fix(ai-engine-burn): run the processor fallback on every core with SIMD (#113)

The crate turns off Burn's default features, which also dropped `rayon` and
`simd`: burn-ndarray was built with `std` only, so the processor fallback ran on
one core with scalar maths. Turning both on gives burn-ndarray `multi-threads`
and `simd`. Measured on the InSPyReNet network (640 x 333 picture, processor,
i5-12600KF, release): 22.8 s -> 6.2 s, masks identical to one pixel by one
level. SAM 2.1's processor timing was not re-measured; the doc says so.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

Cargo.lock

@@ -1103,6 +1103,7 @@ version = "0.22.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7e05d08ef48d738fadf13f8689b634f8b396d5a8cbb27d0a87f61e163d241846"
 dependencies = [
+ "aligned-vec",
  "burn-backend",
  "burn-ir",
  "burn-std",
@@ -1110,9 +1111,11 @@ dependencies = [
  "gemm 0.19.0",
  "half",
  "libm",
+ "macerator",
  "num-traits",
  "once_cell",
  "portable-atomic",
+ "rayon",
 ]
 
 [[package]]
@@ -1167,7 +1170,9 @@ dependencies = [
  "burn-backend",
  "burn-ir",
  "burn-std",
+ "bytemuck",
  "const-random",
+ "itertools 0.15.0",
  "libm",
  "macerator",
  "matrixmultiply",
@@ -1175,6 +1180,8 @@ dependencies = [
  "num-traits",
  "paste",
  "rand 0.10.3",
+ "rayon",
+ "seq-macro",
 ]
 
 [[package]]
@@ -5445,7 +5452,10 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3f607c237553f086e7043417a51df26b2eb899d3caff94e6a67592ff992fedc7"
 dependencies = [
  "autocfg",
+ "num_cpus",
+ "once_cell",
  "rawpointer",
+ "thread-tree",
 ]
 
 [[package]]
@@ -5649,6 +5659,7 @@ dependencies = [
  "portable-atomic",
  "portable-atomic-util",
  "rawpointer",
+ "rayon",
 ]
 
 [[package]]
@@ -8388,6 +8399,15 @@ dependencies = [
  "syn 3.0.7",
 ]
 
+[[package]]
+name = "thread-tree"
+version = "0.3.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ffbd370cb847953a25954d9f63e14824a36113f8c72eecf6eccef5dc4b45d630"
+dependencies = [
+ "crossbeam-channel",
+]
+
 [[package]]
 name = "thread_local"
 version = "1.1.10"

crates/ai-engine-burn/Cargo.toml

@@ -22,8 +22,10 @@ craft-ai-features = { path = "../ai-features", package = "craft-ai-features" }
 # The request handling (clicks, box, the picture kept between requests, the empty-mask answer),
 # picture decoding and resampling are engine-neutral and live there once.
 craft-ai-engine-rten = { path = "../ai-engine-rten", package = "craft-ai-engine-rten" }
-# `wgpu` reaches Vulkan, Metal and DirectX 12 cards; `ndarray` is the processor fallback.
-burn = { version = "=0.22.1", default-features = false, features = ["std", "ndarray", "wgpu"] }
+# `wgpu` reaches Vulkan, Metal and DirectX 12 cards; `ndarray` is the processor fallback. Without
+# `rayon` and `simd` (off with the default features) the processor fallback runs on one core with
+# scalar maths.
+burn = { version = "=0.22.1", default-features = false, features = ["std", "ndarray", "wgpu", "rayon", "simd"] }
 burn-store = "=0.22.1"
 craft-ai-burn-setup = { path = "../ai-burn-setup", package = "craft-ai-burn-setup", features = ["conversion"], optional = true }
 

docs/BURN-ENGINE.md

@@ -79,7 +79,10 @@ python tools/mutation/mutation_check.py be
 - SAM 2.1 tiny only. A different architecture needs its own generated code and runner.
 - Click and box prompts. A description stays on the ComfyUI route.
 - The processor fallback uses Burn's `ndarray` backend, which Burn has marked deprecated in favour
-  of `burn-flex`. Swap it once flex is measured.
+  of `burn-flex`. Swap it once flex is measured. It is built with Burn's `rayon` and `simd`
+  features: the crate turns Burn's default features off, and without these two the processor runs
+  on one core with scalar maths (3.7 times slower on the one network measured, see the pull request
+  that turned them on). The SAM 2.1 timings below were taken before that change.
 - `CRAFT_AI_BURN_DEVICE=cpu` forces the processor even when a graphics card answers. Run on the
   test PC (release, astronaut sample): one click and one box both gave IoU 1.0000 against rten
   (22.5 % selected by both). Timings on the processor: first run 8.70 s (rten 0.71 s), second