Improve annotated plugin #13916

alexanderivrii · 2025-02-24T14:12:45Z

Summary

This PR significantly improves the high-level-synthesis plugin for synthesizing AnnotatedOperations and in particular the synthesis of hierarchical circuits involving controlled subcircuits. Addresses #13566. Should be rebased on top of #13813. Does not need #13870, but serves as a basis to that PR explaining why we want to change the default handling of controlled gates to annotated gates.

Details and comments

Here are the experiments for the following 9 simple quantum circuits:

# circuit1: 6-controlled X
qc = QuantumCircuit(15)
qc.append(XGate().control(6), [0, 1, 2, 3, 4, 5, 6])

# circuit2: 5-controlled CX
qc = QuantumCircuit(15)
qc.append(XGate().control(6), [0, 1, 2, 3, 4, 5, 6])

# circuit3: hierarchical circuit morally equivalent to 5-controlled CX
inner = QuantumCircuit(5)
inner.append(CXGate().control(3, annotated=True), [0, 1, 2, 3, 4])
controlled_inner_gate2 = inner.to_gate().control(2, annotated=True)
qc = QuantumCircuit(15)

# circuit4: hierarchical circuit morally equivalent to 5-controlled CX
inner = QuantumCircuit(5)
inner.append(CXGate().control(3, annotated=False), [0, 1, 2, 3, 4])
controlled_inner_gate2 = inner.to_gate().control(2, annotated= False)
qc = QuantumCircuit(15)

# circuits 5, 6, 7, 8: same as 1, 2, 3, 4 but with Z/CZ gates instead of X/CX

#circuit 9: our old friend controlled QFT-based adder
gate = HalfAdderGate(3).control(2, annotated=True)
qc = QuantumCircuit(gate.num_qubits)
qc.append(gate, qc.qubits)

We are going to transpile these circuits with HighLevelSynthesis(basis_gates=["cx", "u"]).

MAIN:

CIRCUIT1: {'u': 57, 'cx': 30}
CIRCUIT2: {'u': 57, 'cx': 30}
CIRCUIT3: {'u': 849, 'cx': 574}
CIRCUIT4: {'u': 1575, 'cx': 974}
CIRCUIT5: {'u': 59, 'cx': 30}
CIRCUIT6: {'u': 159, 'cx': 78}
CIRCUIT7: {'u': 5719, 'cx': 3626}
CIRCUIT8: {'u': 2549, 'cx': 1586}
CIRCUIT9: {'u': 1607, 'cx': 1310}

THIS PR:

CIRCUIT1: {'u': 57, 'cx': 30}
CIRCUIT2: ({'u': 57, 'cx': 30}
CIRCUIT3: {'u': 57, 'cx': 30}
CIRCUIT4: {'u': 1543, 'cx': 990}
CIRCUIT5: {'u': 59, 'cx': 30}
CIRCUIT6: {'u': 59, 'cx': 30}
CIRCUIT7: {'u': 59, 'cx': 30}
CIRCUIT8: {'u': 1573, 'cx': 1002}
CIRCUIT9: {'u': 539, 'cx': 450}

Takeaways: (1) this PR improves the synthesis for almost all of the examples, with especial gains for hierarchical circuits with controlled operations represented as annotated operations; (2) using annotated operations for representing controlled gates is the right way to go forward, as this easily enables reductions and optimizations not available otherwise.

The interface is now quite ugly, and we also have an extra isinstance check for each gate, we will see if it can be removed.

…AGCircuits

This argument is not needed as the qubit tracker already restricts allowed ancilla qubits.

With various conversions of DAGCircuit to QuantumCircuitData to CircuitData in Rust, it's best to make sure that the global phases appearing in circuits/control-flow subcircuits are tracked correctly

… ops when calls from annotated plugin

…separate file

…fiers; adding test

This leads to significantly improved synthesis results when adding multiple controls to a gate. Previously, first controlling a CX with 2 controls, and then controlling the results with 3 controls, first resulted in expanding the CCCX gates into single- and two-qubits gates, adding controls to each of these gates, and then expanding each of these controlled gates. Now, a single MCX gate will be produced, which will then be expanded using the best MCX synthesis algorithm available.

