@graphty/webgpu-graph-algorithms 0.6.3 → 0.6.4

package/README.md

@@ -3,11 +3,11 @@
 WebGPU-accelerated graph algorithms and layouts over the `@graphty/graph-format` snapshot, for Node
 (Dawn, through the `webgpu` npm package) and browsers (Chromium). One code base, three entry points:
 
-| Entry                                      | Import        | What it gives you                                                                                                                                                                                                                                                                                                                                                                                                                                                                         |
-| ------------------------------------------ | ------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| Entry                                      | Import        | What it gives you                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                            |
+| ------------------------------------------ | ------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
 | `@graphty/webgpu-graph-algorithms`         | the core      | the layouts (`createForceAtlas2`, `createFruchtermanReingold`, `createSpringElectrical`, `seedPositions`), the algorithms (`pageRank`, `personalizedPageRank`, `hits`, `eigenvectorCentrality`, `katzCentrality`, `connectedComponents`, `degree`), `createAccelerator`, `calibrateLayout`, `verifyDevice`, `GpuContext`, `WebGpuGraphError`, `isSoftwareAdapter`, the constants (`EXACT_MAX_NODES`, `FA2_DEFAULTS`, `FR_DEFAULTS`, `SE_DEFAULTS`, `LAYOUT_TUNING_DEFAULTS`, ...) and the option / stats / accelerator types |
-| `@graphty/webgpu-graph-algorithms/node`    | Node only     | `createNodeGpuContext`, `probeNodeWebGpu`, `createNodeGpu` (Dawn), `dawnFlags`                                                                                                                                                                                                                                                                                                                                                                                                            |
-| `@graphty/webgpu-graph-algorithms/browser` | browsers only | `probeBrowserWebGpu`, `requestGpuContext`                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
+| `@graphty/webgpu-graph-algorithms/node`    | Node only     | `createNodeGpuContext`, `probeNodeWebGpu`, `createNodeGpu` (Dawn), `dawnFlags`                                                                                                                                                                                                                                                                                                                                                                                                                                               |
+| `@graphty/webgpu-graph-algorithms/browser` | browsers only | `probeBrowserWebGpu`, `requestGpuContext`                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
 
 **Status: three force layouts and six algorithms, on the exact and the grid repulsion tiers.**
 ForceAtlas2, Fruchterman-Reingold and ngraph's spring-electrical preset run from Node (`run()`) and from a
@@ -650,7 +650,8 @@ pnpm run bench                                                 # every group; ap
 pnpm exec tsx benchmarks/run.ts upload roundtrip layout-exact  # selected groups; --no-save, --runs N, --allow-software
 pnpm exec tsx benchmarks/layout-run.ts --nodes 100000 --edges 1000000   # the end-to-end layout driver (exit 1 on a bad result)
 pnpm run gpu:report > gpu-report.json                          # the adapter report with a 10 s nvidia-smi sample
-pnpm run bench:compare                                         # the last out session vs benchmarks/results/<runner-class>.json (> 3x fails)
+pnpm run bench:compare                                         # the last out session vs benchmarks/results/<runner-class>.json (1.35x and 2.5 ms over the pinned best fails)
+pnpm run bench:append benchmarks/out/<class>.json benchmarks/results/<class>.json   # append the last out session to a baseline (refuses software / incomplete / duplicate sessions)
 ```
 
 The runner class is `<vendor>-<architecture>-driver<major>` (`scripts/runner-class.js`; `GRAPHTY_RUNNER_CLASS` overrides it,
@@ -667,7 +668,12 @@ SM clock at its idle 210 MHz under sparse sub-millisecond dispatches and the ker
 at 10k and at 100k with `repulsion: "exact"`, the same two rows per model and rung as `layout-exact`, tagged `fr` /
 `se`), `layout-grid` (T-6 and T-7: `step(1)` of the grid tier on the grid ladder 32k / 65k / 100k / 262k / 1M in 2D
 and in 3D, the same two rows per rung tagged `grid` with the dimension, plus the `fa2-attraction` pass of the 1M 2D
-iteration from the profiler, and the exact ladder's 1k / 4k / 8k / 16k rungs in 2D for the crossover re-check). The Chromium numbers of T-5 (10k on the exact tier, 100k on the grid tier) come from the
+iteration from the profiler, and the exact ladder's 1k / 4k / 8k / 16k rungs in 2D for the crossover re-check),
+`attraction-scale` (no target: the `fa2-attraction` pass across a working-set ladder, the G4-F16 diagnostic) and `bfs`
+(T-10: `breadthFirstSearch` direction-optimizing and top-down from node 0 of the undirected RMAT tiers 100k / 1M and
+1M / 10M, `sssp` on the same tiers with random weights in [0.1, 10), and `breadthFirstSearch` from a corner of the
+1000 x 1000 grid, all wall end to end including the upload). A baseline session must carry every group:
