CSAPP: 极度舒适的Perflab
最近的lab和pj太多了5555
每天写一点叭…要加油鸭！！
Perflab
共两个函数 rotate & smooth
需要优化性能 CPE越低越好
but竟然没有评分标准的嘛…
5.15：rotate优化至13.9
原始版本：
性能：
改进1：32*32块划分
注意这里调换了原本的RIDX(dim-1-j, i, dim)的i/j顺序，性能提升不少
性能：
改进2：循环4路展开
性能：
改进3：32*32块+指针
but为什么用指针更好呢？应该是避免重复计算地址叭～
这里是我自己的构思，分32*32块
性能：
改进4：32*dim+指针
这是参考大佬的，32*dim一块，直接整行拷走
今日最佳：
几个很奇怪的地方：
1.同样的代码，大佬们跑出来能到20+？？（电脑的问题嘛…
2.指针一次取6个字节，不知道取内存时是不是4字节对齐性能更好，明天继续冲冲冲～
5.17 完结撒花～
发现大家得到的基准都不一样，所以应该是看加速比哒
首先对rotate做了最后的优化，即把所有dp++改为++dp（如果是dp++在汇编中会多用一个寄存器保存加之前的值）
性能：
然后是smooth函数：
原始版本：
void naive_smooth(int dim, pixel *src, pixel *dst)
{
int i, j;
for (i = 0; i < dim; i++)
for (j = 0; j < dim; j++)
dst[RIDX(i, j, dim)] = avg(dim, i, j, src);
}
性能：
改进1：减少对函数的调用，改为define
#define min(a, b) (a < b ? a : b)
#define max(a, b) (a > b ? a : b)
void smooth_inline(int dim, pixel *src, pixel *dst)
{
int i, j, i1, j1, max1, max2, min1, min2;
pixel_sum sum;
pixel tmp;
for (i = 0; i < dim; i++)
for (j = 0; j < dim; j++)
{ initialize_pixel_sum(&sum);
max1=max(i-1, 0); max2=max(j-1, 0);
min1=min(i+1, dim-1);min2=min(j+1, dim-1);
for(i1=max1; i1<=min1; i1++)
for(j1=max2; j1<=min2; j1++)
accumulate_sum(&sum, src[RIDX(i1, j1, dim)]);
assign_sum_to_pixel(&tmp, sum);
dst[RIDX(i, j, dim)]=tmp;
}
}
性能忘记截图了…改进不是很大
改进2：重写avg函数，不调用原始函数，按像素位置分别计算：
void smooth_avg(int dim, pixel *src, pixel *dst)
{
int i, j, tmp=dim+1, tmp2=2*dim-1, tmp3=dim*(dim-2), urc=dim-1, llc=dim*(dim-1), lrc=dim*dim-1, i1=0, j1=0;
//upper left corner
dst[0].red=(src[0].red+src[1].red+src[dim].red+src[tmp].red)>>2;
dst[0].blue=(src[0].blue+src[1].blue+src[dim].blue+src[tmp].blue)>>2;
dst[0].green=(src[0].green+src[1].green+src[dim].green+src[tmp].green)>>2;
//upper right corner
dst[urc].red=(src[urc].red+src[urc-1].red+src[tmp2].red+src[tmp2-1].red)>>2;
dst[urc].blue=(src[urc].blue+src[urc-1].blue+src[tmp2].blue+src[tmp2-1].blue)>>2;
dst[urc].green=(src[urc].green+src[urc-1].green+src[tmp2].green+src[tmp2-1].green)>>2;
//lower left corner
dst[llc].red=(src[llc].red+src[llc+1].red+src[tmp3].red+src[tmp3+1].red)>>2;
dst[llc].blue=(src[llc].blue+src[llc+1].blue+src[tmp3].blue+src[tmp3+1].blue)>>2;
dst[llc].green=(src[llc].green+src[llc+1].green+src[tmp3].green+src[tmp3+1].green)>>2;
//lower right corner
dst[lrc].red=(src[lrc].red+src[lrc-1].red+src[llc-1].red+src[llc-2].red)>>2;
dst[lrc].blue=(src[lrc].blue+src[lrc-1].blue+src[llc-1].blue+src[llc-2].blue)>>2;
