Feature/plume strike vectorization parallel - #121
Merged
andytorrestb merged 7 commits intoJul 19, 2026
Merged
Conversation
Replace the per-face Python loop in compute_plume_strikes() with NumPy operations over all faces per active thruster. The original loop is preserved verbatim as _compute_plume_strikes_scalar() for regression comparison, debugging, and benchmarking. The vectorized core operates on plain serializable inputs (centroids, normals, thruster dicts, config values) so it can later run in process workers. Gas kinetics stays scalar, evaluated only for struck faces. Zero-distance faces and the legacy 3.14 theta constant are preserved so strike counts and struck-face IDs match the scalar path bit-for-bit (verified on all case/rpod/* cases, kinetics arrays exactly equal). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The target mesh is stationary during a JFH run, so jfh_plume_strikes() now computes face centroids once and passes them to compute_plume_strikes() via the optional face_centroids parameter instead of recomputing them for every firing. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
jfh_plume_strikes() gains optional parallel/workers arguments layered over new optional config keys ([exec] parallel, [exec] workers); default behavior remains serial and byte-identical to before. When enabled, independent per-firing strike arrays are computed in a ProcessPoolExecutor (one firing per task). Workers receive only plain serializable inputs -- precomputed centroids, unit normals, thruster dicts, JFH step dicts, and the few needed config scalars -- shipped once per worker via a pool initializer, so memory scales with workers x faces rather than firings x faces. Cumulative arrays (cum_strikes, max_pressures, max_shears, cum_heat_flux_load) are still accumulated serially in firing order and VTK files are written serially by the parent, preserving the firing_data return structure exactly. Invalid parallel/workers values raise clear ValueErrors; runtime parallel failures log a warning and fall back to serial execution. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
plume_unit_test_04 asserts the vectorized compute_plume_strikes() path (with and without precomputed centroids) reproduces the scalar reference exactly -- strike arrays and struck-face IDs bit-for-bit -- for every firing of the 1d_approach (geometry-only) and multi_thrusters_square (Simplified kinetics) cases, plus explicit zero-distance face skipping. plume_integration_test_02 asserts jfh_plume_strikes() default behavior stays serial and return-compatible, that the parallel path (workers=2) matches serial per-firing and cumulative outputs, that workers=1 resolves to serial, and that invalid worker counts raise ValueError. Kinetics arrays use np.allclose(rtol=1e-12, atol=1e-12): observed bit-identical on this platform, tolerance only guards platform BLAS reduction-order differences (rationale documented in each file). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
python scripts/benchmark_plume_strikes.py times the scalar reference vs the vectorized geometry path, and serial vs process-parallel execution, on existing repo cases (default case/rpod/1d_approach; --case, --workers, --repeat, --skip-scalar options). It reports elapsed times and verifies strike equivalence but asserts no speedup thresholds, keeping it stable on shared machines. Measured on this machine: vectorized geometry is ~245x faster than the scalar reference on 1d_approach (10 firings x 3584 faces) and ~24x on multi_thrusters_square with Simplified kinetics. Process-parallel is slower than serial on these small cases because Windows process spawn dominates; it is intended for large meshes/many firings. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Relocate scripts/benchmark_plume_strikes.py to tests/rpod/rpod_integration_test_06.py, following the established rpod test format (unittest.TestCase, header description block, ../case/rpod paths, conftest-derived rpod/integration markers). The benchmark prints elapsed times (visible with pytest -s) for scalar-vs-vectorized geometry on 1d_approach and multi_thrusters_square, and serial-vs-parallel execution with 2 workers; it asserts only exact strike equality, never timing thresholds. No existing test files are modified. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The full-file rewrite in 5f44163 wrote the file with LF endings, while the rest of the repository (including this file on master) uses CRLF. Because core.autocrlf is true with no .gitattributes, the EOL flip made git treat all 137 original lines as deleted-and-re-added, inflating the diff to +417/-137 when the real change is +281/-1. Restoring CRLF (content byte-identical to the previous commit) collapses the diff to the actual change. No functional change. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.