)]}'
{
  "commit": "834517d37c02a12b97dfccd911db82155fcce6ed",
  "tree": "850779f696c2b96286f1da781e7e788599961f45",
  "parents": [
    "5da1715de0469934aa952eae4723ce9de2bbb61a"
  ],
  "author": {
    "name": "Radu Costas",
    "email": "radu.costas@imgtec.com",
    "time": "Thu Aug 27 10:31:33 2026 +0300"
  },
  "committer": {
    "name": "Marge Bot",
    "email": "marge-bot@fdo.invalid",
    "time": "Tue Sep 08 09:26:06 2026 +0000"
  },
  "message": "nir: Skip fminmax preserve_sz lowering if non-zero consts are used\n\nPreviously, in the case where the lower_fminmax_preserve sz was enabled,\nfor an fmin/fmax with the preserve_sz bit enabled where one of the arguments\nwas a non-zero const, the lowering would simply remove the preserve_sz bit\nfrom the instruction and do nothing else.\nThis can cause further nir passes optimizing this instruction to\noptimize without taking that flag into account, eventually leaving the\nsign of zero undefined when one of the operands is +/-0.0\n\nSigned-off-by: Radu Costas \u003cradu.costas@imgtec.com\u003e\nReviewed-by: Georg Lehmann \u003cdadschoorse@gmail.com\u003e\nPart-of: \u003chttps://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/44033\u003e\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "7cad1e7f1a240d8fe92189b21c6849842b01d16e",
      "old_mode": 33188,
      "old_path": "src/compiler/nir/nir_lower_alu.c",
      "new_id": "8f5fd5a55fe16101ec3024e880c037bdde3382df",
      "new_mode": 33188,
      "new_path": "src/compiler/nir/nir_lower_alu.c"
    }
  ]
}