dst[lrc].green=(src[lrc].green+src[lrc-1].green+src[llc-1].green+src[llc-2].green)>>2;
//upper side
for(j=1; j<dim-1; j++)
{dst[j].red=(src[j].red+src[j-1].red+src[j+1].red+src[j+dim].red+src[j+dim-1].red+src[j+dim+1].red)/6;
dst[j].blue=(src[j].blue+src[j-1].blue+src[j+1].blue+src[j+dim].blue+src[j+dim-1].blue+src[j+dim+1].blue)/6;
dst[j].green=(src[j].green+src[j-1].green+src[j+1].green+src[j+dim].green+src[j+dim-1].green+src[j+dim+1].green)/6;}
//lower side
for(j=llc+1; j<lrc; j++)
{dst[j].red=(src[j].red+src[j-1].red+src[j+1].red+src[j-dim].red+src[j-dim-1].red+src[j-dim+1].red)/6;
dst[j].blue=(src[j].blue+src[j-1].blue+src[j+1].blue+src[j-dim].blue+src[j-dim-1].blue+src[j-dim+1].blue)/6;
dst[j].green=(src[j].green+src[j-1].green+src[j+1].green+src[j-dim].green+src[j-dim-1].green+src[j-dim+1].green)/6;}
//left side
for(j=dim; j<llc; j+=dim)
{dst[j].red=(src[j].red+src[j-dim].red+src[j+dim].red+src[j+1].red+src[j-dim+1].red+src[j+dim+1].red)/6;
dst[j].blue=(src[j].blue+src[j-dim].blue+src[j+dim].blue+src[j+1].blue+src[j-dim+1].blue+src[j+dim+1].blue)/6;
dst[j].green=(src[j].green+src[j-dim].green+src[j+dim].green+src[j+1].green+src[j-dim+1].green+src[j+dim+1].green)/6;}
//right side
for(j=2*dim-1; j<lrc; j+=dim)
{dst[j].red=(src[j].red+src[j-1].red+src[j-dim].red+src[j+dim].red+src[j-dim-1].red+src[j+dim-1].red)/6;
dst[j].blue=(src[j].blue+src[j-1].blue+src[j-dim].blue+src[j+dim].blue+src[j-dim-1].blue+src[j+dim-1].blue)/6;
dst[j].green=(src[j].green+src[j-1].green+src[j-dim].green+src[j+dim].green+src[j-dim-1].green+src[j+dim-1].green)/6;}
//others
for(i=1; i<urc; i++)
{i1=i*dim;
for(j=1; j<dim-1; j++)
{j1=i1+j;
dst[j1].red=(src[j1].red+src[j1-1].red+src[j1+1].red+src[j1-dim].red+src[j1-dim-1].red+src[j1-dim+1].red+src[j1+dim].red+src[j1+dim-1].red+src[j1+dim+1].red)/9;
dst[j1].blue=(src[j1].blue+src[j1-1].blue+src[j1+1].blue+src[j1-dim].blue+src[j1-dim-1].blue+src[j1-dim+1].blue+src[j1+dim].blue+src[j1+dim-1].blue+src[j1+dim+1].blue)/9;
dst[j1].green=(src[j1].green+src[j1-1].green+src[j1+1].green+src[j1-dim].green+src[j1-dim-1].green+src[j1-dim+1].green+src[j1+dim].green+src[j1+dim-1].green+src[j1+dim+1].green)/9;}
}
}
性能有大幅提升啦！最后用点小聪明
改进3：中间部分像素2*2划分块，循环展开：
void smooth(int dim, pixel *src, pixel *dst)
{
int i, j, tmp=dim+1, tmp2=2*dim-1, tmp3=dim*(dim-2), urc=dim-1, llc=dim*(dim-1), lrc=dim*dim-1, i1=0, j1=0;
//upper left corner
