* fix(compose): Preserve native CMYK fills through PSD round trips
Share depth-quantized ink samples between rendering and PSD caches.
Use the document profile for RGB fill colours and preserve valid imported
caches. Uncached CMYK fills use CPU rendering.
Cover all sample depths, gradient styles and tiled renders. Fix the real
16-bit gradient round trip and raise the psd-tools floor from 307 to 308.
* perf(raster): Reuse exact 8-bit CMYK conversions within rows
Cache the previous ink bytes and ICC result per row, with alpha decoded
independently. Keep the 16/32-bit conversion loops unchanged.
Matched 1 MP reads improve by 77% for smooth ramps, 33% with dithering
and 10% for random colours. Bitwise regression coverage includes all
depths, alpha and tile boundaries.
* Distance fields: the Euclidean passes run on threads, results unchanged (#2119)
The chiselled bevel fields (#2006) and the precise glow fields are built
from an exact Euclidean distance transform, a few sweeps over the whole
map that all ran on one thread. They now run on scoped OS threads, in
groups of rows or columns handed out from a queue: edt_nearest (the
column pass, then the row pass), sharp_edges (rows), refine_edge_normals
(one edge pixel each) and the final per-pixel maps of the bevel and plain
Euclidean fields. Not rayon, as for the blurs (rows_par): maps are built
from inside rayon tiles, where a waiting worker steals other tiles (#276).
Maps under 65536 pixels stay on the calling thread, wasm always does.
Every unit is computed from its own inputs, in the same arithmetic as
before, so the fields are bit for bit the old ones whatever the thread
count: 216 fingerprints (bevel in and out, plain in and out, bevel_maps on
bars at 0 to 150 degrees, thin bars, discs, rounded rectangles, soft,
patchy and half-opaque masks, 40 stars, tiny and 1 pixel maps) are equal
at 1, 7, 16 and the default threads, and equal to the code before.
edt_nearest returns the nearest pixel as u32 (it was usize), which takes its
working memory from 20 to 16 bytes per pixel; a map past 2^32 pixels or a slice that does
not match its size gets "far" distances instead of a panic.
CPU compositor, 12 MP, bevel size change (ms, bevel_bench), main -> this:
ellipse 474 -> 217 (size 8), 990 -> 318 (250); text 747 -> 214 (8),
1110 -> 267 (250). A distance_field call on a 256x256 map goes from 2.2 to
1.8 ms, on 512x512 from 8.0 to 2.8, on 2048x2048 from 99 to 15.
Tests: par_units hands every unit to the closure once with its own
slices (short last unit, no second slice, units of different sizes, 1 to
64 threads); the nearest-pixel transform equals itself on 1, 2, 3 and 8
threads on awkward sizes and returns a seed at the distance it reports;
degenerate sizes; sharp edges and the four Euclidean fields are bit-equal
on 1, 2, 5 and 16 threads on discs, rounded rectangles, bars at several
angles, patchy opacity and an edge on the border. The nested-rayon
deadlock test (#276) now has a chiselled bevel layer in its document.
* GPU effect maps: the Euclidean fields are one window, not bands (#2119)
A bevel size change rebuilds the distance fields over the whole effect
region (the region grows with the size). The GPU path cut that into bands
of max(128, halo) rows, each computed from a window grown by the halo,
which for the bevel fields is reach + 10: at size 250, 262-row bands from
windows of about 786 rows. Every pixel went through the transform and the
edge refinement about three times, a 12 MP region gave about 12 tasks for
32 threads, and the two bevel fields ran one after the other.
The Euclidean fields (chiselled bevels, precise glows) now come from one
window, and distance_field spreads its own passes over the threads. A
rebuild is the whole region, which is also what the CPU compositor
computes, so the GPU canvas gets exactly the CPU's field: bands resolved
ties between equidistant edge pixels per window, and on a rebuild they
differed from the whole region by up to 0.07 px (ellipse) and 1.3 px (600
discs) at size 250, on a few dozen pixels in 12 MP. A partial update (a
brush dab) still computes the damage grown by the halo, from one window.
The chamfer fields (shadows, strokes, spread) are a sequential sweep and
keep the bands, and so does wasm, where there are no threads to spread
over and bands keep the memory small.
