Problem: seed for random numbers

[tsuyoshi38]

This problem probably exists only on Fugaku (Japanese supercomputer), but I think it may be better to report it, anyway.
I have encountered a problem to start MD (NVT) in generating a seed for random number in rng_module.f90.

In the subroutine get_random_seed, it first tries to read the file "/dev/urandom".
On Fugaku, the OS has the file, but it cannot be read.
(this might be due to the inconsistency between the OS of the front-end and the function "random_seed", though I have not checked.)

The easiest solution for this is to set a positive integer for General.RNGSeed in Conquest_input.

[davidbowler]

Thank you for reporting this. What is the error returned by the operating system? The call to open /dev/urandom should return an error if it cannot be read. It would be helpful to understand this in more detail.

[tsuyoshi38]

The error message is like "An endfile record was detected in a READ statement (unit=10)". So, I guess the size n of rn%seed is not consistent.

[davidbowler]

This is interesting: it may be OS dependent, or something more complicated. I wonder whether part of the problem is that the seed is defined as a 64 bit integer (wide). Could you test the effect of reading a normal integer there and copying it into rn%seed? We could (should?) also introduce a check for a read error on that line, and use the other method in the module (date and time).