sync 2e7068faf55e
Browse files- README.md +71 -0
- build/webgpu/bench.json +132 -0
- build/webgpu/grid-sample.wgsl.jinja +342 -0
- build/webgpu/grid-sample3d.wgsl.jinja +228 -0
- build/webgpu/manifest.json +184 -0
- build/webgpu/metadata.json +19 -0
- build/webgpu/test.json +1965 -0
README.md
CHANGED
|
@@ -1,3 +1,74 @@
|
|
| 1 |
---
|
|
|
|
| 2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
library_name: kernels
|
| 3 |
license: apache-2.0
|
| 4 |
+
tags:
|
| 5 |
+
- kernel
|
| 6 |
+
- webgpu
|
| 7 |
+
- wgsl
|
| 8 |
---
|
| 9 |
+
# ai.onnx.GridSample
|
| 10 |
+
|
| 11 |
+
`ai.onnx` · standard ONNX operator · ONNX opset ≥ 20
|
| 12 |
+
|
| 13 |
+
## Description
|
| 14 |
+
|
| 15 |
+
Samples values from input tensor `X` at positions defined by a flow-field `grid`, producing output `Y` with spatial dimensions taken from `grid`. Grid coordinates are normalized to `[-1, 1]` over the input spatial extent; positions outside this range are handled according to `padding_mode`. Supports spatial `(rank-4, NCHW)` and volumetric `(rank-5, NCDHW)` inputs with `linear`, `nearest`, or `cubic` interpolation.
|
| 16 |
+
|
| 17 |
+
See the [ONNX `GridSample` spec](https://onnx.ai/onnx/operators/onnx__GridSample.html) for the reference semantics.
|
| 18 |
+
|
| 19 |
+
## Inputs
|
| 20 |
+
|
| 21 |
+
| Name | Bind key | Logical dtype | Rank | Shape | Description | Presence |
|
| 22 |
+
| --- | --- | --- | --- | --- | --- | --- |
|
| 23 |
+
| `X` | `x` | `T` | — | — | Input tensor of shape `(N, C, D1, ..., Dr)` whose values are sampled. | required |
|
| 24 |
+
| `grid` | `grid` | `T` | — | — | Flow-field of shape `(N, D1_out, ..., Dr_out, r)` with normalized sampling coordinates in `[-1, 1]`. | required |
|
| 25 |
+
|
| 26 |
+
## Outputs
|
| 27 |
+
|
| 28 |
+
| Name | Bind key | Logical dtype | Rank | Shape | Description | Presence |
|
| 29 |
+
| --- | --- | --- | --- | --- | --- | --- |
|
| 30 |
+
| `Y` | `y` | `T` | same as `grid` | derived; see description | Output tensor of shape `(N, C, D1_out, ..., Dr_out)` containing the interpolated samples. | required |
|
| 31 |
+
|
| 32 |
+
## Attributes
|
| 33 |
+
|
| 34 |
+
Default values (overridable per request):
|
| 35 |
+
|
| 36 |
+
| Attribute | Default | Description |
|
| 37 |
+
| --- | --- | --- |
|
| 38 |
+
| `mode` | `"linear"` | Interpolation method: `linear` (bilinear or trilinear, depending on rank), `nearest`, or `cubic`. Cubic interpolation is supported for rank-4 (2-D spatial) inputs. |
|
| 39 |
+
| `padding_mode` | `"zeros"` | How out-of-bound grid positions are handled: `zeros` pads with 0, `border` clamps to the border value, or `reflection` reflects coordinates back into the valid range. |
|
| 40 |
+
| `align_corners` | `0` | When 1, extrema values `-1` and `1` map to the center of the corner pixels; when 0 (default) they map to the outer edge of corner pixels, making sampling resolution-agnostic. |
|
| 41 |
+
|
| 42 |
+
## Type constraints
|
| 43 |
+
|
| 44 |
+
| Variable | Allowed dtypes |
|
| 45 |
+
| --- | --- |
|
| 46 |
+
| `T` | `float32`, `float16` |
|
| 47 |
+
|
| 48 |
+
## Files
|
| 49 |
+
|
| 50 |
+
- [`metadata.json`](build/webgpu/metadata.json) — kernel metadata (id, digests, provenance)
|
| 51 |
+
- [`manifest.json`](build/webgpu/manifest.json) — the op contract (source of truth)
|
| 52 |
+
- [`test.json`](build/webgpu/test.json) — correctness cases
|
| 53 |
+
- [`bench.json`](build/webgpu/bench.json) — benchmark + tuning cases
|
| 54 |
+
- [`grid-sample.wgsl.jinja`](build/webgpu/grid-sample.wgsl.jinja)
|
| 55 |
+
- [`grid-sample3d.wgsl.jinja`](build/webgpu/grid-sample3d.wgsl.jinja)
|
| 56 |
+
|
| 57 |
+
## Use with `@huggingface/kernels`
|
| 58 |
+
|
| 59 |
+
The loader derives every required output's shape and logical dtype from the manifest contract and this call.
|
| 60 |
+
It then allocates the result tensors automatically.
|
| 61 |
+
|
| 62 |
+
The `version: 1` option selects the published kernel contract; it is independent of any operator opset, contrib `since_version`, or model version.
|
| 63 |
+
|
| 64 |
+
Replace each `*Data` placeholder with a typed array containing the corresponding input data.
|
| 65 |
+
|
| 66 |
+
```js
|
| 67 |
+
import { getKernel } from "@huggingface/kernels";
|
| 68 |
+
|
| 69 |
+
const kernel = await getKernel("webgpu-kernels/ai.onnx.GridSample", { version: 1 });
|
| 70 |
+
const { y } = await kernel({
|
| 71 |
+
x: { data: xData, shape: [1, 1, 2, 2] },
|
| 72 |
+
grid: { data: gridData, shape: [1, 1, 3, 2] },
|
| 73 |
+
});
|
| 74 |
+
```
|
build/webgpu/bench.json
ADDED
|
@@ -0,0 +1,132 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"op": "ai.onnx.GridSample",
|
| 3 |
+
"cases": [
|
| 4 |
+
{
|
| 5 |
+
"name": "1x3x256x256_to_256_linear",
|
| 6 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 7 |
+
"inputs": {
|
| 8 |
+
"x": { "dtype": "float32", "shape": [1, 3, 256, 256] },
|
| 9 |
+
"grid": { "dtype": "float32", "shape": [1, 256, 256, 2] }
|
| 10 |
+
},
|
| 11 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 3, 256, 256] } }
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"name": "nchw_1x32x256x256_linear_zeros_f32_healthy",
|
| 15 |
+
"preset": "smoke",
|
| 16 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 17 |
+
"inputs": {
|
| 18 |
+
"x": { "dtype": "float32", "shape": [1, 32, 256, 256], "dist": "normal", "seed": 811, "scale": 1 },
|
| 19 |
+
"grid": { "dtype": "float32", "shape": [1, 256, 256, 2], "dist": "normal", "seed": 812, "scale": 1 }
|
| 20 |
+
},
|
| 21 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 256, 256] } },
|
| 22 |
+
"bench": {
|
| 23 |
+
"primary": true,
|
| 24 |
+
"metrics": [{ "type": "bandwidth", "value": "(numel(shapes.x) + numel(shapes.grid) + numel(shapes.y)) * 4" }]
|
| 25 |
+
}
|
| 26 |
+
},
|
| 27 |
+
{
|
| 28 |
+
"name": "nchw_1x32x256x256_cubic_zeros_f32_16reads_pathology",
|
| 29 |
+
"preset": "smoke",
|
| 30 |
+
"attrs": { "mode": "cubic", "padding_mode": "zeros", "align_corners": 0 },
|
| 31 |
+
"inputs": {
|
| 32 |
+
"x": { "dtype": "float32", "shape": [1, 32, 256, 256], "dist": "normal", "seed": 821, "scale": 1 },
|
| 33 |
+
"grid": { "dtype": "float32", "shape": [1, 256, 256, 2], "dist": "normal", "seed": 822, "scale": 1 }
|
| 34 |
+
},
|
| 35 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 256, 256] } },
|
| 36 |
+
"bench": {
|
| 37 |
+
"metrics": [{ "type": "bandwidth", "value": "(numel(shapes.x) + numel(shapes.grid) + numel(shapes.y)) * 4" }]
|
| 38 |
+
}
|
| 39 |
+
},
|
| 40 |
+
{
|
| 41 |
+
"name": "nchw_1x32x256x256_linear_reflection_f32_reflectcoord_pathology",
|
| 42 |
+
"preset": "smoke",
|
| 43 |
+
"attrs": { "mode": "linear", "padding_mode": "reflection", "align_corners": 1 },
|
| 44 |
+
"inputs": {
|
| 45 |
+
"x": { "dtype": "float32", "shape": [1, 32, 256, 256], "dist": "normal", "seed": 831, "scale": 1 },
|
| 46 |
+
"grid": { "dtype": "float32", "shape": [1, 256, 256, 2], "dist": "normal", "seed": 832, "scale": 1.5 }
|
| 47 |
+
},
|
| 48 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 32, 256, 256] } },
|
| 49 |
+
"bench": {
|
| 50 |
+
"metrics": [{ "type": "bandwidth", "value": "(numel(shapes.x) + numel(shapes.grid) + numel(shapes.y)) * 4" }]
|
| 51 |
+
}
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"name": "nchw_1x32x256x256_linear_zeros_f16_halfbandwidth_pathology",
|
| 55 |
+
"preset": "smoke",
|
| 56 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 57 |
+
"inputs": {
|
| 58 |
+
"x": { "dtype": "float16", "shape": [1, 32, 256, 256], "dist": "normal", "seed": 841, "scale": 1 },
|
| 59 |
+
"grid": { "dtype": "float16", "shape": [1, 256, 256, 2], "dist": "normal", "seed": 842, "scale": 1 }
|
| 60 |
+
},
|
| 61 |
+
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 256, 256] } },
|
| 62 |
+
"bench": {
|
| 63 |
+
"metrics": [{ "type": "bandwidth", "value": "(numel(shapes.x) + numel(shapes.grid) + numel(shapes.y)) * 2" }]
|
| 64 |
+
}
|
| 65 |
+
},
|
| 66 |
+
{
|
| 67 |
+
"name": "ncdhw_1x4x16x32x32_linear_zeros_f32_volumetric_unbenched_variant",
|
| 68 |
+
"preset": "smoke",
|
| 69 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 70 |
+
"inputs": {
|
| 71 |
+
"x": { "dtype": "float32", "shape": [1, 4, 16, 32, 32], "dist": "normal", "seed": 851, "scale": 1 },
|
| 72 |
+
"grid": { "dtype": "float32", "shape": [1, 16, 32, 32, 3], "dist": "normal", "seed": 852, "scale": 1 }
|
| 73 |
+
},
|
| 74 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 16, 32, 32] } },
|
| 75 |
+
"bench": {
|
| 76 |
+
"metrics": [{ "type": "bandwidth", "value": "(numel(shapes.x) + numel(shapes.grid) + numel(shapes.y)) * 4" }]
|
| 77 |
+
}
|
| 78 |
+
},
|
| 79 |
+
{
|
| 80 |
+
"name": "nchw_1x8x1024x1024_linear_zeros_f32_dispatch_cliff_over16m",
|
| 81 |
+
"preset": "smoke",
|
| 82 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 83 |
+
"inputs": {
|
| 84 |
+
"x": { "dtype": "float32", "shape": [1, 8, 256, 256], "dist": "normal", "seed": 861, "scale": 1 },
|
| 85 |
+
"grid": { "dtype": "float32", "shape": [1, 1024, 1024, 2], "dist": "normal", "seed": 862, "scale": 1 }
|
| 86 |
+
},
|
| 87 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 8, 1024, 1024] } },
|
| 88 |
+
"bench": {
|
| 89 |
+
"metrics": [{ "type": "bandwidth", "value": "(numel(shapes.x) + numel(shapes.grid) + numel(shapes.y)) * 4" }]
|
| 90 |
+
}
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"name": "nchw_1x3x2049x2731_linear_zeros_f32_partial_channel_dispatch_cliff",
|
| 94 |
+
"preset": "stress",
|
| 95 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 96 |
+
"inputs": {
|
| 97 |
+
"x": { "dtype": "float32", "shape": [1, 3, 16, 16], "dist": "normal", "seed": 863, "scale": 1 },
|
| 98 |
+
"grid": { "dtype": "float32", "shape": [1, 2049, 2731, 2], "dist": "normal", "seed": 864, "scale": 0.9 }
|
| 99 |
+
},
|
| 100 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 3, 2049, 2731] } },
|
| 101 |
+
"bench": {
|
| 102 |
+
"metrics": [{ "type": "bandwidth", "value": "(numel(shapes.x) + numel(shapes.grid) + numel(shapes.y)) * 4" }]
|
| 103 |
+
}
|
| 104 |
+
},
|
| 105 |
+
{
|
| 106 |
+
"name": "nchw_cubic_reflection_highC_f32_perchannel_recompute_amp",
|
| 107 |
+
"preset": "stress",
|
| 108 |
+
"attrs": { "mode": "cubic", "padding_mode": "reflection", "align_corners": 0 },
|
| 109 |
+
"inputs": {
|
| 110 |
+
"x": { "dtype": "float32", "shape": [1, 48, 192, 192], "dist": "normal", "seed": 9101, "scale": 1 },
|
| 111 |
+
"grid": { "dtype": "float32", "shape": [1, 320, 320, 2], "dist": "normal", "seed": 9102, "scale": 1.5 }
|
| 112 |
+
},
|
| 113 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 48, 320, 320] } },
|
| 114 |
+
"bench": {
|
| 115 |
+
"metrics": [{ "type": "bandwidth", "value": "(numel(shapes.x) + numel(shapes.grid) + numel(shapes.y)) * 4" }]
|
| 116 |
+
}
|
| 117 |
+
},
|
| 118 |
+
{
|
| 119 |
+
"name": "ncdhw_linear_reflection_f32_trilinear_reflect_amp",
|
| 120 |
+
"preset": "stress",
|
| 121 |
+
"attrs": { "mode": "linear", "padding_mode": "reflection", "align_corners": 0 },
|
| 122 |
+
"inputs": {
|
| 123 |
+
"x": { "dtype": "float32", "shape": [1, 8, 24, 32, 32], "dist": "normal", "seed": 9201, "scale": 1 },
|
| 124 |
+
"grid": { "dtype": "float32", "shape": [1, 48, 48, 48, 3], "dist": "normal", "seed": 9202, "scale": 1.5 }
|
| 125 |
+
},
|
| 126 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 8, 48, 48, 48] } },
|
| 127 |
+
"bench": {
|
| 128 |
+
"metrics": [{ "type": "bandwidth", "value": "(numel(shapes.x) + numel(shapes.grid) + numel(shapes.y)) * 4" }]
|
| 129 |
+
}
|
| 130 |
+
}
|
| 131 |
+
]
|
| 132 |
+
}
|
build/webgpu/grid-sample.wgsl.jinja
ADDED
|
@@ -0,0 +1,342 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{% macro flat_index_2d(name="i", bound="params.count", guardInline=false, note="dispatch-limit") %}
|
| 2 |
+
{% if note == "dispatch-limit" %}
|
| 3 |
+
// 2D-folded flat index: gid.y carries the high bits past the
|
| 4 |
+
// maxComputeWorkgroupsPerDimension dispatch limit (outputs > 16.7M elements).
|
| 5 |
+
{% elif note == "limit" %}
|
| 6 |
+
// 2D-folded flat index: gid.y carries the high bits past the
|
| 7 |
+
// maxComputeWorkgroupsPerDimension limit.
|
| 8 |
+
{% elif note == "device-axis" %}
|
| 9 |
+
// The flat dispatch is folded across x/y at the device's per-axis workgroup
|
| 10 |
+
// limit; gid.y carries the high portion of the output index.
|
| 11 |
+
{% elif note == "vec4-limit" %}
|
| 12 |
+
// 2D-folded flat vec4 index: gid.y carries the high bits past the
|
| 13 |
+
// maxComputeWorkgroupsPerDimension limit (the dispatch caps x and spills into y).
|
| 14 |
+
{% elif note == "element-limit" %}
|
| 15 |
+
// 2D-folded flat element index: gid.y carries the high bits past the
|
| 16 |
+
// maxComputeWorkgroupsPerDimension limit.
|
| 17 |
+
{% elif note == "dispatch" %}
|
| 18 |
+
// 2D-folded flat index: gid.y carries the high bits past the
|
| 19 |
+
// maxComputeWorkgroupsPerDimension dispatch limit.
|
| 20 |
+
{% endif %}
|
| 21 |
+
{% if bound == "" %}
|
| 22 |
+
let {{ name }} = gid.x + gid.y * nwg.x * {{ tunables.WORKGROUP_SIZE }}u;
|
| 23 |
+
{%- elif guardInline %}
|
| 24 |
+
let {{ name }} = gid.x + gid.y * nwg.x * {{ tunables.WORKGROUP_SIZE }}u;
|
| 25 |
+
if ({{ name }} >= {{ bound }}) { return; }
|
| 26 |
+
{%- else %}
|
| 27 |
+
let {{ name }} = gid.x + gid.y * nwg.x * {{ tunables.WORKGROUP_SIZE }}u;
|
| 28 |
+
if ({{ name }} >= {{ bound }}) {
|
| 29 |
+
return;
|
| 30 |
+
}
|
| 31 |
+
{%- endif %}
|
| 32 |
+
{% endmacro %}
|
| 33 |
+
|
| 34 |
+
{% if usesF16 %}
|
| 35 |
+
enable f16;
|
| 36 |
+
{% endif %}
|
| 37 |
+
{{ env.wgsl.resourceDeclarations }}
|
| 38 |
+
|
| 39 |
+
fn denormalize(n: f32, length: u32) -> f32 {
|
| 40 |
+
{% if source.alignCorners %}
|
| 41 |
+
return (n + 1.0) * 0.5 * f32(length - 1u);
|
| 42 |
+
{% else %}
|
| 43 |
+
return ((n + 1.0) * f32(length) - 1.0) * 0.5;
|
| 44 |
+
{% endif %}
|
| 45 |
+
}
|
| 46 |
+
|
| 47 |
+
fn sanitize_coord(v: f32) -> f32 {
|
| 48 |
+
if (!(v >= -2147483000.0 && v <= 2147483000.0)) {
|
| 49 |
+
{% if source.alignCorners %}
|
| 50 |
+
return 0.0;
|
| 51 |
+
{% else %}
|
| 52 |
+
return -0.5;
|
| 53 |
+
{% endif %}
|
| 54 |
+
}
|
| 55 |
+
return v;
|
| 56 |
+
}
|
| 57 |
+
{% if source.paddingMode == "reflection" %}
|
| 58 |
+
|
| 59 |
+
fn reflect_coord(v: f32, lo: f32, hi: f32) -> f32 {
|
| 60 |
+
let range = hi - lo;
|
| 61 |
+
if (range == 0.0) { return lo; }
|
| 62 |
+
var x0 = v;
|
| 63 |
+
if (x0 < lo) {
|
| 64 |
+
let d = lo - x0;
|
| 65 |
+
if (d / range > 2147483000.0) { return lo; }
|
| 66 |
+
let n = i32(floor(d / range));
|
| 67 |
+
let r = d - f32(n) * range;
|
| 68 |
+
x0 = select(hi - r, lo + r, n % 2 == 0);
|
| 69 |
+
} else if (x0 > hi) {
|
| 70 |
+
let d = x0 - hi;
|
| 71 |
+
if (d / range > 2147483000.0) { return hi; }
|
| 72 |
+
let n = i32(floor(d / range));
|
| 73 |
+
let r = d - f32(n) * range;
|
| 74 |
+
x0 = select(lo + r, hi - r, n % 2 == 0);
|
| 75 |
+
}
|
| 76 |
+
return x0;
|
| 77 |
+
}
|
| 78 |
+
|
| 79 |
+
{% endif %}
|
| 80 |
+
{% if source.mode != "nearest" %}
|
| 81 |
+
fn sample_floor(v: f32) -> i32 {
|
| 82 |
+
if (v > 2147483000.0) {
|
| 83 |
+
{% if source.paddingMode == "reflection" %}
|
| 84 |
+
return 0;
|
| 85 |
+
{% else %}
|
| 86 |
+
return 2147483000;
|
| 87 |
+
{% endif %}
|
| 88 |
+
}
|
| 89 |
+
if (v < -2147483000.0) {
|
| 90 |
+
{% if source.paddingMode == "reflection" %}
|
| 91 |
+
return 0;
|
| 92 |
+
{% else %}
|
| 93 |
+
return -2147483000;
|
| 94 |
+
{% endif %}
|
| 95 |
+
}
|
| 96 |
+
return i32(floor(v));
|
| 97 |
+
}
|
| 98 |
+
{%- else %}
|
| 99 |
+
fn sample_round(v: f32) -> i32 {
|
| 100 |
+
if (v > 2147483000.0) {
|
| 101 |
+
{% if source.paddingMode == "reflection" %}
|
| 102 |
+
return 0;
|
| 103 |
+
{% else %}
|
| 104 |
+
return 2147483000;
|
| 105 |
+
{% endif %}
|
| 106 |
+
}
|
| 107 |
+
if (v < -2147483000.0) {
|
| 108 |
+
{% if source.paddingMode == "reflection" %}
|
| 109 |
+
return 0;
|
| 110 |
+
{% else %}
|
| 111 |
+
return -2147483000;
|
| 112 |
+
{% endif %}
|
| 113 |
+
}
|
| 114 |
+
return i32(round(v));
|
| 115 |
+
}
|
| 116 |
+
{%- endif %}
|
| 117 |
+
|
| 118 |
+
|
| 119 |
+
{% if source.mode == "cubic" %}
|
| 120 |
+
fn cubic_coeffs(t: f32) -> vec4<f32> {
|
| 121 |
+
let a = -0.75;
|
| 122 |
+
let x0 = abs(t + 1.0);
|
| 123 |
+
let x1 = abs(t);
|
| 124 |
+
let x2 = abs(1.0 - t);
|
| 125 |
+
let x3 = abs(2.0 - t);
|
| 126 |
+
return vec4<f32>(cubic_one(x0, a), cubic_one(x1, a), cubic_one(x2, a), cubic_one(x3, a));
|
| 127 |
+
}
|
| 128 |
+
|
| 129 |
+
fn cubic_one(x0: f32, a: f32) -> f32 {
|
| 130 |
+
if (x0 <= 1.0) { return (a + 2.0) * x0 * x0 * x0 - (a + 3.0) * x0 * x0 + 1.0; }
|
| 131 |
+
if (x0 < 2.0) { return a * x0 * x0 * x0 - 5.0 * a * x0 * x0 + 8.0 * a * x0 - 4.0 * a; }
|
| 132 |
+
return 0.0;
|
| 133 |
+
}
|
| 134 |
+
|
| 135 |
+
{% endif -%}
|
| 136 |
+
{% if source.channelWidth is not defined %}
|
| 137 |
+
fn pixel(base: u32, h: i32, w: i32) -> f32 {
|
| 138 |
+
{% if source.paddingMode == "zeros" %}
|
| 139 |
+
if (h < 0 || h >= i32(params.inH) || w < 0 || w >= i32(params.inW)) { return 0.0; }
|
| 140 |
+
let hh = u32(h);
|
| 141 |
+
let ww = u32(w);
|
| 142 |
+
{% elif source.paddingMode == "border" %}
|
| 143 |
+
let hh = u32(clamp(h, 0, i32(params.inH) - 1));
|
| 144 |
+
let ww = u32(clamp(w, 0, i32(params.inW) - 1));
|
| 145 |
+
{% else %}
|
| 146 |
+
{% if source.alignCorners %}
|
| 147 |
+
let rh = i32(reflect_coord(f32(h), 0.0, f32(params.inH) - 1.0));
|
| 148 |
+
let rw = i32(reflect_coord(f32(w), 0.0, f32(params.inW) - 1.0));
|
| 149 |
+
{% else %}
|
| 150 |
+
let rh = i32(reflect_coord(f32(h), -0.5, f32(params.inH) - 0.5));
|
| 151 |
+
let rw = i32(reflect_coord(f32(w), -0.5, f32(params.inW) - 0.5));
|
| 152 |
+
{% endif %}
|
| 153 |
+
let hh = u32(clamp(rh, 0, i32(params.inH) - 1));
|
| 154 |
+
let ww = u32(clamp(rw, 0, i32(params.inW) - 1));
|
| 155 |
+
{% endif %}
|
| 156 |
+
return f32(x[base + hh * params.inW + ww]);
|
| 157 |
+
}
|
| 158 |
+
|
| 159 |
+
|
| 160 |
+
{% endif %}
|
| 161 |
+
{% if source.channelWidth is defined %}
|
| 162 |
+
{% set channelVec = "vec" ~ source.channelWidth ~ "<f32>" %}
|
| 163 |
+
{% set components = ["x", "y", "z", "w"] %}
|
| 164 |
+
// A block of NCHW channels shares one grid coordinate. Tail lanes alias the
|
| 165 |
+
// final channel safely because main suppresses their stores.
