[Openmp-commits] [PATCH] D107121: [OpenMP] Fix performance regression reported in bug #51235

[Shilei Tian via Phabricator via Openmp-commits]

tianshilei1992 added a comment.

I didn't see an issue regarding the child task counter.
Increment:

  if (flags->proxy == TASK_PROXY || flags->detachable == TASK_DETACHABLE ||
        flags->hidden_helper ||
        !(taskdata->td_flags.team_serial || taskdata->td_flags.tasking_ser)) {
      KMP_ATOMIC_INC(&parent_task->td_incomplete_child_tasks);
  ...

Decrement:

  if (!(taskdata->td_flags.team_serial || taskdata->td_flags.tasking_ser) ||
          taskdata->td_flags.detachable == TASK_DETACHABLE ||
          taskdata->td_flags.hidden_helper) {
        __kmp_release_deps(gtid, taskdata);
        // Predecrement simulated by "- 1" calculation
  #if KMP_DEBUG
        children = -1 +
  #endif
            KMP_ATOMIC_DEC(&taskdata->td_parent->td_incomplete_child_tasks);
  ...

Another increment of the counter is in `__kmp_task_dup_alloc`, which is for task loop, that is unrelated here. No other increment/decrement.
I did check the counter for the case `gtid`, and after the three tasks are created, the counter is 3, as expected.

https://bugs.llvm.org/show_bug.cgi?id=49816 also reports that one task is missing. So I'm thinking the issue is not introduced by this patch. Can you help double check if w/o the patch you can still observe the failure? If it is not introduced by this patch, I feel we could first get this patch in and then see if we can reproduce the issue in another way.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D107121/new/

https://reviews.llvm.org/D107121