パターン撃退クイックソートを使用する

パターン撃退クイックソート (pattern-defeating quicksort) は、イントロソートを拡張したハイブリッドな比較ソートである。

Rust の標準ライブラリや Go 1.19 以降の sort パッケージなどでも採用されている。

基本はクイックソートだが、次の仕掛けで「悪い入力パターン」を撃退する。

  1. ピボット選択: 小区間は 3 要素の中央値、大きめの区間は Tukey の ninther(9 要素の疑似中央値)を使う。
  2. 等値の扱い: 直前の分割でピボットと等しい値が左端に接しているときは、等値を左へ寄せる分割に切り替え、等値だらけの区間を再帰から外す。
  3. 不均衡の検知: 左右の部分配列が元サイズの 1/8 未満に偏ったら、候補位置を入れ替えてパターンを崩す。偏りが ⌊log₂ n⌋ 回続くとヒープソートへフォールバックする。
  4. ほぼ整列の検知: 分割がほとんど動かなかった区間には、移動回数に上限付きの部分挿入ソートを試し、成功すればその場で打ち切る。
  5. 小区間: 要素数が閾値未満なら挿入ソートで仕上げる。
procedure pdqsort(A, lo, hi, bad_allowed, leftmost)
  while lo <= hi
    size = hi - lo + 1
    if size < INSERTION_THRESHOLD then
      insertion_sort(A, lo, hi)
      return
    choose_pivot_median3_or_ninther(A, lo, hi)
    if not leftmost and A[lo - 1] = A[lo] then
      lo = partition_left(A, lo, hi) + 1
      continue
    pivot_pos, already = partition_right(A, lo, hi)
    l = pivot_pos - lo
    r = hi - pivot_pos
    if l < size / 8 or r < size / 8 then
      if bad_allowed = 0 then
        heapsort(A, lo, hi)
        return
      bad_allowed = bad_allowed - 1
      shuffle_candidates(A, lo, pivot_pos, hi)
    else if already and partial_insertion(A, lo, pivot_pos - 1)
                    and partial_insertion(A, pivot_pos + 1, hi) then
      return
    pdqsort(A, lo, pivot_pos - 1, bad_allowed, leftmost)
    lo = pivot_pos + 1
    leftmost = false

procedure sort(A)
  if length(A) > 1 then
    pdqsort(A, 0, length(A) - 1, floor(log2(length(A))), true)

平均計算量は O(n log n)、最悪もヒープソートへの切り替えにより O(n log n) に抑えられる。一般に不安定である。

類似アルゴリズムとの相違点

イントロソートは再帰深度の上限でヒープソートへ切り替えるのに対し、 パターン撃退クイックソートは「分割の偏り」を数え、加えて等値の片寄せや部分挿入による早期終了、 不均衡時の候補シャッフルで悪パターン自体を崩す。

デュアルピボットクイックソートは 1 回の走査で 3 分割する単一アルゴリズムの改良だが、 パターン撃退版は単一ピボットのままハイブリッド戦略で最悪ケースと現実的な入力パターンの両方に備える。

計算時間量および空間計算量を計測する

Size Average time Maximum time Average memory Maximum memory
256 0.000011 0.000206 62 68
512 0.000024 0.000537 82 88
1024 0.000054 0.000345 58 64
2048 0.000115 0.003080 82 88
4096 0.000247 0.001553 62 68
8192 0.000574 0.006052 58 64
16384 0.001200 0.015301 62 68
32768 0.002701 0.084036 58 64
65536 0.005330 0.022071 57 64
131072 0.010897 0.051939 70 76
262144 0.022784 0.073584 58 64
計測に使用したコードを表示する

set -euo pipefail

WORKDIR="$(mktemp -d)"
trap 'rm -rf "$WORKDIR"' EXIT

cat > "$WORKDIR/Dockerfile" <<'EOF'
FROM rust:1.95.0

WORKDIR /app

RUN mkdir -p src

RUN cat > Cargo.toml <<'CARGO'
[package]
name = "rust-benchmark"
version = "0.1.0"
edition = "2021"