|
| 166 |
+
{% if source.paddingMode != "reflection" or source.mode == "nearest" %}
|
| 167 |
+
fn pixel_channels(n: u32, c0: u32, h: i32, w: i32) -> {{ channelVec }} {
|
| 168 |
+
{% if source.paddingMode == "zeros" %}
|
| 169 |
+
if (h < 0 || h >= i32(params.inH) || w < 0 || w >= i32(params.inW)) { return {{ channelVec }}(0.0); }
|
| 170 |
+
let hh = u32(h);
|
| 171 |
+
let ww = u32(w);
|
| 172 |
+
{% elif source.paddingMode == "border" %}
|
| 173 |
+
let hh = u32(clamp(h, 0, i32(params.inH) - 1));
|
| 174 |
+
let ww = u32(clamp(w, 0, i32(params.inW) - 1));
|
| 175 |
+
{% else %}
|
| 176 |
+
{% if source.alignCorners %}
|
| 177 |
+
let rh = i32(reflect_coord(f32(h), 0.0, f32(params.inH) - 1.0));
|
| 178 |
+
let rw = i32(reflect_coord(f32(w), 0.0, f32(params.inW) - 1.0));
|
| 179 |
+
{% else %}
|
| 180 |
+
let rh = i32(reflect_coord(f32(h), -0.5, f32(params.inH) - 0.5));
|
| 181 |
+
let rw = i32(reflect_coord(f32(w), -0.5, f32(params.inW) - 0.5));
|
| 182 |
+
{% endif %}
|
| 183 |
+
let hh = u32(clamp(rh, 0, i32(params.inH) - 1));
|
| 184 |
+
let ww = u32(clamp(rw, 0, i32(params.inW) - 1));
|
| 185 |
+
{% endif %}
|
| 186 |
+
let plane = params.inH * params.inW;
|
| 187 |
+
let offset = hh * params.inW + ww;
|
| 188 |
+
let nb = n * params.C;
|
| 189 |
+
return {{ channelVec }}(
|
| 190 |
+
{% for lane in range(source.channelWidth) %}
|
| 191 |
+
f32(x[(nb + {% if source.channelTail %}min(c0 + {{ lane }}u, params.C - 1u){% else %}c0 + {{ lane }}u{% endif %}) * plane + offset]){% if not loop.last %},{% else %});{% endif %}
|
| 192 |
+
{% endfor %}
|
| 193 |
+
}
|
| 194 |
+
|
| 195 |
+
{% endif %}
|
| 196 |
+
{% if source.paddingMode == "reflection" %}
|
| 197 |
+
// Reflection gives every tap in one row or column the same resolved coordinate.
|
| 198 |
+
// Resolve each row and column once in main and reuse it for all channel loads.
|
| 199 |
+
{% if source.mode != "nearest" %}
|
| 200 |
+
fn pixel_channels_resolved(n: u32, c0: u32, h: u32, w: u32) -> {{ channelVec }} {
|
| 201 |
+
let plane = params.inH * params.inW;
|
| 202 |
+
let offset = h * params.inW + w;
|
| 203 |
+
let nb = n * params.C;
|
| 204 |
+
return {{ channelVec }}(
|
| 205 |
+
{% for lane in range(source.channelWidth) %}
|
| 206 |
+
f32(x[(nb + {% if source.channelTail %}min(c0 + {{ lane }}u, params.C - 1u){% else %}c0 + {{ lane }}u{% endif %}) * plane + offset]){% if not loop.last %},{% else %});{% endif %}
|
| 207 |
+
{% endfor %}
|
| 208 |
+
}
|
| 209 |
+
|
| 210 |
+
{% endif %}
|
| 211 |
+
{% else %}
|
| 212 |
+
|
| 213 |
+
{% endif %}
|
| 214 |
+
{% endif -%}
|
| 215 |
+
@compute @workgroup_size({{ tunables.WORKGROUP_SIZE }})
|
| 216 |
+
fn main(@builtin(global_invocation_id) gid: vec3<u32>, @builtin(num_workgroups) nwg: vec3<u32>) {
|
| 217 |
+
{{ flat_index_2d("i", "") }}
|
| 218 |
+
{% if source.channelWidth is defined %}
|
| 219 |
+
let out_plane = params.outH * params.outW;
|
| 220 |
+
let channel_blocks = (params.C + {{ source.channelWidth - 1 }}u) / {{ source.channelWidth }}u;
|
| 221 |
+
let block_count = (params.count / params.C) * channel_blocks;
|
| 222 |
+
if (i >= block_count) { return; }
|
| 223 |
+
let spatial = i % out_plane;
|
| 224 |
+
let cb = (i / out_plane) % channel_blocks;
|
| 225 |
+
let n = i / (out_plane * channel_blocks);
|
| 226 |
+
let c = cb * {{ source.channelWidth }}u;
|
| 227 |
+
let grid_base = (n * out_plane + spatial) * 2u;
|
| 228 |
+
let sx = sanitize_coord(denormalize(f32(grid[grid_base]), params.inW));
|
| 229 |
+
let sy = sanitize_coord(denormalize(f32(grid[grid_base + 1u]), params.inH));
|
| 230 |
+
{% if source.mode == "nearest" %}
|
| 231 |
+
let result = pixel_channels(n, c, sample_round(sy), sample_round(sx));
|
| 232 |
+
{% elif source.mode == "cubic" %}
|
| 233 |
+
let x0 = sample_floor(sx) - 1;
|
| 234 |
+
let y0 = sample_floor(sy) - 1;
|
| 235 |
+
let dx = sx - f32(x0 + 1);
|
| 236 |
+
let dy = sy - f32(y0 + 1);
|
| 237 |
+
let cx = cubic_coeffs(dx);
|
| 238 |
+
let cy = cubic_coeffs(dy);
|
| 239 |
+
{% if source.paddingMode == "reflection" %}
|
| 240 |
+
var reflected_y: array<u32, 4>;
|
| 241 |
+
var reflected_x: array<u32, 4>;
|
| 242 |
+
for (var k = 0u; k < 4u; k = k + 1u) {
|
| 243 |
+
{% if source.alignCorners %}
|
| 244 |
+
reflected_y[k] = u32(clamp(i32(reflect_coord(f32(y0 + i32(k)), 0.0, f32(params.inH) - 1.0)), 0, i32(params.inH) - 1));
|
| 245 |
+
reflected_x[k] = u32(clamp(i32(reflect_coord(f32(x0 + i32(k)), 0.0, f32(params.inW) - 1.0)), 0, i32(params.inW) - 1));
|
| 246 |
+
{% else %}
|
| 247 |
+
reflected_y[k] = u32(clamp(i32(reflect_coord(f32(y0 + i32(k)), -0.5, f32(params.inH) - 0.5)), 0, i32(params.inH) - 1));
|
| 248 |
+
reflected_x[k] = u32(clamp(i32(reflect_coord(f32(x0 + i32(k)), -0.5, f32(params.inW) - 0.5)), 0, i32(params.inW) - 1));
|
| 249 |
+
{% endif %}
|
| 250 |
+
}
|
| 251 |
+
{% endif %}
|
| 252 |
+
var result = {{ channelVec }}(0.0);
|
| 253 |
+
for (var r = 0i; r < 4i; r = r + 1i) {
|
| 254 |
+
{% if source.paddingMode == "reflection" %}
|
| 255 |
+
let row = cx.x * pixel_channels_resolved(n, c, reflected_y[u32(r)], reflected_x[0])
|
| 256 |
+
+ cx.y * pixel_channels_resolved(n, c, reflected_y[u32(r)], reflected_x[1])
|
| 257 |
+
+ cx.z * pixel_channels_resolved(n, c, reflected_y[u32(r)], reflected_x[2])
|
| 258 |
+
+ cx.w * pixel_channels_resolved(n, c, reflected_y[u32(r)], reflected_x[3]);
|
| 259 |
+
{% else %}
|
| 260 |
+
let row = cx.x * pixel_channels(n, c, y0 + r, x0)
|
| 261 |
+
+ cx.y * pixel_channels(n, c, y0 + r, x0 + 1)
|
| 262 |
+
+ cx.z * pixel_channels(n, c, y0 + r, x0 + 2)
|
| 263 |
+
+ cx.w * pixel_channels(n, c, y0 + r, x0 + 3);
|
| 264 |
+
{% endif %}
|
| 265 |
+
result = result + cy[u32(r)] * row;
|
| 266 |
+
}
|
| 267 |
+
{% else %}
|
| 268 |
+
let x0 = sample_floor(sx);
|
| 269 |
+
let y0 = sample_floor(sy);
|
| 270 |
+
let wx = sx - f32(x0);
|
| 271 |
+
let wy = sy - f32(y0);
|
| 272 |
+
{% if source.paddingMode == "reflection" %}
|
| 273 |
+
{% if source.alignCorners %}
|
| 274 |
+
let reflected_y0 = u32(clamp(i32(reflect_coord(f32(y0), 0.0, f32(params.inH) - 1.0)), 0, i32(params.inH) - 1));
|
| 275 |
+
let reflected_y1 = u32(clamp(i32(reflect_coord(f32(y0 + 1), 0.0, f32(params.inH) - 1.0)), 0, i32(params.inH) - 1));
|
| 276 |
+
let reflected_x0 = u32(clamp(i32(reflect_coord(f32(x0), 0.0, f32(params.inW) - 1.0)), 0, i32(params.inW) - 1));
|
| 277 |
+
let reflected_x1 = u32(clamp(i32(reflect_coord(f32(x0 + 1), 0.0, f32(params.inW) - 1.0)), 0, i32(params.inW) - 1));
|
| 278 |
+
{% else %}
|
| 279 |
+
let reflected_y0 = u32(clamp(i32(reflect_coord(f32(y0), -0.5, f32(params.inH) - 0.5)), 0, i32(params.inH) - 1));
|
| 280 |
+
let reflected_y1 = u32(clamp(i32(reflect_coord(f32(y0 + 1), -0.5, f32(params.inH) - 0.5)), 0, i32(params.inH) - 1));
|
| 281 |
+
let reflected_x0 = u32(clamp(i32(reflect_coord(f32(x0), -0.5, f32(params.inW) - 0.5)), 0, i32(params.inW) - 1));
|
| 282 |
+
let reflected_x1 = u32(clamp(i32(reflect_coord(f32(x0 + 1), -0.5, f32(params.inW) - 0.5)), 0, i32(params.inW) - 1));
|
| 283 |
+
{% endif %}
|
| 284 |
+
let v00 = pixel_channels_resolved(n, c, reflected_y0, reflected_x0);
|
| 285 |
+
let v01 = pixel_channels_resolved(n, c, reflected_y0, reflected_x1);
|
| 286 |
+
let v10 = pixel_channels_resolved(n, c, reflected_y1, reflected_x0);
|
| 287 |
+
let v11 = pixel_channels_resolved(n, c, reflected_y1, reflected_x1);
|
| 288 |
+
{% else %}
|
| 289 |
+
let v00 = pixel_channels(n, c, y0, x0);
|
| 290 |
+
let v01 = pixel_channels(n, c, y0, x0 + 1);
|
| 291 |
+
let v10 = pixel_channels(n, c, y0 + 1, x0);
|
| 292 |
+
let v11 = pixel_channels(n, c, y0 + 1, x0 + 1);
|
| 293 |
+
{% endif %}
|
| 294 |
+
let result = (1.0 - wy) * ((1.0 - wx) * v00 + wx * v01)
|
| 295 |
+
+ wy * ((1.0 - wx) * v10 + wx * v11);
|
| 296 |
+
{% endif %}
|
| 297 |
+
let out_base = (n * params.C + c) * out_plane + spatial;
|
| 298 |
+
{% for lane in range(source.channelWidth) %}
|
| 299 |
+
{% if source.channelTail and lane > 0 %}
|
| 300 |
+
if (c + {{ lane }}u < params.C) { y[out_base + {{ lane }}u * out_plane] = {{ scalar }}(result.{{ components[lane] }}); }
|
| 301 |
+
{% else %}
|
| 302 |
+
y[out_base + {{ lane }}u * out_plane] = {{ scalar }}(result.{{ components[lane] }});
|
| 303 |
+
{% endif %}
|
| 304 |
+
{% endfor %}
|
| 305 |
+
{% else %}
|
| 306 |
+
if (i >= params.count) { return; }
|
| 307 |
+
let ow = i % params.outW;
|
| 308 |
+
let oh = (i / params.outW) % params.outH;
|
| 309 |
+
let c = (i / (params.outW * params.outH)) % params.C;
|
| 310 |
+
let n = i / (params.outW * params.outH * params.C);
|
| 311 |
+
let grid_base = ((n * params.outH + oh) * params.outW + ow) * 2u;
|
| 312 |
+
let sx = sanitize_coord(denormalize(f32(grid[grid_base]), params.inW));
|
| 313 |
+
let sy = sanitize_coord(denormalize(f32(grid[grid_base + 1u]), params.inH));
|
| 314 |
+
let img_base = (n * params.C + c) * params.inH * params.inW;
|
| 315 |
+
{% if source.mode == "nearest" %}
|
| 316 |
+
let result = pixel(img_base, sample_round(sy), sample_round(sx));
|
| 317 |
+
{% elif source.mode == "cubic" %}
|
| 318 |
+
let x0 = sample_floor(sx) - 1;
|
| 319 |
+
let y0 = sample_floor(sy) - 1;
|
| 320 |
+
let dx = sx - f32(x0 + 1);
|
| 321 |
+
let dy = sy - f32(y0 + 1);
|
| 322 |
+
let cx = cubic_coeffs(dx);
|
| 323 |
+
let cy = cubic_coeffs(dy);
|
| 324 |
+
var rows: vec4<f32>;
|
| 325 |
+
for (var r = 0i; r < 4i; r = r + 1i) {
|
| 326 |
+
rows[u32(r)] = cx.x * pixel(img_base, y0 + r, x0) + cx.y * pixel(img_base, y0 + r, x0 + 1) + cx.z * pixel(img_base, y0 + r, x0 + 2) + cx.w * pixel(img_base, y0 + r, x0 + 3);
|
| 327 |
+
}
|
| 328 |
+
let result = dot(cy, rows);
|
| 329 |
+
{% else %}
|
| 330 |
+
let x0 = sample_floor(sx);
|
| 331 |
+
let y0 = sample_floor(sy);
|
| 332 |
+
let wx = sx - f32(x0);
|
| 333 |
+
let wy = sy - f32(y0);
|
| 334 |
+
let v00 = pixel(img_base, y0, x0);
|
| 335 |
+
let v01 = pixel(img_base, y0, x0 + 1);
|
| 336 |
+
let v10 = pixel(img_base, y0 + 1, x0);
|
| 337 |
+
let v11 = pixel(img_base, y0 + 1, x0 + 1);
|
| 338 |
+
let result = (1.0 - wy) * ((1.0 - wx) * v00 + wx * v01) + wy * ((1.0 - wx) * v10 + wx * v11);
|
| 339 |
+
{% endif %}
|
| 340 |
+
y[i] = {{ scalar }}(result);
|
| 341 |
+
{% endif %}
|
| 342 |
+
}
|
build/webgpu/grid-sample3d.wgsl.jinja
ADDED
|
@@ -0,0 +1,228 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{% macro flat_index_2d(name="i", bound="params.count", guardInline=false, note="dispatch-limit") %}
|
| 2 |
+
{% if note == "dispatch-limit" %}
|
| 3 |
+
// 2D-folded flat index: gid.y carries the high bits past the
|
| 4 |
+
// maxComputeWorkgroupsPerDimension dispatch limit (outputs > 16.7M elements).
|
| 5 |
+
{% elif note == "limit" %}
|
| 6 |
+
// 2D-folded flat index: gid.y carries the high bits past the
|
| 7 |
+
// maxComputeWorkgroupsPerDimension limit.
|
| 8 |
+
{% elif note == "device-axis" %}
|
| 9 |
+
// The flat dispatch is folded across x/y at the device's per-axis workgroup
|
| 10 |
+
// limit; gid.y carries the high portion of the output index.
|
| 11 |
+
{% elif note == "vec4-limit" %}
|
| 12 |
+
// 2D-folded flat vec4 index: gid.y carries the high bits past the
|
| 13 |
+
// maxComputeWorkgroupsPerDimension limit (the dispatch caps x and spills into y).
|
| 14 |
+
{% elif note == "element-limit" %}
|
| 15 |
+
// 2D-folded flat element index: gid.y carries the high bits past the
|
| 16 |
+
// maxComputeWorkgroupsPerDimension limit.
|
| 17 |
+
{% elif note == "dispatch" %}
|
| 18 |
+
// 2D-folded flat index: gid.y carries the high bits past the
|
| 19 |
+
// maxComputeWorkgroupsPerDimension dispatch limit.
