USER-CUDA problem

LAMMPS LAMMPS Mailing List Mirror
1

Dear lamms user

I installed LAMMPS distribution from the following way:

http://lammps.sandia.gov/threads/msg51799.html

I decided to use with cuda package. My graphic card is eligible. The CUDA toolkit have installed and it’s in the correct path (usr/local/cuda).

As mentioned in LAMMPS Manual I tried both steps but faced with following error. I will appreciate to help me.

1st way:
python Make.py -p cuda -cuda mode=single arch=20 -o cuda -a lib-cuda file mpi

Error:
mpicxx -restrict -c tmpauto.cpp
g++: error: unrecognized command line option ‘-restrict’
mpicxx -xHost -c tmpauto.cpp
g++: error: language Host not recognized
g++: error: language Host not recognized
mpicxx -fno-alias -c tmpauto.cpp
g++: error: unrecognized command line option ‘-fno-alias’
mpicxx -ansi-alias -c tmpauto.cpp
g++: error: unrecognized command line option ‘-ansi-alias’
mpicxx -override-limits -c tmpauto.cpp
compile produced no output
Created src/MAKE/MINE/Makefile.auto
Action exe …
make[1]: Entering directory /home/reza/MD/lammps-7Dec15/src/Obj_auto' Makefile.package.settings:5: ../../lib/gpu/Makefile.lammps: No such file or directory make[1]: *** No rule to make target …/…/lib/gpu/Makefile.lammps’. Stop.
make[1]: Leaving directory `/home/reza/MD/lammps-7Dec15/src/Obj_auto’
make: *** [auto] Error 2

ERROR: Unsuccessful “make auto”

2nd way:

…/lib/cuda$ make OPTIONS
make: *** No rule to make target `OPTIONS’. Stop.

…/lib/cuda$ make
/usr/local/cuda/bin/nvcc -lib atom_vec_cuda_cu.o comm_cuda_cu.o compute_temp_cuda_cu.o compute_temp_partial_cuda_cu.o cuda_data_cu.o cuda_cu.o cuda_pair_cu.o cuda_wrapper_cu.o domain_cu.o fft3d_cuda_cu.o fix_addforce_cuda_cu.o fix_aveforce_cuda_cu.o fix_enforce2d_cuda_cu.o fix_freeze_cuda_cu.o fix_gravity_cuda_cu.o fix_nh_cuda_cu.o fix_nve_cuda_cu.o fix_set_force_cuda_cu.o fix_shake_cuda_cu.o fix_temp_berendsen_cuda_cu.o fix_temp_rescale_cuda_cu.o fix_temp_rescale_limit_cuda_cu.o fix_viscous_cuda_cu.o neighbor_cu.o pppm_cuda_cu.o -I/usr/local/cuda/include -DUNIX -DFFT_CUFFT -DCUDA_PRECISION=1 -DCUDA_ARCH=20 -L/usr/local/cuda/lib -L/usr/local/cuda/lib64 -lcufft -o liblammpscuda.a

…/src$ make ubuntu
make[1]: Entering directory /home/reza/MD/lammps-7Dec15/src/Obj_ubuntu' Makefile.package.settings:5: ../../lib/gpu/Makefile.lammps: No such file or directory make[1]: *** No rule to make target …/…/lib/gpu/Makefile.lammps’. Stop.
make[1]: Leaving directory `/home/reza/MD/lammps-7Dec15/src/Obj_ubuntu’
make: *** [ubuntu] Error 2

Regards

Reza

2

Dear lamms user

I installed LAMMPS distribution from the following way:
http://lammps.sandia.gov/threads/msg51799.html

I decided to use with cuda package. My graphic card is eligible. The CUDA

please note that the USER-CUDA package is essentially unmaintained at
this point and bugs are no longer fixed unless they are trivial to
correct.

toolkit have installed and it's in the correct path (usr/local/cuda).
As mentioned in LAMMPS Manual I tried both steps but faced with following
error. I will appreciate to help me.

