一个带LRU的哈希缓存的实现

在做三角网格文件转换的时候需要对从文件中读取的顶点坐标进行缓存,于是索性做了一个哈希缓存以提高访问效率,同时以LRU原则在缓存满的情况下清楚缓存中的单元。在hash表中我用了最简单的取摸的方法,由于原顶点有一个索引而且这种索引连续,所以取摸的方法已经可以达到平均散列。在处理碰撞的时候我使用了链表,而所有在缓存中的单元都按放入缓存的时间串成一个链表,这样在清除单元的时候选择队尾的元素进行清除即可。

/*
    manipulations for the vertex binary file
    there is a least recent used hash cache
    for reading vertices

    the hash bucket with lru list is like this

    _____________        _____          _____
    |bucket_head| ---->  |   | -> .. -> |   | -> Null
    _____________        _____          _____
    |bucket_head| ---->  |   | -> .. -> |   | -> Null
    _____________        
    |bucket_head| ---->  Null
    
    ...
    _____________        _____          _____
    |bucket_head| ---->  |   | -> .. -> |   | -> Null

    while all the nodes in the buckets are linked
    based on the time they are accessed, thus the
    newly accessed node are inserted into the head
    of the lru list, if it has already existed in
    the list, it will be deleted from the list first

    author: waytofall
*/

#include <fstream>
#include <iostream>
using namespace std;

class VertexBinary
{
    typedef struct __Vertex
    {
        float x, y, z;
    } __Vertex;

    typedef struct __CacheUnit
    {
        struct __CacheUnit *bucket_prev; // bucket pointer
        struct __CacheUnit *bucket_next;
        struct __CacheUnit *use_prev; // lru pointer
         struct __CacheUnit *use_next;
        __Vertex vert;
        unsigned int index;
    } __CacheUnit;

public:
    VertexBinary();
    ~VertexBinary();

    /* write */
    int openForWrite(const char* filename);
    int closeWriteFile();
    int writeVertexFloat(float x, float y, float z);

    /* read */
    int initCache(unsigned int size);
    int openForRead(const char* filename);
    int indexedRead(unsigned int index);
    int closeReadFile();
    float getXFloat();
    float getYFloat();
    float getZFloat();

    void writeCacheDebug();

private:
    void insertBucketList(__CacheUnit **head, __CacheUnit *new_unit);
    void deleteBucketList(__CacheUnit **head, __CacheUnit *unit);
    void insertLruList(__CacheUnit *new_unit);
    void deleteLruList(__CacheUnit *unit);
    int indexedReadFromFile(unsigned int index, float &x, float &y, float &z);

private:
    ifstream fin;
    ofstream fout;
    unsigned int cache_size;
    unsigned int cache_count;
    __CacheUnit **cache_bucket;
    __CacheUnit *lru_head;
    __CacheUnit *lru_tail;
    __CacheUnit *current_unit;

public:
    unsigned int read_count;
    unsigned int hit_count;
};
/*
    manipulations for the vertex binary file
    author: waytofall
*/

#include "vert_bin.h"
#include <fstream>
#include <iostream>

// if dump the cache to the file
//#define _WRITE_CACHE_DEBUG

VertexBinary::VertexBinary()
{
    cache_bucket = NULL;
    cache_size = 0;
    cache_count = 0;
}

VertexBinary::~VertexBinary()
{
    if (cache_bucket)
    {
        delete[] cache_bucket;
    }
}

int VertexBinary::openForWrite(const char *filename)
{
    fout.open(filename, ios::out | ios::binary);

    if(fout.good())
        return 1;
    else
        return 0;
}

int VertexBinary::writeVertexFloat(float x, float y, float z)
{
    if (!fout.good())
    {
        return 0;
    }

    fout.write((char*)&x, sizeof(float));
    fout.write((char*)&y, sizeof(float));
    fout.write((char*)&z, sizeof(float));

    return 1;
}

int VertexBinary::openForRead(const char* filename)
{
    fin.open(filename, ios::binary | ios::in);
    if (fin.good())
        return 1;
    else
        return 0;
}

int VertexBinary::indexedRead(unsigned int index)
{
    if (cache_size <= 0)
    {
        cerr << "#error : no cache while indexed reading" << endl;
        return 0;
    }
    
    read_count ++;

    int hash_index = index % cache_size;
    __CacheUnit *hit_unit = NULL;

    // check if the cache hit
    if(cache_bucket[hash_index])
    {
        __CacheUnit *bucket_head;
        for(bucket_head = cache_bucket[hash_index]; bucket_head; bucket_head = bucket_head->bucket_next)
        {
            if (bucket_head->index == index)
            {
                hit_unit = bucket_head;
                break;
            }
        }
    }

    // the the cache has the queried unit
    if (hit_unit)
    {
        deleteLruList(hit_unit);
        insertLruList(hit_unit);
        current_unit = hit_unit;
        hit_count ++;
        return 1;
    }

    // the cache doesn't store the queried unit

    // check if needed to delete the least recent used unit
    if (cache_count == cache_size)
    {
        // delete the tail unit from the two lists
        hit_unit = lru_tail;
        deleteBucketList(&cache_bucket[hit_unit->index % cache_size], hit_unit);
        deleteLruList(hit_unit);
    }
    else
    {
        hit_unit =  new __CacheUnit();
        cache_count ++;
    }

    hit_unit->index = index;
    indexedReadFromFile(index, hit_unit->vert.x, hit_unit->vert.y, hit_unit->vert.z);
    insertBucketList(&cache_bucket[hash_index], hit_unit);
    insertLruList(hit_unit);
    current_unit = hit_unit;

