Struct ThreadPool 

pub struct ThreadPool { /* private fields */ }

Available on crate feature parallel only.

A bounded Rayon pool that PRISM-Q runs inside, leaving the process-wide pool untouched.

Simulation entry points size the global Rayon pool on first use, which an embedding application may not want. Work run through ThreadPool::install uses this pool instead, and the global pool is neither built nor resized, so an application that installed its own keeps it.

Pool width is part of what a result depends on. Dense unitary evolution, seeded terminal sampling, and the stabilizer tableau are bitwise at any width, parallel reductions move by about 1e-12, and compiled (BTS) shot payloads differ between widths because the batched sampler splits shots by pool width. A narrower pool is therefore not a slower route to the same bytes for every result; the per-path contract is the determinism section of the threading architecture page.

Examples

use prism_q::{ThreadPool, run_qasm};

let qasm = r#"
    OPENQASM 3.0;
    include "stdgates.inc";
    qubit[2] q;
    h q[0];
    cx q[0], q[1];
"#;

let pool = ThreadPool::with_threads(2).expect("build pool");
let result = pool.install(|| run_qasm(qasm, 42)).expect("simulation");
let probs = result.probabilities.expect("no probabilities").to_vec();
assert!((probs[0] - 0.5).abs() < 1e-10);

Implementations

impl ThreadPool

pub fn with_threads(threads: usize) -> Result<Self>

Build a pool of threads workers, or of the Rayon default width when threads is 0.

Errors

Returns PrismError::InvalidParameter when the operating system refuses to spawn the workers.

pub fn install<T: Send>(&self, op: impl FnOnce() -> T + Send) -> T

Run op on this pool, blocking the calling thread until it returns.

Every Rayon kernel PRISM-Q reaches from op runs on these workers. Work that escapes op, such as a simulation handed to another thread, falls back to the global pool.

pub fn num_threads(&self) -> usize

Worker count, which is what rayon::current_num_threads reports inside install. Resolves the default width taken by with_threads(0).