[lammps-users] Crash during run

Dear developper
I am using 10 Mar 2021 LAMMPS version on Windows using MICH version recommended onthe web site.
The following crash has occured.

what is your input file and what is the exact command line you are using to start LAMMPS?

Dear Axel,

Dear Axel,

The command line is :
“C:\Program Files\MPICH2\bin\mpiexec.exe” -np 4 “C:\Users\pascal.brault\AppData\Local\LAMMPS 64-bit 10Mar2021-MPI\bin\lmp_mpi.exe” -in He-Al_SiwHeBv2.in
If you agree I prefer to share privately the input file.

no, I don’t agree.

my guess is that you have some loop construct that is not compatible with replicating input over MPI.

axel.

Dear Axel
No problem: Attached is the input file

He-Al_SiwHeBv2.in (4.15 KB)

try replacing “SELF” with the filename of your input.

due to quirks with the CR-LF line endings on Windows, the rewind() library call, required for SELF, will not always work reliably.

Thanks a lot Axel
I will run it with this modification.
Before, I will also install the last LAMMPS version 27May2021.
Best regards
Pascal

SignatureGREMI.png

Hi Axel
It now works with this modification.

I have another question.
If I run the following command (Lammps version is 8 April 2021)

mpirun -np 4 lmp -in in.file

I got the error message when using the command “processors 2 2 1” in the in script.
Specified processors != physical processors
but it works when removing the command “processors 2 2 1”
It only occurs on my linux machine.

What I have to do for solving this issue ?

On the Windows machine both work well. The only difference is the generated grid
for example I got 2x2x1 MPI grid if processors 2 2 1 is included, which is consistent.
but 1x1x4 grid if not included. Which is how lammps find the best grid itself (if I well understand)

Thanks a lot in advance

Best regards
Pascal

Hi Axel
It now works with this modification.

I have another question.
If I run the following command (Lammps version is 8 April 2021)

mpirun -np 4 lmp -in in.file

I got the error message when using the command “processors 2 2 1” in the in script.
Specified processors != physical processors
but it works when removing the command “processors 2 2 1”
It only occurs on my linux machine.

What I have to do for solving this issue ?

look very closely at the output on the Linux machine.
most likely you will have everything printed 4 times and a 1x1x1 processor grid.

the code in question that does the check and the processor assignment has not been changed in a very long time and if you get the error message you describe that is usually the consequence of using an mpirun/mpiexec command from a different MPI library or having a non-parallel LAMMPS executable.

On the Windows machine both work well. The only difference is the generated grid
for example I got 2x2x1 MPI grid if processors 2 2 1 is included, which is consistent.
but 1x1x4 grid if not included. Which is how lammps find the best grid itself (if I well understand)

LAMMPS by default assumes that all volume is filled by atoms and thus will assign subdomains based on volume. but that fails, e.g., for slab configurations.
a way to make this independent from the number of processors is to use the processors command as follows:
processors * * 1

but of course that does not prevent the issue of having inconsistent executable and mpirun/mpiexec

axel.

Hi Axel
I will check this more carefully
Best
Pascal

Hi Axel
It now works with this modification.

I have another question.
If I run the following command (Lammps version is 8 April 2021)

mpirun -np 4 lmp -in in.file

I got the error message when using the command “processors 2 2 1” in the in script.
Specified processors != physical processors
but it works when removing the command “processors 2 2 1”
It only occurs on my linux machine.

What I have to do for solving this issue ?

look very closely at the output on the Linux machine.
most likely you will have everything printed 4 times and a 1x1x1 processor grid.

the code in question that does the check and the processor assignment has not been changed in a very long time and if you get the error message you describe that is usually the consequence of using an mpirun/mpiexec command from a different MPI library or having a non-parallel LAMMPS executable.

On the Windows machine both work well. The only difference is the generated grid
for example I got 2x2x1 MPI grid if processors 2 2 1 is included, which is consistent.
but 1x1x4 grid if not included. Which is how lammps find the best grid itself (if I well understand)

LAMMPS by default assumes that all volume is filled by atoms and thus will assign subdomains based on volume. but that fails, e.g., for slab configurations.
a way to make this independent from the number of processors is to use the processors command as follows:
processors * * 1

but of course that does not prevent the issue of having inconsistent executable and mpirun/mpiexec

axel.