RTX 4080 SUPER, release, 12 MP, ms per size change on the GPU canvas
(bevel_bench from #2118, p50, median of 3 runs, before #2006 / main /
this), ellipse: 90 / 103 / 98 (8), 162 / 188 / 119 (50), 286 / 446 / 219
(250); text: 83 / 139 / 113, 140 / 215 / 130, 222 / 489 / 165. Alone in the
process (--no-cpu) ellipse at 250 is 407 -> 157 and text 468 -> 169.
36 MP ellipse at 250: 779 -> 382. Peak RSS: 12 MP at 250 1604 -> 862 MiB,
36 MP at 250 2987 -> 1726 MiB; at size 8 on 36 MP it is 163 MiB more
(1380 -> 1543). Precise glow fields on 12 MP: 46-84 ms -> 26 ms at reach
52-252 (32 ms at 12).
The window test now checks that the whole region is bit for bit the CPU's
field, and that a too small halo is noticed on a part of the region.
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
An RGB pattern (every built-in one) showed in colour as a pattern fill or
Pattern Overlay in a Grayscale or CMYK document, unlike Edit > Fill, the
export, and Photoshop, which fills in grey but keeps the pattern's colours
(a Grayscale round trip brings them back).
pattern::pattern_rgba projects a pattern's pixels through the document's
mode when painted; the CPU prepared tiles and the GPU pattern textures use
it. Patterns themselves are unchanged.
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
* Layer styles: chiselled bevel facets are flat at any edge angle (#1926)
Chisel Hard and Soft shaded the facets of an edge at 30, 60, 120 or 150
degrees in regular cross-stripes (a comb along the edge), while 0, 45, 90
and 135 degree edges were clean. The height of a chisel is a linear ramp of
the distance to the edge, and that distance was measured to the nearest
pixel centre of a thresholded mask: an oblique edge is a pixel staircase,
so the distance ripples with its period (0.19 px at 60 degrees), and the
shading is a gradient of it, scaled by depth.
Chisel bevels now measure to the edge itself. A sharp anti-aliased edge
pixel (partly covered, touching an empty and a full pixel) places the edge
by its coverage along its own normal (edge_offset inverts the area a
straight edge cuts off a pixel), and a pixel measures to the line of the
edge pixels around the foot of its perpendicular, not to the centre of the
nearest one (edge_distance). The ripple drops to 0.01 px. Hard-edged masks
and soft ramps have no sharp edge pixel and give exactly the plain fields.
The GPU compositor takes the bevel distance fields from the CPU through
new FieldKind::BevelInside / BevelOutside (the glow fields are unchanged);
distance_field takes the reach so values past it are not refined, and the
band halo of the bevel fields grows to reach + 9.
* Chisel bevel: edge normals from a line fit, exact at hard stretches (#1926)
A follow-up to the previous commit, from measuring how far it was from the
best it can do. The 3x3 coverage gradient gives an edge's direction to 1-3
degrees, and the error is systematic, so every distance tilted the same way
and left a 0.010-0.017 px ripple. With the true normals the ripple is 0.002
to 0.004 px, so the normal was the limit.
refine_edge_normals fits a line through the edge points (pixel centre moved
to its edge) of the sharp pixels within 3 px that agree to 25 degrees, and
takes the line's normal where the points really are on a line (spread along
it at least 0.7 px, off it at most 0.1 px). Corners and tight curves keep the
gradient's normal. On straight edges it reaches the ideal-normal floor.
The inside field now seeds from the empty pixels, as the plain field does,
and measures to the sharp pixel next to the nearest empty one, capped by the
plain distance (which never falls short of the true one). A circle's pole,
or a rounded rectangle on whole pixels, has a stretch of edge exactly on a
pixel boundary with no sharp pixel; the distance to the sharp pixels alone
was up to 0.29 px too long there. A mask without sharp pixels still gives
exactly the plain field.
Measured on anti-aliased shapes: ripple on straight edges 0.0105 -> 0.0038
(0.0172 -> 0.0021 at 75 degrees); a disc of radius 70, inside, worst error
0.29 -> 0.02. The band halo of the bevel fields is reach + 9 and the banded
test now sees a halo that is too small (the old one passed at reach + 3).
