we have a cluster with AMD opetron Processor with 4coreX8=32procs each node. I have compiled Lammps with intel compiler and getting amazing performance however when I compiled with with opencc(which is supposed to be better on AMD procs), I am getting very slow performance(almost one half of intel compiler).
Nodes are connected with infiniband and I am using mvapich2. Is there any issue with open64 compiler or I have to make some special flag in MAKE file to get better performance.
I have no idea. This a compile/link question for your box. I suggest
you talk to system people familiar with your box and its hardware
and simply experiment.