IMB and --input, 2017.0.2

4f0drlp7eyj3 — Wed, 19 Apr 2017 15:21:59 GMT

Hi,

I've been running IMB with the following command

mpirun -n $i -genv I_MPI_DEBUG 5 -machinefile ./nodefile src/IMB-MPI1 -npmin $i -input IMB_SELECT_MPI1 -msglen ./msglens

where $i is set to 2, 4, etc., and the file IMB_SELECT_MPI1 reads

--BOF--

Alltoall

Allgather

Allreduce

Sendrecv

Exchange

Uniband

Biband

--EOF--

What happens is that Biband gets run 7 times, (one for each targeted test), rather than different tests as desired:

# List of Benchmarks to run:

# Biband

#---------------------------------------------------

# Benchmarking Biband

# #processes = 4

#---------------------------------------------------

#bytes #repetitions Mbytes/sec Msg/sec

0 1000 0.00 7199701

65536 640 13233.43 211735

524288 80 19708.07 39416

4194304 10 22623.73 5656

#---------------------------------------------------

# Benchmarking Biband

# #processes = 4

...

Any idea why this is happening? Full stdout attached (there are some bogus runs there, but focus on the Biband output).

Thanks; Chris

Hi,

Dmitry_S_Intel — Thu, 27 Apr 2017 14:56:15 GMT

Hi,

I've filed the ticket to the development team. Fix will be available for future releases of Intel(R) MPI Benchmarks.

Dmitry Sivkov, Ph.D.

4f0drlp7eyj3 — Thu, 27 Apr 2017 16:14:38 GMT

Thanks Dmitry!