@photivra/engine 1.5.0 → 2.0.0

package/dist/gpu/sensor-stage-gpu.js

@@ -0,0 +1,317 @@
+// SPDX-License-Identifier: Apache-2.0
+/**
+ * Host side of the `webgpu-f32` sensor stage (#277): domain checks, frame constants, per-site mean
+ * packing, banded dispatch and readback. The float64 reference is `simulatePackedSensorRaw`
+ * (capture/packed-sensor-raw.ts). Internal; the public entry point is the image-kernel accelerator.
+ */
+import { createHostYieldGate } from "../capture/host-yield-cadence.js";
+import { calculatePackedSensorRawPlan, createPackedFixedPattern, packedSensorDiagnostics, packedSensorExtras, simulatePackedSensorRows } from "../capture/packed-sensor-raw.js";
+import { adcTransfer } from "../sensor/raw-readout.js";
+import { ImageKernelExecutionError, webGpuExecution } from "./image-kernel-execution.js";
+import { BUFFER_USAGE, GpuResourceScope, abortedError, linearDispatch, runComputePass } from "./webgpu-runtime.js";
+import { SENSOR_AUX, SENSOR_EXTRAS, SENSOR_MEAN_CATEGORY, SENSOR_OUTPUT, SENSOR_RAW_WGSL, SENSOR_SITE_WORDS, SENSOR_WORKGROUP_SIZE } from "./wgsl-sensor-kernel.js";
+/** Largest physical capacity on the GPU: every realized count up to it is exact in float32. */
+export const SENSOR_GPU_MAXIMUM_CAPACITY = 2 ** 24;
+/** Poisson means must stay below this on the GPU (30-bit integer part of a mean word). */
+export const SENSOR_GPU_MAXIMUM_MEAN = 2 ** 30;
+/** Largest |fixed-pattern readout offset| on the GPU, in electrons and in codes. */
+export const SENSOR_GPU_MAXIMUM_OFFSET_ELECTRONS = 2 ** 28;
+export const SENSOR_GPU_MAXIMUM_OFFSET_CODES = 2 ** 20;
+const CPU_RELATIVE = 2 ** -47;
+const U = 2 ** -24;
+const POISSON_NORMAL_MINIMUM_MEAN = 1e9;
+function checkAbort(signal) {
+    if (signal.aborted)
+        throw abortedError();
+}
+/** CPU reference with host yields and abort, row-chunked; codes equal `simulatePackedSensorRaw`. */
+export async function simulateSensorRawOnCpu(input, abort, yieldControl, execution) {
+    const plan = calculatePackedSensorRawPlan(input);
+    // Snapshot so a caller mutation during host yields cannot change the reference.
+    const photo = input.photoElectrons.slice(), dark = input.darkElectrons.slice(), extras = packedSensorExtras(plan, input, true);
+    const codes = new Uint16Array(plan.siteCount), counts = [0, 0, 0, 0];
+    const rowsPerChunk = Math.max(1, Math.floor(65_536 / plan.region.width)), gate = createHostYieldGate();
+    for (let row = 0; row < plan.region.height; row += rowsPerChunk) {
+        checkAbort(abort.signal);
+        simulatePackedSensorRows(plan, photo, dark, row, Math.min(plan.region.height, row + rowsPerChunk), codes, counts, extras);
+        if (yieldControl && gate.due()) {
+            await yieldControl(abort.signal);
+            gate.restart();
+        }
+    }
+    checkAbort(abort.signal);
+    return { plan, codes, diagnostics: packedSensorDiagnostics(counts), execution, flaggedSiteIndices: new Uint32Array(0) };
+}
+/**
+ * Splits a positive value into four non-negative pieces with at most 12 significant bits each (by
+ * truncation), so that the product of any integer below 2^12 and a piece is exact in float32. Returns
+ * the pieces and the non-negative remainder (below 2^-44 of the value).
+ * @internal
+ */
+export function twelveBitPieces(value) {
+    const pieces = [];
+    let rest = value;
+    for (let j = 0; j < 4; j++) {
+        if (!(rest > 0)) {
+            pieces.push(0);
+            continue;
+        }
+        let e = Math.floor(Math.log2(rest));
+        if (rest / 2 ** e >= 2)
+            e++;
+        if (rest / 2 ** e < 1)
+            e--;
+        const scale = 2 ** (e - 11), piece = Math.floor(rest / scale) * scale;
+        pieces.push(piece);
+        rest -= piece;
+    }
+    return { pieces, residual: rest };
+}
+/**
+ * `L(k) − S(k)` for k < 256, where L is the reference's log k! (the summed logarithms of
+ * `samplePoisson`'s logFactorial) and S(k) = (k + ½) ln k − k + ½ ln 2π.
+ * @internal
+ */
+export function stirlingRemainderTable() {
+    const table = new Float64Array(256);
+    let sum = 0;
+    for (let k = 1; k < 256; k++) {
+        if (k >= 2)
+            sum += Math.log(k);
+        table[k] = sum - ((k + 0.5) * Math.log(k) - k + 0.5 * Math.log(2 * Math.PI));
+    }
+    return table;
+}
+function unsupported(message) {
+    return new ImageKernelExecutionError("unsupported-float32-domain", message);
+}
+/** Checks the float32 execution domain and builds the frame constants (see docs/WEBGPU.md). @internal */
+export function sensorGpuConstants(plan) {
+    const regime = plan.readout.regime, adc = regime.adc, capacity = plan.physicalChargeCapacityElectrons;
+    const gain = regime.systemConversionGainElectronsPerCode.value, inverse = 1 / gain;
+    const threshold = regime.preAdcSaturationElectronEquivalent.value;
+    const sigmas = regime.readNoiseComponents.map(c => c.rmsElectrons.value);
+    const sigmaSum = sigmas.reduce((s, v) => s + v, 0);
+    if (capacity > SENSOR_GPU_MAXIMUM_CAPACITY)
+        throw unsupported("Physical capacity above 2^24 electrons is outside the float32 sensor domain.");
+    // Fixed-pattern factors scale the means; those are checked per site while packing.
+    if (!(plan.maximumPhotoElectrons < SENSOR_GPU_MAXIMUM_MEAN && plan.maximumDarkElectrons < SENSOR_GPU_MAXIMUM_MEAN) ||
+        plan.additionalChargeComponents.some(c => !(c.maximumExpectedElectrons < SENSOR_GPU_MAXIMUM_MEAN))) {
+        throw unsupported("Expected electrons of 2^30 and above are outside the float32 sensor domain.");
+    }
+    if (!(inverse >= 2 ** -60 && inverse <= 2 ** 60) || capacity * inverse > 2 ** 30 ||
+        sigmas.some(s => s !== 0 && !(s >= 2 ** -100 && s <= 2 ** 100)) || 7 * sigmaSum > 2 ** 28 || 7 * sigmaSum * inverse > 2 ** 20) {
+        throw unsupported("Conversion gain or read noise is outside the float32 sensor domain.");
+    }
+    const knots = adc.integralNonLinearity?.knots ?? [];
+    let slope = 0, offset = 0;
+    knots.forEach((k, i) => {
+        offset = Math.max(offset, Math.abs(k.offsetCodes));
+        if (i > 0)
+            slope = Math.max(slope, Math.abs((k.offsetCodes - knots[i - 1].offsetCodes) / (k.code - knots[i - 1].code)));
+    });
+    const aux = new Float32Array(SENSOR_AUX.knots + 2 * knots.length);
+    aux.set(stirlingRemainderTable(), SENSOR_AUX.stirlingRemainder);
+    aux.set(sigmas, SENSOR_AUX.sigma);
+    const inversePieces = twelveBitPieces(inverse), digital = adc.digitalGain === undefined ? null : twelveBitPieces(adc.digitalGain);
+    aux.set(inversePieces.pieces, SENSOR_AUX.inversePieces);
+    if (digital)
+        aux.set(digital.pieces, SENSOR_AUX.digitalPieces);
+    knots.forEach((k, i) => { aux[SENSOR_AUX.knots + i] = k.code; aux[SENSOR_AUX.knots + knots.length + i] = k.offsetCodes; });
+    // Pre-ADC-saturated sites all have the same rounded ADC input; computed once with the reference, before digital gain.
+    const { digitalGain: _omitted, ...withoutDigital } = adc;
+    void _omitted;
+    const saturated = adcTransfer({ ...regime, adc: withoutDigital }, plan.readout.adcMaximumCode, threshold).quantized;
+    const params = new ArrayBuffer(128), words = new Uint32Array(params), ints = new Int32Array(params), floats = new Float32Array(params);
+    const capacityFloor = Math.floor(capacity), thresholdFloor = Math.min(Math.floor(threshold), 2 ** 30);
+    words[2] = plan.region.width;
+    words[3] = plan.nativeWidth;
+    words[4] = plan.region.x;
+    words[5] = plan.region.y;
+    // Mode bits: 1 digital gain; 2 the digital-gain product is exact on both sides (gain has ≤ 12 significant
+    // bits); 4 the ADC input is exact on both sides (no read noise, power-of-two gain, integer capacity, no INL).
+    // In an exact mode both evaluations give the same dyadic value, so a tie is decided identically and needs no flag.
+    const digitalExact = digital !== null && digital.residual === 0 && digital.pieces.slice(1).every(p => p === 0);
+    const poissonComponents = [], deterministicComponents = [];
+    plan.additionalChargeComponents.forEach((c, k) => (c.model === "poisson" ? poissonComponents : deterministicComponents).push(k));
+    const deterministic = deterministicComponents.length > 0, pattern = plan.fixedPatternNoise !== null;
+    const idealExact = sigmas.every(s => s === 0) && Number.isInteger(Math.log2(gain)) && Number.isInteger(capacity) && knots.length === 0 &&
+        !deterministic && !pattern;
+    words[6] = plan.captureSeedUint32;
+    words[7] = sigmas.length;
+    words[8] = knots.length;
+    words[9] = (digital ? 1 : 0) | (digitalExact ? 2 : 0) | (idealExact ? 4 : 0);
+    words[10] = adc.digitalSaturationCode;
+    words[11] = plan.readout.adcMaximumCode;
+    words[12] = capacityFloor;
+    words[13] = thresholdFloor;
+    words[14] = adc.blackLevelCode;
+    ints[15] = Math.max(-(2 ** 30), Math.min(2 ** 30, saturated));
+    floats[16] = capacity - capacityFloor;
+    floats[17] = threshold < 2 ** 30 ? threshold - thresholdFloor : 0;
+    floats[18] = inverse;
+    floats[19] = inversePieces.residual * (1 + 2 ** -20);
+    floats[20] = adc.digitalGain ?? 1;
+    floats[21] = digital ? digital.residual * (1 + 2 ** -20) : 0;
+    floats[22] = slope * (1 + 2 ** -20);
+    floats[23] = offset;
+    const siteWords = SENSOR_SITE_WORDS + 2 * poissonComponents.length + (deterministic ? 2 : 0) + (pattern ? 4 : 0);
+    words[24] = siteWords;
+    words[25] = poissonComponents.length;
+    words[26] = (deterministic ? SENSOR_EXTRAS.deterministic : 0) | (pattern ? SENSOR_EXTRAS.fixedPattern : 0);
+    return { aux, params, siteWords, poissonComponents, deterministicComponents, threshold, gain };
+}
+/** Writes the float64 slack terms of the fixed-pattern bounds, from the region's largest offsets. */
+function setFixedPatternSlack(params, threshold, maximumOffset, maximumOffsetCodes) {
+    const floats = new Float32Array(params);
+    // The reference adds the offset in float64 and the host forms threshold − offset and offset/gain in float64.
+    floats[28] = CPU_RELATIVE * (maximumOffset + Math.min(threshold, 2 ** 30)) * (1 + 2 ** -20);
+    // The offset fraction's float32 rounding, one more unit of |fp| across the ten fraction additions, and offset/gain's float64 rounding.
+    floats[29] = (12 * U + CPU_RELATIVE * maximumOffsetCodes) * (1 + 2 ** -20);
+}
+/** Splits a value into a floor and its float32 fraction. */
+function packSplit(value, words, floats, at) {
+    const whole = Math.floor(value);
+    words[at] = whole;
+    floats[at + 1] = value - whole;
+}
+/** Packs one mean: category in bits 30–31 with the integer part, and the float32 fraction. */
+function packMean(mean, words, floats, at) {
+    if (mean === 0) {
+        words[at] = 0;
+        floats[at + 1] = 0;
+        return;
+    }
+    if (!(mean < SENSOR_GPU_MAXIMUM_MEAN))
+        throw unsupported("Expected electrons of 2^30 and above are outside the float32 sensor domain.");
+    const whole = Math.floor(mean);
+    const category = mean < 30 ? SENSOR_MEAN_CATEGORY.multiplication : mean < POISSON_NORMAL_MINIMUM_MEAN ? SENSOR_MEAN_CATEGORY.transformedRejection
+        : SENSOR_MEAN_CATEGORY.normal;
+    words[at] = ((category << 30) | whole) >>> 0;
+    floats[at + 1] = mean - whole;
+}
+export async function simulateSensorRawOnGpu(runtime, bandBytes, input, abort, yieldControl, pipelines) {
+    const plan = calculatePackedSensorRawPlan(input);
+    const constants = sensorGpuConstants(plan), { aux, params, siteWords, poissonComponents, deterministicComponents, threshold, gain } = constants;
+    const photo = input.photoElectrons.slice(), dark = input.darkElectrons.slice();
+    const additional = (input.additionalChargeComponents ?? []).map(c => c.expectedElectrons.slice());
+    const width = plan.region.width, height = plan.region.height, siteBytes = siteWords * 4;
+    const pattern = createPackedFixedPattern(plan);
+    // Fixed-pattern values are evaluated once, in float64 (exactly the reference's values), before any dispatch.
