深入理解计算机系统 cache lab

实验梗概

这个实验分为Part A和Part B。在Part A里，你要编写一个cache模拟器。在Part B里，你要编写矩阵转置函数。资料:
CMU2015年的cache lab讲义
大佬的lab总结(（；´д｀）ゞ我太菜了，都是借鉴这位大佬的总结)




#Part A
编写cache模拟器：思路：
要使用getopt函数解析命令行: 上面的讲义有写：跳过

上面贴的cache lab讲义里有一些hint，根据这个，就可以大致写出来

定义cache_line的结构，其中valid_bits是有效位，tag是标记位，stamp是时间戳

typedef struct{
	int valid_bits;
	unsigned  tag;
	int stamp;
}cache_line;

定义一个cache[S][E]大小的二维数组(using malloc). 这样cache就模拟好了
下面是cache运行模拟：读入要访问的地址，再在cache里查找是否存在。
handle里有讲I是不用关的。所有现在focus on 其他操作

operator	what	do what
I	instruction load
L	data load	读access cache
S	data store	写access cache
M	data modify	又读又写 access cache两次

所以M直接fall through

	while(fscanf(file," %c %x,%d",&operation,&address,&size)>0){
		switch(operation){
			case 'L':
				update(address);
				break;
			case 'M':
				update(address);
			case 'S':
				update(address);
				break;
		}
		time();
	}

在handle里，block offset是不用关的，所以当得到一个地址时，按照上图的方式的到tag和set就行(分别对应了t_address和s_address)
然后就是正常的cache访问了，如果hit cache，hit次数加一。如果cache没有，就要用下一层缓冲中取出.这就是miss。如果当时的组中，每行都被使用，就要是驱逐evicting

void update(unsigned address){
	unsigned s_address =(address>>b) & ((0xffffffff)>>(32-s));//set`s index
	unsigned t_address = address>>(s+b);                     //tag`s index            
	for(int i=0;i<E;i++){
		if((*(cache+s_address)+i)->tag ==t_address){
			cache[s_address][i].stamp = 0;       //now ,this is used
			hit++;
			return;
		}
	}
	for(int i=0;i<E;i++){
		if(cache[s_address][i].valid_bits == 0){
			cache[s_address][i].tag = t_address;
			cache[s_address][i].valid_bits = 1;
			cache[s_address][i].stamp = 0;       //now ,this is load
			miss++;
			return;
		}
	}
	int max_stamp=0;
	int max_i;
	for(int i=0;i<E;i++){
		if(cache[s_address][i].stamp >max_stamp){
			max_stamp = cache[s_address][i].stamp;
			max_i = i;
		}
	}
	eviction++;
	miss++;
	cache[s_address][max_i].tag = t_address;
	cache[s_address][max_i].stamp = 0;	
}

最后说说时间戳:
每次访问一次cache,都会执行time函数，也就是把stamp的值加一。如果当前cache访问中，地址命中，则该stamp置为0，代表被使用。




完整代码:

#include "cachelab.h"
#include <getopt.h>
#include <stdlib.h>
#include <unistd.h>
#include <stdio.h>
#include <stddef.h>
typedef struct{
	int valid_bits;
	unsigned  tag;
	int stamp;
}cache_line;
char* filepath = NULL;
int s,E,b,S;           // s is set ,E  is line,each line have 2^b bits ,S is 2^s set
int hit=0,miss=0,eviction=0;
cache_line** cache = NULL;
void init(){
	cache = (cache_line**)malloc(sizeof(cache_line*)*S);             //malloc cache[S][E]
	for(int i=0;i<S;i++)
		*(cache+i) = (cache_line*)malloc(sizeof(cache_line)*E);
	for(int i=0;i<S;i++){
		for(int j=0;j<E;j++){
			cache[i][j].valid_bits = 0;      // set all valid_bits is zero
			cache[i][j].tag = 0xffffffff;           //no address
			cache[i][j].stamp = 0;            //time is 0;		
		}	
	}
}
void update(unsigned address){
	unsigned s_address =(address>>b) & ((0xffffffff)>>(32-s));                //set`s index
	unsigned t_address = address>>(s+b);                                 //tag`s index
	for(int i=0;i<E;i++){
		if((*(cache+s_address)+i)->tag ==t_address){
			cache[s_address][i].stamp = 0;       //now ,this is used
			hit++;
			return;
		}
	}
	for(int i=0;i<E;i++){
		if(cache[s_address][i].valid_bits == 0){
			cache[s_address][i].tag = t_address;
			cache[s_address][i].valid_bits = 1;
			cache[s_address][i].stamp = 0;       //now ,this is load
			miss++;
			return;
		}
	}
	int max_stamp=0;
	int max_i;
	for(int i=0;i<E;i++){
		if(cache[s_address][i].stamp >max_stamp){
			max_stamp = cache[s_address][i].stamp;
			max_i = i;
		}
	}
	eviction++;
	miss++;
	cache[s_address][max_i].tag = t_address;
	cache[s_address][max_i].stamp = 0;	
}
void time(){
	for(int i=0;i<S;i++){
		for(int j=0;j<E;j++){
			if(cache[i][j].valid_bits == 1)
				cache[i][j].stamp++;		
		}	
	}
}
int main(int argc,char *argv[])
{
	int opt;         
	while((opt = getopt(argc,argv,"s:E:b:t:")) !=-1){           //parse command line arguments
		switch(opt){
		case 's':
			s=atoi(optarg);
			break;
		case 'E':
			E=atoi(optarg);
			break;
		case 'b':
			b=atoi(optarg);
			break;
		case 't':
			filepath = optarg;
			break;
		}
	}
	S = 1<<s;
	init();
	FILE* file=fopen(filepath,"r");
	if(file == NULL){     // read trace file
		printf("Open file wrong");		
		exit(-1);
	}
	char operation;
	unsigned address;
	int size;	
	while(fscanf(file," %c %x,%d",&operation,&address,&size)>0){
		switch(operation){
			case 'L':
				update(address);
				break;
			case 'M':
				update(address);
			case 'S':
				update(address);
				break;
		}
		time();
	}
	for(int i=0;i<S;i++)                  //free cache[S][E]
		free(*(cache+i));
	free(cache);
	fclose(file);	                //close file	
    printSummary(hit,miss,eviction);
    return 0;
}

