Fold Invokers.checkVarHandleGenericType using const refs by matthewhall2 · Pull Request #23639 · eclipse-openj9/openj9

matthewhall2 · 2026-04-02T14:50:33Z

Modifies getMethodHandleTableEntryIndex to use the asTypeCache when there is no exact MethodType match and the cache is non-null.
folds the call to a const ref load when the KOI is not UNKNOWN

nbhuiyan

Other than the few minor comments, LGTM.

matthewhall2 · 2026-05-12T20:22:29Z

@0xdaryl can you run the tests with eclipse-omr/omr#8209 ?

0xdaryl · 2026-05-12T21:04:02Z

Jenkins test sanity all jdk21 depends eclipse-omr/omr#8209

matthewhall2 · 2026-05-19T19:59:48Z

windows sanity.functional fails look related to #22758 (comment)

windows sanity.openjdk fail was caused by the assertion fail "orphaned const ref", I've pushed the fix for that
@0xdaryl can you run the tests again please?

0xdaryl · 2026-05-19T20:24:38Z

Jenkins test sanity all jdk21 depends eclipse-omr/omr#8209

0xdaryl · 2026-05-20T17:08:11Z

You may need to rebase this and we try again.

matthewhall2 · 2026-05-21T13:51:31Z

rebased

0xdaryl · 2026-05-21T13:54:02Z

Jenkins test sanity xlinux,plinux,alinux,zlinux jdk21 depends eclipse-omr/omr#8209

Starting small, with just the Linux platforms first.

matthewhall2 · 2026-06-02T19:06:29Z

Z failure is from the vector API tests, which Ehsan confirmed was fixed. I did rebase OMR but I guess I need to again.
P failure looks unrelated (ACTION: testng -- Error. Agent communication error: java.net.SocketException: Broken pipe) - hopefully is won't fail after the rebase

matthewhall2 · 2026-06-10T21:43:02Z

@mpirvu or @nbhuiyan can you run the tests again please?

mpirvu · 2026-06-11T14:28:36Z

Jenkins test sanity xlinux,plinux,alinux,zlinux jdk21 depends eclipse-omr/omr#8209

nbhuiyan · 2026-06-11T15:46:40Z

@matthewhall2 I had a few comments in my approval (approved with comments instead of requesting changes as it was just prior to me going on vacation) that have not been addressed yet.

nbhuiyan

Please review the last set of comments.

matthewhall2 · 2026-06-11T17:36:17Z

thanks @nbhuiyan , I somehow missed these. I've now addressed the comments

mpirvu · 2026-06-15T02:24:35Z

Jenkins test sanity xlinux,plinux,alinux,zlinux jdk21 depends eclipse-omr/omr#8209

matthewhall2 · 2026-06-15T15:18:48Z

aarch64 testList_3 fail is #17270
aarch64 failure testList_2 is unclear atm, but there are many issues open with the same "Agent communication error: java.net.SocketException: Broken pipe"

plinux sanity.functional and sanity.openjdk are #24139

mpirvu · 2026-06-16T18:25:35Z

zlinux failures
jdk_collections_0

11:01:11  Assertion failed at /home/jenkins/workspace/Build_JDK21_s390x_linux_Personal/openj9/runtime/compiler/codegen/J9CodeGenerator.cpp:5270: false
11:01:11  VMState: 0x0005ff09
11:01:11  	orphaned const refs: obj3
11:01:11  	the const provenance graph is missing at least one edge

I have seen this when doing perf experiments with constrefs enabled on x86-64.

jdk_int128vector_j9_0, jdk_byte128vector_j9_0, jdk_long128vector_j9_0, jdk_short128vector_j9_0 fail without a clear error message. We have seen this type of failure before: #24071

