anku94 / carp-utils

smallCARP

512 ranks, 1.6M/rank, Magnetic Reconnection ("expert crafted" to produce a magnetic island)

• reconnection.cc - original, unmodified
• reconnection512.cc - modified to dump suitably

bigCARP

512 ranks, 8M/rank, Magnetic Reconnection ("amateurly modified" from smallCARP for desired scale)

• reconnection.cc - original, unmodified
• reconnection512.cc - modified to dump suitably

gen_compressed_traces.py

This is a multithreaded parser that compresses all the eparticle files generated from a VPIC run into logs of energy values.

Given a $vpic_run/particle path, it will generate a $vpic_run/particle.compressed path with an identical directory hierarchy, except the compressed eparticle files will contain an aggregated property, which is currently computed using the formula sqrt(1 + ux^2 + uy^2 + uz^2) - 1

After a VPIC run has completed, point this script to generated particle directory.

Currently, this only handles the eparticle files and ignores the hparticle files. You can tweak the code to parse both.

Running a deck

• Compile vpic407 to get a VPIC executable.
• Run build/vpic reconnection512.cc to get an executable (usually called reconnection.Linux)
• mpirun -h hostfile -n 512 reconnection.Linux
• This will start the simulation. All data will be dumped in the directory containing the executable. (I usually just put the executabe in PanFS and run things there)
• Periodic dumps are collected in particle/T.ts
• After the simulation completes, I point the gen_compressed_traces script at the particle directory. It creates a particle.compressed directory in the same parent, applying an aggregation function to each particle struct, and collecting a log of floats in particle.compressed.