ionq_target, get_ionq_target  # noqa: PLC2701


Not quite sure whether I like that this is importing a private function.

I removed the test.

burgholzer · 2025-05-23T08:04:59Z

src/mqt/bench/targets/devices/__init__.py

+        get_ionq_target("ionq_forte_1"),
+        get_ionq_target("ionq_aria_1"),


But aria 2 still uses the same layout and the same gateset if I am not mistaken.
Given how the device data is averaged anyway, I do not think there is too much value to this number.

If the gateset should really be different between these, then I would actually add the number to the gateset name and still append an underscore with the number of qubits for the device here.

Same holds for the Rigetti device.

src/mqt/bench/targets/devices/ionq.py

burgholzer · 2025-05-23T08:09:26Z

src/mqt/bench/targets/gatesets/ionq.py

+        [0],
+    )
+    u_gate.append(GPI2Gate(0.5 + phi_param / (2 * np.pi)), [0])
+    SessionEquivalenceLibrary.add_equivalence(UGate(theta_param, phi_param, lambda_param), u_gate)


I believe it would be better if these methods would take an equivalence library as a parameter instead of unconditionally adding to the SEL.
Gives a bit more flexibility.

I followed your hint. Additionally, I add the library now only immediately before the transpile method is called. I played around a bit with the EquivalenceLibrary and you can get all available decompositions for a gate but would need to compare the existing ones with the one that might be added to check if it is already in there - which is also not ideal.

burgholzer · 2025-05-23T08:15:14Z

src/mqt/bench/targets/gatesets/rigetti.py

+    SessionEquivalenceLibrary.add_equivalence(UGate(theta, phi, lam), qc)
+
+
+def cx_gate_equivalence() -> None:


Is it really true that a single CX gate takes two iSWAP gates? Seems counterintuitive to me.
Furthermore, the iSWAP gate is a Qiskit standard gate, so this equivalence should not really be needed.

I thinks so, also considering this reference implementation (https://docs.quantum.ibm.com/api/qiskit/qiskit.circuit.library.iSwapGate). Although it is the other way round there. Anyways, I removed this equivalence because it is not necessary.

burgholzer · 2025-05-23T08:17:03Z

src/mqt/bench/targets/gatesets/rigetti.py

+        self.definition = qc
+
+
+def u_gate_equivalence() -> None:


I think it could be worth to add a couple more equivalences here that take fewer gates.
Equivalences for the real (parametrized) RX gate come to mind.
Pretty sure there are others.

I added one for that, although it still has 5 gates (but at least without parameters - not sure how much it improves though). Let me know if you have specific other ones in mind as well.

burgholzer

One final comment, then I believe this should be good to go here 🙌

burgholzer · 2025-05-24T03:08:10Z

src/mqt/bench/targets/gatesets/rigetti.py

+def rx_gate_equivalence(sel: EquivalenceLibrary) -> None:
+    """Add RX(θ) gate equivalence using RX(±π/2) and RZ gates."""
+    theta = Parameter("θ")
+    qr = QuantumRegister(1, "q")
+    qc = QuantumCircuit(qr)
+
+    qc.rz(-np.pi / 2, qr[0])
+    qc.append(RXPI2Gate(), [qr[0]])
+    qc.rz(theta, qr[0])
+    qc.append(RXPI2DgGate(), [qr[0]])
+    qc.rz(np.pi / 2, qr[0])
+
+    sel.add_equivalence(RXGate(theta), qc)


This doesn't really help much I suppose.
What I actually meant with my other comment was to add equivalences from
RXGate(pi/2) to RXPI2Gate and so on.

Is this what you would like to see here?

def rx_gate_equivalences(sel: EquivalenceLibrary) -> None: """Add RX(θ) gate equivalence using RX(±π/2) and RZ gates.""" qr = QuantumRegister(1, "q") qc = QuantumCircuit(qr) qc.append(RXPI2Gate(), [qr[0]]) sel.add_equivalence(RXGate(np.pi / 2), qc) qr = QuantumRegister(1, "q") qc = QuantumCircuit(qr) qc.append(RXPI2DgGate(), [qr[0]]) sel.add_equivalence(RXGate(-np.pi / 2), qc) qr = QuantumRegister(1, "q") qc = QuantumCircuit(qr) qc.append(RXPIGate(), [qr[0]]) sel.add_equivalence(RXGate(np.pi), qc)

If yes: Apparently it is not possible to add these equivalences only for specific angles.

