Merge pull request 'feat(webgpu-rt): atomic pixel accumulator + raysPerPixel — >1 ray per pixel per bounce (#30)' (#31) from claude/issue-30 into master

README.md

@@ -101,7 +101,9 @@ shaded through an intersection shader with an any-hit cut-out.
   bounce loop (`dispatchWorkgroupsIndirect`). TRACE carries zero user
   code (traversal + intersection only); user raygen calls
   `rtEmitPrimaryRay`, and closesthit / miss run in SHADE where they
-  `rtEmitRay` continuation/shadow rays and `rtAccumulate` radiance. An
+  `rtEmitRay` continuation/shadow rays and `rtAccumulate` radiance
+  (atomic — any number of rays per pixel per bounce, e.g. one shadow
+  ray per light). An
   optional Resolve shader tonemaps the linear accumulator. See
   [WAVEFRONT-DESIGN.md](WAVEFRONT-DESIGN.md).
 - **ComputeShader / WebGPUComputeShader** — Tier 1 wrapper used by the

WAVEFRONT-DESIGN.md

@@ -21,12 +21,17 @@ compute pass, dispatch sizes driven by `dispatchWorkgroupsIndirect`.
 - **RESOLVE** (1 thread/pixel, 8×8): reads accum slot, runs user `resolve_main`
   if present else passthrough; writes outImage.
 
-## Buffers (rtState, sized to 2*W*H rays)
+## Buffers (rtState; per-bounce ray capacity = `RTPass::raysPerPixel`·W·H)
 - `wfRaysA`,`wfRaysB`: array<WfRay>, ping/pong. WfRay = origin,tMin,dir,tMax,
-  pixel,flags,cullMask,missIndex,sbtOffset,payloadSlot,kind,_pad.
+  pixel,flags,cullMask,missIndex,sbtOffset,payloadSlot,kind,_pad. Each holds
+  one bounce's rays; `raysPerPixel` (default 1) scales them so closest-hit
+  can emit several rays per pixel in one bounce before rtEmitRay drops.
 - `wfHits`: array<HitResult> (sized = ray capacity).
 - `wfPayload`: array<Payload> — declared in CODEGEN region after user Payload.
-- `wfAccum`: array<vec4<f32>> per pixel (W*H).
+- `wfAccum`: array<atomic<u32>>, 4 slots per pixel (W*H, RGBA as f32 bit
+  patterns — 16 B/pixel). `rtAccumulate` CASes each channel, so a SHADE
+  invocation may emit several rays for the same pixel in one bounce (one
+  shadow ray per light, issue #30) without the accumulates racing.
 - `wfCounters`: atomic counters: emitA, emitB, trace dispatch args, etc.
 - `wfIndirect`: INDIRECT dispatch-args buffer.
 
@@ -74,6 +79,15 @@ maxDepth=1 (primary only). Sponza maxDepth=2 (primary + shadow).
 - [x] device limits (maxBufferSize / maxStorageBufferBindingSize /
       maxComputeWorkgroupsPerDimension) + timestamp-query feature
 - [x] megakernel dead path removed (RT pipeline builds only wavefront)
+- [x] atomic pixel accumulator (#30) — `rtAccumulate` is a per-channel
+      f32 CAS over `array<atomic<u32>>`, lifting the old one-ray-per-pixel-
+      per-bounce cap so closest-hit can emit one shadow ray per light in a
+      single bounce (multi-light shadowing; 3DForts #153). Same 16 B/pixel
+      footprint; uncontended CAS succeeds first try, so single-ray scenes
+      (VulkanTriangle/Sponza/RTStress) are unaffected. Capacity side:
+      `RTPass::raysPerPixel` (default 1) scales the ray/hit/payload buffers
+      so those N rays/pixel actually fit a bounce instead of being dropped
+      by rtEmitRay's capacity guard. Exercised by `examples/RTMultiShadow`.
 - [~] binding packing (Phase 7): SKIPPED — target device reports 64 storage
       buffers/stage (≥12), so the merge is unnecessary (issue makes it
       conditional on <12). NOTE: this only holds because dom-webgpu.js now

additional/dom-webgpu.js

@@ -1714,7 +1714,11 @@ struct BvhNode {
 @group(1) @binding(10) var<storage,read_write> wfRaysA   : array<WfRay>;
 @group(1) @binding(11) var<storage,read_write> wfRaysB   : array<WfRay>;
 @group(1) @binding(12) var<storage,read_write> wfHits    : array<HitResult>;
-@group(1) @binding(13) var<storage,read_write> wfAccum   : array<vec4<f32>>;
+// wfAccum: 4 atomic<u32> slots per pixel (RGBA as f32 bit patterns — same
+// 16 B/pixel footprint as the old array<vec4<f32>>). Atomic so a SHADE
+// invocation may emit N rays for the same pixel in one bounce (e.g. one
+// shadow ray per light) and their rtAccumulate calls don't race (#30).
+@group(1) @binding(13) var<storage,read_write> wfAccum   : array<atomic<u32>>;
 @group(1) @binding(14) var<storage,read_write> wfCounters : array<atomic<u32>>;
 // @group(1) @binding(15) wfPayload : array<Payload>  — emitted by codegen.
 
