[mvapich-discuss] OSU 4.2 OpenSHMEM micro-benchmark: reduce latency test inconsistency

[Jithin Jose]

Hi Sayan,

Thanks for the report.
We will update this benchmark in the next release.

Regards,

--
Jithin Jose


On Fri, Dec 13, 2013 at 8:13 PM, Sayan Ghosh <sayandeep52 at gmail.com> wrote:

> The OSU 4.2 OpenSHMEM micro-benchmark for testing reduce latency has a
> minor inconsistency, for which the code fails (on SGI systems) with the
> following:
>
> LIBSMA ERROR:   The "pWrk" argument passed to _shmem_float_sum_to_all
>                 or shmem_real4_sum_to_all_
>                 is not a symmetric data object.  See the intro_shmem(3)
>                 man page for a list of symmetric data objects.
>
> pWrkF1|2 in osu_oshm_reduce.c should be either global or have symmetric
> allocation.
>
> ...
> float *pWrkF1 = malloc(MAX(nreduce/2+1, _SHMEM_REDUCE_MIN_WRKDATA_SIZE));
> float *pWrkF2 = malloc(MAX(nreduce/2+1, _SHMEM_REDUCE_MIN_WRKDATA_SIZE));
> ...
> if(i%2)
>    shmem_float_sum_to_all(recvbuf, sendbuf, size, 0, 0, numprocs, pWrkF1,
> pSyncRed1);
> else
>    shmem_float_sum_to_all(recvbuf, sendbuf, size, 0, 0, numprocs, pWrkF2,
> pSyncRed2);
> ...
>
> From OpenSHMEM reference manual: pWrk A symmetric work array.
>
> --
> Regards,
> Sayan
>
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mailman.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20131213/90b2ab4b/attachment.html>