oidn-webgpu 0.0.0-stage → 0.6.0

package/lib/gpuTypes.js

@@ -0,0 +1,27 @@
+/**
+ * The WebGPU subset used by the native WGSL runtime, as structural types and
+ * spec-defined flag values. A browser GPUDevice implements these operations,
+ * but its branded TypeScript types need the conversion in nativeUNet.ts;
+ * they also let an embedder supply its own device (for example one that
+ * records the work for replay in a native wgpu host) without lib.dom or
+ * WebGPU globals.
+ */
+/** GPUBufferUsage flag values from the WebGPU specification. */
+export const BufferUsage = {
+    MAP_READ: 0x0001,
+    MAP_WRITE: 0x0002,
+    COPY_SRC: 0x0004,
+    COPY_DST: 0x0008,
+    INDEX: 0x0010,
+    VERTEX: 0x0020,
+    UNIFORM: 0x0040,
+    STORAGE: 0x0080,
+    INDIRECT: 0x0100,
+    QUERY_RESOLVE: 0x0200
+};
+/** GPUMapMode flag values from the WebGPU specification. */
+export const MapMode = {
+    READ: 0x0001,
+    WRITE: 0x0002
+};
+//# sourceMappingURL=gpuTypes.js.map

package/lib/gpuTypes.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"gpuTypes.js","sourceRoot":"","sources":["../src/gpuTypes.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,gEAAgE;AAChE,MAAM,CAAC,MAAM,WAAW,GAAG;IACzB,QAAQ,EAAE,MAAM;IAChB,SAAS,EAAE,MAAM;IACjB,QAAQ,EAAE,MAAM;IAChB,QAAQ,EAAE,MAAM;IAChB,KAAK,EAAE,MAAM;IACb,MAAM,EAAE,MAAM;IACd,OAAO,EAAE,MAAM;IACf,OAAO,EAAE,MAAM;IACf,QAAQ,EAAE,MAAM;IAChB,aAAa,EAAE,MAAM;CACb,CAAC;AAEX,4DAA4D;AAC5D,MAAM,CAAC,MAAM,OAAO,GAAG;IACrB,IAAI,EAAE,MAAM;IACZ,KAAK,EAAE,MAAM;CACL,CAAC"}

package/lib/graphOptimizer.d.ts

@@ -0,0 +1,54 @@
+import type { Conv2DNodeSpec, MaxPool2DNodeSpec, ModelNodeSpec, UNetModelSpec, Upsample2DNodeSpec, UNetModelGraph } from './modelSpec';
+export interface FusedConvPoolNodeSpec {
+    op: 'fusedConvReluMaxPool2d';
+    id: string;
+    input: string;
+    conv: Conv2DNodeSpec;
+    pool: MaxPool2DNodeSpec;
+}
+export interface FusedUpsampleConcatConvNodeSpec {
+    op: 'fusedUpsampleConcatConv2d';
+    id: string;
+    /** Inputs stay in concat order because that order selects weight channels. */
+    inputs: readonly {
+        value: string;
+        upsample?: Upsample2DNodeSpec;
+    }[];
+    conv: Conv2DNodeSpec;
+}
+export type ExecutableModelNode = ModelNodeSpec | FusedConvPoolNodeSpec | FusedUpsampleConcatConvNodeSpec;
+export interface GraphOptimizationOptions {
+    fuseConvPool?: boolean;
+    fuseUpsampleConcatConv?: boolean;
+}
+export interface OptimizedModelGraph {
+    spec: UNetModelSpec;
+    nodes: readonly ExecutableModelNode[];
+    fusions: {
+        convPool: number;
+        upsampleConcatConv: number;
+    };
+}
+/**
+ * Applies topology-only fusions. It never depends on a particular OIDN model
+ * name, so new descriptors automatically benefit from known graph patterns.
+ */
+export declare function optimizeModelGraph(validated: UNetModelGraph, options?: GraphOptimizationOptions): OptimizedModelGraph;
+export interface ModelValueShape {
+    width: number;
+    height: number;
+    channels: number;
+}
+export interface PlannedModelNode {
+    node: ExecutableModelNode;
+    outputShape: ModelValueShape;
+    /** Last planned node that reads the output; output itself uses nodes.length. */
+    lastUse: number;
+}
+export interface ModelExecutionPlan extends OptimizedModelGraph {
+    inputShape: ModelValueShape;
+    valueShapes: ReadonlyMap<string, ModelValueShape>;
+    plannedNodes: readonly PlannedModelNode[];
+}
+/** Resolve all runtime shapes and value lifetimes before allocating GPU data. */
+export declare function planModelExecution(validated: UNetModelGraph, width: number, height: number, options?: GraphOptimizationOptions): ModelExecutionPlan;

package/lib/graphOptimizer.js

@@ -0,0 +1,216 @@
+function nodeInputs(node) {
+    return node.op === 'concat' ? node.inputs : [node.input];
+}
+function buildConsumers(spec) {
+    const consumers = new Map();
+    for (const node of spec.nodes) {
+        for (const input of nodeInputs(node)) {
+            const list = consumers.get(input) ?? [];
+            list.push(node);
+            consumers.set(input, list);
+        }
+    }
+    return consumers;
+}
+// Callers narrow on `op`; node ids are unique, so they compare consumers by id.