+`pnpm run bench:append` refuses one that does not. The Chromium numbers of T-5 (10k on the exact tier, 100k on the grid tier) come from the
 `bench`-tagged browser test (`GRAPHTY_BROWSER_GPU=nvidia node scripts/run-browser-project.js`), which appends its
 session through the Vitest commands bridge. `exactMaxNodes` is re-fixed from the ladder by the rule of plan section 7.8
 (the largest rung under 4 ms per iteration and not slower than the grid tier at the same n -- the `layout-grid` 2D rows,
@@ -686,19 +692,22 @@ are the T-table of plan section 10.4. A missed target is re-fixed by a recorded
 
 Measured on nvidia-lovelace-driver580 (NVIDIA: 580.173.02 580.173.2.0), session 2026-09-20T01:06:48.656Z, medians of 5 runs; Chromium: nvidia / lovelace (nvidia-lovelace-driver0, the description is redacted by Chromium), session 2026-09-16T02:18:11.896Z. The T-5 (100k), T-6 and T-7 rows are from the later session 2026-09-21T06:15:17.027Z (the file's last session, the current baseline, the one the crossover re-check reads; its Chromium session 2026-09-21T05:48:14.190Z), whose other rows are within 1.25x of this table's.
 
-| Id   | What                                                                                                                                     | Target                        | Measured                     |
-| ---- | ---------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------- | ---------------------------- |
-| T-1  | Upload of the 100k / 1M weighted hot prefix (16.4 MB); 1M / 10M (164 MB)                                                                 | <= 10 ms; <= 100 ms           | 5.980 ms; 125.738 ms         |
-| T-2  | `degree` + 400 KB readback at 100k (core resident), Node                                                                                 | <= 2 ms                       | 0.870 ms                     |
-| T-3  | Empty submit + 4-byte `readU32` round trip, Dawn                                                                                         | <= 0.1 ms                     | 0.181 ms                     |
-| T-4  | ForceAtlas2 exact tier, GPU time per iteration (profiler) at 10k; at 16k                                                                 | <= 1 ms; <= 2 ms              | 0.586 ms; 1.053 ms           |
-| T-5  | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 10k, Chromium (Node in brackets)                                             | <= 6 ms                       | 2.400 ms (0.726 ms)          |
-| T-5  | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 100k on the grid tier, Chromium (Node in brackets)                           | <= 12 ms                      | 3.700 ms (2.087 ms)          |
-| T-6  | ForceAtlas2 grid tier, GPU time per iteration (profiler) at 100k 2D; at 1M 2D; at 100k 3D                                                | <= 10 ms; <= 100 ms; <= 20 ms | 0.634 ms; 5.414 ms; 1.306 ms |
-| T-7  | Attraction gather (the `fa2-attraction` pass of the grid tier), GPU time per iteration (profiler) at 1M / 10M                            | <= 15 ms                      | 1.493 ms                     |
-| T-8  | PageRank, 100 iterations, wall end to end including the upload, at 100k / 1M; at 1M / 10M                                                | <= 150 ms; <= 1.5 s           | 17.204 ms; 198.645 ms        |
-| T-9  | Weakly connected components (Afforest), wall end to end including the upload and the label readback, at 1M / 10M (100k / 1M in brackets) | <= 100 ms                     | 145.970 ms (12.613 ms)       |
-| T-14 | Fruchterman-Reingold exact tier, GPU time per iteration (profiler) at 10k; at 100k (`repulsion: "exact"`)                                | recorded                      | 0.617 ms; 16.367 ms          |
+| Id   | What                                                                                                                                                                                                                            | Target                        | Measured                                                                                                                                                                                                                 |
+| ---- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| T-1  | Upload of the 100k / 1M weighted hot prefix (16.4 MB); 1M / 10M (164 MB)                                                                                                                                                        | <= 10 ms; <= 100 ms           | 5.980 ms; 125.738 ms                                                                                                                                                                                                     |
+| T-2  | `degree` + 400 KB readback at 100k (core resident), Node                                                                                                                                                                        | <= 2 ms                       | 0.870 ms                                                                                                                                                                                                                 |
+| T-3  | Empty submit + 4-byte `readU32` round trip, Dawn                                                                                                                                                                                | <= 0.1 ms                     | 0.181 ms                                                                                                                                                                                                                 |
