/*
***6.23
*/
等价于求xr(1 - x)的最大值,由代数知识得x=0.5的时候取得最大。
/*
***6.24
*/
0.5 * 60 / 12000 * 1000 + 60 / 12000 * 1000 /500 + 3 = 5.51ms
/*
***6.25
*/
定位时间 = 3 + 2.5 = 5.5ms
A. 3072 / 500 * 5 + 5.5 = 36.22ms
B. 5.5 * 3072 = 16896ms
6.26,6.27 根据公式C = B * E * S即可推得,此处略之
/*
***6.28
*/
A. 0x1238, 0x1239, 0x123a, 0x123b
B. 0x8a4, 0x8a5, 0x8a6, 0x8a7 0x704, 0x705, 0x706, 0x707
/*
***6.29
*/
A. 0x1bdc, 0x1bdd, 0x1bde, 0x1bdf
B. 0xe34, 0xe35, 0xe36, 0xe37
C. 0x18f0, 0x18f1, 0x18f2, 0x18f3 0xb0, 0xb1, 0xb2, 0xb3
D. 不会命中
/*
***6.30
*/
B. 1.不命中,有效位为0
2.命中。在1中已加载
3.命中。值为D0
/*
***6.31
*/
A. C= E*B*S = 128
B. CO 最后两位
CI 除去最后两位的后三位
CT 前8位
/*
***6.32
*/
A. 0011100011000
B. CO 0x0
CI 0x6
CT 0x38
没有命中
/*
***6.33
*/
A. 1011011101100
B. CO 0x0
CI 0x3
CT 0xB7
没有命中
/*
***6.34
*/
0x1314, 0x1315, 0x1316, 0x1317
0x1794, 0x1795, 0x1796, 0x1797
/*
***6.35
*/
缓存一共有两个块,src[0],src[2],dst[0],dst[2]访问缓存第一个块,src[1],src[3],dst[1],dst[3]访问缓存第二个块。
dst数组
m h m h
m m h m
m h m h
m m h m
src数组
m m m m
m m m m
m m m m
m m m m
/*
***6.36
*/
当总大小为128时,能容纳下src与dst数组中的所有元素
dst数组
m h h h
m h h h
m h h h
m h h h
src数组
m h h h
m h h h
m h h h
m h h h
/*
***6.37
*/
A. x[0][i]和x[1][i]是同一个缓存条目。不命中率为100%,属于交叉不命中
B. 缓存大小可容纳数组所有内容。不命中率为1/8。
C. x[0][i]和x[1][i]加载到不同行。不命中率为1/8.
D. 不能,存在冷不命中。
E. 能。加大块大小能减小冷不命中概率。
/*
***6.38
*/
N=64时:
sumA: 1/4
sumB: 1
sumC: 1/2
N= 60时:
sumA ,sumB,sumC的缓存不命中率均为 1/4
比较难判断的是N = 60时sumB的缓存不命中率(sumC与sumB是一样的),我写了一个函数返回不命中次数,将形参n赋值60即可。
//高速缓存命中率函数,返回不命中次数
int noHitPercentage(int n)
{
//不命中的次数
int result = 0;
//总共要循环的次数
int count;
//存储块的标记位
int a[256];
for(int i =0;i < 256;i++)
{
a[i] = -1;
}
for(int j = 0;j < n;j++)
for(int i = 0;i < n;i++)
{
//求出这个数的相对索引
count = i * n + j;
//求这个索引对应的块号
int blockNo = (count/4) % 256;
//求出标记t
int t = (count/4)/256;
//如果标记位不相等则不明中
if(t != a[blockNo])
{
a[blockNo] = t;
result++;
}
}
return result;
}
/*
***6.39
*/
A. 16 * 16 * 4 = 1024
B. 64
C. 1/16
/*
***6.40
*/
A. 1024
B. 256
C. 1/4
/*
***6.41
*/
A. 1024
B. 64 + 64 = 128
C. 1/8
/*
***6.42
*/
25%
/*
***6.43
*/
25%
/*
***6.44
*/
100%
/*
***6.46
*/
void betterTranspose(int *dst,int *src,int dim)
{
<span style="white-space:pre"> </span>int i, j;
<span style="white-space:pre"> </span>int iCount,jCount;
<span style="white-space:pre"> </span>//以4 * 4 的方阵为单位依次计算,增加了写的缓存命中率,多个元素一起读写还减少了循环开销
<span style="white-space:pre"> </span>for(i = 0;i < dim - 3;i += 4)
<span style="white-space:pre"> </span>{
<span style="white-space:pre"> </span>iCount = i * dim;
<span style="white-space:pre"> </span>for(j = 0;j < dim - 3;j += 4)
<span style="white-space:pre"> </span>{
<span style="white-space:pre"> </span>jCount = j * dim;
<span style="white-space:pre"> </span>dst[jCount + i] = src[iCount + j]; //dst[j][i] = src[i][j]