+function onlyConsumer(consumers, value, op) {
+    const list = consumers.get(value);
+    if (list === undefined || list.length !== 1 || list[0].op !== op)
+        return undefined;
+    return list[0];
+}
+/**
+ * Applies topology-only fusions. It never depends on a particular OIDN model
+ * name, so new descriptors automatically benefit from known graph patterns.
+ */
+export function optimizeModelGraph(validated, options = {}) {
+    const spec = validated.spec;
+    const consumers = buildConsumers(spec);
+    const nodesById = new Map(spec.nodes.map((node) => [node.id, node]));
+    const eliminated = new Set();
+    const fusedAt = new Map();
+    let convPool = 0;
+    let upsampleConcatConv = 0;
+    if (options.fuseConvPool !== false) {
+        for (const node of spec.nodes) {
+            if (node.op !== 'conv2d' || node.activation !== 'relu')
+                continue;
+            const pool = onlyConsumer(consumers, node.id, 'maxPool2d');
+            if (!pool || pool.op !== 'maxPool2d' || pool.size !== 2 || pool.stride !== 2)
+                continue;
+            eliminated.add(node.id);
+            fusedAt.set(pool.id, {
+                op: 'fusedConvReluMaxPool2d',
+                id: pool.id,
+                input: node.input,
+                conv: node,
+                pool
+            });
+            convPool++;
+        }
+    }
+    if (options.fuseUpsampleConcatConv !== false) {
+        for (const node of spec.nodes) {
+            if (node.op !== 'conv2d')
+                continue;
+            const concat = nodesById.get(node.input);
+            if (concat?.op !== 'concat' || concat.inputs.length !== 2)
+                continue;
+            if (onlyConsumer(consumers, concat.id, 'conv2d')?.id !== node.id)
+                continue;
+            // The native blocked layout can remove concat only when the source
+            // boundary is also a vec4 boundary. Other graphs keep the generic ops.
+            const firstInputChannels = validated.channelsByValue.get(concat.inputs[0]);
+            if (firstInputChannels === undefined || firstInputChannels % 4 !== 0) {
+                continue;
+            }
+            const inputs = concat.inputs.map((value) => {
+                const candidate = nodesById.get(value);
+                if (candidate?.op === 'upsample2d' &&
+                    candidate.scale === 2 &&
+                    candidate.mode === 'nearest' &&
+                    onlyConsumer(consumers, candidate.id, 'concat')?.id === concat.id) {
+                    return { value: candidate.input, upsample: candidate };
+                }
+                return { value };
+            });
+            const upsampleCount = inputs.filter((input) => input.upsample).length;
+            if (upsampleCount !== 1)
+                continue;
+            eliminated.add(concat.id);
+            for (const value of concat.inputs) {
+                const candidate = nodesById.get(value);
+                if (candidate?.op === 'upsample2d')
+                    eliminated.add(candidate.id);
+            }
+            fusedAt.set(node.id, {
+                op: 'fusedUpsampleConcatConv2d',
+                id: node.id,
+                inputs,
+                conv: node
+            });
+            upsampleConcatConv++;
+        }
+    }
+    const nodes = [];
+    for (const node of spec.nodes) {
+        const fused = fusedAt.get(node.id);
+        if (fused) {
+            nodes.push(fused);
+        }
+        else if (!eliminated.has(node.id)) {
+            nodes.push(node);
+        }
+    }
+    return {
+        spec,
+        nodes,
+        fusions: { convPool, upsampleConcatConv }
+    };
+}
+function executableInputs(node) {
+    if (node.op === 'concat')
+        return node.inputs;
+    if (node.op === 'fusedUpsampleConcatConv2d') {
+        return node.inputs.map((input) => input.value);
+    }
+    return [node.input];
+}
+function sameSpatialShape(left, right) {
+    return left.width === right.width && left.height === right.height;
+}
