)]}'
{
  "commit": "ab810952eb7c061e37054ddd1dfe0aa033365131",
  "tree": "2c6d25511afd3baed2f9d5f4d397d8f05aab8cd2",
  "parents": [
    "ab1008255e37b5b51a433ed69e04c06300543799"
  ],
  "author": {
    "name": "Martin Jambor",
    "email": "mjambor@suse.cz",
    "time": "Wed Oct 27 14:49:02 2021 +0200"
  },
  "committer": {
    "name": "Martin Jambor",
    "email": "mjambor@suse.cz",
    "time": "Wed Oct 27 15:12:05 2021 +0200"
  },
  "message": "ipa-cp: Use profile counters (or not) based on local availability\n\nThis is a follow-up small patch to address Honza\u0027s review of my\nprevious patch to select saner profile count to base heuristics on.\nCurrently the IPA-CP heuristics switch to PGO-mode only if there are\nPGO counters available for any part of the call graph.  This change\nmakes it to switch to the PGO mode only if any of the incoming edges\nbringing in the constant in question had any ipa-quality counts on\nthem.  Consequently, if a part of the program is built with\n-fprofile-use and another part without, IPA-CP will use\nestimated-frequency-based heuristics for the latter.\n\nI still wonder whether this should only happen with\nflag_profile_partial_training on.  It seems like we\u0027re behaving as if\nit was always on.\n\ngcc/ChangeLog:\n\n2021-10-18  Martin Jambor  \u003cmjambor@suse.cz\u003e\n\n\t* ipa-cp.c (good_cloning_opportunity_p): Decide whether to use\n\tprofile feedback depending on their local availability.\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "4d07a6d0a5861f0205a58d5a3f3a7c01233cf0cd",
      "old_mode": 33188,
      "old_path": "gcc/ipa-cp.c",
      "new_id": "703541d15cc275244ca6418efa3c88f7ae53c84c",
      "new_mode": 33188,
      "new_path": "gcc/ipa-cp.c"
    }
  ]
}