代码

#pragma optimize("unroll-loops")
typedef unsigned u32;
u32 cnt1[1 << 8], cnt2[1 << 8], cnt3[1 << 8], cnt4[1 << 8];
void sort(unsigned *a, int n)
{
	u32 *tmp = new u32[n];
	u32 *end= a + n;

    for (register u32 *i = a; i != a + n; i++)
		++cnt1[*i & 0xff],
		++cnt2[(*i >> 8) & 0xff],
		++cnt3[(*i >> 16) & 0xff],
		++cnt4[*i >> 24];
		
	for(register u32 *i = cnt1 + 1, *j = cnt2 + 1, *k = cnt3 + 1, *l = cnt4 + 1; i <= cnt1 + 0xff; ++i, ++j, ++k, ++l)
		*i += *(i - 1),
		*j += *(j - 1),
		*k += *(k - 1),
		*l += *(l - 1);

	for(register u32 *i= end - 1; i != a - 1; --i)
		tmp[--cnt1[*i & 0xff]] = *i;
	for(register u32 *i= tmp + n - 1; i != tmp - 1; --i)
		a[--cnt2[(*i >> 8) & 0xff]] = *i;
	for(register u32 *i= end - 1; i != a - 1; --i)
		tmp[--cnt3[(*i >> 16) & 0xff]] = *i;
	for(register u32 *i= tmp + n - 1; i != tmp - 1; --i)
		a[--cnt4[*i >> 24]] = *i;
	delete[] tmp;
}

评测结果