|
| 20 |
+
{% endif %}
|
| 21 |
+
{% if bound == "" %}
|
| 22 |
+
let {{ name }} = gid.x + gid.y * nwg.x * {{ tunables.WORKGROUP_SIZE }}u;
|
| 23 |
+
{%- elif guardInline %}
|
| 24 |
+
let {{ name }} = gid.x + gid.y * nwg.x * {{ tunables.WORKGROUP_SIZE }}u;
|
| 25 |
+
if ({{ name }} >= {{ bound }}) { return; }
|
| 26 |
+
{%- else %}
|
| 27 |
+
let {{ name }} = gid.x + gid.y * nwg.x * {{ tunables.WORKGROUP_SIZE }}u;
|
| 28 |
+
if ({{ name }} >= {{ bound }}) {
|
| 29 |
+
return;
|
| 30 |
+
}
|
| 31 |
+
{%- endif %}
|
| 32 |
+
{% endmacro %}
|
| 33 |
+
|
| 34 |
+
{% if usesF16 %}
|
| 35 |
+
enable f16;
|
| 36 |
+
{% endif %}
|
| 37 |
+
{{ env.wgsl.resourceDeclarations }}
|
| 38 |
+
|
| 39 |
+
fn denormalize(n: f32, length: u32) -> f32 {
|
| 40 |
+
{% if source.alignCorners %}
|
| 41 |
+
return (n + 1.0) * 0.5 * f32(length - 1u);
|
| 42 |
+
{% else %}
|
| 43 |
+
return ((n + 1.0) * f32(length) - 1.0) * 0.5;
|
| 44 |
+
{% endif %}
|
| 45 |
+
}
|
| 46 |
+
|
| 47 |
+
fn sanitize_coord(v: f32) -> f32 {
|
| 48 |
+
if (!(v >= -2147483000.0 && v <= 2147483000.0)) {
|
| 49 |
+
{% if source.alignCorners %}
|
| 50 |
+
return 0.0;
|
| 51 |
+
{% else %}
|
| 52 |
+
return -0.5;
|
| 53 |
+
{% endif %}
|
| 54 |
+
}
|
| 55 |
+
return v;
|
| 56 |
+
}
|
| 57 |
+
{% if source.paddingMode == "reflection" %}
|
| 58 |
+
|
| 59 |
+
fn reflect_coord(v: f32, lo: f32, hi: f32) -> f32 {
|
| 60 |
+
let range = hi - lo;
|
| 61 |
+
if (range == 0.0) { return lo; }
|
| 62 |
+
var x0 = v;
|
| 63 |
+
if (x0 < lo) {
|
| 64 |
+
let d = lo - x0;
|
| 65 |
+
if (d / range > 2147483000.0) { return lo; }
|
| 66 |
+
let n = i32(floor(d / range));
|
| 67 |
+
let r = d - f32(n) * range;
|
| 68 |
+
x0 = select(hi - r, lo + r, n % 2 == 0);
|
| 69 |
+
} else if (x0 > hi) {
|
| 70 |
+
let d = x0 - hi;
|
| 71 |
+
if (d / range > 2147483000.0) { return hi; }
|
| 72 |
+
let n = i32(floor(d / range));
|
| 73 |
+
let r = d - f32(n) * range;
|
| 74 |
+
x0 = select(lo + r, hi - r, n % 2 == 0);
|
| 75 |
+
}
|
| 76 |
+
return x0;
|
| 77 |
+
}
|
| 78 |
+
|
| 79 |
+
{% endif %}
|
| 80 |
+
{% if source.mode != "nearest" %}
|
| 81 |
+
fn sample_floor(v: f32) -> i32 {
|
| 82 |
+
if (v > 2147483000.0) {
|
| 83 |
+
{% if source.paddingMode == "reflection" %}
|
| 84 |
+
return 0;
|
| 85 |
+
{% else %}
|
| 86 |
+
return 2147483000;
|
| 87 |
+
{% endif %}
|
| 88 |
+
}
|
| 89 |
+
if (v < -2147483000.0) {
|
| 90 |
+
{% if source.paddingMode == "reflection" %}
|
| 91 |
+
return 0;
|
| 92 |
+
{% else %}
|
| 93 |
+
return -2147483000;
|
| 94 |
+
{% endif %}
|
| 95 |
+
}
|
| 96 |
+
return i32(floor(v));
|
| 97 |
+
}
|
| 98 |
+
{%- else %}
|
| 99 |
+
fn sample_round(v: f32) -> i32 {
|
| 100 |
+
if (v > 2147483000.0) {
|
| 101 |
+
{% if source.paddingMode == "reflection" %}
|
| 102 |
+
return 0;
|
| 103 |
+
{% else %}
|
| 104 |
+
return 2147483000;
|
| 105 |
+
{% endif %}
|
| 106 |
+
}
|
| 107 |
+
if (v < -2147483000.0) {
|
| 108 |
+
{% if source.paddingMode == "reflection" %}
|
| 109 |
+
return 0;
|
| 110 |
+
{% else %}
|
| 111 |
+
return -2147483000;
|
| 112 |
+
{% endif %}
|
| 113 |
+
}
|
| 114 |
+
return i32(round(v));
|
| 115 |
+
}
|
| 116 |
+
{%- endif %}
|
| 117 |
+
|
| 118 |
+
|
| 119 |
+
{% if source.mode == "cubic" %}
|
| 120 |
+
fn cubic_coeffs(t: f32) -> vec4<f32> {
|
| 121 |
+
let a = -0.75;
|
| 122 |
+
let x0 = abs(t + 1.0);
|
| 123 |
+
let x1 = abs(t);
|
| 124 |
+
let x2 = abs(1.0 - t);
|
| 125 |
+
let x3 = abs(2.0 - t);
|
| 126 |
+
return vec4<f32>(cubic_one(x0, a), cubic_one(x1, a), cubic_one(x2, a), cubic_one(x3, a));
|
| 127 |
+
}
|
| 128 |
+
|
| 129 |
+
fn cubic_one(x0: f32, a: f32) -> f32 {
|
| 130 |
+
if (x0 <= 1.0) { return (a + 2.0) * x0 * x0 * x0 - (a + 3.0) * x0 * x0 + 1.0; }
|
| 131 |
+
if (x0 < 2.0) { return a * x0 * x0 * x0 - 5.0 * a * x0 * x0 + 8.0 * a * x0 - 4.0 * a; }
|
| 132 |
+
return 0.0;
|
| 133 |
+
}
|
| 134 |
+
|
| 135 |
+
{% endif -%}
|
| 136 |
+
fn voxel(base: u32, d: i32, h: i32, w: i32) -> f32 {
|
| 137 |
+
{% if source.paddingMode == "zeros" %}
|
| 138 |
+
if (d < 0 || d >= i32(params.inD) || h < 0 || h >= i32(params.inH) || w < 0 || w >= i32(params.inW)) {
|
| 139 |
+
return 0.0;
|
| 140 |
+
}
|
| 141 |
+
let dd = u32(d);
|
| 142 |
+
let hh = u32(h);
|
| 143 |
+
let ww = u32(w);
|
| 144 |
+
{% elif source.paddingMode == "border" %}
|
| 145 |
+
let dd = u32(clamp(d, 0, i32(params.inD) - 1));
|
| 146 |
+
let hh = u32(clamp(h, 0, i32(params.inH) - 1));
|
| 147 |
+
let ww = u32(clamp(w, 0, i32(params.inW) - 1));
|
| 148 |
+
{% else %}
|
| 149 |
+
{% if source.alignCorners %}
|
| 150 |
+
let rd = i32(reflect_coord(f32(d), 0.0, f32(params.inD) - 1.0));
|
| 151 |
+
let rh = i32(reflect_coord(f32(h), 0.0, f32(params.inH) - 1.0));
|
| 152 |
+
let rw = i32(reflect_coord(f32(w), 0.0, f32(params.inW) - 1.0));
|
| 153 |
+
{% else %}
|
| 154 |
+
let rd = i32(reflect_coord(f32(d), -0.5, f32(params.inD) - 0.5));
|
| 155 |
+
let rh = i32(reflect_coord(f32(h), -0.5, f32(params.inH) - 0.5));
|
| 156 |
+
let rw = i32(reflect_coord(f32(w), -0.5, f32(params.inW) - 0.5));
|
| 157 |
+
{% endif %}
|
| 158 |
+
let dd = u32(clamp(rd, 0, i32(params.inD) - 1));
|
| 159 |
+
let hh = u32(clamp(rh, 0, i32(params.inH) - 1));
|
| 160 |
+
let ww = u32(clamp(rw, 0, i32(params.inW) - 1));
|
| 161 |
+
{% endif %}
|
| 162 |
+
return f32(x[((base + dd) * params.inH + hh) * params.inW + ww]);
|
| 163 |
+
}
|
| 164 |
+
|
| 165 |
+
@compute @workgroup_size({{ tunables.WORKGROUP_SIZE }})
|
| 166 |
+
fn main(@builtin(global_invocation_id) gid: vec3<u32>, @builtin(num_workgroups) nwg: vec3<u32>) {
|
| 167 |
+
{{ flat_index_2d(guardInline=true) }}
|
| 168 |
+
|
| 169 |
+
let ow = i % params.outW;
|
| 170 |
+
var t = i / params.outW;
|
| 171 |
+
let oh = t % params.outH;
|
| 172 |
+
t = t / params.outH;
|
| 173 |
+
let od = t % params.outD;
|
| 174 |
+
t = t / params.outD;
|
| 175 |
+
let c = t % params.C;
|
| 176 |
+
let n = t / params.C;
|
| 177 |
+
|
| 178 |
+
let grid_base = (((n * params.outD + od) * params.outH + oh) * params.outW + ow) * 3u;
|
| 179 |
+
let sx = sanitize_coord(denormalize(f32(grid[grid_base]), params.inW));
|
| 180 |
+
let sy = sanitize_coord(denormalize(f32(grid[grid_base + 1u]), params.inH));
|
| 181 |
+
let sz = sanitize_coord(denormalize(f32(grid[grid_base + 2u]), params.inD));
|
| 182 |
+
let img_base = (n * params.C + c) * params.inD;
|
| 183 |
+
|
| 184 |
+
{% if source.mode == "nearest" %}
|
| 185 |
+
let result = voxel(img_base, sample_round(sz), sample_round(sy), sample_round(sx));
|
| 186 |
+
{% elif source.mode == "cubic" %}
|
| 187 |
+
// Tricubic is the separable product of the same 4-tap Keys kernel the 2D path
|
| 188 |
+
// uses, one axis at a time; voxel() already resolves every padding mode, so
|
| 189 |
+
// the 64-tap window needs no boundary handling of its own.
|
| 190 |
+
let x0 = sample_floor(sx) - 1;
|
| 191 |
+
let y0 = sample_floor(sy) - 1;
|
| 192 |
+
let z0 = sample_floor(sz) - 1;
|
| 193 |
+
let cx = cubic_coeffs(sx - f32(x0 + 1));
|
| 194 |
+
let cy = cubic_coeffs(sy - f32(y0 + 1));
|
| 195 |
+
let cz = cubic_coeffs(sz - f32(z0 + 1));
|
| 196 |
+
var result = 0.0;
|
| 197 |
+
for (var kz = 0i; kz < 4i; kz = kz + 1i) {
|
| 198 |
+
var plane = 0.0;
|
| 199 |
+
for (var ky = 0i; ky < 4i; ky = ky + 1i) {
|
| 200 |
+
let row = cx.x * voxel(img_base, z0 + kz, y0 + ky, x0)
|
| 201 |
+
+ cx.y * voxel(img_base, z0 + kz, y0 + ky, x0 + 1)
|
| 202 |
+
+ cx.z * voxel(img_base, z0 + kz, y0 + ky, x0 + 2)
|
| 203 |
+
+ cx.w * voxel(img_base, z0 + kz, y0 + ky, x0 + 3);
|
| 204 |
+
plane = plane + cy[u32(ky)] * row;
|
| 205 |
+
}
|
| 206 |
+
result = result + cz[u32(kz)] * plane;
|
| 207 |
+
}
|
| 208 |
+
{% else %}
|
| 209 |
+
let x0 = sample_floor(sx);
|
| 210 |
+
let y0 = sample_floor(sy);
|
| 211 |
+
let z0 = sample_floor(sz);
|
| 212 |
+
let wx = sx - f32(x0);
|
| 213 |
+
let wy = sy - f32(y0);
|
| 214 |
+
let wz = sz - f32(z0);
|
| 215 |
+
let v000 = voxel(img_base, z0, y0, x0);
|
| 216 |
+
let v001 = voxel(img_base, z0, y0, x0 + 1);
|
| 217 |
+
let v010 = voxel(img_base, z0, y0 + 1, x0);
|
| 218 |
+
let v011 = voxel(img_base, z0, y0 + 1, x0 + 1);
|
| 219 |
+
let v100 = voxel(img_base, z0 + 1, y0, x0);
|
| 220 |
+
let v101 = voxel(img_base, z0 + 1, y0, x0 + 1);
|
| 221 |
+
let v110 = voxel(img_base, z0 + 1, y0 + 1, x0);
|
| 222 |
+
let v111 = voxel(img_base, z0 + 1, y0 + 1, x0 + 1);
|
| 223 |
+
let front = (1.0 - wy) * ((1.0 - wx) * v000 + wx * v001) + wy * ((1.0 - wx) * v010 + wx * v011);
|
| 224 |
+
let back = (1.0 - wy) * ((1.0 - wx) * v100 + wx * v101) + wy * ((1.0 - wx) * v110 + wx * v111);
|
| 225 |
+
let result = (1.0 - wz) * front + wz * back;
|
| 226 |
+
{% endif %}
|
| 227 |
+
y[i] = {{ scalar }}(result);
|
| 228 |
+
}
|
build/webgpu/manifest.json
ADDED
|
@@ -0,0 +1,184 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"domain": "ai.onnx",
|
| 3 |
+
"name": "GridSample",
|
| 4 |
+
"sinceVersion": 20,
|
| 5 |
+
"description": "Samples values from input tensor `X` at positions defined by a flow-field `grid`, producing output `Y` with spatial dimensions taken from `grid`. Grid coordinates are normalized to `[-1, 1]` over the input spatial extent; positions outside this range are handled according to `padding_mode`. Supports spatial `(rank-4, NCHW)` and volumetric `(rank-5, NCDHW)` inputs with `linear`, `nearest`, or `cubic` interpolation.",
|
| 6 |
+
"inputs": [
|
| 7 |
+
{
|
| 8 |
+
"role": "X",
|
| 9 |
+
"dtype": "T",
|
| 10 |
+
"description": "Input tensor of shape `(N, C, D1, ..., Dr)` whose values are sampled."
|
| 11 |
+
},
|
| 12 |
+
{
|
| 13 |
+
"role": "grid",
|
| 14 |
+
"dtype": "T",
|
| 15 |
+
"description": "Flow-field of shape `(N, D1_out, ..., Dr_out, r)` with normalized sampling coordinates in `[-1, 1]`."
|
| 16 |
+
}
|
| 17 |
+
],
|
| 18 |
+
"outputs": [
|
| 19 |
+
{
|
| 20 |
+
"role": "Y",
|
| 21 |
+
"dtype": "T",
|
| 22 |
+
"rank": "ranks.grid",
|
| 23 |
+
"description": "Output tensor of shape `(N, C, D1_out, ..., Dr_out)` containing the interpolated samples.",
|
| 24 |
+
"shape": "prefix(shapes.X, 2) + prefix(suffix(shapes.grid, 1), ranks.grid - 2)"
|
| 25 |
+
}
|
| 26 |
+
],
|
| 27 |
+
"attributes": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 28 |
+
"attributeDescriptions": {
|
| 29 |
+
"mode": "Interpolation method: `linear` (bilinear or trilinear, depending on rank), `nearest`, or `cubic`. Cubic interpolation is supported for rank-4 (2-D spatial) inputs.",
|
| 30 |
+
"padding_mode": "How out-of-bound grid positions are handled: `zeros` pads with 0, `border` clamps to the border value, or `reflection` reflects coordinates back into the valid range.",
|
| 31 |
+
"align_corners": "When 1, extrema values `-1` and `1` map to the center of the corner pixels; when 0 (default) they map to the outer edge of corner pixels, making sampling resolution-agnostic."
|
| 32 |
+
},
|
| 33 |
+
"attributeConstraints": {
|
| 34 |
+
"mode": { "values": ["linear", "nearest", "cubic"] },
|
| 35 |
+
"padding_mode": { "values": ["zeros", "border", "reflection"] },
|
| 36 |
+
"align_corners": { "values": [0, 1] }
|
| 37 |
+
},
|
| 38 |
+
"typeConstraints": { "T": ["float32", "float16"] },
|
| 39 |
+
"args": {
|
| 40 |
+
"x": { "kind": "tensor", "semantic": "X", "role": "input" },
|
| 41 |
+
"grid": { "kind": "tensor", "semantic": "grid", "role": "input" },
|
| 42 |
+
"y": { "kind": "tensor", "semantic": "Y", "role": "output" }
|
| 43 |
+
},
|
| 44 |
+
"tunables": { "WORKGROUP_SIZE": 256 },
|
| 45 |
+
"derive": {
|
| 46 |
+
"wave32Adapter": "has(device.adapterInfo, \"subgroupMinSize\") and has(device.adapterInfo, \"subgroupMaxSize\") and device.adapterInfo.subgroupMinSize == 32 and device.adapterInfo.subgroupMaxSize == 32",
|
| 47 |
+
"reportedNonWave32Adapter": "not wave32Adapter and (has(device.adapterInfo, \"subgroupMinSize\") or has(device.adapterInfo, \"subgroupMaxSize\"))",
|
| 48 |
+
"rank4Ok": "ranks.X == 4 and ranks.grid == 4 and ranks.Y == 4 and dim(shapes.grid, 0) == dim(shapes.X, 0) and dim(shapes.grid, 3) == 2 and dim(shapes.Y, 0) == dim(shapes.X, 0) and dim(shapes.Y, 1) == dim(shapes.X, 1) and dim(shapes.Y, 2) == dim(shapes.grid, 1) and dim(shapes.Y, 3) == dim(shapes.grid, 2) and f16Ok(dtypes.T)",
|
| 49 |
+
"channelWidth": 4
|
| 50 |
+
},
|
| 51 |
+
"constants": { "scalar": "dtypes.T", "usesF16": "dtypes.T == \"f16\"" },
|
| 52 |
+
"bindingSets": {
|
| 53 |
+
"rank4": [
|
| 54 |
+
{ "name": "x", "arg": "x", "semantic": "X", "buffer": { "type": "read-only-storage" }, "elementType": "$scalar" },
|
| 55 |
+
{
|
| 56 |
+
"name": "grid",
|
| 57 |
+
"arg": "grid",
|
| 58 |
+
"semantic": "grid",
|
| 59 |
+
"buffer": { "type": "read-only-storage" },
|
| 60 |
+
"elementType": "$scalar"
|
| 61 |
+
},
|
| 62 |
+
{ "name": "y", "arg": "y", "semantic": "Y", "buffer": { "type": "storage" }, "elementType": "$scalar" },
|
| 63 |
+
{
|
| 64 |
+
"name": "params",
|
| 65 |
+
"semantic": "kernel.params",
|
| 66 |
+
"buffer": { "type": "uniform" },
|
| 67 |
+
"struct": {
|
| 68 |
+
"name": "Params",
|
| 69 |
+
"fields": [
|
| 70 |
+
{ "name": "count", "type": "u32", "value": "numel(shapes.Y)" },
|
| 71 |
+
{ "name": "C", "type": "u32", "value": "dim(shapes.X, 1)" },
|
| 72 |
+
{ "name": "inH", "type": "u32", "value": "dim(shapes.X, 2)" },
|
| 73 |
+
{ "name": "inW", "type": "u32", "value": "dim(shapes.X, 3)" },
|
| 74 |
+
{ "name": "outH", "type": "u32", "value": "dim(shapes.Y, 2)" },
|
| 75 |
+
{ "name": "outW", "type": "u32", "value": "dim(shapes.Y, 3)" }
|
| 76 |
+
]
|
| 77 |
+
}
|
| 78 |
+
}
|
| 79 |
+
]
|
| 80 |
+
},
|
| 81 |
+
"variants": [
|
| 82 |
+
{
|
| 83 |
+
"id": "nchw_rank4_channel_vector",
|
| 84 |
+
"priority": 5,
|
| 85 |
+
"when": ["rank4Ok", "not reportedNonWave32Adapter", "dim(shapes.X, 1) >= 2"],
|
| 86 |
+
"passes": [
|
| 87 |
+
{
|
| 88 |
+
"id": "main",
|
| 89 |
+
"name": "GridSample.ChannelX4",
|
| 90 |
+
"source": {
|
| 91 |
+
"shader": "grid-sample.wgsl.jinja",
|
| 92 |
+
"inputs": {
|
| 93 |
+
"mode": "attrs.mode",
|
| 94 |
+
"paddingMode": "attrs.padding_mode",
|
| 95 |
+
"alignCorners": "attrs.align_corners != 0",
|
| 96 |
+
"channelWidth": "channelWidth",
|
| 97 |
+
"channelTail": "dim(shapes.X, 1) % channelWidth != 0"
|
| 98 |
+
}
|
| 99 |
+
},
|
| 100 |
+
"bindings": "rank4",
|
| 101 |
+
"dispatch": {
|
| 102 |
+
"threads": "dim(shapes.Y, 0) * ceilDiv(dim(shapes.Y, 1), channelWidth) * dim(shapes.Y, 2) * dim(shapes.Y, 3)",
|
| 103 |
+
"workgroupSize": "tunables.WORKGROUP_SIZE"
|
| 104 |
+
}
|
| 105 |
+
}
|
| 106 |
+
]
|
| 107 |
+
},
|
| 108 |
+
{
|
| 109 |
+
"id": "nchw_rank4",
|
| 110 |
+
"when": ["rank4Ok"],
|
| 111 |
+
"passes": [
|
| 112 |
+
{
|
| 113 |
+
"id": "main",
|
| 114 |
+
"name": "GridSample",
|
| 115 |
+
"source": {
|
| 116 |
+
"shader": "grid-sample.wgsl.jinja",
|
| 117 |
+
"inputs": {
|
| 118 |
+
"mode": "attrs.mode",
|
| 119 |
+
"paddingMode": "attrs.padding_mode",
|
| 120 |
+
"alignCorners": "attrs.align_corners != 0"
|
| 121 |
+
}
|
| 122 |
+
},
|
| 123 |
+
"bindings": "rank4",
|
| 124 |
+
"dispatch": { "threads": "numel(shapes.Y)", "workgroupSize": "tunables.WORKGROUP_SIZE" }
|
| 125 |
+
}
|
| 126 |
+
]
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"id": "ncdhw_rank5",
|
| 130 |
+
"priority": 10,
|
| 131 |
+
"when": ["ranks.X == 5", "ranks.grid == 5", "ranks.Y == 5", "dim(shapes.grid, 0) == dim(shapes.X, 0)", "dim(shapes.grid, 4) == 3", "dim(shapes.Y, 0) == dim(shapes.X, 0)", "dim(shapes.Y, 1) == dim(shapes.X, 1)", "dim(shapes.Y, 2) == dim(shapes.grid, 1)", "dim(shapes.Y, 3) == dim(shapes.grid, 2)", "dim(shapes.Y, 4) == dim(shapes.grid, 3)", "(attrs.mode == \"linear\" or attrs.mode == \"nearest\" or attrs.mode == \"cubic\")", "(attrs.padding_mode == \"zeros\" or attrs.padding_mode == \"border\" or attrs.padding_mode == \"reflection\")", "f16Ok(dtypes.T)"],
|
| 132 |
+
"passes": [
|
| 133 |
+
{
|
| 134 |
+
"id": "main",
|
| 135 |
+
"name": "GridSample.Volumetric",
|
| 136 |
+
"source": {
|
| 137 |
+
"shader": "grid-sample3d.wgsl.jinja",
|
| 138 |
+
"inputs": {
|
| 139 |
+
"mode": "attrs.mode",
|
| 140 |
+
"paddingMode": "attrs.padding_mode",
|
| 141 |
+
"alignCorners": "attrs.align_corners != 0"
|
| 142 |
+
}
|
| 143 |
+
},
|
| 144 |
+
"bindings": [
|
| 145 |
+
{
|
| 146 |
+
"name": "x",
|
| 147 |
+
"arg": "x",
|
| 148 |
+
"semantic": "X",
|
| 149 |
+
"buffer": { "type": "read-only-storage" },
|
| 150 |
+
"elementType": "$scalar"
|
| 151 |
+
},
|
| 152 |
+
{
|
| 153 |
+
"name": "grid",
|
| 154 |
+
"arg": "grid",
|
| 155 |
+
"semantic": "grid",
|
| 156 |
+
"buffer": { "type": "read-only-storage" },
|
| 157 |
+
"elementType": "$scalar"
|
| 158 |
+
},
|
| 159 |
+
{ "name": "y", "arg": "y", "semantic": "Y", "buffer": { "type": "storage" }, "elementType": "$scalar" },
|
| 160 |
+
{
|
| 161 |
+
"name": "params",
|
| 162 |
+
"semantic": "kernel.params",
|
| 163 |
+
"buffer": { "type": "uniform" },
|
| 164 |
+
"struct": {
|
| 165 |
+
"name": "Params",
|
| 166 |
+
"fields": [
|
| 167 |
+
{ "name": "count", "type": "u32", "value": "numel(shapes.Y)" },
|
| 168 |
+
{ "name": "C", "type": "u32", "value": "dim(shapes.X, 1)" },
|
| 169 |
+
{ "name": "inD", "type": "u32", "value": "dim(shapes.X, 2)" },
|
| 170 |
+
{ "name": "inH", "type": "u32", "value": "dim(shapes.X, 3)" },
|
| 171 |
+
{ "name": "inW", "type": "u32", "value": "dim(shapes.X, 4)" },
|
| 172 |
+
{ "name": "outD", "type": "u32", "value": "dim(shapes.Y, 2)" },
|
| 173 |
+
{ "name": "outH", "type": "u32", "value": "dim(shapes.Y, 3)" },
|
| 174 |
+
{ "name": "outW", "type": "u32", "value": "dim(shapes.Y, 4)" }
|
| 175 |
+
]
|
| 176 |
+
}
|
| 177 |
+
}
|
| 178 |
+
],
|
| 179 |
+
"dispatch": { "threads": "numel(shapes.Y)", "workgroupSize": "tunables.WORKGROUP_SIZE" }
|
| 180 |
+
}
|
| 181 |
+
]
|
| 182 |
+
}
|
| 183 |
+
]
|
| 184 |
+
}
|
build/webgpu/metadata.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"name": "ai.onnx.GridSample",
|
| 3 |
+
"id": "_ai_onnx_gridsample_webgpu_3892192",
|
| 4 |
+
"version": 1,
|
| 5 |
+
"license": "Apache-2.0",
|
| 6 |
+
"backend": { "type": "webgpu" },
|
| 7 |
+
"digest": {
|
| 8 |
+
"algorithm": "sha256",
|
| 9 |
+
"files": {
|
| 10 |
+
"bench.json": "G6tTb9EsyvpKUZLsd4qAik9iHOhPJGsUBR3ah5o7L+w=",
|
| 11 |
+
"grid-sample.wgsl.jinja": "vGFCLAzhkxtkeC071yc3MrcZa3C5mqPT4+ZbjxJn5Zo=",
|
| 12 |
+
"grid-sample3d.wgsl.jinja": "cSEW1lJIrYaiamCfxoP1KBWbzqcRfe+00AtqpeXRUNA=",
|
| 13 |
+
"manifest.json": "iZOT6Iz9u/5yFWpFbpO+1EgSPRoW9YPluK9w2unD2wk=",
|
| 14 |
+
"test.json": "iD0xQONckmqQHJylOBRTC7z57/IvRn28fEdFgYX1TNI="
|
| 15 |
+
}
|
| 16 |
+
},
|
| 17 |
+
"provenance": { "kernel": { "sha": "2e7068faf55e7f43df740015f6d1ee49391a41c5", "dirty": false } },
|
| 18 |
+
"webgpu": { "manifestSpec": "1.0", "specialized": true, "opPath": "ops/ai.onnx.GridSample" }
|
| 19 |
+
}
|
build/webgpu/test.json
ADDED
|
@@ -0,0 +1,1965 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"op": "ai.onnx.GridSample",
|
| 3 |
+
"fixtureArrays": {
|
| 4 |
+
"onnx_backend_gridsample_volumetric_input_grid": [-1, -1, -1, -1, -0.5, 0.30000001192092896, -0.5, -0.5, -0.5, 1, -0.6000000238418579, -1, -0.20000000298023224, -0.20000000298023224, -0.20000000298023224, 0.4000000059604645, 0.20000000298023224, 0.6000000238418579, 0, 0, 0, -1, 0, 0, 0, 0, 0, -1, 1, 0, -0.20000000298023224, -0.20000000298023224, -0.20000000298023224, 1, 0.4000000059604645, -0.20000000298023224, 0.5, 0.5, 0.5, -1, -0.800000011920929, 0.800000011920929, 1, 1, 1, 0.4000000059604645, 0.6000000238418579, -0.30000001192092896],
|
| 5 |
+
"grid_sample_grid_16": [-1, -0.800000011920929, -0.6000000238418579, -0.5, -0.10000000149011612, -0.20000000298023224, 0.699999988079071, 0, 0, 0.4000000059604645, 0.20000000298023224, -0.20000000298023224, -0.30000001192092896, 0.5, -1, 1],
|
| 6 |
+
"onnx_backend_far_coords_zeros_padding_input_grid": [-10, -10, -5, -5, -0.2, -0.2, 10, 10, 10, 10, -0.2, -0.2, 5, 5, 10, 10],
|
| 7 |
+
"onnx_backend_gridsample_input_x": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
|
| 8 |
+
"onnx_backend_gridsample_aligncorners_true_input_grid": [-1, -1, -0.5, -0.5, -0.20000000298023224, -0.20000000298023224, 0, 0, 0, 0, -0.20000000298023224, -0.20000000298023224, 0.5, 0.5, 1, 1],
|
| 9 |
+
"onnx_backend_gridsample_border_padding_input_grid": [-10, -10, -5, -5, -0.20000000298023224, -0.20000000298023224, 10, 10, 10, 10, -0.20000000298023224, -0.20000000298023224, 5, 5, 10, 10]
|
| 10 |
+
},
|
| 11 |
+
"cases": [
|
| 12 |
+
{
|
| 13 |
+
"name": "channel_x4_cubic_reflection_c16_f32",
|
| 14 |
+
"attrs": { "mode": "cubic", "padding_mode": "reflection", "align_corners": 0 },
|
| 15 |
+
"inputs": {
|
| 16 |
+
"x": {
|
| 17 |
+
"dtype": "float32",
|
| 18 |
+
"shape": [1, 16, 4, 4],
|
| 19 |
+
"data": { "kind": "fillFloat32", "scale": 0.5, "sinStep": 0.17, "cosStep": 0.09 }
|
| 20 |
+
},
|
| 21 |
+
"grid": { "dtype": "float32", "shape": [1, 3, 3, 2], "data": { "kind": "linspace", "start": -1.4, "end": 1.4 } }
|
| 22 |
+
},
|
| 23 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 16, 3, 3], "tolerance": 0.0001, "relTolerance": 0.0001 } }
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"name": "channel_x4_cubic_reflection_align_corners_c4",
|
| 27 |
+
"provenance": {
|
| 28 |
+
"notes": "Cubic reflection padding with align_corners on over four channels checks the aligned reflection bounds 0 .. dim-1 on the channel-cooperative path."
