perf: multi shared_preloads runtime measurement — 1.14x at hst x 4 exposures#60
Merged
Merged
Conversation
…hst x 4 exposures Mesh-geometry-only sharing as designed (PyAutoLens#599 D1): the ~12% saving is the per-exposure image-mesh ray-trace + Delaunay triangulation; consistency (identical source-pixel grid across exposures) is the primary win. Includes the NaN adapt-cache re-run trap in OPTIMIZATION_NOTES. 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Phase 3/4 (autolens_workspace#260): the multi shared_preloads runtime measurement — sibling of
likelihood_runtime/datacube/shared_preloads.py.Headline (recorded in
results/runtime/multi/+ OPTIMIZATION_NOTES): 1.14× per-likelihood at hst × 4 identical exposures (unshared 53279.5 ms → shared 46747.4 ms, local CPU fp64, Hilbert-1500 + Delaunay). As the design predicted (PyAutoLens#599 D1): only the mesh geometry is shareable for imaging — the ~12% saved is the per-exposure image-mesh ray-trace + Delaunay triangulation; the primary win is consistency (identical source-pixel grid across exposures).Also records the re-run trap: a crashed JAX run can poison the gitignored
lensed_source.fitsadapt cache with in-mask NaNs (qhull "Points cannot contain NaN") — delete and regenerate.Scripts Changed
likelihood_runtime/multi/shared_preloads.py(new; ruff clean;AUTOLENS_PROFILING_SMOKE=1short-circuit)likelihood_runtime/OPTIMIZATION_NOTES.md+results/runtime/multi/shared_preloads_hst_local_cpu_fp64.stdoutAutonomy
--autosafe (gate on issue #260; Heart YELLOW set unchanged from in-session ack).🤖 Generated with Claude Code