* Chisel bevel: sparse sharp-edge storage, no full-map copies (#1926)
Measured before sending the fix: the new
bevel fields made a bevel size change in the GPU canvas about twice as slow
(12 MP: 84 -> 161 ms per step), and nearly all of a step is the CPU effect
maps. The cost of the normal refinement was memory, not arithmetic: a
Vec<Option<SharpEdge>> is 16 bytes for every pixel, refine_edge_normals
copied the whole map twice for a few thousand edge pixels, and
local_coverage cloned the whole map and read nine pixels for each.
SharpEdges keeps a 4 byte slot per pixel and a short list of the edge
pixels. The candidates are found by a row by row pass over which pixels are
covered (an edge pixel is covered and has an empty neighbour), and the
coverage is worked out where it is needed by the same formula as
local_coverage. The refinement works on the list. Nothing else changes:
the fields are bit-identical (FNV fingerprints of 27 maps at 5 reaches and
of whole bevel_maps at 3 sizes, 216 of 216 equal), and the rendered PNGs of
the issue scene and a realistic scene are byte-identical.
Release, 12 MP ellipse, size 8, one thread: new fields 408 -> 351 ms (old
fields 290); the sharp pass 48 -> 19 ms and the PCA step 32 -> 1 ms.
GPU canvas, ms per size change, main / before / after: 12 MP ellipse
84 / 161 / 96, 300 stars 122 / 246 / 166, text 91 / 184 / 139, 36 MP
ellipse 295 / 522 / 336. What is left on edge-heavy content is the
neighbour search and edge_distance, which this commit does not touch.
A layer with Transparency Shapes Layer off takes the whole canvas as its
effect region, so on a 1920 x 103675 page each such stroked layer built
~2.4 GB of effect maps, and a render held them for every layer at once.
Every edit's damage rect rebuilt them: 13 s and 11+ GB for three layers,
past 24 GB for ten.
When such a layer's effects are local (no gradient strokes, Precise
glows, chiselled bevels or noise) and it is composited within the
rectangle being rendered (each band of render_bands, or a damage rect),
build its maps over that rectangle padded by twice the effect reach, and
let them go when the next band starts. Composited over anything wider (a
group's effect shape renders its children over the group's whole region)
it keeps the whole region. The pixels are unchanged: a test renders every
local effect kind, masked and unmasked, inside a group with far-reaching
effects and clipped to a base with a wide glow, at 8/16/32 bits, in bands,
parts and tiles from a cold effect cache, and compares bit for bit with
the whole render.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Outer Glow now gets the same fill-dependent knockout as Drop Shadow, on CPU and GPU, so a Fill 0%
shape with a glow no longer paints the glow across its interior. Fixes#1953.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Levels and Curves composed their master and per-channel records in opposite orders (#975). The
Photoshop-authored corpus decides it: channel first, then master. Levels now does that, bringing
psd-tools levels_rgb.psd from 0.447 to 0.012 max error; Curves (curves_rgb.psd passes) is unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Concurrent requests for the same cold combined mask each rasterized and feathered the whole
image. A per-key OnceLock slot now lets one caller build while the others wait for the same
tiles; different keys build independently outside the cache mutex, and the result is admitted
through the byte-budgeted cache with the generation taken at lookup, so a purge mid-build still
wins. The vector mask renders sequentially inside the slot so a Rayon worker can't wait on itself.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The combined-mask cache was capped only at 64 entries, so large feathered masks could retain a
lot of memory, and Edit > Purge > All left them allocated. The cache is now also bounded to
512 MiB of accounted pixel bytes (result plus the pinned input), Purge All clears it and reports
its bytes, and a purge generation stops a build that started earlier from refilling the cache.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Linear Dodge (Add) and Divide keep values above 1 in 32-bit float documents, clipped only at f32::MAX. Integer depths still clip at 1.
- CPU: add psblend::HDR and generic::blend_rgb_hdr / blend_channel_hdr; render sets the flag per document depth.
- GPU: add the F_HDR pass flag, and compose.wgsl uses it for the same clip bound.
- Non-separable modes are unchanged.
- Tests: colour unit tests, compose depth tests, and the GPU parity test float_documents_blend_past_white.