+| T-4  | ForceAtlas2 exact tier, GPU time per iteration (profiler) at 10k; at 16k                                                                                                                                                        | <= 1 ms; <= 2 ms              | 0.586 ms; 1.053 ms                                                                                                                                                                                                       |
+| T-5  | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 10k, Chromium (Node in brackets)                                                                                                                                    | <= 6 ms                       | 2.400 ms (0.726 ms)                                                                                                                                                                                                      |
+| T-5  | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 100k on the grid tier, Chromium (Node in brackets)                                                                                                                  | <= 12 ms                      | 3.700 ms (2.087 ms)                                                                                                                                                                                                      |
+| T-6  | ForceAtlas2 grid tier, GPU time per iteration (profiler) at 100k 2D; at 1M 2D; at 100k 3D                                                                                                                                       | <= 10 ms; <= 100 ms; <= 20 ms | 0.634 ms; 5.414 ms; 1.306 ms                                                                                                                                                                                             |
+| T-7  | Attraction gather (the `fa2-attraction` pass of the grid tier), GPU time per iteration (profiler) at 1M / 10M                                                                                                                   | <= 15 ms                      | 1.493 ms                                                                                                                                                                                                                 |
+| T-8  | PageRank, 100 iterations, wall end to end including the upload, at 100k / 1M; at 1M / 10M                                                                                                                                       | <= 150 ms; <= 1.5 s           | 17.204 ms; 198.645 ms                                                                                                                                                                                                    |
+| T-9  | Weakly connected components (Afforest), wall end to end including the upload and the label readback, at 1M / 10M (100k / 1M in brackets)                                                                                        | <= 100 ms                     | 145.970 ms (12.613 ms)                                                                                                                                                                                                   |
+| T-14 | Fruchterman-Reingold exact tier, GPU time per iteration (profiler) at 10k; at 100k (`repulsion: "exact"`)                                                                                                                       | recorded                      | 0.617 ms; 16.367 ms                                                                                                                                                                                                      |
+| T-10 | BFS direction-optimizing (`breadthFirstSearch`), wall end to end including the upload, from node 0 of the undirected RMAT at 1M / 10M (577,681 nodes reached, 6 levels; 100k / 1M in brackets; top-down 233.064 ms / 97.259 ms) | <= 100 ms                     | 237.388 ms (101.183 ms) in the baseline session; 156.560 ms (16.731 ms) re-measured 2026-09-25 under load after the direct-dispatch change, top-down 88.606 ms -- see below and G8-F5 / G8-F21 of `docs/decisions/G8.md` |
+| T-10 | BFS on the 1000 x 1000 grid from a corner (1,999 levels; 64 `mapAsync` calls = ceil(levels / 32) + 1), wall including the upload                                                                                                | <= 1.5 s                      | 5680.192 ms in the baseline session; 427.235 ms re-measured 2026-09-25 under load after the direct-dispatch change, MET -- see below and G8-F5 of `docs/decisions/G8.md`                                                 |
+| T-10 | `sssp` (the near-far queue), wall end to end including the upload, random f32 weights in [0.1, 10), at 1M / 10M (100k / 1M in brackets)                                                                                         | recorded                      | 310.351 ms (118.064 ms); 271.040 ms (50.871 ms) re-measured 2026-09-25 under load                                                                                                                                        |
 
 Three rows miss their target in this session: the 1M / 10M upload (125.7 ms against 100 ms, the open owner decision of
 `docs/decisions/G1.md` section 7), the empty-submit round trip (0.181 ms against 0.1 ms: the row is measured after the
@@ -712,6 +721,26 @@ four iterations, which would time one pull and call it a hundred. The T-14 row i
 session 2026-09-20T19:25:37.311Z (the file's last session, the current baseline), whose other rows are within 1.08x of
 this table's; the spring-electrical preset measures 0.648 ms at 10k and 18.154 ms at 100k in the same session.