Part B

要求

编写矩阵转置函数，要求cache miss最小，分别会用这三种矩阵来测试:
$32*32 $,cache miss $<600$
$64\ast 64$,cache miss$<2000$
$61\ast 67$,cache miss$<3000$
测试的cache是$2^5$组，一行,block size为$2^5$bytes
强烈要看上面讲义：讲义里讲的就是这个Part的做法.

我的做法

简单来说：就是分块。
我没有分析,想看分析的可以看我上面贴的大佬的分析.(因为我也不会＞﹏＜)

32*32

分为8*8块。

char transpose1[] = "Transpose 1";
void transpose11(int M, int N, int A[N][M], int B[M][N])
{
	int i,j;
	int i1;
	int val1,val2,val3,val4,val5,val6,val7,val8;
	for(i=0;i<32;i+=8){
		for(j=0;j<32;j+=8){
			for(i1=i;i1<i+8;i1++){
					val1 = A[i1][j+0];
					val2 = A[i1][j+1];
					val3 = A[i1][j+2];
					val4 = A[i1][j+3];
					val5 = A[i1][j+4];
					val6 = A[i1][j+5];
					val7 = A[i1][j+6];
					val8 = A[i1][j+7];
					B[j+0][i1] = val1;
					B[j+1][i1] = val2;
					B[j+2][i1] = val3;
					B[j+3][i1] = val4;
					B[j+4][i1] = val5;
					B[j+5][i1] = val6;
					B[j+6][i1] = val7;
					B[j+7][i1] = val8;	
			}
		}
	}
}

64*64

分为4*4块

char transpose2[]="Transpose 2";
void transpose22(int M,int N,int A[N][M],int B[M][N])
{
	int i,j;
	int i1;
	int val1,val2,val3,val4,val5,val6,val7,val8;
	for(i=0;i<64;i+=4){
		for(j=0;j<64;j+=4){
			for(i1=i;i1<i+4;i1+=2){
				val1=A[i1][j+0];
				val2=A[i1][j+1];
				val3=A[i1][j+2];
				val4=A[i1][j+3];
				val5=A[i1+1][j+0];
				val6=A[i1+1][j+1];
				val7=A[i1+1][j+2];
				val8=A[i1+1][j+3];
				B[j+0][i1]=val1;
				B[j+1][i1]=val2;
				B[j+2][i1]=val3;
				B[j+3][i1]=val4;
				B[j+0][i1+1]=val5;
				B[j+1][i1+1]=val6;
				B[j+2][i1+1]=val7;
				B[j+3][i1+1]=val8;		
			}
		}
	}
}

61*67

分成16*16块

char transpose3[]="Transpose 3";
void transpose33(int M,int N,int A[N][M],int B[M][N])
{
	int ii,jj,i,j,val0,val1,val2,val3,val4,val5,val6,val7;
	for(ii = 0; ii + 16 < N; ii += 16)
		for(jj = 0; jj + 16 < M; jj += 16)
		{
			for(i = ii; i < ii + 16; i++)
			{
				val0 = A[i][jj + 0];
				val1 = A[i][jj + 1];
				val2 = A[i][jj + 2];
				val3 = A[i][jj + 3];
				val4 = A[i][jj + 4];
				val5 = A[i][jj + 5];
				val6 = A[i][jj + 6];
				val7 = A[i][jj + 7];
				B[jj + 0][i] = val0;
				B[jj + 1][i] = val1;
				B[jj + 2][i] = val2;
				B[jj + 3][i] = val3;
				B[jj + 4][i] = val4;
				B[jj + 5][i] = val5;
				B[jj + 6][i] = val6;
				B[jj + 7][i] = val7;

				val0 = A[i][jj + 8];
				val1 = A[i][jj + 9];
				val2 = A[i][jj + 10];
				val3 = A[i][jj + 11];
				val4 = A[i][jj + 12];
				val5 = A[i][jj + 13];
				val6 = A[i][jj + 14];
				val7 = A[i][jj + 15];
				B[jj + 8][i] = val0;
				B[jj + 9][i] = val1;
				B[jj + 10][i] = val2;
				B[jj + 11][i] = val3;
				B[jj + 12][i] = val4;
				B[jj + 13][i] = val5;
				B[jj + 14][i] = val6;
				B[jj + 15][i] = val7;

			}
		}
	for(i = ii; i < N; i++)
		for(j = 0; j < M; j++)
			B[j][i] = A[i][j];
	for(i = 0; i < ii; i++)
		for(j = jj; j < M; j++)
			B[j][i] = A[i][j];		
}

最后提交的函数:
把上面三个函数总结在一起即可。

char transpose_submit_desc[] = "Transpose submission";
void transpose_submit(int M, int N, int A[N][M], int B[M][N])
{
	if(M==32)
		return transpose11(32,32,A,B);
	else if(M==64)
		return transpose22(64,64,A,B);
	else if(M==61)
		return transpose33(M,N,A,B);
}

完结撒花