Release 0.13.1

The Malvar "R at green in R row" kernel weights the two greens two
sites away along the row at -1 and the pair up and down the column at
+1/2. The shader had the two swapped, in the comment as well as the
code, so the transcription checked against itself. Both sum to zero
and reconstruct a flat patch exactly, which is all the tests fed it.

On an edge the correction at green sites is half strength and the
false colour doubles: 0.375 against 0.19 on a grey step, and a
blue/yellow zipper around every clipped highlight at 1:1. The other
three kernels and the CFA tables were right.

A grey vertical step now runs through the pass; the transposed kernel
fails it at 0.375.

Cargo.lock

@@ -1221,7 +1221,7 @@ checksum = "f27ae1dd37df86211c42e150270f82743308803d90a6f6e6651cd730d5e1732f"
 
 [[package]]
 name = "darkroom-android"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "android_logger",
  "dr-plat",
@@ -1234,7 +1234,7 @@ dependencies = [
 
 [[package]]
 name = "darkroom-desktop"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "anyhow",
  "dr-plat",
@@ -1408,7 +1408,7 @@ checksum = "d8b14ccef22fc6f5a8f4d7d768562a182c04ce9a3b3157b91390b52ddfdf1a76"
 
 [[package]]
 name = "dr-bench"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "anyhow",
  "dr-catalog",
@@ -1425,7 +1425,7 @@ dependencies = [
 
 [[package]]
 name = "dr-catalog"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "dr-face",
  "dr-plat",
@@ -1440,7 +1440,7 @@ dependencies = [
 
 [[package]]
 name = "dr-decode"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "dr-types",
  "env_logger",
@@ -1454,7 +1454,7 @@ dependencies = [
 
 [[package]]
 name = "dr-export"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "dr-decode",
  "dr-gpu",
@@ -1473,7 +1473,7 @@ dependencies = [
 
 [[package]]
 name = "dr-face"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "dr-inference-engine",
  "env_logger",
@@ -1486,7 +1486,7 @@ dependencies = [
 
 [[package]]
 name = "dr-film"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "log",
  "serde",
@@ -1495,7 +1495,7 @@ dependencies = [
 
 [[package]]
 name = "dr-gpu"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "bytemuck",
  "dr-decode",
@@ -1513,7 +1513,7 @@ dependencies = [
 
 [[package]]
 name = "dr-inference-engine"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "libloading",
  "log",
@@ -1527,7 +1527,7 @@ dependencies = [
 
 [[package]]
 name = "dr-ingest"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "dr-plat",
  "dr-types",
@@ -1539,7 +1539,7 @@ dependencies = [
 
 [[package]]
 name = "dr-lens"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "lensfun",
  "log",
@@ -1547,7 +1547,7 @@ dependencies = [
 
 [[package]]
 name = "dr-pano"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "dr-decode",
  "dr-inference-engine",
@@ -1561,7 +1561,7 @@ dependencies = [
 
 [[package]]
 name = "dr-pipeline"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "dr-types",
  "log",
@@ -1570,7 +1570,7 @@ dependencies = [
 
 [[package]]
 name = "dr-plat"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "android-native-keyring-store",
  "dr-types",
@@ -1586,7 +1586,7 @@ dependencies = [
 
 [[package]]
 name = "dr-preset-xmp"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "dr-pipeline",
  "log",
@@ -1596,7 +1596,7 @@ dependencies = [
 
 [[package]]
 name = "dr-segment"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "dr-inference-engine",
  "env_logger",
@@ -1609,7 +1609,7 @@ dependencies = [
 
 [[package]]
 name = "dr-sync"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "async-trait",
  "dr-plat",
@@ -1623,7 +1623,7 @@ dependencies = [
 
 [[package]]
 name = "dr-sync-folder"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "async-trait",
  "dr-sync",
@@ -1635,7 +1635,7 @@ dependencies = [
 
 [[package]]
 name = "dr-sync-nextcloud"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "async-trait",
  "dr-decode",
@@ -1657,7 +1657,7 @@ dependencies = [
 
