{
  "op": "ai.onnx.Expand",
  "tunableSpace": { "WORKGROUP_SIZE": [64, 128, 256] },
  "cases": [
    {
      "name": "expand-f32-1m",
      "preset": "smoke",
      "vars": { "dtype": "float32", "count": 1048576 },
      "inputs": {
        "input": { "shape": [1, 1024], "dtype": "float32", "dist": "normal", "seed": 860, "scale": 1 },
        "shape": { "shape": [2], "dtype": "uint32", "data": { "kind": "values", "values": [1024, 1024] } }
      },
      "outputs": { "output": { "shape": [1024, 1024], "dtype": "float32" } },
      "bench": {
        "primary": true,
        "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }]
      }
    },
    {
      "name": "expand-u32-bool-like-1m",
      "preset": "edge",
      "vars": { "dtype": "uint32", "count": 1048576 },
      "inputs": {
        "input": { "shape": [1, 1024], "dtype": "uint32", "dist": "linearMod", "mod": 2 },
        "shape": { "shape": [2], "dtype": "uint32", "data": { "kind": "values", "values": [1024, 1024] } }
      },
      "outputs": { "output": { "shape": [1024, 1024], "dtype": "uint32" } },
      "bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] }
    },
    {
      "name": "expand-f32-scalar-fallback-1m",
      "preset": "stress",
      "vars": { "dtype": "float32", "count": 1049600 },
      "inputs": {
        "input": { "shape": [1, 1025], "dtype": "float32", "dist": "normal", "seed": 42, "scale": 1 },
        "shape": { "shape": [2], "dtype": "uint32", "data": { "kind": "values", "values": [1024, 1025] } }
      },
      "outputs": { "output": { "shape": [1024, 1025], "dtype": "float32" } },
      "bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] }
    }
  ]
}