+    const patterns = pattern === null ? null : new Float64Array(plan.siteCount * 3);
+    if (patterns !== null) {
+        let maximumOffset = 0, maximumOffsetCodes = 0;
+        for (let row = 0, i = 0; row < height; row++)
+            for (let column = 0; column < width; column++, i++) {
+                const p = pattern(column, row), offsetCodes = p.readoutOffsetElectrons / gain;
+                patterns[i * 3] = p.photoResponseGainFactor;
+                patterns[i * 3 + 1] = p.darkCurrentFactor;
+                patterns[i * 3 + 2] = p.readoutOffsetElectrons;
+                maximumOffset = Math.max(maximumOffset, Math.abs(p.readoutOffsetElectrons));
+                maximumOffsetCodes = Math.max(maximumOffsetCodes, Math.abs(offsetCodes));
+            }
+        if (!(maximumOffset <= SENSOR_GPU_MAXIMUM_OFFSET_ELECTRONS && maximumOffsetCodes <= SENSOR_GPU_MAXIMUM_OFFSET_CODES)) {
+            throw unsupported("Fixed-pattern readout offsets are outside the float32 sensor domain.");
+        }
+        setFixedPatternSlack(params, threshold, maximumOffset, maximumOffsetCodes);
+    }
+    const rows = Math.max(1, Math.min(height, Math.floor(bandBytes / (width * siteBytes))));
+    if (rows * width * siteBytes > runtime.maximumStorageBindingBytes) {
+        throw new ImageKernelExecutionError("insufficient-limits", "One sensor band exceeds the device's storage binding limit.");
+    }
+    const pipeline = await pipelines("photivra-sensor-raw", SENSOR_RAW_WGSL);
+    const codes = new Uint16Array(plan.siteCount), counts = [0, 0, 0, 0];
+    const packed = new ArrayBuffer(rows * width * siteBytes), packedWords = new Uint32Array(packed), packedFloats = new Float32Array(packed);
+    const packedInts = new Int32Array(packed);
+    const words = new Uint32Array(params);
+    let flagged = 0;
+    const flaggedIndices = [];
+    const scope = new GpuResourceScope(runtime);
+    try {
+        const auxBuffer = scope.upload("photivra-sensor-aux", aux, BUFFER_USAGE.STORAGE);
+        const gate = createHostYieldGate();
+        for (let row0 = 0; row0 < height; row0 += rows) {
+            checkAbort(abort.signal);
+            const row1 = Math.min(height, row0 + rows), count = (row1 - row0) * width, first = row0 * width;
+            for (let i = 0; i < count; i++) {
+                const site = first + i, at = i * siteWords;
+                // The reference's own float64 products of the means and the fixed-pattern factors.
+                packMean(patterns === null ? photo[site] : photo[site] * patterns[site * 3], packedWords, packedFloats, at);
+                packMean(patterns === null ? dark[site] : dark[site] * patterns[site * 3 + 1], packedWords, packedFloats, at + 2);
+                let cursor = at + 4;
+                for (const k of poissonComponents) {
+                    packMean(additional[k][site], packedWords, packedFloats, cursor);
+                    cursor += 2;
+                }
+                if (deterministicComponents.length > 0) {
+                    let sum = 0;
+                    for (const k of deterministicComponents)
+                        sum += additional[k][site];
+                    if (!(sum < SENSOR_GPU_MAXIMUM_MEAN))
+                        throw unsupported("Deterministic charge of 2^30 electrons and above is outside the float32 sensor domain.");
+                    packSplit(sum, packedInts, packedFloats, cursor);
+                    cursor += 2;
+                }
+                if (patterns !== null) {
+                    const offset = patterns[site * 3 + 2], limit = threshold - offset;
+                    if (limit >= 2 ** 30) {
+                        packedInts[cursor] = 2 ** 30;
+                        packedFloats[cursor + 1] = 0;
+                    }
+                    else
+                        packSplit(limit, packedInts, packedFloats, cursor);
+                    packSplit(offset / gain, packedInts, packedFloats, cursor + 2);
+                }
+            }
+            words[0] = count;
+            words[1] = row0;
+            const band = new GpuResourceScope(runtime);
+            try {
+                const means = band.upload("photivra-sensor-means", new Uint32Array(packed, 0, count * siteWords), BUFFER_USAGE.STORAGE);
+                const out = band.buffer("photivra-sensor-codes", count * 4, BUFFER_USAGE.STORAGE | BUFFER_USAGE.COPY_SRC);
+                const uniform = band.upload("photivra-sensor-params", new Uint8Array(params), BUFFER_USAGE.UNIFORM);
+                const bindGroup = runtime.device.createBindGroup({ layout: pipeline.getBindGroupLayout(0), entries: [
+                        { binding: 0, resource: { buffer: uniform } }, { binding: 1, resource: { buffer: means } },
+                        { binding: 2, resource: { buffer: auxBuffer } }, { binding: 3, resource: { buffer: out } }
+                    ] });
+                const [data] = await runComputePass(runtime, band, pipeline, bindGroup, linearDispatch(count, SENSOR_WORKGROUP_SIZE, runtime.maximumWorkgroupsPerDimension), [{ buffer: out, bytes: count * 4 }], abort.signal);
+                const result = new Uint32Array(data, 0, count);
+                for (let i = 0; i < count; i++) {
+                    const word = result[i], flags = word >>> SENSOR_OUTPUT.flagShift;
+                    codes[first + i] = word & SENSOR_OUTPUT.codeMask;
+                    if (flags & 1)
+                        counts[0]++;
+                    if (flags & 2)
+                        counts[1]++;
+                    if (flags & 4)
+                        counts[2]++;
+                    if (flags & 8)
+                        counts[3]++;
+                    if (word & SENSOR_OUTPUT.uncertainBit) {
+                        flagged++;
+                        flaggedIndices.push(first + i);
+                    }
+                }
+            }
+            finally {
+                band.release();
+            }
+            if (yieldControl && gate.due()) {
+                await yieldControl(abort.signal);
+                gate.restart();
+            }
+        }
+    }
+    catch (error) {
+        codes.fill(0);
+        throw error;
+    }
+    finally {
+        scope.release();
+    }
+    checkAbort(abort.signal);
+    const execution = webGpuExecution({ kind: "exact-except-flagged-sites", flaggedSiteCount: flagged, noiseModel: { ...plan.noiseModel },
+        seedSchedule: plan.seedSchedule }, flagged);
+    return { plan, codes, diagnostics: packedSensorDiagnostics(counts), execution, flaggedSiteIndices: Uint32Array.from(flaggedIndices) };
+}
+//# sourceMappingURL=sensor-stage-gpu.js.map

package/dist/gpu/sensor-stage-gpu.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"sensor-stage-gpu.js","sourceRoot":"","sources":["../../src/gpu/sensor-stage-gpu.ts"],"names":[],"mappings":"AAAA,sCAAsC;AAEtC;;;;GAIG;AACH,OAAO,EAAE,mBAAmB,EAAE,MAAM,kCAAkC,CAAC;AACvE,OAAO,EAAE,4BAA4B,EAAE,wBAAwB,EAAE,uBAAuB,EAAE,kBAAkB,EAAE,wBAAwB,EAC/E,MAAM,iCAAiC,CAAC;AAC/F,OAAO,EAAE,WAAW,EAAsC,MAAM,0BAA0B,CAAC;AAC3F,OAAO,EAAE,yBAAyB,EAAE,eAAe,EAA6B,MAAM,6BAA6B,CAAC;AAEpH,OAAO,EAAE,YAAY,EAAE,gBAAgB,EAAE,YAAY,EAAE,cAAc,EAAE,cAAc,EAAsB,MAAM,qBAAqB,CAAC;AACvI,OAAO,EAAE,UAAU,EAAE,aAAa,EAAE,oBAAoB,EAAE,aAAa,EAAE,eAAe,EAAE,iBAAiB,EAAE,qBAAqB,EAAE,MAAM,yBAAyB,CAAC;AAKpK,+FAA+F;AAC/F,MAAM,CAAC,MAAM,2BAA2B,GAAG,CAAC,IAAI,EAAE,CAAC;AACnD,0FAA0F;AAC1F,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC,IAAI,EAAE,CAAC;AAC/C,oFAAoF;AACpF,MAAM,CAAC,MAAM,mCAAmC,GAAG,CAAC,IAAI,EAAE,CAAC;AAC3D,MAAM,CAAC,MAAM,+BAA+B,GAAG,CAAC,IAAI,EAAE,CAAC;AACvD,MAAM,YAAY,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;AAC9B,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;AACnB,MAAM,2BAA2B,GAAG,GAAG,CAAC;AAExC,SAAS,UAAU,CAAC,MAAmB;IACrC,IAAI,MAAM,CAAC,OAAO;QAAE,MAAM,YAAY,EAAE,CAAC;AAC3C,CAAC;AAED,oGAAoG;AACpG,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAAC,KAA2B,EAAE,KAAsB,EAAE,YAA0B,EAC1H,SAA+B;IAC/B,MAAM,IAAI,GAAG,4BAA4B,CAAC,KAAK,CAAC,CAAC;IACjD,gFAAgF;IAChF,MAAM,KAAK,GAAG,KAAK,CAAC,cAAc,CAAC,KAAK,EAAE,EAAE,IAAI,GAAG,KAAK,CAAC,aAAa,CAAC,KAAK,EAAE,EAAE,MAAM,GAAG,kBAAkB,CAAC,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,CAAC;IAC/H,MAAM,KAAK,GAAG,IAAI,WAAW,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;IACrE,MAAM,YAAY,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,GAAG,mBAAmB,EAAE,CAAC;IACvG,KAAK,IAAI,GAAG,GAAG,CAAC,EAAE,GAAG,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,GAAG,IAAI,YAAY,EAAE,CAAC;QAChE,UAAU,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QACzB,wBAAwB,CAAC,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,GAAG,GAAG,YAAY,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;QAC1H,IAAI,YAAY,IAAI,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC;YAAC,MAAM,YAAY,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;YAAC,IAAI,CAAC,OAAO,EAAE,CAAC;QAAC,CAAC;IACvF,CAAC;IACD,UAAU,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IACzB,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,WAAW,EAAE,uBAAuB,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,kBAAkB,EAAE,IAAI,WAAW,CAAC,CAAC,CAAC,EAAE,CAAC;AAC1H,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,eAAe,CAAC,KAAa;IAC3C,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,IAAI,IAAI,GAAG,KAAK,CAAC;IACjB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3B,IAAI,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,EAAE,CAAC;YAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YAAC,SAAS;QAAC,CAAC;QAC9C,IAAI,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;QACpC,IAAI,IAAI,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC;YAAE,CAAC,EAAE,CAAC;QAC5B,IAAI,IAAI,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC;YAAE,CAAC,EAAE,CAAC;QAC3B,MAAM,KAAK,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,GAAG,KAAK,CAAC,GAAG,KAAK,CAAC;QACtE,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACnB,IAAI,IAAI,KAAK,CAAC;IAChB,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;AACpC,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,sBAAsB;IACpC,MAAM,KAAK,GAAG,IAAI,YAAY,CAAC,GAAG,CAAC,CAAC;IACpC,IAAI,GAAG,GAAG,CAAC,CAAC;IACZ,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7B,IAAI,CAAC,IAAI,CAAC;YAAE,GAAG,IAAI,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAC/B,KAAK,CAAC,CAAC,CAAC,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC;IAC/E,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,WAAW,CAAC,OAAe;IAClC,OAAO,IAAI,yBAAyB,CAAC,4BAA4B,EAAE,OAAO,CAAC,CAAC;AAC9E,CAAC;AAKD,yGAAyG;AACzG,MAAM,UAAU,kBAAkB,CAAC,IAAyB;IAC1D,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,GAAG,MAAM,CAAC,GAAG,EAAE,QAAQ,GAAG,IAAI,CAAC,+BAA+B,CAAC;IACtG,MAAM,IAAI,GAAG,MAAM,CAAC,oCAAoC,CAAC,KAAK,EAAE,OAAO,GAAG,CAAC,GAAG,IAAI,CAAC;IACnF,MAAM,SAAS,GAAG,MAAM,CAAC,kCAAkC,CAAC,KAAK,CAAC;IAClE,MAAM,MAAM,GAAG,MAAM,CAAC,mBAAmB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC;IACzE,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;IACnD,IAAI,QAAQ,GAAG,2BAA2B;QAAE,MAAM,WAAW,CAAC,8EAA8E,CAAC,CAAC;IAC9I,mFAAmF;IACnF,IAAI,CAAC,CAAC,IAAI,CAAC,qBAAqB,GAAG,uBAAuB,IAAI,IAAI,CAAC,oBAAoB,GAAG,uBAAuB,CAAC;QAC9G,IAAI,CAAC,0BAA0B,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,wBAAwB,GAAG,uBAAuB,CAAC,CAAC,EAAE,CAAC;QACvG,MAAM,WAAW,CAAC,6EAA6E,CAAC,CAAC;IACnG,CAAC;IACD,IAAI,CAAC,CAAC,OAAO,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,OAAO,IAAI,CAAC,IAAI,EAAE,CAAC,IAAI,QAAQ,GAAG,OAAO,GAAG,CAAC,IAAI,EAAE;QAC5E,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC,GAAG,QAAQ,GAAG,CAAC,IAAI,EAAE,IAAI,CAAC,GAAG,QAAQ,GAAG,OAAO,GAAG,CAAC,IAAI,EAAE,EAAE,CAAC;QAClI,MAAM,WAAW,CAAC,qEAAqE,CAAC,CAAC;IAC3F,CAAC;IACD,MAAM,KAAK,GAAG,GAAG,CAAC,oBAAoB,EAAE,KAAK,IAAI,EAAE,CAAC;IACpD,IAAI,KAAK,GAAG,CAAC,EAAE,MAAM,GAAG,CAAC,CAAC;IAC1B,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACrB,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC;QACnD,IAAI,CAAC,GAAG,CAAC;YAAE,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,WAAW,GAAG,KAAK,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,GAAG,KAAK,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAC5H,CAAC,CAAC,CAAC;IACH,MAAM,GAAG,GAAG,IAAI,YAAY,CAAC,UAAU,CAAC,KAAK,GAAG,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC;IAClE,GAAG,CAAC,GAAG,CAAC,sBAAsB,EAAE,EAAE,UAAU,CAAC,iBAAiB,CAAC,CAAC;IAChE,GAAG,CAAC,GAAG,CAAC,MAAM,EAAE,UAAU,CAAC,KAAK,CAAC,CAAC;IAClC,MAAM,aAAa,GAAG,eAAe,CAAC,OAAO,CAAC,EAAE,OAAO,GAAG,GAAG,CAAC,WAAW,KAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,eAAe,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC;IAClI,GAAG,CAAC,GAAG,CAAC,aAAa,CAAC,MAAM,EAAE,UAAU,CAAC,aAAa,CAAC,CAAC;IACxD,IAAI,OAAO;QAAE,GAAG,CAAC,GAAG,CAAC,OAAO,CAAC,MAAM,EAAE,UAAU,CAAC,aAAa,CAAC,CAAC;IAC/D,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,GAAG,GAAG,CAAC,UAAU,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,UAAU,CAAC,KAAK,GAAG,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3H,sHAAsH;IACtH,MAAM,EAAE,WAAW,EAAE,QAAQ,EAAE,GAAG,cAAc,EAAE,GAAG,GAAG,CAAC;IACzD,KAAK,QAAQ,CAAC;IACd,MAAM,SAAS,GAAG,WAAW,CAAC,EAAE,GAAG,MAAM,EAAE,GAAG,EAAE,cAAc,EAAmC,EAAE,IAAI,CAAC,OAAO,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC,SAAS,CAAC;IACrJ,MAAM,MAAM,GAAG,IAAI,WAAW,CAAC,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,WAAW,CAAC,MAAM,CAAC,EAAE,IAAI,GAAG,IAAI,UAAU,CAAC,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,YAAY,CAAC,MAAM,CAAC,CAAC;IACvI,MAAM,aAAa,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC;IACtG,KAAK,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC;IAAC,KAAK,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,WAAW,CAAC;IAAC,KAAK,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC;IAAC,KAAK,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC;IAC9G,0GAA0G;IAC1G,8GAA8G;IAC9G,mHAAmH;IACnH,MAAM,YAAY,GAAG,OAAO,KAAK,IAAI,IAAI,OAAO,CAAC,QAAQ,KAAK,CAAC,IAAI,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC;IAC/G,MAAM,iBAAiB,GAAa,EAAE,EAAE,uBAAuB,GAAa,EAAE,CAAC;IAC/E,IAAI,CAAC,0BAA0B,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,uBAAuB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;IACjI,MAAM,aAAa,GAAG,uBAAuB,CAAC,MAAM,GAAG,CAAC,EAAE,OAAO,GAAG,IAAI,CAAC,iBAAiB,KAAK,IAAI,CAAC;IACpG,MAAM,UAAU,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,IAAI,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,MAAM,CAAC,SAAS,CAAC,QAAQ,CAAC,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QACpI,CAAC,aAAa,IAAI,CAAC,OAAO,CAAC;IAC7B,KAAK,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,iBAAiB,CAAC;IAAC,KAAK,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IAAC,KAAK,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC;IACrF,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC7E,KAAK,CAAC,EAAE,CAAC,GAAG,GAAG,CAAC,qBAAqB,CAAC;IAAC,KAAK,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC;IAAC,KAAK,CAAC,EAAE,CAAC,GAAG,aAAa,CAAC;IAAC,KAAK,CAAC,EAAE,CAAC,GAAG,cAAc,CAAC;IACtI,KAAK,CAAC,EAAE,CAAC,GAAG,GAAG,CAAC,cAAc,CAAC;IAAC,IAAI,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,SAAS,CAAC,CAAC,CAAC;IAC9F,MAAM,CAAC,EAAE,CAAC,GAAG,QAAQ,GAAG,aAAa,CAAC;IAAC,MAAM,CAAC,EAAE,CAAC,GAAG,SAAS,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,SAAS,GAAG,cAAc,CAAC,CAAC,CAAC,CAAC,CAAC;IACzG,MAAM,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC;IAAC,MAAM,CAAC,EAAE,CAAC,GAAG,aAAa,CAAC,QAAQ,GAAG,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC3E,MAAM,CAAC,EAAE,CAAC,GAAG,GAAG,CAAC,WAAW,IAAI,CAAC,CAAC;IAAC,MAAM,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,QAAQ,GAAG,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAChG,MAAM,CAAC,EAAE,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAAC,MAAM,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC;IACzD,MAAM,SAAS,GAAG,iBAAiB,GAAG,CAAC,GAAG,iBAAiB,CAAC,MAAM,GAAG,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACjH,KAAK,CAAC,EAAE,CAAC,GAAG,SAAS,CAAC;IAAC,KAAK,CAAC,EAAE,CAAC,GAAG,iBAAiB,CAAC,MAAM,CAAC;IAC5D,KAAK,CAAC,EAAE,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC,CAAC,aAAa,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,aAAa,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3G,OAAO,EAAE,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,iBAAiB,EAAE,uBAAuB,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;AACjG,CAAC;AAED,qGAAqG;AACrG,SAAS,oBAAoB,CAAC,MAAmB,EAAE,SAAiB,EAAE,aAAqB,EAAE,kBAA0B;IACrH,MAAM,MAAM,GAAG,IAAI,YAAY,CAAC,MAAM,CAAC,CAAC;IACxC,6GAA6G;IAC7G,MAAM,CAAC,EAAE,CAAC,GAAG,YAAY,GAAG,CAAC,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,SAAS,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC5F,uIAAuI;IACvI,MAAM,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,GAAG,YAAY,GAAG,kBAAkB,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAC7E,CAAC;AAED,4DAA4D;AAC5D,SAAS,SAAS,CAAC,KAAa,EAAE,KAAiB,EAAE,MAAoB,EAAE,EAAU;IACnF,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;IAChC,KAAK,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC;IAAC,MAAM,CAAC,EAAE,GAAG,CAAC,CAAC,GAAG,KAAK,GAAG,KAAK,CAAC;AACpD,CAAC;AAED,8FAA8F;AAC9F,SAAS,QAAQ,CAAC,IAAY,EAAE,KAAkB,EAAE,MAAoB,EAAE,EAAU;IAClF,IAAI,IAAI,KAAK,CAAC,EAAE,CAAC;QAAC,KAAK,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC;QAAC,MAAM,CAAC,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QAAC,OAAO;IAAC,CAAC;IAC9D,IAAI,CAAC,CAAC,IAAI,GAAG,uBAAuB,CAAC;QAAE,MAAM,WAAW,CAAC,6EAA6E,CAAC,CAAC;IACxI,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC/B,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAE,CAAC,CAAC,CAAC,oBAAoB,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI,GAAG,2BAA2B,CAAC,CAAC,CAAC,oBAAoB,CAAC,oBAAoB;QAC/I,CAAC,CAAC,oBAAoB,CAAC,MAAM,CAAC;IAChC,KAAK,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,QAAQ,IAAI,EAAE,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;IAC7C,MAAM,CAAC,EAAE,GAAG,CAAC,CAAC,GAAG,IAAI,GAAG,KAAK,CAAC;AAChC,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAAC,OAAsB,EAAE,SAAiB,EAAE,KAA2B,EAAE,KAAsB,EACzI,YAA0B,EAAE,SAAyB;IACrD,MAAM,IAAI,GAAG,4BAA4B,CAAC,KAAK,CAAC,CAAC;IACjD,MAAM,SAAS,GAAG,kBAAkB,CAAC,IAAI,CAAC,EAAE,EAAE,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,iBAAiB,EAAE,uBAAuB,EAAE,SAAS,EAAE,IAAI,EAAE,GAAG,SAAS,CAAC;IAChJ,MAAM,KAAK,GAAG,KAAK,CAAC,cAAc,CAAC,KAAK,EAAE,EAAE,IAAI,GAAG,KAAK,CAAC,aAAa,CAAC,KAAK,EAAE,CAAC;IAC/E,MAAM,UAAU,GAAG,CAAC,KAAK,CAAC,0BAA0B,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,iBAAiB,CAAC,KAAK,EAAE,CAAC,CAAC;IAClG,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,SAAS,GAAG,SAAS,GAAG,CAAC,CAAC;IACxF,MAAM,OAAO,GAAG,wBAAwB,CAAC,IAAI,CAAC,CAAC;IAC/C,6GAA6G;IAC7G,MAAM,QAAQ,GAAG,OAAO,KAAK,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,YAAY,CAAC,IAAI,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC;IAChF,IAAI,QAAQ,KAAK,IAAI,EAAE,CAAC;QACtB,IAAI,aAAa,GAAG,CAAC,EAAE,kBAAkB,GAAG,CAAC,CAAC;QAC9C,KAAK,IAAI,GAAG,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,GAAG,GAAG,MAAM,EAAE,GAAG,EAAE;YAAE,KAAK,IAAI,MAAM,GAAG,CAAC,EAAE,MAAM,GAAG,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC;gBACjG,MAAM,CAAC,GAAG,OAAQ,CAAC,MAAM,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,CAAC,CAAC,sBAAsB,GAAG,IAAI,CAAC;gBAC/E,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,uBAAuB,CAAC;gBAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,iBAAiB,CAAC;gBAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,sBAAsB,CAAC;gBACvI,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,aAAa,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC;gBAAC,kBAAkB,GAAG,IAAI,CAAC,GAAG,CAAC,kBAAkB,EAAE,IAAI,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC;YACxJ,CAAC;QACD,IAAI,CAAC,CAAC,aAAa,IAAI,mCAAmC,IAAI,kBAAkB,IAAI,+BAA+B,CAAC,EAAE,CAAC;YACrH,MAAM,WAAW,CAAC,sEAAsE,CAAC,CAAC;QAC5F,CAAC;QACD,oBAAoB,CAAC,MAAM,EAAE,SAAS,EAAE,aAAa,EAAE,kBAAkB,CAAC,CAAC;IAC7E,CAAC;IACD,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,IAAI,CAAC,KAAK,CAAC,SAAS,GAAG,CAAC,KAAK,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC;IACxF,IAAI,IAAI,GAAG,KAAK,GAAG,SAAS,GAAG,OAAO,CAAC,0BAA0B,EAAE,CAAC;QAClE,MAAM,IAAI,yBAAyB,CAAC,qBAAqB,EAAE,6DAA6D,CAAC,CAAC;IAC5H,CAAC;IACD,MAAM,QAAQ,GAAG,MAAM,SAAS,CAAC,qBAAqB,EAAE,eAAe,CAAC,CAAC;IACzE,MAAM,KAAK,GAAG,IAAI,WAAW,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;IACrE,MAAM,MAAM,GAAG,IAAI,WAAW,CAAC,IAAI,GAAG,KAAK,GAAG,SAAS,CAAC,EAAE,WAAW,GAAG,IAAI,WAAW,CAAC,MAAM,CAAC,EAAE,YAAY,GAAG,IAAI,YAAY,CAAC,MAAM,CAAC,CAAC;IACzI,MAAM,UAAU,GAAG,IAAI,UAAU,CAAC,MAAM,CAAC,CAAC;IAC1C,MAAM,KAAK,GAAG,IAAI,WAAW,CAAC,MAAM,CAAC,CAAC;IACtC,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,MAAM,cAAc,GAAa,EAAE,CAAC;IACpC,MAAM,KAAK,GAAG,IAAI,gBAAgB,CAAC,OAAO,CAAC,CAAC;IAC5C,IAAI,CAAC;QACH,MAAM,SAAS,GAAG,KAAK,CAAC,MAAM,CAAC,qBAAqB,EAAE,GAAG,EAAE,YAAY,CAAC,OAAO,CAAC,CAAC;QACjF,MAAM,IAAI,GAAG,mBAAmB,EAAE,CAAC;QACnC,KAAK,IAAI,IAAI,GAAG,CAAC,EAAE,IAAI,GAAG,MAAM,EAAE,IAAI,IAAI,IAAI,EAAE,CAAC;YAC/C,UAAU,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;YACzB,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,IAAI,GAAG,IAAI,CAAC,EAAE,KAAK,GAAG,CAAC,IAAI,GAAG,IAAI,CAAC,GAAG,KAAK,EAAE,KAAK,GAAG,IAAI,GAAG,KAAK,CAAC;YAChG,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC/B,MAAM,IAAI,GAAG,KAAK,GAAG,CAAC,EAAE,EAAE,GAAG,CAAC,GAAG,SAAS,CAAC;gBAC3C,mFAAmF;gBACnF,QAAQ,CAAC,QAAQ,KAAK,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAE,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAE,GAAG,QAAQ,CAAC,IAAI,GAAG,CAAC,CAAE,EAAE,WAAW,EAAE,YAAY,EAAE,EAAE,CAAC,CAAC;gBAC/G,QAAQ,CAAC,QAAQ,KAAK,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAE,GAAG,QAAQ,CAAC,IAAI,GAAG,CAAC,GAAG,CAAC,CAAE,EAAE,WAAW,EAAE,YAAY,EAAE,EAAE,GAAG,CAAC,CAAC,CAAC;gBACrH,IAAI,MAAM,GAAG,EAAE,GAAG,CAAC,CAAC;gBACpB,KAAK,MAAM,CAAC,IAAI,iBAAiB,EAAE,CAAC;oBAAC,QAAQ,CAAC,UAAU,CAAC,CAAC,CAAE,CAAC,IAAI,CAAE,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,CAAC,CAAC;oBAAC,MAAM,IAAI,CAAC,CAAC;gBAAC,CAAC;gBACvH,IAAI,uBAAuB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;oBACvC,IAAI,GAAG,GAAG,CAAC,CAAC;oBACZ,KAAK,MAAM,CAAC,IAAI,uBAAuB;wBAAE,GAAG,IAAI,UAAU,CAAC,CAAC,CAAE,CAAC,IAAI,CAAE,CAAC;oBACtE,IAAI,CAAC,CAAC,GAAG,GAAG,uBAAuB,CAAC;wBAAE,MAAM,WAAW,CAAC,wFAAwF,CAAC,CAAC;oBAClJ,SAAS,CAAC,GAAG,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,CAAC,CAAC;oBACjD,MAAM,IAAI,CAAC,CAAC;gBACd,CAAC;gBACD,IAAI,QAAQ,KAAK,IAAI,EAAE,CAAC;oBACtB,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,GAAG,CAAC,GAAG,CAAC,CAAE,EAAE,KAAK,GAAG,SAAS,GAAG,MAAM,CAAC;oBACnE,IAAI,KAAK,IAAI,CAAC,IAAI,EAAE,EAAE,CAAC;wBAAC,UAAU,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC;wBAAC,YAAY,CAAC,MAAM,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