 [[package]]
 name = "dr-thumbs"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "dr-types",
  "jpeg-encoder",
@@ -1669,7 +1669,7 @@ dependencies = [
 
 [[package]]
 name = "dr-types"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "serde",
  "serde_json",
@@ -1678,7 +1678,7 @@ dependencies = [
 
 [[package]]
 name = "dr-ui"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "anyhow",
  "async-trait",
@@ -1720,7 +1720,7 @@ dependencies = [
 
 [[package]]
 name = "dr-xmp"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "dr-types",
  "log",
@@ -7021,7 +7021,7 @@ checksum = "8df9b6e13f2d32c91b9bd719c00d1958837bc7dec474d94952798cc8e69eeec3"
 
 [[package]]
 name = "traceability"
-version = "0.13.0"
+version = "0.13.1"
 dependencies = [
  "anyhow",
  "serde",

Cargo.toml

@@ -29,7 +29,7 @@ members = [
 ]
 
 [workspace.package]
-version = "0.13.0"
+version = "0.13.1"
 edition = "2021"
 rust-version = "1.92"
 license = "GPL-3.0-or-later"

apps/darkroom-android/src/lib.rs

@@ -333,7 +333,7 @@ fn unpack_bundled_models(app: &slint::android::AndroidApp) {
     // The int8 forms beside the three detectors are what the Hexagon runs
     // (docs/inference.md §5); the engine loads the sibling when the probe
     // chose that rung and ignores it otherwise.
-    const BUNDLED: [(&std::ffi::CStr, &str); 13] = [
+    const BUNDLED: [(&std::ffi::CStr, &str); 14] = [
         (c"models/scrfd_500m_640.onnx", "scrfd_500m_640.onnx"),
         (
             c"models/scrfd_500m_640.int8.onnx",

apps/darkroom-desktop/src/main.rs

@@ -83,10 +83,14 @@ fn main() -> anyhow::Result<()> {
 /// `DARKROOM_ORT_DIR` is for a developer pointing at a runtime that is not
 /// installed — the wheel's `capi` directory, say. Then beside the executable
 /// and in the package's private library directory, for a package that
-/// bundles its own; then the Flatpak prefix; then the system library
+/// bundles its own; then the user's own `runtime/` beside the models, where
-/// directory, for a distribution that ships ONNX Runtime as a package of its
+/// `tools/fetch-desktop-runtime.sh` puts one; then the Flatpak prefix; then
-/// own. A system copy whose GPU providers do not load is not a problem: the
+/// the system library directory, for a distribution that ships ONNX Runtime
-/// probe builds a real session before believing a provider.
+/// as a package of its own. The user's copy outranks the system's because
+/// the system's is the one most likely to be built without the GPU
+/// providers, or against the wrong cuDNN — and a system copy whose providers
+/// do not load is not a problem, only a slower app: the probe builds a real
+/// session before believing a provider.
 fn runtime_dirs() -> Vec<PathBuf> {
     let mut dirs = Vec::new();
     if let Some(dir) = std::env::var_os("DARKROOM_ORT_DIR") {
@@ -98,6 +102,7 @@ fn runtime_dirs() -> Vec<PathBuf> {
             dirs.push(bin.join("../lib/darkroom"));
         }
     }
+    dirs.push(dr_ui::inference::user_runtime_dir());
     #[cfg(target_os = "linux")]
     dirs.extend([
         PathBuf::from("/app/lib/darkroom"),

