| { |
| "domain": "ai.onnx", |
| "name": "BitwiseNot", |
| "sinceVersion": 18, |
| "inputs": { "x": { "onnx": "X", "dtype": "T" } }, |
| "outputs": { "y": { "onnx": "Y", "dtype": "T", "rank": "ranks.x", "shape": "shapes.x" } }, |
| "typeConstraints": { "T": ["uint32", "int32", "int16", "uint8", "int8"] }, |
| "tunables": { "WORKGROUP_SIZE": { "default": 256 } }, |
| "derive": { "shapeContract": "numel(shapes.x) == numel(shapes.y)", "logicalDtype": "tensorDtypes.y" }, |
| "when": ["shapeContract"], |
| "variants": [ |
| { |
| "id": "elementwise_vec4", |
| "priority": 20, |
| "when": ["numel(shapes.x) % 4 == 0"], |
| "derive": { "vectorScalar": "\"vec4<\" ~ dtypes.T ~ \">\"" }, |
| "passes": [ |
| { |
| "id": "main", |
| "name": "BitwiseNot.Vec4", |
| "shader": "bitwise-not.wgsl.jinja", |
| "derive": { "vectorizedSpec": true }, |
| "bindings": [ |
| { "arg": "x", "elementType": "$vectorScalar" }, |
| { "arg": "y", "elementType": "$vectorScalar" }, |
| { "name": "params", "struct": [{ "name": "count", "type": "u32", "value": "numel(shapes.y) / 4" }] } |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((numel(shapes.y) / 4), (tunables.WORKGROUP_SIZE)), 65535)", |
| "y": "ceilDiv(ceilDiv((numel(shapes.y) / 4), (tunables.WORKGROUP_SIZE)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| }, |
| { |
| "id": "elementwise", |
| "passes": [ |
| { |
| "id": "main", |
| "name": "BitwiseNot", |
| "shader": "bitwise-not.wgsl.jinja", |
| "derive": { "vectorizedSpec": false }, |
| "bindings": [ |
| "x", |
| "y", |
| { "name": "params", "struct": [{ "name": "count", "type": "u32", "value": "numel(shapes.y)" }] } |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((numel(shapes.y)), (tunables.WORKGROUP_SIZE)), 65535)", |
| "y": "ceilDiv(ceilDiv((numel(shapes.y)), (tunables.WORKGROUP_SIZE)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| } |
|
|
