)]}'
{
  "commit": "f55fa5ee4eebad77d9271198436f27123340ed52",
  "tree": "7c4b341e2688591226afd0e21fb5fd42e2f545d7",
  "parents": [
    "6d691cecc715c3b9985d22a72932b07d0d82a8c6"
  ],
  "author": {
    "name": "Jerome Jiang",
    "email": "jianj@google.com",
    "time": "Thu Jun 11 14:08:33 2026 -0400"
  },
  "committer": {
    "name": "Jerome Jiang",
    "email": "jianj@google.com",
    "time": "Mon Jun 29 12:24:26 2026 -0700"
  },
  "message": "Optimize update_coeff_* functions in txb_rdopt.c\n\nThis change is bitexact.\n\nAvoid calling get_coeff_dist() with value\u003d0 (which computes dist0)\nwhen qmatrix \u003d\u003d NULL.\n\nThis is achieved by rewriting the RD cost comparison using relative\noffset distortions compared to dist0 (dist_diff_0 \u003d dist - dist0\nand dist_diff_low_0 \u003d dist_low - dist0).\n\nSince dist0 * 128 cancels out on both sides of the inequality:\n  rd_low \u003c rd\n  \u003d\u003e RDCOST(rdmult, rate_low, dist_low) \u003c RDCOST(rdmult, rate, dist)\n  \u003d\u003e RDCOST(rdmult, rate_low, dist_diff_low_0) \u003c\n     RDCOST(rdmult, rate, dist_diff_0)\n\nWe can compute the exact same RD cost comparison with perfect\nbit-exactness using the standard RDCOST macro, while completely\nskipping the computation of dist0 when qmatrix is NULL.\n\nSpeed up:\n\nspeed 1: 0.096%\nspeed 2: 0.118%\nspeed 3: 0.115%\nspeed 4: 0.025%\nspeed 5: 0.022%\n\nChange-Id: I360628cfcde67709255bd53a37396cca252ac069\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "fb0ef12860a1124d4df15a4496351a843d3b49bb",
      "old_mode": 33188,
      "old_path": "av1/encoder/txb_rdopt.c",
      "new_id": "14c903eba1a53827fcd019cd4a0b70454c8c2c80",
      "new_mode": 33188,
      "new_path": "av1/encoder/txb_rdopt.c"
    }
  ]
}