[lammps-users] Compiling the GPU library in LAMMPS

Hi,

I am trying to compile LAMMPS (1May10) with the gpu package in a Linux system. So, as per instructions at http://lammps.sandia.gov/doc/Section_start.html#2_3_3, I am trying to compile the "gpu" library under $LAMMPS_SOURCE/lib/gpu.

This is what I get.

[email protected]...:/usr/local/src/prakash/lammps/lammps-1May10/lib/

make -f Makefile.nvidia

nvcc -I/usr/local/cuda/include -DUNIX -O3 -Xptxas -v --use_fast_math -arch=sm_10 -D_SINGLE_SINGLE -o nvc_device.o -c nvc_device.cu
ptxas info : Compiling entry function '__dummy_entry__'
ptxas info : Used 0 registers
In file included from nvc_device.cu:96:
/usr/include/c++/4.2.1/x86_64-suse-linux/bits/gthr-default.h:96: error: ?int __gthrw_pthread_cancel(pthread_t)? used but never defined
make: *** [nvc_device.o] Error 255

Does anyone know what is going on?

Thanks,
Prakash

Hello all,

I just figured that once I upgraded my CUDA toolkit from 2.1 to 3.0, "gpu" library compiles just fine.

Regards,
Prakash