@@ -1735,11 +1739,29 @@ fn _wfCurCount() -> u32 {
     return min(raw, wfParams.rayCapacity);
 }
 
-// Add linear radiance to the pixel this SHADE/GENERATE thread owns. Safe
+// Atomic float add via compare-exchange on the f32 bit pattern. WGSL has
-// without atomics: at most one ray per pixel per bounce, and bounces run
+// no native atomic<f32>, so each channel CASes until its read-add-write
-// in separate passes (implicit barrier between them).
+// wins. Uncontended (the common ≤1 ray/pixel/bounce case) the exchange
+// succeeds first try; contention only occurs when several rays for the
+// same pixel resolve in one SHADE pass.
+fn _wfAtomicAddF32(slot: u32, v: f32) {
+    var old = atomicLoad(&wfAccum[slot]);
+    loop {
+        let sum = bitcast<u32>(bitcast<f32>(old) + v);
+        let r = atomicCompareExchangeWeak(&wfAccum[slot], old, sum);
+        if (r.exchanged) { break; }
+        old = r.old_value;
+    }
+}
+
+// Add linear radiance to the pixel this SHADE/GENERATE thread owns. Atomic,
+// so a closest-hit may emit any number of rays for the same pixel within a
+// single bounce (multi-light shadowing) and their accumulates won't race.
 fn rtAccumulate(rgb: vec3<f32>) {
-    wfAccum[_wfPixel] = wfAccum[_wfPixel] + vec4<f32>(rgb, 0.0);
+    let base = _wfPixel * 4u;
+    _wfAtomicAddF32(base + 0u, rgb.x);
+    _wfAtomicAddF32(base + 1u, rgb.y);
+    _wfAtomicAddF32(base + 2u, rgb.z);
 }
 
 // raygen → emit the pixel's primary ray. Bounce 0's current buffer is
@@ -3030,8 +3052,9 @@ function wfNextPow2(n) {
     return p;
 }
 
-function ensureWavefrontBuffers(W, H) {
+function ensureWavefrontBuffers(W, H, raysPerPixel) {
-    const cap = W * H;
+    const rpp = Math.max(1, raysPerPixel | 0);
+    const cap = W * H * rpp;   // per-bounce ray capacity (= WfParams.rayCapacity)
     rtState.wf = rtState.wf || { cap: 0 };
     const wf = rtState.wf;
     if (wf.cap === cap && wf.raysA) return wf;
@@ -3041,7 +3064,8 @@ function ensureWavefrontBuffers(W, H) {
     wf.raysA    = device.createBuffer({ size: cap * 64,                 usage: S, label: "wf-raysA" });
     wf.raysB    = device.createBuffer({ size: cap * 64,                 usage: S, label: "wf-raysB" });
     wf.hits     = device.createBuffer({ size: cap * 112,                usage: S, label: "wf-hits" });
-    wf.accum    = device.createBuffer({ size: cap * 16,                 usage: S, label: "wf-accum" });
+    // Accumulator is per *pixel*, not per ray: 4 atomic<u32> (16 B).
+    wf.accum    = device.createBuffer({ size: W * H * 16,               usage: S, label: "wf-accum" });
     wf.payload  = device.createBuffer({ size: 2 * cap * WF_PAYLOAD_BYTES, usage: S, label: "wf-payload" });
     wf.counters = device.createBuffer({ size: 64,
         usage: GPUBufferUsage.STORAGE | GPUBufferUsage.COPY_DST | GPUBufferUsage.COPY_SRC, label: "wf-counters" });
@@ -3270,7 +3294,8 @@ function wfUserBindGroups(pipe, handlesPtr, handlesCount) {
 
 env.wgpuDispatchRT = (pipelineHandle, pushPtr, pushBytes,
                       tlasBufHandle, instanceCount, gx, gy,
-                      handlesPtr, handlesCount, maxDepth, outTexHandle) => {
+                      handlesPtr, handlesCount, maxDepth, outTexHandle,
+                      raysPerPixel) => {
     if (!state.encoder) return;
     const pipe = rtPipelines.get(pipelineHandle);
     const tlas = buffers.get(tlasBufHandle);
@@ -3296,9 +3321,9 @@ env.wgpuDispatchRT = (pipelineHandle, pushPtr, pushBytes,
         return;
     }
     const W = state.width, H = state.height;
-    const cap = W * H;
     const depth = Math.max(1, maxDepth | 0);
-    const wf = ensureWavefrontBuffers(W, H);
+    const wf = ensureWavefrontBuffers(W, H, raysPerPixel);
+    const cap = wf.cap;   // per-bounce ray capacity = raysPerPixel·W·H
 
