Port cx_cz_lnn to rust

[gadial]

Summary

Ports the synth_cx_cz_depth_line_my method to Rust.

Closes #12231

Details and comments

This is synthesis pass for constructing circuits given by two matrices, representing CX and CZ circuits. The code generating the instruction list was ported to Rust in a straightforward manner; construction of the circuit itself is done in Python.

A quick benchmarking was performed by using random_invertible_binary_matrix(n, seed=seed) of the CX matrix and np.triu(np.random.randint(0, 2, (n, n)), k=1) for the CZ matrix, and timing the application of synth_cx_cz_depth_line_my on the result, showing consistent improvement for the rust version.

╒═════════════════════╤═════════════╤═════════════╤═══════════════════════╕
│ Test Case           │      python │        rust │   Ratio (python/rust) │
╞═════════════════════╪═════════════╪═════════════╪═══════════════════════╡
│ 5x5 (seed 42)       │  0.0032567  │ 5.85556e-05 │               55.6173 │
├─────────────────────┼─────────────┼─────────────┼───────────────────────┤
│ 10x10 (seed 42)     │  0.00927253 │ 8.74519e-05 │              106.03   │
├─────────────────────┼─────────────┼─────────────┼───────────────────────┤
│ 20x20 (seed 55)     │  0.0447065  │ 0.000269413 │              165.94   │
├─────────────────────┼─────────────┼─────────────┼───────────────────────┤
│ 50x50 (seed 13)     │  0.472343   │ 0.00473766  │               99.6997 │
├─────────────────────┼─────────────┼─────────────┼───────────────────────┤
│ 60x60 (seed 55)     │  0.825494   │ 0.00885715  │               93.2008 │
├─────────────────────┼─────────────┼─────────────┼───────────────────────┤
│ 70x70 (seed 13)     │  1.19549    │ 0.0150931   │               79.2078 │
├─────────────────────┼─────────────┼─────────────┼───────────────────────┤
│ 100x100 (seed 1089) │  3.91824    │ 0.0512372   │               76.4727 │
├─────────────────────┼─────────────┼─────────────┼───────────────────────┤
│ 120x120 (seed 17)   │  7.31543    │ 0.100896    │               72.5043 │
├─────────────────────┼─────────────┼─────────────┼───────────────────────┤
│ 150x150 (seed 99)   │ 16.1747     │ 0.234127    │               69.0851 │
╘═════════════════════╧═════════════╧═════════════╧═══════════════════════╛

[qiskit-bot]

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

• @Qiskit/terra-core

[gadial]

I see now I missed the % 4 here and it wasn't caught by testing; I'll extend the tests to ensure this is caught.

[gadial]

It seems this cannot have any real effect since the values in the phase schedule are always computed modulo 4 so the code is technically correct, although I'll add the % 4 since it's better to be on the safe side.

[coveralls]

Pull Request Test Coverage Report for Build 13386025321

Details

• 180 of 183 (98.36%) changed or added relevant lines in 4 files are covered.
• 21 unchanged lines in 5 files lost coverage.
• Overall coverage increased (+0.009%) to 88.189%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
crates/accelerate/src/synthesis/linear/lnn.rs	4	5	80.0%
crates/accelerate/src/synthesis/linear_phase/cx_cz_depth_lnn.rs	170	172	98.84%

Files with Coverage Reduction	New Missed Lines	%
crates/qasm2/src/expr.rs	1	94.23%
crates/accelerate/src/unitary_synthesis.rs	3	94.29%
crates/accelerate/src/synthesis/linear/lnn.rs	5	96.94%
crates/qasm2/src/lex.rs	6	91.48%
crates/qasm2/src/parse.rs	6	97.61%

Totals
Change from base Build 13376164595:	0.009%
Covered Lines:	78820
Relevant Lines:	89376

---

💛 - Coveralls

[ShellyGarion]

Could you perhaps also compare the time improvement of the function synth_clifford_depth_lnn now that all of the internal functions (synth_cnot_depth_line_kms, synth_cz_depth_line_mr, synth_cx_cz_depth_line_my) were ported to rust? (compare qiskit 1.2 vs. qiskit 2.0).

qiskit/qiskit/synthesis/clifford/clifford_decompose_layers.py

Line 417 in bd30693

def synth_clifford_depth_lnn(cliff):

[gadial]

Could you perhaps also compare the time improvement of the function synth_clifford_depth_lnn now that all of the internal functions (synth_cnot_depth_line_kms, synth_cz_depth_line_mr, synth_cx_cz_depth_line_my) were ported to rust? (compare qiskit 1.2 vs. qiskit 2.0).

qiskit/qiskit/synthesis/clifford/clifford_decompose_layers.py

Line 417 in bd30693

def synth_clifford_depth_lnn(cliff):

When running on random cliffords, I'm getting a nice improvement but I suspect there's more we can do.