|
| 29 |
+
},
|
| 30 |
+
"attrs": { "mode": "cubic", "padding_mode": "reflection", "align_corners": 1 },
|
| 31 |
+
"inputs": {
|
| 32 |
+
"x": {
|
| 33 |
+
"dtype": "float32",
|
| 34 |
+
"shape": [1, 4, 4, 4],
|
| 35 |
+
"data": { "kind": "fillFloat32", "scale": 0.5, "sinStep": 0.17, "cosStep": 0.09 }
|
| 36 |
+
},
|
| 37 |
+
"grid": { "dtype": "float32", "shape": [1, 3, 3, 2], "data": { "kind": "linspace", "start": -1.4, "end": 1.4 } }
|
| 38 |
+
},
|
| 39 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 3, 3], "tolerance": 0.0001, "relTolerance": 0.0001 } }
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"name": "channel_x4_linear_reflection_c4",
|
| 43 |
+
"provenance": {
|
| 44 |
+
"notes": "Bilinear reflection padding with align_corners off over four channels checks the half-pixel reflection bounds on the channel-cooperative path."
|
| 45 |
+
},
|
| 46 |
+
"attrs": { "mode": "linear", "padding_mode": "reflection", "align_corners": 0 },
|
| 47 |
+
"inputs": {
|
| 48 |
+
"x": {
|
| 49 |
+
"dtype": "float32",
|
| 50 |
+
"shape": [1, 4, 4, 5],
|
| 51 |
+
"data": { "kind": "fillFloat32", "scale": 0.5, "sinStep": 0.23, "cosStep": 0.11 }
|
| 52 |
+
},
|
| 53 |
+
"grid": { "dtype": "float32", "shape": [1, 3, 4, 2], "data": { "kind": "linspace", "start": -1.6, "end": 1.4 } }
|
| 54 |
+
},
|
| 55 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 4, 3, 4], "tolerance": 0.0001, "relTolerance": 0.0001 } }
|
| 56 |
+
},
|
| 57 |
+
{
|
| 58 |
+
"name": "channel_x4_linear_border_c8_f32",
|
| 59 |
+
"attrs": { "mode": "linear", "padding_mode": "border", "align_corners": 1 },
|
| 60 |
+
"inputs": {
|
| 61 |
+
"x": {
|
| 62 |
+
"dtype": "float32",
|
| 63 |
+
"shape": [1, 8, 5, 5],
|
| 64 |
+
"data": { "kind": "fillFloat32", "scale": 0.4, "sinStep": 0.13, "cosStep": 0.07 }
|
| 65 |
+
},
|
| 66 |
+
"grid": { "dtype": "float32", "shape": [1, 4, 4, 2], "data": { "kind": "linspace", "start": -1.2, "end": 1.2 } }
|
| 67 |
+
},
|
| 68 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 8, 4, 4], "tolerance": 0.00001, "relTolerance": 0.00001 } }
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"name": "linear_zeros_exact_subnormal_pixel_gpu_gap",
|
| 72 |
+
"skipGpu": {
|
| 73 |
+
"category": "permanent",
|
| 74 |
+
"reason": "Portable WGSL floating-point semantics do not guarantee preservation of the subnormal values required by this fixture. Backend evidence: Metal flushes denormals to zero in the ALU; bilinear/trilinear interpolation arithmetic flushes the subnormal pixel/voxel on GPU. Permanent FTZ limitation."
|
| 75 |
+
},
|
| 76 |
+
"provenance": {
|
| 77 |
+
"notes": "Exact corner sampling with align_corners=1 should copy the positive subnormal source pixel through the bilinear path."
|
| 78 |
+
},
|
| 79 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 1 },
|
| 80 |
+
"inputs": {
|
| 81 |
+
"x": {
|
| 82 |
+
"dtype": "float32",
|
| 83 |
+
"shape": [1, 1, 2, 2],
|
| 84 |
+
"data": { "kind": "values", "values": [1e-40, 0.0, 0.0, 0.0] }
|
| 85 |
+
},
|
| 86 |
+
"grid": { "dtype": "float32", "shape": [1, 1, 1, 2], "data": { "kind": "values", "values": [-1.0, -1.0] } }
|
| 87 |
+
},
|
| 88 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 1], "tolerance": 0 } }
|
| 89 |
+
},
|
| 90 |
+
{
|
| 91 |
+
"name": "linear_zeros_exact_subnormal_voxel_gpu_gap",
|
| 92 |
+
"skipGpu": {
|
| 93 |
+
"category": "permanent",
|
| 94 |
+
"reason": "Portable WGSL floating-point semantics do not guarantee preservation of the subnormal values required by this fixture. Backend evidence: Metal flushes denormals to zero in the ALU; bilinear/trilinear interpolation arithmetic flushes the subnormal pixel/voxel on GPU. Permanent FTZ limitation."
|
| 95 |
+
},
|
| 96 |
+
"provenance": {
|
| 97 |
+
"notes": "Rank-5 companion: exact corner sampling should copy the positive subnormal source voxel through the trilinear path."
|
| 98 |
+
},
|
| 99 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 1 },
|
| 100 |
+
"inputs": {
|
| 101 |
+
"x": {
|
| 102 |
+
"dtype": "float32",
|
| 103 |
+
"shape": [1, 1, 2, 2, 2],
|
| 104 |
+
"data": { "kind": "values", "values": [1e-40, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0] }
|
| 105 |
+
},
|
| 106 |
+
"grid": {
|
| 107 |
+
"dtype": "float32",
|
| 108 |
+
"shape": [1, 1, 1, 1, 3],
|
| 109 |
+
"data": { "kind": "values", "values": [-1.0, -1.0, -1.0] }
|
| 110 |
+
}
|
| 111 |
+
},
|
| 112 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 1, 1], "tolerance": 0 } }
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"name": "dispatch_cliff_nchw_over_16m_elements",
|
| 116 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 117 |
+
"inputs": {
|
| 118 |
+
"x": { "dtype": "float32", "shape": [1, 256, 8, 8] },
|
| 119 |
+
"grid": {
|
| 120 |
+
"dtype": "float32",
|
| 121 |
+
"shape": [1, 256, 256, 2],
|
| 122 |
+
"data": { "kind": "linspace", "start": -1.0, "end": 1.0 }
|
| 123 |
+
}
|
| 124 |
+
},
|
| 125 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 256, 256, 256], "tolerance": 0.0001 } }
|
| 126 |
+
},
|
| 127 |
+
{
|
| 128 |
+
"name": "dispatch_cliff_ncdhw_over_16m_elements",
|
| 129 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 130 |
+
"inputs": {
|
| 131 |
+
"x": { "dtype": "float32", "shape": [1, 64, 8, 8, 8] },
|
| 132 |
+
"grid": {
|
| 133 |
+
"dtype": "float32",
|
| 134 |
+
"shape": [1, 64, 64, 64, 3],
|
| 135 |
+
"data": { "kind": "linspace", "start": -1.0, "end": 1.0 }
|
| 136 |
+
}
|
| 137 |
+
},
|
| 138 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 64, 64, 64, 64], "tolerance": 0.0001 } }
|
| 139 |
+
},
|
| 140 |
+
{
|
| 141 |
+
"name": "linear_zeros_align_corners",
|
| 142 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 1 },
|
| 143 |
+
"inputs": {
|
| 144 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2], "data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0] } },
|
| 145 |
+
"grid": {
|
| 146 |
+
"dtype": "float32",
|
| 147 |
+
"shape": [1, 2, 2, 2],
|
| 148 |
+
"data": { "kind": "values", "values": [-1.0, -1.0, 1.0, -1.0, -1.0, 1.0, 1.0, 1.0] }
|
| 149 |
+
}
|
| 150 |
+
},
|
| 151 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 2] } },
|
| 152 |
+
"tolerance": 0.000001
|
| 153 |
+
},
|
| 154 |
+
{
|
| 155 |
+
"name": "ort_linear_zeros_align_corners_rank4",
|
| 156 |
+
"provenance": {
|
| 157 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_gen.py",
|
| 158 |
+
"test": "GridSampleTest.test_grid_sample_16_4D_bilinear_zeros_align_corners",
|
| 159 |
+
"notes": "Generated ORT fixture materialized in grid_sample_test.cc; opset-20 spelling uses mode=linear for the 4D bilinear case."
|
| 160 |
+
},
|
| 161 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 1 },
|
| 162 |
+
"inputs": {
|
| 163 |
+
"x": {
|
| 164 |
+
"dtype": "float32",
|
| 165 |
+
"shape": [2, 2, 3, 2],
|
| 166 |
+
"data": {
|
| 167 |
+
"kind": "values",
|
| 168 |
+
"values": [0.294201, 0.797322, 1.264215, 0.935492, 0.545464, -1.537389, 0.312439, 0.74006, -0.575326, -1.432532, -0.666175, 1.017438, -2.241368, 0.437349, -0.555362, -0.057943, 0.658583, 0.992938, -0.206548, -0.244841, -0.380599, 1.131112, -0.090205, -0.8979]
|
| 169 |
+
}
|
| 170 |
+
},
|
| 171 |
+
"grid": {
|
| 172 |
+
"dtype": "float32",
|
| 173 |
+
"shape": [2, 3, 2, 2],
|
| 174 |
+
"data": {
|
| 175 |
+
"kind": "values",
|
| 176 |
+
"values": [0.595248, -1.096726, -0.214731, -0.891773, -0.512023, 0.432352, -0.852156, 0.446072, 1.018534, 0.078706, -0.799785, -0.429942, 0.262037, -0.914782, 0.596172, -1.089444, -1.153552, -1.165993, -0.243436, 0.80692, -1.135775, 0.997425, -0.480027, 0.351461]
|
| 177 |
+
}
|
| 178 |
+
}
|
| 179 |
+
},
|
| 180 |
+
"outputs": { "y": { "dtype": "float32", "shape": [2, 2, 3, 2], "tolerance": 0.00001 } }
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"name": "nearest_border",
|
| 184 |
+
"attrs": { "mode": "nearest", "padding_mode": "border", "align_corners": 1 },
|
| 185 |
+
"inputs": {
|
| 186 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2], "data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0] } },
|
| 187 |
+
"grid": {
|
| 188 |
+
"dtype": "float32",
|
| 189 |
+
"shape": [1, 1, 3, 2],
|
| 190 |
+
"data": { "kind": "values", "values": [-2.0, -2.0, 0.0, 0.0, 2.0, 2.0] }
|
| 191 |
+
}
|
| 192 |
+
},
|
| 193 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 3] } }
|
| 194 |
+
},
|
| 195 |
+
{
|
| 196 |
+
"name": "linear_f16",
|
| 197 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 198 |
+
"inputs": {
|
| 199 |
+
"x": { "dtype": "float16", "shape": [1, 1, 2, 2], "data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0] } },
|
| 200 |
+
"grid": {
|
| 201 |
+
"dtype": "float16",
|
| 202 |
+
"shape": [1, 1, 2, 2],
|
| 203 |
+
"data": { "kind": "values", "values": [0.0, 0.0, 1.0, 0.0] }
|
| 204 |
+
}
|
| 205 |
+
},
|
| 206 |
+
"outputs": { "y": { "dtype": "float16", "shape": [1, 1, 1, 2] } },
|
| 207 |
+
"tolerance": 0.002
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"name": "nearest_reflection_extreme_coords_ort",
|
| 211 |
+
"provenance": {
|
| 212 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 213 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_nearest_reflection_extreme_coords"
|
| 214 |
+
},
|
| 215 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 0 },
|
| 216 |
+
"inputs": {
|
| 217 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2], "data": { "kind": "constant", "value": 1.0 } },
|
| 218 |
+
"grid": {
|
| 219 |
+
"dtype": "float32",
|
| 220 |
+
"shape": [1, 1, 2, 2],
|
| 221 |
+
"data": { "kind": "values", "values": [10000000000.0, 10000000000.0, -10000000000.0, -10000000000.0] }
|
| 222 |
+
}
|
| 223 |
+
},
|
| 224 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 2], "tolerance": 0.000001 } }
|
| 225 |
+
},
|
| 226 |
+
{
|
| 227 |
+
"name": "linear_reflection_far_coords_ort",
|
| 228 |
+
"provenance": {
|
| 229 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 230 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_bilinear_reflection_extreme_coords",
|
| 231 |
+
"notes": "Uses smaller finite coordinates than ORT's extreme-coordinate regression while preserving the reflection boundary path."
|
| 232 |
+
},
|
| 233 |
+
"attrs": { "mode": "linear", "padding_mode": "reflection", "align_corners": 0 },
|
| 234 |
+
"inputs": {
|
| 235 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2], "data": { "kind": "constant", "value": 1.0 } },
|
| 236 |
+
"grid": {
|
| 237 |
+
"dtype": "float32",
|
| 238 |
+
"shape": [1, 1, 2, 2],
|
| 239 |
+
"data": { "kind": "values", "values": [5.0, 5.0, -5.0, -5.0] }
|
| 240 |
+
}
|
| 241 |
+
},
|
| 242 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 2], "tolerance": 0.000001 } }
|
| 243 |
+
},
|
| 244 |
+
{
|
| 245 |
+
"name": "cubic_reflection_extreme_coords_ort",
|
| 246 |
+
"provenance": {
|
| 247 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 248 |
+
"test": "GridSampleCustomTestFloatOnly.test_grid_sample_20_4D_cubic_reflection_extreme_coords"
|
| 249 |
+
},
|
| 250 |
+
"attrs": { "mode": "cubic", "padding_mode": "reflection", "align_corners": 0 },
|
| 251 |
+
"inputs": {
|
| 252 |
+
"x": { "dtype": "float32", "shape": [1, 1, 4, 4], "data": { "kind": "constant", "value": 1.0 } },
|
| 253 |
+
"grid": {
|
| 254 |
+
"dtype": "float32",
|
| 255 |
+
"shape": [1, 1, 1, 2],
|
| 256 |
+
"data": { "kind": "values", "values": [10000000000.0, -10000000000.0] }
|
| 257 |
+
}
|
| 258 |
+
},
|
| 259 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 1], "tolerance": 0.0001 } }
|
| 260 |
+
},
|
| 261 |
+
{
|
| 262 |
+
"name": "nearest_reflection_extreme_coords_rank5",
|
| 263 |
+
"provenance": {
|
| 264 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 265 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_5D_nearest_reflection_extreme_coords",
|
| 266 |
+
"notes": "Covers ONNX 5D/volumetric nearest sampling with reflection padding."
|
| 267 |
+
},
|
| 268 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 0 },
|
| 269 |
+
"inputs": {
|
| 270 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2, 2], "data": { "kind": "constant", "value": 1.0 } },
|
| 271 |
+
"grid": {
|
| 272 |
+
"dtype": "float32",
|
| 273 |
+
"shape": [1, 1, 1, 2, 3],
|
| 274 |
+
"data": {
|
| 275 |
+
"kind": "values",
|
| 276 |
+
"values": [10000000000.0, 10000000000.0, 10000000000.0, -10000000000.0, -10000000000.0, -10000000000.0]
|
| 277 |
+
}
|
| 278 |
+
}
|
| 279 |
+
},
|
| 280 |
+
"outputs": {
|
| 281 |
+
"y": {
|
| 282 |
+
"dtype": "float32",
|
| 283 |
+
"shape": [1, 1, 1, 1, 2],
|
| 284 |
+
"tolerance": 0.000001,
|
| 285 |
+
"data": { "kind": "values", "values": [1.0, 1.0] }
|
| 286 |
+
}
|
| 287 |
+
}
|
| 288 |
+
},
|
| 289 |
+
{
|
| 290 |
+
"name": "linear_zeros_mixed_bounds_right_bottom_ort",
|
| 291 |
+
"provenance": {
|
| 292 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 293 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_linear_zeros_mixed_bounds_right_bottom",
|
| 294 |
+
"notes": "Projection onto a 3x3 source image; it preserves ORT's right/bottom zero-padding boundary behavior."
|
| 295 |
+
},
|
| 296 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 297 |
+
"inputs": {
|
| 298 |
+
"x": {
|
| 299 |
+
"dtype": "float32",
|
| 300 |
+
"shape": [1, 1, 3, 3],
|
| 301 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0] }
|
| 302 |
+
},
|
| 303 |
+
"grid": {
|
| 304 |
+
"dtype": "float32",
|
| 305 |
+
"shape": [1, 1, 4, 2],
|
| 306 |
+
"data": { "kind": "values", "values": [1.0, 1.0, 0.8, 1.0, 1.0, 0.8, 1.2, 1.2] }
|
| 307 |
+
}
|
| 308 |
+
},
|
| 309 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 4], "tolerance": 0.000001 } }
|
| 310 |
+
},
|
| 311 |
+
{
|
| 312 |
+
"name": "linear_border_batch2_channels2_ort",
|
| 313 |
+
"provenance": {
|
| 314 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 315 |
+
"test": "GridSampleTest.test_grid_sample_20_4D_bilinear_border_no_align_corners",
|
| 316 |
+
"notes": "Projection using deterministic generated data to cover batch and channel indexing in the same border/no-align mode."
|
| 317 |
+
},
|
| 318 |
+
"attrs": { "mode": "linear", "padding_mode": "border", "align_corners": 0 },
|
| 319 |
+
"inputs": {
|
| 320 |
+
"x": {
|
| 321 |
+
"dtype": "float32",
|
| 322 |
+
"shape": [2, 2, 3, 2],
|
| 323 |
+
"data": { "kind": "fillFloat32", "scale": 0.7, "sinStep": 0.13, "cosStep": 0.29 }
|
| 324 |
+
},
|
| 325 |
+
"grid": {
|
| 326 |
+
"dtype": "float32",
|
| 327 |
+
"shape": [2, 2, 2, 2],
|
| 328 |
+
"data": {
|
| 329 |
+
"kind": "values",
|
| 330 |
+
"values": [-1.1, -0.9, 0.25, -0.25, 0.8, 0.6, 1.2, 1.1, -0.4, 0.7, 0.0, 0.0, 0.9, -1.2, -1.3, 1.3]
|
| 331 |
+
}
|
| 332 |
+
}
|
| 333 |
+
},
|
| 334 |
+
"outputs": { "y": { "dtype": "float32", "shape": [2, 2, 2, 2], "tolerance": 0.000001 } }
|
| 335 |
+
},
|
| 336 |
+
{
|
| 337 |
+
"name": "cubic_zeros_fractional_multichannel",
|
| 338 |
+
"attrs": { "mode": "cubic", "padding_mode": "zeros", "align_corners": 1 },
|
| 339 |
+
"inputs": {
|
| 340 |
+
"x": {
|
| 341 |
+
"dtype": "float32",
|
| 342 |
+
"shape": [1, 2, 4, 4],
|
| 343 |
+
"data": { "kind": "fillFloat32", "scale": 0.5, "sinStep": 0.17, "cosStep": 0.11 }
|
| 344 |
+
},
|
| 345 |
+
"grid": {
|
| 346 |
+
"dtype": "float32",
|
| 347 |
+
"shape": [1, 2, 2, 2],
|
| 348 |
+
"data": { "kind": "values", "values": [-0.5, -0.5, 0.25, -0.25, 0.75, 0.5, 1.2, -1.2] }
|
| 349 |
+
}
|
| 350 |
+
},
|
| 351 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 2, 2, 2], "tolerance": 0.0001 } }
|
| 352 |
+
},
|
| 353 |
+
{
|
| 354 |
+
"name": "ort_linear_border_align_corners_compact",
|
| 355 |
+
"provenance": {
|
| 356 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 357 |
+
"test": "GridSampleTest.test_grid_sample_16_4D_bilinear_border_align_corners",
|
| 358 |
+
"notes": "Compact opset-20 projection using mode=linear for ORT's generated 4D bilinear border align-corners case."
