When merging in [JDK-8373343](https://bugs.openjdk.org/browse/JDK-8373343), we adapted the code to handle an additional verification case for Valhalla specific `AddP` shapes (`case #10`). It was wrongly assumed that when a `Phi` is present, it must be this new case. However, `case #3` could also contain a phi as seen in `is_captured_store_address()`:
https://github.com/openjdk/valhalla/blob/ae263e906af1e94bd07c6786c190816e9c104e17/src/hotspot/share/opto/escape.cpp#L3940-L3958 I propose to revert back to the original assert and just additionally add a check for `case #10`. Thanks, Christian ------------- Commit messages: - 8378292: [lworld] C2 fails assert(in1->uncast()->Opcode() == Op_LoadKlass && in2->uncast()->Opcode() == Op_LoadKlass) failed: must be LoadKlass Changes: https://git.openjdk.org/valhalla/pull/2144/files Webrev: https://webrevs.openjdk.org/?repo=valhalla&pr=2144&range=00 Issue: https://bugs.openjdk.org/browse/JDK-8378292 Stats: 35 lines in 2 files changed: 16 ins; 11 del; 8 mod Patch: https://git.openjdk.org/valhalla/pull/2144.diff Fetch: git fetch https://git.openjdk.org/valhalla.git pull/2144/head:pull/2144 PR: https://git.openjdk.org/valhalla/pull/2144
