代码

#include <bits/stdc++.h>

using namespace std;

void sort(uint* a, int __n) {
    size_t n = __n;
    uint buc[4][256] = {};
    uint* b = (uint*)malloc(n * sizeof(uint));
    for (size_t i = 0; i < n; i++) {
        buc[0][a[i] & 255]++;
        buc[1][a[i] >> 8 & 255]++;
        buc[2][a[i] >> 16 & 255]++;
        buc[3][a[i] >> 24]++;
    }
    for (size_t k = 0; k < 4; k++) {
        uint32_t offset = 0;
        for (size_t i = 0; i < 256; i++)
            swap(buc[k][i], offset), offset += buc[k][i];
    }
    for (size_t i = 0; i < n; i++) b[buc[0][a[i] & 255]++] = a[i];
    for (size_t i = 0; i < n; i++) a[buc[1][b[i] >> 8 & 255]++] = b[i];
    for (size_t i = 0; i < n; i++) b[buc[2][a[i] >> 16 & 255]++] = a[i];
    for (size_t i = 0; i < n; i++) a[buc[3][b[i] >> 24 & 255]++] = b[i];
}

评测结果