Yeah. Something like that is what I would have expected.
How's that not supported? Or what is the error you are seeing?

It leads to an internal Qiskit Error in qiskit.transpiler.passes.basis.basis_translator.BasisTranslator.run: AttributeError: 'float' object has no attribute '_uuid' because for a parameterized gate, Qiskit apparently expects a parameter instead of just a specific angle.

Hm.. unfortunate..
That's something that I would consider opening up an issue on the Qiskit tracker for.
Such equivalences could be extremely helpful.

Any how.. I suppose this PR can be merged then 😌

🎨 WIP for IonQ Device handling

8f97eea

nquetschlich added the enhancement New feature or request label May 22, 2025

github-project-automation bot added this to MQT Applications May 22, 2025

github-project-automation bot moved this to In Progress in MQT Applications May 22, 2025

nquetschlich added 2 commits May 22, 2025 16:58

Merge branch 'main' into update_ionq_devices

ea45fa2

handled linter warnings

33a23f5

burgholzer requested changes May 22, 2025

View reviewed changes

nquetschlich added 8 commits May 23, 2025 07:40

♻️ refactored module structrue

60575ae

♻️ refactored ionq handling

a0799c1

♻️ refactored gateset handling

9831391

♻️ refactored the IonQ gate definitions

1031198

✨ introduced the new IonQ gates also to the native gates level

c44b965

💡 added copyrights

0a2e568

🎨 improved ionq gateset handling

64b96d3

🎨 simplified gate definition

8d54a14

update Rigetti Device and Gateset accordingly to IonQ one

3ac576f

nquetschlich changed the title ~~🎨 IonQ Device Update~~ 🎨 Update of IonQ and Rigetti Devices May 23, 2025

nquetschlich added 4 commits May 23, 2025 09:30

✅ added test to increase coverage

e6279f7

🚨

58b8de4

🎨 renamed custom rigetti gates

a27e24f

🔥 removed unnecessary array definition fur custom gates

bc53846

nquetschlich marked this pull request as ready for review May 23, 2025 08:02

🎨 improved equivalence handling

d9e8084

burgholzer requested changes May 23, 2025

View reviewed changes

nquetschlich added 5 commits May 23, 2025 14:23

🎨 use relative imports instead of absolute ones

b13e824

🎨 renamed custom rigetti gates

db651b3

🎨 adjusted the naming scheme of devices

e300c9b

🎨 removed tests that used an internal method

f1666f4

🔥 removed unnecessary equivalence

2eda174

nquetschlich added 3 commits May 23, 2025 15:26

✨ added a rx decomposition

371a77e

✨ improved the equivalence adding logic

4c5666e

✨ removed comment

9645895

nquetschlich requested a review from burgholzer May 23, 2025 14:23

burgholzer requested changes May 24, 2025

View reviewed changes

nquetschlich added 2 commits May 24, 2025 09:53

Merge branch 'main' into update_ionq_devices

f2b5fee

Merge branch 'main' into update_ionq_devices

4b1f0ee

burgholzer approved these changes May 25, 2025

View reviewed changes

nquetschlich merged commit cb0c53b into main May 25, 2025
17 of 18 checks passed

nquetschlich deleted the update_ionq_devices branch May 25, 2025 16:44

github-project-automation bot moved this from In Progress to Done in MQT Applications May 25, 2025

		get_ionq_target("ionq_forte_1"),
		get_ionq_target("ionq_aria_1"),

		from qiskit.transpiler import InstructionProperties, Target

		from mqt.bench.targets.gatesets.ionq import GPI2Gate, GPIGate, MSGate, ZZGate, add_ionq_equivalences

		SessionEquivalenceLibrary.add_equivalence(UGate(theta, phi, lam), qc)


		def cx_gate_equivalence() -> None:

🎨 Update of IonQ and Rigetti Devices #570

🎨 Update of IonQ and Rigetti Devices #570

Uh oh!

Conversation

nquetschlich commented May 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

burgholzer left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

codecov bot commented May 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

burgholzer left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nquetschlich May 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

burgholzer left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

nquetschlich commented May 22, 2025 •

edited

Loading

codecov bot commented May 23, 2025 •

edited

Loading

nquetschlich May 23, 2025 •

edited

Loading