Skip to content

docs(inversion): two-stage curvature kernel docstring — dispatch is on pix_pixels; two-stage wins on Delaunay too (autolens_profiling#274) #557

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Jammy2211 merged 1 commit into from
Sep 17, 2026
Merged

docs(inversion): two-stage curvature kernel docstring — dispatch is on pix_pixels; two-stage wins on Delaunay too (autolens_profiling#274) #557

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 14 additions & 5 deletions autoarray/inversion/inversion/imaging_numba/inversion_imaging_numba_util.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -774,11 +774,20 @@ def curvature_matrix_via_sparse_operator_two_stage_from(
The trade is `sum_pairs u0 * u1` scattered RMWs for
`sum_pairs u1` L1 scatters plus `(sum_data u0 + data_pixels) * pix_pixels`
dense ops. It is therefore *not* unconditionally faster: it wins when the
source space is small relative to the PSF overlap and the mappings are wide
(bilinear, u0 = 4), and loses when the source space is large and the mappings
narrow (barycentric Delaunay, u0 ~ 1.55). `curvature_matrix_via_sparse_operator_from`
chooses between the two from the measured geometry -- see
`CURVATURE_TWO_STAGE_COST_RATIO_THRESHOLD`.
source space is small enough for the accumulator and a row of `F` to stay
cache-resident and the mappings are wide (bilinear, u0 = 4: 2.9x at the HST
rectangular fiducial), and it was expected to lose for narrow barycentric
mappings on a large source space (Delaunay, u0 ~ 1.55). Measured, it does
not lose there either: on the HST Delaunay N=1500 fixed-light cell (u0 =
1.60, one thread, autolens_profiling#274, RAL job 343394) the two-stage
kernel is 1.20x faster than the direct loop at the site (85.0 vs 102.1 ms)
and the direct loop costs 6.4 % more on the whole likelihood call; a
touched-index variant of stage 2, which scatters over and re-zeroes only the
indices stage 1 touched, is slower than the plain two-stage form as well
(97.0 ms), because a 1500-long fp64 AXPY is 12 KB and L1-resident while the
index bookkeeping is not free. `curvature_matrix_via_sparse_operator_from`
therefore selects on the source pixel count alone -- two-stage at or below
`CURVATURE_TWO_STAGE_MAX_PIX_PIXELS` (4096), the direct loop above it.

Parameters and return value are exactly those of
`curvature_matrix_via_sparse_operator_direct_from`.
Expand Down
Loading