Cross initialization error from pair_aniso.cpp when compiling with cmake and Invalid atom_style error when trying to fix this

[mce13]

Hi there,

thanks a lot for the lovely package, which I'm trying to install/get to work on our HPC.

Following your installation guide make DNA_mpi gave me

g++: error: unrecognized command-line option '-restrict'; did you mean '-Wrestrict'?

Changing '-restrict' to '-Wrestrict' did not help but resulted in a different error

g++: error: language host not recognized,

which I could not really make sense of.

Therefore, I decided to compile using cmake as described here, which gave me

/usr/users/mengeho/lammps/src/pair_aniso.cpp:204:14: error: jump to case label [-fpermissive]
         case SPHERE_ELLIPSE:
              ^~~~~~~~~~~~~~
/usr/users/mengeho/lammps/src/pair_aniso.cpp:190:18: note:   crosses initialization of ‘double r’
           double r = sqrt(rsq);

So, there seems to be a problem with the switch statement beginning line 179 in pair_aniso.cpp. I tried to fix this in two different ways:

1. I put the contents of each case in brackets to limit the scope of the variable.

2. I declared variable r before the switch statement.

With either approach compliation worked fine and an executable lmp was buildt. However, when I used this either on the demo file in.run or on the chemically_specific_12N_165NRL_HREMD/in.hremd_breathing I got

ERROR: Invalid atom_style command (src/atom_vec.cpp:75)
Last command: atom_style   hybrid ellipsoid angle charge

I suppose this could be related to the changes I made to pair_aniso.cpp, although I'm not sure about this.

Do you have any suggestion how I could get my installation to work (with or without cmake) or how I could fix the Invalid atom_style error?

Your help would be much appreciated,

Cheers,

Maik

[sef43]

Hi @mce13

The problem is because the provided Makefile is for Intel compilers only. It looks like you are using gcc compilers. You will need to do some minor edits to the Makefile: Delete -restrict -xhost on both the CCFLAGS line and LINKFLAGS line. They are both intel compiler specific options which cause g++ to throw errors.

Yes some {} might be needed, i think the older compilers I used did not throw this as an error.

The invalid atom style will probably be because the other packages have not been installed correctly by cmake, e.g. asphere.
Best bet is to edit the makefile to be gcc compatible and follow the install guide as closely as possible.

[mce13]

Hi @sef43 ,

thanks a lot for your reply and sorry for the slow response. This is how we got it to work using cmake in the end. Thought it might be useful to other people as well.

1. Fix the switch statement beginning line 179 in pair_aniso.cpp by putting the contents of each case in curly brackets as above.
2. Clone a copy of LAMMPS and checkout stable version 3rd March 2020.
3. Copy the code from lammps_custom_code including the modified pair_aniso.cpp into lammps/src.
4. Build LAMMPS using cmake as described here i.e.:

cd lammps                
mkdir build; cd build 
cmake ../cmake  

In the resulting file CMakeCache.txt in the build folder most Packages are set to OFF by default. Switch this to ON for ASPHERE, MOLECULE and RIGID, i.e.

//Build ASPHERE Package
PKG_ASPHERE:BOOL=ON
...
//Build MOLECULE Package
PKG_MOLECULE:BOOL=ON
...
//Build RIGID Package
PKG_RIGID:BOOL=ON

These were lines 381, 438, and 465 in my version of the CMAKECache.txt.

Now run cmake --build.

Worked fine for me with the gcc compilers.

Thanks a lot for your help once more and best wishes,

Maik