[mvapich-discuss] Questions in OSU OpenSHMEM Microbenchmarks

Naveen Ravichandrasekaran nravi.research at gmail.com
Tue Dec 19 13:40:55 EST 2017


Hello,
If I'm understanding the all-reduce tests in the OSU OpenSHMEM
Microbenchmark correctly, the allocation for symmetric work arrays are done
over the number of elements and not the size of the work arrays.

In OSU-MB-5.4 - osu_oshm_reduce.c;
    float *pWrkF1 = (float *)shmem_malloc(MAX(nreduce/2+1,
_SHMEM_REDUCE_MIN_WRKDATA_SIZE));
    float *pWrkF2 = (float *)shmem_malloc(MAX(nreduce/2+1,
_SHMEM_REDUCE_MIN_WRKDATA_SIZE));

shouldn't this be?
    float *pWrkF1 = (float *)shmem_malloc(sizeof(float) * MAX(nreduce/2+1,
_SHMEM_REDUCE_MIN_WRKDATA_SIZE));
    float *pWrkF2 = (float *)shmem_malloc(sizeof(float) * MAX(nreduce/2+1,
_SHMEM_REDUCE_MIN_WRKDATA_SIZE));

- Naveen Ravichandrasekaran
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20171219/4f868c2e/attachment.html>


More information about the mvapich-discuss mailing list