Hacker‘s Delight中的Trick

Hacker’s Delight 里的各种技巧真有意思。

乘法

Chapter8，32位符号整型乘积的高32位：

int mulhs(int u, int v) {
	unsigned int u0, v0, w0;
	int u1, v1, w1, w2, t;
	u0 = u & 0xFFFF; u1 = u >> 16;
	v0 = v & 0xFFFF; v1 = v >> 16;
	w0 = u0*v0;
	t = u1*v0 + (w0 >> 16);
	w1 = t & 0xFFFF;
	w2 = t >> 16;
	w1 = u0*v1 + w1;
	return u1*v1 + w2 + (w1 >> 16);
}

unsigned int mulhs(unsigned int u, unsigned int v) {
	unsigned int u0, v0, w0;
	unsigned int u1, v1, w1, w2, t;
	u0 = u & 0xFFFF; u1 = u >> 16;
	v0 = v & 0xFFFF; v1 = v >> 16;
	w0 = u0*v0;
	t = u1*v0 + (w0 >> 16);
	w1 = t & 0xFFFF;
	w2 = t >> 16;
	w1 = u0*v1 + w1;
	return u1*v1 + w2 + (w1 >> 16);
}

拼凑一下可以算出完整的64位乘积：

long long int mul32(int u, int v)
{
	int h = mulhs(u, v);
	int l = u * v;
	return ((long long int)h << 32 | l);
}

long long unsigned int mul32(unsigned int u, unsigned int v)
{
	unsigned int h = mulhs(u, v);
	unsigned int l = u * v;
	return ((long long unsigned int)h << 32 | l);
}

后面的习题提到低32位的运算用有符号或者无符号都是相同的，测试了一下确实如此。

for a 32×32 ⇒ 64 bit multiplication, the low-order 32 bits of the product are the same whether the operands are interpreted as signed or unsigned integers.

这种把长位宽的运算拆成高低两部分分别计算的方法，在用SIMD做优化的时候可能会有用，因为一些SIMD指令集并没有现成的长位宽运算指令，只好自行拼凑了。