EXIT CODE: 255: when I ran lammps with GPU package with pair_style coul/long

LAMMPS LAMMPS Mailing List Mirror
1

Dear Kohlmeyer:

Thanks a lot, your advice is very helpful and works fine. I also figure out another way to work round this. Given I adopt hybrid pair style, if I set the

2

Dear Kohlmeyer:

Thanks a lot, your advice is very helpful and works fine. I also figure
out another way to work round this. Given I adopt hybrid pair style, if I
set the epsilon and sigma parameters of lj/cut/coul/long/gpu to be both
zero then * * lj/cut/coul/long/gpu will be the same as coul/long/gpu. Do
you think this is adoptable?

​yes, this is just as well. you'll have to benchmark, which of the two
approaches is actually faster.

BTW: it would be sufficient to only set epsilon to zero.​

​axel.​

3

Hi Zhang,

can you try apply the attached patch (i.e. use the command “patch -p1 < patch-coul-long-gpu.diff” at the same level with src/, lib/ doc/, etc.), rebuild the GPU library in lib/gpu, and then rebuild the LAMMPS binary with the new libgpu.a?

Axel, I have submitted a pull request for this issue on github. Let me know if that works.

Thanks,
-Trung

patch-coul-long-gpu.diff (6.71 KB)

4

… ​
rebuild the GPU library in lib/gpu, and then go to src/, “make package-update” so that src/pair_coul_long_gpu.cpp is updated, and then rebuild the LAMMPS binary…

5

Hi Zhang,

can you try apply the attached patch (i.e. use the command "patch -p1 <
patch-coul-long-gpu.diff" at the same level with src/, lib/ doc/, etc.),
rebuild the GPU library in lib/gpu, and then rebuild the LAMMPS binary with
the new libgpu.a?

Axel, I have submitted a pull request for this issue on github. Let me
know if that works.

​thanks trung!

it works very nicely.

​axel.​

6

Dear Trung Nguyen:

Thanks very much!

The method you proposed also works very well for me, and I have compared the results, and they are in accordance with with other methods. Your work is great and should be adopted by next version.