USER-CUDA can't write restart file

Hi all,

I am using CUDA package doing simulation of 67000 atoms with eam/fs/cuda pair_style. When I was trying to write restart file the following error message comes out:

malloc.c:3096: sYSMALLOc: Assertion `(old_top == (((mbinptr) (((char *) &((av)->bins[((1) - 1) * 2])) - __builtin_offsetof (struct malloc_chunk, fd)))) && old_size == 0) || ((unsigned long) (old_size) >= (unsigned long)((((__builtin_offsetof (struct malloc_chunk, fd_nextsize))+((2 * (sizeof(size_t))) - 1)) & ~((2 * (sizeof(size_t))) - 1))) && ((old_top)->size & 0x1) && ((unsigned long)old_end & pagemask) == 0)’ failed


I am running lammps on Ubuntu 11.10 with one GTX 560 graphic card and one i7 cpu.

Here is the input file:

units metal

boundary p p p
atom_style atomic

read_data Fe-5.lammps

mass 1 55.845
pair_style eam/fs/cuda

pair_coeff * * Fe_mm.eam.fs Fe

timestep 0.01

thermo 1000

thermo_style custom step temp pe etotal press vol

fix 1 all nve/cuda

dump atoms all custom 100 fe-5 id x y z

run 1000

unfix 1

write_restart restartfe-5

Thank you all!



just wanted to let you know, that I reproduced the problem (apparently not occuring with the simple eam style just with the eam/fs). I'll try to fix that now.


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

See the 10 Feb patch.

Thanks Christian,


that issue probably affects all regular eam and related (eim?) potentials, too.


Hi all

no the issue is USER-CUDA specific. I use one more coefficient (making it 8 instead of 7) for some of the splines. The 3rd is just replicated. Thats because I load them as float4 or double4 from texture cache, and there are situations where you either need coefficients 1,2,3,4 or 5,6,7,3.

Also the /fs and /alloy are derived from the pure cuda style so that should fix the issue for all three of them.


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