 
+The three T-10 rows are the `bfs` group of session 2026-09-25T10:19:11.536Z (the file's last session, the current
+baseline; `webgpu` 0.4.0, driver 580.173.02, load average 1.7-2.1, medians of 5). Both T-10 targets are MISSED on the
+reference card, by 2.4x and 3.8x, and the cost is not in the kernels: a traversal costs about 2.85 ms per LEVEL
+whatever the level's size or the submit cadence (the 200 x 200 grid, 399 levels, at 32, 8 and 1 levels per submit
+alike) plus about 80 ms per CALL (a karate BFS, 34 nodes and 3 levels, takes 90 ms), and with Dawn's `skip_validation`
+toggle the same karate call takes 3.0 ms and the 399-level grid 27 ms. About 97 % of the wall time is Dawn's own
+validation of `dispatchWorkgroupsIndirect`, which the frontier design pays nine times a level; the baseline records
+the default runtime because that is what a consumer gets, and the toggle is unsafe. The decision -- re-fix T-10 to
+the class, fold the per-level indirect dispatches, or dispatch directly where the host already knows a count -- is
+recorded in `docs/decisions/G8.md`.
+
+That decision was taken on 2026-09-25, before the merge: every level kernel is now a DIRECT grid-stride dispatch
+gated by a path word the device-side selector writes, and no traversal driver dispatches indirectly
+(`design/decisions/2026-09-25-frontier-kernels-dispatch-directly.md`; the measurement of Dawn's per-indirect-dispatch
+cost, about 0.4 ms each in Node and Chromium alike, is `design/webgpu/dawn-indirect-dispatch-validation-cost.md`).
+Re-measured on the same card with `--no-save --runs 3` at load average about 20 (so an upper bound, and not appended
+to the baseline): a 34-node BFS 7.2 ms (from 89.3 with the core resident), the grid row 427 ms (MET), the 1M / 10M
+auto row 156.6 ms and the top-down row 88.6 ms. The auto row now misses only because the direction-optimizing
+choice itself is slower than top-down on that graph, which the validation cost had hidden (G8-F21).
+
 The exact curve (the `layout-exact` group: 2D, E = 10n, seeded G(n, m), one simulation per rung; ms / iteration from the profiler):
 
 | n     | ms / iteration | step(1) wall (ms) | pairs / s |
@@ -741,19 +770,20 @@ the grid tier's is 2.8x at 100k 2D and 5.7x at 1M 2D while the attraction pass a
 
 Measured on gpu-linux-t4 (NVIDIA: 580.126.20 580.126.20.0), session 2026-09-20T02:29:33.210Z (run 35483512705), medians of 5 runs; Chromium: nvidia / turing (nvidia-turing-driver0, the description is redacted by Chromium), session 2026-09-20T02:28:10.676Z. The T-5 (100k), T-6 and T-7 rows are from the later session 2026-09-22T20:21:53.814Z (GPU lane run 35775999450 on commit 1d2d4dcf, the file's last session and the current baseline of this class; its Chromium session 2026-09-22T20:19:30.117Z).
 
-| Id   | What                                                                                                                                     | Target                        | Measured                                                                                                         |
-| ---- | ---------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------- | ---------------------------------------------------------------------------------------------------------------- |
-| T-1  | Upload of the 100k / 1M weighted hot prefix (16.4 MB); 1M / 10M (164 MB)                                                                 | <= 10 ms; <= 100 ms           | 14.338 ms; 256.418 ms                                                                                            |
-| T-2  | `degree` + 400 KB readback at 100k (core resident), Node                                                                                 | <= 2 ms                       | 2.957 ms                                                                                                         |
-| T-3  | Empty submit + 4-byte `readU32` round trip, Dawn                                                                                         | <= 0.1 ms                     | 1.296 ms                                                                                                         |
-| T-4  | ForceAtlas2 exact tier, GPU time per iteration (profiler) at 10k; at 16k                                                                 | <= 1 ms; <= 2 ms              | 0.972 ms; 1.953 ms                                                                                               |