+/** Resolve all runtime shapes and value lifetimes before allocating GPU data. */
+export function planModelExecution(validated, width, height, options) {
+    if (!Number.isInteger(width) || width <= 0 || !Number.isInteger(height) || height <= 0) {
+        throw new Error(`Invalid model input size ${width}x${height}`);
+    }
+    const graph = optimizeModelGraph(validated, options);
+    const inputShape = { width, height, channels: validated.inputChannels };
+    const valueShapes = new Map([
+        [validated.spec.input, inputShape]
+    ]);
+    const outputShapes = [];
+    const shapeOf = (value, nodeId) => {
+        const shape = valueShapes.get(value);
+        if (!shape)
+            throw new Error(`Planned node ${nodeId} reads missing value ${value}`);
+        return shape;
+    };
+    for (const node of graph.nodes) {
+        let outputShape;
+        if (node.op === 'conv2d') {
+            const input = shapeOf(node.input, node.id);
+            outputShape = {
+                width: input.width,
+                height: input.height,
+                channels: validated.convChannels.get(node.id).outputChannels
+            };
+        }
+        else if (node.op === 'maxPool2d') {
+            const input = shapeOf(node.input, node.id);
+            outputShape = {
+                width: Math.ceil(input.width / 2),
+                height: Math.ceil(input.height / 2),
+                channels: input.channels
+            };
+        }
+        else if (node.op === 'upsample2d') {
+            const input = shapeOf(node.input, node.id);
+            outputShape = {
+                width: input.width * 2,
+                height: input.height * 2,
+                channels: input.channels
+            };
+        }
+        else if (node.op === 'concat') {
+            const inputs = node.inputs.map((value) => shapeOf(value, node.id));
+            if (inputs.some((shape) => !sameSpatialShape(shape, inputs[0]))) {
+                throw new Error(`Concat ${node.id} has mismatched spatial shapes`);
+            }
+            outputShape = {
+                width: inputs[0].width,
+                height: inputs[0].height,
+                channels: inputs.reduce((sum, shape) => sum + shape.channels, 0)
+            };
+        }
+        else if (node.op === 'fusedConvReluMaxPool2d') {
+            const input = shapeOf(node.input, node.id);
+            outputShape = {
+                width: Math.ceil(input.width / 2),
+                height: Math.ceil(input.height / 2),
+                channels: validated.convChannels.get(node.conv.id).outputChannels
+            };
+        }
+        else {
+            const inputs = node.inputs.map((input) => {
+                const source = shapeOf(input.value, node.id);
+                return input.upsample
+                    ? { ...source, width: source.width * 2, height: source.height * 2 }
+                    : source;
+            });
+            if (inputs.some((shape) => !sameSpatialShape(shape, inputs[0]))) {
+                throw new Error(`Fused decoder ${node.id} has mismatched spatial shapes`);
+            }
+            outputShape = {
+                width: inputs[0].width,
+                height: inputs[0].height,
+                channels: validated.convChannels.get(node.conv.id).outputChannels
+            };
+        }
+        valueShapes.set(node.id, outputShape);
+        outputShapes.push(outputShape);
+    }
+    const lastUses = new Map();
+    graph.nodes.forEach((node, index) => {
+        for (const input of executableInputs(node))
+            lastUses.set(input, index);
+    });
+    lastUses.set(validated.spec.output, graph.nodes.length);
+    const plannedNodes = graph.nodes.map((node, index) => ({
+        node,
+        outputShape: outputShapes[index],
+        lastUse: lastUses.get(node.id) ?? index
+    }));
+    return { ...graph, inputShape, valueShapes, plannedNodes };
+}
+//# sourceMappingURL=graphOptimizer.js.map