Co-authored-by: hideoneric <hideoneric@users.noreply.github.com>
A crop with Delete Cropped Pixels off keeps what it cuts away (and a moved layer can reach past
the edges), but the Crop tool drew only the canvas, so a frame couldn't be pulled back out over
those pixels by eye. Photoshop 25.4 (observed) grows the view to every layer's bounds and the
frame from the first press on the frame: hidden pixels under the shield, transparency as the
checkerboard, no canvas border or shadow, until the crop is committed or cancelled.
- crop_ui: `CropState::editing`, set by the press (before egui's drag threshold) and cleared when
a new default frame is made (commit, Esc) or the tool is left; `shows_beyond_canvas`.
- canvas: `draw_beyond_canvas` composites only the ring between the canvas and every layer's
pixels (`extra_cmds::reveal_all_bounds`, shared with Image > Reveal All), once per document
state, at most MAX_TEXTURE texels per side (from a downsampled proxy past that), and draws it
with the checkerboard out to the frame under the canvas, reaching a point under its edge so the
edge's anti-aliasing has no seam. The drop shadow and the Line border are off meanwhile.
- compose: `render_reduced_rect`, the reduced render of any area of the document plane.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A Satin size or a Bevel & Emboss size or softness of 1e308 reached the
box blur unclamped: the kernel radius saturated to ~2^62 taps and the
render panicked with "capacity overflow"; 1e6 ran for minutes. Drop
Shadow survived because its reach is clamped in margin(). The blur width
is now capped at MAX_REACH (and NaN is 1) in one place, box_width, which
both the CPU tent and the GPU's tent_kernel go through, so values from
commands and from files are bounded alike.
Claude-Session: https://claude.ai/code/session_01UHgFEHEsefoQoBParH47JA
Co-authored-by: Claude <noreply@anthropic.com>
* Advanced Blending: knockout, blend as group, transparency shapes, masks hide effects (part of #215)
Blending Options > Advanced Blending now works end to end:
- doc: AdvancedBlending on every layer (knockout none/shallow/deep, blend
interior effects as group, blend clipped layers as group, transparency
shapes layer, layer/vector mask hides effects), saved in .pcraft.
- PSD and layered TIFF: knko, infx, clbl, tsly, lmgm, vmgm read into the
model and written back (in place, or added when not at the defaults);
the TIFF byte-order transcoder handles lmgm/vmgm as byte flags.
- CPU compositor (compose/src/advanced.rs): shallow and deep knockouts
(isolated groups and clipping groups stop a deep knockout), clipped
layers blended individually, effects over the whole layer when
Transparency Shapes is off, effects built from an unmasked copy when a
mask hides them, interior effects under fill opacity.
- GPU: documents using Advanced Blending fall back to the CPU compositor.
- engine: layer.layerStyle.blendingOptions takes channels, knockout and the
five switches; inspect reports them; Copy/Paste Layer Style carries them.
- UI: the Blending Options page laid out like Photoshop's (General,
Advanced Blending, Blend If with split sliders), translated.
psd-tools corpus oracle 229 -> 236 (seven knockout files now pass; floor
raised); other corpora and all round trips unchanged. Scorecard
FILE-215-16 -> partial.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Polish catalog rows for Advanced Blending; regenerate docs/scorecard.md
pl.tsv landed on main and the coverage test needs every catalog to translate new UI strings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Undo damage: compare Layer.advanced in layer_pixels_equal; regenerate scorecard
#810's exhaustive Layer destructure must classify the new advanced-blending
field; it changes the composite, so it is compared.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Regenerate docs/scorecard.md
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: leon <claude.udder842@passmail.net>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Saturation −100 greyed to the HSL lightness, so a yellow (255, 255, 0) became grey 128 where
Photoshop shows near-white: a yellow logo with Brightness/Contrast +150 and Vibrance › Saturation
−100 clipped to it stayed yellow under Lighter Color. Measured black-box on Photoshop 25.4
(16-bit runs of 7,400-colour charts and HSV sweeps; sRGB, Adobe RGB and ProPhoto documents), both
sliders work in linear light through the document tone curve, Vibrance first, then Saturation:
- Saturation moves each colour from the grey 0.288 R + 0.712 G (blue has no weight, in every
profile) by 1 + s, clipping per channel: within 0.07/255.
- Negative Vibrance is an HSV change in linear light keeping the hue, in closed form: 0.2/255 on
average, the deepest shadows (V < 0.05) drift.
