The horizontal box pass gave each thread 16+ consecutive pixels to slide
a running sum over, so a workgroup's threads read pixels 16 apart. On a
Radeon 780M (DX12) that pass took 9-15x as long as the vertical one, and
it sits under every GPU Gaussian and guided filter (NR, texture, dehaze,
clarity/base). Each thread now sums one pixel's 2r+1 window: adjacent
threads read adjacent pixels. Same 64x4 workgroups, parameters and
bindings; box_v unchanged.
Measured on the Radeon 780M (6000x4000): guided filter at sigma 1.75
828 -> 146 ms, colour-NR Gaussian 495 -> 104 ms. Full-size heavy-edit
export of the CC0 Sony ARW: GPU 3.6-4.0 s -> 1.9-2.0 s (CPU 1.9-2.1 s);
GPU-vs-CPU output distance unchanged.
Adds gaussian_matches_cpu (GPU blur vs lightcraft_raster::blur::gaussian,
1-3 channels, partial workgroups, windows wider than the image) and
per-pass rows to bench_kernels.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
AVIF HDR export takes hdrTransfer = pq | hlg (export dialog: HDR curve;
MCP export schema lists it). HLG writes CICP transfer 18 and no clli box.
The window logs whether it shows HDR; today always SDR (eframe configures
an 8-bit surface), with the platform / surface format / headroom / opt-out
rules unit-tested. Also satisfies clippy 1.99's chunks_exact_to_as_chunks
in spots.rs.
Part of #868.
Sources: ITU-R BT.2100 (HLG OETF, inverse OOTF with system gamma 1.2),
ITU-R BT.2408 (203 cd/m2 reference white at 75 % HLG); own design for
the display decision.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Photos that tie on the sort key (equal ratings, one import batch, a
burst within a second) now sort by file name before id, in
Catalog::query, Recently Added and auto-stack. Split out of #641.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
All 38 files decode from sensor data and develop at full size (9520x6336 and
6336x9520) without error; mean CIE76 dE to the camera JPEG is 6.3-13.9 at 400 px.
The files are not committed and the evidence entry carries no path. Leica is now
14 verified / 47 unverified, the Leica bucket on a verified body 5% -> 21%; the
weighted shares (85% decode, 42% on a verified body) do not move.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
On merge: the HDR histogram is applied in develop_reporting (#712 split develop), is skipped for
draft renders (a second render per drag tick), and HdrBins::bin clamps a hand-built split so it
can't underflow (test hdr_bins_with_any_split_stay_in_range).
photo.editExternal takes the selection (or ids, at most 50): each photo gets its own -Edit copy
stacked on its original, the copies are selected and returned in copies (path / id / original
still name the first). Photo ▸ Edit In opens each copy in the editor.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BKWB9TCnvzJHV6xMLAj2J4
On merge: Lightroom's default profiles (Adobe Standard, Adobe Color), named by nearly every sidecar,
give no camera-profile warning; other profiles still do (test default_lightroom_profiles_are_not_reported).
The import help string keeps main's onDeleted wording and gains warnings.
On merge: the quick preview path orients like the whole-file path (preview_orientation with the
raw's header prefix, so a portrait raw whose embedded JPEG carries no orientation is not sideways),
and its doc links point at preview_window.
Turning on AI Denoise without a model opens Settings > AI Denoise and
keeps a pending request, so the switch showed checked and Settings
offered "Cancel pending action". Closing the setup with no model
download or installation running now drops that request: the switch
goes back off and nothing claims to be pending. A download started
before closing still turns AI Denoise on when it finishes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BKWB9TCnvzJHV6xMLAj2J4
File > Import Profiles & Presets... now builds one toast for both halves
(a second toast replaced the first, so a .cube-only import said "No new
presets"). The preset half is skipped when no preset files were picked,
and profile.import's failures are shown with their reason. profile.import
also reports .cube files and zips it couldn't read instead of dropping
them silently. The command's result carries the profile half under
"profiles".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BKWB9TCnvzJHV6xMLAj2J4
On merge: metadata.downloadPlaceNames accepts plain http only for the loopback address itself
(http://127.0.0.1.example.com/ and http://127.0.0.1@host/ are refused; test
download_base_is_https_or_loopback_only).