|
| 359 |
+
},
|
| 360 |
+
"attrs": { "mode": "linear", "padding_mode": "border", "align_corners": 1 },
|
| 361 |
+
"inputs": {
|
| 362 |
+
"x": {
|
| 363 |
+
"dtype": "float32",
|
| 364 |
+
"shape": [1, 1, 2, 3],
|
| 365 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0] }
|
| 366 |
+
},
|
| 367 |
+
"grid": {
|
| 368 |
+
"dtype": "float32",
|
| 369 |
+
"shape": [1, 2, 3, 2],
|
| 370 |
+
"data": { "kind": "values", "values": [-1.5, -1.0, 0.0, 0.0, 1.5, 1.0, -1.0, 1.2, 0.5, -1.2, 1.0, 0.0] }
|
| 371 |
+
}
|
| 372 |
+
},
|
| 373 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 3], "tolerance": 0.00001 } }
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"name": "ort_linear_reflection_align_corners_compact",
|
| 377 |
+
"provenance": {
|
| 378 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 379 |
+
"test": "GridSampleTest.test_grid_sample_20_4D_bilinear_reflection_align_corners",
|
| 380 |
+
"notes": "Compact opset-20 projection using mode=linear for ORT's generated 4D bilinear reflection align-corners case."
|
| 381 |
+
},
|
| 382 |
+
"attrs": { "mode": "linear", "padding_mode": "reflection", "align_corners": 1 },
|
| 383 |
+
"inputs": {
|
| 384 |
+
"x": {
|
| 385 |
+
"dtype": "float32",
|
| 386 |
+
"shape": [1, 1, 3, 3],
|
| 387 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 4.0, 8.0, 16.0, 32.0, 64.0, 128.0, 256.0] }
|
| 388 |
+
},
|
| 389 |
+
"grid": {
|
| 390 |
+
"dtype": "float32",
|
| 391 |
+
"shape": [1, 2, 2, 2],
|
| 392 |
+
"data": { "kind": "values", "values": [-1.4, -1.4, 1.4, 1.4, 0.25, -0.5, -2.2, 0.6] }
|
| 393 |
+
}
|
| 394 |
+
},
|
| 395 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 2], "tolerance": 0.00001 } }
|
| 396 |
+
},
|
| 397 |
+
{
|
| 398 |
+
"name": "ort_cubic_border_no_align_corners_compact",
|
| 399 |
+
"provenance": {
|
| 400 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 401 |
+
"test": "GridSampleTest.test_grid_sample_20_4D_bicubic_border_no_align_corners",
|
| 402 |
+
"notes": "Compact projection of ORT's generated 4D bicubic border no-align case; opset-20 spelling uses mode=cubic."
|
| 403 |
+
},
|
| 404 |
+
"attrs": { "mode": "cubic", "padding_mode": "border", "align_corners": 0 },
|
| 405 |
+
"inputs": {
|
| 406 |
+
"x": {
|
| 407 |
+
"dtype": "float32",
|
| 408 |
+
"shape": [1, 1, 3, 3],
|
| 409 |
+
"data": { "kind": "values", "values": [-2.0, -1.0, 0.0, 1.0, 3.0, 5.0, 8.0, 13.0, 21.0] }
|
| 410 |
+
},
|
| 411 |
+
"grid": {
|
| 412 |
+
"dtype": "float32",
|
| 413 |
+
"shape": [1, 2, 2, 2],
|
| 414 |
+
"data": { "kind": "values", "values": [0.2, -0.6, 1.2, 1.2, -1.1, 0.0, 0.0, 0.8] }
|
| 415 |
+
}
|
| 416 |
+
},
|
| 417 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 2], "tolerance": 0.00001 } }
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"name": "ort_cubic_reflection_align_corners_compact",
|
| 421 |
+
"provenance": {
|
| 422 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 423 |
+
"test": "GridSampleTest.test_grid_sample_20_4D_bicubic_reflection_align_corners",
|
| 424 |
+
"notes": "Compact projection of ORT's generated 4D bicubic reflection align-corners case; opset-20 spelling uses mode=cubic."
|
| 425 |
+
},
|
| 426 |
+
"attrs": { "mode": "cubic", "padding_mode": "reflection", "align_corners": 1 },
|
| 427 |
+
"inputs": {
|
| 428 |
+
"x": {
|
| 429 |
+
"dtype": "float32",
|
| 430 |
+
"shape": [1, 1, 3, 3],
|
| 431 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 0.0, 2.0, 4.0, 2.0, 0.0, 1.0, 0.0] }
|
| 432 |
+
},
|
| 433 |
+
"grid": {
|
| 434 |
+
"dtype": "float32",
|
| 435 |
+
"shape": [1, 2, 2, 2],
|
| 436 |
+
"data": { "kind": "values", "values": [-1.25, -1.25, 1.25, -0.25, -0.6, 1.3, 0.3, 0.4] }
|
| 437 |
+
}
|
| 438 |
+
},
|
| 439 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 2], "tolerance": 0.00001 } }
|
| 440 |
+
},
|
| 441 |
+
{
|
| 442 |
+
"name": "nearest_reflection_dim1_align_corners_ort",
|
| 443 |
+
"provenance": {
|
| 444 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 445 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_nearest_reflection_dim1_align_corners",
|
| 446 |
+
"notes": "Extends ORT's dim-1 reflection case to a width-3 input so the same row-degenerate path also checks horizontal sampling."
|
| 447 |
+
},
|
| 448 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 1 },
|
| 449 |
+
"inputs": {
|
| 450 |
+
"x": { "dtype": "float32", "shape": [1, 1, 1, 3], "data": { "kind": "values", "values": [10.0, 20.0, 30.0] } },
|
| 451 |
+
"grid": {
|
| 452 |
+
"dtype": "float32",
|
| 453 |
+
"shape": [1, 1, 5, 2],
|
| 454 |
+
"data": { "kind": "values", "values": [-2.0, -2.0, -1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 2.0, 2.0] }
|
| 455 |
+
}
|
| 456 |
+
},
|
| 457 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 5], "tolerance": 0.000001 } }
|
| 458 |
+
},
|
| 459 |
+
{
|
| 460 |
+
"name": "onnx_backend_far_coords_zeros_padding",
|
| 461 |
+
"provenance": {
|
| 462 |
+
"source": "cmake/external/onnx/onnx/backend/test/case/node/gridsample.py",
|
| 463 |
+
"test": "test_gridsample_zeros_padding"
|
| 464 |
+
},
|
| 465 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 466 |
+
"inputs": {
|
| 467 |
+
"x": {
|
| 468 |
+
"dtype": "float32",
|
| 469 |
+
"shape": [1, 1, 3, 2],
|
| 470 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 471 |
+
},
|
| 472 |
+
"grid": {
|
| 473 |
+
"dtype": "float32",
|
| 474 |
+
"shape": [1, 2, 4, 2],
|
| 475 |
+
"data": {
|
| 476 |
+
"kind": "values",
|
| 477 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_far_coords_zeros_padding_input_grid" }
|
| 478 |
+
}
|
| 479 |
+
}
|
| 480 |
+
},
|
| 481 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } }
|
| 482 |
+
},
|
| 483 |
+
{
|
| 484 |
+
"name": "onnx_backend_far_coords_border_padding",
|
| 485 |
+
"provenance": {
|
| 486 |
+
"source": "cmake/external/onnx/onnx/backend/test/case/node/gridsample.py",
|
| 487 |
+
"test": "test_gridsample_border_padding"
|
| 488 |
+
},
|
| 489 |
+
"attrs": { "mode": "linear", "padding_mode": "border", "align_corners": 0 },
|
| 490 |
+
"inputs": {
|
| 491 |
+
"x": {
|
| 492 |
+
"dtype": "float32",
|
| 493 |
+
"shape": [1, 1, 3, 2],
|
| 494 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 495 |
+
},
|
| 496 |
+
"grid": {
|
| 497 |
+
"dtype": "float32",
|
| 498 |
+
"shape": [1, 2, 4, 2],
|
| 499 |
+
"data": {
|
| 500 |
+
"kind": "values",
|
| 501 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_far_coords_zeros_padding_input_grid" }
|
| 502 |
+
}
|
| 503 |
+
}
|
| 504 |
+
},
|
| 505 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } }
|
| 506 |
+
},
|
| 507 |
+
{
|
| 508 |
+
"name": "onnx_backend_far_coords_reflection_padding",
|
| 509 |
+
"provenance": {
|
| 510 |
+
"source": "cmake/external/onnx/onnx/backend/test/case/node/gridsample.py",
|
| 511 |
+
"test": "test_gridsample_reflection_padding"
|
| 512 |
+
},
|
| 513 |
+
"attrs": { "mode": "linear", "padding_mode": "reflection", "align_corners": 0 },
|
| 514 |
+
"inputs": {
|
| 515 |
+
"x": {
|
| 516 |
+
"dtype": "float32",
|
| 517 |
+
"shape": [1, 1, 3, 2],
|
| 518 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 519 |
+
},
|
| 520 |
+
"grid": {
|
| 521 |
+
"dtype": "float32",
|
| 522 |
+
"shape": [1, 2, 4, 2],
|
| 523 |
+
"data": {
|
| 524 |
+
"kind": "values",
|
| 525 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_far_coords_zeros_padding_input_grid" }
|
| 526 |
+
}
|
| 527 |
+
}
|
| 528 |
+
},
|
| 529 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } }
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"name": "onnx_backend_cubic_mode",
|
| 533 |
+
"provenance": {
|
| 534 |
+
"source": "cmake/external/onnx/onnx/backend/test/case/node/gridsample.py",
|
| 535 |
+
"test": "test_gridsample_bicubic"
|
| 536 |
+
},
|
| 537 |
+
"attrs": { "mode": "cubic", "padding_mode": "zeros", "align_corners": 0 },
|
| 538 |
+
"inputs": {
|
| 539 |
+
"x": {
|
| 540 |
+
"dtype": "float32",
|
| 541 |
+
"shape": [1, 1, 3, 2],
|
| 542 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 543 |
+
},
|
| 544 |
+
"grid": {
|
| 545 |
+
"dtype": "float32",
|
| 546 |
+
"shape": [1, 2, 4, 2],
|
| 547 |
+
"data": {
|
| 548 |
+
"kind": "values",
|
| 549 |
+
"values": [-1.0, -1.0, -0.5, -0.5, -0.2, -0.2, 0.0, 0.0, 0.0, 0.0, -0.2, -0.2, 0.5, 0.5, 1.0, 1.0]
|
| 550 |
+
}
|
| 551 |
+
}
|
| 552 |
+
},
|
| 553 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0002 } }
|
| 554 |
+
},
|
| 555 |
+
{
|
| 556 |
+
"name": "ort_nearest_zeros_align",
|
| 557 |
+
"provenance": {
|
| 558 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 559 |
+
"test": "GridSampleTest.test_grid_sample_16_4D_nearest_zeros_align_corners"
|
| 560 |
+
},
|
| 561 |
+
"attrs": { "mode": "nearest", "padding_mode": "zeros", "align_corners": 1 },
|
| 562 |
+
"inputs": {
|
| 563 |
+
"x": {
|
| 564 |
+
"dtype": "float32",
|
| 565 |
+
"shape": [2, 2, 3, 2],
|
| 566 |
+
"data": {
|
| 567 |
+
"kind": "values",
|
| 568 |
+
"values": [-1.12584, -1.15236, -0.250579, -0.433879, 0.84871, 0.692009, -0.316013, -2.115219, 0.468096, -0.157712, 1.44366, 0.266049, 0.166455, 0.874382, -0.143474, -0.111609, 0.931827, 1.259009, 2.004981, 0.053737, 0.618057, -0.412802, -0.841065, -2.316042]
|
| 569 |
+
}
|
| 570 |
+
},
|
| 571 |
+
"grid": {
|
| 572 |
+
"dtype": "float32",
|
| 573 |
+
"shape": [2, 3, 2, 2],
|
| 574 |
+
"data": {
|
| 575 |
+
"kind": "values",
|
| 576 |
+
"values": [0.06311, -0.61522, 0.203022, -1.120434, -0.867079, -0.618636, 0.757125, 0.703586, -0.532194, -0.043299, 0.767473, 1.19296, 0.476259, 0.162111, 0.804584, -0.706563, 0.223613, -0.930367, -0.831703, -0.6199, 0.542968, 0.482592, -0.710823, 0.362529]
|
| 577 |
+
}
|
| 578 |
+
}
|
| 579 |
+
},
|
| 580 |
+
"outputs": { "y": { "dtype": "float32", "shape": [2, 2, 3, 2], "tolerance": 0 } }
|
| 581 |
+
},
|
| 582 |
+
{
|
| 583 |
+
"name": "ort_nearest_zeros_no_align",
|
| 584 |
+
"provenance": {
|
| 585 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 586 |
+
"test": "GridSampleTest.test_grid_sample_16_4D_nearest_zeros_no_align_corners"
|
| 587 |
+
},
|
| 588 |
+
"attrs": { "mode": "nearest", "padding_mode": "zeros", "align_corners": 0 },
|
| 589 |
+
"inputs": {
|
| 590 |
+
"x": {
|
| 591 |
+
"dtype": "float32",
|
| 592 |
+
"shape": [2, 2, 3, 2],
|
| 593 |
+
"data": {
|
| 594 |
+
"kind": "values",
|
| 595 |
+
"values": [-0.569248, 0.919971, 1.110816, 1.289874, -1.478174, 2.567233, -0.47312, 0.335551, -0.003304, -0.534441, 1.168688, 0.394503, 1.941462, 0.791498, -0.020252, -0.43717, -1.535287, -0.412679, 0.966303, 1.624783, -0.365619, -1.30244, 0.099403, 0.441822]
|
| 596 |
+
}
|
| 597 |
+
},
|
| 598 |
+
"grid": {
|
| 599 |
+
"dtype": "float32",
|
| 600 |
+
"shape": [2, 3, 2, 2],
|
| 601 |
+
"data": {
|
| 602 |
+
"kind": "values",
|
| 603 |
+
"values": [-1.143118, -0.021569, -0.903671, -0.925628, -0.06612, 0.180174, -0.491436, 0.712053, -0.730247, 1.088844, 0.82236, -1.01194, -0.298661, 0.054147, 0.175081, 0.284609, 0.470914, 0.07188, -0.585515, 0.567827, -1.151099, -0.711248, -0.300396, -0.584536]
|
| 604 |
+
}
|
| 605 |
+
}
|
| 606 |
+
},
|
| 607 |
+
"outputs": { "y": { "dtype": "float32", "shape": [2, 2, 3, 2], "tolerance": 0 } }
|
| 608 |
+
},
|
| 609 |
+
{
|
| 610 |
+
"name": "ort_nearest_border_align",
|
| 611 |
+
"provenance": {
|
| 612 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 613 |
+
"test": "GridSampleTest.test_grid_sample_16_4D_nearest_border_align_corners"
|
| 614 |
+
},
|
| 615 |
+
"attrs": { "mode": "nearest", "padding_mode": "border", "align_corners": 1 },
|
| 616 |
+
"inputs": {
|
| 617 |
+
"x": {
|
| 618 |
+
"dtype": "float32",
|
| 619 |
+
"shape": [2, 2, 3, 2],
|
| 620 |
+
"data": {
|
| 621 |
+
"kind": "values",
|
| 622 |
+
"values": [-0.883376, -0.418913, -0.804826, 0.56561, 0.610365, 0.466884, 1.950657, -1.063099, -0.829367, -1.407257, 1.626847, 0.172273, -1.611502, -0.479448, -0.143351, -0.317295, 0.573655, 0.997931, 0.543609, 0.078804, 0.86286, -0.01949, 0.991047, -0.777735]
|
| 623 |
+
}
|
| 624 |
+
},
|
| 625 |
+
"grid": {
|
| 626 |
+
"dtype": "float32",
|
| 627 |
+
"shape": [2, 3, 2, 2],
|
| 628 |
+
"data": {
|
| 629 |
+
"kind": "values",
|
| 630 |
+
"values": [-1.08007, -0.080985, 1.055303, -0.48947, 1.083604, 0.434584, -1.082953, 0.759237, -0.138473, -0.535688, 0.959584, -0.969714, 0.128766, -0.251242, 0.856935, 0.334973, 0.576606, 0.423791, -0.28857, -0.252367, -0.988898, 0.650213, 0.952774, 0.82107]
|
| 631 |
+
}
|
| 632 |
+
}
|
| 633 |
+
},
|
| 634 |
+
"outputs": { "y": { "dtype": "float32", "shape": [2, 2, 3, 2], "tolerance": 0 } }
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"name": "ort_nearest_border_no_align",
|
| 638 |
+
"provenance": {
|
| 639 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 640 |
+
"test": "GridSampleTest.test_grid_sample_16_4D_nearest_border_no_align_corners"
|
| 641 |
+
},
|
| 642 |
+
"attrs": { "mode": "nearest", "padding_mode": "border", "align_corners": 0 },
|
| 643 |
+
"inputs": {
|
| 644 |
+
"x": {
|
| 645 |
+
"dtype": "float32",
|
| 646 |
+
"shape": [2, 2, 3, 2],
|
| 647 |
+
"data": {
|
| 648 |
+
"kind": "values",
|
| 649 |
+
"values": [-0.55963, 0.533472, 0.406887, 0.394587, 0.171511, 0.876045, -0.287087, 1.02164, 0.438649, -0.010704, 1.338354, -0.279405, -0.551834, -2.889061, -1.509981, 1.024115, 0.195393, -0.737109, 1.700101, 0.346216, 0.971125, 1.45025, -0.051909, -0.628431]
|
| 650 |
+
}
|
| 651 |
+
},
|
| 652 |
+
"grid": {
|
| 653 |
+
"dtype": "float32",
|
| 654 |
+
"shape": [2, 3, 2, 2],
|
| 655 |
+
"data": {
|
| 656 |
+
"kind": "values",
|
| 657 |
+
"values": [0.149807, 1.074831, 0.734055, -0.758657, 0.538205, -0.848275, -0.50859, 0.352947, 0.396231, 0.900274, -0.386299, 0.001921, 0.617788, -1.160511, 0.867577, -0.992307, 0.016539, -0.20402, -0.632008, 0.158605, 0.992302, -0.350783, -0.712433, -0.443807]
|
| 658 |
+
}
|
| 659 |
+
}
|
| 660 |
+
},
|
| 661 |
+
"outputs": { "y": { "dtype": "float32", "shape": [2, 2, 3, 2], "tolerance": 0 } }
|
| 662 |
+
},
|
| 663 |
+
{
|
| 664 |
+
"name": "ort_nearest_reflection_align",
|
| 665 |
+
"provenance": {
|
| 666 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 667 |
+
"test": "GridSampleTest.test_grid_sample_16_4D_nearest_reflection_align_corners"
|
| 668 |
+
},
|
| 669 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 1 },
|
| 670 |
+
"inputs": {
|
| 671 |
+
"x": {
|
| 672 |
+
"dtype": "float32",
|
| 673 |
+
"shape": [2, 2, 3, 2],
|
| 674 |
+
"data": {
|
| 675 |
+
"kind": "values",
|
| 676 |
+
"values": [-0.039373, -0.801472, -0.495544, -0.361514, 0.585113, -1.156007, -0.143365, -0.194741, -0.906885, -0.591838, 0.150785, -1.041149, -0.720534, -2.214754, -0.68373, 0.516358, 0.792848, 0.083228, 0.4228, -1.868747, -1.105713, 0.143731, 0.583597, 1.348155]
|
| 677 |
+
}
|
| 678 |
+
},
|
| 679 |
+
"grid": {
|
| 680 |
+
"dtype": "float32",
|
| 681 |
+
"shape": [2, 3, 2, 2],
|
| 682 |
+
"data": {
|
| 683 |
+
"kind": "values",
|
| 684 |
+
"values": [0.829854, -0.893309, 0.491599, -0.403504, -0.578962, 0.215574, -0.623348, 0.276486, 0.235657, -0.890987, 0.199798, 0.511115, 0.474997, -0.151054, -0.983745, -0.184985, 0.416769, -0.437853, 0.455497, 0.799155, -0.626582, 0.011834, 0.496199, 0.094053]
|
| 685 |
+
}
|
| 686 |
+
}
|
| 687 |
+
},
|
| 688 |
+
"outputs": { "y": { "dtype": "float32", "shape": [2, 2, 3, 2], "tolerance": 0 } }
|
| 689 |
+
},
|
| 690 |
+
{
|
| 691 |
+
"name": "ort_nearest_reflection_no_align",
|
| 692 |
+
"provenance": {
|
| 693 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 694 |
+
"test": "GridSampleTest.test_grid_sample_16_4D_nearest_reflection_no_align_corners"
|
| 695 |
+
},
|
| 696 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 0 },
|
| 697 |
+
"inputs": {
|
| 698 |
+
"x": {
|
| 699 |
+
"dtype": "float32",
|
| 700 |
+
"shape": [2, 2, 3, 2],
|
| 701 |
+
"data": {
|
| 702 |
+
"kind": "values",
|
| 703 |
+
"values": [-0.12923, -0.054595, 0.408347, 1.126366, 1.935057, 1.007685, 1.004642, -0.43352, -0.562711, -0.832754, -1.395545, -0.399295, -0.30994, -0.056062, 0.517413, -1.596237, 0.35696, -2.297482, -0.871083, -1.674028, 0.563055, -1.435067, 0.7194, -1.370747]
|
| 704 |
+
}
|
| 705 |
+
},
|
| 706 |
+
"grid": {
|
| 707 |
+
"dtype": "float32",
|
| 708 |
+
"shape": [2, 3, 2, 2],
|
| 709 |
+
"data": {
|
| 710 |
+
"kind": "values",
|
| 711 |
+
"values": [-0.81191, -1.183845, -0.963667, 0.947364, 0.649243, 1.125859, 0.961345, -1.071655, -0.818917, -0.193899, -0.779319, 0.833276, -0.907209, -0.585482, -1.15931, -0.681295, 0.986973, 0.982512, 0.859005, 0.926553, 1.067024, -0.307276, 0.528003, 1.069117]
|
| 712 |
+
}
|
| 713 |
+
}
|
| 714 |
+
},
|
| 715 |
+
"outputs": { "y": { "dtype": "float32", "shape": [2, 2, 3, 2], "tolerance": 0 } }
|
| 716 |
+
},
|
| 717 |
+
{
|
| 718 |
+
"name": "ort_custom_linear_zeros_mixed_bounds_right_bottom",
|
| 719 |
+
"provenance": {
|
| 720 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 721 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_linear_zeros_mixed_bounds_right_bottom"
|
| 722 |
+
},
|
| 723 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 724 |
+
"inputs": {
|
| 725 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2], "data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0] } },
|
| 726 |
+
"grid": {
|
| 727 |
+
"dtype": "float32",
|
| 728 |
+
"shape": [1, 2, 2, 2],
|
| 729 |
+
"data": { "kind": "values", "values": [0.0, 0.0, 0.9, 0.0, 0.0, 0.9, 0.9, 0.9] }
|
| 730 |
+
}
|
| 731 |
+
},
|
| 732 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 2], "tolerance": 0.000001 } }
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"name": "ort_custom_linear_zeros_mixed_bounds_left_top",
|
| 736 |
+
"provenance": {
|
| 737 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 738 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_linear_zeros_mixed_bounds_left_top"
|
| 739 |
+
},
|
| 740 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 741 |
+
"inputs": {
|
| 742 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2], "data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0] } },
|
| 743 |
+
"grid": {
|
| 744 |
+
"dtype": "float32",
|
| 745 |
+
"shape": [1, 2, 2, 2],
|
| 746 |
+
"data": { "kind": "values", "values": [0.0, 0.0, -0.9, 0.0, 0.0, -0.9, -0.9, -0.9] }
|
| 747 |
+
}
|
| 748 |
+
},
|
| 749 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 2], "tolerance": 0.000001 } }
|
| 750 |
+
},
|
| 751 |
+
{
|
| 752 |
+
"name": "ort_custom_nearest_reflection_nan_coords",
|
| 753 |
+
"provenance": {
|
| 754 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 755 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_nearest_reflection_nan_coords"
|
| 756 |
+
},
|
| 757 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 0 },
|
| 758 |
+
"inputs": {
|
| 759 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2], "data": { "kind": "constant", "value": 1.0 } },
|
| 760 |
+
"grid": {
|
| 761 |
+
"dtype": "float32",
|
| 762 |
+
"shape": [1, 1, 2, 2],
|
| 763 |
+
"data": { "kind": "values", "values": ["NaN", "NaN", 0.0, "NaN"] }
|
| 764 |
+
}
|
| 765 |
+
},
|
| 766 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 2], "tolerance": 0.000001 } }
|
| 767 |
+
},
|
| 768 |
+
{
|
| 769 |
+
"name": "ort_custom_linear_reflection_infinity_coords",
|
| 770 |
+
"provenance": {
|
| 771 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 772 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_bilinear_reflection_infinity_coords"
|
| 773 |
+
},
|
| 774 |
+
"attrs": { "mode": "linear", "padding_mode": "reflection", "align_corners": 0 },
|
| 775 |
+
"inputs": {
|
| 776 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2], "data": { "kind": "constant", "value": 1.0 } },
|
| 777 |
+
"grid": {
|
| 778 |
+
"dtype": "float32",
|
| 779 |
+
"shape": [1, 1, 2, 2],
|
| 780 |
+
"data": { "kind": "values", "values": ["Infinity", "-Infinity", "-Infinity", "Infinity"] }
|
| 781 |
+
}
|
| 782 |
+
},
|
| 783 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 2], "tolerance": 0.000001 } }
|
| 784 |
+
},
|
| 785 |
+
{
|
| 786 |
+
"name": "ort_custom_linear_reflection_extreme_coords",
|
| 787 |
+
"provenance": {
|
| 788 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 789 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_bilinear_reflection_extreme_coords"
|
| 790 |
+
},
|
| 791 |
+
"attrs": { "mode": "linear", "padding_mode": "reflection", "align_corners": 0 },
|
| 792 |
+
"inputs": {
|
| 793 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2], "data": { "kind": "constant", "value": 1.0 } },
|
| 794 |
+
"grid": {
|
| 795 |
+
"dtype": "float32",
|
| 796 |
+
"shape": [1, 1, 2, 2],
|
| 797 |
+
"data": {
|
| 798 |
+
"kind": "values",
|
| 799 |
+
"values": [100000000000000000000.0, 100000000000000000000.0, -100000000000000000000.0, -100000000000000000000.0]
|
| 800 |
+
}
|
| 801 |
+
}
|
| 802 |
+
},
|
| 803 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 2], "tolerance": 0.000001 } }
|
| 804 |
+
},
|
| 805 |
+
{
|
| 806 |
+
"name": "ort_custom_linear_border_extreme_coords",
|
| 807 |
+
"provenance": {
|
| 808 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 809 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_bilinear_border_extreme_coords"
|
| 810 |
+
},
|
| 811 |
+
"attrs": { "mode": "linear", "padding_mode": "border", "align_corners": 0 },
|
| 812 |
+
"inputs": {
|
| 813 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2], "data": { "kind": "constant", "value": 1.0 } },
|
| 814 |
+
"grid": {
|
| 815 |
+
