)]}'
{
  "commit": "9517176b8702ff1eae2077041c844c4d59dd8371",
  "tree": "35fa4b14d20ae982f36399af05fbd5744dcd371d",
  "parents": [
    "e1e516f3b800f2098f824e9f87d7c263cf460846"
  ],
  "author": {
    "name": "Frank Galligan",
    "email": "fgalligan@google.com",
    "time": "Fri Oct 30 12:07:14 2026 -0700"
  },
  "committer": {
    "name": "Frank Galligan",
    "email": "fgalligan@google.com",
    "time": "Mon Nov 09 15:10:28 2026 -0800"
  },
  "message": "av1: Optimize avg_cdf_symbol and AC dequant\n\n- Specialize avg_cdf_symbol for the common (wt_left \u003d\u003d 3, wt_tr \u003d\u003d 1)\n  row-multithreading CDF update using shift-and-average arithmetic that\n  auto-vectorizes with 16-bit SIMD instructions.\n- Use dequant[1] directly in update_coeff_simple since it is only\n  called on AC coefficients (si \u003e 0).\n\nBenchmarking results (incremental vs prev CL / cumulative vs baseline):\n- Stock avifenc still-image (Default 1T): -0.45% / -18.78% encode time\n- Stock avifenc still-image (Default 8T): -0.50% / -15.05% encode time\n- 48-image multi-rate corpus (Full Suite 1T): -0.55% / -18.73% time\n- Bit-exact output (1T and 8T)\n\nTAG\u003dagy\nCONV\u003db929a630-a6cd-45a0-979c-1767c30178ee\nChange-Id: Ib997b3791b37da2622a0237c9ac0c51e0b7af90b\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "0dc7ef74b2e9804e2654a76c6bad9bf1f87b8dca",
      "old_mode": 33188,
      "old_path": "av1/encoder/encodeframe_utils.c",
      "new_id": "58cb0ea0677efcc44820ecff04bbd8e980058eb4",
      "new_mode": 33188,
      "new_path": "av1/encoder/encodeframe_utils.c"
    },
    {
      "type": "modify",
      "old_id": "f03217165118d446f080e12f87246535d54ff8ed",
      "old_mode": 33188,
      "old_path": "av1/encoder/txb_rdopt.c",
      "new_id": "4b09adea621362bf771e2e27ee9c66574c0541ab",
      "new_mode": 33188,
      "new_path": "av1/encoder/txb_rdopt.c"
    }
  ]
}