Dear lammps developers/users,
I am trying to compile my lammps with my gpu package, which is successfully installed.
I have successfully built a gpu package (as in the manual) and have a libgpu.a in my lib directory.
Here are the relevant parts of my makefile in the src directory:
gpu_SYSINC =
gpu_SYSLIB = -lcudart -lcuda
gpu_SYSPATH = -L/usr/local/Cluster-Apps/cuda/5.0/lib
However, when I compile lammps with include yes-gpu, compilation terminates in this error:
mpiicc -O3 -xHOST -fno-alias -ip -unroll0 -DMPICH_IGNORE_CXX_SEEK -DLAMMPS_GZIP -DFFT_FFTW -I/usr/local/Cluster-Apps/intel/mkl/10.3.10.319/composer_xe_2011_sp1.10.319/mkl/include/fftw -M pair_born_coul_long_gpu.cpp > pair_born_coul_long_gpu.d
pair_born_coul_long_gpu.h(23): catastrophic error: cannot open source file “pair_born_coul_long.h”
#include “pair_born_coul_long.h”
^
make[1]: *** [pair_born_coul_long_gpu.d] Error 4
Could anyone possibly point out where I am going wrong? Thank you very much in advanced.
With best wishes,
Anna
please do a "make clean-all" and "make purge" and then try again.
axel.
Dear Axel,
Thank you for your quick response, I did this and I am still getting the same error, unfortunately!
with best wishes,
Anna
Dear Axel,
Thank you for your quick response, I did this and I am still getting the
same error, unfortunately!
well, it works for me with the latest git version.
you did install the KSPACE package, right?
axel.
Dear Axel,
thank you! Now that I explicitly stated make yes-gpu and make yes-kspce, I get the following error:
mkl/lib/intel64/libfftw2xc_intel.a -lcudart -lcuda -lstdc++ -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -liomp5 -lpthread -o ../lmp_darwin
ipo: warning #11010: file format not recognized for /usr/local/Cluster-Apps/cuda/5.0/lib/libcudart.so
ld: *skipping incompatible /usr/local/Cluster-Apps/cuda/5.0/lib/libcudart.so when searching for -lcudart**
**ld: cannot find -lcudart*
make[1]: *** [../lmp_darwin] Error 1
There appears to be an error with lcudart...
thanks again!
Dear Axel,
thank you! Now that I explicitly stated make yes-gpu and make yes-kspce, I
get the following error:
mkl/lib/intel64/libfftw2xc_intel.a -lcudart -lcuda -lstdc++
-lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -liomp5 -lpthread -o
../lmp_darwin
ipo: warning #11010: file format not recognized for
/usr/local/Cluster-Apps/cuda/5.0/lib/libcudart.so
ld: skipping incompatible /usr/local/Cluster-Apps/cuda/5.0/lib/libcudart.so
when searching for -lcudart
ld: cannot find -lcudart
make[1]: *** [../lmp_darwin] Error 1
There appears to be an error with lcudart...
you need to check if the cuda library is of the same "bitness" as your compile.
this error message happens, if you try to link a 32-bit library to a
64-bit executable.
isn't there a directory /usr/local/Cluster-Apps/cuda/5.0/lib64/ ?
axel.
Dear Axel,
Thank you so much, it does work with lib64 indeed - I have learned so much from your help over the years!
With best wishes,
Anna