[lammps-users] CUDA Bus Error

Hello,

I am trying a simple flex-spc water atom simulation using lammps with CUDA, but I get this error:

LAMMPS (28 Jun 2010)
Scanning data file …
2 = max bonds/atom
1 = max angles/atom
Reading data file …
orthogonal box = (-1.553 -1.696 -1.696) to (11.877 11.453 11.498)
1 by 1 by 1 processor grid
84 atoms
56 bonds
28 angles
Finding 1-2 1-3 1-4 neighbors …
2 = max # of 1-2 neighbors
1 = max # of 1-3 neighbors
1 = max # of 1-4 neighbors
2 = max # of special neighbors
[gyan:07626] *** Process received signal ***
[gyan:07626] Signal: Bus error (7)
[gyan:07626] Signal code: (128)
[gyan:07626] Failing at address: (nil)
[gyan:07626] [ 0] /lib/libpthread.so.0(+0xf8f0) [0x7f33b3d0a8f0]
[gyan:07626] [ 1] /opt/lammps/lmp_cuda_openmpi(_ZN12NVC_ConstMatIfE10safe_allocEmmP16textureReference+0x96) [0x6ba396]
[gyan:07626] [ 2] /opt/lammps/lmp_cuda_openmpi(_ZN13LJ_GPU_MemoryIffE4initEiiPPdS2_S2_S2_S2_S2_S2_S2_S1_iiii+0x251) [0x6bcaf1]
[gyan:07626] [ 3] /opt/lammps/lmp_cuda_openmpi(_Z11lj_gpu_initRiiPPdS1_S1_S1_S1_S1_S1_S1_S0_S0_S0_ddii+0x12c) [0x6b7c3c]
[gyan:07626] [ 4] /opt/lammps/lmp_cuda_openmpi(_ZN9LAMMPS_NS12PairLJCutGPU10init_styleEv+0x2e1) [0x63ddf1]
[gyan:07626] [ 5] /opt/lammps/lmp_cuda_openmpi(_ZN9LAMMPS_NS4Pair4initEv+0xaf) [0x61736f]
[gyan:07626] [ 6] /opt/lammps/lmp_cuda_openmpi(_ZN9LAMMPS_NS5Force4initEv+0x56) [0x5991d6]
[gyan:07626] [ 7] /opt/lammps/lmp_cuda_openmpi(_ZN9LAMMPS_NS6LAMMPS4initEv+0x17) [0x5aed17]
[gyan:07626] [ 8] /opt/lammps/lmp_cuda_openmpi(_ZN9LAMMPS_NS3Run7commandEiPPc+0x9ab) [0x68bd9b]
[gyan:07626] [ 9] /opt/lammps/lmp_cuda_openmpi(_ZN9LAMMPS_NS5Input15execute_commandEv+0x285d) [0x5a7bed]
[gyan:07626] [10] /opt/lammps/lmp_cuda_openmpi(_ZN9LAMMPS_NS5Input4fileEv+0x2e8) [0x5a4c78]
[gyan:07626] [11] /opt/lammps/lmp_cuda_openmpi(main+0xad) [0x5b22cd]
[gyan:07626] [12] /lib/libc.so.6(__libc_start_main+0xfd) [0x7f33b3997c4d]
[gyan:07626] [13] /opt/lammps/lmp_cuda_openmpi() [0x45bc09]
[gyan:07626] *** End of error message ***
Bus error

Machine Details: Ubuntu 10.04, gcc4.4.3, Cuda 3.1, GTS 250 card, Core2Duo, lammps complied with intel compilers11.1.064 and openmp
The NVIDIA sdk examples compile and run correctly.
The input file is (runs correctly without pair style gpu)

Init

units real
dimension 3
newton off

processors px,py,pz (let lammps choose)

boundary p p p
atom_style full
read_data spc-water-10-10-10.data
angle_style harmonic
angle_coeff 1 37.5 113.47
bond_style harmonic
bond_coeff 1 529.583 1.012
pair_style lj/cut/gpu one/node 0 8
pair_coeff 1 1 0.1550 3.166
pair_coeff 1 2 0.0 0.0
pair_coeff 2 2 0.0 0.0
timestep 1.0
dump 1 all atom 500 trajectory.lmptrj
fix 1 all nvt temp 300.0 300.0 100.0
thermo_style multi
thermo 100.0
run 500000

Sincerely,
Ritwik.

Ritwik,

LAMMPS’s GPU modules don’t yet handle the exclusion lists that would be needed for simulating water. We’re working on capabilities along these lines. See: http://code.google.com/p/gpulammps/ You might try using some pre-release code — the USER-CUDA package for LAMMPS that can be downloaded from that site.

Paul