RFR: 8340183: Shenandoah: Incorrect match for clone barrier in is_gc_barrier_node
Aleksey Shipilev
shade at openjdk.org
Mon Sep 16 10:40:16 UTC 2024
The name of the call we emit is "shenandoah_clone":
https://github.com/openjdk/jdk/blob/545951889c1ea68646be600decaf2bf4c049600b/src/hotspot/share/gc/shenandoah/c2/shenandoahBarrierSetC2.cpp#L806
...yet we test for "shenandoah_clone_barrier" here:
https://github.com/openjdk/jdk/blob/545951889c1ea68646be600decaf2bf4c049600b/src/hotspot/share/gc/shenandoah/c2/shenandoahBarrierSetC2.cpp#L688
I think we are better off polling the call target instead of relying on call name. This change also eliminates `shenandoah_cas_obj` matcher, for which we do not have the emitted call ever since we started doing CAS expansions inline.
Additional testing:
- [x] Linux x86_64 server fastdebug, `hotspot_gc_shenandoah`
- [ ] Linux x86_64 server fastdebug, `all` with `-XX:+UseShenandoahGC`
-------------
Commit messages:
- Fix
Changes: https://git.openjdk.org/jdk/pull/21014/files
Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=21014&range=00
Issue: https://bugs.openjdk.org/browse/JDK-8340183
Stats: 20 lines in 3 files changed: 6 ins; 10 del; 4 mod
Patch: https://git.openjdk.org/jdk/pull/21014.diff
Fetch: git fetch https://git.openjdk.org/jdk.git pull/21014/head:pull/21014
PR: https://git.openjdk.org/jdk/pull/21014
More information about the hotspot-gc-dev
mailing list