On Tue, 30 Jun 2026 08:30:01 GMT, Tobias Hartmann <[email protected]> wrote:
>> The `PushInlineTypeDown` implementation from >> [JDK-8302217](https://bugs.openjdk.org/browse/JDK-8302217) rewires `PhiNode` >> uses to a new `InlineTypeNode` but only adds direct users to the IGVN >> worklist. Nodes whose transformations depend on deeper input type changes >> might not be re-enqueued. >> >> In the failing case, the offending shape looks like this: >> ```CmpL(OrL(CastP2X(phi_or_inline_type), CastP2X(other)), 0L)``` >> >> After the Phi input is replaced by a scalarized inline type or a not-null >> value, >> [CmpLNode::Ideal](https://github.com/openjdk/valhalla/blob/a0dc49c20ee6d20e925a107b0d3dabc7e0ba8863/src/hotspot/share/opto/subnode.cpp#L899-L922) >> can fold the corresponding `CastP2X` to the null marker or `1L`. Without >> notifying users through the normal IGVN replacement path, >> `VerifyIterativeGVN` will report a missed `CmpL` idealization. >> >> The fix makes `replace_in_uses` notify users before rewiring, and uses it >> from `PushInlineTypeDown`. >> >> Thanks, >> Tobias >> >> --------- >> - [x] I confirm that I make this contribution in accordance with the >> [OpenJDK Interim AI Policy](https://openjdk.org/legal/ai). > > Tobias Hartmann has updated the pull request incrementally with one > additional commit since the last revision: > > Update > test/hotspot/jtreg/compiler/valhalla/inlinetypes/TestPushThroughPhiNotification.java > > Co-authored-by: Christian Hagedorn <[email protected]> src/hotspot/share/opto/phaseX.cpp line 2493: > 2491: for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) { > 2492: Node* u = n->fast_out(i); > 2493: if (u != n) { I see that `replace_edge()` will bail out in the `u == n` case, but we still call `rehash_node_delayed` and delete the hash which is probably not be necessary. But I guess that does not hurt much. ------------- PR Review Comment: https://git.openjdk.org/valhalla/pull/2593#discussion_r3497254736