     // ── Per-pass WfParams ring. queue.writeBuffer lands before submit, so
     // we can't mutate the uniform between passes — instead we pre-write one

examples/README.md

@@ -106,3 +106,15 @@ barrier WebGPU only provides between submits (or between passes), never
 within a single compute pass. WebGPU/DOM only; the same chain is wireable
 on Vulkan today via an offscreen HDR heap image + a composite `RenderPass`
 (the present path records passes generically and barriers between them).
+
+### [RTMultiShadow](RTMultiShadow/)
+Multi-light shadowing through the wavefront RT pipeline (issue #30). Five
+pillars on a checkered ground, four colored point lights; the closest-hit
+emits one shadow ray **per light** from the same invocation, so up to four
+rays per pixel resolve in a single SHADE pass. Exercises both halves of
+the >1-ray-per-pixel-per-bounce widening: the atomic `rtAccumulate`
+(per-channel f32 CAS — concurrent same-pixel adds don't race) and
+`RTPass::raysPerPixel` (scales the ray/hit/payload buffers so the
+per-light emits aren't capacity-dropped). Any regression shows up as
+flickering dark noise or a missing shadow color in the overlap regions.
+WebGPU/DOM only.

examples/RTMultiShadow/closesthit.wgsl

@@ -0,0 +1,82 @@
+// RTMultiShadow closest-hit (runs in SHADE). The multi-light counterpart
+// of RTStress: EVERY light emits its own shadow ray from this single
+// invocation, so several rays for the same pixel resolve in the next SHADE
+// pass — exactly the contention the atomic rtAccumulate exists for (#30).
+// Before the atomic accumulator their rtAccumulate calls raced (lost
+// updates → flickering dark noise); before RTPass::raysPerPixel the extra
+// rays were silently dropped by the capacity guard. The host sets
+// raysPerPixel = LIGHT_COUNT so every emit fits the bounce.
+//
+// Payload declared here so the assembler sees it before wfPayload / SHADE.
+struct Payload {
+    color:     vec3<f32>,   // shadow ray: pending direct contribution
+    shadowRay: u32,         // 0 primary, 1 shadow
+};
+
+// Point lights, color premultiplied with intensity; 1/d² falloff at shade
+// time. Four distinct hues so each occluder casts four separable shadows —
+// any accumulator race or dropped shadow ray is immediately visible as
+// noise / a missing color in the overlap regions.
+const LIGHT_COUNT: u32 = 4u;
+struct Light {
+    pos:   vec3<f32>,
+    color: vec3<f32>,
+};
+var<private> LIGHTS: array<Light, 4> = array<Light, 4>(
+    Light(vec3<f32>( 14.0,  9.0,   2.0), vec3<f32>(250.0, 205.0, 140.0)),  // warm white
+    Light(vec3<f32>(-13.0,  8.0,   7.0), vec3<f32>(235.0,  45.0,  30.0)),  // red
+    Light(vec3<f32>(  3.0,  8.0, -14.0), vec3<f32>( 55.0, 225.0, 105.0)),  // green
+    Light(vec3<f32>( -5.0, 10.0,  13.0), vec3<f32>( 65.0, 105.0, 250.0)),  // blue
+);
+
+const AMBIENT_COLOR: vec3<f32> = vec3<f32>(0.030, 0.034, 0.045);
+
+// Ground (customIndex 0) is a subtle checker so the colored shadows read;
+// pillars hash their instance index like RTStress.
+fn surfaceAlbedo(customIndex: u32, worldPos: vec3<f32>) -> vec3<f32> {
+    if (customIndex == 0u) {
+        let cx = u32(floor(worldPos.x * 0.25 + 100.0));
+        let cz = u32(floor(worldPos.z * 0.25 + 100.0));
+        return mix(vec3<f32>(0.60), vec3<f32>(0.76), f32((cx + cz) & 1u));
+    }
+    let h = customIndex * 2654435761u;
+    return vec3<f32>(
+        0.45 + 0.5 * f32((h >>  0u) & 255u) / 255.0,
+        0.45 + 0.5 * f32((h >>  8u) & 255u) / 255.0,
+        0.45 + 0.5 * f32((h >> 16u) & 255u) / 255.0);
+}
+
+fn closesthit_main(ray: RayDesc, hit: HitInfo, payload: ptr<function, Payload>) {
+    let meshRec = meshRecords[tlasEntries[hit.instanceId].blasMeshIdx];
+    let verts = _rtFetchTri(meshRec, hit.primitiveId);
+    let nObj = normalize(cross(verts[1] - verts[0], verts[2] - verts[0]));
+    let nWorld = normalize(vec3<f32>(
+        dot(hit.objectToWorldR0.xyz, nObj),
+        dot(hit.objectToWorldR1.xyz, nObj),
+        dot(hit.objectToWorldR2.xyz, nObj)));
+
+    let worldPos = ray.origin + ray.direction * hit.t;
+    let nFacing  = select(-nWorld, nWorld, dot(nWorld, -ray.direction) > 0.0);
+    let albedo   = surfaceAlbedo(hit.customIndex, worldPos);
+
+    rtAccumulate(albedo * AMBIENT_COLOR);
+
+    // One shadow ray PER LIGHT from this one closest-hit invocation. All of
+    // them carry the same pixel; the ones that miss (light visible) each
+    // rtAccumulate their light's contribution in the same SHADE pass.
+    let shadowOrigin = worldPos + nFacing * 0.05;
+    for (var i: u32 = 0u; i < LIGHT_COUNT; i = i + 1u) {
+        let toLight = LIGHTS[i].pos - shadowOrigin;
+        let dist    = length(toLight);
+        let dir     = toLight / dist;
+        let nDotL   = dot(nFacing, dir);
+        if (nDotL <= 0.0) { continue; }
+        var sp: Payload;
+        sp.color     = albedo * LIGHTS[i].color * (nDotL / (dist * dist));
+        sp.shadowRay = 1u;
+        // tMax stops at the light so geometry beyond it can't occlude.
+        rtEmitRay(shadowOrigin, 0.01, dir, dist,
+                  RT_FLAG_SKIP_CLOSEST_HIT | RT_FLAG_TERMINATE_ON_FIRST_HIT,
+                  0xFFu, 0u, 0u, sp);
+    }
+}

