Update geometry
- Implement universe cell
- Reimplement the continuously moving surface. Verified on moving planar surfaces. Need verification for quadratic surfaces (next PR).
- The user interface changes: see C5G7-TDX in
examplesand moving pellet test intest/regressioninterface (next PR will update the docs) - The new implementation supports moving quadric surfaces too (still need verification).
- The user interface changes: see C5G7-TDX in
- Major refactoring on geometry-related functions.
- Temporarily remove caching. The current Numba caching mechanics does not work with the refactoring of kernel functions into submodules. We had a plan to reimplement more robust caching mechanics. So, rather than trying to fix the caching mechanics that were soon to be replaced, we removed the caching option altogether and will work on the plan in the next PR.
The iQMC Numba tests failed... But they passed on my local machine... Can you please check if they pass on your end, @clemekay? Thanks!
I cloned your update-geometry branch on Dane, created a fresh MC/DC install from it, checked out an interactive node, and executed python run.py --mode=numba from MCDC/test/regression/. The four iqmc tests that failed on github (iqmc_cooper2, iqmc_kornreich_pi, iqmc_reed, and iqmc_sood_pi) also failed for me with a Segmentation fault.
Interestingly, I originally accidentally ran the tests before executing bash patch_numba.sh and test iqmc_sood_pi passed, but it failed after patching numba. This is my pip list, do all of these versions match what you've got locally?
The recent merge of dev has the planned caching capabilities per #250.
Looks like loosening the coincidence tolerance from 1E-12 to 1E-10 causes inf_shem361_td to fail.
I just reran the test, and the failing MPI-Python test went away...
Merging it now as discussed in the dev meeting on 11/12/2024.