Portland compiler fails trying to allocate an unexpectedly large amount of memory: issue with large arrays

GMIN jobs on large systems (e.g. 5000 LJ atoms) can sometimes fail with memory allocation errors like this one:

 0: ALLOCATE: 18446744071814864320 bytes requested; not enough memory

This seems rather a lot of memory, and more than GMIN actually requires. It's probably caused by one or more arrays that's larger than 2GB (in my case, what triggered the error was the W work array in mymylbfgs.f, a double precision array of dimension 300035000 for the 5000-atom test job). Different pgi compilers behave differently in this case: on volkhan, pgi/64/7* through pgi/64/11* give the allocate error, but pgi/64/12* work "normally" (for a short test job).

Adding the compiler flags "-mcmodel=medium -Mlarge_arrays" is a good fix for this with the earlier pgi compilers.