Final version of flow36 DNS solver for turbulent channel flow with dipersed solid particles; No-slip boundary condition only;
In the source_code you can find the CPU only version of the solver, parallelized according to a 2D decomposition (pencils) with MPI; no test case is implemented yet; use the available prints to generate one (simpler cases)
LPT: tracers, finite spheres and prolate spheroids are currently implemented, 2-way coupling according to the ERPP method is being tested.
To launch copy the proper Makefile from the folder makefiles, make all and then launch by mpirun/mpiexec -np ** ./surf36.exe, where ** is the product nycpu*nzcpu.
The GPU version is in the folder /source_code/gpu/source_code/; currently, it is a serial version based on a local data decomposition that is specified in the input.f90 file. Larger the local slabs, higher the achievable memory bandwitdh during the memory copies. The GPU version is parallelized according to a 1D (slabs) decomposition between nodes, whilst each slab is parallelized between multiple GPUs on the same node in the other direction, so that actually we are talking about local pencils (not yet implemented).
Remember to create the subfolders /source_code/results/backup before running