    #ifdef _WRITE_CACHE_DEBUG
    writeCacheDebug();
    #endif

    return 1;
}

int VertexBinary::initCache(unsigned int size)
{
    if(size <= 0)
        return 0;

    cache_size = size;
    cache_bucket = new __CacheUnit*[cache_size];

    for(unsigned int i = 0; i < cache_size; i ++)
    {
        cache_bucket[i] = NULL;
    }

    cache_count = 0;
    lru_head = NULL;
    lru_tail = NULL;

    read_count = 0; 
    hit_count = 0;

    return 1;
}

// insert from head
void VertexBinary::insertBucketList(__CacheUnit **head, __CacheUnit *new_unit)
{
    new_unit->bucket_prev = NULL;
    new_unit->bucket_next = *head;
    if (*head)
    {
        (*head)->bucket_prev = new_unit;
    }
    (*head) = new_unit;
}

// delete from any place
void VertexBinary::deleteBucketList(__CacheUnit **head, __CacheUnit *unit)
{
    if(*head == NULL)
        return;

    if(*head == unit)
    {
        *head = unit->bucket_next;
    }
    if (unit->bucket_prev)
    {
        unit->bucket_prev->bucket_next = unit->bucket_next;
    }
    if (unit->bucket_next)
    {
        unit->bucket_next->bucket_prev = unit->bucket_prev;
    }
}

void VertexBinary::insertLruList(__CacheUnit *new_unit)
{
    new_unit->use_prev = NULL;
    new_unit->use_next = lru_head;
    if (lru_head)
    {
        lru_head->use_prev = new_unit;
    }

    if (lru_head == NULL)
    {
        lru_tail = new_unit;
    }

    lru_head = new_unit;
}

// delete the lru list from the tail
void VertexBinary::deleteLruList(__CacheUnit *unit)
{
    if (lru_head == NULL)
        return;

    if (unit->use_next)
    {
        unit->use_next->use_prev = unit->use_prev;
    }
    if (unit->use_prev)
    {
        unit->use_prev->use_next = unit->use_next;
    }
    
    if (lru_head == unit)
    {
        lru_head = unit->use_next;
    }
    if (lru_tail == unit)
    {
        lru_tail = unit->use_prev;
    }
}

int VertexBinary::indexedReadFromFile(unsigned int index, float &x, float &y, float &z)
{
    if (!fin.good())
    {
        return 0;
    }

    fin.seekg(index * 3 * sizeof(float));
    fin.read((char*)&x, sizeof(float));
    fin.read((char*)&y, sizeof(float));
    fin.read((char*)&z, sizeof(float));

    return 1;
}

int VertexBinary::closeWriteFile()
{
    fout.close();

    return 1;
}

int VertexBinary::closeReadFile()
{
    fin.close();

    return 1;
}

float VertexBinary::getXFloat()
{
    return current_unit->vert.x;
}

float VertexBinary::getYFloat()
{
    return current_unit->vert.y;
}

float VertexBinary::getZFloat()
{
    return current_unit->vert.z;
}

void VertexBinary::writeCacheDebug()
{
    int i, c1, c2;
    using namespace std;
    ofstream fout ("cache_dump.txt", ios::out | ios::app);
    __CacheUnit *pUnit;

    for(i = 0, c1 = 0; i < cache_size; i ++)
    {
        if (cache_bucket[i] == NULL)
            continue;

        fout << "bucket #" << i << ": ";
        for(pUnit = cache_bucket[i], c2 = 0; pUnit; pUnit = pUnit->bucket_next, c2 ++)
        {
            fout << pUnit->index << " ";
        }
        fout << "count: " << c2 << " " << endl;
        c1 += c2;
    }

    fout << "total count: " << c1 << endl << "lru list: ";
    for(c1 = 0, pUnit = lru_head; pUnit; pUnit = pUnit->use_next, c1 ++)
    {
        fout << pUnit->index << " ";
    }

    fout << endl << "lru count: " << c1 << endl << endl;

    fout.close();
}

一个大小为20的cache的打印结果

bucket #0: 420 count: 1 
bucket #1: 421 261 301 count: 3 
bucket #2: 422 count: 1 
bucket #5: 425 count: 1 
bucket #6: 306 446 386 count: 3 
bucket #8: 368 448 count: 2 
bucket #9: 409 count: 1 
bucket #10: 410 390 count: 2 
bucket #11: 411 count: 1 
bucket #12: 412 count: 1 
bucket #14: 434 count: 1 
bucket #16: 436 count: 1 
bucket #17: 297 count: 1 
bucket #19: 259 count: 1 
total count: 20
lru list: 412 306 409 368 410 422 420 421 436 434 259 261 446 448 411 386 390 297 425 301 
lru count: 20



____________________________
本博客文章主要供博主学习交流用,所有描述、代码无法保证准确性,如有问题可以留言共同讨论。
原文地址:https://www.cnblogs.com/waytofall/p/2532536.html