core/dr-gpu/src/demosaic.rs

@@ -1347,6 +1347,53 @@ mod tests {
         }
     }
 
+    #[test]
+    fn a_grey_step_edge_stays_grey() {
+        // A flat patch cannot tell the Malvar kernels from any other set of
+        // weights that sum to zero. An edge can. A grey vertical step, so
+        // every photosite records the same profile, must come back with the
+        // three channels close together on both sides; any spread is false
+        // colour from interpolating across the edge.
+        //
+        // The bound is set by the paper's kernels, which peak at 0.19 here.
+        // With the ±2 terms of the green-site kernels transposed — the bug
+        // this test was written against — the peak is 0.375.
+        let Some(ctx) = ctx() else { return };
+        let d = Demosaicer::new(&ctx).expect("demosaicer");
+
+        let size = 32u32;
+        let white = 16383u16;
+        let mut raw = flat_cfa(CfaPattern::Rggb, size, [0, 0, 0], 0, white);
+        for y in 0..size {
+            for x in size / 2..size {
+                raw.data[(y * size + x) as usize] = white;
+            }
+        }
+
+        let img = d.run(&raw).expect("demosaic");
+        let px = read_rgba(&ctx, &img);
+        let (w, _) = img.size();
+
+        let mut worst = (0.0f32, 0u32, 0u32);
+        for y in 2..size - 2 {
+            for x in 2..size - 2 {
+                let p = px[(y * w + x) as usize];
+                let spread = (p[0] - p[1]).abs().max((p[2] - p[1]).abs());
+                if spread > worst.0 {
+                    worst = (spread, x, y);
+                }
+            }
+        }
+        assert!(
+            worst.0 < 0.25,
+            "false colour of {} at ({}, {}) on a grey edge — the green-site \
+             kernels are interpolating across the edge",
+            worst.0,
+            worst.1,
+            worst.2
+        );
+    }
+
     #[test]
     fn output_is_free_of_nan_and_negatives() {
         // f16 NaN propagates silently through every later stage; a negative

core/dr-gpu/src/shaders/demosaic.wgsl

@@ -160,12 +160,18 @@ fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
         // Green is measured. Red and blue are interpolated from their own
         // axis, with a correction from the green Laplacian.
         //
-        // Malvar "G at R/B locations" kernels, transposed per axis:
+        // Malvar "R at green in R row" kernel, and its transpose:
-        //   chroma along the row:  (5c + 4(w1+e1) - (nw+ne+sw+se) - (n2+s2) + 0.5(w2+e2)) / 8
+        //   chroma along the row:  (5c + 4(w1+e1) - (nw+ne+sw+se) - (w2+e2) + 0.5(n2+s2)) / 8
+        //
+        // The -1 goes on the two greens *along* the chroma axis and the +0.5
+        // on the pair across it. Transposed, both kernels still sum to zero
+        // and reconstruct a flat patch exactly, but on an edge the correction
+        // at green sites is half strength and the false colour doubles: a
+        // blue/yellow zipper around every clipped highlight.
         let along_row =
-            (5.0 * c + 4.0 * (w1 + e1) - diag1 - vert2 + 0.5 * horiz2) * 0.125;
+            (5.0 * c + 4.0 * (w1 + e1) - diag1 - horiz2 + 0.5 * vert2) * 0.125;
         let along_col =
-            (5.0 * c + 4.0 * (n1 + s1) - diag1 - horiz2 + 0.5 * vert2) * 0.125;
+            (5.0 * c + 4.0 * (n1 + s1) - diag1 - vert2 + 0.5 * horiz2) * 0.125;
 
         let red_horizontal = red_is_horizontal(gid.x, gid.y);
         let r = select(along_col, along_row, red_horizontal);

core/dr-inference-engine/src/lib.rs

@@ -155,6 +155,8 @@ pub struct Status {
     /// Engines compiled and engines wanted, for a compiling rung; `(0, 0)`
     /// otherwise.
     pub engines: (usize, usize),
+    /// Every rung above the selected one that was tried, and why it lost.
+    pub failed: Vec<(Rung, String)>,
 }
 
 impl Status {
@@ -399,6 +401,7 @@ pub fn status() -> Status {
         runtime: api::runtime(),
         rung,
         reason: s.cache.reason.clone(),
+        failed: s.cache.failed.clone(),
         probing: s.probing,
         engines: if rung.compiles() {
             (s.cache.compiled.len(), s.wanted)