"dtype": "float32",
|
| 816 |
+
"shape": [1, 1, 2, 2],
|
| 817 |
+
"data": {
|
| 818 |
+
"kind": "values",
|
| 819 |
+
"values": [100000000000000000000.0, 100000000000000000000.0, -100000000000000000000.0, -100000000000000000000.0]
|
| 820 |
+
}
|
| 821 |
+
}
|
| 822 |
+
},
|
| 823 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 2], "tolerance": 0.000001 } }
|
| 824 |
+
},
|
| 825 |
+
{
|
| 826 |
+
"name": "ort_custom_linear_zeros_extreme_nan_inf_coords",
|
| 827 |
+
"provenance": {
|
| 828 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 829 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_linear_zeros_extreme_nan_inf_coords"
|
| 830 |
+
},
|
| 831 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 832 |
+
"inputs": {
|
| 833 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2], "data": { "kind": "constant", "value": 1.0 } },
|
| 834 |
+
"grid": {
|
| 835 |
+
"dtype": "float32",
|
| 836 |
+
"shape": [1, 1, 3, 2],
|
| 837 |
+
"data": {
|
| 838 |
+
"kind": "values",
|
| 839 |
+
"values": ["NaN", "NaN", "Infinity", "-Infinity", 100000000000000000000.0, -100000000000000000000.0]
|
| 840 |
+
}
|
| 841 |
+
}
|
| 842 |
+
},
|
| 843 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 3], "tolerance": 0.000001 } }
|
| 844 |
+
},
|
| 845 |
+
{
|
| 846 |
+
"name": "ort_custom_cubic_reflection_nan_inf_coords",
|
| 847 |
+
"provenance": {
|
| 848 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 849 |
+
"test": "GridSampleCustomTestFloatOnly.test_grid_sample_20_4D_cubic_reflection_nan_inf_coords"
|
| 850 |
+
},
|
| 851 |
+
"attrs": { "mode": "cubic", "padding_mode": "reflection", "align_corners": 0 },
|
| 852 |
+
"inputs": {
|
| 853 |
+
"x": { "dtype": "float32", "shape": [1, 1, 4, 4], "data": { "kind": "constant", "value": 1.0 } },
|
| 854 |
+
"grid": {
|
| 855 |
+
"dtype": "float32",
|
| 856 |
+
"shape": [1, 1, 3, 2],
|
| 857 |
+
"data": { "kind": "values", "values": ["NaN", "NaN", "Infinity", "-Infinity", "-Infinity", "Infinity"] }
|
| 858 |
+
}
|
| 859 |
+
},
|
| 860 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 3], "tolerance": 0.0001 } }
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"name": "ort_custom_nearest_reflection_align_corners_extreme",
|
| 864 |
+
"provenance": {
|
| 865 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 866 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_nearest_reflection_align_corners_extreme"
|
| 867 |
+
},
|
| 868 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 1 },
|
| 869 |
+
"inputs": {
|
| 870 |
+
"x": {
|
| 871 |
+
"dtype": "float32",
|
| 872 |
+
"shape": [1, 1, 3, 3],
|
| 873 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0] }
|
| 874 |
+
},
|
| 875 |
+
"grid": {
|
| 876 |
+
"dtype": "float32",
|
| 877 |
+
"shape": [1, 1, 2, 2],
|
| 878 |
+
"data": { "kind": "values", "values": [100000000000000000000.0, 100000000000000000000.0, "NaN", "-Infinity"] }
|
| 879 |
+
}
|
| 880 |
+
},
|
| 881 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 2], "tolerance": 0.000001 } }
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"name": "ort_custom_nearest_reflection_negative_only_extreme",
|
| 885 |
+
"provenance": {
|
| 886 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 887 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_nearest_reflection_negative_only_extreme"
|
| 888 |
+
},
|
| 889 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 0 },
|
| 890 |
+
"inputs": {
|
| 891 |
+
"x": {
|
| 892 |
+
"dtype": "float32",
|
| 893 |
+
"shape": [1, 1, 3, 3],
|
| 894 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0] }
|
| 895 |
+
},
|
| 896 |
+
"grid": {
|
| 897 |
+
"dtype": "float32",
|
| 898 |
+
"shape": [1, 1, 2, 2],
|
| 899 |
+
"data": {
|
| 900 |
+
"kind": "values",
|
| 901 |
+
"values": [-100000000000000000000.0, -100000000000000000000.0, "-Infinity", "-Infinity"]
|
| 902 |
+
}
|
| 903 |
+
}
|
| 904 |
+
},
|
| 905 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 2], "tolerance": 0.000001 } }
|
| 906 |
+
},
|
| 907 |
+
{
|
| 908 |
+
"name": "ort_custom_nearest_reflection_mixed_normal_pathological",
|
| 909 |
+
"provenance": {
|
| 910 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 911 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_4D_nearest_reflection_mixed_normal_pathological"
|
| 912 |
+
},
|
| 913 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 1 },
|
| 914 |
+
"inputs": {
|
| 915 |
+
"x": {
|
| 916 |
+
"dtype": "float32",
|
| 917 |
+
"shape": [1, 1, 3, 3],
|
| 918 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0] }
|
| 919 |
+
},
|
| 920 |
+
"grid": {
|
| 921 |
+
"dtype": "float32",
|
| 922 |
+
"shape": [1, 1, 4, 2],
|
| 923 |
+
"data": {
|
| 924 |
+
"kind": "values",
|
| 925 |
+
"values": [0.0, 0.0, -1.0, -1.0, "NaN", "NaN", 100000000000000000000.0, -100000000000000000000.0]
|
| 926 |
+
}
|
| 927 |
+
}
|
| 928 |
+
},
|
| 929 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 4], "tolerance": 0.000001 } }
|
| 930 |
+
},
|
| 931 |
+
{
|
| 932 |
+
"name": "onnx_backend_gridsample",
|
| 933 |
+
"provenance": { "source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample" },
|
| 934 |
+
"inputs": {
|
| 935 |
+
"x": {
|
| 936 |
+
"dtype": "float32",
|
| 937 |
+
"shape": [1, 1, 4, 4],
|
| 938 |
+
"data": { "kind": "values", "values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_input_x" } }
|
| 939 |
+
},
|
| 940 |
+
"grid": {
|
| 941 |
+
"dtype": "float32",
|
| 942 |
+
"shape": [1, 6, 6, 2],
|
| 943 |
+
"data": {
|
| 944 |
+
"kind": "values",
|
| 945 |
+
"values": [-1.0, -1.0, -0.6000000238418579, -1.0, -0.20000000298023224, -1.0, 0.20000000298023224, -1.0, 0.6000000238418579, -1.0, 1.0, -1.0, -1.0, -0.6000000238418579, -0.6000000238418579, -0.6000000238418579, -0.20000000298023224, -0.6000000238418579, 0.20000000298023224, -0.6000000238418579, 0.6000000238418579, -0.6000000238418579, 1.0, -0.6000000238418579, -1.0, -0.20000000298023224, -0.6000000238418579, -0.20000000298023224, -0.20000000298023224, -0.20000000298023224, 0.20000000298023224, -0.20000000298023224, 0.6000000238418579, -0.20000000298023224, 1.0, -0.20000000298023224, -1.0, 0.20000000298023224, -0.6000000238418579, 0.20000000298023224, -0.20000000298023224, 0.20000000298023224, 0.20000000298023224, 0.20000000298023224, 0.6000000238418579, 0.20000000298023224, 1.0, 0.20000000298023224, -1.0, 0.6000000238418579, -0.6000000238418579, 0.6000000238418579, -0.20000000298023224, 0.6000000238418579, 0.20000000298023224, 0.6000000238418579, 0.6000000238418579, 0.6000000238418579, 1.0, 0.6000000238418579, -1.0, 1.0, -0.6000000238418579, 1.0, -0.20000000298023224, 1.0, 0.20000000298023224, 1.0, 0.6000000238418579, 1.0, 1.0, 1.0]
|
| 946 |
+
}
|
| 947 |
+
}
|
| 948 |
+
},
|
| 949 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 6, 6], "tolerance": 0.0001 } },
|
| 950 |
+
"attrs": { "align_corners": 0, "mode": "linear", "padding_mode": "zeros" }
|
| 951 |
+
},
|
| 952 |
+
{
|
| 953 |
+
"name": "onnx_backend_gridsample_aligncorners_true",
|
| 954 |
+
"provenance": { "source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_aligncorners_true" },
|
| 955 |
+
"inputs": {
|
| 956 |
+
"x": {
|
| 957 |
+
"dtype": "float32",
|
| 958 |
+
"shape": [1, 1, 3, 2],
|
| 959 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 960 |
+
},
|
| 961 |
+
"grid": {
|
| 962 |
+
"dtype": "float32",
|
| 963 |
+
"shape": [1, 2, 4, 2],
|
| 964 |
+
"data": {
|
| 965 |
+
"kind": "values",
|
| 966 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_aligncorners_true_input_grid" }
|
| 967 |
+
}
|
| 968 |
+
}
|
| 969 |
+
},
|
| 970 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 971 |
+
"attrs": { "align_corners": 1, "mode": "linear" }
|
| 972 |
+
},
|
| 973 |
+
{
|
| 974 |
+
"name": "onnx_backend_gridsample_bicubic",
|
| 975 |
+
"provenance": { "source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_bicubic" },
|
| 976 |
+
"inputs": {
|
| 977 |
+
"x": {
|
| 978 |
+
"dtype": "float32",
|
| 979 |
+
"shape": [1, 1, 3, 2],
|
| 980 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 981 |
+
},
|
| 982 |
+
"grid": {
|
| 983 |
+
"dtype": "float32",
|
| 984 |
+
"shape": [1, 2, 4, 2],
|
| 985 |
+
"data": {
|
| 986 |
+
"kind": "values",
|
| 987 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_aligncorners_true_input_grid" }
|
| 988 |
+
}
|
| 989 |
+
}
|
| 990 |
+
},
|
| 991 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 992 |
+
"attrs": { "mode": "cubic" }
|
| 993 |
+
},
|
| 994 |
+
{
|
| 995 |
+
"name": "onnx_backend_gridsample_bicubic_align_corners_0_additional_1",
|
| 996 |
+
"provenance": {
|
| 997 |
+
"source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_bicubic_align_corners_0_additional_1"
|
| 998 |
+
},
|
| 999 |
+
"inputs": {
|
| 1000 |
+
"x": {
|
| 1001 |
+
"dtype": "float32",
|
| 1002 |
+
"shape": [1, 1, 3, 2],
|
| 1003 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 1004 |
+
},
|
| 1005 |
+
"grid": {
|
| 1006 |
+
"dtype": "float32",
|
| 1007 |
+
"shape": [1, 2, 4, 2],
|
| 1008 |
+
"data": { "kind": "values", "values": { "$ref": "#/fixtureArrays/grid_sample_grid_16" } }
|
| 1009 |
+
}
|
| 1010 |
+
},
|
| 1011 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 1012 |
+
"attrs": { "align_corners": 0, "mode": "cubic" }
|
| 1013 |
+
},
|
| 1014 |
+
{
|
| 1015 |
+
"name": "onnx_backend_gridsample_bicubic_align_corners_1_additional_1",
|
| 1016 |
+
"provenance": {
|
| 1017 |
+
"source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_bicubic_align_corners_1_additional_1"
|
| 1018 |
+
},
|
| 1019 |
+
"inputs": {
|
| 1020 |
+
"x": {
|
| 1021 |
+
"dtype": "float32",
|
| 1022 |
+
"shape": [1, 1, 3, 2],
|
| 1023 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 1024 |
+
},
|
| 1025 |
+
"grid": {
|
| 1026 |
+
"dtype": "float32",
|
| 1027 |
+
"shape": [1, 2, 4, 2],
|
| 1028 |
+
"data": { "kind": "values", "values": { "$ref": "#/fixtureArrays/grid_sample_grid_16" } }
|
| 1029 |
+
}
|
| 1030 |
+
},
|
| 1031 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 1032 |
+
"attrs": { "align_corners": 1, "mode": "cubic" }
|
| 1033 |
+
},
|
| 1034 |
+
{
|
| 1035 |
+
"name": "onnx_backend_gridsample_bilinear",
|
| 1036 |
+
"provenance": { "source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_bilinear" },
|
| 1037 |
+
"inputs": {
|
| 1038 |
+
"x": {
|
| 1039 |
+
"dtype": "float32",
|
| 1040 |
+
"shape": [1, 1, 3, 2],
|
| 1041 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 1042 |
+
},
|
| 1043 |
+
"grid": {
|
| 1044 |
+
"dtype": "float32",
|
| 1045 |
+
"shape": [1, 2, 4, 2],
|
| 1046 |
+
"data": {
|
| 1047 |
+
"kind": "values",
|
| 1048 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_aligncorners_true_input_grid" }
|
| 1049 |
+
}
|
| 1050 |
+
}
|
| 1051 |
+
},
|
| 1052 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 1053 |
+
"attrs": { "mode": "linear" }
|
| 1054 |
+
},
|
| 1055 |
+
{
|
| 1056 |
+
"name": "onnx_backend_gridsample_bilinear_align_corners_0_additional_1",
|
| 1057 |
+
"provenance": {
|
| 1058 |
+
"source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_bilinear_align_corners_0_additional_1"
|
| 1059 |
+
},
|
| 1060 |
+
"inputs": {
|
| 1061 |
+
"x": {
|
| 1062 |
+
"dtype": "float32",
|
| 1063 |
+
"shape": [1, 1, 3, 2],
|
| 1064 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 1065 |
+
},
|
| 1066 |
+
"grid": {
|
| 1067 |
+
"dtype": "float32",
|
| 1068 |
+
"shape": [1, 2, 4, 2],
|
| 1069 |
+
"data": { "kind": "values", "values": { "$ref": "#/fixtureArrays/grid_sample_grid_16" } }
|
| 1070 |
+
}
|
| 1071 |
+
},
|
| 1072 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 1073 |
+
"attrs": { "align_corners": 0, "mode": "linear" }
|
| 1074 |
+
},
|
| 1075 |
+
{
|
| 1076 |
+
"name": "onnx_backend_gridsample_bilinear_align_corners_1_additional_1",
|
| 1077 |
+
"provenance": {
|
| 1078 |
+
"source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_bilinear_align_corners_1_additional_1"
|
| 1079 |
+
},
|
| 1080 |
+
"inputs": {
|
| 1081 |
+
"x": {
|
| 1082 |
+
"dtype": "float32",
|
| 1083 |
+
"shape": [1, 1, 3, 2],
|
| 1084 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 1085 |
+
},
|
| 1086 |
+
"grid": {
|
| 1087 |
+
"dtype": "float32",
|
| 1088 |
+
"shape": [1, 2, 4, 2],
|
| 1089 |
+
"data": { "kind": "values", "values": { "$ref": "#/fixtureArrays/grid_sample_grid_16" } }
|
| 1090 |
+
}
|
| 1091 |
+
},
|
| 1092 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 1093 |
+
"attrs": { "align_corners": 1, "mode": "linear" }
|
| 1094 |
+
},
|
| 1095 |
+
{
|
| 1096 |
+
"name": "onnx_backend_gridsample_border_padding",
|
| 1097 |
+
"provenance": { "source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_border_padding" },
|
| 1098 |
+
"inputs": {
|
| 1099 |
+
"x": {
|
| 1100 |
+
"dtype": "float32",
|
| 1101 |
+
"shape": [1, 1, 3, 2],
|
| 1102 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 1103 |
+
},
|
| 1104 |
+
"grid": {
|
| 1105 |
+
"dtype": "float32",
|
| 1106 |
+
"shape": [1, 2, 4, 2],
|
| 1107 |
+
"data": {
|
| 1108 |
+
"kind": "values",
|
| 1109 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_border_padding_input_grid" }
|
| 1110 |
+
}
|
| 1111 |
+
}
|
| 1112 |
+
},
|
| 1113 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 1114 |
+
"attrs": { "padding_mode": "border" }
|
| 1115 |
+
},
|
| 1116 |
+
{
|
| 1117 |
+
"name": "onnx_backend_gridsample_nearest",
|
| 1118 |
+
"provenance": { "source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_nearest" },
|
| 1119 |
+
"inputs": {
|
| 1120 |
+
"x": {
|
| 1121 |
+
"dtype": "float32",
|
| 1122 |
+
"shape": [1, 1, 3, 2],
|
| 1123 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 1124 |
+
},
|
| 1125 |
+
"grid": {
|
| 1126 |
+
"dtype": "float32",
|
| 1127 |
+
"shape": [1, 2, 4, 2],
|
| 1128 |
+
"data": {
|
| 1129 |
+
"kind": "values",
|
| 1130 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_aligncorners_true_input_grid" }
|
| 1131 |
+
}
|
| 1132 |
+
}
|
| 1133 |
+
},
|
| 1134 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 1135 |
+
"attrs": { "mode": "nearest" }
|
| 1136 |
+
},
|
| 1137 |
+
{
|
| 1138 |
+
"name": "onnx_backend_gridsample_nearest_align_corners_0_additional_1",
|
| 1139 |
+
"provenance": {
|
| 1140 |
+
"source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_nearest_align_corners_0_additional_1"
|
| 1141 |
+
},
|
| 1142 |
+
"inputs": {
|
| 1143 |
+
"x": {
|
| 1144 |
+
"dtype": "float32",
|
| 1145 |
+
"shape": [1, 1, 3, 2],
|
| 1146 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 1147 |
+
},
|
| 1148 |
+
"grid": {
|
| 1149 |
+
"dtype": "float32",
|
| 1150 |
+
"shape": [1, 2, 4, 2],
|
| 1151 |
+
"data": { "kind": "values", "values": { "$ref": "#/fixtureArrays/grid_sample_grid_16" } }
|
| 1152 |
+
}
|
| 1153 |
+
},
|
| 1154 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 1155 |
+
"attrs": { "align_corners": 0, "mode": "nearest" }
|
| 1156 |
+
},
|
| 1157 |
+
{
|
| 1158 |
+
"name": "onnx_backend_gridsample_nearest_align_corners_1_additional_1",
|
| 1159 |
+
"provenance": {
|
| 1160 |
+
"source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_nearest_align_corners_1_additional_1"
|
| 1161 |
+
},
|
| 1162 |
+
"inputs": {
|
| 1163 |
+
"x": {
|
| 1164 |
+
"dtype": "float32",
|
| 1165 |
+
"shape": [1, 1, 3, 2],
|
| 1166 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 1167 |
+
},
|
| 1168 |
+
"grid": {
|
| 1169 |
+
"dtype": "float32",
|
| 1170 |
+
"shape": [1, 2, 4, 2],
|
| 1171 |
+
"data": { "kind": "values", "values": { "$ref": "#/fixtureArrays/grid_sample_grid_16" } }
|
| 1172 |
+
}
|
| 1173 |
+
},
|
| 1174 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 1175 |
+
"attrs": { "align_corners": 1, "mode": "nearest" }
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"name": "onnx_backend_gridsample_reflection_padding",
|
| 1179 |
+
"provenance": { "source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_reflection_padding" },
|
| 1180 |
+
"inputs": {
|
| 1181 |
+
"x": {
|
| 1182 |
+
"dtype": "float32",
|
| 1183 |
+
"shape": [1, 1, 3, 2],
|
| 1184 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 1185 |
+
},
|
| 1186 |
+
"grid": {
|
| 1187 |
+
"dtype": "float32",
|
| 1188 |
+
"shape": [1, 2, 4, 2],
|
| 1189 |
+
"data": {
|
| 1190 |
+
"kind": "values",
|
| 1191 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_border_padding_input_grid" }
|
| 1192 |
+
}
|
| 1193 |
+
}
|
| 1194 |
+
},
|
| 1195 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 1196 |
+
"attrs": { "padding_mode": "reflection" }
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"name": "onnx_backend_gridsample_zeros_padding",
|
| 1200 |
+
"provenance": { "source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_zeros_padding" },
|
| 1201 |
+
"inputs": {
|
| 1202 |
+
"x": {
|
| 1203 |
+
"dtype": "float32",
|
| 1204 |
+
"shape": [1, 1, 3, 2],
|
| 1205 |
+
"data": { "kind": "values", "values": [0.0, 1.0, 2.0, 3.0, 4.0, 5.0] }
|
| 1206 |
+
},
|
| 1207 |
+
"grid": {
|
| 1208 |
+
"dtype": "float32",
|
| 1209 |
+
"shape": [1, 2, 4, 2],
|
| 1210 |
+
"data": {
|
| 1211 |
+
"kind": "values",
|
| 1212 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_border_padding_input_grid" }
|
| 1213 |
+
}
|
| 1214 |
+
}
|
| 1215 |
+
},
|
| 1216 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4], "tolerance": 0.0001 } },
|
| 1217 |
+
"attrs": { "padding_mode": "zeros" }
|
| 1218 |
+
},
|
| 1219 |
+
{
|
| 1220 |
+
"name": "onnx_backend_gridsample_volumetric_bilinear_align_corners_0",
|
| 1221 |
+
"provenance": {
|
| 1222 |
+
"source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_volumetric_bilinear_align_corners_0"
|
| 1223 |
+
},
|
| 1224 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 1225 |
+
"inputs": {
|
| 1226 |
+
"x": {
|
| 1227 |
+
"dtype": "float32",
|
| 1228 |
+
"shape": [1, 1, 3, 2, 2],
|
| 1229 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0] }
|
| 1230 |
+
},
|
| 1231 |
+
"grid": {
|
| 1232 |
+
"dtype": "float32",
|
| 1233 |
+
"shape": [1, 2, 4, 2, 3],
|
| 1234 |
+
"data": {
|
| 1235 |
+
"kind": "values",
|
| 1236 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_volumetric_input_grid" }
|
| 1237 |
+
}
|
| 1238 |
+
}
|
| 1239 |
+
},
|
| 1240 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4, 2], "tolerance": 0.00001 } }
|
| 1241 |
+
},
|
| 1242 |
+
{
|
| 1243 |
+
"name": "onnx_backend_gridsample_volumetric_bilinear_align_corners_1",
|
| 1244 |
+
"provenance": {
|
| 1245 |
+
"source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_volumetric_bilinear_align_corners_1"
|
| 1246 |
+
},
|
| 1247 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 1 },
|
| 1248 |
+
"inputs": {
|
| 1249 |
+
"x": {
|
| 1250 |
+
"dtype": "float32",
|
| 1251 |
+
"shape": [1, 1, 3, 2, 2],
|
| 1252 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0] }
|
| 1253 |
+
},
|
| 1254 |
+
"grid": {
|
| 1255 |
+
"dtype": "float32",
|
| 1256 |
+
"shape": [1, 2, 4, 2, 3],
|
| 1257 |
+
"data": {
|
| 1258 |
+
"kind": "values",
|
| 1259 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_volumetric_input_grid" }
|
| 1260 |
+
}
|
| 1261 |
+
}
|
| 1262 |
+
},
|
| 1263 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4, 2], "tolerance": 0.00001 } }
|
| 1264 |
+
},
|
| 1265 |
+
{
|
| 1266 |
+
"name": "onnx_backend_gridsample_volumetric_nearest_align_corners_0",
|
| 1267 |
+
"provenance": {
|
| 1268 |
+
"source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_volumetric_nearest_align_corners_0"
|
| 1269 |
+
},
|
| 1270 |
+
"attrs": { "mode": "nearest", "padding_mode": "zeros", "align_corners": 0 },
|
| 1271 |
+
"inputs": {
|
| 1272 |
+
"x": {
|
| 1273 |
+
"dtype": "float32",
|
| 1274 |
+
"shape": [1, 1, 3, 2, 2],
|
| 1275 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0] }
|
| 1276 |
+
},
|
| 1277 |
+
"grid": {
|
| 1278 |
+
"dtype": "float32",
|
| 1279 |
+
"shape": [1, 2, 4, 2, 3],
|
| 1280 |
+
"data": {
|
| 1281 |
+
"kind": "values",
|
| 1282 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_volumetric_input_grid" }
|
| 1283 |
+
}
|
| 1284 |
+
}
|
| 1285 |
+
},
|
| 1286 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4, 2], "tolerance": 0 } }
|
| 1287 |
+
},
|
| 1288 |
+
{
|
| 1289 |
+
"name": "onnx_backend_gridsample_volumetric_nearest_align_corners_1",
|
| 1290 |
+
"provenance": {
|
| 1291 |
+
"source": "cmake/external/onnx/onnx/backend/test/data/node/test_gridsample_volumetric_nearest_align_corners_1"
|
| 1292 |
+
},
|
| 1293 |
+
"attrs": { "mode": "nearest", "padding_mode": "zeros", "align_corners": 1 },
|
| 1294 |
+
"inputs": {
|
| 1295 |
+
"x": {
|
| 1296 |
+
"dtype": "float32",
|
| 1297 |
+
"shape": [1, 1, 3, 2, 2],
|
| 1298 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0] }
|
| 1299 |
+
},
|
| 1300 |
+
"grid": {
|
| 1301 |
+
"dtype": "float32",
|
| 1302 |
+
"shape": [1, 2, 4, 2, 3],
|
| 1303 |
+
"data": {
|
| 1304 |
+
"kind": "values",
|
| 1305 |
+
"values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_volumetric_input_grid" }
|
| 1306 |
+
}
|
| 1307 |
+
}
|
| 1308 |
+
},
|
| 1309 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 4, 2], "tolerance": 0 } }
|
| 1310 |
+
},
|
| 1311 |
+
{
|
| 1312 |
+
"name": "ort_5d_nearest_zeros_align_corners_simple",
|
| 1313 |
+
"provenance": {
|
| 1314 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1315 |
+
"test": "GridSampleTest.test_grid_sample_20_5D_nearest_zeros_align_corners",
|
| 1316 |
+
"notes": "Compact node-level projection of ORT's generated 5D nearest zeros case."
