{
  "domain": "ai.onnx",
  "name": "Mod",
  "sinceVersion": 13,
  "inputs": { "a": { "onnx": "A", "dtype": "T" }, "b": { "onnx": "B", "dtype": "T" } },
  "outputs": {
    "c": { "onnx": "C", "dtype": "T", "rank": "max(ranks.a, ranks.b)", "shape": "broadcastShape(shapes.a, shapes.b)" }
  },
  "attributes": { "fmod": { "default": 0 } },
  "attributeConstraints": { "fmod": { "values": [0, 1] } },
  "typeConstraints": { "T": ["float32", "float16", "int32", "uint32", "int16", "int8", "uint8"] },
  "tunables": { "WORKGROUP_SIZE": { "default": 256 } },
  "derive": {
    "floatFmodContract": "attrs.fmod != 0 or not (dtypes.T == \"f32\" or dtypes.T == \"f16\")",
    "scalar": "dtypes.T",
    "isU32": "dtypes.T == \"u32\"",
    "isI32": "dtypes.T == \"i32\"",
    "fmod": "attrs.fmod != 0",
    "vectorScalar": "\"vec4<\" ~ dtypes.T ~ \">\""
  },
  "when": ["f16Ok(dtypes.T)", "floatFmodContract"],
  "bindings": {
    "a": { "elementType": "$vectorScalar" },
    "c": { "elementType": "$vectorScalar" },
    "params": { "struct": [{ "name": "count", "type": "u32", "value": "numel(shapes.c) / 4" }] }
  },
  "variants": [
    {
      "id": "last_axis_broadcast_vec4",
      "priority": 25,
      "when": ["ranks.c >= 1", "sameShape(shapes.a, shapes.c)", "ranks.b == 1", "dim(shapes.b, 0) == dim(shapes.c, ranks.c - 1)", "numel(shapes.c) > 0", "numel(shapes.c) % 4 == 0"],
      "passes": [
        {
          "id": "main",
          "name": "Mod.LastAxisBroadcastVec4",
          "shader": "mod-vec4.wgsl.jinja",
          "derive": { "perAxisB": true, "axisDimSpec": "dim(shapes.b, 0)" },
          "bindings": ["a", "b", "c", "params"],
          "dispatch": {
            "x": "min(ceilDiv((numel(shapes.c) / 4), (tunables.WORKGROUP_SIZE)), 65535)",
            "y": "ceilDiv(ceilDiv((numel(shapes.c) / 4), (tunables.WORKGROUP_SIZE)), 65535)",
            "z": 1
          }
        }
      ]
    },
    {
      "id": "scalar_b_vec4",
      "priority": 30,
      "when": ["sameShape(shapes.a, shapes.c)", "numel(shapes.b) == 1", "numel(shapes.c) > 0", "numel(shapes.c) % 4 == 0"],
      "passes": [
        {
          "id": "main",
          "name": "Mod.ScalarBVec4",
          "shader": "mod-vec4.wgsl.jinja",
          "derive": { "scalarB": true },
          "bindings": ["a", { "arg": "b", "elementType": "$scalar", "length": 1 }, "c", "params"],
          "dispatch": {
            "x": "min(ceilDiv((numel(shapes.c) / 4), (tunables.WORKGROUP_SIZE)), 65535)",
            "y": "ceilDiv(ceilDiv((numel(shapes.c) / 4), (tunables.WORKGROUP_SIZE)), 65535)",
            "z": 1
          }
        }
      ]
    },
    {
      "id": "same_shape_vec4",
      "priority": 26,
      "when": ["sameShape(shapes.a, shapes.c)", "sameShape(shapes.b, shapes.c)", "numel(shapes.c) > 0", "numel(shapes.c) % 4 == 0"],
      "passes": [
        {
          "id": "main",
          "name": "Mod.Vec4",
          "shader": "mod-vec4.wgsl.jinja",
          "bindings": ["a", { "arg": "b", "elementType": "$vectorScalar" }, "c", "params"],
          "dispatch": {
            "x": "min(ceilDiv((numel(shapes.c) / 4), (tunables.WORKGROUP_SIZE)), 65535)",
            "y": "ceilDiv(ceilDiv((numel(shapes.c) / 4), (tunables.WORKGROUP_SIZE)), 65535)",
            "z": 1
          }
        }
      ]
    },
    {
      "id": "broadcast",
      "when": ["ranks.a >= 0", "ranks.b >= 0", "ranks.c >= 0", "ranks.a <= ranks.c", "ranks.b <= ranks.c"],
      "passes": [
        {
          "id": "main",
          "name": "Mod",
          "shader": "mod.wgsl.jinja",
          "derive": {
            "aShape": "shapes.a",
            "bShape": "shapes.b",
            "cShape": "shapes.c",
            "aRank": "ranks.a",
            "bRank": "ranks.b",
            "cRank": "ranks.c"
          },
          "bindings": [
            { "arg": "a" },
            "b",
            { "arg": "c" },
            { "name": "params", "struct": [{ "name": "count", "type": "u32", "value": "numel(shapes.c)" }] }
          ],
          "dispatch": {
            "x": "min(ceilDiv((numel(shapes.c)), (tunables.WORKGROUP_SIZE)), 65535)",
            "y": "ceilDiv(ceilDiv((numel(shapes.c)), (tunables.WORKGROUP_SIZE)), 65535)",
            "z": 1
          }
        }
      ]
    }
  ]
}