- Positive Vibrance: the brightness term is exact; the saturation boost (stronger for dull and
dark colours, damped on reds and skin) is a fit, 0.4 (+10) to 5 (+100) /255 on average.
CPU (compose::adjust::vibrance_px) and GPU (compose.wgsl) share the constants; the parity test
gains a negative Vibrance and Saturation −100. photocraft-corpus adjustments/rgb8|rgb16/
vibrance.psd (Vibrance +60, Saturation −15): 25.6/255 mean, 89 max → 0.38 mean, 5 max.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Rendering set the thread-local LAB_MIX to the document's value around each
tile, then set it to false. A nested rayon job (a gradient fill, say) can
let the same thread render another tile in the middle of an outer one;
clearing the flag there made the rest of the outer Lab tile mix in sRGB.
The issue's reproduction (a 4096 px Lab document with a gradient fill and
two 50 % fills, 60 renders, RAYON_NUM_THREADS=4) showed whole 256 px tiles
differing between identical renders.
psblend::with_lab_mix sets the flag, runs the closure, and restores the
previous value with a drop guard, as with_cmyk_space does for the CMYK
space. The three set/clear pairs in render_tiled_with use it.
The new test sets the flag, renders Lab and RGB documents on the single-
and multi-tile paths, and checks the flag is what it was. It failed before
the fix. The issue's reproduction, run twice each way on macOS with four
threads: before, 6 and 2 of 60 renders had one whole tile wrong (65536 px);
after, 0 of 60 both times.
Fill 0 % + an Outside stroke is Photoshop's "outline only" look, but on pixel
layers the stroke lay beneath the whole interior at full coverage, relying on
the layer composited on top to hide it. With the fill lowered the interior
showed the stroke colour (a solid disc instead of a ring).
The share beneath an interior pixel is now (1 - a) / (1 - c), with a the
shape's coverage and c the layer's alpha after fill: 1 under an opaque
layer (100 % fill renders exactly as before, so PSD oracles don't move),
1 - a at 0 % fill. CPU compositor and the wgpu shader agree (the layer's
alpha now rides along the coverage seed for pixel layers too).
Tests: compose regression at 8/16/32-bit (0 %, 50 %, 100 % fill), and a GPU
parity case at fill 0 and 0.5 (fails without the shader change).
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* CPU effect maps: key fill layers by their cached pixels, drop opacity/visible
The layer-effect map cache keyed a Fill layer by its Fill spec alone.
But what shapes the maps is the layer's cached pixels whenever they
still match the fill (what a PSD import stores and Image > Transform /
crop / canvas ops rewrite in place, without touching the spec): after
such a rewrite the CPU compositor kept serving the pre-rewrite maps, so
flattens and the CPU canvas showed the drop shadow of the pre-rewrite
shape while the GPU canvas was correct - until budget eviction or a
purge. The key now hashes the cache surface's tile fingerprint while it
matches the fill, like the GPU's fx::shape_key/group_key already did.
layer.opacity and layer.visible leave the key: the maps depend only on
fill_opacity (opacity is applied when the maps are composited,
visibility only skips the lookup), so an opacity slider drag or a
visibility toggle rebuilt every shadow and glow map over the whole
layer for nothing. Again matching the GPU key.
* Effect-map key: keep child visibility/opacity/blend for groups
A styled group's maps come from its children's composite, so hiding a
child or changing its opacity (or blend, clipping) must change the key;
only the top layer's own visible/opacity stay out. Adds a warm-vs-purged
render test for a group with a drop shadow.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Effect-map test: name the edit tuple type (clippy type_complexity)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
vector::Paint::from_fill discarded every gradient field beyond stops,
angle, scale and style, so Photoshop files whose shape fills use opacity
stops, colour midpoints, a centre offset (Ofst), dither or no
"Align with layer" rendered with the wrong ramp, alpha, position or
banding. Paint::Gradient now carries the full ramp (sorted stops,
midpoints, opacity stops, dither) and the compositor's geometry
(centre offset, whole-pixel end-point snapping, pixel-corner sampling,
the align frame choice: shape bounds or canvas), with shape strokes
painting through the same path.