examples/RTMultiShadow/main.cpp

@@ -0,0 +1,210 @@
+// RTMultiShadow — multi-light shadowing through the wavefront RT pipeline
+// (issue #30). Five pillars on a checkered ground, lit by four colored
+// point lights; the closest-hit emits one shadow ray PER LIGHT from the
+// same invocation, so up to four rays per pixel resolve in a single SHADE
+// pass. That requires both halves of #30:
+//   - atomic rtAccumulate — the concurrent per-pixel adds don't race;
+//   - RTPass::raysPerPixel — the ray/hit/payload buffers hold 4·W·H rays
+//     per bounce, so none of the per-light emits get capacity-dropped.
+// Each pillar casting four differently-colored shadows is the visual
+// proof; any lost accumulate (race) or dropped ray (capacity) shows up as
+// flickering dark noise / a missing shadow color.
+//
+// WebGPU/DOM only — the wavefront tracer is the WebGPU software RT path.
+
+#ifndef CRAFTER_GRAPHICS_WINDOW_DOM
+int main() { return 0; }   // native path is hardware RT; out of scope here
+#else
+
+import Crafter.Graphics;
+import Crafter.Math;
+import Crafter.Event;
+import std;
+
+using namespace Crafter;
+namespace fs = std::filesystem;
+
+namespace {
+    // Must match LIGHT_COUNT in closesthit.wgsl — it is the raysPerPixel
+    // budget the RTPass is configured with.
+    constexpr std::uint32_t kLightCount = 4;
+
+    struct CameraGPU {
+        float origin[3];   float pad0;
+        float right[3];    float tanHalf;
+        float up[3];       float aspect;
+        float forward[3];  float pad1;
+    };
+    static_assert(sizeof(CameraGPU) == 64);
+
+    // Axis-aligned box: 8 corners between mn and mx, same winding as the
+    // RTStress unit cube.
+    std::array<Vector<float, 3, 3>, 8> BoxVerts(float mnx, float mny, float mnz,
+                                                float mxx, float mxy, float mxz) {
+        return {{
+            {mnx, mny, mnz}, {mxx, mny, mnz}, {mxx, mxy, mnz}, {mnx, mxy, mnz},
+            {mnx, mny, mxz}, {mxx, mny, mxz}, {mxx, mxy, mxz}, {mnx, mxy, mxz},
+        }};
+    }
+    // Mesh::Build takes mutable spans, so this can't be constexpr.
+    std::array<std::uint32_t, 36> kBoxIndices {{
+        0,1,2, 0,2,3,   5,4,7, 5,7,6,   4,0,3, 4,3,7,
+        1,5,6, 1,6,2,   4,5,1, 4,1,0,   3,2,6, 3,6,7,
+    }};
+}
+
+int main() {
+    std::println("[RTMultiShadow] {} lights, one shadow ray per light per pixel", kLightCount);
+
+    Device::Initialize();
+    static Window window(1280, 720, "RTMultiShadow");
+    auto cmd = window.StartInit();
+
+    DescriptorHeapWebGPU heap;
+    heap.Initialize(/*images*/ 1, /*buffers*/ 2, /*samplers*/ 1);
+
+    std::array<WebGPUShader, 4> shaders {{
+        WebGPUShader(fs::path("raygen.wgsl"),     "raygen_main",     WebGPURTStage::Raygen),
+        WebGPUShader(fs::path("miss.wgsl"),       "miss_main",       WebGPURTStage::Miss),
+        WebGPUShader(fs::path("closesthit.wgsl"), "closesthit_main", WebGPURTStage::ClosestHit),
+        WebGPUShader(fs::path("resolve.wgsl"),    "resolve_main",    WebGPURTStage::Resolve),
+    }};
+    ShaderBindingTableWebGPU sbt;
+    sbt.Init(shaders);
+
+    std::array<RTShaderGroup, 1> raygenGroups {{ { .type = RTShaderGroupType::General,           .generalShader = 0 } }};