<span style="white-space:pre"> </span>dst[jCount + i + 1] = src[iCount + dim + j]; //dst[j][i + 1] = src[i + 1][j]
<span style="white-space:pre"> </span>dst[jCount + i + 2] = src[iCount + dim * 2 + j]; //dst[j][i + 2] = src[i + 2][j]
<span style="white-space:pre"> </span>dst[jCount + i + 3] = src[iCount + dim * 3 + j]; //dst[j][i + 3] = src[i + 3][j]
<span style="white-space:pre"> </span>dst[jCount + dim + i] = src[iCount + j + 1]; //dst[j + 1][i] = src[i][j + 1]
<span style="white-space:pre"> </span>dst[jCount + dim + i + 1] = src[iCount + dim + j + 1]; //dst[j + 1][i + 1] = src[i + 1][j + 1]
<span style="white-space:pre"> </span>dst[jCount + dim + i + 2] = src[iCount + dim * 2 + j + 1]; //dst[j + 1][i + 2] = src[i + 2][j + 1]
<span style="white-space:pre"> </span>dst[jCount + dim + i + 3] = src[iCount + dim * 3 + j + 1]; //dst[j + 1][i + 3] = src[i + 3][j + 1]
<span style="white-space:pre"> </span>dst[jCount + dim * 2 + i] = src[iCount + j + 2]; //dst[j + 2][i] = src[i][j + 2]
<span style="white-space:pre"> </span>dst[jCount + dim * 2 + i + 1] = src[iCount + dim + j + 2]; //dst[j + 2][i + 1] = src[i + 1][j + 2]
<span style="white-space:pre"> </span>dst[jCount + dim * 2 + i + 2] = src[iCount + dim * 2 + j + 2]; //dst[j + 2][i + 2] = src[i + 2][j + 2]
<span style="white-space:pre"> </span>dst[jCount + dim * 2+ i + 3] = src[iCount + dim * 3 + j + 2]; //dst[j + 2][i + 3] = src[i + 3][j + 2]
<span style="white-space:pre"> </span>dst[jCount + dim * 3 + i] = src[iCount + j + 3]; //dst[j + 3][i] = src[i][j + 3]
<span style="white-space:pre"> </span>dst[jCount + dim * 3 + i + 1] = src[iCount + dim + j + 3]; //dst[j + 3][i + 1] = src[i + 1][j + 3]
<span style="white-space:pre"> </span>dst[jCount + dim * 3 + i + 2] = src[iCount + dim * 2 + j + 3]; //dst[j + 3][i + 2] = src[i + 2][j + 3]
<span style="white-space:pre"> </span>dst[jCount + dim * 3 + i + 3] = src[iCount + dim * 3 + j + 3]; //dst[j + 3][i + 3] = src[i + 3][j + 3]
<span style="white-space:pre"> </span>
<span style="white-space:pre"> </span>}
<span style="white-space:pre"> </span>}
<span style="white-space:pre"> </span>//记录当前行和列的索引,以便执行完剩余的项
<span style="white-space:pre"> </span>int curIndex = i;
<span style="white-space:pre"> </span>//处理剩余项,简单的交换处理
<span style="white-space:pre"> </span>for(i = 0;i < curIndex;i++)
<span style="white-space:pre"> </span>for(j = curIndex;j < dim;j++)
<span style="white-space:pre"> </span>{
<span style="white-space:pre"> </span>dst[j * dim + i] = src[i * dim + j];
<span style="white-space:pre"> </span>}
<span style="white-space:pre"> </span>for(i = curIndex;i < dim;i++)
<span style="white-space:pre"> </span>for(j = 0;j < dim;j++)
