[zeromq-dev] ØMQ VSM data alignment

[Matt Weinstein]

On Aug 4, 2010, at 1:37 PM, Andrew Hume wrote:

> this sort of thing can get problematic quickly.
> the combinatorial explosion of cache-line design, memory design,
> language alignment rules, architectures allowing non-aligned  
> references
> (but punishing you with slowness) makes testing hard.
>
> on the other hand, starting on a 8 byte boundary is always a good  
> idea in modern architectures.
>
> better style is to use a byte stream so you won't get bitten by LSB/ 
> MSB issues
> or packets coming from other machine architectures.
>

Yes, but I want to do this:

	zmq::message_t foo;

	s.recv(&foo);
	my_struct const* pstruct = reinterpret_cast<my_struct  
const*>(foo.data());

	cout << "Works like magic: " << pstruct->member << ". All your  
processor are belong to Intel! (*)" << endl;

That's the no-copy approach :-)

I can do it with malloc() without issue or penalty...

(*) There are no other architectures :-)

> andrew
>
> On Aug 4, 2010, at 1:09 PM, Pieter Hintjens wrote:
>
>> Matt,
>>
>> I don't think this has been covered.  It would be sensible to have
>> vsm_data aligned but it's something that is IMO worth testing rather
>> than doing on faith.
>>
>> -Pieter
>>
>>> Sorry if this has been covered before.
>>> I've been using zmq to transport raw structures, and I've noticed  
>>> that
>>> zeromq apparently does not provide appropriate alignment for  
>>> messages in
>>> VSMs, and maybe for larger messages as well if you're using the  
>>> front of the
>>> content buffer (I haven't looked :-( )
>>> Here are the offsets:
>>>
>>> 0x00    void *content;
>>> 0x08    unsigned char flags;
>>> 0x09    unsigned char vsm_size;
>>> 0x0a    unsigned char vsm_data [ZMQ_MAX_VSM_SIZE];
>>>
>>> Shouldn't vsm_data be padded to 0x10 on 64 bit architectures (and  
>>> maybe
>>> others)?
>>> RSVP,
>>> Thanks,
>>> Matt
>>>
>>> _______________________________________________
>>> zeromq-dev mailing list
>>> zeromq-dev at lists.zeromq.org
>>> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>>>
>>>
>> _______________________________________________
>> zeromq-dev mailing list
>> zeromq-dev at lists.zeromq.org
>> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>>
>
> ------------------
> Andrew Hume  (best -> Telework) +1 732-886-1886
> andrew at research.att.com  (Work) +1 973-360-8651
> AT&T Labs - Research; member of USENIX and LOPSA
>
>
>
> _______________________________________________
> zeromq-dev mailing list
> zeromq-dev at lists.zeromq.org
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20100804/7fc59f74/attachment.htm>