[assistant-librarian]

Updated all pinned TheRock workflow and source refs to 5c67705 due to submodule bump

Baseline run ID: 35596222200

[therock-pr-bot]

🤖 Bump PR detected (author @assistant-librarian[bot]). All policy checks are auto-approved for automated dependency bumps.

[tony-davis]

CI review: tracked / out-of-scope breakdown

Verdict: safe to merge. This bump adds no new CI red. There are zero rocm-libraries-scoped failures in this run, and the one real failure is a pre-existing rocm-systems-owned timeout that reproduces identically on the previous bump PR.

Note on merge state: mergeable: MERGEABLE, merge_state_status: BLOCKED. No rebase is needed (this does not conflict with develop); it's blocked only because the aggregate Multi-Arch CI Summary check is red off the back of the hip-tests job below, so it needs someone with override rights to merge.

Per-failure disposition (rocm-libraries scope only)

None. All rocm-libraries targets passed, including all 9 hipBLASLt, 16 MIOpen, 3 libhipcxx, and 2 hipkernelprovider checks.

rocm-systems-owned reds (visibility only, not filed, not ours)

Job	Target / Arch	Signature
Windows::release / Test hip-tests (ROCR) (shard 3/4)	hip-tests (gfx110X-all)	test-timeout: The action 'Test' has timed out after 120 minutes at test 487/1000

Detail, in case it's useful to the owning team: this is not a hard failure, the shard just doesn't finish. Six tests hang to their 600s+ CTest timeout (Unit_hipStreamWaitValue_Default - uint32_t at 656s, Unit_hipLaunchHostFunc_Diffpriority at 652s, Unit_hipSharedQueueAnyOrderOverlap_IntraStreamOrderAcrossActivation at 626s, Unit_hipExtStreamCreateWithCUMask_Functionality at 656s, Unit_hipEventElapsedTime_Verify_Capture at 624s, Unit_hipEventMGpuMThreads_1 at 624s), and separately every Unit_Layered*Texture / Unit_surf* case takes a flat ~78s.

The same job timed out with the same signature on the previous bump PR #12354 (both shard 3/4 and shard 4/4, reaching 500/1000 and 537/1000), which pins this as pre-existing rather than introduced here.

This looks related to but distinct from ROCm/rocm-systems#11673: that issue is the PAL shard 1/4 with a segfault in Unit_deviceAllocationAfterDeviceReset driving cascading timeouts. There's no segfault in this ROCR run, so I think it may be a separate problem, and I couldn't find an existing issue covering the ROCR-side hang. Flagging for the rocm-systems team rather than filing, since it's not this repo's component.

Also for the record: four jobs show CANCELLED in run 35653250677. Those were superseded by the rerun in 35655834675 and are not failures.

What this bump actually brings in

TheRock 90d12f3 -> 5c67705 (12 commits). This is not just a workflow-reference update; three submodules moved underneath it.

Submodule	Repo	Commits	Notable changes
rocm-libraries	ROCm/rocm-libraries	4	TensileLite enables PrefetchGL2 with GSU (#11851); rocKE GDN decode kernel and dispatch for gfx950 (#12172); hipSPARSELt TDM sparse GSU-offset and metadata batch-stride fix on gfx1250 (#12339); MIOpen removes gfx803-tuned solvers and stale kernel sources (#11761)
rocm-systems	ROCm/rocm-systems	7	Mostly rocjitsu: packed FP32 source selections on the SIMD path (#11825), simulated streaming stores and integer VOPD pairs (#11679), generation-based GPU discovery refactor, GPU VM and queue lifetime hardening. Plus rocprofv3 PC-sampling integration tests (#8097) and a TheRock workflow pin (#11898)
compiler/amd-llvm	ROCm/llvm-project	1	OpenMP build fix: remove hwloc INSTALL_INTERFACE on the omp target (#218923)

None of these plausibly explains a Windows hip-tests stream/event hang on gfx110X, which supports the pre-existing read above.

Project 231 sweep

Checked the three open ci-red items; none are close-able on this run's evidence, so no status changes are proposed:

• [Issue] Linux : hipkernelprovider — hip-kernel-provider-external-integration_comprehensive_suite hangs, 30-min timeout #11529 (hipkernelprovider comprehensive-suite hang): this PR runs only the standard shard 1/1 hipkernelprovider job (passed on both Windows gfx110X-all and Linux gfx94X-dcgpu), not the nightly comprehensive suite the issue tracks. Inconclusive, leaving open.
• [Issue] Windows nightly: libhipcxx_amdclang comprehensive tests fail — 20 failed / 12 unresolved libhip++ tests TheRock#7126 (Windows gfx1151 libhipcxx_amdclang comprehensive): this PR runs libhipcxx only on Windows gfx110X-all and Linux gfx94X-dcgpu (all passed), not gfx1151 comprehensive. Inconclusive, leaving open.
• MIOpen CI: system find-DB changes can merge without DBSync running on the affected architectures #12013 (MIOpen system find-DB can merge without DBSync): a gating-policy gap, not a red test, so a CI run can't confirm or refute it. Leaving open.

Writes made in this review

• Filed: none (the only failure is rocm-systems-owned and pre-existing).
• Updated: none.
• Closed: none.
• Jira: none.

One caveat: the bot has since opened #12413 (TheRock e9ae059), which will supersede this PR. As of this review #12413 is still in its build stages with no test results, so this review stands on its own; if #12398 sits long enough for #12413 to finish, reviewing that one instead may be the better move.

[tony-davis]

Approving after CI analysis.

[assistant-librarian]

TheRock Submodule Bump Activity

Newest first

• 2026-09-24 — Included in TheRock via ROCm/TheRock#8444.