| { |
| "domain": "ai.onnx", |
| "name": "EyeLike", |
| "sinceVersion": 22, |
| "inputs": { "input": { "dtype": "T1", "rank": 2 } }, |
| "outputs": { "output": { "dtype": "T2", "rank": "ranks.input", "shape": "shapes.input" } }, |
| "attributes": { "k": { "default": 0 }, "dtype": {} }, |
| "typeConstraints": { |
| "T1": ["float32", "float16", "uint32", "int32", "int16", "uint8", "int8", "bool"], |
| "T2": ["float32", "float16", "uint32", "int32", "int16", "uint8", "int8", "bool"] |
| }, |
| "tunables": { "WORKGROUP_SIZE": { "default": 256 } }, |
| "derive": { |
| "wave32Adapter": "has(device.adapterInfo, \"subgroupMinSize\") and has(device.adapterInfo, \"subgroupMaxSize\") and device.adapterInfo.subgroupMinSize == 32 and device.adapterInfo.subgroupMaxSize == 32", |
| "reportedNonWave32Adapter": "not wave32Adapter and (has(device.adapterInfo, \"subgroupMinSize\") or has(device.adapterInfo, \"subgroupMaxSize\"))", |
| "dtypeContract": "(has(attrs, \"dtype\") and attrs.dtype == onnxDtypeCode(logicalDtypes.T2)) or (not has(attrs, \"dtype\") and logicalDtypes.T2 == logicalDtypes.T1)", |
| "packedClearPreferred": "numel(shapes.output) % 4 == 0 or not (reportedNonWave32Adapter and device.features.has(\"subgroups\"))", |
| "outScalar": "dtypes.T2", |
| "diagonalK": "max(min(attrs.k, dim(shapes.output, 1)), 0 - dim(shapes.output, 0))" |
| }, |
| "when": ["dtypeContract", "ranks.input == 2", "ranks.output == 2", "dim(shapes.output, 0) == dim(shapes.input, 0)", "dim(shapes.output, 1) == dim(shapes.input, 1)", "f16Ok(tensorDtypes.output)"], |
| "variants": [ |
| { |
| "id": "rank2_vec4", |
| "priority": 20, |
| "when": ["dim(shapes.output, 1) > 0", "packedClearPreferred"], |
| "derive": { |
| "tailSafe": "numel(shapes.output) % 4 != 0", |
| "outVector": "\"vec4<\" ~ dtypes.T2 ~ \">\"", |
| "clearElement": "dtypes.T2 if numel(shapes.output) % 4 != 0 else (\"vec4<\" ~ dtypes.T2 ~ \">\")", |
| "diagonalRows": "max(0, min(dim(shapes.output, 0), dim(shapes.output, 1) - diagonalK) - max(0, 0 - diagonalK))" |
| }, |
| "passes": [ |
| { |
| "id": "clear", |
| "name": "EyeLike.ClearVec4", |
| "shader": "eyelike-clear-vec4.wgsl.jinja", |
| "derive": { "count4": "ceilDiv(numel(shapes.output), 4)", "count": "numel(shapes.output)" }, |
| "bindings": [{ "arg": "output", "elementType": "$clearElement" }], |
| "dispatch": { |
| "x": "min(ceilDiv((ceilDiv(numel(shapes.output), 4)), (tunables.WORKGROUP_SIZE)), 65535)", |
| "y": "ceilDiv(ceilDiv((ceilDiv(numel(shapes.output), 4)), (tunables.WORKGROUP_SIZE)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "diagonal", |
| "name": "EyeLike.Diagonal", |
| "shader": "eyelike-diagonal.wgsl.jinja", |
| "bindings": [ |
| "output", |
| { |
| "name": "params", |
| "struct": [ |
| { "name": "rows", "type": "u32", "value": "dim(shapes.output, 0)" }, |
| { "name": "cols", "type": "u32", "value": "dim(shapes.output, 1)" }, |
| { "name": "k", "type": "i32", "value": "diagonalK" } |
| ] |
| } |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((max(1, diagonalRows)), (tunables.WORKGROUP_SIZE)), 65535)", |
| "y": "ceilDiv(ceilDiv((max(1, diagonalRows)), (tunables.WORKGROUP_SIZE)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| }, |
| { |
| "id": "rank2", |
| "passes": [ |
| { |
| "id": "main", |
| "name": "EyeLike", |
| "shader": "eyelike.wgsl.jinja", |
| "bindings": [ |
| "output", |
| { |
| "name": "params", |
| "struct": [ |
| { "name": "count", "type": "u32", "value": "numel(shapes.output)" }, |
| { "name": "cols", "type": "u32", "value": "dim(shapes.output, 1)" }, |
| { "name": "k", "type": "i32", "value": "diagonalK" } |
| ] |
| } |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((numel(shapes.output)), (tunables.WORKGROUP_SIZE)), 65535)", |
| "y": "ceilDiv(ceilDiv((numel(shapes.output)), (tunables.WORKGROUP_SIZE)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| } |
|
|
