Skip to content

Conversation

@pdziekan
Copy link
Contributor

@pdziekan pdziekan commented Mar 16, 2022

Refined cells at the MPI boundary are cut in half by the MPI boundary.
Both MPI ranks calculated refined points between the domains.

TODO:

  • refinee w 1D, 2D
  • ijk_ref i ijk_r2r w 1D i 2D
  • refinee_global? Czy niepotrzebne?
  • logika refinement dla wektorow
  • samo liczenie refined fields: intepolacja (tri-linear?) predkosci, rekonstrukcja fraktalna th i rv
  • przygotuj rysunek siatki: Arakawa-C + sharedmem / distmem decomposition + grid refinement (do papieru i do pomocy)
  • jak liczyc wplyw mikrofizyki liczonej na refined grid na th i rv? sumowac po malych komorkach do duzej komorki? trzeba uwzglednic rozne objetosci refined komorek. Refined cells z MPI sa przeciete na pol, trzeba uwzglednic ich mniejsza objetosc w libcloudph++
  • doesn't compile if reconstruction is not done (e.g. mpi_adv_3d test fails, because of errors in hdf5 without refinement)
  • unit tests

ISSUES:

  • Reconstruction fails if threads have small slices, e.g. UWLCM moist thermal simulation gives wild values of reconstructed theta for:
    OMP_NUM_THREADS=7 ./uwlcm --case=moist_thermal --dt=1 --micro=lgrngn --nt=500 --nx=51 --ny=51 --nz=51 --outdir=out --outfreq=50 --backend=CUDA --sd_conc=10 --coal=0 --cond=0 --n_fra_iter=1
    however works well for less threads:
    OMP_NUM_THREADS=6 ./uwlcm --case=moist_thermal --dt=1 --micro=lgrngn --nt=500 --nx=51 --ny=51 --nz=51 --outdir=out --outfreq=50 --backend=CUDA --sd_conc=10 --coal=0 --cond=0 --n_fra_iter=1
    or for finer refined grid:
    OMP_NUM_THREADS=8 ./uwlcm --case=moist_thermal --dt=1 --micro=lgrngn --nt=500 --nx=51 --ny=51 --nz=51 --outdir=out --outfreq=50 --backend=CUDA --sd_conc=10 --coal=0 --cond=0 --n_fra_iter=2

@pdziekan pdziekan changed the title [WIP] Fractal reconstruction [obsolete?] Fractal reconstruction May 9, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant