I am trying to upgrade some legacy f77 codes to analyse molecular dynamics trajectories. I have had some success with modern fortran with OpenACC directives. My aim now is to read the LAMMPS trajectory files which are usually big, about a few GBs, using GPU. First, I would like to know if this makes sense? If yes, then, how to do it efficiently?
The trajectory file usually contains system configuration for multiple time steps. The data structure has multiple blocks of following structure: a time stamp, a few lines about number of particles and box dimensions and then position, velocity etc. for all particles at that time. This block structure repeats for successive time steps.
I would really appreciate an suggestions regarding this. Thank you!!