static inline void* memcpy_aligned(void* dest, const void* src, size_t count) {
    char* pDest = (char*) dest;
    char* pSrc = (char*) src;
    while (pSrc < ((char*) src) + count && (intptr_t(pSrc) & (sizeof(long) - 1))) {
        *pDest++ = *pSrc++;
    }

    while (pSrc + sizeof(long) - 1 < ((char*) src) + count) {
        *(long*) pDest = *(long*) pSrc;
        pSrc += sizeof(long);
        pDest += sizeof(long);
    }
    
    return dest;
           ^~~~ 可能导致少拷贝数据。

}

void* memcpy(void* dest, const void* src, size_t count) {

    // 如果 src 和 dest 可以互相对齐，则使用快速拷贝。
    if (!((intptr_t(dest) ^ intptr_t(src)) & (sizeof(long) - 1))) {
        return memcpy_aligned(dest, src, count);
    }

    // 单字节拷贝。
    char* pDest = (char*) dest;
    char* pSrc = (char*) src;
    while (pSrc < ((char*) src) + count) {
        *pDest++ = *pSrc++;
    }

    return dest;
}