Error while compiling lammps with GPU package

LAMMPS LAMMPS Mailing List Mirror
1

Hello,

I’m trying to install GPU package in lammps-30Jul16. I’m using 1xGeForce GT 730 with a compute capability of 3.5 (compute capability according to this link: https://developer.nvidia.com/cuda-gpus).

I compiled lmp_mpi with mpich2 and fftw successfully. Now for GPU compilation, I tried both methods,

2

Hello,

I'm trying to install GPU package in lammps-30Jul16. I'm using 1xGeForce GT
730 with a compute capability of 3.5 (compute capability according to this
link: CUDA GPUs - Compute Capability | NVIDIA Developer).

with this kind of GPU, it is unlikely you'll see any significant
speedup, if at all.

I compiled lmp_mpi with mpich2 and fftw successfully. Now for GPU
compilation, I tried both methods,
__________________________________________________________

Method-1:
In ~/lammps-30Jul16/src, I used the command "./Make.py -p gpu -gpu
mode=single arch=35 -o gpu -a lib-gpu file mpi" and I got the following
error

Installing packages ...
Action lib-gpu ...
building gpu library ...
nvcc -I/usr/local/cuda/include -DUNIX -O3 -Xptxas -v --use_fast_math
-Icudpp_mini -arch=sm_35 -D_SINGLE_SINGLE --cubin -DNV_KERNEL -o atom.cubin
lal_atom.cu
make: nvcc: Command not found
Nvidia.makefile:158: recipe for target 'atom.cubin' failed
make: *** [atom.cubin] Error 127

ERROR: Unsuccessful build of lib/gpu library

which means, that your PATH environment variable is not set up
correctly for CUDA

__________________________________________________________

Method-2:
In ~/lammps-30Jul16/lib/gpu, I used "sudo make -f Makefile.linux" and got
the following error
/usr/local/cuda/bin/nvcc -I/usr/local/cuda/include -DUNIX -O3 -Xptxas -v
--use_fast_math -DLAMMPS_SMALLBIG -Icudpp_mini -arch=sm_35 -D_SINGLE_SINGLE
--cubin -DNV_KERNEL -o atom.cubin lal_atom.cu
gcc: error trying to exec 'cc1plus': execvp: No such file or directory
Nvidia.makefile:158: recipe for target 'atom.cubin' failed
make: *** [atom.cubin] Error 1

no idea, why this happens, but using "sudo" for simply compiling
software, is a very, *very* bad idea.

__________________________________________________________

Below is the makefile that I used (Makefile.linux) location:
~/lammps-30Jul16/lib/gpu (Used for method-2)

EXTRAMAKE = Makefile.lammps.standard

ifeq ($(CUDA_HOME),)
CUDA_HOME = /usr/local/cuda
endif

NVCC = /usr/local/cuda/bin/nvcc

CUDA_ARCH = -arch=sm_35

LMP_INC = -DLAMMPS_SMALLBIG

CUDA_PRECISION = -D_SINGLE_SINGLE

CUDA_INCLUDE = -I$(CUDA_HOME)/include
CUDA_LIB = -L$(CUDA_HOME)/lib64
CUDA_OPTS = -DUNIX -O3 -Xptxas -v --use_fast_math $(LMP_INC)

CUDR_CPP = mpic++ -DMPI_GERYON -DUCL_NO_EXIT -DMPICH_IGNORE_CXX_SEEK
-DOMPI_SKIP_MPICXX=1 -fPIC
CUDR_OPTS = -O2 $(LMP_INC) # -xHost -no-prec-div -ansi-alias

BIN_DIR = ./
OBJ_DIR = ./
LIB_DIR = ./
AR = ar
BSH = /bin/sh

CUDPP_OPT = -DUSE_CUDPP -Icudpp_mini

include Nvidia.makefile
__________________________________________________________

I already checked archives and I couldn't find an exact answer for the
problem that I'm facing. Please help

my suggestion is to abandon this. there is too little to gain.

axel.