[mvapich-discuss] OSU MPI one-sided micro-benchmark typo (?)

[Jeff Hammond]

Unlike MPI_Send though, I don't think zero-byte RMA has any effect
because RMA comm routines are, by design, non-synchronizing.  It's
perfectly reasonable to test it though from the standpoint of software
overhead and to make sure the trivial case is correct.  Not all
implementations handle count=0 properly.

Jeff

On Thu, Dec 5, 2013 at 6:44 AM, sreeram potluri <potluri.2 at osu.edu> wrote:
> Sayan,
>
> All the latency tests in OMB do start from size 0B. Its not a typo.
>
> Best
> Sreeram Potluri
>
>
> On Wed, Dec 4, 2013 at 7:31 PM, Sayan Ghosh <sayandeep52 at gmail.com> wrote:
>>
>> In the latest OSU Micro-benchmarks 4.2 (MPI one-sided), I notice a
>> probable typo, causing this to show up in stdout for some of the benchmarks:
>>
>> # OSU One Sided MPI_Get latency Test
>> # Size            Latency (us)
>> 0                         1.74
>> ...
>>
>> I guess size/count could be started from 1 and not 0
>>
>> < for (count = 0; count <= MAX_SIZE / sizeof(int); count = (count ? count
>> << 1 : 1)) {
>> >  for (count = 1; count <= MAX_SIZE / sizeof(int); count = count << 1) {
>>
>>
>> --
>> Regards,
>> Sayan
>>
>> _______________________________________________
>> mvapich-discuss mailing list
>> mvapich-discuss at cse.ohio-state.edu
>> http://mailman.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>>
>
>
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mailman.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>



-- 
Jeff Hammond
jeff.science at gmail.com