Tensor Profile

The tensor profile targets numeric blocks, region-oriented payloads, and structured numeric results.

Good-fit scenarios

1. Real-time enhanced rendering.
2. Image or video enhancement pipelines with structured numeric output.
3. Any result that needs shape, layout, and dtype together for interpretation.

Reading path

1. Start with Descriptor Common Header to understand shared fields such as profile_id, schema_id, stream_semantics, offset, and length.
2. Continue with Schema and Body to see which fields tensor adds on top of the shared descriptor.
3. Finish with Payload Frame to see how the real binary tensor chunk is anchored by the descriptor and body.

Packet skeleton

Mock Dump Example

This is not a byte-accurate wire dump. It is a JSON-style mock dump that keeps the whole structure readable while the child pages break down each part in more detail.

{
  "message_type": "RESULT_PUSH",
  "common_header": {
    "version_major": 1,
    "wire_format": 0,
    "msg_type": "RESULT_PUSH",
    "header_len": 40,
    "meta_len": 32,
    "body_len": 80,
    "session_id": "0x00000012",
    "frame_id": "0x0000048a",
    "view_id": "0x00000007",
    "route_id": "0x00000002",
    "trace_id": "0x4f9c9b31b26d40d2"
  },
  "fixed_metadata": {
    "result_status": "partial",
    "flow_scope": "operation",
    "flow_credit_delta": 2
  },
  "typed_payload_descriptor": {
    "profile_id": "tensor",
    "schema_id": "im.render.tile.v1",
    "schema_version": 1,
    "stream_semantics": "region_partial",
    "offset": 0,
    "length": 3145728,
    "descriptor_flags": ["degraded_allowed", "coverage_present"]
  },
  "profile_body": {
    "tensor": {
      "dtype": "f16",
      "shape": [1, 3, 512, 512],
      "layout": "nchw",
      "coverage": {
        "tile_x": 12,
        "tile_y": 7,
        "width": 512,
        "height": 512
      }
    }
  },
  "payload_frame": {
    "tensor_chunk": "<1536 KiB binary tensor payload>"
  }
}

Semantics people care about first

1. partial means the current numeric result is already usable but not final.
2. degraded means the result is still consumable but quality, precision, or provenance changed.
3. stale_reuse means the result comes from reused older context rather than a fresh full computation.