-| T-5  | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 10k, Chromium (Node in brackets)                                             | <= 6 ms                       | 2.600 ms (1.359 ms)                                                                                              |
-| T-5  | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 100k on the grid tier, Chromium (Node in brackets)                           | <= 12 ms                      | 8.100 ms (5.314 ms)                                                                                              |
-| T-6  | ForceAtlas2 grid tier, GPU time per iteration (profiler) at 100k 2D; at 1M 2D; at 100k 3D                                                | <= 10 ms; <= 100 ms; <= 20 ms | 1.790 ms; 31.057 ms; 3.954 ms                                                                                    |
-| T-7  | Attraction gather (the `fa2-attraction` pass of the grid tier), GPU time per iteration (profiler) at 1M / 10M                            | <= 15 ms                      | 18.165 ms -- MISSED: 21 % over the target on this card (1.493 ms on the dev box; G4-F16 of docs/decisions/G4.md) |
-| T-8  | PageRank, 100 iterations, wall end to end including the upload, at 100k / 1M; at 1M / 10M                                                | <= 150 ms; <= 1.5 s           | 45.461 ms; 1092.799 ms                                                                                           |
-| T-9  | Weakly connected components (Afforest), wall end to end including the upload and the label readback, at 1M / 10M (100k / 1M in brackets) | <= 100 ms                     | 293.079 ms (28.544 ms)                                                                                           |
-| T-14 | Fruchterman-Reingold exact tier, GPU time per iteration (profiler) at 10k; at 100k (`repulsion: "exact"`)                                | recorded                      | 0.942 ms; 54.232 ms (the spring preset 1.051 ms; 60.706 ms)                                                      |
+| Id   | What                                                                                                                                     | Target                                   | Measured                                                                                                                                                                               |
+| ---- | ---------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| T-1  | Upload of the 100k / 1M weighted hot prefix (16.4 MB); 1M / 10M (164 MB)                                                                 | <= 10 ms; <= 100 ms                      | 14.338 ms; 256.418 ms                                                                                                                                                                  |
+| T-2  | `degree` + 400 KB readback at 100k (core resident), Node                                                                                 | <= 2 ms                                  | 2.957 ms                                                                                                                                                                               |
+| T-3  | Empty submit + 4-byte `readU32` round trip, Dawn                                                                                         | <= 0.1 ms                                | 1.296 ms                                                                                                                                                                               |
+| T-4  | ForceAtlas2 exact tier, GPU time per iteration (profiler) at 10k; at 16k                                                                 | <= 1 ms; <= 2 ms                         | 0.972 ms; 1.953 ms                                                                                                                                                                     |
+| T-5  | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 10k, Chromium (Node in brackets)                                             | <= 6 ms                                  | 2.600 ms (1.359 ms)                                                                                                                                                                    |