<span style="white-space:pre"> </span>{
<span style="white-space:pre"> </span>dst[j * dim + i] = src[i * dim + j];
<span style="white-space:pre"> </span>}
}
/*
***6.47
*/
void better_col_convert(int *G,int dim)
{
int i, j;
int iCount,jCount;
//以4 * 4 的方阵为单位依次计算,增加了写的缓存命中率,多个元素一起读写还减少了循环开销
for(i = 0;i < dim - 3;i += 4)
{
iCount = i * dim;
for(j = 0;j < dim - 3;j += 4)
{
jCount = j * dim;
G[jCount + i] = G[iCount + j] || G[jCount + i]; //G[j][i] = G[i][j] || G[j][i]
G[jCount + i + 1] = G[iCount + dim + j] || G[jCount + i + 1]; //G[j][i + 1] = G[i + 1][j] || G[j][i + 1]
G[jCount + i + 2] = G[iCount + dim * 2 + j] || G[jCount + i + 2]; //G[j][i + 2] = G[i + 2][j] || G[j][i + 2]
G[jCount + i + 3] = G[iCount + dim * 3 + j] || G[jCount + i + 3]; //G[j][i + 3] = G[i + 3][j] || G[j][i + 3]
G[jCount + dim + i] = G[iCount + j + 1] || G[jCount + dim + i]; //G[j + 1][i] = G[i][j + 1] || G[j + 1][i]
G[jCount + dim + i + 1] = G[iCount + dim + j + 1] || G[jCount + dim + i + 1]; //G[j + 1][i + 1] = G[i + 1][j + 1] || G[j +1][i + 1]
G[jCount + dim + i + 2] = G[iCount + dim * 2 + j + 1] || G[jCount + dim + i + 2]; //G[j + 1][i + 2] = G[i + 2][j + 1] || G[j +1][i + 2]
G[jCount + dim + i + 3] = G[iCount + dim * 3 + j + 1] || G[jCount + dim + i + 3]; //G[j + 1][i + 3] = G[i + 3][j + 1] || G[j + 1][i + 3]
G[jCount + dim * 2 + i] = G[iCount + j + 2] || G[jCount + dim * 2 + i]; //G[j + 2][i] = G[i][j + 2] || G[j +2][i]
G[jCount + dim * 2 + i + 1] = G[iCount + dim + j + 2] || G[jCount + dim * 2 + i +1]; //G[j + 2][i + 1] = G[i + 1][j + 2] || G[j +2][i + 1]
G[jCount + dim * 2 + i + 2] = G[iCount + dim * 2 + j + 2] || G[jCount + dim * 2 + i + 2]; //G[j + 2][i + 2] = G[i + 2][j + 2] || G[j +2][i + 2]
G[jCount + dim * 2+ i + 3] = G[iCount + dim * 3 + j + 2] || G[jCount + dim * 2 + i + 3]; //G[j + 2][i + 3] = G[i + 3][j + 2] || G[j + 2][i + 3]
G[jCount + dim * 3 + i] = G[iCount + j + 3] || G[jCount + dim * 3 + i]; //G[j + 3][i] = G[i][j + 3] || G[j +3][i]
G[jCount + dim * 3 + i + 1] = G[iCount + dim + j + 3] || G[jCount + dim * 3 + i + 1]; //G[j + 3][i + 1] = G[i + 1][j + 3] || G[j +3][i + 1]
G[jCount + dim * 3 + i + 2] = G[iCount + dim * 2 + j + 3] || G[jCount + dim * 3 + i + 2]; //G[j + 3][i + 2] = G[i + 2][j + 3] || G[j + 3][i + 2]
G[jCount + dim * 3 + i + 3] = G[iCount + dim * 3 + j + 3] || G[jCount + dim * 3 + i + 3]; //G[j + 3][i + 3] = G[i + 3][j + 3] || G[j + 3][i + 3]
}
}
//记录当前行和列的索引,以便执行完剩余的项
int curIndex = i;
//处理剩余项,简单的交换处理
for(i = 0;i < curIndex;i++)
for(j = curIndex;j < dim;j++)
{
G[j * dim + i] = G[i * dim + j] || G[j * dim + i];
}
for(i = curIndex;i < dim;i++)
for(j = 0;j < dim;j++)
{
G[j * dim + i] = G[i * dim + j] || G[j * dim + i];
}
}