1st way:
python Make.py -p cuda -cuda mode=single arch=20 -o cuda -a lib-cuda file
mpi

Error:
mpicxx -restrict -c tmpauto.cpp
g++: error: unrecognized command line option ‘-restrict’
mpicxx -xHost -c tmpauto.cpp
g++: error: language Host not recognized
g++: error: language Host not recognized
mpicxx -fno-alias -c tmpauto.cpp
g++: error: unrecognized command line option ‘-fno-alias’
mpicxx -ansi-alias -c tmpauto.cpp
g++: error: unrecognized command line option ‘-ansi-alias’
mpicxx -override-limits -c tmpauto.cpp
compile produced no output

these should all be harmless and are just tests to determine supported
flags for your specific compiler/MPI combination.

Created src/MAKE/MINE/Makefile.auto
Action exe ...
make[1]: Entering directory `/home/reza/MD/lammps-7Dec15/src/Obj_auto'
Makefile.package.settings:5: ../../lib/gpu/Makefile.lammps: No such file or
directory
make[1]: *** No rule to make target `../../lib/gpu/Makefile.lammps'. Stop.
make[1]: Leaving directory `/home/reza/MD/lammps-7Dec15/src/Obj_auto'
make: *** [auto] Error 2

ERROR: Unsuccessful "make auto"

this has nothing to do with USER-CUDA, but the GPU package. i won't
recommend to use the Make.py script before understanding the build
process of LAMMPS in general. the main purpose of Make.py is to
automate the build process. for that you should first be able to build
LAMMPS the conventional way with the minimum number of packages
installed (make no-all; make mpi).

2nd way:

../lib/cuda$ make OPTIONS
make: *** No rule to make target `OPTIONS'. Stop.

this is due to not reading the README file properly and take its
instructions literally without paying attention to the explanations.

../lib/cuda$ make
/usr/local/cuda/bin/nvcc -lib atom_vec_cuda_cu.o comm_cuda_cu.o
compute_temp_cuda_cu.o compute_temp_partial_cuda_cu.o cuda_data_cu.o
cuda_cu.o cuda_pair_cu.o cuda_wrapper_cu.o domain_cu.o fft3d_cuda_cu.o
fix_addforce_cuda_cu.o fix_aveforce_cuda_cu.o fix_enforce2d_cuda_cu.o
fix_freeze_cuda_cu.o fix_gravity_cuda_cu.o fix_nh_cuda_cu.o
fix_nve_cuda_cu.o fix_set_force_cuda_cu.o fix_shake_cuda_cu.o
fix_temp_berendsen_cuda_cu.o fix_temp_rescale_cuda_cu.o
fix_temp_rescale_limit_cuda_cu.o fix_viscous_cuda_cu.o neighbor_cu.o
pppm_cuda_cu.o -I/usr/local/cuda/include -DUNIX -DFFT_CUFFT
-DCUDA_PRECISION=1 -DCUDA_ARCH=20 -L/usr/local/cuda/lib
-L/usr/local/cuda/lib64 -lcufft -o liblammpscuda.a

this is a successful compile with the default settings.

../src$ make ubuntu
make[1]: Entering directory `/home/reza/MD/lammps-7Dec15/src/Obj_ubuntu'
Makefile.package.settings:5: ../../lib/gpu/Makefile.lammps: No such file or
directory
make[1]: *** No rule to make target `../../lib/gpu/Makefile.lammps'. Stop.
make[1]: Leaving directory `/home/reza/MD/lammps-7Dec15/src/Obj_ubuntu'
make: *** [ubuntu] Error 2

as before, the problem is due to having installed the GPU package
without building the corresponding library and has nothing to do with
USER-CUDA. i suggest you re-read the installation instructions in the
LAMMPS manual and the corresponding README files again with more care,
and first try to build a LAMMPS executable with no packages installed
and then gradually install additional packages (and perform the
necessary steps to install/configure/compile auxiliary external
libraries and tools as well as the corresponding libraries in the lib
folder.

axel.