…or control modifiers

… and Z gates

This is a simplified version from the unused code from the OptimizeAnnotated transpiler pass. It's especially useful for controlled QFT-adder and similar circuits.

This allows to let the default QFT-based adder plugins to create the inverse QFTGate as an AnnotatedOperation, allowing to exploit optimizations in the HighLevelSynthesis transpiler pass

qiskit-bot · 2025-02-24T14:12:52Z

One or more of the following people are relevant to this code:

@Cryoris
@Qiskit/terra-core
@ajavadia

…ed gates from add_control, but with the twist that some other high-level gate names are also supported; fixing failing tests that are synthesized differently (and actually better) using the new flow

…lier synthesis algorithm, and not the synthesis of the underlying QFT gate.

coveralls · 2025-02-25T08:44:17Z

Pull Request Test Coverage Report for Build 13516621530

Details

732 of 783 (93.49%) changed or added relevant lines in 10 files are covered.
33 unchanged lines in 7 files lost coverage.
Overall coverage increased (+0.03%) to 88.008%

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
qiskit/transpiler/passes/synthesis/plugin.py	1	2	50.0%
qiskit/transpiler/passes/synthesis/high_level_synthesis.py	62	64	96.88%
qiskit/transpiler/passes/synthesis/annotated_synthesis.py	123	133	92.48%
crates/accelerate/src/high_level_synthesis.rs	513	551	93.1%

Files with Coverage Reduction	New Missed Lines	%
qiskit/transpiler/passes/synthesis/plugin.py	1	84.34%
crates/circuit/src/dag_circuit.rs	3	86.07%
crates/qasm2/src/lex.rs	3	93.23%
crates/circuit/src/circuit_instruction.rs	4	90.16%
crates/qasm2/src/parse.rs	6	97.15%
crates/accelerate/src/high_level_synthesis.rs	7	90.27%
crates/circuit/src/dag_node.rs	9	74.55%

Totals
Change from base Build 13512719904:	0.03%
Covered Lines:	77980
Relevant Lines:	88606

💛 - Coveralls

ShellyGarion

That's a great PR! see some comments and suggestions below.

ShellyGarion · 2025-02-26T12:11:52Z

qiskit/circuit/add_control.py

@@ -109,10 +109,10 @@ def control(

    global_phase = 0

-    basis = ["p", "u", "x", "z", "y", "h", "sx", "sxdg", "rx", "ry", "rz", "cx"]
+    basis = ["p", "u", "x", "z", "y", "h", "sx", "sxdg", "rx", "ry", "rz", "cx", "cz"]


we should check (in a follow-up PR) what happens with other controlled gates, such as:
cy, ch, cp, crx, cry, crz, cs, csdg, cswap, csx, csxdg, cu,
as well as ccx, ccz, c3x, c4x, c3sx.
Will adding them here improve the total CX cost?

btw, I really don't like the fact that we're copying this array in different part of the code.
I would suggest to set it somewhere and use it, see e.g. the CollectClifford transpiler pass:

qiskit/qiskit/transpiler/passes/optimization/collect_cliffords.py

Line 77 in f6608c2

clifford_gate_names = (

ShellyGarion · 2025-02-26T12:16:16Z

qiskit/circuit/add_control.py

@@ -187,7 +187,7 @@ def apply_basic_controlled_gate(circuit, gate, controls, target):

    This implements multi-control operations for the following basis gates:

-        ["p", "u", "x", "z", "y", "h", "sx", "sxdg", "rx", "ry", "rz", "cx"]
+        ["p", "u", "x", "z", "y", "h", "sx", "sxdg", "rx", "ry", "rz", "cx", "cz"]


see my comment about this array above.

ShellyGarion · 2025-02-26T12:19:03Z

qiskit/circuit/add_control.py

@@ -230,6 +230,14 @@ def apply_basic_controlled_gate(circuit, gate, controls, target):
            controls[:] + [target[0]],  # CX has two targets
            target[1],
        )
+    elif gate.name == "cz":


btw, what is the improvement in cx-cost of adding cz here?

