jdk_vector passed but then failed

[pshipton]

https://openj9-jenkins.osuosl.org/job/Test_openjdk23_j9_sanity.openjdk_aarch64_mac_Nightly_testList_0/121 https://openj9-jenkins.osuosl.org/job/Test_openjdk23_j9_sanity.openjdk_ppc64le_linux_Nightly_testList_0/79 jdk_vector_byte64_j9_0

22:16:11  Test results: passed: 1
22:16:11  Report written to /Users/jenkins/workspace/Test_openjdk23_j9_sanity.openjdk_aarch64_mac_Nightly_testList_0/aqa-tests/TKG/output_17375144638283/jdk_vector_byte64_j9_0/report/html/report.html
22:16:11  Results written to /Users/jenkins/workspace/Test_openjdk23_j9_sanity.openjdk_aarch64_mac_Nightly_testList_0/aqa-tests/TKG/output_17375144638283/jdk_vector_byte64_j9_0/work
22:16:11  grep #VECTOR API /Users/jenkins/workspace/Test_openjdk23_j9_sanity.openjdk_aarch64_mac_Nightly_testList_0/aqa-tests/TKG/../TKG/output_17375144638283/jdk_vector_byte64_j9_0/work/jdk/incubator/vector/Byte64VectorTests.jtr
22:16:11  -----------------------------------
22:16:11  jdk_vector_byte64_j9_0_FAILED

There are probably more failures in the builds, but they haven't finished and I won't be in on Wed to check them.

Changes https://github.com/eclipse-openj9/openj9/compare/4f87f3f4402...351d40a670b https://github.com/eclipse-openj9/openj9-omr/compare/6bfb20f781c...1c2b08d3399 https://github.com/eclipse-openj9/openj9/pull/20972

@gita-omr @knn-k

[pshipton]

See all the failures in https://openj9-jenkins.osuosl.org/job/Pipeline-Build-Test-JDK21/473/, failing on every platform. Example: https://openj9-jenkins.osuosl.org/job/Test_openjdk21_j9_sanity.openjdk_x86-64_linux_Nightly_testList_0/298 jdk_vector_double128_j9_0 jdk_vector_byte64_j9_0 https://openj9-jenkins.osuosl.org/job/Test_openjdk21_j9_sanity.openjdk_x86-64_linux_Nightly_testList_1/298 jdk_vector_short128_j9_0 jdk_vector_long128_j9_0 https://openj9-jenkins.osuosl.org/job/Test_openjdk21_j9_sanity.openjdk_x86-64_linux_Nightly_testList_2/298 jdk_vector_float128_j9_0 jdk_vector_int128_j9_0 jdk_vector_byte128_j9_0

This needs to be fixed or the problem change reverted asap. We can't handle triage of the builds with all these failures.

@hzongaro fyi

[knn-k]

Ah, I only ran "test sanity.functional" in PR #20972, and some tests in openjdk are failing.

@gita-omr Would you open a PR for reverting #20972, please?

[pshipton]

I used the revert button in the PR. Reverted via https://github.com/eclipse-openj9/openj9/pull/20996

[pshipton]

Ran a test build on jdk21 amac, and the test didn't fail. https://openj9-jenkins.osuosl.org/job/Pipeline-Build-Test-Personal/629/

[gita-omr]

Sorry, forgot to run these tests! Hopefully, this did not cause too much trouble.

[jdmpapin]

This has been happening recently in JITServer PR builds. In particular, I've seen it in #21992 and #21998. The former is merged, but it is (as I anticipated) unrelated. I'm able to reproduce the problem locally using an older tip of master (commits listed by -version in the abridged output):

$ make _jdk_vector_double128_j9_0 TEST_FLAG=JITAAS
...
JITServer is ready to accept incoming requests
openjdk version "21.0.8-internal" 2025-07-15
OpenJDK Runtime Environment (build 21.0.8-internal-adhoc.jdmpapin.openj9-openjdk-jdk21)
Eclipse OpenJ9 VM (build master-7848c4459a, JRE 21 Linux amd64-64-Bit Compressed References 20250611_000000 (JIT enabled, AOT enabled)
OpenJ9   - 7848c4459a
OMR      - 2e58de247
JCL      - 48ce4d6293b based on jdk-21.0.8+5)
...
Test results: passed: 1
Report written to /[...]/aqa-tests/TKG/output_17496591985815/jdk_vector_double128_j9_0/report/html/report.html
Results written to /[...]/aqa-tests/TKG/output_17496591985815/jdk_vector_double128_j9_0/work
grep #VECTOR API /[...]/aqa-tests/TKG/../TKG/output_17496591985815/jdk_vector_double128_j9_0/work/jdk/incubator/vector/Double128VectorTests.jtr
-----------------------------------
jdk_vector_double128_j9_0_FAILED
-----------------------------------

This looks like a failure to vectorize, or perhaps a failure to even run the vector API expansion pass. Along with this I've also seen timeouts in the same tests, but I think they're related. A timeout could be the result of the test running much more slowly than anticipated as a result of a failure to vectorize.