[MPICH] IMB Exchange

[Sylvain Jeaugey]

Hi Rajeev,

As you mention, MPI implementations make progress on all requests, but is 
the Exchange algorithm really MPI compliant ? I mean, since all 
applications do this kind of things, MPI implementations have no choice 
but to support this behaviour, but is there really something in the 
standard about making progress on all requests when we call MPI_Wait on 
only one request ?

Sylvain

On Fri, 16 Nov 2007, Rajeev Thakur wrote:

> You still need to call the Waitall though -- to deallocate the request
> objects etc.
>
>
>> -----Original Message-----
>> From: Calin Iaru [mailto:calin at dolphinics.no]
>> Sent: Friday, November 16, 2007 6:23 AM
>> To: Rajeev Thakur
>> Cc: mpich-discuss at mcs.anl.gov
>> Subject: Re: [MPICH] IMB Exchange
>>
>> Thanks Rajeev. It's certainly useful. So MPI_Waitall is not
>> actually needed.
>>
>> Rajeev Thakur wrote:
>>> It is ok to call the waitall after the receives. While
>> blocked in the
>>> receive, the MPI implementation will cause progress on other
>>> previously issued communication. It cannot just make
>> progress on the
>>> currently issued receive.
>>>
>>> Rajeev
>>>
>>>
>>>> -----Original Message-----
>>>> From: owner-mpich-discuss at mcs.anl.gov
>>>> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Calin Iaru
>>>> Sent: Thursday, November 15, 2007 6:01 AM
>>>> To: mpich-discuss at mcs.anl.gov
>>>> Subject: [MPICH] IMB Exchange
>>>>
>>>> I am looking at IMB Excahnge implementation which is:
>>>>   - asynchronous send to left and right rank
>>>>   - blocking receive from left and right
>>>>   - Waitall
>>>> What I don't understand is why is blocking receive followed by
>>>> Waitall.
>>>> If all nodes are doing async send which are not guaranteed to
>>>> complete, then the receive operation is not satisfied. Let me know
>>>> your opinion. I would expect that waitall is after async send.
>>>>
>>>>
>>>>
>>>
>>>
>>
>>
>
>