compiling with openmpi 2018
getting an error on csr.c in metis routines. it's just a warning, but my parallel config is not running parallel, so I'm wondering if this is the problem.
CC GKlib/libmetis_a-csr.o GKlib/csr.c(796): warning #3180: unrecognized OpenMP #pragma #pragma omp parallel private(i, j, ncand, rsum, tsum, cand) ^ GKlib/csr.c(800): warning #3180: unrecognized OpenMP #pragma #pragma omp for schedule(static) |
I got rid of the warning with --with-metis-cppflags="-fopenmp"
Now to see if it will run in parallel |
It won't run the partitioning step now. It gets to "Read Grid File Information", lists the number of points, and then dies, saying "could not launch executable"
I'm using a pbs scheduler, and have all my path and environment set up properly i'm pretty sure. |
What version of openmpi are you using specifically (x.y.z)?
How are you compiling the code? Before using a scheduler, have you tried running on a local machine? |
openmpi_3.0.1_intel_2018
intel_2018.2.199 Haven't tried serial since it's really of no use to me and I can't run interactive on our main cluster. Right now, it seems to be a path issue to whatever is called to do the grid partitioning. But when I had that step working, it would load one cpu up with the entire parmetis job and eventually die due to maxxing out the memory. |
Solved
I went with mpt instead of openmpi and it is working!
configure --prefix=/u/tjentink/SU2 --exec-prefix=/u/tjentink/SU2 --docdir=/u/tjentink/SU2 CXXFLAGS=-O2 --enable-metis --enable-parmetis --enable-cgns --enable-mpi --with-cc=mpicc --with-cxx=mpicxx --enable-tecio --libdir=/opt/hpe/hpc/mpt/mpt-2.16 |
All times are GMT -4. The time now is 14:59. |