Hi all LAMMPS users
I have trouble executing the TAD example (in the example folder). I did not change the input and the program runs but incomplete. It seems that the problem is due to parallel processing because it gives a “job aborted” message as follows:
in.tad (2.75 KB)
log.lammps (245 Bytes)
log.lammps.0 (72.4 KB)
please try applying the following changes to the sources, recompile, run again, and let us know if that resolves the issue.
thanks,
axel.
diff --git a/src/REPLICA/neb.cpp b/src/REPLICA/neb.cpp
index 295e97dee8…3914e2b7d0 100644
— a/src/REPLICA/neb.cpp
+++ b/src/REPLICA/neb.cpp
@@ -59,6 +59,7 @@ NEB::NEB(LAMMPS *lmp, double etol_in, double ftol_in, int n1steps_in,
n1steps = n1steps_in;
n2steps = n2steps_in;
nevery = nevery_in;
// replica info
diff --git a/src/REPLICA/tad.cpp b/src/REPLICA/tad.cpp
index a849e7a242…f9ea044757 100644
— a/src/REPLICA/tad.cpp
+++ b/src/REPLICA/tad.cpp
@@ -43,7 +43,10 @@ using namespace LAMMPS_NS;
/* ---------------------------------------------------------------------- */
-TAD::TAD(LAMMPS *lmp) : Command(lmp) {}
+TAD::TAD(LAMMPS *lmp) : Command(lmp)
+{
/* ---------------------------------------------------------------------- */
Many thanks Dear axel for your reply,
However, I have this problem with Windows version and i’m interested in a solution to this. But I will also try the Linux version to see if there is the same problem and check what you said to resolve it (I will share the result).
For verifying this issue you need to compile LAMMPS from source code and apply the changes I sent. There are no precompiled binary versions that have this fix (yet).
There are two possible ways to compile LAMMPS on windows:
you can install the Cygwin toolchain which provides a compiler toolchain and system library that lets you compile sources meant for unix-like operating systems on windows: https://docs.lammps.org/Build.html
you can install the windows subsystem for linux and then do a regular linux style compilation inside of a “bash” shell session. https://docs.lammps.org/Howto_wsl.html
Axel.
Dear axel
I compiled a Stable version (29 Oct 2020) and corresponding packages on ubuntu. Actually, there was no need to apply the changes you sent because It worked without them (however, using replica=5 gave the same “job aborted” error, but other replicas (e.g. 3,6,8) were ok).
you have the same bug in all versions, but it does not always manifest because that is how “uninitialized data” bugs work. specifically on Linux machines memory that is allocated for the first time, will be consistently zeroed out, which is the reason why those kinds of bugs can go unnoticed for a long time. then when memory is reused, it depends on what kind of data has been in that location before. that can be very random and thus you get random crashes. those changes I sent out are needed. the question is whether they are sufficient.
axel.
Dear axel
Thanks for the great tips you made on this issue. However, what I meant by mentioning the compiled version (stable version 29 Oct 2020) was to give more details about what I had done, not to emphasize the specific version. Actually, you are right, this was the case in all versions (at least all of the ones I reviewed). I will definitely make the changes you mentioned, however, I hope the Windows version will be updated as well.
Dear axel
Thanks for the great tips you made on this issue. However, what I meant by mentioning the compiled version (stable version 29 Oct 2020) was to give more details about what I had done, not to emphasize the specific version. Actually, you are right, this was the case in all versions (at least all of the ones I reviewed). I will definitely make the changes you mentioned, however, I hope the Windows version will be updated as well.
windows versions will be updated when there is time to do it. usually after a patch release, but sometimes there is so much else to do, that it can take longer. sometimes even until the next patch release is done. the windows packages were conceived to be used at LAMMPS tutorials, but then we decided to make them available to everybody, and we can keep them around because the process is mostly scripted. but every once in a while things break or tools get updated and then this needs to be fixed and tested on windows machines, and that is where everything slows down, since it is not very often that LAMMPS developers use windows.
we would be happy to offload this task to somebody else that has more interest in windows and more energy and time to do more testing and make improvements.
but until somebody sufficiently competent and interested volunteers, people are stuck with how it currently is.
sorry,
axel.