[profile.release]
lto = true
codegen-units = 1
panic = "abort"
CARGO

RUN cat > src/main.rs <<'RUST'
use std::{
    env,
    process::Command,
    time::{Duration, Instant},
};
const MIN_POWER: u32 = 8;
const MAX_POWER: u32 = 18;
const RUNS: usize = 8192;
fn insertion_sort(a: &mut [usize]) {
    for i in 1..a.len() {
        let mut j = i;
        while j > 0 && a[j - 1] > a[j] {
            a.swap(j - 1, j);
            j -= 1;
        }
    }
}

fn sift_down(a: &mut [usize], mut root: usize, end: usize) {
    loop {
        let child = root * 2 + 1;
        if child > end {
            break;
        }
        let mut swap_idx = child;
        if child < end && a[child] < a[child + 1] {
            swap_idx = child + 1;
        }
        if a[root] >= a[swap_idx] {
            break;
        }
        a.swap(root, swap_idx);
        root = swap_idx;
    }
}

fn heap_sort(a: &mut [usize]) {
    if a.len() <= 1 {
        return;
    }
    for start in (0..a.len() / 2).rev() {
        sift_down(a, start, a.len() - 1);
    }
    for end in (1..a.len()).rev() {
        a.swap(0, end);
        sift_down(a, 0, end - 1);
    }
}



const INSERTION_SORT_THRESHOLD: usize = 24;
const NINTHER_THRESHOLD: usize = 128;
const PARTIAL_INSERTION_SORT_LIMIT: usize = 8;

fn floor_log2(n: usize) -> usize {
    usize::BITS as usize - n.leading_zeros() as usize - 1
}

fn sort2(a: &mut [usize], i: usize, j: usize) {
    if a[j] < a[i] {
        a.swap(i, j);
    }
}

fn sort3(a: &mut [usize], i: usize, j: usize, k: usize) {
    sort2(a, i, j);
    sort2(a, j, k);
    sort2(a, i, j);
}

fn partial_insertion_sort_range(a: &mut [usize], lo: usize, hi: usize) -> bool {
    if hi <= lo {
        return true;
    }
    let mut limit = 0usize;
    for i in (lo + 1)..=hi {
        if a[i] < a[i - 1] {
            let tmp = a[i];
            let mut j = i;
            loop {
                a[j] = a[j - 1];
                j -= 1;
                if j == lo || !(tmp < a[j - 1]) {
                    break;
                }
            }
            a[j] = tmp;
            limit += i - j;
            if limit > PARTIAL_INSERTION_SORT_LIMIT {
                return false;
            }
        }
    }
    true
}

/// Partition `[lo, hi]` around pivot at `lo`. Equals go to the right.
/// Returns `(pivot_pos, already_partitioned)`.
fn partition_right_range(a: &mut [usize], lo: usize, hi: usize) -> (usize, bool) {
    let pivot = a[lo];
    let mut first = lo;
    let mut last = hi + 1;

    loop {
        first += 1;
        if !(a[first] < pivot) {
            break;
        }
    }

    if first - 1 == lo {
        loop {
            if first >= last {
                break;
            }
            last -= 1;
            if a[last] < pivot {
                break;
            }
        }
    } else {
        loop {
            last -= 1;
            if a[last] < pivot {
                break;
            }
        }
    }

    let already_partitioned = first >= last;
    while first < last {
        a.swap(first, last);
        loop {
            first += 1;
            if !(a[first] < pivot) {
                break;
            }
        }
        loop {
            last -= 1;
            if a[last] < pivot {
                break;
            }
        }
    }

    let pivot_pos = first - 1;
    a[lo] = a[pivot_pos];
    a[pivot_pos] = pivot;
    (pivot_pos, already_partitioned)
}

