I have a parallel program in Fortran using coarrays and I have detected a different behavior in the memory when I run it in shared mode or distributed mode. When I use coarray=shared the used memory along the execution is constant but when I use coarray=distributed the memory starts to grow at each iteration and finally the program finish with an out of memory error. I have isolated the problem in my code and I have wrote a sample code (attached) that reproduce this behavior. I am using Intel Fortran Compiler for Cluster v18, but the effect is similar with versions 17 and 15.
Not sure if it is related to your problem, but the OpenCoarrays runtime gives the following runtime error with your code:
libcaf_mpi RUNTIME ERROR: Cannot convert type 5 kind 4 to type 1 kind 2 Program received signal SIGABRT: Process abort signal.
Intel now wants you to request support at their Online Service Center (OSC) for issues with their software products, I suggest you submit an incident at the OSC: