代码

#include <bits/stdc++.h>

using u32 = unsigned int;
using u64 = unsigned long long;

void sort(u32 *a, int n) {
  static u32 cnt[256];
  u32 *b = new u32[n];
  for (int _ = 0; _ < 32; _ += 8) {
    memset(cnt, 0, sizeof(cnt));
#pragma GCC unroll(8)
    for (int i = 0; i < n; i++)
      cnt[a[i] >> _ & 255]++;
    for (int i = 0; i < 255; i++)
      cnt[i + 1] += cnt[i];
    for (int i = n - 1; ~i; i--)
      b[--cnt[a[i] >> _ & 255]] = a[i];
    memcpy(a, b, n * sizeof(u32));
  }
}

评测结果