Tune LitinskiTransformation pass

[mtreinish]

Summary

This commit optimizes the implementation of the LitinskiTransformation pass. The improvements fall into a couple of categories, but most of the runtime performance improvements will come from removing unnecessary allocations.

From a quick glance the Clifford struct looks like it could also use a similar once over, mostly to get it to leverage Qiskit's types more effectively if nothing else. But I kept the changes there minimal for this commit and only modified the method used by The LitinskiTransformation directly.

Details and comments

[qiskit-bot]

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

• @Qiskit/terra-core

[mtreinish]

To benchmark this I used the benchmark @alexanderivrii described in the PR summary on #14753 with this script:

import statistics
import time

from qiskit.converters import circuit_to_dag
from qiskit.circuit.library import QFTGate
from qiskit.transpiler.passes import LitinskiTransformation
from qiskit.transpiler import Target
from qiskit import transpile, QuantumCircuit

litinski = LitinskiTransformation()

i = 2
while i <= 1024:

    qft = QuantumCircuit(i)
    qft.append(QFTGate(i), qft.qubits)
    target = Target.from_configuration(["cx", "sx", "rz"], i)
    tqft = transpile(qft, target=target)
    dag = circuit_to_dag(tqft, False)
    times = []
    for _ in range(10):
        start = time.perf_counter()
        litinski.run(dag)
        stop = time.perf_counter()
        times.append(stop - start)
    print(f"nqubits: {i}, runtime: {statistics.geometric_mean(times)}")
    i = i * 2

It yielded these results on my local system:

[coveralls]

Pull Request Test Coverage Report for Build 19208500951

Details

• 145 of 146 (99.32%) changed or added relevant lines in 2 files are covered.
• 3 unchanged lines in 3 files lost coverage.
• Overall coverage increased (+0.03%) to 88.188%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
crates/transpiler/src/passes/litinski_transformation.rs	124	125	99.2%

Files with Coverage Reduction	New Missed Lines	%
crates/circuit/src/parameter/symbol_expr.rs	1	73.27%
crates/qasm2/src/lex.rs	1	92.54%
crates/transpiler/src/passes/litinski_transformation.rs	1	96.43%

Totals
Change from base Build 19206691463:	0.03%
Covered Lines:	94298
Relevant Lines:	106928

---

💛 - Coveralls

[alexanderivrii]

Thanks @mtreinish! I have just now run your script for the most problematic case of QFTGate(1024) (with the only addition of also printing the runtime for each individual run) and I see no significant difference between the main branch, #15217 (which extends the support to PauliProductMeasurement instructions and also refactors the code a bit, though this refactoring is orthogonal to your changes), and this PR. For each of the 3 code versions it takes about 43 seconds to run the transform. For reference, this is done on my mac laptop and python 3.13.

In any case, for the interest of helping @Cryoris to prepare for QDC, I would suggest merging #15217 first, and then merging your PR.