fix(rocr): load thunk libraries from the runtime directory

[fcui-amd]

Motivation

rocminfo

WSL environment detected.
Cannot load librocdxg.so, failed:librocdxg.so: cannot open shared object file: No such file or directory
GetExportAddress failed: /opt/venv/lib/python3.14/site-packages/_rocm_sdk_devel/bin/../lib/libhsa-runtime64.so.1: undefined symbol: hsaKmtOpenKFD
hsa_init Failed, possibly no supported GPU devices

Technical Details

Add a fallback path for loading thunk libraries when the normal platform library search fails.
The fallback locates the directory containing the currently loaded ROCR runtime library and attempts to load the thunk library from the same directory.

Issue Tracking

Test Plan

rocminfo pass without specified LD_LIBRARY_PATH setting

Test Result

rocminfo pass

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/rocm-systems/blob/develop/CONTRIBUTING.md.

[therock-pr-bot]

❌ PR Check — Action Required

Check	Status	Details
🌿 Branch Name	✅ Pass	—
📝 PR Title/Description	✅ Pass	—
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	❌ Fail	Error: Source/code files changed without an accompanying unit test. Expected: add at least one test file named like test_<name>.py / test_<name>.cpp (or <name>_test.*). Current: code file(s) changed: projects/rocr-runtime/runtime/hsa-runtime/core/runtime/thunk_loader.cpp, projects/rocr-runtime/runtime/hsa-runtime/core/util/lnx/os_linux.cpp, projects/rocr-runtime/runtime/hsa-runtime/core/util/os.h, projects/rocr-runtime/runtime/hsa-runtime/core/util/win/os_win.cpp; no test file found
🔎 pre-commit	⏳ Pending	⏳ Still running…
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—

⚠️ 1 policy check(s) failed. Please address the issues above before this PR can be Reviewed.

🚫 Please fix the failed policies

• ❌ Unit Test

The Not ready to Review label was added to this PR. Once all policies pass, the label is removed automatically.

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

[therock-pr-bot]

🚫 Please fix the failed policies before requesting reviews.

The following policy checks failed:

• ❌ Unit Test

The Not ready to Review label has been added to this PR.
Once all policies pass, the label will be removed automatically.

[kentrussell]

Thunk part is fine. Will let David/Chris give the approval for the ROCr side

[cfreeamd]

🤖 AI-assisted pre-review (for @cfreeamd)

The adjacent-path loader fallback in thunk_loader.cpp and the new os::GetAdjacentLibraryPath implementations (Linux dladdr, Windows GetModuleHandleExA/GetModuleFileNameA) look correct — self-locating via the runtime's own address, with sane truncation handling on Windows and a null-guard on DlError(). Two inline notes below (one Question on the bundled .ver exports, one Nit on the success log). Noting the PR is labeled "Not ready to Review," so this is early feedback.

[cfreeamd]

🤖 AI-assisted pre-review (for @cfreeamd)

[Question] These three exports (hsaKmtGetKernelQueueId, hsaKmtGetCoreRuntimeInfo, hsaKmtGetCoreDeviceInfo, mirrored in librocdxg.ver) come from the separate "export missing libhsakmt symbols" commit and are unrelated to the adjacent-path loader change. Two things worth confirming: (a) are all three actually defined in libhsakmt? A .ver entry for an undefined symbol is a silent no-op. (b) Is widening the public export surface intended as part of this PR, or would it be cleaner to split it out? The ABI change and the loader change have independent risk/review profiles.

[fcui-amd]

Sorry, this commit belongs to another PR and was included here by mistake.

[cfreeamd]

🤖 AI-assisted pre-review (for @cfreeamd)

[Nit] When the library is loaded via the adjacent-path fallback, this success message still prints library_name (the original name), so the log won't reveal that the fallback path was actually used. Consider logging the path that succeeded to aid diagnosis. (Also a pre-existing typo here while the line is nearby: "successully" → "successfully".)