Extend arraycmp length child

[Spencer-Comin]

Change all uses of arraycmp to take a 64 bit length child instead of 32 bits.

Relies on eclipse-omr/omr#7313

[hzongaro]

In addition to the review comments, I noticed there is code in CICSTransform2ArrayCmpIfs that still seems to be using a 32-bit length for an arraycmplen.

[Spencer-Comin]

Ran some internal tests. I saw a couple failures, but the failures look unrelated to me.

JDK21 tests:

https://hyc-runtimes-jenkins.swg-devops.com/job/Pipeline-Build-Test-Personal/22751/
https://hyc-runtimes-jenkins.swg-devops.com/job/Pipeline-Build-Test-Personal/22779/ to account for infra failures

x86-64_linux sanity.functional failure:

• cmdLineTester_criu_nonPortableRestore_6 - looks like cmdLineTester_criu_nonPortableRestore testGetProcessRestoreStartTime getProcessRestoreStartTime is more than getLastRestoreTime #19050

s390x_linux sanity.system failure:

• DaaLoadTest_daa1_CS_5m_0 - looks like DaaLoadTest_daa1 BigInteger would overflow supported range #19666

JDK8 vmfarm tests:

http://vmfarm.rtp.raleigh.ibm.com/build_info.php?build_id=72852

• LoZ failing gdb test - illegal instruction, looks like it's caused by a wild branch. This shouldn't be an arraycmp failure mode.
• Windows failing jtreg tests - cannot find file, possibly an infra issue? Also not likely to be an arraycmp failure mode.

[Spencer-Comin]

As a note, this PR must be merged in coordination with reverting the 64-bit arraycmp revert in OMR: eclipse-omr/omr#7380

[hzongaro]

@Spencer-Comin, may I ask you to open the new PR to reapply eclipse-omr/omr#7313?

[hzongaro]

Looks good. Thanks!

[Spencer-Comin]

@Spencer-Comin, may I ask you to open the new PR to reapply eclipse/omr#7313?

Done! eclipse-omr/omr#7388

[hzongaro]

Jenkins test sanity.functional,sanity.openjdk all jdk8,jdk11,jdk17,jdk21 depends eclipse-omr/omr#7388

[hzongaro]

xLinux failures appear to be infrastructure-related. Rerunning. The runs that are listed as "pending" appear to have actually aborted. I'll rerun those as well.

The JDK 21 sanity.openjdk s390x Linux failures appear to be due to #18908. That issue had been closed recently as it couldn't be reproduced any longer.

Jenkins test sanity.functional,sanity.openjdk xlinunux jdk8,jdk11,jdk21 depends eclipse-omr/omr#7388

[hzongaro]

Fixing the typo:

Jenkins test sanity.functional,sanity.openjdk xlinux jdk8,jdk11,jdk21 depends eclipse-omr/omr#7388

[hzongaro]

Jenkins test sanity.openjdk xlinux jdk17 depends eclipse-omr/omr#7388

[hzongaro]

Jenkins test sanity.functional,sanity.openjdk win jdk8,jdk11,jdk17 depends eclipse-omr/omr#7388

[hzongaro]

Jenkins test sanity.openjdk win jdk21 depends eclipse-omr/omr#7388

[hzongaro]

Jenkins test sanity.functional,sanity.openjdk aix jdk21 depends eclipse-omr/omr#7388

[hzongaro]

There are some build failures in JDK 21 builds for aix, xLinux and x86-64 Windows. Two of them appear to be happening during Value Propagation - I haven't looked into the third. I looked at a jitdump for the Windows failure, and there didn't appear to be any arraycmp operations in the method that was being compiled. However, we don't seem to have seen any other recent build failures similar to this.

@Spencer-Comin, may I ask you to take a look at these build failures to determine whether they could be in any way related to your pull requests?

[hzongaro]

@Spencer-Comin, it looks like OMR acceptance builds are hitting this same problem. For instance, see https://openj9-jenkins.osuosl.org/view/Build_OMR/job/Build_JDK21_ppc64_aix_OMR/140/consoleFull

[hzongaro]

Looking at some of the other failures, there are some infrastructure-related problems, but others appear to be similar to the failures we saw JDK 21 testing. Once those problems have been resolved, I'll rerun the testing.

[Spencer-Comin]

I looked into the windows failures and reran the jdk11 sanity.functional (https://hyc-runtimes-jenkins.swg-devops.com/job/Pipeline-Build-Test-Personal/22949/) (branches rebased to more recent openj9/omr). I also ran a control test without my changes (https://hyc-runtimes-jenkins.swg-devops.com/job/Pipeline-Build-Test-Personal/22950/). The failing tests are the same between my changes and the control.

[hzongaro]

I looked into the windows failures and reran the jdk11 sanity.functional

Right - thanks for looking at that. I should have clarified earlier that because we're seeing similar failures with the OMR Acceptance builds, the failures are almost certainly unrelated to your changes.

Those failures are still under investigation. Once they're resolved, I'll rerun testing on your changes. In the meanwhile, those failures are creating too much noise to feel entirely safe about merging your changes.

[hzongaro]

Most of the failures we were seeing were caused by unrelated changes and should now be resolved. Rerunning testing, as this will require a coordinate merge.

Jenkins test sanity.functional,sanity.openjdk all jdk8,jdk11,jdk17,jdk21 depends eclipse-omr/omr#7388

[hzongaro]

Restarting AIX test runs that were aborted:

Jenkins test sanity.functional,sanity.openjdk aix jdk8,jdk11 depends eclipse-omr/omr#7388

[hzongaro]

Jenkins test sanity.openjdk aix jdk17 depends eclipse-omr/omr#7388

[hzongaro]

JDK 8 x86-64 macOS sanity.functional test failure looks like an infrastructure problem. Rerunning:

Jenkins test sanity.functional xmac jdk8 depends eclipse-omr/omr#7388

[hzongaro]

JDK 17 sanity.functional aix testing looks like it was aborted. Rerunning.

Looking through the failures on other platforms, they all appear to be due to known issues:

• JDK 11 sanity.openjdk Windows failure appears to be due to known issue jdk_util_j9_0 RuntimeException: Unexpected bundle: java.util.PropertyResourceBundle@16d73892 #9124
• JDK 17 sanity.openjdk aarch64 Linux failure looks like an instance of known issue OpenJDK NullPointerException: Cannot invoke "java.util.concurrent.atomic.AtomicLong.getAndAdd(long)" because "java.util.concurrent.ThreadLocalRandom.seeder" is null #16609
• JDK 8 sanity.functional x86-64 Linux failure looks like an instance of known issue testJitserverArguments, Test SSL success condition - failed to configure SSL ecdh #18599
• JDK 8 sanity.openjdk x86-64 Linux failure looks like an instance of known issue OpenJDK java/security/KeyRep/Serial - SecurityException: PBE test failed #17746

Jenkins test sanity.functional aix jdk17 depends eclipse-omr/omr#7388

[pshipton]

There are only two AIX machines left functioning atm. I removed AIX from nightly builds, but not the weekend testing. You might be able to get a PR build through once the weekend testing finishes or times out.

https://github.ibm.com/runtimes/infrastructure/issues/9553

[hzongaro]

Despite the problem running the AIX JDK 17 sanity.functional testing, I think this change and the change in OMR on which it depends have been well exercised. Merging.