╒══════════════════════╤════════════╤════════════╤═══════════════════════╕
│ Test Case            │     python │       rust │   Ratio (python/rust) │
╞══════════════════════╪════════════╪════════════╪═══════════════════════╡
│ 2 qubits (seed 42)   │ 0.00163498 │ 0.00252638 │              0.647163 │
├──────────────────────┼────────────┼────────────┼───────────────────────┤
│ 5 qubits (seed 42)   │ 0.00278735 │ 0.00405846 │              0.686801 │
├──────────────────────┼────────────┼────────────┼───────────────────────┤
│ 10 qubits (seed 42)  │ 0.00726461 │ 0.00962257 │              0.754955 │
├──────────────────────┼────────────┼────────────┼───────────────────────┤
│ 20 qubits (seed 42)  │ 0.029776   │ 0.0239356  │              1.24401  │
├──────────────────────┼────────────┼────────────┼───────────────────────┤
│ 40 qubits (seed 42)  │ 0.14392    │ 0.0663365  │              2.16955  │
├──────────────────────┼────────────┼────────────┼───────────────────────┤
│ 80 qubits (seed 42)  │ 0.940864   │ 0.265452   │              3.54439  │
├──────────────────────┼────────────┼────────────┼───────────────────────┤
│ 100 qubits (seed 42) │ 1.98206    │ 0.446445   │              4.43966  │
├──────────────────────┼────────────┼────────────┼───────────────────────┤
│ 150 qubits (seed 42) │ 8.59436    │ 1.27764    │              6.72675  │
╘══════════════════════╧════════════╧════════════╧═══════════════════════╛

[gadial]

Doing quick profiling of 1.2 vs 2.0:

1.2:

ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000   25.077   25.077 clifford_decompose_layers.py:416(synth_clifford_depth_lnn)
        1    0.000    0.000   25.077   25.077 clifford_decompose_layers.py:67(synth_clifford_layers)
        1    0.007    0.007   12.975   12.975 clifford_decompose_layers.py:248(_decompose_graph_state)
        1    3.998    3.998   12.813   12.813 cz_depth_lnn.py:121(synth_cz_depth_line_mr)
        1    0.002    0.002    8.967    8.967 clifford_decompose_layers.py:311(_decompose_hadamard_free)
        1    0.009    0.009    8.917    8.917 cx_cz_depth_lnn.py:218(synth_cx_cz_depth_line_my)
    18772    8.034    0.000    8.034    0.000 {method 'count' of 'list' objects}
        1    2.971    2.971    3.983    3.983 cx_cz_depth_lnn.py:119(_update_phase_schedule)
        1    0.000    0.000    3.421    3.421 linear_depth_lnn.py:210(_optimize_cx_circ_depth_5n_line)
        1    0.053    0.053    3.354    3.354 linear_depth_lnn.py:113(_matrix_to_north_west)
    49238    3.225    0.000    3.242    0.000 linear_depth_lnn.py:89(_in_linear_combination)
     13/6    0.000    0.000    2.801    0.467 clifford.py:147(__init__)
        1    0.000    0.000    2.801    2.801 clifford.py:670(from_circuit)
      7/1    0.232    0.033    2.800    2.800 clifford_circuits.py:25(_append_circuit)
 152282/6    0.226    0.000    2.800    0.467 clifford_circuits.py:53(_append_operation)
   152424    0.474    0.000    1.888    0.000 quantumcircuit.py:2405(_append_standard_gate)
   119340    0.062    0.000    1.595    0.000 quantumcircuit.py:5303(cx)
   119340    0.909    0.000    1.548    0.000 clifford_circuits.py:401(_append_cx)
        1    0.041    0.041    0.864    0.864 cx_cz_depth_lnn.py:163(_apply_phase_to_nw_circuit)
  5933932    0.610    0.000    0.610    0.000 {built-in method builtins.min}

2.0:

ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000    4.967    4.967 clifford_decompose_layers.py:417(synth_clifford_depth_lnn)
        1    0.000    0.000    4.967    4.967 clifford_decompose_layers.py:67(synth_clifford_layers)
     13/6    0.000    0.000    3.085    0.514 clifford.py:149(__init__)
        1    0.000    0.000    3.084    3.084 clifford.py:672(from_circuit)
      7/1    0.243    0.035    3.084    3.084 clifford_circuits.py:25(_append_circuit)
 152254/6    0.244    0.000    3.083    0.514 clifford_circuits.py:53(_append_operation)
   119172    1.010    0.000    1.735    0.000 clifford_circuits.py:401(_append_cx)
        1    0.007    0.007    0.797    0.797 clifford_decompose_layers.py:249(_decompose_graph_state)
        1    0.000    0.000    0.792    0.792 clifford_decompose_layers.py:312(_decompose_hadamard_free)
        1    0.000    0.000    0.765    0.765 cx_cz_depth_lnn.py:36(synth_cx_cz_depth_line_my)
        1    0.764    0.764    0.764    0.764 {py_synth_cx_cz_depth_line_my}
        1    0.000    0.000    0.649    0.649 cz_depth_lnn.py:33(synth_cz_depth_line_mr)
        1    0.648    0.648    0.648    0.648 {synth_cz_depth_line_mr}
   279257    0.200    0.000    0.462    0.000 clifford.py:285(z)
   850215    0.209    0.000    0.408    0.000 base_operator.py:91(num_qubits)
   291626    0.153    0.000    0.299    0.000 clifford.py:276(x)
   152254    0.096    0.000    0.285    0.000 clifford_circuits.py:48(<listcomp>)
586413/586379    0.113    0.000    0.227    0.000 {built-in method builtins.isinstance}
   850215    0.199    0.000    0.199    0.000 op_shape.py:105(num_qubits)
   272820    0.132    0.000    0.189    0.000 quantumcircuit.py:3127(find_bit)

It seems the bottleneck now is handling the large number of append_operation applications.