DFPC的BDI压缩代码详解

最新推荐文章于 2024-01-07 02:09:27 发布

原创最新推荐文章于 2024-01-07 02:09:27 发布 · 687 阅读

0 ·

CC 4.0 BY-SA版权

内存同时被 2 个专栏收录

41 篇文章

订阅专栏

系统领域

26 篇文章

订阅专栏

本文深入探讨了一种缓存行压缩算法的具体实现过程，该算法通过对数据块进行多级基底压缩，寻找最佳压缩方案，以减少内存占用。文章详细讲解了如何将数据从字节转换为单词，对比不同压缩策略，选择最优压缩结果，并将其应用到NVMain请求中。

bool FRFCFS::BDICompress (NVMainRequest *request, uint64_t _blockSize, bool flag )
{
    //blocksize就是req的data的size，一般是64
    uint64_t * values = convertByte2Word(request, flag, _blockSize, 8);
    //把64个指向uint8的转换成8个word,一个word64bits
    //也就是论文中说的8*8bytes
    uint64_t bestCSize = _blockSize;
    uint64_t currCSize = _blockSize;
    uint64_t i, pos, bestPos;
    uint64_t words[35];
    uint64_t wordPos[35]; //0~8 chars
    
    uint64_t currWords[35];
    uint64_t currWordPos[35]; //0~8 chars
    bool comFlag = false;
    bestPos = 16;
    //看一条cacheline内的差别，比较value[0]~value[8]
    if( isSameValuePackable( values, _blockSize / 8))
    {
        currCSize = 8;
    }
    if(bestCSize > currCSize)
    {
        //这里我也不太清楚到底是哪种，大概是全0或者全相同中的一种
        //bestPos就=2，word[1]为value高32bits，word[2]低32bits
        bestCSize = currCSize;
        bestPos = bestCSize / 4;
        //假设这个value 1-8是一样的, bestPos = 8/4 =2
        words[0] = 0x0;
        wordPos[0] = 1;
        for(i = 0; i < bestPos; i++)
        {
            words[i+1] = (values[i/2] >> (32*(1-i%2))) & 0xFFFFFFFF;
            wordPos[i+1] = 8;
        }
        //words[1] = value[0]高32位
        //words[2] = value[0]低32位
        bestPos++;
        //然后bestPos=3
    }
    //传入values,8,1,8,....
    //总共有size = 8个，blimit=1,bsize=8为不压缩的大小
    currCSize = multBaseCompression( values, _blockSize / 8, 1, 8, currWords, currWordPos, pos);
    if(bestCSize > currCSize)
    {
        bestCSize = currCSize;
        bestPos = pos;
        words[0] = 0x1;
        wordPos[0] = 1;
        for(i = 0; i < bestPos; i++)
        {
            words[i+1] = currWords[i];
            wordPos[i+1] = currWordPos[i];
        }
        bestPos++;
    }
    currCSize = multBaseCompression( values, _blockSize / 8, 2, 8, currWords, currWordPos, pos);
    if(bestCSize > currCSize)
    {
        bestCSize = currCSize;
        bestPos = pos;
        words[0] = 0x2;
        wordPos[0] = 1;
        for(i = 0; i < bestPos; i++)
        {
            words[i+1] = currWords[i];
            wordPos[i+1] = currWordPos[i];
        }
        bestPos++;
    }
    currCSize = multBaseCompression( values, _blockSize / 8, 4, 8, currWords, currWordPos, pos);
    if(bestCSize > currCSize)
    {
        bestCSize = currCSize;
        bestPos = pos;
        words[0] = 0x3;
        wordPos[0] = 1;
        for(i = 0; i < bestPos; i++)
        {
            words[i+1] = currWords[i];
            wordPos[i+1] = currWordPos[i];
        }
        bestPos++;
    }
    free(values);
    values = convertByte2Word(request, flag, _blockSize, 4);
    if( isSameValuePackable( values, _blockSize / 4))
    {
        currCSize = 4;
    }
    if(bestCSize > currCSize)
    {
        bestCSize = currCSize;
        bestPos = bestCSize / 4;
        words[0] = 0x4;
        wordPos[0] = 1;
        for(i = 0; i < bestPos; i++)
        {
            words[i+1] = (values[i/2] >> (32*(1-i%2))) & 0xFFFFFFFF;
            wordPos[i+1] = 8;
        }
        bestPos++;
    }
    currCSize = multBaseCompression( values, _blockSize / 4, 1, 4, currWords, currWordPos, pos);
    if(bestCSize > currCSize)
    {
        bestCSize = currCSize;
        bestPos = pos;
        words[0] = 0x5;
        wordPos[0] = 1;
        for(i = 0; i < bestPos; i++)
        {
            words[i+1] = currWords[i];
            wordPos[i+1] = currWordPos[i];
        }
        bestPos++;
    }
    currCSize = multBaseCompression( values, _blockSize / 4, 2, 4, currWords, currWordPos, pos);
    if(bestCSize > currCSize)
    {
        bestCSize = currCSize;
        bestPos = pos;
        words[0] = 0x6;
        wordPos[0] = 1;
        for(i = 0; i < bestPos; i++)
        {
            words[i+1] = currWords[i];
            wordPos[i+1] = currWordPos[i];
        }
        bestPos++;
    }
    free(values);
    values = convertByte2Word(request, flag, _blockSize, 2);
    currCSize = multBaseCompression( values, _blockSize / 2, 1, 2, currWords, currWordPos, pos);
    if(bestCSize > currCSize)
    {
        bestCSize = currCSize;
        bestPos = pos;
        words[0] = 0x7;
        wordPos[0] = 1;
        for(i = 0; i < bestPos; i++)
        {
            words[i+1] = currWords[i];
            wordPos[i+1] = currWordPos[i];
        }
        bestPos++;
    }
    free(values);
    values = NULL;
    
    if(bestCSize < _blockSize)
    {
        comFlag = true;
        Word2Byte(request, flag, bestPos, bestCSize, words, wordPos);
    }
    
    return comFlag;

}