代码

unsigned long long A[1<<26];
void sort(unsigned *a, int n) {
	for (int i=0; i<n; ++i) asm("bts %1, %0": "=m"(*A): "r"(a[i]));
	for (int i=0; i<1<<26; ++i) {
		unsigned long long t = A[i];
		while (t) {
			*a++ = i * 64u + __builtin_ctzll(t); t &= t-1; }}
}

评测结果