On Thu, 6 Nov 2025 07:36:34 GMT, Manuel Hässig <[email protected]> wrote:
>> ## Problem Analysis >> >> The stress test `compiler/valhalla/inlinetypes/TestAcmpWithUnstableIf.java`, >> which uses the `StressUnstableIfTrap` flag, failed intermittently with the >> assert "no node with a side effect" during C2 compilation. I tracked down >> the origin of the failure to the raw store of the unstable if trap stress >> counter that was missing a memory edge to the backedge phi and thus had no >> side effect in the loop, which lead to the aforementioned assert. During >> parsing, the missing memory edge gets discarded with as vestigial parsing >> state when `do_if()` ends up `stopped()`. However, the effect of the stress >> counter should still be wired back into the backedge since it will be >> incremented in the next iteration. >> >> ## Patch Description >> >> To prevent the stress counters memory state being lost to >> `PreserveJVMState`, I pass it to the caller over an out-parameter and >> `set_memory` with that node, similar to what is already done for control. >> This is only necessary for the two last invocation of `do_if()` in >> `do_acmp()` where they are actually allowed to trap. Further, this PR >> reenables `StressUnstableIfTrap` in >> `compiler/valhalla/inlinetypes/TestAcmpWithUnstableIf.java`. >> >> ## Testing >> >> - [x] tier1,tier2,tier3 plus internal stress testing > > Manuel Hässig has updated the pull request incrementally with one additional > commit since the last revision: > > Review Christian @mhaessig Your change (at version 9e09aa23584438396e8a50b049693018620efd69) is now ready to be sponsored by a Committer. ------------- PR Comment: https://git.openjdk.org/valhalla/pull/1716#issuecomment-3496033789
