[mpich-discuss] allgatherv

Halim Amer aamer at anl.gov
Mon May 9 12:52:04 CDT 2016


Never mind, Rajeev was right; I missed the medium messages with powers 
of two case.

--Halim

On 5/9/16 12:45 PM, Halim Amer wrote:
> Looking at the code, for long messages the same ring algorithm is used
> for long/medium messages regardless of whether the number of processes
> is a power of two or not. The comment is wrong. We will fix it, but in
> general one shouldn't rely solely on code comments and should look at
> the code itself.
>
> --Halim
> www.mcs.anl.gov/~aamer
>
> On 5/9/16 12:21 PM, Jen B wrote:
>> Hello,
>>
>> I am looking at allgatherv.c from:
>>
>> git.mpich.org/mpich.git/tree/HEAD:/src/mpi/coll
>> <http://git.mpich.org/mpich.git/tree/HEAD:/src/mpi/coll>
>>
>> It says the following:
>>
>> "For short or medium-size messages and power-of-two no. of processes,
>> we use the recursive doubling algorithm.
>> ...
>> For long messages or medium-size messages and non-power-of-two
>> no. of processes, we use a ring algorithm.
>> ..."
>>
>> Could someone please tell me what algorithm is used when we have "long
>> messages and power-of-two no. of processes"? Thanks a lot!
>>
>> Best,
>> Jen
>>
>>
>> _______________________________________________
>> discuss mailing list     discuss at mpich.org
>> To manage subscription options or unsubscribe:
>> https://lists.mpich.org/mailman/listinfo/discuss
>>
> _______________________________________________
> discuss mailing list     discuss at mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss
_______________________________________________
discuss mailing list     discuss at mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss


More information about the discuss mailing list