|
| 1317 |
+
},
|
| 1318 |
+
"attrs": { "mode": "nearest", "padding_mode": "zeros", "align_corners": 1 },
|
| 1319 |
+
"inputs": {
|
| 1320 |
+
"x": {
|
| 1321 |
+
"dtype": "float32",
|
| 1322 |
+
"shape": [1, 1, 2, 2, 2],
|
| 1323 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0] }
|
| 1324 |
+
},
|
| 1325 |
+
"grid": {
|
| 1326 |
+
"dtype": "float32",
|
| 1327 |
+
"shape": [1, 1, 1, 3, 3],
|
| 1328 |
+
"data": { "kind": "values", "values": [-1.0, -1.0, -1.0, 1.0, 1.0, 1.0, 3.0, 0.0, 0.0] }
|
| 1329 |
+
}
|
| 1330 |
+
},
|
| 1331 |
+
"outputs": {
|
| 1332 |
+
"y": {
|
| 1333 |
+
"dtype": "float32",
|
| 1334 |
+
"shape": [1, 1, 1, 1, 3],
|
| 1335 |
+
"tolerance": 0,
|
| 1336 |
+
"data": { "kind": "values", "values": [1.0, 8.0, 0.0] }
|
| 1337 |
+
}
|
| 1338 |
+
}
|
| 1339 |
+
},
|
| 1340 |
+
{
|
| 1341 |
+
"name": "ort_5d_nearest_zeros_no_align_corners_simple",
|
| 1342 |
+
"provenance": {
|
| 1343 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1344 |
+
"test": "GridSampleTest.test_grid_sample_20_5D_nearest_zeros_no_align_corners",
|
| 1345 |
+
"notes": "Uses normalized coordinates that map exactly to both volume corners when align_corners=0."
|
| 1346 |
+
},
|
| 1347 |
+
"attrs": { "mode": "nearest", "padding_mode": "zeros", "align_corners": 0 },
|
| 1348 |
+
"inputs": {
|
| 1349 |
+
"x": {
|
| 1350 |
+
"dtype": "float32",
|
| 1351 |
+
"shape": [1, 1, 2, 2, 2],
|
| 1352 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0] }
|
| 1353 |
+
},
|
| 1354 |
+
"grid": {
|
| 1355 |
+
"dtype": "float32",
|
| 1356 |
+
"shape": [1, 1, 1, 3, 3],
|
| 1357 |
+
"data": { "kind": "values", "values": [-0.5, -0.5, -0.5, 0.5, 0.5, 0.5, 1.5, 0.0, 0.0] }
|
| 1358 |
+
}
|
| 1359 |
+
},
|
| 1360 |
+
"outputs": {
|
| 1361 |
+
"y": {
|
| 1362 |
+
"dtype": "float32",
|
| 1363 |
+
"shape": [1, 1, 1, 1, 3],
|
| 1364 |
+
"tolerance": 0,
|
| 1365 |
+
"data": { "kind": "values", "values": [1.0, 8.0, 0.0] }
|
| 1366 |
+
}
|
| 1367 |
+
}
|
| 1368 |
+
},
|
| 1369 |
+
{
|
| 1370 |
+
"name": "ort_5d_linear_zeros_align_corners_simple",
|
| 1371 |
+
"provenance": {
|
| 1372 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1373 |
+
"test": "GridSampleTest.test_grid_sample_20_5D_bilinear_zeros_align_corners",
|
| 1374 |
+
"notes": "Opset-20 5D bilinear is represented as mode=linear."
|
| 1375 |
+
},
|
| 1376 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 1 },
|
| 1377 |
+
"inputs": {
|
| 1378 |
+
"x": {
|
| 1379 |
+
"dtype": "float32",
|
| 1380 |
+
"shape": [1, 1, 2, 2, 2],
|
| 1381 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0] }
|
| 1382 |
+
},
|
| 1383 |
+
"grid": {
|
| 1384 |
+
"dtype": "float32",
|
| 1385 |
+
"shape": [1, 1, 1, 3, 3],
|
| 1386 |
+
"data": { "kind": "values", "values": [-1.0, -1.0, -1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0] }
|
| 1387 |
+
}
|
| 1388 |
+
},
|
| 1389 |
+
"outputs": {
|
| 1390 |
+
"y": {
|
| 1391 |
+
"dtype": "float32",
|
| 1392 |
+
"shape": [1, 1, 1, 1, 3],
|
| 1393 |
+
"tolerance": 0.000001,
|
| 1394 |
+
"data": { "kind": "values", "values": [1.0, 4.5, 8.0] }
|
| 1395 |
+
}
|
| 1396 |
+
}
|
| 1397 |
+
},
|
| 1398 |
+
{
|
| 1399 |
+
"name": "ort_5d_linear_zeros_no_align_corners_simple",
|
| 1400 |
+
"provenance": {
|
| 1401 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1402 |
+
"test": "GridSampleTest.test_grid_sample_22_5D_bilinear_zeros_no_align_corners",
|
| 1403 |
+
"notes": "Uses exact corner/center coordinates for align_corners=0."
|
| 1404 |
+
},
|
| 1405 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 1406 |
+
"inputs": {
|
| 1407 |
+
"x": {
|
| 1408 |
+
"dtype": "float32",
|
| 1409 |
+
"shape": [1, 1, 2, 2, 2],
|
| 1410 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0] }
|
| 1411 |
+
},
|
| 1412 |
+
"grid": {
|
| 1413 |
+
"dtype": "float32",
|
| 1414 |
+
"shape": [1, 1, 1, 3, 3],
|
| 1415 |
+
"data": { "kind": "values", "values": [-0.5, -0.5, -0.5, 0.0, 0.0, 0.0, 0.5, 0.5, 0.5] }
|
| 1416 |
+
}
|
| 1417 |
+
},
|
| 1418 |
+
"outputs": {
|
| 1419 |
+
"y": {
|
| 1420 |
+
"dtype": "float32",
|
| 1421 |
+
"shape": [1, 1, 1, 1, 3],
|
| 1422 |
+
"tolerance": 0.000001,
|
| 1423 |
+
"data": { "kind": "values", "values": [1.0, 4.5, 8.0] }
|
| 1424 |
+
}
|
| 1425 |
+
}
|
| 1426 |
+
},
|
| 1427 |
+
{
|
| 1428 |
+
"name": "ort_custom_5d_nearest_reflection_extreme_coords",
|
| 1429 |
+
"provenance": {
|
| 1430 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 1431 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_5D_nearest_reflection_extreme_coords"
|
| 1432 |
+
},
|
| 1433 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 0 },
|
| 1434 |
+
"inputs": {
|
| 1435 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2, 2], "data": { "kind": "constant", "value": 1.0 } },
|
| 1436 |
+
"grid": {
|
| 1437 |
+
"dtype": "float32",
|
| 1438 |
+
"shape": [1, 1, 1, 2, 3],
|
| 1439 |
+
"data": {
|
| 1440 |
+
"kind": "values",
|
| 1441 |
+
"values": [10000000000.0, 10000000000.0, 10000000000.0, -10000000000.0, -10000000000.0, -10000000000.0]
|
| 1442 |
+
}
|
| 1443 |
+
}
|
| 1444 |
+
},
|
| 1445 |
+
"outputs": {
|
| 1446 |
+
"y": {
|
| 1447 |
+
"dtype": "float32",
|
| 1448 |
+
"shape": [1, 1, 1, 1, 2],
|
| 1449 |
+
"tolerance": 0,
|
| 1450 |
+
"data": { "kind": "values", "values": [1.0, 1.0] }
|
| 1451 |
+
}
|
| 1452 |
+
}
|
| 1453 |
+
},
|
| 1454 |
+
{
|
| 1455 |
+
"name": "ort_5d_nearest_border_no_align_corners",
|
| 1456 |
+
"provenance": {
|
| 1457 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1458 |
+
"test": "GridSampleTest.test_grid_sample_20_5D_nearest_border_no_align_corners",
|
| 1459 |
+
"notes": "Compact projection of ORT's generated 5D nearest border/no-align case."
|
| 1460 |
+
},
|
| 1461 |
+
"attrs": { "mode": "nearest", "padding_mode": "border", "align_corners": 0 },
|
| 1462 |
+
"inputs": {
|
| 1463 |
+
"x": {
|
| 1464 |
+
"dtype": "float32",
|
| 1465 |
+
"shape": [1, 1, 2, 2, 2],
|
| 1466 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0] }
|
| 1467 |
+
},
|
| 1468 |
+
"grid": {
|
| 1469 |
+
"dtype": "float32",
|
| 1470 |
+
"shape": [1, 1, 1, 3, 3],
|
| 1471 |
+
"data": { "kind": "values", "values": [-2.0, -2.0, -2.0, 0.5, 0.5, 0.5, 2.0, 2.0, 2.0] }
|
| 1472 |
+
}
|
| 1473 |
+
},
|
| 1474 |
+
"outputs": {
|
| 1475 |
+
"y": {
|
| 1476 |
+
"dtype": "float32",
|
| 1477 |
+
"shape": [1, 1, 1, 1, 3],
|
| 1478 |
+
"tolerance": 0,
|
| 1479 |
+
"data": { "kind": "values", "values": [1.0, 8.0, 8.0] }
|
| 1480 |
+
}
|
| 1481 |
+
}
|
| 1482 |
+
},
|
| 1483 |
+
{
|
| 1484 |
+
"name": "ort_5d_nearest_reflection_align_corners",
|
| 1485 |
+
"provenance": {
|
| 1486 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1487 |
+
"test": "GridSampleTest.test_grid_sample_20_5D_nearest_reflection_align_corners",
|
| 1488 |
+
"notes": "Compact projection of ORT's generated 5D nearest reflection/align-corners case."
|
| 1489 |
+
},
|
| 1490 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 1 },
|
| 1491 |
+
"inputs": {
|
| 1492 |
+
"x": {
|
| 1493 |
+
"dtype": "float32",
|
| 1494 |
+
"shape": [1, 1, 2, 2, 2],
|
| 1495 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0] }
|
| 1496 |
+
},
|
| 1497 |
+
"grid": {
|
| 1498 |
+
"dtype": "float32",
|
| 1499 |
+
"shape": [1, 1, 1, 3, 3],
|
| 1500 |
+
"data": { "kind": "values", "values": [-2.0, -2.0, -2.0, 0.0, 0.0, 0.0, 2.0, 2.0, 2.0] }
|
| 1501 |
+
}
|
| 1502 |
+
},
|
| 1503 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 1, 3], "tolerance": 0 } }
|
| 1504 |
+
},
|
| 1505 |
+
{
|
| 1506 |
+
"name": "ort_5d_linear_border_align_corners",
|
| 1507 |
+
"provenance": {
|
| 1508 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1509 |
+
"test": "GridSampleTest.test_grid_sample_22_5D_bilinear_border_align_corners",
|
| 1510 |
+
"notes": "Compact opset-20 projection using mode=linear for ORT's generated 5D bilinear border/align-corners case."
|
| 1511 |
+
},
|
| 1512 |
+
"attrs": { "mode": "linear", "padding_mode": "border", "align_corners": 1 },
|
| 1513 |
+
"inputs": {
|
| 1514 |
+
"x": {
|
| 1515 |
+
"dtype": "float32",
|
| 1516 |
+
"shape": [1, 1, 2, 2, 2],
|
| 1517 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0] }
|
| 1518 |
+
},
|
| 1519 |
+
"grid": {
|
| 1520 |
+
"dtype": "float32",
|
| 1521 |
+
"shape": [1, 1, 1, 3, 3],
|
| 1522 |
+
"data": { "kind": "values", "values": [-2.0, -2.0, -2.0, 0.0, 0.0, 0.0, 2.0, 2.0, 2.0] }
|
| 1523 |
+
}
|
| 1524 |
+
},
|
| 1525 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 1, 3], "tolerance": 0.000001 } }
|
| 1526 |
+
},
|
| 1527 |
+
{
|
| 1528 |
+
"name": "ort_5d_linear_reflection_no_align_corners",
|
| 1529 |
+
"provenance": {
|
| 1530 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1531 |
+
"test": "GridSampleTest.test_grid_sample_22_5D_bilinear_reflection_no_align_corners",
|
| 1532 |
+
"notes": "Compact opset-20 projection using mode=linear for ORT's generated 5D bilinear reflection/no-align case."
|
| 1533 |
+
},
|
| 1534 |
+
"attrs": { "mode": "linear", "padding_mode": "reflection", "align_corners": 0 },
|
| 1535 |
+
"inputs": {
|
| 1536 |
+
"x": {
|
| 1537 |
+
"dtype": "float32",
|
| 1538 |
+
"shape": [1, 1, 2, 2, 2],
|
| 1539 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0] }
|
| 1540 |
+
},
|
| 1541 |
+
"grid": {
|
| 1542 |
+
"dtype": "float32",
|
| 1543 |
+
"shape": [1, 1, 1, 3, 3],
|
| 1544 |
+
"data": { "kind": "values", "values": [-2.0, -2.0, -2.0, 0.0, 0.0, 0.0, 2.0, 2.0, 2.0] }
|
| 1545 |
+
}
|
| 1546 |
+
},
|
| 1547 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 1, 3], "tolerance": 0.000001 } }
|
| 1548 |
+
},
|
| 1549 |
+
{
|
| 1550 |
+
"name": "ort_5d_nearest_border_align_corners",
|
| 1551 |
+
"provenance": {
|
| 1552 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1553 |
+
"test": "GridSampleTest.test_grid_sample_20_5D_nearest_border_align_corners",
|
| 1554 |
+
"notes": "Compact 5D border-padding projection of ORT's generated align-corners case."
|
| 1555 |
+
},
|
| 1556 |
+
"attrs": { "mode": "nearest", "padding_mode": "border", "align_corners": 1 },
|
| 1557 |
+
"inputs": {
|
| 1558 |
+
"x": {
|
| 1559 |
+
"dtype": "float32",
|
| 1560 |
+
"shape": [1, 1, 2, 2, 2],
|
| 1561 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0] }
|
| 1562 |
+
},
|
| 1563 |
+
"grid": {
|
| 1564 |
+
"dtype": "float32",
|
| 1565 |
+
"shape": [1, 1, 1, 4, 3],
|
| 1566 |
+
"data": { "kind": "values", "values": [-2.0, -2.0, -2.0, 2.0, 2.0, 2.0, 0.2, -0.2, 0.6, -0.6, 0.6, -0.2] }
|
| 1567 |
+
}
|
| 1568 |
+
},
|
| 1569 |
+
"outputs": {
|
| 1570 |
+
"y": {
|
| 1571 |
+
"dtype": "float32",
|
| 1572 |
+
"shape": [1, 1, 1, 1, 4],
|
| 1573 |
+
"tolerance": 0,
|
| 1574 |
+
"data": { "kind": "values", "values": [1.0, 8.0, 6.0, 3.0] }
|
| 1575 |
+
}
|
| 1576 |
+
}
|
| 1577 |
+
},
|
| 1578 |
+
{
|
| 1579 |
+
"name": "ort_custom_5d_nearest_reflection_nan_inf_coords",
|
| 1580 |
+
"provenance": {
|
| 1581 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 1582 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_5D_nearest_reflection_nan_inf_coords"
|
| 1583 |
+
},
|
| 1584 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 0 },
|
| 1585 |
+
"inputs": {
|
| 1586 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2, 2], "data": { "kind": "constant", "value": 1.0 } },
|
| 1587 |
+
"grid": {
|
| 1588 |
+
"dtype": "float32",
|
| 1589 |
+
"shape": [1, 1, 1, 2, 3],
|
| 1590 |
+
"data": { "kind": "values", "values": ["NaN", "Infinity", "-Infinity", "Infinity", "NaN", "-Infinity"] }
|
| 1591 |
+
}
|
| 1592 |
+
},
|
| 1593 |
+
"outputs": {
|
| 1594 |
+
"y": {
|
| 1595 |
+
"dtype": "float32",
|
| 1596 |
+
"shape": [1, 1, 1, 1, 2],
|
| 1597 |
+
"tolerance": 0,
|
| 1598 |
+
"data": { "kind": "values", "values": [1.0, 1.0] }
|
| 1599 |
+
}
|
| 1600 |
+
}
|
| 1601 |
+
},
|
| 1602 |
+
{
|
| 1603 |
+
"name": "ort_linear_zeros_no_align_corners_compact",
|
| 1604 |
+
"provenance": {
|
| 1605 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1606 |
+
"test": "GridSampleTest.test_grid_sample_16_4D_bilinear_zeros_no_align_corners",
|
| 1607 |
+
"notes": "Opset-20 spelling uses mode=linear for ORT's generated 4D bilinear no-align case."
|
| 1608 |
+
},
|
| 1609 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 1610 |
+
"inputs": {
|
| 1611 |
+
"x": {
|
| 1612 |
+
"dtype": "float32",
|
| 1613 |
+
"shape": [1, 1, 3, 3],
|
| 1614 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0] }
|
| 1615 |
+
},
|
| 1616 |
+
"grid": {
|
| 1617 |
+
"dtype": "float32",
|
| 1618 |
+
"shape": [1, 2, 3, 2],
|
| 1619 |
+
"data": { "kind": "values", "values": [-1.0, -1.0, 0.0, 0.0, 1.0, 1.0, -1.2, 0.4, 0.4, -1.2, 1.2, 1.2] }
|
| 1620 |
+
}
|
| 1621 |
+
},
|
| 1622 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 3], "tolerance": 0.000001 } }
|
| 1623 |
+
},
|
| 1624 |
+
{
|
| 1625 |
+
"name": "ort_linear_reflection_no_align_corners_compact",
|
| 1626 |
+
"provenance": {
|
| 1627 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1628 |
+
"test": "GridSampleTest.test_grid_sample_16_4D_bilinear_reflection_no_align_corners",
|
| 1629 |
+
"notes": "Compact projection of ORT's 4D bilinear reflection no-align coverage."
