I am experimenting an issue on the cluster with SpecMPI using Intel Compilerv12 / Intel MPI. Description of the Cluster: Network: Mellanox IB-QDR Node: 2 x Intel Xeon E5640 with 12 GB memory MPI: Intel MPI v4.0.2.003 OFED: 1.5.1 Compiler Intel v11 and Intel v12
I am running SpecMPI on 64 cores using the command: mpiexec.hydra -f hosts-intel -n $ranks -perhost=8 ....
CC = mpiicc CXX = mpiicpc FC = mpiifort CPP = mpiicpc
Description of the problem: When I use the Intel Compiler v12 (12.0.3), using the following compilation flags inside the config file of SpecMPI for 104.milc and 130.socorro: COPTIMIZE = -O3 -xSSE4.2 -no-prec-div FOPTIMIZE = -O3 -xSSE4.2 -no-prec-div
The results of 104.milc are incorrect. All the rest is ok.
If I use the Intel Compiler v11 (11.1.069). All the results are correct.
Removing the compilation flag -xSSE4.2 resolved the issue but created a new one. Using the compilation flags inside the config file of SpecMPI :
COPTIMIZE = -O3 -no-prec-div
FOPTIMIZE = -O3 -no-prec-div
With Intel Compiler v12 (12.0.3), all is running fine except Socorro that crashed With Intel Compiler v11 (11.1.069), all is running fine without any problem.
I was wondering if you were aware of this issue or you were able to reproduce it. I don't know if the issue is coming from Compiler12, Spec benchs or Intel MPI + compiler v12.
FYI, I found the reason of the issue. I am posting here to inform everybody. SPEC had to update the validation file for 104.milc due to compiler Intel compiler v12. So, you need to use the last version of Spec MPI (2.0.1) when you want to run Spec MPI with Intel compiler 12.