+| T-5  | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 100k on the grid tier, Chromium (Node in brackets)                           | <= 12 ms                                 | 8.100 ms (5.314 ms)                                                                                                                                                                    |
+| T-6  | ForceAtlas2 grid tier, GPU time per iteration (profiler) at 100k 2D; at 1M 2D; at 100k 3D                                                | <= 10 ms; <= 100 ms; <= 20 ms            | 1.790 ms; 31.057 ms; 3.954 ms                                                                                                                                                          |
+| T-7  | Attraction gather (the `fa2-attraction` pass of the grid tier), GPU time per iteration (profiler) at 1M / 10M                            | <= 15 ms                                 | 18.165 ms -- MISSED: 21 % over the target on this card (1.493 ms on the dev box; G4-F16 of docs/decisions/G4.md)                                                                       |
+| T-8  | PageRank, 100 iterations, wall end to end including the upload, at 100k / 1M; at 1M / 10M                                                | <= 150 ms; <= 1.5 s                      | 45.461 ms; 1092.799 ms                                                                                                                                                                 |
+| T-9  | Weakly connected components (Afforest), wall end to end including the upload and the label readback, at 1M / 10M (100k / 1M in brackets) | <= 100 ms                                | 293.079 ms (28.544 ms)                                                                                                                                                                 |
+| T-14 | Fruchterman-Reingold exact tier, GPU time per iteration (profiler) at 10k; at 100k (`repulsion: "exact"`)                                | recorded                                 | 0.942 ms; 54.232 ms (the spring preset 1.051 ms; 60.706 ms)                                                                                                                            |
+| T-10 | BFS direction-optimizing at 1M / 10M RMAT; the 1000 x 1000 grid; `sssp` at 1M / 10M (the `bfs` group)                                    | recorded (the contract is the dev box's) | not yet recorded: the lane has not run a branch that carries the `bfs` group; the first `gpu.yml` run of it is appended with `pnpm run bench:append` (G8-F3 of `docs/decisions/G8.md`) |
 
 The exact curve (the `layout-exact` group: 2D, E = 10n, seeded G(n, m), one simulation per rung; ms / iteration from the profiler):
 

package/dist/browser.js

@@ -1,4 +1,4 @@
-import { G as GpuContext, W as WebGpuGraphError } from "./chunks/context-BXqgCifx.js";
+import { G as GpuContext, W as WebGpuGraphError } from "./chunks/context-Dvq-Cc6v.js";
 function navigatorGpu() {
   if (typeof navigator === "undefined") {
     return void 0;

package/dist/chunks/{context-BXqgCifx.js → context-Dvq-Cc6v.js}

@@ -88,6 +88,12 @@ const GRID_HUB_CELL = 1024;
 const GRID_EXTENT_FLOOR = 1e-6;
 const GRID_BBOX_MARGIN = 1.01;
 const GRID_SORT_BITS = 24;
+const MAX_LEVELS_PER_SUBMIT = 32;
+const FUSED_FRONTIER_MAX = 4096;
+const FRONTIER_CANDIDATES = 7;
+const BEAMER_BETA = 24;
+const SSSP_DELTA_FACTOR = 32;
+const F32_INF_BITS = 2139095040;
 const PASSTHROUGH_FORMAT_CODES = Object.freeze(["E_GPU_INELIGIBLE", "E_UNKNOWN_NODE", "E_UNKNOWN_COLUMN", "E_COLUMN_LENGTH"]);
 const EMPTY_DETAILS = Object.freeze({});
 class WebGpuGraphError extends Error {
@@ -883,6 +889,7 @@ const PRELUDE_WGSL = (
   `// ---- prelude: constants, standard overrides, helpers (every module receives this text first)
 const INVALID_INDEX: u32 = ${INVALID_INDEX}u;
 const U32_MAX: u32 = ${U32_MAX}u;
+const F32_INF_BITS: u32 = ${F32_INF_BITS}u;
 const MAX_WORKGROUPS_PER_DIM: u32 = ${MAX_WORKGROUPS_PER_DIM}u;
 const FA2_DIST_FLOOR: f32 = ${wgslF32Literal(FA2_DISTANCE_FLOOR)};
 const FA2_DIST_FLOOR_SQ: f32 = ${wgslF32Literal(FA2_DISTANCE_FLOOR_SQ)};
@@ -969,6 +976,21 @@ fn wg_reduce_u32(v: u32, lid: u32, op: u32) -> u32 {
     workgroupBarrier();
     return total;
 }
+fn wg_scan_u32(v: u32, lid: u32) -> u32 {
+    workgroupBarrier();