The vector crate cannot depend on compose, so the ramp and geometry
mirror effects::gradient_units/gradient_t, fill_layout::gradient_layout
and gradient_fill::{Ramp, midpoint_remap, dither}. A parity test in
compose (which sees both) renders each non-default field, style and
frame as a shape layer and as a fill layer and requires equal pixels;
from_fill binds every Fill::Gradient field — no ".." — so a new field
fails to compile until the shape renderer honours it.
- Color Lookup: a stored document's `size` is untrusted. `size³ × 3` overflowed for sizes of
2²² and up: debug builds panicked, release builds wrapped past the table-length guard and
indexed out of bounds. One checked `photocraft_doc::adjust::lut3d_len` now backs the CPU
compositor, the GPU planner and the PSD writer.
- Quick Selection: a stroke point far off the canvas (`select.quick {"points":[[3e9,0]]}`)
saturated to i32::MAX and then overflowed on `+ 1`.
- Link Layers: `max + 1` over stored link group ids overflowed at u64::MAX (debug panic, release
wrap that could merge unrelated layers). A fresh id is now the smallest unused one when the
largest is taken; a selection's first group is read with `first()`.
- Vanishing Point clone footprint: the same `as i32 + 1` saturates (hardening; no reproducer).
* Photo Filter and gradient glows render like Photoshop (#442, #443)
Photo Filter multiplied encoded RGB by the filter colour and rescaled to Rec.601 luma. Photoshop
multiplies the pixel's D50 XYZ (relative to white) by the filter colour's, mixed by density, and
Preserve Luminosity restores the encoded luminosity like the Luminosity blend mode (SetLum +
ClipColor); 32-bit documents keep luminance Y. rgb16 photo-filter.psd now matches within 0.6/255
(was 125/255); rgb8 mean error 0.3/255 (was 20.8).
Outer and Inner Glow gradients were laid out across the canvas (a vertical linear gradient), so
the colour bands ran top to bottom. Photoshop colours the glow at strength v with the gradient at
1 - v (first stop at the edge) and the glow is opaque from v = range^2. outer-glow-gradient.psd:
mean 1.4/255, max 9.9 (was 252).
CPU and GPU paths updated together; photoshop oracle floor 132 -> 133.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Merge origin/main; regenerate docs/scorecard.md
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
The live on-canvas filter preview runs the real command on a downsampled
proxy of the document, but the proxy dropped the selection, so on documents
large enough to need a proxy (over ~5 MP) the preview filtered the whole
layer while OK only changed the selection. The proxy now downsamples the
selection with the pixels, so the preview is limited to the (feathered)
selection exactly like the result.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
* layer style: instance-aware engine apply (layer.layerStyle.replace) with param overlays, contour presets and factory defaults
* layer style: instance-based dialog state preserves every effect; Preview checkbox gates the live preview
* layer style: Styles swatches and New Style… save the dialog pending state (style.presets.new effects param)
* layer style: Contour picker and Noise slider for shadows and glows (CPU-rendered noise; GPU falls back to the CPU for noisy effects)
* layer style: Make Default / Reset to Default per effect kind (user defaults persist with the presets)
* layer style: double-stroke band regression (manga case) and control-protocol docs for the instance-based dialog
* layer style: make Styles a top-level tab displaying preset swatches in the right panel
* i18n: translate the Layer Style preset strings in de and pt-br
The German and Brazilian Portuguese catalogs landed on main after this
branch added "Style {name} saved", "Make Default" and "Reset to
Default"; every_tl_literal_is_translated needs them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* presets: move the stray BTreeMap import to the top of the module
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Shapes: a hairline vector stroke no longer darkens the shape's edge pixels
A stroked shape with layer effects (or clipped layers) is drawn as our
fill plus a vector stroke fitted to Photoshop's own pixels: where those
pixels reach past our fill, the stroke takes the alpha the fill doesn't
explain, `(pixels - fill) / (1 - fill)`. It kept our stroke's colour for
that alpha. When Photoshop's fill covers more of an edge pixel than ours
(shapes snapped to whole pixels with Align Edges, our path a fraction off
the grid), the fitted alpha is mostly Photoshop's fill, so a 0.25 px grey
border on a white card was painted as a solid grey line: the edge came out
at the stroke's full colour (122) instead of Photoshop's 220.
