Building USER-CUDA

LAMMPS LAMMPS Mailing List Mirror
1

Hi,
I have been having some problems with the USER-CUDA building,
I did this:

  1. sudo apt-get install g++ mpi-default-bin mpi-default-dev fftw-dev
    2 wget http://lammps.sandia.gov/tars/lammps.tar.gz
    3 tar xvzf lammps.tar.gz
    4 cd lammps-27Oct11/src
    5 make openmpi
    6 make yes-USER-CUDA
    7 cd …/lib/cuda/
    8 make arch=21 cufft=1
    9 cd …/…/src/
    10 make yes-USER-CUDA
    11 make openmpi

It gives me the ERROR that is reported at the attachment.
I don’t know what’s the problem.

Error USER-CUDA (6.27 KB)

2

Hi

the problem is the linker doesnt find the libcuda, thats a library which comes with the driver (as opposed to libcudart, which comes with the CUDA-Toolkit [default path /usr/local/cuda], and the liblammpscuda which is compiled in lammps/lib/cuda).

My guess is that on the machine on which you try to compile LAMMPS with the USER-CUDA package, you dont have a gpu-driver installed but only the cuda-toolkit.

Cheers
Christian

-------- Original-Nachricht --------