package/lib/graphOptimizer.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"graphOptimizer.js","sourceRoot":"","sources":["../src/graphOptimizer.ts"],"names":[],"mappings":"AA+CA,SAAS,UAAU,CAAC,IAAmB;IACrC,OAAO,IAAI,CAAC,EAAE,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;AAC3D,CAAC;AAED,SAAS,cAAc,CAAC,IAAmB;IACzC,MAAM,SAAS,GAAG,IAAI,GAAG,EAA2B,CAAC;IACrD,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;QAC9B,KAAK,MAAM,KAAK,IAAI,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;YACrC,MAAM,IAAI,GAAG,SAAS,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC;YACxC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAChB,SAAS,CAAC,GAAG,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;QAC7B,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,gFAAgF;AAChF,SAAS,YAAY,CACnB,SAA+C,EAC/C,KAAa,EACb,EAAuB;IAEvB,MAAM,IAAI,GAAG,SAAS,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;IAClC,IAAI,IAAI,KAAK,SAAS,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE;QAAE,OAAO,SAAS,CAAC;IACnF,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC;AACjB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,kBAAkB,CAChC,SAAyB,EACzB,UAAoC,EAAE;IAEtC,MAAM,IAAI,GAAG,SAAS,CAAC,IAAI,CAAC;IAC5B,MAAM,SAAS,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;IACvC,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,EAAE,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC;IACrE,MAAM,UAAU,GAAG,IAAI,GAAG,EAAU,CAAC;IACrC,MAAM,OAAO,GAAG,IAAI,GAAG,EAA+B,CAAC;IACvD,IAAI,QAAQ,GAAG,CAAC,CAAC;IACjB,IAAI,kBAAkB,GAAG,CAAC,CAAC;IAE3B,IAAI,OAAO,CAAC,YAAY,KAAK,KAAK,EAAE,CAAC;QACnC,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;YAC9B,IAAI,IAAI,CAAC,EAAE,KAAK,QAAQ,IAAI,IAAI,CAAC,UAAU,KAAK,MAAM;gBAAE,SAAS;YACjE,MAAM,IAAI,GAAG,YAAY,CAAC,SAAS,EAAE,IAAI,CAAC,EAAE,EAAE,WAAW,CAAC,CAAC;YAC3D,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,EAAE,KAAK,WAAW,IAAI,IAAI,CAAC,IAAI,KAAK,CAAC,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC;gBAAE,SAAS;YAEvF,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACxB,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,EAAE;gBACnB,EAAE,EAAE,wBAAwB;gBAC5B,EAAE,EAAE,IAAI,CAAC,EAAE;gBACX,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,IAAI,EAAE,IAAI;gBACV,IAAI;aACL,CAAC,CAAC;YACH,QAAQ,EAAE,CAAC;QACb,CAAC;IACH,CAAC;IAED,IAAI,OAAO,CAAC,sBAAsB,KAAK,KAAK,EAAE,CAAC;QAC7C,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;YAC9B,IAAI,IAAI,CAAC,EAAE,KAAK,QAAQ;gBAAE,SAAS;YACnC,MAAM,MAAM,GAAG,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YACzC,IAAI,MAAM,EAAE,EAAE,KAAK,QAAQ,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,KAAK,CAAC;gBAAE,SAAS;YACpE,IAAI,YAAY,CAAC,SAAS,EAAE,MAAM,CAAC,EAAE,EAAE,QAAQ,CAAC,EAAE,EAAE,KAAK,IAAI,CAAC,EAAE;gBAAE,SAAS;YAC3E,mEAAmE;YACnE,uEAAuE;YACvE,MAAM,kBAAkB,GAAG,SAAS,CAAC,eAAe,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;YAC3E,IAAI,kBAAkB,KAAK,SAAS,IAAI,kBAAkB,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;gBACrE,SAAS;YACX,CAAC;YAED,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;gBACzC,MAAM,SAAS,GAAG,SAAS,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;gBACvC,IACE,SAAS,EAAE,EAAE,KAAK,YAAY;oBAC9B,SAAS,CAAC,KAAK,KAAK,CAAC;oBACrB,SAAS,CAAC,IAAI,KAAK,SAAS;oBAC5B,YAAY,CAAC,SAAS,EAAE,SAAS,CAAC,EAAE,EAAE,QAAQ,CAAC,EAAE,EAAE,KAAK,MAAM,CAAC,EAAE,EACjE,CAAC;oBACD,OAAO,EAAE,KAAK,EAAE,SAAS,CAAC,KAAK,EAAE,QAAQ,EAAE,SAAS,EAAE,CAAC;gBACzD,CAAC;gBACD,OAAO,EAAE,KAAK,EAAE,CAAC;YACnB,CAAC,CAAC,CAAC;YACH,MAAM,aAAa,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC;YACtE,IAAI,aAAa,KAAK,CAAC;gBAAE,SAAS;YAElC,UAAU,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;YAC1B,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;gBAClC,MAAM,SAAS,GAAG,SAAS,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;gBACvC,IAAI,SAAS,EAAE,EAAE,KAAK,YAAY;oBAAE,UAAU,CAAC,GAAG,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC;YACnE,CAAC;YACD,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,EAAE;gBACnB,EAAE,EAAE,2BAA2B;gBAC/B,EAAE,EAAE,IAAI,CAAC,EAAE;gBACX,MAAM;gBACN,IAAI,EAAE,IAAI;aACX,CAAC,CAAC;YACH,kBAAkB,EAAE,CAAC;QACvB,CAAC;IACH,CAAC;IAED,MAAM,KAAK,GAA0B,EAAE,CAAC;IACxC,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;QAC9B,MAAM,KAAK,GAAG,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACnC,IAAI,KAAK,EAAE,CAAC;YACV,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACpB,CAAC;aAAM,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC;YACpC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACnB,CAAC;IACH,CAAC;IAED,OAAO;QACL,IAAI;QACJ,KAAK;QACL,OAAO,EAAE,EAAE,QAAQ,EAAE,kBAAkB,EAAE;KAC1C,CAAC;AACJ,CAAC;AAqBD,SAAS,gBAAgB,CAAC,IAAyB;IACjD,IAAI,IAAI,CAAC,EAAE,KAAK,QAAQ;QAAE,OAAO,IAAI,CAAC,MAAM,CAAC;IAC7C,IAAI,IAAI,CAAC,EAAE,KAAK,2BAA2B,EAAE,CAAC;QAC5C,OAAO,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;IACjD,CAAC;IACD,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;AACtB,CAAC;AAED,SAAS,gBAAgB,CACvB,IAAqB,EACrB,KAAsB;IAEtB,OAAO,IAAI,CAAC,KAAK,KAAK,KAAK,CAAC,KAAK,IAAI,IAAI,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC;AACpE,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,kBAAkB,CAChC,SAAyB,EACzB,KAAa,EACb,MAAc,EACd,OAAkC;IAElC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,MAAM,IAAI,CAAC,EAAE,CAAC;QACvF,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,IAAI,MAAM,EAAE,CAAC,CAAC;IACjE,CAAC;IAED,MAAM,KAAK,GAAG,kBAAkB,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;IACrD,MAAM,UAAU,GAAG,EAAE,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,SAAS,CAAC,aAAa,EAAE,CAAC;IACxE,MAAM,WAAW,GAAG,IAAI,GAAG,CAA0B;QACnD,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,EAAE,UAAU,CAAC;KACnC,CAAC,CAAC;IACH,MAAM,YAAY,GAAsB,EAAE,CAAC;IAE3C,MAAM,OAAO,GAAG,CAAC,KAAa,EAAE,MAAc,EAAE,EAAE;QAChD,MAAM,KAAK,GAAG,WAAW,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;QACrC,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,gBAAgB,MAAM,wBAAwB,KAAK,EAAE,CAAC,CAAC;QACnF,OAAO,KAAK,CAAC;IACf,CAAC,CAAC;IAEF,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC;QAC/B,IAAI,WAA4B,CAAC;QACjC,IAAI,IAAI,CAAC,EAAE,KAAK,QAAQ,EAAE,CAAC;YACzB,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC;YAC3C,WAAW,GAAG;gBACZ,KAAK,EAAE,KAAK,CAAC,KAAK;gBAClB,MAAM,EAAE,KAAK,CAAC,MAAM;gBACpB,QAAQ,EAAE,SAAS,CAAC,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAE,CAAC,cAAc;aAC9D,CAAC;QACJ,CAAC;aAAM,IAAI,IAAI,CAAC,EAAE,KAAK,WAAW,EAAE,CAAC;YACnC,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC;YAC3C,WAAW,GAAG;gBACZ,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC;gBACjC,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC;gBACnC,QAAQ,EAAE,KAAK,CAAC,QAAQ;aACzB,CAAC;QACJ,CAAC;aAAM,IAAI,IAAI,CAAC,EAAE,KAAK,YAAY,EAAE,CAAC;YACpC,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC;YAC3C,WAAW,GAAG;gBACZ,KAAK,EAAE,KAAK,CAAC,KAAK,GAAG,CAAC;gBACtB,MAAM,EAAE,KAAK,CAAC,MAAM,GAAG,CAAC;gBACxB,QAAQ,EAAE,KAAK,CAAC,QAAQ;aACzB,CAAC;QACJ,CAAC;aAAM,IAAI,IAAI,CAAC,EAAE,KAAK,QAAQ,EAAE,CAAC;YAChC,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC;YACnE,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,gBAAgB,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;gBAChE,MAAM,IAAI,KAAK,CAAC,UAAU,IAAI,CAAC,EAAE,gCAAgC,CAAC,CAAC;YACrE,CAAC;YACD,WAAW,GAAG;gBACZ,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,KAAK;gBACtB,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM;gBACxB,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;aACjE,CAAC;QACJ,CAAC;aAAM,IAAI,IAAI,CAAC,EAAE,KAAK,wBAAwB,EAAE,CAAC;YAChD,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC;YAC3C,WAAW,GAAG;gBACZ,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC;gBACjC,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC;gBACnC,QAAQ,EAAE,SAAS,CAAC,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAE,CAAC,cAAc;aACnE,CAAC;QACJ,CAAC;aAAM,CAAC;YACN,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;gBACvC,MAAM,MAAM,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC;gBAC7C,OAAO,KAAK,CAAC,QAAQ;oBACnB,CAAC,CAAC,EAAE,GAAG,MAAM,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,GAAG,CAAC,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE;oBACnE,CAAC,CAAC,MAAM,CAAC;YACb,CAAC,CAAC,CAAC;YACH,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,gBAAgB,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;gBAChE,MAAM,IAAI,KAAK,CAAC,iBAAiB,IAAI,CAAC,EAAE,gCAAgC,CAAC,CAAC;YAC5E,CAAC;YACD,WAAW,GAAG;gBACZ,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