)]}'
{
  "commit": "c501b07be392027178f0a522bcc14836ea8e6f06",
  "tree": "76fc19c7d4610aa3deefd911520abe66bb1b237e",
  "parents": [
    "2e87bb93226ff97be167f617abf91db7e43153d8"
  ],
  "author": {
    "name": "Satheesh Kumar",
    "email": "satheesh.kumar@ittiam.com",
    "time": "Wed Mar 11 19:54:28 2026 +0530"
  },
  "committer": {
    "name": "Yunqing Wang",
    "email": "yunqingwang@google.com",
    "time": "Mon Mar 16 14:08:18 2026 -0700"
  },
  "message": "Improve av1_convolve_y_sr_general_avx2()\n\nThis patch introduces the following optimizations:\n- Added a specialized width wise handling for 2-tap filtering, with\n  separate handling for subpel_y_qn \u003d 8\n- Improved the existing 4/6/8-tap filtering by optimizations based on\n  width\n\nThe scaling of av1_convolve_y_sr_avx2() wrt existing implementation is\nprovided below:\n\nBlockSize  2tap   4tap   6tap   8tap\n2x2        1.07   0.95    -      -\n4x4        1.00   0.96    -      -\n8x8        0.91   0.91   0.98   0.94\n16x16      0.96   0.97   0.95   1.03\n32x32      0.98   0.99   1.44   1.67\n64x64      1.00   0.98   1.45   1.71\n128x128    0.98   0.98   1.48   1.71\n\nChange-Id: Ie2fb8a1a26652d748351a6b3bd6edb5d4513c0bd\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "76a6ca0c5515d9a0df2cfd61d605e9ca28c5a85f",
      "old_mode": 33188,
      "old_path": "aom_dsp/x86/convolve_avx2.h",
      "new_id": "ffafa13319e067e52b83f9686cbb59eb63344093",
      "new_mode": 33188,
      "new_path": "aom_dsp/x86/convolve_avx2.h"
    },
    {
      "type": "modify",
      "old_id": "005d666395859a989175cf19b43f4e30e70a99bb",
      "old_mode": 33188,
      "old_path": "av1/common/x86/convolve_avx2.c",
      "new_id": "209dee75e5e66b90a9c845c2cb72c0d09a5e1aee",
      "new_mode": 33188,
      "new_path": "av1/common/x86/convolve_avx2.c"
    }
  ]
}