)]}'
{
  "commit": "065f9d377a532e514cd5027d9e12bcb8281758a4",
  "tree": "86584e73a83f1d82985ac5d6bea838a2c365b8f7",
  "parents": [
    "96958d1ecc073f7deb84c99e675437a7a6457566"
  ],
  "author": {
    "name": "Gerda Zsejke More",
    "email": "gerdazsejke.more@arm.com",
    "time": "Fri May 31 12:26:26 2024 +0200"
  },
  "committer": {
    "name": "James Zern",
    "email": "jzern@google.com",
    "time": "Tue Jun 04 23:21:06 2024 +0000"
  },
  "message": "Add Neon I8MM horiz 2x1 scale spec. impl for convolve_2d_scale\n\nAV1 has a limit on the scale ratio, specifically, the reference\nresolution cannot be more than 2 times the source resolution in any\ndimension. Given that the algorithm uses higher precision\n(1/1024-pel) for the step size (chapter 7.11.3.4. [1]), the\nhorizontal scaling function can be easily optimised for this specific\ncase. The indices of the source pixel to be interpolated are\ncalculated using the (subpel_qn + x * step) \u003e\u003e 1024 equation, which\ncan be simplified if step is a multiple of 1024.\n\nAdd implementation that specialises on x_step_qn equals to 2048, that\ngives an uplift of around 28% when a 2x1 scaling is applied.\n\n[1]https://aomediacodec.github.io/av1-spec/av1-spec.pdf\n\nChange-Id: I87b0730d0a75c534813f154555cbdb473b445438\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "ab215ca637160e5be5caae5762bffdc516a5c4c0",
      "old_mode": 33188,
      "old_path": "av1/common/arm/av1_convolve_scale_neon_i8mm.c",
      "new_id": "42de38f461bb0859d335d0f7db29f6feb4c75986",
      "new_mode": 33188,
      "new_path": "av1/common/arm/av1_convolve_scale_neon_i8mm.c"
    }
  ]
}