,KAAK;gBACtB,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM;gBACxB,QAAQ,EAAE,SAAS,CAAC,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAE,CAAC,cAAc;aACnE,CAAC;QACJ,CAAC;QAED,WAAW,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,EAAE,WAAW,CAAC,CAAC;QACtC,YAAY,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;IACjC,CAAC;IAED,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC3C,KAAK,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE;QAClC,KAAK,MAAM,KAAK,IAAI,gBAAgB,CAAC,IAAI,CAAC;YAAE,QAAQ,CAAC,GAAG,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;IACzE,CAAC,CAAC,CAAC;IACH,QAAQ,CAAC,GAAG,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IAExD,MAAM,YAAY,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC;QACrD,IAAI;QACJ,WAAW,EAAE,YAAY,CAAC,KAAK,CAAC;QAChC,OAAO,EAAE,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,KAAK;KACxC,CAAC,CAAC,CAAC;IAEJ,OAAO,EAAE,GAAG,KAAK,EAAE,UAAU,EAAE,WAAW,EAAE,YAAY,EAAE,CAAC;AAC7D,CAAC"}

package/lib/hdr.d.ts

@@ -0,0 +1,31 @@
+/**
+ * Convert 4 byte uint8 buffer to 3 channels float data
+ * @param rgbe input uint8 buffer
+ * @param float output float data
+ */
+declare function rgbe2float(rgbe: Uint8Array, float: Float32Array): void;
+/**
+ * Convert 3 channels float data to 4 byte uint8 buffer
+ * @param float input float data
+ * @param rgbe output uint8 buffer
+ */
+declare function float2rgbe(float: Float32Array, rgbe: Uint8Array): void;
+/**
+ * Write float data to RGBE(.hdr) file buffer
+ * @param w image width
+ * @param h image height
+ * @param data float data, RGB 3 channels.
+ * @returns file buffer
+ */
+declare function write_hdr(w: number, h: number, data: Float32Array): Uint8Array;
+/**
+ * Read a RGBE(.hdr) file from buffer
+ * @param uint8 RGBE(.hdr) file buffer
+ * @returns Failure reason or resolved data.
+ */
+declare function read_hdr(uint8: Uint8Array): string | {
+    rgbFloat: Float32Array;
+    width: number;
+    height: number;
+};
+export { read_hdr as readHDR, write_hdr as writeHDR, float2rgbe, rgbe2float };

package/lib/hdr.js

@@ -0,0 +1,340 @@
+// https://github.com/bbbbx/hdr.js/blob/main/src/hdr.ts
+function frexp(v) {
+    v = Number(v);
+    const result = { f: v, e: 0 };
+    if (v !== 0 && Number.isFinite(v)) {
+        const absV = Math.abs(v);
+        const log2 = Math.log2 ||
+            function log2(n) {
+                return Math.log(n) * Math.LOG2E;
+            };
+        // Math.pow(2, -exp) === Infinity when exp <= -1024
+        let e = Math.max(-1023, Math.floor(log2(absV)) + 1);
+        let f = absV * Math.pow(2.0, -e);
+        while (f >= 1.0) {
+            f *= 0.5;
+            e++;
+        }
+        while (f < 0.5) {
+            f *= 2;
+            e--;
+        }
+        if (v < 0) {
+            f = -f;
+        }
+        result.f = f;
+        result.e = e;
+    }
+    return result;
+}
+function ldexp(f, e) {
+    return f * Math.pow(2.0, e);
+}
+/**
+ * Convert 4 byte uint8 buffer to 3 channels float data
+ * @param rgbe input uint8 buffer
+ * @param float output float data
+ */
+function rgbe2float(rgbe, float) {
+    if (rgbe[3] !== 0) {
+        const f1 = ldexp(1.0, rgbe[3] - (128 + 8));
+        float[0] = rgbe[0] * f1;
+        float[1] = rgbe[1] * f1;
+        float[2] = rgbe[2] * f1;
+    }
+    else {
+        float[0] = float[1] = float[2] = 0.0;
+    }
+}
+/**
+ * Convert 3 channels float data to 4 byte uint8 buffer
+ * @param float input float data
+ * @param rgbe output uint8 buffer
+ */
+function float2rgbe(float, rgbe) {
+    const red = float[0];
+    const green = float[1];
+    const blue = float[2];
+    const v = Math.max(red, Math.max(green, blue));
+    if (v < 1e-32) {
+        rgbe[0] = rgbe[1] = rgbe[2] = rgbe[3] = 0;
+    }
+    else {
+        const { f, e } = frexp(v);
+        const s = (f / v) * 256.0;
+        rgbe[0] = red * s;
+        rgbe[1] = green * s;
+        rgbe[2] = blue * s;
+        rgbe[3] = e + 128;
+    }
+}
+/**
+ * Write float data to RGBE(.hdr) file buffer
+ * @param w image width
+ * @param h image height
+ * @param data float data, RGB 3 channels.