+    std::array<RTShaderGroup, 1> missGroups   {{ { .type = RTShaderGroupType::General,           .generalShader = 1 } }};
+    std::array<RTShaderGroup, 1> hitGroups    {{ { .type = RTShaderGroupType::TrianglesHitGroup, .closestHitShader = 2 } }};
+
+    // One user binding: the camera storage buffer at @group(3).
+    std::array<UICustomBinding, 1> bindings {{
+        { .group = 3, .binding = 0, .kind = UICustomBindingKind::Buffer, .pushOffset = 0 },
+    }};
+
+    PipelineRTWebGPU pipeline;
+    pipeline.Init(cmd, raygenGroups, missGroups, hitGroups, sbt, bindings);
+
+    // ── Meshes: a large ground slab and a pillar (origin at its base). ──
+    static auto groundVerts = BoxVerts(-30.0f, -1.0f, -30.0f, 30.0f, 0.0f, 30.0f);
+    static auto pillarVerts = BoxVerts(-0.8f,   0.0f,  -0.8f,  0.8f, 6.0f,  0.8f);
+    static Mesh ground, pillar;
+    ground.Build(groundVerts, kBoxIndices, cmd);
+    pillar.Build(pillarVerts, kBoxIndices, cmd);
+
+    // ── Camera buffer + handle array. ─────────────────────────────────
+    WebGPUBuffer<CameraGPU, true> cameraBuf;
+    cameraBuf.Create(1);
+    static std::array<std::uint32_t, 1> userHandles { cameraBuf.handle };
+
+    // ── Instances: ground (customIndex 0) + five pillars. ─────────────
+    struct Placement { float x, z; };
+    static constexpr std::array<Placement, 5> kPillars {{
+        { 0.0f, 0.0f }, { 5.0f, 5.0f }, { -5.0f, 5.0f }, { 5.0f, -5.0f }, { -5.0f, -5.0f },
+    }};
+    static std::vector<RenderingElement3D> renderers;
+    renderers.reserve(1 + kPillars.size());
+    auto addInstance = [&](std::uint64_t blasAddr, float x, float z) {
+        renderers.emplace_back();
+        RenderingElement3D& r = renderers.back();
+        auto& tx = r.instance.transform.matrix;
+        tx[0][0] = 1; tx[0][1] = 0; tx[0][2] = 0; tx[0][3] = x;
+        tx[1][0] = 0; tx[1][1] = 1; tx[1][2] = 0; tx[1][3] = 0;
+        tx[2][0] = 0; tx[2][1] = 0; tx[2][2] = 1; tx[2][3] = z;
+        r.instance.instanceCustomIndex                    = static_cast<std::uint32_t>(renderers.size() - 1);
+        r.instance.mask                                   = 0xFF;
+        r.instance.instanceShaderBindingTableRecordOffset = 0;
+        r.instance.flags                                  = kRTGeometryInstanceForceOpaque;
+        r.instance.accelerationStructureReference         = blasAddr;
+        RenderingElement3D::Add(&r);
+    };
+    addInstance(ground.blasAddr, 0.0f, 0.0f);
+    for (const auto& p : kPillars) addInstance(pillar.blasAddr, p.x, p.z);
+    RenderingElement3D::BuildTLAS(cmd, 0);
+
+    window.descriptorHeap = &heap;
+    window.FinishInit();
+
+    RTPass rtPass(&pipeline);
+    rtPass.handlesPtr   = userHandles.data();
+    rtPass.handlesCount = static_cast<std::uint32_t>(userHandles.size());
+    rtPass.maxDepth     = 2;            // primary + shadow
+    rtPass.raysPerPixel = kLightCount;  // one shadow ray per light per pixel
+    window.passes.push_back(&rtPass);
+
+    // ── Free camera framing the pillars from above one corner. ────────
+    struct CamState {
+        Vector<float, 3, 4> position;
+        float yaw;
+        float pitch;
+    } cam {
+        Vector<float, 3, 4>{ 16.0f, 13.0f, 16.0f },