The fitted stroke now takes the colour that, drawn Normal over our fill,
gives back Photoshop's pixel (un-compositing the premultiplied colour),
and keeps our colour only where the fitted alpha is negligible. Where
Photoshop's pixels agree with our fill this is our stroke's colour, as
before.
Test: hairline_vector_stroke_keeps_photoshops_edge_pixel_colour (edge
pixel 0.478 before, 0.86 after). The compose, gpu, io and engine tests
and the psd/io/engine corpus tests pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Shapes: neutral wording in the hairline-stroke fix; test it at 8, 16 and 32 bits
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: EthanSuquet <292885762+EthanSuquet@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Tanish Manku <tmanku.dev@gmail.com>
* 32-bit documents: Levels don't clip the input and output ranges
Photoshop's float Levels keep the curve going past the black and white points
and use a plain power for the gamma. rgb32 and gray32 levels.psd now match
within 0.5/255 (was 10/255).
* Raise the Photoshop oracle floor to 132
* Levels: destructive Image › Adjustments › Levels doesn't clip in 32-bit either; neutral comment wording
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* 32-bit documents: Exposure works on the stored linear values
Their samples are linear already, so going through the 2.2 curve applied
the gamma twice. rgb32 and gray32 exposure.psd now match Photoshop within
0.5/255 (was 69/255 and 44/255).
* Raise the Photoshop oracle floor to 130
blend_text_gamma_setting switched the compositor's process-wide text gamma
(psblend::set_text_gamma) to 1.0 and back while other engine unit tests ran
in parallel threads. turns_at_every_depth_and_mode flattens a document with
a type layer before a turn and again after edit.undo; when the toggle landed
between the two flattens, anti-aliased glyph edges blended under different
gammas and the bit-exact comparison failed (seen once on Windows CI).
Move the gamma test to crates/engine/tests/text_gamma.rs (its own process),
document the hazard on set_text_gamma, and make the undo assertion report
the turn, depth, mode, differing pixel count, max difference and first pixel.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Reserve up to 64 MiB of converted pattern pixels and 64 entries for
one render call, including pending initializers and all export bands.
Initialize outside the map lock on the requesting CMYK-scoped worker.
Budget exhaustion retains the existing uncached conversion path;
the retained-payload budget is not a cap on transient fallback buffers.
Keep OnceLock geometry initialization sequential so Rayon work stealing
cannot re-enter an unfinished slot. Raster coverage and scratch remain
local to each requested rectangle; export bands share only geometry.
Keep full-region neighbourhood effect maps while rendering content only for the output rectangle. Compare complete tiled output at U8, U16 and F32 depths rather than three sampled pixels.
Keep effect and fill sampling separate: empty effects use a gray ramp,
while empty fills stay transparent and interpolate Color.alpha.
Preparation stays inside the active document CMYK scope.
Evidence:
- 24 MP RGB gradient overlay: warm median 362.752 ms to 235.813 ms.
- 24 MP RGB gradient fill: warm median 129.264 ms to 117.642 ms.
- Five release samples, eight Rayon workers, matching pixel digests.
Pass-through groups snapshot only for mixing or visible clipping. Adjustments retain one independent mutable result and read the untouched original backdrop pixel before replacing it.
* fix(ui): brush and eraser show the real stroke while drawing
While dragging, the Brush and Eraser drew a hard, flat egui polyline at
the brush width; the soft brush only appeared on release.
The engine's new `brush_cmds::LiveStroke` shares `paint.stroke`'s setup
and renders incrementally with `StrokeRenderer` onto a copy of the
document. The canvas shows that copy and recomposites only each step's
damage (preview key = stroke key + step). The commit passes the live
stroke's seed, so its pixels match the preview, and reuses the canvas so
only the stroke rect refreshes.
Downsampled CPU textures (documents over 4096 px, the Navigator) now
update partially too, aligned to the downsample factor, instead of
re-flattening the whole document.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Live stroke on the large-image stack: reduced CPU textures, pen tilt, display key
Adapts #64's live Brush/Eraser stroke to #71 (banded/reduced canvas), #66
(colour-managed display) and #74 (stylus tilt):
- compose: `render_reduced_damage` renders just the output pixels of
`render_reduced` whose source area meets a damage rect, with bit-identical
values (same spans, same summation order). The CPU canvas texture of a
document over 4096 px now updates partially from it (strokes, live steps,
any damaged edit) instead of #64's factor-aligned `downsample`, which
didn't match #71's area-reduced texture; uploads still go through
`display_image` (document -> monitor profile).