+    wg_scratch_u[lid] = v;
+    workgroupBarrier();
+    for (var s = 1u; s < WG; s = s * 2u) {                       // Hillis-Steele inclusive scan (scan-block's body); uniform: every lane runs every round
+        var t = 0u;
+        if (lid >= s) { t = wg_scratch_u[lid - s]; }
+        workgroupBarrier();
+        wg_scratch_u[lid] = wg_scratch_u[lid] + t;
+        workgroupBarrier();
+    }
+    let inclusive = wg_scratch_u[lid];
+    workgroupBarrier();                                          // the scratch is free for the next call
+    return inclusive;
+}
 fn wg_reduce_f32(v: f32, lid: u32, op: u32) -> f32 { return wg_reduce_vec4(vec4f(v, 0.0, 0.0, 0.0), lid, op).x; }`
 );
 const REDUCE_HELPERS_SUBGROUP_WGSL = (
@@ -1048,9 +1070,27 @@ fn wg_reduce_u32(v: u32, lid: u32, op: u32) -> u32 {
     workgroupBarrier();
     return total;
 }
+fn wg_scan_u32(v: u32, lid: u32) -> u32 {
+    workgroupBarrier();
+    if (lid == 0u) { atomicStore(&sg_counter, 0u); }
+    workgroupBarrier();
+    let inSub = subgroupExclusiveAdd(v) + v;                     // the lane's inclusive sum within its subgroup
+    let total = subgroupAdd(v);
+    let key = subgroupMin(lid);                                   // the subgroup's identity: its smallest local id (D16)
+    var slot = 0u;
+    if (subgroupElect()) { slot = atomicAdd(&sg_counter, 1u); }
+    slot = subgroupBroadcast(slot, 0u);
+    if (subgroupElect()) { sg_val_u[slot] = total; sg_key[slot] = key; }
+    workgroupBarrier();
+    let count = atomicLoad(&sg_counter);
+    var carry = 0u;                                               // the totals of every subgroup whose key is below this lane's: LANE order, whatever order the slots were taken in
+    for (var k = 0u; k < count; k = k + 1u) { if (sg_key[k] < key) { carry = carry + sg_val_u[k]; } }
+    workgroupBarrier();                                          // the slots are free for the next call
+    return inSub + carry;
+}
 fn wg_reduce_f32(v: f32, lid: u32, op: u32) -> f32 { return wg_reduce_vec4(vec4f(v, 0.0, 0.0, 0.0), lid, op).x; }`
 );
-const REDUCE_HELPER_NAMES = ["wg_reduce_f32", "wg_reduce_u32", "wg_reduce_vec4"];
+const REDUCE_HELPER_NAMES = ["wg_reduce_f32", "wg_reduce_u32", "wg_reduce_vec4", "wg_scan_u32"];
 const WGSL_RESERVED_WORDS = Object.freeze(
   `NULL Self abstract active alignas alignof as asm asm_fragment async attribute auto await become cast catch class
     co_await co_return co_yield coherent column_major common compile compile_fragment concept const_cast consteval
@@ -3444,12 +3484,17 @@ class GpuContext {
   }
 }
 export {
-  ARC_WINDOW_ALIGN as A,
+  SE_SCALE_REFERENCE_NODES as A,
   BufferUsage as B,
+  ARC_WINDOW_ALIGN as C,
+  PASSTHROUGH_FORMAT_CODES as D,
   EXACT_MAX_NODES as E,
-  FA2_DEFAULTS as F,
+  FRONTIER_CANDIDATES as F,
   GpuContext as G,
+  STORAGE_ALIGN as H,
   INDIRECT_ARGS_STRIDE as I,
+  WORKGROUP_SIZE as J,
+  isSoftwareAdapter as K,
   LAYOUT_TUNING_DEFAULTS as L,
   MAX_WORKGROUPS_PER_DIM as M,
   PARTIAL_BYTES as P,
@@ -3460,28 +3505,29 @@ export {
   WebGpuGraphError as W,
   WGSL_RESERVED_WORDS as a,
   U32_MAX as b,
-  GRID_COARSEST_SIDE as c,
+  MAX_LEVELS_PER_SUBMIT as c,
   deviceLostError as d,
-  GRID_MIN_SIDE as e,
-  GRID_SORT_BITS as f,
-  MAX_ITERATIONS_PER_STEP as g,
-  MAX_1D_ITEMS as h,
+  FUSED_FRONTIER_MAX as e,
+  BEAMER_BETA as f,
+  SSSP_DELTA_FACTOR as g,
+  F32_INF_BITS as h,
   isWebGpuGraphError as i,
-  hasErrorCode as j,
-  FA2_FLAG_FIRST as k,
-  GRID_HUB_CELL as l,
-  GRID_BBOX_MARGIN as m,
-  GRID_EXTENT_FLOOR as n,
-  FR_ADAPTIVE_MAX_ITERATIONS as o,
-  FR_START_TEMPERATURE as p,
-  FA2_FLAG_ADAPTIVE as q,
-  FR_REHEAT_FRACTION as r,
-  FR_DEFAULTS as s,
-  SE_DEFAULTS as t,
-  SE_SCALE_REFERENCE_NODES as u,
-  PASSTHROUGH_FORMAT_CODES as v,
-  STORAGE_ALIGN as w,
-  WORKGROUP_SIZE as x,
-  isSoftwareAdapter as y
+  GRID_COARSEST_SIDE as j,
+  GRID_MIN_SIDE as k,
+  GRID_SORT_BITS as l,
+  FA2_DEFAULTS as m,
+  MAX_ITERATIONS_PER_STEP as n,
+  MAX_1D_ITEMS as o,
+  hasErrorCode as p,
+  FA2_FLAG_FIRST as q,
+  GRID_HUB_CELL as r,
+  GRID_BBOX_MARGIN as s,
+  GRID_EXTENT_FLOOR as t,
+  FR_ADAPTIVE_MAX_ITERATIONS as u,
+  FR_START_TEMPERATURE as v,
+  FA2_FLAG_ADAPTIVE as w,
+  FR_REHEAT_FRACTION as x,
+  FR_DEFAULTS as y,
+  SE_DEFAULTS as z
 };
-//# sourceMappingURL=context-BXqgCifx.js.map
+//# sourceMappingURL=context-Dvq-Cc6v.js.map