11:12:57  #VECTOR API: Vectorized using vstoreiVector128Int32 in Int128VectorTests.withInt128VectorTests(Ljava/util/function/IntFunction;)V at scorching 
11:12:57  java.lang.Exception: config failures: 0, test failures: 1
11:12:57  	at com.sun.javatest.regtest.agent.TestNGRunner.main(TestNGRunner.java:111)
11:12:57  	at com.sun.javatest.regtest.agent.TestNGRunner.main(TestNGRunner.java:63)
11:12:57  	at java.base/jdk.internal.reflect.DirectMethodHandleAccessor.invoke(DirectMethodHandleAccessor.java:103)
11:12:57  	at java.base/java.lang.reflect.Method.invoke(Method.java:586)
11:12:57  	at com.sun.javatest.regtest.agent.MainWrapper$MainTask.run(MainWrapper.java:138)
11:12:57  	at java.base/java.lang.Thread.run(Thread.java:1600)

mpirvu · 2026-06-16T18:27:17Z

Jenkins test sanity win,aix,xmac,amac jdk21 depends eclipse-omr/omr#8209

matthewhall2 · 2026-06-16T18:32:22Z

@mpirvu

I have seen this when doing perf experiments with constrefs enabled on x86-64

have you seen this without my changes? I'm currently running internal tests with c40c05b cherry-picked to see if that fixes it, as I don't think it's the changes in this PR

mpirvu · 2026-06-16T18:56:51Z

have you seen this without my changes? I'm currently running internal tests with c40c05b cherry-picked to see if that fixes it, as I don't think it's the changes in this PR

@nbhuiyan Testing from this PR hits the same assert I was hitting when running AcmeAir/Daytrader with your prototype. Did that prototype include this change?

nbhuiyan · 2026-06-16T19:15:40Z

Testing from this PR hits the same assert I was hitting when running AcmeAir/Daytrader with your prototype. Did that prototype include this change?

The prototype I shared with you included the change in eclipse-omr/omr#8209, which then results in the assertion failure due to missing CPG edges. That needs to be addressed first so that the tests pass. I am working on that.

- Modifies getMethodHandleTableEntryIndex to use the asTypeCache when there is no exact MethodType match and the cache is non-null. - folds the call to a const ref load when the KOI is not UNKNOWN Signed-off-by: Matthew Hall <matthew.hall3@outlook.com>

matthewhall2 · 2026-06-18T13:48:27Z

rebase was needed to fix the mac build errors

mpirvu · 2026-06-18T17:03:25Z

Jenkins test sanity all jdk21 depends eclipse-omr/omr#8209

mpirvu · 2026-06-19T14:46:41Z

windows had one SCC test fail due to not being able to destroy the SCC. This is a known issue.
openjdk on zlinux had the usual vectorAPI failures: jdk_int128vector_j9_0, jdk_byte128vector_j9_0, jdk_long128vector_j9_0, jdk_short128vector_j9_0:

18:22:07  java.lang.Exception: config failures: 0, test failures: 1
18:22:07  	at com.sun.javatest.regtest.agent.TestNGRunner.main(TestNGRunner.java:111)
18:22:07  	at com.sun.javatest.regtest.agent.TestNGRunner.main(TestNGRunner.java:63)
18:22:07  	at java.base/jdk.internal.reflect.DirectMethodHandleAccessor.invoke(DirectMethodHandleAccessor.java:103)
18:22:07  	at java.base/java.lang.reflect.Method.invoke(Method.java:586)
18:22:07  	at com.sun.javatest.regtest.agent.MainWrapper$MainTask.run(MainWrapper.java:138)
18:22:07  	at java.base/java.lang.Thread.run(Thread.java:1600)

mpirvu · 2026-06-19T14:47:16Z

Since all test failures are accounted for and all review comments have been addresses, this PR is ready to be merged.