+        0.0f, 0.0f,
+    };
+    {
+        // Aim at the scene centre, slightly above the ground.
+        Vector<float, 3, 4> d { -cam.position.x, 2.0f - cam.position.y, -cam.position.z };
+        const float len = std::sqrt(d.x*d.x + d.y*d.y + d.z*d.z);
+        cam.yaw   = std::atan2(d.z, d.x);
+        cam.pitch = std::asin(d.y / len);
+    }
+
+    Input::Map inputMap;
+    Input::Action& moveAct = inputMap.AddAction("Move", Input::ActionType::Vector2);
+    Input::Action& lookAct = inputMap.AddAction("Look", Input::ActionType::Vector2);
+    moveAct.bindings = { Input::WASDBind{
+        Key(CrafterKeys::W), Key(CrafterKeys::S), Key(CrafterKeys::A), Key(CrafterKeys::D) } };
+    lookAct.bindings = { Input::MouseDeltaBind{ 1.0f } };
+    inputMap.Attach(window);
+
+    const float kMoveSpeed = 14.0f;
+    const float kLookSens  = 0.05f;
+    const float kDt        = 1.0f / 60.0f;
+
+    static int frames = 0;
+    EventListener<void> camTick(&window.onBeforeUpdate, [&]() {
+        inputMap.Tick();
+        cam.yaw   += lookAct.vector2.x * kLookSens;
+        cam.pitch -= lookAct.vector2.y * kLookSens;
+        cam.pitch  = std::clamp(cam.pitch, -1.55f, 1.55f);
+
+        const float cp = std::cos(cam.pitch), sp = std::sin(cam.pitch);
+        const float cy = std::cos(cam.yaw),   sy = std::sin(cam.yaw);
+        Vector<float, 3, 4> forward { cp * cy, sp, cp * sy };
+        Vector<float, 3, 4> worldUp { 0.0f, 1.0f, 0.0f };
+        Vector<float, 3, 4> right { forward.y*worldUp.z - forward.z*worldUp.y,
+                                    forward.z*worldUp.x - forward.x*worldUp.z,
+                                    forward.x*worldUp.y - forward.y*worldUp.x };
+        const float rLen = std::sqrt(right.x*right.x + right.y*right.y + right.z*right.z);
+        right.x /= rLen; right.y /= rLen; right.z /= rLen;
+        Vector<float, 3, 4> up { right.y*forward.z - right.z*forward.y,
+                                 right.z*forward.x - right.x*forward.z,
+                                 right.x*forward.y - right.y*forward.x };
+
+        const float dx = moveAct.vector2.x * kMoveSpeed * kDt;
+        const float dy = moveAct.vector2.y * kMoveSpeed * kDt;
+        cam.position.x += right.x*dx + forward.x*dy;
+        cam.position.y += right.y*dx + forward.y*dy;
+        cam.position.z += right.z*dx + forward.z*dy;
+
+        CameraGPU& g = cameraBuf.value[0];
+        g.origin[0]=cam.position.x; g.origin[1]=cam.position.y; g.origin[2]=cam.position.z; g.pad0=0;
+        g.right[0]=right.x;   g.right[1]=right.y;   g.right[2]=right.z;
+        g.up[0]=up.x;         g.up[1]=up.y;         g.up[2]=up.z;
+        g.forward[0]=forward.x; g.forward[1]=forward.y; g.forward[2]=forward.z;
+        g.aspect  = float(window.width) / float(window.height);
+        g.tanHalf = std::tan(70.0f * 3.14159265f / 360.0f);
+        g.pad1    = 0;
+        cameraBuf.FlushDevice();
+
+        if (++frames >= 60) {
+            std::println("[RTMultiShadow] {} lights x {} pillars rendering", kLightCount, kPillars.size());
+            frames = 0;
+        }
+    });
+
+    window.Render();
+    window.StartUpdate();
+    window.StartSync();
+    return 0;
+}
+#endif