;oBAAC,CAAC;;wBAAM,SAAS,CAAC,KAAK,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,CAAC,CAAC;oBAC9I,SAAS,CAAC,MAAM,GAAG,IAAI,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,GAAG,CAAC,CAAC,CAAC;gBACjE,CAAC;YACH,CAAC;YACD,KAAK,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC;YAAC,KAAK,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC;YAClC,MAAM,IAAI,GAAG,IAAI,gBAAgB,CAAC,OAAO,CAAC,CAAC;YAC3C,IAAI,CAAC;gBACH,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,uBAAuB,EAAE,IAAI,WAAW,CAAC,MAAM,EAAE,CAAC,EAAE,KAAK,GAAG,SAAS,CAAC,EAAE,YAAY,CAAC,OAAO,CAAC,CAAC;gBACxH,MAAM,GAAG,GAAG,IAAI,CAAC,MAAM,CAAC,uBAAuB,EAAE,KAAK,GAAG,CAAC,EAAE,YAAY,CAAC,OAAO,GAAG,YAAY,CAAC,QAAQ,CAAC,CAAC;gBAC1G,MAAM,OAAO,GAAG,IAAI,CAAC,MAAM,CAAC,wBAAwB,EAAE,IAAI,UAAU,CAAC,MAAM,CAAC,EAAE,YAAY,CAAC,OAAO,CAAC,CAAC;gBACpG,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,eAAe,CAAC,EAAE,MAAM,EAAE,QAAQ,CAAC,kBAAkB,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE;wBAClG,EAAE,OAAO,EAAE,CAAC,EAAE,QAAQ,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,EAAE,EAAE,OAAO,EAAE,CAAC,EAAE,QAAQ,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE;wBAC1F,EAAE,OAAO,EAAE,CAAC,EAAE,QAAQ,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,EAAE,EAAE,EAAE,OAAO,EAAE,CAAC,EAAE,QAAQ,EAAE,EAAE,MAAM,EAAE,GAAG,EAAE,EAAE;qBAAC,EAAE,CAAC,CAAC;gBACjG,MAAM,CAAC,IAAI,CAAC,GAAG,MAAM,cAAc,CAAC,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,SAAS,EAAE,cAAc,CAAC,KAAK,EAAE,qBAAqB,EAAE,OAAO,CAAC,6BAA6B,CAAC,EACzJ,CAAC,EAAE,MAAM,EAAE,GAAG,EAAE,KAAK,EAAE,KAAK,GAAG,CAAC,EAAE,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;gBACrD,MAAM,MAAM,GAAG,IAAI,WAAW,CAAC,IAAK,EAAE,CAAC,EAAE,KAAK,CAAC,CAAC;gBAChD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,EAAE,CAAC,EAAE,EAAE,CAAC;oBAC/B,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAE,EAAE,KAAK,GAAG,IAAI,KAAK,aAAa,CAAC,SAAS,CAAC;oBAClE,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,IAAI,GAAG,aAAa,CAAC,QAAQ,CAAC;oBACjD,IAAI,KAAK,GAAG,CAAC;wBAAE,MAAM,CAAC,CAAC,CAAE,EAAE,CAAC;oBAC5B,IAAI,KAAK,GAAG,CAAC;wBAAE,MAAM,CAAC,CAAC,CAAE,EAAE,CAAC;oBAC5B,IAAI,KAAK,GAAG,CAAC;wBAAE,MAAM,CAAC,CAAC,CAAE,EAAE,CAAC;oBAC5B,IAAI,KAAK,GAAG,CAAC;wBAAE,MAAM,CAAC,CAAC,CAAE,EAAE,CAAC;oBAC5B,IAAI,IAAI,GAAG,aAAa,CAAC,YAAY,EAAE,CAAC;wBAAC,OAAO,EAAE,CAAC;wBAAC,cAAc,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;oBAAC,CAAC;gBACvF,CAAC;YACH,CAAC;oBAAS,CAAC;gBACT,IAAI,CAAC,OAAO,EAAE,CAAC;YACjB,CAAC;YACD,IAAI,YAAY,IAAI,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC;gBAAC,MAAM,YAAY,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;gBAAC,IAAI,CAAC,OAAO,EAAE,CAAC;YAAC,CAAC;QACvF,CAAC;IACH,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACd,MAAM,KAAK,CAAC;IACd,CAAC;YAAS,CAAC;QACT,KAAK,CAAC,OAAO,EAAE,CAAC;IAClB,CAAC;IACD,UAAU,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IACzB,MAAM,SAAS,GAAG,eAAe,CAAC,EAAE,IAAI,EAAE,4BAA4B,EAAE,gBAAgB,EAAE,OAAO,EAAE,UAAU,EAAE,EAAE,GAAG,IAAI,CAAC,UAAU,EAAE;QACnI,YAAY,EAAE,IAAI,CAAC,YAAY,EAAE,EAAE,OAAO,CAAC,CAAC;IAC9C,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,WAAW,EAAE,uBAAuB,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,kBAAkB,EAAE,WAAW,CAAC,IAAI,CAAC,cAAc,CAAC,EAAE,CAAC;AACxI,CAAC","sourcesContent":["// SPDX-License-Identifier: Apache-2.0\n\n/**\n * Host side of the `webgpu-f32` sensor stage (#277): domain checks, frame constants, per-site mean\n * packing, banded dispatch and readback. The float64 reference is `simulatePackedSensorRaw`\n * (capture/packed-sensor-raw.ts). Internal; the public entry point is the image-kernel accelerator.\n */\nimport { createHostYieldGate } from \"../capture/host-yield-cadence.js\";\nimport { calculatePackedSensorRawPlan, createPackedFixedPattern, packedSensorDiagnostics, packedSensorExtras, simulatePackedSensorRows,\n  type PackedSensorRawInput, type PackedSensorRawPlan } from \"../capture/packed-sensor-raw.js\";\nimport { adcTransfer, type SensorReadoutConversionRegime } from \"../sensor/raw-readout.js\";\nimport { ImageKernelExecutionError, webGpuExecution, type ImageKernelExecution } from \"./image-kernel-execution.js\";\nimport type { AcceleratedPackedSensorRawResult } from \"./image-kernel-accelerator.js\";\nimport { BUFFER_USAGE, GpuResourceScope, abortedError, linearDispatch, runComputePass, type WebGpuRuntime } from \"./webgpu-runtime.js\";\nimport { SENSOR_AUX, SENSOR_EXTRAS, SENSOR_MEAN_CATEGORY, SENSOR_OUTPUT, SENSOR_RAW_WGSL, SENSOR_SITE_WORDS, SENSOR_WORKGROUP_SIZE } from \"./wgsl-sensor-kernel.js\";\n\ntype PipelineSource = (label: string, code: string) => Promise<GPUComputePipeline>;\ntype YieldControl = ((signal: AbortSignal) => Promise<void>) | undefined;\n\n/** Largest physical capacity on the GPU: every realized count up to it is exact in float32. */\nexport const SENSOR_GPU_MAXIMUM_CAPACITY = 2 ** 24;\n/** Poisson means must stay below this on the GPU (30-bit integer part of a mean word). */\nexport const SENSOR_GPU_MAXIMUM_MEAN = 2 ** 30;\n/** Largest |fixed-pattern readout offset| on the GPU, in electrons and in codes. */\nexport const SENSOR_GPU_MAXIMUM_OFFSET_ELECTRONS = 2 ** 28;\nexport const SENSOR_GPU_MAXIMUM_OFFSET_CODES = 2 ** 20;\nconst CPU_RELATIVE = 2 ** -47;\nconst U = 2 ** -24;\nconst POISSON_NORMAL_MINIMUM_MEAN = 1e9;\n\nfunction checkAbort(signal: AbortSignal): void {\n  if (signal.aborted) throw abortedError();\n}\n\n/** CPU reference with host yields and abort, row-chunked; codes equal `simulatePackedSensorRaw`. */\nexport async function simulateSensorRawOnCpu(input: PackedSensorRawInput, abort: AbortController, yieldControl: YieldControl,\n  execution: ImageKernelExecution): Promise<AcceleratedPackedSensorRawResult> {\n  const plan = calculatePackedSensorRawPlan(input);\n  // Snapshot so a caller mutation during host yields cannot change the reference.\n  const photo = input.photoElectrons.slice(), dark = input.darkElectrons.slice(), extras = packedSensorExtras(plan, input, true);\n  const codes = new Uint16Array(plan.siteCount), counts = [0, 0, 0, 0];\n  const rowsPerChunk = Math.max(1, Math.floor(65_536 / plan.region.width)), gate = createHostYieldGate();\n  for (let row = 0; row < plan.region.height; row += rowsPerChunk) {\n    checkAbort(abort.signal);\n    simulatePackedSensorRows(plan, photo, dark, row, Math.min(plan.region.height, row + rowsPerChunk), codes, counts, extras);\n    if (yieldControl && gate.due()) { await yieldControl(abort.signal); gate.restart(); }\n  }\n  checkAbort(abort.signal);\n  return { plan, codes, diagnostics: packedSensorDiagnostics(counts), execution, flaggedSiteIndices: new Uint32Array(0) };\n}\n\n/**\n * Splits a positive value into four non-negative pieces with at most 12 significant bits each (by\n * truncation), so that the product of any integer below 2^12 and a piece is exact in float32. Returns\n * the pieces and the non-negative remainder (below 2^-44 of the value).\n * @internal\n */\nexport function twelveBitPieces(value: number): { pieces: number[]; residual: number } {\n  const pieces: number[] = [];\n  let rest = value;\n  for (let j = 0; j < 4; j++) {\n    if (!(rest > 0)) { pieces.push(0); continue; }\n    let e = Math.floor(Math.log2(rest));\n    if (rest / 2 ** e >= 2) e++;\n    if (rest / 2 ** e < 1) e--;\n    const scale = 2 ** (e - 11), piece = Math.floor(rest / scale) * scale;\n    pieces.push(piece);\n    rest -= piece;\n  }\n  return { pieces, residual: rest };\n}\n\n/**\n * `L(k) − S(k)` for k < 256, where L is the reference's log k! (the summed logarithms of\n * `samplePoisson`'s logFactorial) and S(k) = (k + ½) ln k − k + ½ ln 2π.\n * @internal\n */\nexport function stirlingRemainderTable(): Float64Array {\n  const table = new Float64Array(256);\n  let sum = 0;\n  for (let k = 1; k < 256; k++) {\n    if (k >= 2) sum += Math.log(k);\n    table[k] = sum - ((k + 0.5) * Math.log(k) - k + 0.5 * Math.log(2 * Math.PI));\n  }\n  return table;\n}\n\nfunction unsupported(message: string): ImageKernelExecutionError {\n  return new ImageKernelExecutionError(\"unsupported-float32-domain\", message);\n}\n\ninterface SensorConstants { aux: Float32Array<ArrayBuffer>; params: ArrayBuffer; siteWords: number; poissonComponents: number[];\n  deterministicComponents: number[]; threshold: number; gain: number }\n\n/** Checks the float32 execution domain and builds the frame constants (see docs/WEBGPU.md). @internal */\nexport function sensorGpuConstants(plan: PackedSensorRawPlan): SensorConstants {\n  const regime = plan.readout.regime, adc = regime.adc, capacity = plan.physicalChargeCapacityElectrons;\n  const gain = regime.systemConversionGainElectronsPerCode.value, inverse = 1 / gain;\n  const threshold = regime.preAdcSaturationElectronEquivalent.value;\n  const sigmas = regime.readNoiseComponents.map(c => c.rmsElectrons.value);\n  const sigmaSum = sigmas.reduce((s, v) => s + v, 0);\n  if (capacity > SENSOR_GPU_MAXIMUM_CAPACITY) throw unsupported(\"Physical capacity above 2^24 electrons is outside the float32 sensor domain.\");\n  // Fixed-pattern factors scale the means; those are checked per site while packing.\n  if (!(plan.maximumPhotoElectrons < SENSOR_GPU_MAXIMUM_MEAN && plan.maximumDarkElectrons < SENSOR_GPU_MAXIMUM_MEAN) ||\n      plan.additionalChargeComponents.some(c => !(c.maximumExpectedElectrons < SENSOR_GPU_MAXIMUM_MEAN))) {\n    throw unsupported(\"Expected electrons of 2^30 and above are outside the float32 sensor domain.\");\n  }\n  if (!(inverse >= 2 ** -60 && inverse <= 2 ** 60) || capacity * inverse > 2 ** 30 ||\n      sigmas.some(s => s !== 0 && !(s >= 2 ** -100 && s <= 2 ** 100)) || 7 * sigmaSum > 2 ** 28 || 7 * sigmaSum * inverse > 2 ** 20) {\n    throw unsupported(\"Conversion gain or read noise is outside the float32 sensor domain.\");\n  }\n  const knots = adc.integralNonLinearity?.knots ?? [];\n  let slope = 0, offset = 0;\n  knots.forEach((k, i) => {\n    offset = Math.max(offset, Math.abs(k.offsetCodes));\n    if (i > 0) slope = Math.max(slope, Math.abs((k.offsetCodes - knots[i - 1]!.offsetCodes) / (k.code - knots[i - 1]!.code)));\n  });\n  const aux = new Float32Array(SENSOR_AUX.knots + 2 * knots.length);\n  aux.set(stirlingRemainderTable(), SENSOR_AUX.stirlingRemainder);\n  aux.set(sigmas, SENSOR_AUX.sigma);\n  const inversePieces = twelveBitPieces(inverse), digital = adc.digitalGain === undefined ? null : twelveBitPieces(adc.digitalGain);\n  aux.set(inversePieces.pieces, SENSOR_AUX.inversePieces);\n  if (digital) aux.set(digital.pieces, SENSOR_AUX.digitalPieces);\n  knots.forEach((k, i) => { aux[SENSOR_AUX.knots + i] = k.code; aux[SENSOR_AUX.knots + knots.length + i] = k.offsetCodes; });\n  // Pre-ADC-saturated sites all have the same rounded ADC input; computed once with the reference, before digital gain.\n  const { digitalGain: _omitted, ...withoutDigital } = adc;\n  void _omitted;\n  const saturated = adcTransfer({ ...regime, adc: withoutDigital } as SensorReadoutConversionRegime, plan.readout.adcMaximumCode, threshold).quantized;\n  const params = new ArrayBuffer(128), words = new Uint32Array(params), ints = new Int32Array(params), floats = new Float32Array(params);\n  const capacityFloor = Math.floor(capacity), thresholdFloor = Math.min(Math.floor(threshold), 2 ** 30);\n  words[2] = plan.region.width; words[3] = plan.nativeWidth; words[4] = plan.region.x; words[5] = plan.region.y;\n  // Mode bits: 1 digital gain; 2 the digital-gain product is exact on both sides (gain has ≤ 12 significant\n  // bits); 4 the ADC input is exact on both sides (no read noise, power-of-two gain, integer capacity, no INL).