dst[0].red=(src[0].red+src[1].red+src[dim].red+src[tmp].red)>>2;
dst[0].blue=(src[0].blue+src[1].blue+src[dim].blue+src[tmp].blue)>>2;
dst[0].green=(src[0].green+src[1].green+src[dim].green+src[tmp].green)>>2;
//upper right corner
dst[urc].red=(src[urc].red+src[urc-1].red+src[tmp2].red+src[tmp2-1].red)>>2;
dst[urc].blue=(src[urc].blue+src[urc-1].blue+src[tmp2].blue+src[tmp2-1].blue)>>2;
dst[urc].green=(src[urc].green+src[urc-1].green+src[tmp2].green+src[tmp2-1].green)>>2;
//lower left corner
dst[llc].red=(src[llc].red+src[llc+1].red+src[tmp3].red+src[tmp3+1].red)>>2;
dst[llc].blue=(src[llc].blue+src[llc+1].blue+src[tmp3].blue+src[tmp3+1].blue)>>2;
dst[llc].green=(src[llc].green+src[llc+1].green+src[tmp3].green+src[tmp3+1].green)>>2;
//lower right corner
dst[lrc].red=(src[lrc].red+src[lrc-1].red+src[llc-1].red+src[llc-2].red)>>2;
dst[lrc].blue=(src[lrc].blue+src[lrc-1].blue+src[llc-1].blue+src[llc-2].blue)>>2;
dst[lrc].green=(src[lrc].green+src[lrc-1].green+src[llc-1].green+src[llc-2].green)>>2;
//upper side
for(j=1; j<dim-1; ++j)
{dst[j].red=(src[j].red+src[j-1].red+src[j+1].red+src[j+dim].red+src[j+dim-1].red+src[j+dim+1].red)/6;
dst[j].blue=(src[j].blue+src[j-1].blue+src[j+1].blue+src[j+dim].blue+src[j+dim-1].blue+src[j+dim+1].blue)/6;
dst[j].green=(src[j].green+src[j-1].green+src[j+1].green+src[j+dim].green+src[j+dim-1].green+src[j+dim+1].green)/6;}
//lower side
for(j=llc+1; j<lrc; ++j)
{dst[j].red=(src[j].red+src[j-1].red+src[j+1].red+src[j-dim].red+src[j-dim-1].red+src[j-dim+1].red)/6;
dst[j].blue=(src[j].blue+src[j-1].blue+src[j+1].blue+src[j-dim].blue+src[j-dim-1].blue+src[j-dim+1].blue)/6;
dst[j].green=(src[j].green+src[j-1].green+src[j+1].green+src[j-dim].green+src[j-dim-1].green+src[j-dim+1].green)/6;}
//left side
for(j=dim; j<llc; j+=dim)
{dst[j].red=(src[j].red+src[j-dim].red+src[j+dim].red+src[j+1].red+src[j-dim+1].red+src[j+dim+1].red)/6;
dst[j].blue=(src[j].blue+src[j-dim].blue+src[j+dim].blue+src[j+1].blue+src[j-dim+1].blue+src[j+dim+1].blue)/6;
dst[j].green=(src[j].green+src[j-dim].green+src[j+dim].green+src[j+1].green+src[j-dim+1].green+src[j+dim+1].green)/6;}
//right side
for(j=2*dim-1; j<lrc; j+=dim)
{dst[j].red=(src[j].red+src[j-1].red+src[j-dim].red+src[j+dim].red+src[j-dim-1].red+src[j+dim-1].red)/6;
dst[j].blue=(src[j].blue+src[j-1].blue+src[j-dim].blue+src[j+dim].blue+src[j-dim-1].blue+src[j+dim-1].blue)/6;
