prism-q 0.35.0

Fast Rust quantum circuit simulator. OpenQASM 3.0, multiple backends, AVX2 SIMD kernels, optional CUDA and MPI, QEC tooling, Python bindings.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135

//! Three-path timing for the GPU crossover work: CPU direct, GPU direct
//! (`.with_gpu(ctx)`), GPU dispatched (`simulate(...).gpu(ctx)`).
//!
//! Runs each path three times per (circuit, qubits) pair and reports the minimum.
//! The "GPU direct" path bypasses fusion + decomposition + crossover; the "GPU
//! dispatched" path goes through them.
//!
//! Build + run (`cargo run --release --features "parallel gpu" --example bench_gpu_dispatch`).

use std::time::Instant;

use prism_q::backend::Backend;
use prism_q::circuits::{independent_bell_pairs, random_circuit};
use prism_q::{BackendKind, Circuit, StatevectorBackend, sim};

use prism_q::gpu::GpuContext;

const SEED: u64 = 0xDEAD_BEEF;
const REPS: usize = 3;

fn time_cpu_direct(circuit: &Circuit) -> f64 {
    let mut best = f64::INFINITY;
    for _ in 0..REPS {
        let mut b = StatevectorBackend::new(42);
        b.init(circuit.num_qubits, circuit.num_classical_bits)
            .unwrap();
        let start = Instant::now();
        for inst in &circuit.instructions {
            b.apply(inst).unwrap();
        }
        let _ = b.probabilities().unwrap();
        let t = start.elapsed().as_secs_f64();
        if t < best {
            best = t;
        }
    }
    best
}

fn time_cpu_dispatched(circuit: &Circuit) -> f64 {
    let mut best = f64::INFINITY;
    for _ in 0..REPS {
        let start = Instant::now();
        let _ = sim::simulate(circuit)
            .backend(BackendKind::Statevector)
            .seed(42)
            .run()
            .unwrap();
        let t = start.elapsed().as_secs_f64();
        if t < best {
            best = t;
        }
    }
    best
}

fn time_gpu_direct(circuit: &Circuit, ctx: &std::sync::Arc<GpuContext>) -> f64 {
    let mut best = f64::INFINITY;
    for _ in 0..REPS {
        let mut b = StatevectorBackend::new(42).with_gpu(ctx.clone());
        b.init(circuit.num_qubits, circuit.num_classical_bits)
            .unwrap();
        let start = Instant::now();
        for inst in &circuit.instructions {
            b.apply(inst).unwrap();
        }
        let _ = b.probabilities().unwrap();
        let t = start.elapsed().as_secs_f64();
        if t < best {
            best = t;
        }
    }
    best
}

fn time_gpu_dispatched(circuit: &Circuit, ctx: &std::sync::Arc<GpuContext>) -> f64 {
    let mut best = f64::INFINITY;
    for _ in 0..REPS {
        let start = Instant::now();
        let _ = sim::simulate(circuit)
            .gpu(ctx.clone())
            .seed(42)
            .run()
            .unwrap();
        let t = start.elapsed().as_secs_f64();
        if t < best {
            best = t;
        }
    }
    best
}

fn row(label: &str, circuit: Circuit, ctx: &std::sync::Arc<GpuContext>) {
    let cpu_d = time_cpu_direct(&circuit) * 1e3;
    let cpu_r = time_cpu_dispatched(&circuit) * 1e3;
    let gpu_d = time_gpu_direct(&circuit, ctx) * 1e3;
    let gpu_r = time_gpu_dispatched(&circuit, ctx) * 1e3;
    println!(
        "{:<18} {:>10.3} {:>10.3} {:>10.3} {:>10.3} {:>10.2}x",
        label,
        cpu_d,
        cpu_r,
        gpu_d,
        gpu_r,
        gpu_d / gpu_r,
    );
}

fn main() {
    let ctx = match GpuContext::new(0) {
        Ok(c) => c,
        Err(e) => {
            eprintln!("no usable GPU: {e}");
            return;
        }
    };

    println!(
        "{:<18} {:>10} {:>10} {:>10} {:>10} {:>10}",
        "config", "cpu_d(ms)", "cpu_r(ms)", "gpu_d(ms)", "gpu_r(ms)", "gpu_d/gpu_r"
    );
    println!("{}", "-".repeat(82));

    // Non-decomposable random circuits across the crossover boundary.
    for n in &[10usize, 14, 16, 18, 20] {
        row(
            &format!("random {n}q d10"),
            random_circuit(*n, 10, SEED),
            &ctx,
        );
    }

    // Decomposable circuit: 8 independent 2q Bell pairs at 16q total.
    row("bell_pairs 16q", independent_bell_pairs(8), &ctx);
}