\n  // In an exact mode both evaluations give the same dyadic value, so a tie is decided identically and needs no flag.\n  const digitalExact = digital !== null && digital.residual === 0 && digital.pieces.slice(1).every(p => p === 0);\n  const poissonComponents: number[] = [], deterministicComponents: number[] = [];\n  plan.additionalChargeComponents.forEach((c, k) => (c.model === \"poisson\" ? poissonComponents : deterministicComponents).push(k));\n  const deterministic = deterministicComponents.length > 0, pattern = plan.fixedPatternNoise !== null;\n  const idealExact = sigmas.every(s => s === 0) && Number.isInteger(Math.log2(gain)) && Number.isInteger(capacity) && knots.length === 0 &&\n    !deterministic && !pattern;\n  words[6] = plan.captureSeedUint32; words[7] = sigmas.length; words[8] = knots.length;\n  words[9] = (digital ? 1 : 0) | (digitalExact ? 2 : 0) | (idealExact ? 4 : 0);\n  words[10] = adc.digitalSaturationCode; words[11] = plan.readout.adcMaximumCode; words[12] = capacityFloor; words[13] = thresholdFloor;\n  words[14] = adc.blackLevelCode; ints[15] = Math.max(-(2 ** 30), Math.min(2 ** 30, saturated));\n  floats[16] = capacity - capacityFloor; floats[17] = threshold < 2 ** 30 ? threshold - thresholdFloor : 0;\n  floats[18] = inverse; floats[19] = inversePieces.residual * (1 + 2 ** -20);\n  floats[20] = adc.digitalGain ?? 1; floats[21] = digital ? digital.residual * (1 + 2 ** -20) : 0;\n  floats[22] = slope * (1 + 2 ** -20); floats[23] = offset;\n  const siteWords = SENSOR_SITE_WORDS + 2 * poissonComponents.length + (deterministic ? 2 : 0) + (pattern ? 4 : 0);\n  words[24] = siteWords; words[25] = poissonComponents.length;\n  words[26] = (deterministic ? SENSOR_EXTRAS.deterministic : 0) | (pattern ? SENSOR_EXTRAS.fixedPattern : 0);\n  return { aux, params, siteWords, poissonComponents, deterministicComponents, threshold, gain };\n}\n\n/** Writes the float64 slack terms of the fixed-pattern bounds, from the region's largest offsets. */\nfunction setFixedPatternSlack(params: ArrayBuffer, threshold: number, maximumOffset: number, maximumOffsetCodes: number): void {\n  const floats = new Float32Array(params);\n  // The reference adds the offset in float64 and the host forms threshold − offset and offset/gain in float64.\n  floats[28] = CPU_RELATIVE * (maximumOffset + Math.min(threshold, 2 ** 30)) * (1 + 2 ** -20);\n  // The offset fraction's float32 rounding, one more unit of |fp| across the ten fraction additions, and offset/gain's float64 rounding.\n  floats[29] = (12 * U + CPU_RELATIVE * maximumOffsetCodes) * (1 + 2 ** -20);\n}\n\n/** Splits a value into a floor and its float32 fraction. */\nfunction packSplit(value: number, words: Int32Array, floats: Float32Array, at: number): void {\n  const whole = Math.floor(value);\n  words[at] = whole; floats[at + 1] = value - whole;\n}\n\n/** Packs one mean: category in bits 30–31 with the integer part, and the float32 fraction. */\nfunction packMean(mean: number, words: Uint32Array, floats: Float32Array, at: number): void {\n  if (mean === 0) { words[at] = 0; floats[at + 1] = 0; return; }\n  if (!(mean < SENSOR_GPU_MAXIMUM_MEAN)) throw unsupported(\"Expected electrons of 2^30 and above are outside the float32 sensor domain.\");\n  const whole = Math.floor(mean);\n  const category = mean < 30 ? SENSOR_MEAN_CATEGORY.multiplication : mean < POISSON_NORMAL_MINIMUM_MEAN ? SENSOR_MEAN_CATEGORY.transformedRejection\n    : SENSOR_MEAN_CATEGORY.normal;\n  words[at] = ((category << 30) | whole) >>> 0;\n  floats[at + 1] = mean - whole;\n}\n\nexport async function simulateSensorRawOnGpu(runtime: WebGpuRuntime, bandBytes: number, input: PackedSensorRawInput, abort: AbortController,\n  yieldControl: YieldControl, pipelines: PipelineSource): Promise<AcceleratedPackedSensorRawResult> {\n  const plan = calculatePackedSensorRawPlan(input);\n  const constants = sensorGpuConstants(plan), { aux, params, siteWords, poissonComponents, deterministicComponents, threshold, gain } = constants;\n  const photo = input.photoElectrons.slice(), dark = input.darkElectrons.slice();\n  const additional = (input.additionalChargeComponents ?? []).map(c => c.expectedElectrons.slice());\n  const width = plan.region.width, height = plan.region.height, siteBytes = siteWords * 4;\n  const pattern = createPackedFixedPattern(plan);\n  // Fixed-pattern values are evaluated once, in float64 (exactly the reference's values), before any dispatch.\n  const patterns = pattern === null ? null : new Float64Array(plan.siteCount * 3);\n  if (patterns !== null) {\n    let maximumOffset = 0, maximumOffsetCodes = 0;\n    for (let row = 0, i = 0; row < height; row++) for (let column = 0; column < width; column++, i++) {\n      const p = pattern!(column, row), offsetCodes = p.readoutOffsetElectrons / gain;\n      patterns[i * 3] = p.photoResponseGainFactor; patterns[i * 3 + 1] = p.darkCurrentFactor; patterns[i * 3 + 2] = p.readoutOffsetElectrons;\n      maximumOffset = Math.max(maximumOffset, Math.abs(p.readoutOffsetElectrons)); maximumOffsetCodes = Math.max(maximumOffsetCodes, Math.abs(offsetCodes));\n    }\n    if (!(maximumOffset <= SENSOR_GPU_MAXIMUM_OFFSET_ELECTRONS && maximumOffsetCodes <= SENSOR_GPU_MAXIMUM_OFFSET_CODES)) {\n      throw unsupported(\"Fixed-pattern readout offsets are outside the float32 sensor domain.\");\n    }\n    setFixedPatternSlack(params, threshold, maximumOffset, maximumOffsetCodes);\n  }\n  const rows = Math.max(1, Math.min(height, Math.floor(bandBytes / (width * siteBytes))));\n  if (rows * width * siteBytes > runtime.maximumStorageBindingBytes) {\n    throw new ImageKernelExecutionError(\"insufficient-limits\", \"One sensor band exceeds the device's storage binding limit.\");\n  }\n  const pipeline = await pipelines(\"photivra-sensor-raw\", SENSOR_RAW_WGSL);\n  const codes = new Uint16Array(plan.siteCount), counts = [0, 0, 0, 0];\n  const packed = new ArrayBuffer(rows * width * siteBytes), packedWords = new Uint32Array(packed), packedFloats = new Float32Array(packed);\n  const packedInts = new Int32Array(packed);\n  const words = new Uint32Array(params);\n  let flagged = 0;\n  const flaggedIndices: number[] = [];\n  const scope = new GpuResourceScope(runtime);\n  try {\n    const auxBuffer = scope.upload(\"photivra-sensor-aux\", aux, BUFFER_USAGE.STORAGE);\n    const gate = createHostYieldGate();\n    for (let row0 = 0; row0 < height; row0 += rows) {\n      checkAbort(abort.signal);\n      const row1 = Math.min(height, row0 + rows), count = (row1 - row0) * width, first = row0 * width;\n      for (let i = 0; i < count; i++) {\n        const site = first + i, at = i * siteWords;\n        // The reference's own float64 products of the means and the fixed-pattern factors.\n        packMean(patterns === null ? photo[site]! : photo[site]! * patterns[site * 3]!, packedWords, packedFloats, at);\n        packMean(patterns === null ? dark[site]! : dark[site]! * patterns[site * 3 + 1]!, packedWords, packedFloats, at + 2);\n        let cursor = at + 4;\n        for (const k of poissonComponents) { packMean(additional[k]![site]!, packedWords, packedFloats, cursor); cursor += 2; }\n        if (deterministicComponents.length > 0) {\n          let sum = 0;\n          for (const k of deterministicComponents) sum += additional[k]![site]!;\n          if (!(sum < SENSOR_GPU_MAXIMUM_MEAN)) throw unsupported(\"Deterministic charge of 2^30 electrons and above is outside the float32 sensor domain.\");\n          packSplit(sum, packedInts, packedFloats, cursor);\n          cursor += 2;\n        }\n        if (patterns !== null) {\n          const offset = patterns[site * 3 + 2]!, limit = threshold - offset;\n          if (limit >= 2 ** 30) { packedInts[cursor] = 2 ** 30; packedFloats[cursor + 1] = 0; } else packSplit(limit, packedInts, packedFloats, cursor);\n          packSplit(offset / gain, packedInts, packedFloats, cursor + 2);\n        }\n      }\n      words[0] = count; words[1] = row0;\n      const band = new GpuResourceScope(runtime);\n      try {\n        const means = band.upload(\"photivra-sensor-means\", new Uint32Array(packed, 0, count * siteWords), BUFFER_USAGE.STORAGE);\n        const out = band.buffer(\"photivra-sensor-codes\", count * 4, BUFFER_USAGE.STORAGE | BUFFER_USAGE.COPY_SRC);\n        const uniform = band.upload(\"photivra-sensor-params\", new Uint8Array(params), BUFFER_USAGE.UNIFORM);\n        const bindGroup = runtime.device.createBindGroup({ layout: pipeline.getBindGroupLayout(0), entries: [\n          { binding: 0, resource: { buffer: uniform } }, { binding: 1, resource: { buffer: means } },\n          { binding: 2, resource: { buffer: auxBuffer } }, { binding: 3, resource: { buffer: out } }] });\n        const [data] = await runComputePass(runtime, band, pipeline, bindGroup, linearDispatch(count, SENSOR_WORKGROUP_SIZE, runtime.maximumWorkgroupsPerDimension),\n          [{ buffer: out, bytes: count * 4 }], abort.signal);\n        const result = new Uint32Array(data!, 0, count);\n        for (let i = 0; i < count; i++) {\n          const word = result[i]!, flags = word >>> SENSOR_OUTPUT.flagShift;\n          codes[first + i] = word & SENSOR_OUTPUT.codeMask;\n          if (flags & 1) counts[0]!++;\n          if (flags & 2) counts[1]!++;\n          if (flags & 4) counts[2]!++;\n          if (flags & 8) counts[3]!++;\n          if (word & SENSOR_OUTPUT.uncertainBit) { flagged++; flaggedIndices.push(first + i); }\n        }\n      } finally {\n        band.release();\n      }\n      if (yieldControl && gate.due()) { await yieldControl(abort.signal); gate.restart(); }\n    }\n  } catch (error) {\n    codes.fill(0);\n    throw error;\n  } finally {\n    scope.release();\n  }\n  checkAbort(abort.signal);\n  const execution = webGpuExecution({ kind: \"exact-except-flagged-sites\", flaggedSiteCount: flagged, noiseModel: { ...plan.noiseModel },\n    seedSchedule: plan.seedSchedule }, flagged);\n  return { plan, codes, diagnostics: packedSensorDiagnostics(counts), execution, flaggedSiteIndices: Uint32Array.from(flaggedIndices) };\n}\n"]}

package/dist/gpu/temporal-sum-gpu.d.ts

@@ -0,0 +1,23 @@
+import { type TemporalExposureSumInput, type TemporalExposureSumPlan } from "../motion/temporal-exposure-sum.js";
+import { type ImageKernelExecution, type ImageKernelRelativeErrorBound } from "./image-kernel-execution.js";
+import { type WebGpuRuntime } from "./webgpu-runtime.js";
+import type { TemporalExposureSumResult } from "./image-kernel-accelerator.js";
+type PipelineSource = (label: string, code: string) => Promise<GPUComputePipeline>;
+type YieldControl = ((signal: AbortSignal) => Promise<void>) | undefined;
+export declare const TEMPORAL_WORKGROUP_EDGE = 16;
+/**
+ * A-priori float32 bound for every output (all terms are non-negative): each term carries at most K + 7
+ * roundings (sample, bilinear weight and their product, three inner additions, row weight and its
+ * product, K − 1 outer additions), so |v_f32 − v_ref| ≤ (γ_{K+8} + 2^-40)·v_ref + K·(8·w_max + 2)·2^-126 + R.
