CSAPP 第三版 第五章 家庭作业and so on

CSAPP 第三版 第五章 家庭作业
自己做的 仅供参考 可能出现错误

注：5.18 5.19 mark一下

5.13

A. 略
B. 浮点加法的延迟界限，CPE 3.00
C. 整数加法的延迟界限，CPE 3.00
D. 浮点数乘法与关键路径上的浮点数加法并行，浮点数乘法功能单元容量>1，所以浮点数乘法不会成为关键路径的阻碍。而关键路径上只有浮点加法，浮点数加法的延迟界限为3。

5.14

程序如下：

/* Inner product. Accumulate in temporary */  
void inner5(vec_ptr u, vec_ptr v, data_t *dest) {
	long i;
	long length = vec_length(u);
	long limit = length - 5;
	data_t *udata = get_ver_start(u);
	data_t *vdata = get_vec_start(v);
	data_t sum = (data_t)0;

	for (i = 0; i < limit; i += 6) {
		sum = sum + udata[i] * vdata[i] + 
			udata[i + 1] * vdata[i + 1] + 
			udata[i + 2] * vdata[i + 2] + 
			udata[i + 3] * vdata[i + 3] + 
			udata[i + 4] * vdata[i + 4] + 
			udata[i + 5] * vdata[i + 5];
	}
	for (; i < length; i++) {
		sum = sum + udata[i] * vdata[i];
	}
	*dest = sum;
}

A. 当加法的功能单元和乘法的功能单元全都处于满流水的状态CPE最低，即合并运算达到吞吐量下界。对于整数运算，加法的吞吐量下界为0.5，乘法的吞吐量下界为1.0，故CPE=max{0.5,1.0}；对于浮点数运算，加法的吞吐量下界是1.0，乘法的吞吐量下界是0.5，故CPE=max{1.0,0.5}=1.0。综上，CPE的下界是1.0。
B. 即使进行了 6 x 1 循环展开，但是还是要依次进行6次浮点加法，并没有减少内存读写的次数和流水线的发生，算下来单个元素还是需要3个时钟周期。

5.15

/* Inner product. Accumulate in temporary */  
void inner6(vec_ptr u, vec_ptr v, data_t *dest) {
	long i;
	long length = vec_length(u);
	long limit = length - 5;
	data_t *udata = get_ver_start(u);
	data_t *vdata = get_vec_start(v);
	data_t sum0 = (data_t)0;
	data_t sum1 = (data_t)0;
	data_t sum2 = (data_t)0;
	data_t sum3 = (data_t)0;
	data_t sum4 = (data_t)0;
	data_t sum5 = (data_t)0;

	for (i = 0; i < limit; i += 6) {
		sum0 = sum0 + udata[i] * vdata[i];
		sum1 = sum1 + udata[i + 1] * vdata[i + 1];
		sum2 = sum2 + udata[i + 2] * vdata[i + 2];
		sum3 = sum3 + udata[i + 3] * vdata[i + 3];
		sum4 = sum4 + udata[i + 4] * vdata[i + 4];
		sum5 = sum5 + udata[i + 5] * vdata[i + 5];
	}
	for (; i < length; i++) {
		sum0 = sum0 + udata[i] * vdata[i];
	}
	*dest = sum0 + sum1 + sum2 + sum3 + sum4 + sum5;
}

加载器的数量为2，而每个元素的循环就需要两个加载器。

5.16

/* Inner product. Accumulate in temporary */  
void inner7(vec_ptr u, vec_ptr v, data_t *dest) {
	long i;
	long length = vec_length(u);
	long limit = length - 5;
	data_t *udata = get_ver_start(u);
	data_t *vdata = get_vec_start(v);
	data_t sum = (data_t)0;
	
	for (i = 0; i < limit; i += 6) {
		sum = sum + (udata[i] * vdata[i] +
			(udata[i + 1] * vdata[i + 1] +
			(udata[i + 2] * vdata[i + 2] +
			(udata[i + 3] * vdata[i + 3] +
			(udata[i + 4] * vdata[i + 4] +
			udata[i + 5] * vdata[i + 5])))));
	}
	for (; i < length; i++) {
		sum = sum + udata[i] * vdata[i];
	}
	*dest = sum;
}

5.17

void *new_memset(void *s, int c, size_t n) {
	unsigned long w;
	unsigned char *lw = (unsigned char *)&w;
	size_t cnt = 0;
	while (cnt < K) {
		*lw++ = (unsigned char)c;
		cnt++;
	}
	
	size_t i;
	unsigned char *schar = s;
	for (i = 0; (size_t)schar % K != 0 || i == n; i++) {
		*schar++ = (unsigned char)c;
	}
	

	size_t limit = n - K + 1;
	for (; i < limit && (int)limit > 0; i += K) {
		*(unsigned long *)schar = w;
		schar += K;
	}

	for (; i < n; i++) {
		*schar++ = (unsigned char)c;
	}
	return s;
}

5.18

double poly_6_3a(double a[], double x, long degree) {
  long i = 1;
  double result0 = a[0];
  double result1 = 0;
  double result2 = 0;

  double xpwr0 = x;
  double xpwr1 = x * x * x;
  double xpwr2 = x * x * x * x * x;

  double xpwr_step = x * x * x * x * x * x;
  for (; i < degree - 5; i+=6) {
    result0 = result0 + (a[i] * xpwr0 + a[i + 1] * xpwr0 * x);
    result1 = result1 + (a[i + 2] * xpwr1 + a[i + 3] * xpwr1 * x);
    result2 = result2 + (a[i + 4] * xpwr2 + a[i + 5] * xpwr2 * x);
    xpwr0 *= xpwr_step;
    xpwr1 *= xpwr_step;
    xpwr2 *= xpwr_step;
  }

  for (; i <= degree; i++) {
    result0 = result0 + a[i] * xpwr0;
    xpwr *= x;
  }
  return result0 + result1 + result2;
}

5.19

void psum_4_1a(float a[], float p[], long n) {
	long i;
	float tmp0, tmp1, tmp2, tmp3 = 0;

	for (i = 0; i < n - 3; i += 4) {
		tmp0 = tmp3 + a[i];
		tmp1 = tmp0 + a[i + 1];
		tmp2 = tmp1 + a[i + 2];
		tmp3 = tmp2 + a[i + 3];

		p[i] = tmp0;
		p[i + 1] = tmp1;
		p[i + 2] = tmp2;
		p[i + 3] = tmp3;
	}
	for (; i < n; i++) {
		tmp3 += a[i];
		p[i] = tmp3;
	}
}