+ * @returns file buffer
+ */
+function write_hdr(w, h, data) {
+    const s = [];
+    const comp = 3;
+    write_hdr_core(s, w, h, comp, data);
+    return new Uint8Array(s);
+}
+function write_hdr_core(s, x, y, comp, data) {
+    if (y <= 0 || x <= 0 || !data) {
+        return;
+    }
+    const header = '#?RADIANCE\n' +
+        'FORMAT=32-bit_rle_rgbe\n' +
+        'EXPOSURE=1.0\n' +
+        '\n' +
+        '-Y ' +
+        y +
+        ' +X ' +
+        x +
+        '\n';
+    header.split('').forEach((c) => {
+        const charCode = c.charCodeAt(0);
+        s.push(charCode);
+    });
+    const scratch = new Uint8Array(x * 4);
+    const stbi__flip_vertically_on_write = false;
+    for (let i = 0; i < y; i++) {
+        write_hdr_scanline(s, x, comp, scratch, data.subarray(comp * x * (stbi__flip_vertically_on_write ? y - 1 - i : i)));
+    }
+}
+function write_hdr_scanline(s, width, ncomp, scratch, scanline) {
+    const scanlineheader = new Uint8Array([2, 2, 0, 0]);
+    const rgbe = new Uint8Array(4);
+    const linear = new Float32Array(3);
+    let x;
+    scanlineheader[2] = (width & 0xff00) >> 8;
+    scanlineheader[3] = width & 0x00ff;
+    /* skip RLE for images too small or large */
+    if (width < 8 || width >= 32768) {
+        for (x = 0; x < width; x++) {
+            switch (ncomp) {
+                case 4: /* fallthrough */
+                case 3:
+                    linear[2] = scanline[x * ncomp + 2];
+                    linear[1] = scanline[x * ncomp + 1];
+                    linear[0] = scanline[x * ncomp + 0];
+                    break;
+                default:
+                    linear[0] = linear[1] = linear[2] = scanline[x * ncomp + 0];
+                    break;
+            }
+            float2rgbe(linear, rgbe);
+            s.push(rgbe[0], rgbe[1], rgbe[2], rgbe[3]);
+        }
+    }
+    else {
+        let c, r;
+        /* encode into scratch buffer */
+        for (x = 0; x < width; x++) {
+            switch (ncomp) {
+                case 4: /* fallthrough */
+                case 3:
+                    linear[2] = scanline[x * ncomp + 2];
+                    linear[1] = scanline[x * ncomp + 1];
+                    linear[0] = scanline[x * ncomp + 0];
+                    break;
+                default:
+                    linear[0] = linear[1] = linear[2] = scanline[x * ncomp + 0];
+                    break;
+            }
+            float2rgbe(linear, rgbe);
+            scratch[x + width * 0] = rgbe[0];
+            scratch[x + width * 1] = rgbe[1];
+            scratch[x + width * 2] = rgbe[2];
+            scratch[x + width * 3] = rgbe[3];
+        }
+        s.push(scanlineheader[0], scanlineheader[1], scanlineheader[2], scanlineheader[3]);
+        /* RLE each component separately */
+        for (c = 0; c < 4; c++) {
+            const comp = scratch.subarray(width * c);
+            x = 0;
+            while (x < width) {
+                // find first run
+                r = x;
+                while (r + 2 < width) {
+                    if (comp[r] === comp[r + 1] && comp[r] === comp[r + 2])
+                        break;
+                    ++r;
+                }
+                if (r + 2 >= width)
+                    r = width;
+                // dump up to first run
+                while (x < r) {
+                    let len = r - x;
+                    if (len > 128)
+                        len = 128;
+                    write_dump_data(s, len, comp.subarray(x));
+                    x += len;
+                }
+                // if there's a run, output it
+                if (r + 2 < width) {
+                    // same test as what we break out of in search loop, so only true if we break'd
+                    // find next byte after run
+                    while (r < width && comp[r] == comp[x])
+                        ++r;
+                    // output run up to r
+                    while (x < r) {
+                        let len = r - x;
+                        if (len > 127)
+                            len = 127;
+                        write_run_data(s, len, comp[x]);
+                        x += len;
+                    }
+                }
+            }
+        }
+    }
+}
+function write_dump_data(s, length, data) {
+    const lengthbyte = length & 0xff;
+    if (!(length <= 128))
+        throw new Error('length is greater than 128');
+    s.push(lengthbyte);
+    for (let i = 0; i < length; i++) {
+        s.push(data[i]);
+    }
+}
+function write_run_data(s, length, databyte) {
+    const lengthbyte = (length + 128) & 0xff;
+    if (!(length + 128 <= 255))
+        throw new Error('length is greater than 128');
+    s.push(lengthbyte, databyte);
+}
+function testMagic(uint8, signature) {
+    const l = signature.length;
+    for (let i = 0; i < l; i++) {
+        if (uint8[i] !== signature.charCodeAt(i)) {
+            return false;
+        }
+    }
+    return true;
+}
+function isHdr(uint8) {
+    let r = testMagic(uint8, '#?RADIANCE\n');
+    if (!r) {
+        r = testMagic(uint8, '#?RGBE\n');
+    }
+    return r;
+}
+const MAX_HEADER_LENGTH = 1024 * 10;
+const MAX_DIMENSIONS = 1 << 24;
+/**
+ * Read a RGBE(.hdr) file from buffer
+ * @param uint8 RGBE(.hdr) file buffer
+ * @returns Failure reason or resolved data.
