On Mon, 18 May 2026 15:15:00 GMT, Marc Chevalier <[email protected]> wrote:
> [C2: Deep recursion with > cmovP_regNode::bottom_type](https://bugs.openjdk.org/browse/JDK-8379667) > removed > > bool InstructForm::captures_bottom_type(FormDict &globals) const > > But in Valhalla, we had a special case on top for `CastI2N` > > https://github.com/openjdk/valhalla/blob/ef03842c0d869c7a88407a41102dacaa6a1707fa/src/hotspot/share/adlc/formssel.cpp#L781 > > which was removed in the merge https://github.com/openjdk/valhalla/pull/2414 > (see > https://github.com/openjdk/valhalla/pull/2414/changes#diff-6faeae52c6c7f235b9f763fb9b018d072e61133a1d4a56fee82fc42311f2c4f3L781) > > When the changed was backed out by [[BACKOUT] C2: Deep recursion with > cmovP_regNode::bottom_type](https://bugs.openjdk.org/browse/JDK-8384281) and > integrated in Valhalla by https://github.com/openjdk/valhalla/pull/2440 this > special case wasn't restored (see > https://github.com/openjdk/valhalla/pull/2440/changes#diff-6faeae52c6c7f235b9f763fb9b018d072e61133a1d4a56fee82fc42311f2c4f3R772-R812). > > The result is that CastI2N lose their types during matching, and then, a load > from such a pointer can't be identified to be strict final (when loading a > field from a value object), which causes the addition of wrong precedence > edges in GCM, making the graph too cyclic. > > Thanks, > Marc > > --------- > - [x] I confirm that I make this contribution in accordance with the [OpenJDK > Interim AI Policy](https://openjdk.org/legal/ai). Looks good to me, too. ------------- Marked as reviewed by qamai (Committer). PR Review: https://git.openjdk.org/valhalla/pull/2449#pullrequestreview-4316466542