/// Partition `[lo, hi]` around pivot at `lo`. Equals go to the left.
fn partition_left_range(a: &mut [usize], lo: usize, hi: usize) -> usize {
    let pivot = a[lo];
    let mut first = lo;
    let mut last = hi + 1;

    loop {
        last -= 1;
        if !(pivot < a[last]) {
            break;
        }
    }

    if last + 1 == hi + 1 {
        loop {
            if first >= last {
                break;
            }
            first += 1;
            if pivot < a[first] {
                break;
            }
        }
    } else {
        loop {
            first += 1;
            if pivot < a[first] {
                break;
            }
        }
    }

    while first < last {
        a.swap(first, last);
        loop {
            last -= 1;
            if !(pivot < a[last]) {
                break;
            }
        }
        loop {
            first += 1;
            if pivot < a[first] {
                break;
            }
        }
    }

    a[lo] = a[last];
    a[last] = pivot;
    last
}

fn pattern_defeating_quick_sort_loop(
    a: &mut [usize],
    mut lo: usize,
    hi: usize,
    mut bad_allowed: usize,
    mut leftmost: bool,
) {
    while lo <= hi {
        let size = hi - lo + 1;
        if size < INSERTION_SORT_THRESHOLD {
            insertion_sort(&mut a[lo..=hi]);
            return;
        }

        let s2 = size / 2;
        if size > NINTHER_THRESHOLD {
            sort3(a, lo, lo + s2, hi);
            sort3(a, lo + 1, lo + (s2 - 1), hi - 1);
            sort3(a, lo + 2, lo + (s2 + 1), hi - 2);
            sort3(a, lo + (s2 - 1), lo + s2, lo + (s2 + 1));
            a.swap(lo, lo + s2);
        } else {
            sort3(a, lo + s2, lo, hi);
        }

        if !leftmost && !(a[lo - 1] < a[lo]) {
            lo = partition_left_range(a, lo, hi) + 1;
            if lo > hi {
                return;
            }
            continue;
        }

        let (pivot_pos, already_partitioned) = partition_right_range(a, lo, hi);
        let l_size = pivot_pos - lo;
        let r_size = hi - pivot_pos;
        let highly_unbalanced = l_size < size / 8 || r_size < size / 8;

        if highly_unbalanced {
            if bad_allowed == 0 {
                heap_sort(&mut a[lo..=hi]);
                return;
            }
            bad_allowed -= 1;

            if l_size >= INSERTION_SORT_THRESHOLD {
                a.swap(lo, lo + l_size / 4);
                a.swap(pivot_pos - 1, pivot_pos - l_size / 4);
                if l_size > NINTHER_THRESHOLD {
                    a.swap(lo + 1, lo + (l_size / 4 + 1));
                    a.swap(lo + 2, lo + (l_size / 4 + 2));
                    a.swap(pivot_pos - 2, pivot_pos - (l_size / 4 + 1));
                    a.swap(pivot_pos - 3, pivot_pos - (l_size / 4 + 2));
                }
            }

            if r_size >= INSERTION_SORT_THRESHOLD {
                a.swap(pivot_pos + 1, pivot_pos + (1 + r_size / 4));
                a.swap(hi, hi + 1 - r_size / 4);
                if r_size > NINTHER_THRESHOLD {
                    a.swap(pivot_pos + 2, pivot_pos + (2 + r_size / 4));
                    a.swap(pivot_pos + 3, pivot_pos + (3 + r_size / 4));
                    a.swap(hi - 1, hi - (r_size / 4));
                    a.swap(hi - 2, hi - (1 + r_size / 4));
                }
            }
        } else if already_partitioned
            && (pivot_pos == lo || partial_insertion_sort_range(a, lo, pivot_pos - 1))
            && (pivot_pos >= hi || partial_insertion_sort_range(a, pivot_pos + 1, hi))
        {
            return;
        }

        if pivot_pos > lo {
            pattern_defeating_quick_sort_loop(a, lo, pivot_pos - 1, bad_allowed, leftmost);
        }
        lo = pivot_pos + 1;
        leftmost = false;
        if lo > hi {
            return;
        }
    }
}

fn pattern_defeating_quick_sort(a: &mut [usize]) {
    if a.len() <= 1 {
        return;
    }
    let bad_allowed = floor_log2(a.len());
    pattern_defeating_quick_sort_loop(a, 0, a.len() - 1, bad_allowed, true);
}


fn benchmark_sort(array: &mut [usize]) {

    pattern_defeating_quick_sort(array);