ShellyGarion · 2025-02-26T12:32:43Z

qiskit/transpiler/passes/synthesis/annotated_synthesis.py

+        if not isinstance(high_level_object, AnnotatedOperation):
+            return None
+
+        # Combine the modifiers. If the were no modifiers, or the modifiers magically canceled out,


the --> there

ShellyGarion · 2025-02-26T12:34:41Z

qiskit/transpiler/passes/synthesis/annotated_synthesis.py

+        # (including annotated operations) as the HighLevelSynthesis transpiler pass will
+        # recursively re-synthesize this circuit, However, we should always guarantee that some
+        # progress is made.
+        basis = {


see my comment on the array above

ShellyGarion · 2025-02-26T12:44:11Z

qiskit/transpiler/passes/synthesis/annotated_synthesis.py

+    """
+    res = False
+
+    if inst1.qubits != inst2.qubits or inst1.clbits != inst2.clbits or inst1.params != inst2.params:


in standard gates, usually the params of the inverse would be different than the params of the original gate,
for example: the inverse of rz(t) is rz(-t).
it does make sense however to check that both params have the same number of params.

ShellyGarion · 2025-02-26T12:51:56Z

qiskit/transpiler/passes/synthesis/annotated_synthesis.py

+    """
+    Decomposes a circuit ``A`` into 3 sub-circuits ``P``, ``Q``, ``R`` such that
+    ``A = P -- Q -- R`` and ``R = P^{-1}``.
+


can you do it more efficiently if you generate the circuit "P - R" and check if some transpiler passes return it to identity?
e.g. InverseCancellation, CommutativeInverseCancellation, CommutativeCancellation, RemoveIdentityEquivalent

ShellyGarion · 2025-02-26T12:55:20Z

test/python/transpiler/test_high_level_synthesis.py

+
+class TestHighLevelSynthesisQuality(QiskitTestCase):
+    """Test the "quality" of circuits produced by HighLevelSynthesis."""
+


Are there tests that use several modifiers? the QFT-adder should use both conjuage and control, right?

perhaps this example is also relevant for the conjugate modifier (since ccx gate is self-conjugate):
#6740

ShellyGarion · 2025-02-26T12:57:37Z

test/python/transpiler/test_high_level_synthesis.py

+        qc = QuantumCircuit(gate.num_qubits)
+        qc.append(gate, qc.qubits)
+        qct = HighLevelSynthesis(basis_gates=["cx", "u"], qubits_initially_zero=False)(qc)
+        self.assertLessEqual(qct.count_ops()["cx"], 450)


is this bound optimal? (assuming that the QFT requires 198). what is the formula for the CX count?
is the bound of 198 for QFT optimal? is there a formula for the CX count?

alexanderivrii added 30 commits December 11, 2024 15:40

Moving all internal fields into a separate class

d0b4137

making all the inner functions to be global

1f5a694

moving handling annotated operations to plugin

2f2e140

The interface is now quite ugly, and we also have an extra isinstance check for each gate, we will see if it can be removed.

improving control-modified synthesis plugin

7db3493

Merge branch 'main' into hls_reorg

ec4f4e8

Changing internal functions to work with QuantumCircuits instead of D…

0a37fee

…AGCircuits

renaming

a74f68e

Removing unnecessary argument use_ancillas

3fe0454

This argument is not needed as the qubit tracker already restricts allowed ancilla qubits.

progress

71fbf57

minor

89b33ac

removing context

76e10a3

minor

5deb316

removing unused argument

663a937

removing an obsolete statement

19bdbde

minor cleanup

8a73f4f

minor

0a6950b

improvements to run method

bbc2175

cleanup

a02ed03

another pass over the run method + black

53c7dee

minor cleanup

23d2ef7

pass over synthesize_operation

3f2904b

more cleanup

09f5c15

pass over HLS::run

1022ec4

cleanup

56272b6

pass over annotated plugin

2d3e800

cleanup

5ea2ccb

improving comment

1ffeba7

fixing pylint

39e552b

remove print statements in tests

919b549

fmt

9f2ef8d

alexanderivrii added 17 commits February 13, 2025 10:06

pylint

9468bb9

clippy

903d54d

Adding tests for HLS tracking global phase.