- ui-egui: the live stroke feeds each point's pen tilt/rotation (the same
values `Stylus::stroke_points` gives the commit), so tilt-driven brushes
preview what they commit; the first points come from the drag after
`Stylus::begin_stroke`.
- Preview keys keep #66's display key folded in; `damage_since` un-folds
the cached keys to compare against the live stroke's steps, and the commit
resets the caches to the document's key (raw 0 ^ display key).
- Stroke counter is masked so live keys stay in their 2^44 band.
6000x4000, 120 px soft brush, release: LiveStroke::push 0.72 ms mean
(p95 1.0), GPU damage refresh 1.6 ms mean (p95 4.7); CPU reduced-texture
rect 0.75 ms vs 135 ms for the full 3000x2000 reduction it replaces.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
* Compose: render in place and in bands; area-reduced and proxy thumbnails
render_tiled wrote every tile into a Vec and then copied them into the output:
two full-size float copies (6.3 GB for a 14000² document). Tiles are now
written straight into the output band by band.
New streaming APIs so large documents never need a full-size float
composite: render_bands (sink per band of ~8 MP), render_reduced (banded
area average, bit-identical to the old flatten-then-average thumbnail) and
flatten_to_surface (Flatten Image, Merge Visible, Copy Merged and the other
flattened-surface users).
thumbnail() reduces a downsampled proxy (moved from ui-egui to
compose::proxy, now also scaling spot channels) for documents over 16 MP
when that is faithful (no layer effects or vector masks), so navigator,
channel and histogram thumbnails of a 196 MP document cost milliseconds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* GPU canvas: request the adapter's real limits; banded CPU fallback; navigator thumbnail
egui-wgpu asks for an 8192 px texture limit, so the wgpu compositor rejected
every document wider or taller than 8192 px and each refresh fell back to a
full CPU flatten. The desktop and web apps now request the adapter's own
max_texture_dimension_2d and max_buffer_size (gpu_canvas::use_adapter_limits);
the benches use the same limits.
The CPU fallback (documents beyond the GPU limit, or features it lacks) is
GpuCanvas::refresh: it renders and uploads the composite band by band
instead of a full float composite plus a full RGBA8 copy. The CPU canvas
texture is reduced in bands too, and the Navigator on the GPU canvas uses a
cached thumbnail instead of a full-resolution CPU composite per revision.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* IO: banded export and import; no full-size copies in encoders; larger decode limits
- Flat export renders the composite in bands and quantises each band; alpha
is dropped in place when every pixel is opaque. The single-layer native
path copies surface rows a band at a time (no f32 read_region and no
per-pixel Vec). Buffers come from try_reserve: Err, not an abort.
- PSD export encodes the merged image band by band into its planes (no
flatten, no second plane copy); near-opaque documents are re-encoded
without the white matte so the output is unchanged. Layer interleaved
bytes are freed before their planes are compressed.
- Flat import converts in ~32 MB bands (and not at all when already RGBA);
the PSD merged-image un-matte works per tile row.
- Encoders borrow the image when no conversion is needed (Image::converted)
instead of cloning it; PNG only copies 16-bit data for the byte swap.
- Decode limits: 2^30 pixels and 8 GiB on 64-bit targets (a 20000² JPEG
was refused at 2^28).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Filters: write tile results in memory-bounded groups
apply_tiled collected every tile's float result before writing any (3.1 GB
for a 14000² layer). Results are now produced and written in groups of about
256 MB (at least one tile per core).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* History: bound pixel memory by Preferences > Performance > Memory Usage
History::max_bytes caps the current document's tiles plus the tiles only
history holds (layers, masks, alpha channels); History::trim drops the
oldest states beyond it after every recorded step and when preferences
change, always keeping the last step undoable. The engine sets it from
memory_usage_mb, which was unused.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Large-image benchmark (14k / 20k): open, refresh, thumbnails, filter, stroke, saves
One operation per run on a cached synthetic photo, with the GPU device created
the way the app creates it, so /usr/bin/time -l reports that operation's peak
memory. Documented in docs/development.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Rene Leonhardt