Buliding LAMMPS with user-cuda package

Dear All,

I have built LAMMPS using make serial command without optional package before successfully. Now, I am going to build LAMMPS with user-cuda package. To do this, first, I installed the package successfully. However, when I’m trying to build LAMMPS using make serial command, I will get following error:

make[1]: Entering directory /home/he67l/lammpsnew/src/Obj_serial' g++ -g -O3 -DLAMMPS_GZIP -I../../lib/cuda -DLMP_USER_CUDA -I../STUBS -I/usr/local/cuda/i nclude -DUNIX -DFFT_CUFFT -DCUDA_PRECISION=2 -DCUDA_ARCH=20 -c ../fft3d_cuda.cpp In file included from ../fft3d_cuda.cpp:45:0: ../fft3d_cuda.h:80:26: **fatal error: cuda_runtime.h: No such file or directory** compilation terminated. make[1]: *** [fft3d_cuda.o] Error 1 make[1]: Leaving directory /home/he67l/lammpsnew/src/Obj_serial’
make: *** [serial] Error 2

Apparently, cuda_runtime.h file is not available. I am wondering if anybody can help me how to approach the problem?

Thank you,
Hamed.

Dear All,

I have built LAMMPS using make serial command without optional package
before successfully. Now, I am going to build LAMMPS with user-cuda package.
To do this, first, I installed the package successfully. However, when I'm
trying to build LAMMPS using make serial command, I will get following
error:

the USER-CUDA package is unmaintained and its use is deprecated.
please consider using the GPU or KOKKOS package instead.

make[1]: Entering directory `/home/he67l/lammpsnew/src/Obj_serial'
g++ -g -O3 -DLAMMPS_GZIP -I../../lib/cuda -DLMP_USER_CUDA -I../STUBS
-I/usr/local/cuda/i nclude -DUNIX -DFFT_CUFFT
-DCUDA_PRECISION=2 -DCUDA_ARCH=20 -c ../fft3d_cuda.cpp
In file included from ../fft3d_cuda.cpp:45:0:
../fft3d_cuda.h:80:26: fatal error: cuda_runtime.h: No such file or
directory
compilation terminated.
make[1]: *** [fft3d_cuda.o] Error 1
make[1]: Leaving directory `/home/he67l/lammpsnew/src/Obj_serial'
make: *** [serial] Error 2

Apparently, cuda_runtime.h file is not available. I am wondering if anybody
can help me how to approach the problem?

a) cuda_runtime.h is part of the cude toolkit. do you have this
installed and is it working?

b) did you properly configure and build the lammpscuda library in the
lib/cuda folder?
this is a pre-requisite to building the USER-CUDA packages and - if
done correctly - will set up the paths to the cuda toolkit properly.

this is discussed in detail in the LAMMPS manual and the README file in lib/cuda

however, let me remind you again, that the USER-CUDA package is
deprecated, unsupported and has known and unfixed bugs.

axel.

Hello Axel,

Thanks for your help. The reason I want to build LAMMPS with user-cuda is that I'm working on an input file for simulation of water. In the input file, there is a fix shake command which I am thinking is part of user-cuda package and I always get error the unknown fix style. How can I run this input file without installing user-cuda? Please advice.

Thanks,
Hamed.

Hello Axel,

Thanks for your help. The reason I want to build LAMMPS with user-cuda is that I'm working on an input file for simulation of water. In the input file, there is a fix shake command which I am thinking is part of user-cuda package and I always get error the unknown fix style. How can I run this input file without installing user-cuda? Please advice.

no. your thinking is wrong.
fix shake is part of the RIGID package. check out the documentation:
http://lammps.sandia.gov/doc/fix_shake.html (see under Restrictions).
USER-CUDA contains fix shake/cuda which is specifically written for
use with other styles in the USER-CUDA packages.

axel.