Energy mismatching in CPU and CUDA modes for lj/gromacs/coul/gromacs

We’ve tested that the potential energy calculated using pure CPUs and CUDA mode is not the same. We’ve noticed that in 2010 there is a thread talking about the bug about lj/gromacs/coul/gromacs is not adding correctly the energy shift term. Similar to Steve’s early reply in the thread, in present case, only the PE value is affected, while the dynamics is not altered. Therefor, we consider it might be the same problem in the command lj/gromacs/coul/gromacs/cuda in USER-CUDA package.

Chen Chen

Christian can look into this.