examples/RTMultiShadow/miss.wgsl

@@ -0,0 +1,14 @@
+// RTMultiShadow miss (runs in SHADE). Shadow miss → that light is visible
+// from the surface, so add its pending contribution; up to LIGHT_COUNT of
+// these resolve for the same pixel in one pass (atomic rtAccumulate, #30).
+// Primary miss → near-black night sky so the colored lighting carries the
+// frame.
+fn miss_main(ray: RayDesc, payload: ptr<function, Payload>) {
+    if ((*payload).shadowRay == 1u) {
+        rtAccumulate((*payload).color);
+        return;
+    }
+    let t = clamp(ray.direction.y * 0.5 + 0.5, 0.0, 1.0);
+    rtAccumulate(mix(vec3<f32>(0.010, 0.012, 0.022),
+                     vec3<f32>(0.030, 0.040, 0.075), t));
+}

examples/RTMultiShadow/project.cpp

@@ -0,0 +1,46 @@
+import std;
+import Crafter.Build;
+namespace fs = std::filesystem;
+using namespace Crafter;
+
+extern "C" Configuration CrafterBuildProject(std::span<const std::string_view> args) {
+    bool isWasm = false;
+    for (std::string_view a : args) {
+        if (a.starts_with("--target=") && a.find("wasm") != std::string_view::npos) {
+            isWasm = true;
+            break;
+        }
+    }
+
+    std::vector<std::string> graphicsArgs(args.begin(), args.end());
+    Configuration* graphics = LocalProject({
+        .projectFile = "../../project.cpp",
+        .args = graphicsArgs,
+    });
+
+    Configuration cfg;
+    cfg.path = "./";
+    cfg.name = "RTMultiShadow";
+    cfg.outputName = "RTMultiShadow";
+    cfg.type = ConfigurationType::Executable;
+    if (isWasm) {
+        cfg.target = "wasm32-wasip1";
+        cfg.defines.push_back({"CRAFTER_GRAPHICS_WINDOW_DOM", ""});
+        cfg.compileFlags.push_back("-msimd128");
+    }
+    ApplyStandardArgs(cfg, args);
+    cfg.dependencies = { graphics };
+
+    std::array<fs::path, 0> ifaces = {};
+    std::array<fs::path, 1> impls = { "main" };
+    cfg.GetInterfacesAndImplementations(ifaces, impls);
+
+    if (isWasm) {
+        cfg.files.emplace_back(fs::path("raygen.wgsl"));
+        cfg.files.emplace_back(fs::path("closesthit.wgsl"));
+        cfg.files.emplace_back(fs::path("miss.wgsl"));
+        cfg.files.emplace_back(fs::path("resolve.wgsl"));
+        EnableWasiBrowserRuntime(cfg);
+    }
+    return cfg;
+}

examples/RTMultiShadow/raygen.wgsl

@@ -0,0 +1,35 @@
+// RTMultiShadow raygen (runs in GENERATE). Host-driven pinhole camera at
+// @group(3) (groups 0..2 are reserved by the wavefront pipeline:
+// 0 = WfParams, 1 = data heaps, 2 = indirect args).
+struct Camera {
+    origin:  vec3<f32>,
+    pad0:    f32,
+    right:   vec3<f32>,
+    tanHalf: f32,
+    up:      vec3<f32>,
+    aspect:  f32,
+    forward: vec3<f32>,
+    pad1:    f32,
+};
+@group(3) @binding(0) var<storage, read> camera : Camera;
+
+fn raygen_main(gid: vec3<u32>) {
+    if (gid.x >= wfParams.surfaceW || gid.y >= wfParams.surfaceH) { return; }
+
+    let pixelf = vec2<f32>(f32(gid.x), f32(gid.y));
+    let res    = vec2<f32>(f32(wfParams.surfaceW), f32(wfParams.surfaceH));
+    let uv     = (pixelf + vec2<f32>(0.5)) / res;
+    let ndc    = uv * 2.0 - vec2<f32>(1.0);
+
+    let direction = normalize(
+        camera.right   * (ndc.x  * camera.aspect * camera.tanHalf) +
+        camera.up      * (-ndc.y * camera.tanHalf) +
+        camera.forward);
+
+    var p: Payload;
+    p.color     = vec3<f32>(0.0);
+    p.shadowRay = 0u;
+
+    rtEmitPrimaryRay(camera.origin, 0.01, direction, 100000.0,
+                     0u, 0xFFu, 0u, 0u, p);
+}

examples/RTMultiShadow/resolve.wgsl

@@ -0,0 +1,7 @@
+// RTMultiShadow RESOLVE-stage tonemap: Reinhard + gamma 2.2 over the
+// linear accumulator. Registered as a WebGPURTStage::Resolve shader.
+fn resolve_main(coord: vec2<u32>, hdr: vec4<f32>) -> vec4<f32> {
+    let mapped = hdr.rgb / (hdr.rgb + vec3<f32>(1.0));
+    let g      = pow(mapped, vec3<f32>(1.0 / 2.2));
+    return vec4<f32>(g, 1.0);
+}