+ * The 2^-40 covers the float64 reference's own rounding; the 2^-126 term covers subnormal flush-to-zero of
+ * samples, products and sums. R covers weights below the float32 normal range, which the host conversion may round
+ * to zero or a subnormal (|f32(w) − w| ≤ w) and the GPU may flush, putting their whole term at risk:
+ * R = 2·S_max·Σ_k (max_y ρ_k(y) + w_max·Σ_j β_{k,j}), where ρ_k(y) and β_{k,j} are the row and bilinear weights in
+ * (0, 2^-126) (zero otherwise) and the factor 2 covers the conversion and the later roundings. R = 0 when no such
+ * weight exists.
+ */
+export declare function temporalExposureSumErrorBound(plan: TemporalExposureSumPlan, rowWeights: Float32Array | Float64Array): ImageKernelRelativeErrorBound;
+export declare const TEMPORAL_SUM_WGSL = "\nstruct Params { width: u32, height: u32, nodes: u32, out_row0: u32, out_rows: u32, in_row0: u32, pad0: u32, pad1: u32 };\nstruct Node { ox: i32, oy: i32, pad0: i32, pad1: i32, b: vec4<f32> };\n@group(0) @binding(0) var<uniform> params: Params;\n@group(0) @binding(1) var<storage, read> source: array<f32>;\n@group(0) @binding(2) var<storage, read> nodes: array<Node>;\n@group(0) @binding(3) var<storage, read> weights: array<f32>;\n@group(0) @binding(4) var<storage, read_write> result: array<f32>;\nfn sample(sx: i32, sy: i32) -> f32 { return source[u32(sy - i32(params.in_row0)) * params.width + u32(sx)]; }\n@compute @workgroup_size(16, 16)\nfn main(@builtin(global_invocation_id) gid: vec3<u32>) {\n  let x = i32(gid.x);\n  let local_y = gid.y;\n  if (gid.x >= params.width || local_y >= params.out_rows) { return; }\n  let y = i32(params.out_row0 + local_y);\n  let w = i32(params.width);\n  let h = i32(params.height);\n  var sum = 0.0;\n  for (var k = 0u; k < params.nodes; k = k + 1u) {\n    let node = nodes[k];\n    let sx = x + node.ox;\n    let sy = y + node.oy;\n    let left = sx >= 0 && sx < w;\n    let right = sx + 1 >= 0 && sx + 1 < w;\n    let top = sy >= 0 && sy < h;\n    let bottom = sy + 1 >= 0 && sy + 1 < h;\n    var inner = 0.0;\n    if (top && left) { inner = inner + node.b.x * sample(sx, sy); }\n    if (top && right) { inner = inner + node.b.y * sample(sx + 1, sy); }\n    if (bottom && left) { inner = inner + node.b.z * sample(sx, sy + 1); }\n    if (bottom && right) { inner = inner + node.b.w * sample(sx + 1, sy + 1); }\n    sum = sum + weights[k * params.out_rows + local_y] * inner;\n  }\n  result[local_y * params.width + gid.x] = sum;\n}\n";
+export declare function sumTemporalExposureOnCpu(plan: TemporalExposureSumPlan, input: TemporalExposureSumInput, abort: AbortController, yieldControl: YieldControl, execution: ImageKernelExecution): Promise<TemporalExposureSumResult>;
+export declare function sumTemporalExposureOnGpu(runtime: WebGpuRuntime, bandBytes: number, plan: TemporalExposureSumPlan, input: TemporalExposureSumInput, abort: AbortController, yieldControl: YieldControl, pipelines: PipelineSource): Promise<TemporalExposureSumResult>;
+export {};

package/dist/gpu/temporal-sum-gpu.js

@@ -0,0 +1,150 @@
+// SPDX-License-Identifier: Apache-2.0
+/**
+ * `webgpu-f32` and CPU-reference execution of the temporal exposure sum (#277, step 3). The equation
+ * and summation order are those of `sumTemporalExposureRows` (motion/temporal-exposure-sum.ts).
+ * Internal; the public entry point is the image-kernel accelerator.
+ */
+import { createHostYieldGate } from "../capture/host-yield-cadence.js";
+import { sumTemporalExposureRows } from "../motion/temporal-exposure-sum.js";
+import { ImageKernelExecutionError, float32Gamma, webGpuExecution } from "./image-kernel-execution.js";
+import { BUFFER_USAGE, GpuResourceScope, abortedError, runComputePass } from "./webgpu-runtime.js";
+export const TEMPORAL_WORKGROUP_EDGE = 16;
+const MAXIMUM_TAPS_PER_DISPATCH = 2 ** 30;
+const FLOAT32_SUBNORMAL_LIMIT = 2 ** -126;
+/**
+ * A-priori float32 bound for every output (all terms are non-negative): each term carries at most K + 7
+ * roundings (sample, bilinear weight and their product, three inner additions, row weight and its
+ * product, K − 1 outer additions), so |v_f32 − v_ref| ≤ (γ_{K+8} + 2^-40)·v_ref + K·(8·w_max + 2)·2^-126 + R.
+ * The 2^-40 covers the float64 reference's own rounding; the 2^-126 term covers subnormal flush-to-zero of
+ * samples, products and sums. R covers weights below the float32 normal range, which the host conversion may round
+ * to zero or a subnormal (|f32(w) − w| ≤ w) and the GPU may flush, putting their whole term at risk:
+ * R = 2·S_max·Σ_k (max_y ρ_k(y) + w_max·Σ_j β_{k,j}), where ρ_k(y) and β_{k,j} are the row and bilinear weights in
+ * (0, 2^-126) (zero otherwise) and the factor 2 covers the conversion and the later roundings. R = 0 when no such
+ * weight exists.