dst[j].green=(src[j].green+src[j-1].green+src[j-dim].green+src[j+dim].green+src[j-dim-1].green+src[j+dim-1].green)/6;}
//others
for(i=1; i<urc-1; i+=2)
{i1=i*dim;
for(j=1; j<dim-2; j+=2)
{j1=i1+j;
dst[j1].red=(src[j1].red+src[j1-1].red+src[j1+1].red+src[j1-dim].red+src[j1-dim-1].red+src[j1-dim+1].red+src[j1+dim].red+src[j1+dim-1].red+src[j1+dim+1].red)/9;
dst[j1].blue=(src[j1].blue+src[j1-1].blue+src[j1+1].blue+src[j1-dim].blue+src[j1-dim-1].blue+src[j1-dim+1].blue+src[j1+dim].blue+src[j1+dim-1].blue+src[j1+dim+1].blue)/9;
dst[j1].green=(src[j1].green+src[j1-1].green+src[j1+1].green+src[j1-dim].green+src[j1-dim-1].green+src[j1-dim+1].green+src[j1+dim].green+src[j1+dim-1].green+src[j1+dim+1].green)/9;
j1=j1+1;
dst[j1].red=(src[j1].red+src[j1-1].red+src[j1+1].red+src[j1-dim].red+src[j1-dim-1].red+src[j1-dim+1].red+src[j1+dim].red+src[j1+dim-1].red+src[j1+dim+1].red)/9;
dst[j1].blue=(src[j1].blue+src[j1-1].blue+src[j1+1].blue+src[j1-dim].blue+src[j1-dim-1].blue+src[j1-dim+1].blue+src[j1+dim].blue+src[j1+dim-1].blue+src[j1+dim+1].blue)/9;
dst[j1].green=(src[j1].green+src[j1-1].green+src[j1+1].green+src[j1-dim].green+src[j1-dim-1].green+src[j1-dim+1].green+src[j1+dim].green+src[j1+dim-1].green+src[j1+dim+1].green)/9;
}
i1=(i+1)*dim;
for(j=1; j<dim-2; j+=2)
{j1=i1+j;
dst[j1].red=(src[j1].red+src[j1-1].red+src[j1+1].red+src[j1-dim].red+src[j1-dim-1].red+src[j1-dim+1].red+src[j1+dim].red+src[j1+dim-1].red+src[j1+dim+1].red)/9;
dst[j1].blue=(src[j1].blue+src[j1-1].blue+src[j1+1].blue+src[j1-dim].blue+src[j1-dim-1].blue+src[j1-dim+1].blue+src[j1+dim].blue+src[j1+dim-1].blue+src[j1+dim+1].blue)/9;
dst[j1].green=(src[j1].green+src[j1-1].green+src[j1+1].green+src[j1-dim].green+src[j1-dim-1].green+src[j1-dim+1].green+src[j1+dim].green+src[j1+dim-1].green+src[j1+dim+1].green)/9;
j1=j1+1;
dst[j1].red=(src[j1].red+src[j1-1].red+src[j1+1].red+src[j1-dim].red+src[j1-dim-1].red+src[j1-dim+1].red+src[j1+dim].red+src[j1+dim-1].red+src[j1+dim+1].red)/9;
dst[j1].blue=(src[j1].blue+src[j1-1].blue+src[j1+1].blue+src[j1-dim].blue+src[j1-dim-1].blue+src[j1-dim+1].blue+src[j1+dim].blue+src[j1+dim-1].blue+src[j1+dim+1].blue)/9;
dst[j1].green=(src[j1].green+src[j1-1].green+src[j1+1].green+src[j1-dim].green+src[j1-dim-1].green+src[j1-dim+1].green+src[j1+dim].green+src[j1+dim-1].green+src[j1+dim+1].green)/9;
}
}
}
性能：
最终结果：
这种没有明确评分的lab…就很虚( •̥́ ˍ •̀ू )
助教手下留情～
相比起前两个lab，感觉这个回到了上学期还在学基础理论的时候…
谁让机智的老师跳着上呢～
加油冲冲冲！