)]}'
{
  "commit": "80264aff67d041e1a4ca7f4fd123079ff99c975e",
  "tree": "52ceba95631c3081dacb0785ac182ab9ea015467",
  "parents": [
    "d5dd40709cd47d4227ab10ac5c3c80c8375f25ae"
  ],
  "author": {
    "name": "Connal de Souza",
    "email": "connaldesouza@google.com",
    "time": "Mon Nov 16 22:18:29 2026 -0800"
  },
  "committer": {
    "name": "Copybara-Service",
    "email": "copybara-worker@google.com",
    "time": "Mon Nov 16 22:20:16 2026 -0800"
  },
  "message": "Copy short heads and tails inline in CRC32C copies.\n\nIn `ExtendWithPclmul`, copying the entire tail after the multi-stream loop with\n`std::memcpy` wrote remaining full 64-byte cache lines with temporal stores even\nin non-temporal mode. In addition, the head alignment copy and the tail copy\nused variable-length `std::memcpy` calls, which compile to out-of-line calls\nthat measurably slow down shorter copies.\n\nThis improves performance slightly (1-5%) in most configurations, and significantly (10-15%) in a few where the lower register pressure from being a leaf function saves a few stack spills and fills.\n\nPiperOrigin-RevId: 1023406402\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "3a2a2d3a7514ea7422c297724ac68c1eed60cbde",
      "old_mode": 33188,
      "old_path": "absl/crc/internal/crc_x86_arm_combined.cc",
      "new_id": "99df29d99287b6a498db615221015c70f7062d63",
      "new_mode": 33188,
      "new_path": "absl/crc/internal/crc_x86_arm_combined.cc"
    }
  ]
}