|
| 1630 |
+
},
|
| 1631 |
+
"attrs": { "mode": "linear", "padding_mode": "reflection", "align_corners": 0 },
|
| 1632 |
+
"inputs": {
|
| 1633 |
+
"x": {
|
| 1634 |
+
"dtype": "float32",
|
| 1635 |
+
"shape": [1, 1, 3, 3],
|
| 1636 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 4.0, 8.0, 16.0, 32.0, 64.0, 128.0, 256.0] }
|
| 1637 |
+
},
|
| 1638 |
+
"grid": {
|
| 1639 |
+
"dtype": "float32",
|
| 1640 |
+
"shape": [1, 2, 2, 2],
|
| 1641 |
+
"data": { "kind": "values", "values": [-1.4, -1.4, 1.4, 1.4, -0.2, 0.6, 2.2, -0.6] }
|
| 1642 |
+
}
|
| 1643 |
+
},
|
| 1644 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 2], "tolerance": 0.00001 } }
|
| 1645 |
+
},
|
| 1646 |
+
{
|
| 1647 |
+
"name": "ort_cubic_zeros_no_align_corners_compact",
|
| 1648 |
+
"provenance": {
|
| 1649 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1650 |
+
"test": "GridSampleTest.test_grid_sample_20_4D_bicubic_zeros_no_align_corners",
|
| 1651 |
+
"notes": "Opset-20 spelling uses mode=cubic for ORT's generated 4D bicubic zeros no-align case."
|
| 1652 |
+
},
|
| 1653 |
+
"attrs": { "mode": "cubic", "padding_mode": "zeros", "align_corners": 0 },
|
| 1654 |
+
"inputs": {
|
| 1655 |
+
"x": {
|
| 1656 |
+
"dtype": "float32",
|
| 1657 |
+
"shape": [1, 1, 4, 4],
|
| 1658 |
+
"data": { "kind": "values", "values": { "$ref": "#/fixtureArrays/onnx_backend_gridsample_input_x" } }
|
| 1659 |
+
},
|
| 1660 |
+
"grid": {
|
| 1661 |
+
"dtype": "float32",
|
| 1662 |
+
"shape": [1, 2, 2, 2],
|
| 1663 |
+
"data": { "kind": "values", "values": [-0.6, -0.6, 0.4, -0.2, 1.2, 1.2, -1.1, 0.3] }
|
| 1664 |
+
}
|
| 1665 |
+
},
|
| 1666 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 2, 2], "tolerance": 0.0001 } }
|
| 1667 |
+
},
|
| 1668 |
+
{
|
| 1669 |
+
"name": "ort_linear_border_rank5_no_align_corners",
|
| 1670 |
+
"provenance": {
|
| 1671 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test.cc",
|
| 1672 |
+
"test": "GridSampleTest.test_grid_sample_22_5D_bilinear_border_no_align_corners",
|
| 1673 |
+
"notes": "Valid 5D trilinear border-padding case."
|
| 1674 |
+
},
|
| 1675 |
+
"attrs": { "mode": "linear", "padding_mode": "border", "align_corners": 0 },
|
| 1676 |
+
"inputs": {
|
| 1677 |
+
"x": {
|
| 1678 |
+
"dtype": "float32",
|
| 1679 |
+
"shape": [1, 1, 2, 2, 2],
|
| 1680 |
+
"data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0] }
|
| 1681 |
+
},
|
| 1682 |
+
"grid": {
|
| 1683 |
+
"dtype": "float32",
|
| 1684 |
+
"shape": [1, 1, 1, 3, 3],
|
| 1685 |
+
"data": { "kind": "values", "values": [-1.2, -1.2, -1.2, 0.0, 0.0, 0.0, 1.2, 1.2, 1.2] }
|
| 1686 |
+
}
|
| 1687 |
+
},
|
| 1688 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1, 1, 3], "tolerance": 0.000001 } }
|
| 1689 |
+
},
|
| 1690 |
+
{
|
| 1691 |
+
"name": "ort_linear_reflection_rank5_extreme",
|
| 1692 |
+
"provenance": {
|
| 1693 |
+
"source": "onnxruntime/test/providers/cpu/tensor/grid_sample_test_custom.cc",
|
| 1694 |
+
"test": "GridSampleCustomTest.test_grid_sample_20_5D_linear_reflection_extreme_coords",
|
| 1695 |
+
"notes": "Valid 5D trilinear reflection-padding case with extreme coordinates."
|
| 1696 |
+
},
|
| 1697 |
+
"attrs": { "mode": "linear", "padding_mode": "reflection", "align_corners": 0 },
|
| 1698 |
+
"inputs": {
|
| 1699 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2, 2], "data": { "kind": "constant", "value": 1.0 } },
|
| 1700 |
+
"grid": {
|
| 1701 |
+
"dtype": "float32",
|
| 1702 |
+
"shape": [1, 1, 1, 2, 3],
|
| 1703 |
+
"data": {
|
| 1704 |
+
"kind": "values",
|
| 1705 |
+
"values": [100000000000000000000.0, 100000000000000000000.0, 100000000000000000000.0, -100000000000000000000.0, -100000000000000000000.0, -100000000000000000000.0]
|
| 1706 |
+
}
|
| 1707 |
+
}
|
| 1708 |
+
},
|
| 1709 |
+
"outputs": {
|
| 1710 |
+
"y": {
|
| 1711 |
+
"dtype": "float32",
|
| 1712 |
+
"shape": [1, 1, 1, 1, 2],
|
| 1713 |
+
"data": { "kind": "values", "values": [1.0, 1.0] },
|
| 1714 |
+
"tolerance": 0.000001
|
| 1715 |
+
}
|
| 1716 |
+
}
|
| 1717 |
+
},
|
| 1718 |
+
{
|
| 1719 |
+
"name": "empty_input_zero_dim",
|
| 1720 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 1 },
|
| 1721 |
+
"inputs": {
|
| 1722 |
+
"x": { "dtype": "float32", "shape": [0, 1, 2, 2], "data": { "kind": "values", "values": [] } },
|
| 1723 |
+
"grid": { "dtype": "float32", "shape": [0, 2, 2, 2], "data": { "kind": "values", "values": [] } }
|
| 1724 |
+
},
|
| 1725 |
+
"outputs": { "y": { "dtype": "float32", "shape": [0, 1, 2, 2], "tolerance": 0 } }
|
| 1726 |
+
},
|
| 1727 |
+
{
|
| 1728 |
+
"name": "ort_caseB_empty",
|
| 1729 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 1 },
|
| 1730 |
+
"inputs": {
|
| 1731 |
+
"x": { "dtype": "float32", "shape": [1, 1, 0, 2], "data": { "kind": "values", "values": [] } },
|
| 1732 |
+
"grid": {
|
| 1733 |
+
"dtype": "float32",
|
| 1734 |
+
"shape": [1, 2, 2, 2],
|
| 1735 |
+
"data": { "kind": "values", "values": [-1.0, -1.0, 1.0, -1.0, -1.0, 1.0, 1.0, 1.0] }
|
| 1736 |
+
}
|
| 1737 |
+
},
|
| 1738 |
+
"outputs": {
|
| 1739 |
+
"y": {
|
| 1740 |
+
"dtype": "float32",
|
| 1741 |
+
"shape": [1, 1, 2, 2],
|
| 1742 |
+
"data": { "kind": "values", "values": [0.0, 0.0, 0.0, 0.0] },
|
| 1743 |
+
"tolerance": 0.001
|
| 1744 |
+
}
|
| 1745 |
+
}
|
| 1746 |
+
},
|
| 1747 |
+
{
|
| 1748 |
+
"name": "f16_linear_zeros_feature_warp",
|
| 1749 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 1750 |
+
"inputs": {
|
| 1751 |
+
"x": {
|
| 1752 |
+
"dtype": "float16",
|
| 1753 |
+
"shape": [1, 32, 32, 32],
|
| 1754 |
+
"data": { "kind": "fillFloat32", "scale": 0.5, "sinStep": 0.07, "cosStep": 0.13 }
|
| 1755 |
+
},
|
| 1756 |
+
"grid": {
|
| 1757 |
+
"dtype": "float16",
|
| 1758 |
+
"shape": [1, 32, 32, 2],
|
| 1759 |
+
"data": { "kind": "linspace", "start": -1.0, "end": 1.0 }
|
| 1760 |
+
}
|
| 1761 |
+
},
|
| 1762 |
+
"outputs": { "y": { "dtype": "float16", "shape": [1, 32, 32, 32], "tolerance": 0.02, "relTolerance": 0.02 } }
|
| 1763 |
+
},
|
| 1764 |
+
{
|
| 1765 |
+
"name": "f16_linear_border_align_corners_stn",
|
| 1766 |
+
"attrs": { "mode": "linear", "padding_mode": "border", "align_corners": 1 },
|
| 1767 |
+
"inputs": {
|
| 1768 |
+
"x": {
|
| 1769 |
+
"dtype": "float16",
|
| 1770 |
+
"shape": [1, 16, 24, 24],
|
| 1771 |
+
"data": { "kind": "fillFloat32", "scale": 0.4, "sinStep": 0.11, "cosStep": 0.19 }
|
| 1772 |
+
},
|
| 1773 |
+
"grid": {
|
| 1774 |
+
"dtype": "float16",
|
| 1775 |
+
"shape": [1, 24, 24, 2],
|
| 1776 |
+
"data": { "kind": "linspace", "start": -1.2, "end": 1.2 }
|
| 1777 |
+
}
|
| 1778 |
+
},
|
| 1779 |
+
"outputs": { "y": { "dtype": "float16", "shape": [1, 16, 24, 24], "tolerance": 0.02, "relTolerance": 0.02 } }
|
| 1780 |
+
},
|
| 1781 |
+
{
|
| 1782 |
+
"name": "f16_volumetric_linear_zeros_3d_warp",
|
| 1783 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 1784 |
+
"inputs": {
|
| 1785 |
+
"x": {
|
| 1786 |
+
"dtype": "float16",
|
| 1787 |
+
"shape": [1, 4, 8, 16, 16],
|
| 1788 |
+
"data": { "kind": "fillFloat32", "scale": 0.5, "sinStep": 0.05, "cosStep": 0.09 }
|
| 1789 |
+
},
|
| 1790 |
+
"grid": {
|
| 1791 |
+
"dtype": "float16",
|
| 1792 |
+
"shape": [1, 8, 16, 16, 3],
|
| 1793 |
+
"data": { "kind": "linspace", "start": -1.0, "end": 1.0 }
|
| 1794 |
+
}
|
| 1795 |
+
},
|
| 1796 |
+
"outputs": { "y": { "dtype": "float16", "shape": [1, 4, 8, 16, 16], "tolerance": 0.03, "relTolerance": 0.03 } }
|
| 1797 |
+
},
|
| 1798 |
+
{
|
| 1799 |
+
"name": "f16_nearest_reflection_warp",
|
| 1800 |
+
"attrs": { "mode": "nearest", "padding_mode": "reflection", "align_corners": 0 },
|
| 1801 |
+
"inputs": {
|
| 1802 |
+
"x": {
|
| 1803 |
+
"dtype": "float16",
|
| 1804 |
+
"shape": [1, 8, 28, 28],
|
| 1805 |
+
"data": { "kind": "fillFloat32", "scale": 0.6, "sinStep": 0.13, "cosStep": 0.23 }
|
| 1806 |
+
},
|
| 1807 |
+
"grid": {
|
| 1808 |
+
"dtype": "float16",
|
| 1809 |
+
"shape": [1, 28, 28, 2],
|
| 1810 |
+
"data": { "kind": "linspace", "start": -1.5, "end": 1.5 }
|
| 1811 |
+
}
|
| 1812 |
+
},
|
| 1813 |
+
"outputs": { "y": { "dtype": "float16", "shape": [1, 8, 28, 28], "tolerance": 0.02, "relTolerance": 0.02 } }
|
| 1814 |
+
},
|
| 1815 |
+
{
|
| 1816 |
+
"name": "empty_output_zero_grid_spatial_nonempty_inputs",
|
| 1817 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 1818 |
+
"inputs": {
|
| 1819 |
+
"x": {
|
| 1820 |
+
"dtype": "float32",
|
| 1821 |
+
"shape": [1, 2, 3, 4],
|
| 1822 |
+
"data": { "kind": "fillFloat32", "sinStep": 0.7, "scale": 1.0, "offset": 0.0 }
|
| 1823 |
+
},
|
| 1824 |
+
"grid": {
|
| 1825 |
+
"dtype": "float32",
|
| 1826 |
+
"shape": [1, 0, 4, 2],
|
| 1827 |
+
"data": { "kind": "fillFloat32", "sinStep": 0.3, "scale": 1.0, "offset": 0.0 }
|
| 1828 |
+
}
|
| 1829 |
+
},
|
| 1830 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 2, 0, 4] } }
|
| 1831 |
+
},
|
| 1832 |
+
{
|
| 1833 |
+
"name": "linear_zeros_fold_boundary_partial_last_row_over16m",
|
| 1834 |
+
"attrs": { "mode": "linear", "padding_mode": "zeros", "align_corners": 0 },
|
| 1835 |
+
"inputs": {
|
| 1836 |
+
"x": {
|
| 1837 |
+
"dtype": "float32",
|
| 1838 |
+
"shape": [1, 3, 16, 16],
|
| 1839 |
+
"data": { "kind": "fillFloat32", "sinStep": 0.13, "scale": 1.0, "offset": 0.0 }
|
| 1840 |
+
},
|
| 1841 |
+
"grid": {
|
| 1842 |
+
"dtype": "float32",
|
| 1843 |
+
"shape": [1, 2049, 2731, 2],
|
| 1844 |
+
"data": { "kind": "fillFloat32", "sinStep": 0.017, "scale": 0.9, "offset": 0.0 }
|
| 1845 |
+
}
|
| 1846 |
+
},
|
| 1847 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 3, 2049, 2731] } }
|
| 1848 |
+
},
|
| 1849 |
+
{
|
| 1850 |
+
"name": "cubic_reflection_channel_quad_tail_compact",
|
| 1851 |
+
"provenance": {
|
| 1852 |
+
"source": "ONNX GridSample-20 cubic reflection semantics",
|
| 1853 |
+
"notes": "Exercises channel-cooperative sampling with a non-multiple-of-four channel tail and out-of-range coordinates."
|
| 1854 |
+
},
|
| 1855 |
+
"attrs": { "mode": "cubic", "padding_mode": "reflection", "align_corners": 0 },
|
| 1856 |
+
"inputs": {
|
| 1857 |
+
"x": {
|
| 1858 |
+
"dtype": "float32",
|
| 1859 |
+
"shape": [1, 5, 4, 5],
|
| 1860 |
+
"data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.29, "scale": 1.0 }
|
| 1861 |
+
},
|
| 1862 |
+
"grid": { "dtype": "float32", "shape": [1, 3, 4, 2], "data": { "kind": "linspace", "start": -1.6, "end": 1.4 } }
|
| 1863 |
+
},
|
| 1864 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 5, 3, 4], "tolerance": 0.0001, "relTolerance": 0.0001 } }
|
| 1865 |
+
},
|
| 1866 |
+
{
|
| 1867 |
+
"name": "rank5_cubic_border_partition_of_unity",
|
| 1868 |
+
"provenance": {
|
| 1869 |
+
"notes": "A constant field: every interpolation returns the constant, so this sees only that the tricubic weights sum to one on each axis. It cannot tell cubic from linear or catch a wrong tap offset — rank5_cubic_border_pinned_taps is the case that does."
|
| 1870 |
+
},
|
| 1871 |
+
"attrs": { "mode": "cubic", "padding_mode": "border", "align_corners": 0 },
|
| 1872 |
+
"inputs": {
|
| 1873 |
+
"x": { "dtype": "float32", "shape": [1, 1, 2, 2, 2], "data": { "kind": "constant", "value": 2.0 } },
|
| 1874 |
+
"grid": {
|
| 1875 |
+
"dtype": "float32",
|
| 1876 |
+
"shape": [1, 1, 1, 1, 3],
|
| 1877 |
+
"data": { "kind": "values", "values": [0.0, 0.0, 0.0] }
|
| 1878 |
+
}
|
| 1879 |
+
},
|
| 1880 |
+
"outputs": {
|
| 1881 |
+
"y": {
|
| 1882 |
+
"dtype": "float32",
|
| 1883 |
+
"shape": [1, 1, 1, 1, 1],
|
| 1884 |
+
"tolerance": 0.000001,
|
| 1885 |
+
"data": { "kind": "values", "values": [2.0] }
|
| 1886 |
+
}
|
| 1887 |
+
}
|
| 1888 |
+
},
|
| 1889 |
+
{
|
| 1890 |
+
"name": "rank5_cubic_border_pinned_taps",
|
| 1891 |
+
"provenance": {
|
| 1892 |
+
"source": "ONNX GridSample cubic definition (Keys kernel, cubic_coeff_a = -0.75), evaluated independently",
|
| 1893 |
+
"notes": "Ground truth computed independently from the separable 4-tap definition, so neither the kernel nor the reference is its own oracle. Sample points are dyadic and the field is in quarters, so the values are exact in binary; two of the four points push taps past the volume, which border clamps. Trilinear on the same points gives different answers, so a cubic path that silently ran as linear would fail here."
|
| 1894 |
+
},
|
| 1895 |
+
"attrs": { "mode": "cubic", "padding_mode": "border", "align_corners": 0 },
|
| 1896 |
+
"inputs": {
|
| 1897 |
+
"x": {
|
| 1898 |
+
"dtype": "float32",
|
| 1899 |
+
"shape": [1, 2, 4, 4, 4],
|
| 1900 |
+
"data": {
|
| 1901 |
+
"kind": "values",
|
| 1902 |
+
"values": [-2.75, -2.5, -2.25, -2.0, -1.75, -1.5, -1.25, -1.0, -0.75, -0.5, -0.25, 0.0, 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75, 2.0, 2.25, 2.5, 2.75, -2.75, -2.5, -2.25, -2.0, -1.75, -1.5, -1.25, -1.0, -0.75, -0.5, -0.25, 0.0, 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75, 2.0, 2.25, 2.5, 2.75, -2.75, -2.5, -2.25, -2.0, -1.75, -1.5, -1.25, -1.0, -0.75, -0.5, -0.25, 0.0, 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75, 2.0, 2.25, 2.5, 2.75, -2.75, -2.5, -2.25, -2.0, -1.75, -1.5, -1.25, -1.0, -0.75, -0.5, -0.25, 0.0, 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75, 2.0, 2.25, 2.5, 2.75, -2.75, -2.5, -2.25, -2.0, -1.75, -1.5, -1.25, -1.0, -0.75, -0.5, -0.25, 0.0, 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75, 2.0, 2.25, 2.5, 2.75, -2.75, -2.5, -2.25, -2.0, -1.75, -1.5, -1.25, -1.0, -0.75, -0.5, -0.25, 0.0, 0.25]
|
| 1903 |
+
}
|
| 1904 |
+
},
|
| 1905 |
+
"grid": {
|
| 1906 |
+
"dtype": "float32",
|
| 1907 |
+
"shape": [1, 1, 2, 2, 3],
|
| 1908 |
+
"data": { "kind": "values", "values": [-0.5, 0.5, 0.0, 1.0, -1.0, 0.25, 0.25, 0.75, -0.75, -0.25, 0.0, 0.5] }
|
| 1909 |
+
}
|
| 1910 |
+
},
|
| 1911 |
+
"outputs": {
|
| 1912 |
+
"y": {
|
| 1913 |
+
"dtype": "float32",
|
| 1914 |
+
"shape": [1, 2, 1, 2, 2],
|
| 1915 |
+
"tolerance": 0.00001,
|
| 1916 |
+
"relTolerance": 0.00001,
|
| 1917 |
+
"data": {
|
| 1918 |
+
"kind": "values",
|
| 1919 |
+
"values": [0.0352783203125, 0.1796875, 0.75, 0.48046875, 0.6776123046875, -1.0703125, -0.5, -1.359130859375]
|
| 1920 |
+
}
|
| 1921 |
+
}
|
| 1922 |
+
}
|
| 1923 |
+
},
|
| 1924 |
+
{
|
| 1925 |
+
"name": "rank5_cubic_zeros_align_corners",
|
| 1926 |
+
"provenance": {
|
| 1927 |
+
"notes": "Cubic with zeros padding and align_corners, so the 4-tap window reads outside the volume and must contribute nothing there. The grid scale puts roughly half the sample points out of range."
|
| 1928 |
+
},
|
| 1929 |
+
"attrs": { "mode": "cubic", "padding_mode": "zeros", "align_corners": 1 },
|
| 1930 |
+
"inputs": {
|
| 1931 |
+
"x": {
|
| 1932 |
+
"dtype": "float32",
|
| 1933 |
+
"shape": [1, 2, 5, 5, 5],
|
| 1934 |
+
"data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 1.5 }
|
| 1935 |
+
},
|
| 1936 |
+
"grid": {
|
| 1937 |
+
"dtype": "float32",
|
| 1938 |
+
"shape": [1, 2, 2, 2, 3],
|
| 1939 |
+
"data": { "kind": "fillFloat32", "sinStep": 0.37, "cosStep": 0.19, "scale": 1.3 }
|
| 1940 |
+
}
|
| 1941 |
+
},
|
| 1942 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 2, 2, 2, 2], "tolerance": 0.0001, "relTolerance": 0.0001 } }
|
| 1943 |
+
},
|
| 1944 |
+
{
|
| 1945 |
+
"name": "rank5_cubic_reflection",
|
| 1946 |
+
"provenance": {
|
| 1947 |
+
"notes": "Cubic with reflection padding: the out-of-range taps fold back inside, which is the one padding mode whose resolved index depends on align_corners as well as the bound."
|
| 1948 |
+
},
|
| 1949 |
+
"attrs": { "mode": "cubic", "padding_mode": "reflection", "align_corners": 0 },
|
| 1950 |
+
"inputs": {
|
| 1951 |
+
"x": {
|
| 1952 |
+
"dtype": "float32",
|
| 1953 |
+
"shape": [1, 2, 5, 5, 5],
|
| 1954 |
+
"data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.41, "scale": 1.5 }
|
| 1955 |
+
},
|
| 1956 |
+
"grid": {
|
| 1957 |
+
"dtype": "float32",
|
| 1958 |
+
"shape": [1, 2, 2, 2, 3],
|
| 1959 |
+
"data": { "kind": "fillFloat32", "sinStep": 0.37, "cosStep": 0.19, "scale": 1.3 }
|
| 1960 |
+
}
|
| 1961 |
+
},
|
| 1962 |
+
"outputs": { "y": { "dtype": "float32", "shape": [1, 2, 2, 2, 2], "tolerance": 0.0001, "relTolerance": 0.0001 } }
|
| 1963 |
+
}
|
| 1964 |
+
]
|
| 1965 |
+
}
|