+ */
+function read_hdr(uint8) {
+    let header = '';
+    let pos = 0;
+    if (!isHdr(uint8)) {
+        return 'Corrupt HDR image.';
+    }
+    // read header
+    while (!header.match(/\n\n[^\n]+\n/g) && pos < MAX_HEADER_LENGTH) {
+        header += String.fromCharCode(uint8[pos++]);
+    }
+    // check format
+    const format = header.match(/FORMAT=(.*)$/m)?.[1];
+    if (format !== '32-bit_rle_rgbe') {
+        return 'Unsupported HDR format: ' + format;
+    }
+    // parse resolution
+    const rez = header.split(/\n/).reverse()[1].split(' ');
+    if (rez[0] !== '-Y' || rez[2] !== '+X') {
+        return 'Unsupported HDR format';
+    }
+    const width = Number.parseFloat(rez[3]);
+    const height = Number.parseFloat(rez[1]);
+    if (width > MAX_DIMENSIONS || height > MAX_DIMENSIONS) {
+        return 'Very large image (corrupt?)';
+    }
+    let i, j;
+    let c1 = uint8[pos];
+    let c2 = uint8[pos + 1];
+    let len = uint8[pos + 2];
+    // not run-length encoded, so we have to actually use THIS data as a decoded
+    // pixel (note this can't be a valid pixel--one of RGB must be >= 128)
+    const notRLE = c1 !== 2 || c2 !== 2 || !!(len & 0x80); // not run-length encoded
+    const hdrData = new Float32Array(width * height * 3);
+    if (width < 8 || width >= 32768 || notRLE) {
+        // 32768: 2^15
+        // Read flat data
+        for (j = 0; j < height; ++j) {
+            for (i = 0; i < width; ++i) {
+                const rgbe = uint8.subarray(pos, pos + 4);
+                pos += 4;
+                const start = (j * width + i) * 3;
+                rgbe2float(rgbe, hdrData.subarray(start, start + 3));
+            }
+        }
+    }
+    else {
+        // Read RLE-encoded data
+        let scanline;
+        let c1;
+        let c2;
+        let len;
+        for (let j = 0; j < height; j++) {
+            c1 = uint8[pos++];
+            c2 = uint8[pos++];
+            len = uint8[pos++];
+            if (c1 !== 2 || c2 !== 2 || len & 0x80) {
+                return 'Invalid scanline';
+            }
+            len = len << 8;
+            len |= uint8[pos++];
+            if (len !== width) {
+                return 'invalid decoded scanline length';
+            }
+            if (!scanline) {
+                scanline = new Uint8Array(width * 4);
+            }
+            let count;
+            let value;
+            for (let k = 0; k < 4; k++) {
+                let nLeft;
+                i = 0;
+                while ((nLeft = width - i) > 0) {
+                    count = uint8[pos++];
+                    if (count > 128) {
+                        // is RUN
+                        value = uint8[pos++];
+                        count -= 128;
+                        if (count > nLeft) {
+                            return 'bad RLE data in HDR';
+                        }
+                        for (let z = 0; z < count; z++) {
+                            scanline[i++ * 4 + k] = value;
+                        }
+                    }
+                    else {
+                        // is DUMP
+                        if (count > nLeft) {
+                            return 'bad RLE data in HDR';
+                        }
+                        for (let z = 0; z < count; z++) {
+                            scanline[i++ * 4 + k] = uint8[pos++];
+                        }
+                    }
+                }
+            }
+            for (let i = 0; i < width; i++) {
+                rgbe2float(scanline.subarray(i * 4), hdrData.subarray((j * width + i) * 3));
+            }
+        }
+    }
+    return {
+        rgbFloat: hdrData,
+        width: width,
+        height: height
+    };
+}
+export { read_hdr as readHDR, write_hdr as writeHDR, float2rgbe, rgbe2float };
+//# sourceMappingURL=hdr.js.map