代码

#include <cstring>
void sort(unsigned *a, int n) {
    static unsigned h0[4096], h1[4096], h2[256];
    static unsigned tmp[1<<16];
    for (int i = 0; i < 4096; i++) { h0[i] = 0; h1[i] = 0; }
    for (int i = 0; i < 256; i++) h2[i] = 0;
    for (int i = 0; i < n; i++) { unsigned v = a[i]; h0[v & 4095]++; h1[(v >> 12) & 4095]++; h2[v >> 24]++; }
    unsigned s = 0;
    for (int i = 0; i < 4096; i++) { unsigned c = h0[i]; h0[i] = s; s += c; }
    s = 0; for (int i = 0; i < 4096; i++) { unsigned c = h1[i]; h1[i] = s; s += c; }
    s = 0; for (int i = 0; i < 256; i++) { unsigned c = h2[i]; h2[i] = s; s += c; }
    for (int i = 0; i < n; i++) { unsigned v = a[i]; tmp[h0[v & 4095]++] = v; }
    for (int i = 0; i < n; i++) { unsigned v = tmp[i]; a[h1[(v >> 12) & 4095]++] = v; }
    for (int i = 0; i < n; i++) { unsigned v = a[i]; tmp[h2[v >> 24]++] = v; }
    memcpy(a, tmp, n * 4);
}

评测结果