+ */
+export function temporalExposureSumErrorBound(plan, rowWeights) {
+    const tiny = (w) => (w < FLOAT32_SUBNORMAL_LIMIT ? w : 0);
+    let atRisk = 0;
+    for (let k = 0; k < plan.nodeCount; k++) {
+        let largestRow = 0;
+        for (let y = 0; y < plan.height; y++)
+            largestRow = Math.max(largestRow, tiny(rowWeights[k * plan.height + y]));
+        atRisk += largestRow + plan.maximumWeight * plan.nodes[k].weights.reduce((sum, b) => sum + tiny(b), 0);
+    }
+    return { kind: "relative-to-reference", relative: float32Gamma(plan.nodeCount + 8) + 2 ** -40,
+        absolute: plan.nodeCount * (8 * plan.maximumWeight + 2) * FLOAT32_SUBNORMAL_LIMIT + 2 * plan.maximumSampleMagnitude * atRisk };
+}
+export const TEMPORAL_SUM_WGSL = /* wgsl */ `
+struct Params { width: u32, height: u32, nodes: u32, out_row0: u32, out_rows: u32, in_row0: u32, pad0: u32, pad1: u32 };
+struct Node { ox: i32, oy: i32, pad0: i32, pad1: i32, b: vec4<f32> };
+@group(0) @binding(0) var<uniform> params: Params;
+@group(0) @binding(1) var<storage, read> source: array<f32>;
+@group(0) @binding(2) var<storage, read> nodes: array<Node>;
+@group(0) @binding(3) var<storage, read> weights: array<f32>;
+@group(0) @binding(4) var<storage, read_write> result: array<f32>;
+fn sample(sx: i32, sy: i32) -> f32 { return source[u32(sy - i32(params.in_row0)) * params.width + u32(sx)]; }
+@compute @workgroup_size(${TEMPORAL_WORKGROUP_EDGE}, ${TEMPORAL_WORKGROUP_EDGE})
+fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
+  let x = i32(gid.x);
+  let local_y = gid.y;
+  if (gid.x >= params.width || local_y >= params.out_rows) { return; }
+  let y = i32(params.out_row0 + local_y);
+  let w = i32(params.width);
+  let h = i32(params.height);
+  var sum = 0.0;
+  for (var k = 0u; k < params.nodes; k = k + 1u) {
+    let node = nodes[k];
+    let sx = x + node.ox;
+    let sy = y + node.oy;
+    let left = sx >= 0 && sx < w;
+    let right = sx + 1 >= 0 && sx + 1 < w;
+    let top = sy >= 0 && sy < h;
+    let bottom = sy + 1 >= 0 && sy + 1 < h;
+    var inner = 0.0;
+    if (top && left) { inner = inner + node.b.x * sample(sx, sy); }
+    if (top && right) { inner = inner + node.b.y * sample(sx + 1, sy); }
+    if (bottom && left) { inner = inner + node.b.z * sample(sx, sy + 1); }
+    if (bottom && right) { inner = inner + node.b.w * sample(sx + 1, sy + 1); }
+    sum = sum + weights[k * params.out_rows + local_y] * inner;
+  }
+  result[local_y * params.width + gid.x] = sum;
+}
+`;
+function checkAbort(abort) {
+    if (abort.signal.aborted)
+        throw abortedError();
+}
+export async function sumTemporalExposureOnCpu(plan, input, abort, yieldControl, execution) {
+    const samples = input.samples.slice(), weights = input.rowWeights.slice();
+    const values = new Float64Array(plan.width * plan.height);
+    const rowsPerChunk = Math.max(1, Math.floor(2_000_000 / Math.max(1, plan.width * plan.nodeCount * 4))), gate = createHostYieldGate();
+    for (let y = 0; y < plan.height; y += rowsPerChunk) {
+        checkAbort(abort);
+        const end = Math.min(plan.height, y + rowsPerChunk);
+        sumTemporalExposureRows(plan, samples, weights, y, end, values.subarray(y * plan.width, end * plan.width));
+        if (yieldControl && gate.due()) {
+            await yieldControl(abort.signal);
+            gate.restart();
+        }
+    }
+    checkAbort(abort);
+    return { plan, values, execution };
+}
+export async function sumTemporalExposureOnGpu(runtime, bandBytes, plan, input, abort, yieldControl, pipelines) {
+    const { width, height, nodes } = plan;
+    const errorBound = temporalExposureSumErrorBound(plan, input.rowWeights);
+    const samples = new Float32Array(input.samples), weights = new Float32Array(input.rowWeights);
+    const minimumOffset = Math.min(...nodes.map(n => n.offsetY)), maximumOffset = Math.max(...nodes.map(n => n.offsetY)) + 1;
+    const span = maximumOffset - minimumOffset, rowBytes = width * 4;
+    let rows = Math.min(height, Math.floor(bandBytes / rowBytes) - span, Math.floor(MAXIMUM_TAPS_PER_DISPATCH / (width * plan.nodeCount * 4)));
+    if (rows < 1) {
+        if ((1 + span) * rowBytes > runtime.maximumStorageBindingBytes) {
+            throw new ImageKernelExecutionError("insufficient-limits", "One temporal band exceeds the device's storage binding limit.");
+        }
+        rows = 1;
+    }
+    const nodeBuffer = new ArrayBuffer(nodes.length * 32), nodeInts = new Int32Array(nodeBuffer), nodeFloats = new Float32Array(nodeBuffer);
+    nodes.forEach((n, k) => { nodeInts[k * 8] = n.offsetX; nodeInts[k * 8 + 1] = n.offsetY; nodeFloats.set(n.weights, k * 8 + 4); });
+    const values = new Float32Array(width * height);
+    const pipeline = await pipelines("photivra-temporal-sum", TEMPORAL_SUM_WGSL);
+    const scope = new GpuResourceScope(runtime);
+    try {
+        const nodeData = scope.upload("photivra-temporal-nodes", new Uint8Array(nodeBuffer), BUFFER_USAGE.STORAGE);
+        const gate = createHostYieldGate();
+        for (let y0 = 0; y0 < height; y0 += rows) {
+            checkAbort(abort);
+            const y1 = Math.min(height, y0 + rows), count = y1 - y0;
+            const in0 = Math.max(0, Math.min(height, y0 + minimumOffset)), in1 = Math.max(in0, Math.min(height, y1 - 1 + maximumOffset + 1));
+            const bandWeights = new Float32Array(nodes.length * count);
+            for (let k = 0; k < nodes.length; k++)
+                bandWeights.set(weights.subarray(k * height + y0, k * height + y1), k * count);
+            const band = new GpuResourceScope(runtime);
+            try {
+                const source = band.upload("photivra-temporal-band", samples.subarray(in0 * width, in1 * width), BUFFER_USAGE.STORAGE);
+                const rowWeights = band.upload("photivra-temporal-weights", bandWeights, BUFFER_USAGE.STORAGE);
+                const result = band.buffer("photivra-temporal-result", count * rowBytes, BUFFER_USAGE.STORAGE | BUFFER_USAGE.COPY_SRC);
+                const uniform = band.upload("photivra-temporal-params", new Uint32Array([width, height, nodes.length, y0, count, in0, 0, 0]), BUFFER_USAGE.UNIFORM);
+                const bindGroup = runtime.device.createBindGroup({ layout: pipeline.getBindGroupLayout(0), entries: [
+                        { binding: 0, resource: { buffer: uniform } }, { binding: 1, resource: { buffer: source } }, { binding: 2, resource: { buffer: nodeData } },
+                        { binding: 3, resource: { buffer: rowWeights } }, { binding: 4, resource: { buffer: result } }
+                    ] });
+                const groups = [Math.ceil(width / TEMPORAL_WORKGROUP_EDGE), Math.ceil(count / TEMPORAL_WORKGROUP_EDGE)];
+                const [data] = await runComputePass(runtime, band, pipeline, bindGroup, groups, [{ buffer: result, bytes: count * rowBytes }], abort.signal);
+                values.set(new Float32Array(data, 0, count * width), y0 * width);
+            }
+            finally {
+                band.release();
+            }
+            if (yieldControl && gate.due()) {
+                await yieldControl(abort.signal);
+                gate.restart();
+            }
+        }
+    }
+    finally {
+        scope.release();
+    }
+    checkAbort(abort);
+    return { plan, values, execution: webGpuExecution(errorBound, 0) };
+}
+//# sourceMappingURL=temporal-sum-gpu.js.map

package/dist/gpu/temporal-sum-gpu.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"temporal-sum-gpu.js","sourceRoot":"","sources":["../../src/gpu/temporal-sum-gpu.ts"],"names":[],"mappings":"AAAA,sCAAsC;AAEtC;;;;GAIG;AACH,OAAO,EAAE,mBAAmB,EAAE,MAAM,kCAAkC,CAAC;AACvE,OAAO,EAAE,uBAAuB,EAA+D,MAAM,oCAAoC,CAAC;AAC1I,OAAO,EAAE,yBAAyB,EAAE,YAAY,EAAE,eAAe,EAAiE,MAAM,6BAA6B,CAAC;AACtK,OAAO,EAAE,YAAY,EAAE,gBAAgB,EAAE,YAAY,EAAE,cAAc,EAAsB,MAAM,qBAAqB,CAAC;AAMvH,MAAM,CAAC,MAAM,uBAAuB,GAAG,EAAE,CAAC;AAC1C,MAAM,yBAAyB,GAAG,CAAC,IAAI,EAAE,CAAC;AAC1C,MAAM,uBAAuB,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC;AAE1C;;;;;;;;;;GAUG;AACH,MAAM,UAAU,6BAA6B,CAAC,IAA6B,EAAE,UAAuC;IAClH,MAAM,IAAI,GAAG,CAAC,CAAS,EAAU,EAAE,CAAC,CAAC,CAAC,GAAG,uBAAuB,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1E,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,SAAS,EAAE,CAAC,EAAE,EAAE,CAAC;QACxC,IAAI,UAAU,GAAG,CAAC,CAAC;QACnB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE;YAAE,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,IAAI,CAAC,UAAU,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,CAAE,CAAC,CAAC,CAAC;QAChH,MAAM,IAAI,UAAU,GAAG,IAAI,CAAC,aAAa,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,CAAE,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAC1G,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,uBAAuB,EAAE,QAAQ,EAAE,YAAY,CAAC,IAAI,CAAC,SAAS,GAAG,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE;QAC3F,QAAQ,EAAE,IAAI,CAAC,SAAS,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,aAAa,GAAG,CAAC,CAAC,GAAG,uBAAuB,GAAG,CAAC,GAAG,IAAI,CAAC,sBAAsB,GAAG,MAAM,EAAE,CAAC;AACnI,CAAC;AAED,MAAM,CAAC,MAAM,iBAAiB,GAAG,UAAU,CAAC;;;;;;;;;2BASjB,uBAAuB,KAAK,uBAAuB;;;;;;;;;;;;;;;;;;;;;;;;;;CA0B7E,CAAC;AAEF,SAAS,UAAU,CAAC,KAAsB;IACxC,IAAI,KAAK,CAAC,MAAM,CAAC,OAAO;QAAE,MAAM,YAAY,EAAE,CAAC;AACjD,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,wBAAwB,CAAC,IAA6B,EAAE,KAA+B,EAAE,KAAsB,EACnI,YAA0B,EAAE,SAA+B;IAC3D,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,OAAO,GAAG,KAAK,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC;IAC1E,MAAM,MAAM,GAAG,IAAI,YAAY,CAAC,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC;IAC1D,MAAM,YAAY,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,GAAG,mBAAmB,EAAE,CAAC;IACrI,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,IAAI,YAAY,EAAE,CAAC;QACnD,UAAU,CAAC,KAAK,CAAC,CAAC;QAClB,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC,GAAG,YAAY,CAAC,CAAC;QACpD,uBAAuB,CAAC,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,GAAG,EAAE,MAAM,CAAC,QAAQ,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,EAAE,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;QAC3G,IAAI,YAAY,IAAI,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC;YAAC,MAAM,YAAY,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;YAAC,IAAI,CAAC,OAAO,EAAE,CAAC;QAAC,CAAC;IACvF,CAAC;IACD,UAAU,CAAC,KAAK,CAAC,CAAC;IAClB,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC;AACrC,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,wBAAwB,CAAC,OAAsB,EAAE,SAAiB,EAAE,IAA6B,EAAE,KAA+B,EACtJ,KAAsB,EAAE,YAA0B,EAAE,SAAyB;IAC7E,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,GAAG,IAAI,CAAC;IACtC,MAAM,UAAU,GAAG,6BAA6B,CAAC,IAAI,EAAE,KAAK,CAAC,UAAU,CAAC,CAAC;IACzE,MAAM,OAAO,GAAG,IAAI,YAAY,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,OAAO,GAAG,IAAI,YAAY,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;IAC9F,MAAM,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,EAAE,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC;IACzH,MAAM,IAAI,GAAG,aAAa,GAAG,aAAa,EAAE,QAAQ,GAAG,KAAK,GAAG,CAAC,CAAC;IACjE,IAAI,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,IAAI,CAAC,KAAK,CAAC,SAAS,GAAG,QAAQ,CAAC,GAAG,IAAI,EAAE,IAAI,CAAC,KAAK,CAAC,yBAAyB,GAAG,CAAC,KAAK,GAAG,IAAI,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3I,IAAI,IAAI,GAAG,CAAC,EAAE,CAAC;QACb,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,QAAQ,GAAG,OAAO,CAAC,0BAA0B,EAAE,CAAC;YAC/D,MAAM,IAAI,yBAAyB,CAAC,qBAAqB,EAAE,+DAA+D,CAAC,CAAC;QAC9H,CAAC;QACD,IAAI,GAAG,CAAC,CAAC;IACX,CAAC;IACD,MAAM,UAAU,GAAG,IAAI,WAAW,CAAC,KAAK,CAAC,MAAM,GAAG,EAAE,CAAC,EAAE,QAAQ,GAAG,IAAI,UAAU,CAAC,UAAU,CAAC,EAAE,UAAU,GAAG,IAAI,YAAY,CAAC,UAAU,CAAC,CAAC;IACxI,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,GAAG,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACjI,MAAM,MAAM,GAAG,IAAI,YAAY,CAAC,KAAK,GAAG,MAAM,CAAC,CAAC;IAChD,MAAM,QAAQ,GAAG,MAAM,SAAS,CAAC,uBAAuB,EAAE,iBAAiB,CAAC,CAAC;IAC7E,MAAM,KAAK,GAAG,IAAI,gBAAgB,CAAC,OAAO,CAAC,CAAC;IAC5C,IAAI,CAAC;QACH,MAAM,QAAQ,GAAG,KAAK,CAAC,MAAM,CAAC,yBAAyB,EAAE,IAAI,UAAU,CAAC,UAAU,CAAC,EAAE,YAAY,CAAC,OAAO,CAAC,CAAC;QAC3G,MAAM,IAAI,GAAG,mBAAmB,EAAE,CAAC;QACnC,KAAK,IAAI,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,MAAM,EAAE,EAAE,IAAI,IAAI,EAAE,CAAC;YACzC,UAAU,CAAC,KAAK,CAAC,CAAC;YAClB,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