}

fn is_non_decreasing(a: &[usize]) -> bool {
    a.windows(2).all(|w| w[0] <= w[1])
}

fn same_multiset(a: &[usize], b: &[usize]) -> bool {
    if a.len() != b.len() {
        return false;
    }

    let mut left = a.to_vec();
    let mut right = b.to_vec();
    left.sort_unstable();
    right.sort_unstable();
    left == right
}

fn check_correctness_case(label: &str, mut input: Vec<usize>) {
    let original = input.clone();

    benchmark_sort(&mut input);

    if !is_non_decreasing(&input) {
        panic!("correctness case {}: output is not sorted", label);
    }

    if !same_multiset(&input, &original) {
        panic!("correctness case {}: elements were lost or added", label);
    }
}

fn few_unique_values(size: usize, unique: usize, seed: u64) -> Vec<usize> {
    let mut state = seed;

    (0..size)
        .map(|_| {
            state ^= state << 13;
            state ^= state >> 7;
            state ^= state << 17;
            (state as usize % unique) + 1
        })
        .collect()
}

fn run_correctness_checks() {
    check_correctness_case("empty", vec![]);
    check_correctness_case("single", vec![42]);
    check_correctness_case("duplicates", vec![3, 1, 3, 2, 1, 2]);
    check_correctness_case("sorted", vec![1, 2, 3, 4, 5]);
    check_correctness_case("reverse", vec![5, 4, 3, 2, 1]);
    check_correctness_case("all_equal", vec![7, 7, 7, 7]);
    check_correctness_case("skewed_range", vec![1_000_000, 2, 1_000_001, 1, 999_999]);
    // Static-buffer Grail skips the in-buffer build when key collection is sparse
    // (ideal_buffer = false). Exercising that path catches regressions in buffer gating.
    check_correctness_case(
        "few_keys_len16",
        vec![2, 2, 2, 2, 2, 2, 2, 2, 4, 3, 1, 2, 3, 4, 1, 4],
    );
    for seed in 0..32 {
        check_correctness_case(
            &format!("few_keys_len32_seed_{seed}"),
            few_unique_values(32, 4, seed),
        );
    }
}


fn shuffled(size: usize, seed: u64) -> Vec<usize> {
    let mut v: Vec<usize> = (1..=size).collect();

    let mut state = seed;

    for i in (1..size).rev() {
        state ^= state << 13;
        state ^= state >> 7;
        state ^= state << 17;

        let j = (state as usize) % (i + 1);

        v.swap(i, j);
    }

    v
}

fn memory_usage_kb() -> usize {
    // VmHWM (peak RSS, KiB). Reported memory subtracts a per-size baseline that only
    // holds the input array, so the table reflects auxiliary space during sorting.
    let contents = std::fs::read_to_string("/proc/self/status")
        .unwrap_or_default();

    for line in contents.lines() {
        if let Some(rest) = line.strip_prefix("VmHWM:") {
            let kb = rest
                .split_whitespace()
                .next()
                .unwrap_or("0")
                .parse::<usize>()
                .unwrap_or(0);

            return kb;
        }
    }

    0
}

fn micros(d: Duration) -> u128 {
    d.as_micros()
}

fn input_array(size: usize, seed: u64) -> Vec<usize> {
    shuffled(size, seed)
}

fn run_baseline(size: usize) -> usize {
    let _hold = input_array(size, 1);
    memory_usage_kb()
}

fn run_once(size: usize, seed: usize) -> (u128, usize) {
    let mut array = input_array(size, seed as u64);

    let start = Instant::now();

    benchmark_sort(&mut array);

    let elapsed = start.elapsed();
    let mem = memory_usage_kb();

    let expected: Vec<usize> = (1..=size).collect();
    if array != expected {
        panic!(
            "sort failed with seed {} for size {}",
            seed,
            size
        );
    }

