void *old_memcpy(void *dest, const void *src, size_t len)
{
        const u8int *sp = (const u8int *)src;
        u8int *dp = (u8int *)dest;
        for(; len > 0; len--) *dp++ = *sp++;
}

void *memcpy(void *dest, const void *src, size_t len)
{
        unsigned int *bdp = (unsigned int *)dest;
        const unsigned int *bsp = (const unsigned int *) src;
        size_t biglen = len / sizeof(unsigned int);
        len = len % sizeof(unsigned int);
        // Move memory in big chunks first ;) wr're faster like that
        for(; biglen != 0; biglen--) *bdp++ = *bsp++;

        const u8int *sp = (const u8int *)bsp;
        u8int *dp = (u8int *)bdp;
        for(; len != 0; len--) *dp++ = *sp++;
        return dest;
}