)]}'
{
  "commit": "3892a19d949dd19dee79f1172a65468939d1bd62",
  "tree": "81081da64768aa95d6a9b7ba17cb02cf3c5b54ed",
  "parents": [
    "0b37a983743b9c74b2b5a65e660c2edc3f670866"
  ],
  "author": {
    "name": "Lin Zheng",
    "email": "linzhen@google.com",
    "time": "Wed Sep 09 23:29:29 2026 +0000"
  },
  "committer": {
    "name": "Wan-Teh Chang",
    "email": "wtc@google.com",
    "time": "Fri Sep 11 07:35:39 2026 -0700"
  },
  "message": "Use correct VMAF Gussian blur filter in the unit test\n\nIn commit 8e7b6a567d, the VmafPreprocessing unit test introduced\nan 8-tap kGaussFilter with a positive tap 0, and NEON_I8MM and SVE2\ntriggered an underflow in the 8-tap horizontal i8mm kernel (which\nassumes negative tap 0).\n\nThe actual filter used by tune vmaf mode has an zero taps 0 and 7, so we\nupdate kGaussFilter in the VmafPreprocessing unit test to match the real\nfilter in tune_vmaf.c, which allows NEON_I8MM and SVE2 to take the\nvalid 6-tap convolution path and pass.\n\nAlso allocate 9 elements with 16-byte alignment matching gauss_filter\nin tune_vmaf.c to avoid global-buffer-overflow in prepare_coeffs_6t().\n\nBug: 558679217\nChange-Id: Ic00d31e5e56fbaee2854f819728358b77444e2b4\n(cherry picked from commit 91197c6f1c9eb50831113c92571c9688b865730e)\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "bbc80152c78f6e3ac014819cfd43832852ef0635",
      "old_mode": 33188,
      "old_path": "test/av1_convolve_test.cc",
      "new_id": "4b444bdff0866235a3b5bb41dfeedc6fe62b4746",
      "new_mode": 33188,
      "new_path": "test/av1_convolve_test.cc"
    }
  ]
}