640e267

With various conversions of DAGCircuit to QuantumCircuitData to CircuitData in Rust, it's best to make sure that the global phases appearing in circuits/control-flow subcircuits are tracked correctly

cleanup

7fadaeb

removing unnecessary changes to other files

bcf62b5

Merge branch 'main' into port-hls-to-rust

e78348b

improving Rust HLS interface; skipping synthesis of already supported…

ad24bf4

… ops when calls from annotated plugin

Moving the default plugin for synthesizing annotated operations to a …

721212e

…separate file

Fixing synthesis of an annotated operation with an empty list of modi…

d318ab9

…fiers; adding test

Adding tests that verify that HLS correctly combines control states f…

bac9056

…or control modifiers

improving handling of CZ-gates; adding quality tests for controlled X…

9d6d711

… and Z gates

clippy

b74b8b8

Simple version of conjugate reduction.

adcbb26

This is a simplified version from the unused code from the OptimizeAnnotated transpiler pass. It's especially useful for controlled QFT-adder and similar circuits.

Adding annotated argument to adder_qft_d00

130c24a

This allows to let the default QFT-based adder plugins to create the inverse QFTGate as an AnnotatedOperation, allowing to exploit optimizations in the HighLevelSynthesis transpiler pass

adjusting HLS plugins + adding quality test for controlled qft adder

a52274d

preliminary release notes

3cd9a20

alexanderivrii added this to the 2.0.0 milestone Feb 24, 2025

alexanderivrii requested a review from a team as a code owner February 24, 2025 14:12

1ucian0 linked an issue Feb 24, 2025 that may be closed by this pull request

Improve synthesis of controlled circuits with controlled gates within #13566

Open

alexanderivrii added 5 commits February 24, 2025 22:42

Fully adapting solution for synthesizing base operation of an annotat…

a5d550f

…ed gates from add_control, but with the twist that some other high-level gate names are also supported; fixing failing tests that are synthesized differently (and actually better) using the new flow

merge main branch

06bec86

fix conflicts after merge

145e722

Fixing test to check for gates that should belong to the given multip…

bbfd6f2

…lier synthesis algorithm, and not the synthesis of the underlying QFT gate.

Adding comments

37c2746

ShellyGarion assigned alexanderivrii and ShellyGarion Feb 26, 2025

ShellyGarion reviewed Feb 26, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Improve annotated plugin #13916

Improve annotated plugin #13916

alexanderivrii commented Feb 24, 2025

qiskit-bot commented Feb 24, 2025

coveralls commented Feb 25, 2025

ShellyGarion left a comment

ShellyGarion Feb 26, 2025

ShellyGarion Feb 26, 2025

ShellyGarion Feb 26, 2025

ShellyGarion Feb 26, 2025

ShellyGarion Feb 26, 2025

ShellyGarion Feb 26, 2025

ShellyGarion Feb 26, 2025

ShellyGarion Feb 26, 2025

ShellyGarion Feb 26, 2025

ShellyGarion Feb 26, 2025

ShellyGarion Feb 26, 2025 •

edited

Loading


		class TestHighLevelSynthesisQuality(QiskitTestCase):
		"""Test the "quality" of circuits produced by HighLevelSynthesis."""

Improve annotated plugin #13916

Are you sure you want to change the base?

Improve annotated plugin #13916

Conversation

alexanderivrii commented Feb 24, 2025

Summary

Details and comments

qiskit-bot commented Feb 24, 2025

coveralls commented Feb 25, 2025

Pull Request Test Coverage Report for Build 13516621530

Details

💛 - Coveralls

ShellyGarion 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

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

ShellyGarion Feb 26, 2025 • edited Loading

Choose a reason for hiding this comment

ShellyGarion Feb 26, 2025 •

edited

Loading