implementations/Crafter.Graphics-PipelineRTWebGPU.cpp

@@ -270,7 +270,13 @@ void PipelineRTWebGPU::Init(WebGPUCommandEncoderRef                 /*cmd*/,
     wgsl += "fn wfGenerate(@builtin(global_invocation_id) gid: vec3<u32>) {\n";
     wgsl += "    if (gid.x >= wfParams.surfaceW || gid.y >= wfParams.surfaceH) { return; }\n";
     wgsl += "    let pixel = gid.y * wfParams.surfaceW + gid.x;\n";
-    wgsl += "    wfAccum[pixel] = vec4<f32>(0.0, 0.0, 0.0, 0.0);\n";
+    // wfAccum is 4 atomic<u32> slots per pixel (f32 bit patterns; see the
+    // bindings prelude JS-side). bitcast<u32>(0.0) == 0u, so plain zeroes.
+    wgsl += "    let accumBase = pixel * 4u;\n";
+    wgsl += "    atomicStore(&wfAccum[accumBase + 0u], 0u);\n";
+    wgsl += "    atomicStore(&wfAccum[accumBase + 1u], 0u);\n";
+    wgsl += "    atomicStore(&wfAccum[accumBase + 2u], 0u);\n";
+    wgsl += "    atomicStore(&wfAccum[accumBase + 3u], 0u);\n";
     wgsl += "    _wfPixel = pixel;\n";
     wgsl += "    ";
     wgsl += raygenEntryFn;
@@ -300,7 +306,16 @@ void PipelineRTWebGPU::Init(WebGPUCommandEncoderRef                 /*cmd*/,
     wgsl += "fn wfResolve(@builtin(global_invocation_id) gid: vec3<u32>) {\n";
     wgsl += "    if (gid.x >= wfParams.surfaceW || gid.y >= wfParams.surfaceH) { return; }\n";
     wgsl += "    let pixel = gid.y * wfParams.surfaceW + gid.x;\n";
-    wgsl += "    let outc = runResolve(gid.xy, wfAccum[pixel]);\n";
+    // Reassemble the vec4<f32> the resolve hook expects from the pixel's 4
+    // atomic<u32> accumulator slots (no contention here — RESOLVE runs in
+    // its own pass — but atomics may only be accessed atomically in WGSL).
+    wgsl += "    let accumBase = pixel * 4u;\n";
+    wgsl += "    let hdr = vec4<f32>(\n";
+    wgsl += "        bitcast<f32>(atomicLoad(&wfAccum[accumBase + 0u])),\n";
+    wgsl += "        bitcast<f32>(atomicLoad(&wfAccum[accumBase + 1u])),\n";
+    wgsl += "        bitcast<f32>(atomicLoad(&wfAccum[accumBase + 2u])),\n";
+    wgsl += "        bitcast<f32>(atomicLoad(&wfAccum[accumBase + 3u])));\n";
+    wgsl += "    let outc = runResolve(gid.xy, hdr);\n";
     wgsl += "    textureStore(outImage, vec2<i32>(i32(gid.x), i32(gid.y)), outc);\n";
     wgsl += "}\n";
 

interfaces/Crafter.Graphics-RTPass.cppm

@@ -103,6 +103,13 @@ export namespace Crafter {
         // its own composite→swapchain pass afterwards. Ignored (0) for the
         // default canvas path.
         std::uint32_t     outTexHandle = 0;
+        // Per-bounce ray budget as a multiple of the pixel count. The
+        // wavefront ray/hit/payload buffers hold raysPerPixel·W·H rays, so
+        // a closest-hit may emit up to raysPerPixel rays per pixel within
+        // one bounce (e.g. one shadow ray per light — rtAccumulate is
+        // atomic, issue #30) before rtEmitRay starts dropping. Memory
+        // scales linearly; keep at 1 for single-ray-per-pixel pipelines.
+        std::uint32_t     raysPerPixel = 1;
 
         RTPass(PipelineRTWebGPU* p) : pipeline(p) {}
 
@@ -121,7 +128,8 @@ export namespace Crafter {
                 handlesPtr,
                 static_cast<std::int32_t>(handlesCount),
                 static_cast<std::int32_t>(maxDepth),
-                outTexHandle);
+                outTexHandle,
+                static_cast<std::int32_t>(raysPerPixel));
         }
     };
 }

interfaces/Crafter.Graphics-WebGPU.cppm

@@ -222,6 +222,11 @@ namespace Crafter::WebGPU {
     // `outTexHandle` is the destination texture for an HDR-output pipeline
     // (one loaded with hdrOutputFormat != 0); ignored (pass 0) for the
     // default canvas path.
+    // `raysPerPixel` scales the wavefront ray/hit/payload buffers to
+    // raysPerPixel·W·H rays per bounce, so closest-hit can emit up to that
+    // many rays per pixel within one bounce (e.g. one shadow ray per light)
+    // without rtEmitRay dropping past capacity. 1 = the historical
+    // single-ray-per-pixel footprint.
     __attribute__((import_module("env"), import_name("wgpuDispatchRT")))
     extern "C" void wgpuDispatchRT(std::uint32_t pipelineHandle,
                                    const void* pushPtr, std::int32_t pushBytes,
@@ -230,7 +235,8 @@ namespace Crafter::WebGPU {
                                    std::int32_t  gx, std::int32_t gy,
                                    const void* handlesPtr, std::int32_t handlesCount,
                                    std::int32_t maxDepth,
-                                   std::uint32_t outTexHandle);
+                                   std::uint32_t outTexHandle,
+                                   std::int32_t  raysPerPixel);
 
     // GPU TLAS-build dispatch. Two sequential compute passes:
     //   1. tlasBuildMain — per-instance world AABB + identity permutation