如何让我的boxblur代码更快地运行？

Question

我为灰色图像模糊写了一个盒子模糊代码，但结果比我预期的要慢得多。如何改善其表现？

void boxfilter(int* const srcImg, int* const desImg, const int mask, const int nrows, const int ncols)

{
    LL  *buffer,  *rowImg, *rowPtr;
    int *srcPtr, *desPtr;
    LL pre, sum;
    int row, col, i, len, r;
    const int MAX_ROW_COL = nrows > ncols ? nrows : ncols;
    const int MASK_SIZE = mask*mask;
    LL *headL, *tailL;
    int *head, *tail;
    r = mask / 2;
    rowImg = (LL *)malloc(sizeof(LL)*nrows*ncols);
    buffer = (LL *)malloc(sizeof(LL)*(MAX_ROW_COL + mask));
    len = nrows + 2 * r;
    srcPtr = srcImg;
    rowPtr = rowImg;

    for (col = 0; col < ncols; ++col)
    {
        srcPtr = srcImg + col;
        head = srcPtr + r*ncols;
        tail = srcPtr + (nrows - 2 - r)*ncols;

        for (i = 0; i < r; ++i)
        {
            buffer[i] = *head;
            buffer[len - 1 - i] = *tail;
            head += ncols;
            tail -= ncols;
        }
        for (i = r; i < len - r; ++i)
        {
            buffer[i] = *srcPtr;
            srcPtr += ncols;
        }
        sum = buffer[0];
        for (i = 1; i < mask; ++i)
        {
            sum += buffer[i];
        }

        rowPtr = rowImg + col;
        *rowPtr = sum;
        pre = sum;
        headL = buffer;
        tailL = buffer + mask;
        for (i = mask; i < len; ++i, ++headL, ++tailL)
        {
            *rowPtr = pre;
            rowPtr += ncols;
            pre = pre - *headL + *tailL;
        }

    }

    len = ncols + 2 * r;
    rowPtr = rowImg;
    desPtr = desImg;

    for (row = 0; row < nrows; ++row)
    {
        headL = rowPtr + r;
        tailL = rowPtr + (ncols - 2 - r);
        for (i = 0; i < r; ++i)
        {
            buffer[i] = *headL;
            buffer[len - 1 - i] = *tailL;
            ++headL;
            --tailL;
        }
        for (i = r; i < len - r; ++i)
        {
            buffer[i] = *rowPtr;
            ++rowPtr;
        }
        sum = buffer[0];
        for (i = 1; i < mask; ++i)
        {
            sum += buffer[i];
        }
        *desPtr = sum/MASK_SIZE;
        ++desPtr;
        pre = sum;

        headL = buffer;
        tailL = buffer + mask;
        for (i = mask; i < len; ++i, ++headL, ++tailL)
        {
            *desPtr = pre/MASK_SIZE;
            ++desPtr;
            pre = pre - *headL + *tailL;
        }
    }
    free(rowImg);
    free(buffer);
    return;

}