matthewhall2 requested a review from dsouzai as a code owner April 2, 2026 14:50

matthewhall2 force-pushed the fold_vh_checkGenericType branch 3 times, most recently from 9eadcd7 to 6a72ea1 Compare April 2, 2026 16:44

matthewhall2 requested a review from nbhuiyan April 2, 2026 16:45

hzongaro added the comp:jit label Apr 8, 2026

hzongaro added this to Adopt OpenJDK MethodHandles Apr 8, 2026

github-project-automation Bot moved this to In progress in Adopt OpenJDK MethodHandles Apr 8, 2026

nbhuiyan requested changes Apr 10, 2026

View reviewed changes

Comment thread runtime/compiler/env/VMJ9.cpp Outdated

Comment thread runtime/compiler/env/VMJ9.cpp Outdated

Comment thread runtime/compiler/optimizer/MethodHandleTransformer.cpp Outdated

matthewhall2 force-pushed the fold_vh_checkGenericType branch from 6a72ea1 to 0b27535 Compare April 14, 2026 19:29

matthewhall2 requested a review from nbhuiyan April 14, 2026 19:31

nbhuiyan approved these changes Apr 15, 2026

View reviewed changes

Comment thread runtime/compiler/env/VMJ9.cpp

Comment thread runtime/compiler/optimizer/MethodHandleTransformer.cpp Outdated

Comment thread runtime/compiler/optimizer/MethodHandleTransformer.cpp

Comment thread runtime/compiler/optimizer/MethodHandleTransformer.cpp

matthewhall2 force-pushed the fold_vh_checkGenericType branch from 0b27535 to e09d7af Compare May 19, 2026 19:58

matthewhall2 force-pushed the fold_vh_checkGenericType branch from e09d7af to 6f0b6c4 Compare May 21, 2026 13:42

matthewhall2 force-pushed the fold_vh_checkGenericType branch 2 times, most recently from a114640 to f7b44cc Compare June 2, 2026 19:06

mpirvu removed the request for review from dsouzai June 4, 2026 17:44

mpirvu self-assigned this Jun 11, 2026

nbhuiyan requested changes Jun 11, 2026

View reviewed changes

matthewhall2 force-pushed the fold_vh_checkGenericType branch 2 times, most recently from 647f284 to 6ee1978 Compare June 11, 2026 17:29

matthewhall2 requested a review from nbhuiyan June 11, 2026 17:36

nbhuiyan approved these changes Jun 12, 2026

View reviewed changes

matthewhall2 force-pushed the fold_vh_checkGenericType branch from 6ee1978 to f5605da Compare June 18, 2026 13:47

mpirvu merged commit 21a124f into eclipse-openj9:master Jun 19, 2026
25 of 28 checks passed

github-project-automation Bot moved this from In progress to Done in Adopt OpenJDK MethodHandles Jun 19, 2026

Uh oh!

Conversation

matthewhall2 commented Apr 2, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

nbhuiyan left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

matthewhall2 commented May 12, 2026

Uh oh!

0xdaryl commented May 12, 2026

Uh oh!

matthewhall2 commented May 19, 2026

Uh oh!

0xdaryl commented May 19, 2026

Uh oh!

0xdaryl commented May 20, 2026

Uh oh!

matthewhall2 commented May 21, 2026

Uh oh!

0xdaryl commented May 21, 2026

Uh oh!

matthewhall2 commented Jun 2, 2026

Uh oh!

matthewhall2 commented Jun 10, 2026

Uh oh!

mpirvu commented Jun 11, 2026

Uh oh!

nbhuiyan commented Jun 11, 2026

Uh oh!

nbhuiyan left a comment

Choose a reason for hiding this comment

Uh oh!

matthewhall2 commented Jun 11, 2026

Uh oh!

mpirvu commented Jun 15, 2026

Uh oh!

matthewhall2 commented Jun 15, 2026

Uh oh!

mpirvu commented Jun 16, 2026

Uh oh!

mpirvu commented Jun 16, 2026

Uh oh!

matthewhall2 commented Jun 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mpirvu commented Jun 16, 2026

Uh oh!

nbhuiyan commented Jun 16, 2026

Uh oh!

matthewhall2 commented Jun 18, 2026

Uh oh!

mpirvu commented Jun 18, 2026

Uh oh!

mpirvu commented Jun 19, 2026

Uh oh!

mpirvu commented Jun 19, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

matthewhall2 commented Jun 16, 2026 •

edited

Loading