,GAAG,IAAI,CAAC,EAAE,KAAK,GAAG,EAAE,GAAG,EAAE,CAAC;YACxD,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,GAAG,aAAa,CAAC,CAAC,EAAE,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,EAAE,GAAG,CAAC,GAAG,aAAa,GAAG,CAAC,CAAC,CAAC,CAAC;YACjI,MAAM,WAAW,GAAG,IAAI,YAAY,CAAC,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,CAAC;YAC3D,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE;gBAAE,WAAW,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,MAAM,GAAG,EAAE,EAAE,CAAC,GAAG,MAAM,GAAG,EAAE,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,CAAC;YACtH,MAAM,IAAI,GAAG,IAAI,gBAAgB,CAAC,OAAO,CAAC,CAAC;YAC3C,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,wBAAwB,EAAE,OAAO,CAAC,QAAQ,CAAC,GAAG,GAAG,KAAK,EAAE,GAAG,GAAG,KAAK,CAAC,EAAE,YAAY,CAAC,OAAO,CAAC,CAAC;gBACvH,MAAM,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,2BAA2B,EAAE,WAAW,EAAE,YAAY,CAAC,OAAO,CAAC,CAAC;gBAC/F,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,0BAA0B,EAAE,KAAK,GAAG,QAAQ,EAAE,YAAY,CAAC,OAAO,GAAG,YAAY,CAAC,QAAQ,CAAC,CAAC;gBACvH,MAAM,OAAO,GAAG,IAAI,CAAC,MAAM,CAAC,0BAA0B,EAAE,IAAI,WAAW,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,YAAY,CAAC,OAAO,CAAC,CAAC;gBACpJ,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,eAAe,CAAC,EAAE,MAAM,EAAE,QAAQ,CAAC,kBAAkB,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE;wBAClG,EAAE,OAAO,EAAE,CAAC,EAAE,QAAQ,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,EAAE,EAAE,OAAO,EAAE,CAAC,EAAE,QAAQ,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,EAAE,EAAE,EAAE,OAAO,EAAE,CAAC,EAAE,QAAQ,EAAE,EAAE,MAAM,EAAE,QAAQ,EAAE,EAAE;wBAC3I,EAAE,OAAO,EAAE,CAAC,EAAE,QAAQ,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,EAAE,EAAE,EAAE,OAAO,EAAE,CAAC,EAAE,QAAQ,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,EAAE;qBAAC,EAAE,CAAC,CAAC;gBACrG,MAAM,MAAM,GAAqB,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,uBAAuB,CAAC,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,uBAAuB,CAAC,CAAC,CAAC;gBAC1H,MAAM,CAAC,IAAI,CAAC,GAAG,MAAM,cAAc,CAAC,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,GAAG,QAAQ,EAAE,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;gBAC7I,MAAM,CAAC,GAAG,CAAC,IAAI,YAAY,CAAC,IAAK,EAAE,CAAC,EAAE,KAAK,GAAG,KAAK,CAAC,EAAE,EAAE,GAAG,KAAK,CAAC,CAAC;YACpE,CAAC;oBAAS,CAAC;gBACT,IAAI,CAAC,OAAO,EAAE,CAAC;YACjB,CAAC;YACD,IAAI,YAAY,IAAI,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC;gBAAC,MAAM,YAAY,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;gBAAC,IAAI,CAAC,OAAO,EAAE,CAAC;YAAC,CAAC;QACvF,CAAC;IACH,CAAC;YAAS,CAAC;QACT,KAAK,CAAC,OAAO,EAAE,CAAC;IAClB,CAAC;IACD,UAAU,CAAC,KAAK,CAAC,CAAC;IAClB,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,eAAe,CAAC,UAAU,EAAE,CAAC,CAAC,EAAE,CAAC;AACrE,CAAC","sourcesContent":["// SPDX-License-Identifier: Apache-2.0\n\n/**\n * `webgpu-f32` and CPU-reference execution of the temporal exposure sum (#277, step 3). The equation\n * and summation order are those of `sumTemporalExposureRows` (motion/temporal-exposure-sum.ts).\n * Internal; the public entry point is the image-kernel accelerator.\n */\nimport { createHostYieldGate } from \"../capture/host-yield-cadence.js\";\nimport { sumTemporalExposureRows, type TemporalExposureSumInput, type TemporalExposureSumPlan } from \"../motion/temporal-exposure-sum.js\";\nimport { ImageKernelExecutionError, float32Gamma, webGpuExecution, type ImageKernelExecution, type ImageKernelRelativeErrorBound } from \"./image-kernel-execution.js\";\nimport { BUFFER_USAGE, GpuResourceScope, abortedError, runComputePass, type WebGpuRuntime } from \"./webgpu-runtime.js\";\nimport type { TemporalExposureSumResult } from \"./image-kernel-accelerator.js\";\n\ntype PipelineSource = (label: string, code: string) => Promise<GPUComputePipeline>;\ntype YieldControl = ((signal: AbortSignal) => Promise<void>) | undefined;\n\nexport const TEMPORAL_WORKGROUP_EDGE = 16;\nconst MAXIMUM_TAPS_PER_DISPATCH = 2 ** 30;\nconst FLOAT32_SUBNORMAL_LIMIT = 2 ** -126;\n\n/**\n * A-priori float32 bound for every output (all terms are non-negative): each term carries at most K + 7\n * roundings (sample, bilinear weight and their product, three inner additions, row weight and its\n * product, K − 1 outer additions), so |v_f32 − v_ref| ≤ (γ_{K+8} + 2^-40)·v_ref + K·(8·w_max + 2)·2^-126 + R.\n * The 2^-40 covers the float64 reference's own rounding; the 2^-126 term covers subnormal flush-to-zero of\n * samples, products and sums. R covers weights below the float32 normal range, which the host conversion may round\n * to zero or a subnormal (|f32(w) − w| ≤ w) and the GPU may flush, putting their whole term at risk:\n * R = 2·S_max·Σ_k (max_y ρ_k(y) + w_max·Σ_j β_{k,j}), where ρ_k(y) and β_{k,j} are the row and bilinear weights in\n * (0, 2^-126) (zero otherwise) and the factor 2 covers the conversion and the later roundings. R = 0 when no such\n * weight exists.\n */\nexport function temporalExposureSumErrorBound(plan: TemporalExposureSumPlan, rowWeights: Float32Array | Float64Array): ImageKernelRelativeErrorBound {\n  const tiny = (w: number): number => (w < FLOAT32_SUBNORMAL_LIMIT ? w : 0);\n  let atRisk = 0;\n  for (let k = 0; k < plan.nodeCount; k++) {\n    let largestRow = 0;\n    for (let y = 0; y < plan.height; y++) largestRow = Math.max(largestRow, tiny(rowWeights[k * plan.height + y]!));\n    atRisk += largestRow + plan.maximumWeight * plan.nodes[k]!.weights.reduce((sum, b) => sum + tiny(b), 0);\n  }\n  return { kind: \"relative-to-reference\", relative: float32Gamma(plan.nodeCount + 8) + 2 ** -40,\n    absolute: plan.nodeCount * (8 * plan.maximumWeight + 2) * FLOAT32_SUBNORMAL_LIMIT + 2 * plan.maximumSampleMagnitude * atRisk };\n}\n\nexport const TEMPORAL_SUM_WGSL = /* wgsl */ `\nstruct Params { width: u32, height: u32, nodes: u32, out_row0: u32, out_rows: u32, in_row0: u32, pad0: u32, pad1: u32 };\nstruct Node { ox: i32, oy: i32, pad0: i32, pad1: i32, b: vec4<f32> };\n@group(0) @binding(0) var<uniform> params: Params;\n@group(0) @binding(1) var<storage, read> source: array<f32>;\n@group(0) @binding(2) var<storage, read> nodes: array<Node>;\n@group(0) @binding(3) var<storage, read> weights: array<f32>;\n@group(0) @binding(4) var<storage, read_write> result: array<f32>;\nfn sample(sx: i32, sy: i32) -> f32 { return source[u32(sy - i32(params.in_row0)) * params.width + u32(sx)]; }\n@compute @workgroup_size(${TEMPORAL_WORKGROUP_EDGE}, ${TEMPORAL_WORKGROUP_EDGE})\nfn main(@builtin(global_invocation_id) gid: vec3<u32>) {\n  let x = i32(gid.x);\n  let local_y = gid.y;\n  if (gid.x >= params.width || local_y >= params.out_rows) { return; }\n  let y = i32(params.out_row0 + local_y);\n  let w = i32(params.width);\n  let h = i32(params.height);\n  var sum = 0.0;\n  for (var k = 0u; k < params.nodes; k = k + 1u) {\n    let node = nodes[k];\n    let sx = x + node.ox;\n    let sy = y + node.oy;\n    let left = sx >= 0 && sx < w;\n    let right = sx + 1 >= 0 && sx + 1 < w;\n    let top = sy >= 0 && sy < h;\n    let bottom = sy + 1 >= 0 && sy + 1 < h;\n    var inner = 0.0;\n    if (top && left) { inner = inner + node.b.x * sample(sx, sy); }\n    if (top && right) { inner = inner + node.b.y * sample(sx + 1, sy); }\n    if (bottom && left) { inner = inner + node.b.z * sample(sx, sy + 1); }\n    if (bottom && right) { inner = inner + node.b.w * sample(sx + 1, sy + 1); }\n    sum = sum + weights[k * params.out_rows + local_y] * inner;\n  }\n  result[local_y * params.width + gid.x] = sum;\n}\n`;\n\nfunction checkAbort(abort: AbortController): void {\n  if (abort.signal.aborted) throw abortedError();\n}\n\nexport async function sumTemporalExposureOnCpu(plan: TemporalExposureSumPlan, input: TemporalExposureSumInput, abort: AbortController,\n  yieldControl: YieldControl, execution: ImageKernelExecution): Promise<TemporalExposureSumResult> {\n  const samples = input.samples.slice(), weights = input.rowWeights.slice();\n  const values = new Float64Array(plan.width * plan.height);\n  const rowsPerChunk = Math.max(1, Math.floor(2_000_000 / Math.max(1, plan.width * plan.nodeCount * 4))), gate = createHostYieldGate();\n  for (let y = 0; y < plan.height; y += rowsPerChunk) {\n    checkAbort(abort);\n    const end = Math.min(plan.height, y + rowsPerChunk);\n    sumTemporalExposureRows(plan, samples, weights, y, end, values.subarray(y * plan.width, end * plan.width));\n    if (yieldControl && gate.due()) { await yieldControl(abort.signal); gate.restart(); }\n  }\n  checkAbort(abort);\n  return { plan, values, execution };\n}\n\nexport async function sumTemporalExposureOnGpu(runtime: WebGpuRuntime, bandBytes: number, plan: TemporalExposureSumPlan, input: TemporalExposureSumInput,\n  abort: AbortController, yieldControl: YieldControl, pipelines: PipelineSource): Promise<TemporalExposureSumResult> {\n  const { width, height, nodes } = plan;\n  const errorBound = temporalExposureSumErrorBound(plan, input.rowWeights);\n  const samples = new Float32Array(input.samples), weights = new Float32Array(input.rowWeights);\n  const minimumOffset = Math.min(...nodes.map(n => n.offsetY)), maximumOffset = Math.max(...nodes.map(n => n.offsetY)) + 1;\n  const span = maximumOffset - minimumOffset, rowBytes = width * 4;\n  let rows = Math.min(height, Math.floor(bandBytes / rowBytes) - span, Math.floor(MAXIMUM_TAPS_PER_DISPATCH / (width * plan.nodeCount * 4)));\n  if (rows < 1) {\n    if ((1 + span) * rowBytes > runtime.maximumStorageBindingBytes) {\n      throw new ImageKernelExecutionError(\"insufficient-limits\", \"One temporal band exceeds the device's storage binding limit.\");\n    }\n    rows = 1;\n  }\n  const nodeBuffer = new ArrayBuffer(nodes.length * 32), nodeInts = new Int32Array(nodeBuffer), nodeFloats = new Float32Array(nodeBuffer);\n  nodes.forEach((n, k) => { nodeInts[k * 8] = n.offsetX; nodeInts[k * 8 + 1] = n.offsetY; nodeFloats.set(n.weights, k * 8 + 4); });\n  const values = new Float32Array(width * height);\n  const pipeline = await pipelines(\"photivra-temporal-sum\", TEMPORAL_SUM_WGSL);\n  const scope = new GpuResourceScope(runtime);\n  try {\n    const nodeData = scope.upload(\"photivra-temporal-nodes\", new Uint8Array(nodeBuffer), BUFFER_USAGE.STORAGE);\n    const gate = createHostYieldGate();\n    for (let y0 = 0; y0 < height; y0 += rows) {\n      checkAbort(abort);\n      const y1 = Math.min(height, y0 + rows), count = y1 - y0;\n      const in0 = Math.max(0, Math.min(height, y0 + minimumOffset)), in1 = Math.max(in0, Math.min(height, y1 - 1 + maximumOffset + 1));\n      const bandWeights = new Float32Array(nodes.length * count);\n      for (let k = 0; k < nodes.length; k++) bandWeights.set(weights.subarray(k * height + y0, k * height + y1), k * count);\n      const band = new GpuResourceScope(runtime);\n      try {\n        const source = band.upload(\"photivra-temporal-band\", samples.subarray(in0 * width, in1 * width), BUFFER_USAGE.STORAGE);\n        const rowWeights = band.upload(\"photivra-temporal-weights\", bandWeights, BUFFER_USAGE.STORAGE);\n        const result = band.buffer(\"photivra-temporal-result\", count * rowBytes, BUFFER_USAGE.STORAGE | BUFFER_USAGE.COPY_SRC);\n        const uniform = band.upload(\"photivra-temporal-params\", new Uint32Array([width, height, nodes.length, y0, count, in0, 0, 0]), BUFFER_USAGE.UNIFORM);\n        const bindGroup = runtime.device.createBindGroup({ layout: pipeline.getBindGroupLayout(0), entries: [\n          { binding: 0, resource: { buffer: uniform } }, { binding: 1, resource: { buffer: source } }, { binding: 2, resource: { buffer: nodeData } },\n          { binding: 3, resource: { buffer: rowWeights } }, { binding: 4, resource: { buffer: result } }] });\n        const groups: [number, number] = [Math.ceil(width / TEMPORAL_WORKGROUP_EDGE), Math.ceil(count / TEMPORAL_WORKGROUP_EDGE)];\n        const [data] = await runComputePass(runtime, band, pipeline, bindGroup, groups, [{ buffer: result, bytes: count * rowBytes }], abort.signal);\n        values.set(new Float32Array(data!, 0, count * width), y0 * width);\n      } finally {\n        band.release();\n      }\n      if (yieldControl && gate.due()) { await yieldControl(abort.signal); gate.restart(); }\n    }\n  } finally {\n    scope.release();\n  }\n  checkAbort(abort);\n  return { plan, values, execution: webGpuExecution(errorBound, 0) };\n}\n"]}