| --- |
| library_name: kernels |
| license: apache-2.0 |
| tags: |
| - kernel |
| - webgpu |
| - wgsl |
| --- |
| # ai.onnx.Scatter |
|
|
| `ai.onnx` · standard ONNX operator · ONNX opset ≥ 11 |
|
|
| ## Description |
|
|
| Copies `data` and overwrites positions identified by `indices` with values from `updates` along a single `axis`. For each element in `updates`, the index along `axis` comes from the corresponding `indices` value, while all other dimension indices come from the element's own position. Deprecated in favor of ScatterElements, which provides identical behavior. |
|
|
| See the [ONNX `Scatter` spec](https://onnx.ai/onnx/operators/onnx__Scatter.html) for the reference semantics. |
|
|
| ## Inputs |
|
|
| | Name | Logical dtype | Rank | Shape | Description | Presence | |
| | --- | --- | --- | --- | --- | --- | |
| | `data` | `T` | — | — | The input tensor to copy; the output has the same shape. | required | |
| | `indices` | `I` | — | — | Integer tensor of the same rank as `data`, specifying the target index along `axis` for each update. | required | |
| | `updates` | `T` | — | — | Values to scatter into the output; must have the same rank and shape as `indices`. | required | |
|
|
| ## Outputs |
|
|
| | Name | Logical dtype | Rank | Shape | Description | Presence | |
| | --- | --- | --- | --- | --- | --- | |
| | `output` | `T` | same as `data` | same as `data` | Copy of `data` with entries overwritten at positions specified by `indices`. | required | |
|
|
| ## Attributes |
|
|
| Default values (overridable per request): |
|
|
| | Attribute | Default | Description | |
| | --- | --- | --- | |
| | `axis` | `0` | The axis of `data` along which scatter indices are applied. Negative values count from the end; defaults to `0` (the outer-most axis). | |
|
|
| ## Type constraints |
|
|
| | Variable | Allowed dtypes | |
| | --- | --- | |
| | `T` | `float32`, `float16`, `int32`, `int16`, `int8`, `uint32`, `uint8`, `bool` | |
| | `I` | `int32` | |
|
|
| ## Files |
|
|
| - [`metadata.json`](build/webgpu/metadata.json) — kernel metadata (id, digests, per-variant templates, provenance) |
| - [`manifest.json`](build/webgpu/manifest.json) — the op contract (source of truth) |
| - [`test.json`](build/webgpu/test.json) — correctness cases |
| - [`bench.json`](build/webgpu/bench.json) — benchmark cases |
| - [`scatter-flat-copy.wgsl.jinja`](build/webgpu/scatter-flat-copy.wgsl.jinja) |
| - [`scatter.wgsl.jinja`](build/webgpu/scatter.wgsl.jinja) |
|
|
| ## Use with `@huggingface/kernels` |
|
|
| ```sh |
| npm install --save-exact @huggingface/kernels@0.0.1-preview.3 |
| ``` |
|
|
| Required output shapes and logical data types are inferred from the supplied inputs and attributes; result tensors are allocated automatically. |
|
|
| The `version: 1` option selects the published kernel contract; it is independent of any operator opset, contrib `since_version`, or model version. |
| It follows the `v1` branch as fixes land. To pin exact artifact bytes, pass a 40-character commit `revision` instead of `version`. |
|
|
| Replace each `*Data` placeholder with a typed array containing the corresponding input data. |
|
|
| ```js |
| import { getKernel } from "@huggingface/kernels"; |
| |
| const kernel = await getKernel("webgpu-kernels/ai.onnx.Scatter", { version: 1 }); |
| const { output } = await kernel({ |
| data: { data: dataData, shape: [1, 3] }, |
| indices: { data: indicesData, shape: [3, 3] }, |
| updates: { data: updatesData, shape: [3, 3] }, |
| }); |
| ``` |
|
|