packaging/PKGBUILD

@@ -4,7 +4,7 @@
 # makes `makepkg -si` in this directory install what you are actually working
 # on. Swap `source` for a tagged tarball when there is something to release.
 pkgname=darkroom
-pkgver=0.13.0
+pkgver=0.13.1
 # Back to 1 with the version: a new pkgver is a new archive name, so there is
 # nothing for makepkg to reuse and nothing for a release number to disambiguate.
 pkgrel=1

tools/fetch-desktop-runtime.sh

@@ -0,0 +1,32 @@
+#!/usr/bin/env bash
+# Put a GPU-capable ONNX Runtime where the desktop app looks for one
+# (docs/inference.md §3): `runtime/` beside the models in the user data
+# directory, ahead of the system library.
+#
+#   ./tools/fetch-desktop-runtime.sh [DEST]
+#
+# The source is the `onnxruntime-gpu` wheel: the one build that carries the
+# CUDA *and* TensorRT providers against the cuDNN and TensorRT majors current
+# on this machine. Distribution packages tend to have neither — Arch's
+# `onnxruntime-opt-cuda` is built without TensorRT and against cuDNN 8 — and
+# the probe rejects them correctly and leaves the app on the CPU provider,
+# which is what this script exists to fix. Nothing NVIDIA is bundled here:
+# the providers load CUDA, cuDNN and TensorRT from the system, and if those
+# are missing the probe says so and the app stays on the CPU.
+set -euo pipefail
+DEST="${1:-${XDG_DATA_HOME:-${HOME}/.local/share}/darkroom/runtime}"
+WORK="$(mktemp -d -p /var/tmp fetch-desktop-runtime.XXXXXX)"
+trap 'rm -rf "${WORK}"' EXIT
+
+echo "==> downloading the onnxruntime-gpu wheel"
+uv venv --python 3.12 "${WORK}/venv" >/dev/null
+VIRTUAL_ENV="${WORK}/venv" uv pip install --quiet onnxruntime-gpu
+CAPI="$(find "${WORK}/venv" -type d -path '*/onnxruntime/capi' | head -1)"
+[[ -n "${CAPI}" ]] || { echo "error: no capi directory in the wheel" >&2; exit 1; }
+
+mkdir -p "${DEST}"
+# The runtime and its provider libraries; not the Python binding.
+cp "${CAPI}"/libonnxruntime.so* "${CAPI}"/libonnxruntime_providers_*.so "${DEST}/"
+echo "==> runtime in ${DEST}:"
+ls -1 "${DEST}" | sed 's/^/    /'
+echo "    (the app finds it on its next launch; Settings › About › Inference says what it chose)"

ui/dr-ui/src/inference.rs

@@ -60,6 +60,18 @@ pub fn init(runtime_dirs: Vec<PathBuf>) {
     crate::memory::evict_at(crate::memory::Tier::Gpu, dr_inference_engine::release_all);
 }
 
+/// Where a person can put a runtime by hand: `runtime/` beside the models,
+/// searched before any system library. The system copy on the reference
+/// desktop is built without TensorRT and against the wrong cuDNN, and a
+/// working one is four files from the `onnxruntime-gpu` wheel; this is
+/// where they go, and `tools/fetch-desktop-runtime.sh` puts them there.
+pub fn user_runtime_dir() -> PathBuf {
+    crate::library::shared_face_models_dir()
+        .parent()
+        .map(|p| p.join("runtime"))
+        .unwrap_or_else(|| PathBuf::from("runtime"))
+}
+
 /// Which form the current backend loads `detector` in, given the files on
 /// this device — the fact `faces.model_id` has to carry (§7).
 ///
@@ -94,8 +106,18 @@ pub fn about_lines() -> (String, String) {
             status.engines.0,
             status.engines.1
         )
-    } else {
+    } else if status.failed.is_empty() {
         status.reason
+    } else {
+        // Every rung that was tried and why it lost, not only the first:
+        // "TensorRT: not enabled in this build" says nothing about why CUDA
+        // was not taken instead.
+        status
+            .failed
+            .iter()
+            .map(|(rung, why)| format!("{}: {why}", rung.label()))
+            .collect::<Vec<_>>()
+            .join(" · ")
     };
     (line, detail)
 }