)]}'
{
  "commit": "9300ad63ce1165f3119e0fc680a12c0daf072ede",
  "tree": "0bc98eb5f9659cae8bdccf6ec715c11b31bfa6f4",
  "parents": [
    "82d69d832a42e7d877046d651a6e5e4b3f1ffadd"
  ],
  "author": {
    "name": "Deepa K G",
    "email": "deepa.kg@ittiam.com",
    "time": "Tue Mar 10 08:54:20 2026 +0530"
  },
  "committer": {
    "name": "Yunqing Wang",
    "email": "yunqingwang@google.com",
    "time": "Mon Mar 16 15:45:49 2026 -0700"
  },
  "message": "Fix sf \u0027reuse_compound_type_decision\u0027\n\nThe speed feature \u0027reuse_compound_type_decision\u0027\nreuses the compound type RD decision when the\nreference frames and motion vectors match. When\nmatch is found, RD cost is calculated using the\nstored stats. However, for COMPOUND_AVERAGE, rate\nand distortion are not backed up resulting in\nearly termination of the matched mode.\n\nThe sf \u0027prune_comp_type_by_model_rd\u0027 is supposed to\nprune approximate RD evaluation of COMPOUND_DIFFWTD\nbased on modelled RD of COMPOUND_AVERAGE. However, as\nmodelled RD is not calculated for COMPOUND_AVERAGE,\nthe sf \u0027prune_comp_type_by_model_rd\u0027 does not prune\napproximate RD evaluation of COMPOUND_DIFFWTD.\n\nThis patch adds backup of rate and distortion and\ncalculates the modelled RD for COMPOUND_AVERAGE and\nthus fixes the above issues.\n\nSTATS_CHANGED for speed\u003e\u003d1\n\nChange-Id: I8adf28ec5f78cd74cd52cb985a1c1d56133aa480\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "778c3f1cf60ca8eb15e6875e193d3f17bb872c3d",
      "old_mode": 33188,
      "old_path": "av1/encoder/compound_type.c",
      "new_id": "f8364f1cf9da5b829699a9935fb47c4200c98d84",
      "new_mode": 33188,
      "new_path": "av1/encoder/compound_type.c"
    }
  ]
}