Performance improvements for collect_2q_blocks

[IvanIsCoding]

Summary

Closes Qiskit/rustworkx#265

This PR contains two core improvements that boost the performance of collect_2q_blocks:

• Use new functions from retworkx in quantum_successors and quantum_predecessors that are more efficient
• Add is_successor and is_predecessor function to DAGCircuit to efficiently check for successors/predecessors instead of getting all successors/predecessors and then doing a check

It also contains two minor performance optimizations that are open to debate:

• Replace qargs with _qargs, because of @property overhead when getting qargs
• Replace op with _op, idem

Details and comments

This PR also upgrades retworkx to version 0.9

[IvanIsCoding]

I benchmarked the changes and I got results ranging from 1.3x to 1.5x times faster.

Below is the code for one of the benchmarks. It is a Quantum Volume Circuit, unrolled to use cx and u3 gates. I got a 1.45x speedup on the transpiler pass on that circuit.

import timeit
from qiskit import QuantumRegister, QuantumCircuit
from qiskit.circuit.library import QuantumVolume
from qiskit.converters import circuit_to_dag
from qiskit.transpiler.passes.optimization import Collect2qBlocks
from qiskit.compiler import transpile

N = 100
q = QuantumRegister(N, 'q')
qc = QuantumCircuit(q)
qc.append(
	QuantumVolume(N, N//2, seed=2, classical_permutation=False), q
)

qc = transpile(qc, optimization_level=0, basis_gates=['u1', 'u2', 'u3', 'cx'])
dag = circuit_to_dag(qc)
c2q = Collect2qBlocks()

print(
	timeit.timeit("c2q.run(dag)", globals=globals(), number=10)
)

Time without optimizations: 3.6710015139997267
Time with optimizations: 2.5208510480006225

[mtreinish]

Can you add an upgrade release note here about the bump in the minimum required retworkx version. We should call that out in the release notes as a thing that might require end user manual updates when upgrading to the next terra release.

[IvanIsCoding]

Can you add an upgrade release note here about the bump in the minimum required retworkx version. We should call that out in the release notes as a thing that might require end user manual updates when upgrading to the next terra release.

I added the release note. This is no longer WIP, it can be merged

[mtreinish]

LGTM, just one comment with some suggestions on the release notes. Other than that I think this is ready to go.

[mtreinish]

I also ran the asv collect_2q_blocks benchmark on this and it yielded this result:

Benchmarks that have improved:

       before           after         ratio
     [a2193a5c]       [e2935d73]
     <main>       <IvanIsCoding-improve_collect2q>
-      44.3±0.2ms       30.4±0.2ms     0.69  passes.PassBenchmarks.time_collect_2q_blocks(14, 1024)
-      65.8±0.3ms       44.9±0.6ms     0.68  passes.PassBenchmarks.time_collect_2q_blocks(20, 1024)
-      15.4±0.1ms       10.3±0.3ms     0.67  passes.PassBenchmarks.time_collect_2q_blocks(5, 1024)

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.
PERFORMANCE INCREASED.

[mtreinish]

I also ran a quick benchmark on https://github.com/Qiskit/qiskit/blob/master/test/benchmarks/qasm/test_eoh_qasm.qasm (which is different load on the pass because it's all 2q, so one giant block) with:

import time
import statistics

from qiskit.transpiler.passes import Collect2qBlocks
from qiskit.circuit import QuantumCircuit
from qiskit.converters import circuit_to_dag

qc = QuantumCircuit.from_qasm_file('test_eoh_qasm.qasm')
dag = circuit_to_dag(qc)
collect_pass = Collect2qBlocks()
times = []
for i in range(20):
    start = time.time()
    collect_pass.run(dag)
    stop = time.time()
    times.append(stop - start)
print("Mean run time: %s" % statistics.mean(times))

which returns 0.015 seconds on main and 0.01 seconds with this PR.

[IvanIsCoding]

I commited the suggestions to the release note, the need for retworkx upgrade should be clearer now

I also ran the asv collect_2q_blocks benchmark on this and it yielded this result:

Benchmarks that have improved:

       before           after         ratio
     [a2193a5c]       [e2935d73]
     <main>       <IvanIsCoding-improve_collect2q>
-      44.3±0.2ms       30.4±0.2ms     0.69  passes.PassBenchmarks.time_collect_2q_blocks(14, 1024)
-      65.8±0.3ms       44.9±0.6ms     0.68  passes.PassBenchmarks.time_collect_2q_blocks(20, 1024)
-      15.4±0.1ms       10.3±0.3ms     0.67  passes.PassBenchmarks.time_collect_2q_blocks(5, 1024)

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.
PERFORMANCE INCREASED.

Cool, the 1.3x - 1.5x speed up remained consistent accross the benchmarks.

[mtreinish]

Oops, thanks for fixing the whitespace issue in my release note suggestion (sigh getting whitespace right in the github editor is so irritating).