    (micros(elapsed), mem)
}

fn run_baseline_child(args: &[String]) {
    let size = args[2].parse::<usize>().expect("invalid size");
    let mem = run_baseline(size);
    println!("{}", mem);
}

fn run_child(args: &[String]) {
    let size = args[2].parse::<usize>().expect("invalid size");
    let seed = args[3].parse::<usize>().expect("invalid seed");
    let (elapsed_us, mem) = run_once(size, seed);
    println!("{} {}", elapsed_us, mem);
}

fn main() {
    let args: Vec<String> = env::args().collect();
    if args.get(1).is_some_and(|arg| arg == "--baseline-once") {
        run_baseline_child(&args);
        return;
    }
    if args.get(1).is_some_and(|arg| arg == "--run-once") {
        run_child(&args);
        return;
    }

    run_correctness_checks();

    println!(
        "| {:>10} | {:>15} | {:>15} | {:>15} | {:>15} |",
        "Size",
        "Average time",
        "Maximum time",
        "Average memory",
        "Maximum memory"
    );

    println!(
        "|{:-<11}:|{:-<16}:|{:-<16}:|{:-<16}:|{:-<16}:|",
        "",
        "",
        "",
        "",
        ""
    );

    for power in MIN_POWER..=MAX_POWER {
        let size = 1usize << power;

        let baseline_output = Command::new(env::current_exe().expect("failed to find current executable"))
            .arg("--baseline-once")
            .arg(size.to_string())
            .output()
            .expect("failed to run benchmark baseline process");

        if !baseline_output.status.success() {
            panic!(
                "benchmark baseline process failed: {}",
                String::from_utf8_lossy(&baseline_output.stderr)
            );
        }

        let baseline_stdout = String::from_utf8(baseline_output.stdout)
            .expect("baseline process returned non-UTF-8 output");
        let baseline_mem = baseline_stdout
            .split_whitespace()
            .next()
            .expect("missing baseline memory usage")
            .parse::<usize>()
            .expect("invalid baseline memory usage");

        let mut total_time: u128 = 0;
        let mut max_time: u128 = 0;

        let mut total_mem: usize = 0;
        let mut max_mem: usize = 0;

        for seed in 1..=RUNS {
            let output = Command::new(env::current_exe().expect("failed to find current executable"))
                .arg("--run-once")
                .arg(size.to_string())
                .arg(seed.to_string())
                .output()
                .expect("failed to run benchmark child process");

            if !output.status.success() {
                panic!(
                    "benchmark child process failed: {}",
                    String::from_utf8_lossy(&output.stderr)
                );
            }

            let stdout = String::from_utf8(output.stdout)
                .expect("child process returned non-UTF-8 output");
            let mut fields = stdout.split_whitespace();
            let elapsed_us = fields
                .next()
                .expect("missing elapsed time")
                .parse::<u128>()
                .expect("invalid elapsed time");
            let mem = fields
                .next()
                .expect("missing memory usage")
                .parse::<usize>()
                .expect("invalid memory usage");

            total_time += elapsed_us;

            if elapsed_us > max_time {
                max_time = elapsed_us;
            }

            let aux_mem = mem.saturating_sub(baseline_mem);

            total_mem += aux_mem;

            if aux_mem > max_mem {
                max_mem = aux_mem;
            }
        }

        let avg_time = total_time / RUNS as u128;
        let avg_mem = total_mem / RUNS;

        println!(
            "| {:>10} | {:>15} | {:>15} | {:>15} | {:>15} |",
            size,
            format!("{}.{:06}", avg_time / 1_000_000, avg_time % 1_000_000),
            format!("{}.{:06}", max_time / 1_000_000, max_time % 1_000_000),
            avg_mem,
            max_mem
        );
    }
}
RUST

RUN cargo build --release

CMD ["./target/release/rust-benchmark"]
EOF

docker build -t rust-benchmark "$WORKDIR"
docker run --rm --init rust-benchmark