[benrichard-amd]

Motivation

Part of ROCm/TheRock#5518

Tests were failing to build on ppc64le architecture because of how dead code that is pruned on x86, is not pruned on ppc64le.

Technical Details

The MicroEnqueue host test (#10861) compiles enqueue.cc against stub implementations. enqueue.cc references the global netTransport (defined in src/transport/net.cc) but the test does not link against net.cc. The code path that uses netTransport is unreachable, so on x86 --gc-sections eliminates the reference and linking succeeds.

On ppc64le, the ELFv2 ABI uses a TOC (Table of Contents) for global variable access. The compiler emits a TOC entry for netTransport at compile time, and this TOC entry creates a hard symbol reference that survives --gc-sections — the linker can discard the code section but not the TOC relocation. This causes:

ld.lld: error: undefined symbol: netTransport
>>> referenced by enqueue-test.cc
>>>               test/host/CMakeFiles/rccl-UnitTestsMicroEnqueue.dir/enqueue-test.cc.o:(.toc+0x28)

The fix adds a zero-initialized definition of netTransport in transport_stubs.cc. It is never called — it only satisfies the TOC reference.

Issue Tracking

Test plan

• Full TheRock ppc64le build succeeds with this patch applied
• No impact on x86 builds (the stub is benign — netTransport was already undefined and GC'd)

Submission Checklist

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

[therock-pr-bot]

✅ All Policy Checks Passed

Check	Status	Details
📝 PR Description	✅ Pass	—
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	⚠️ Warning	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/rccl/test/host/fakes/transport_stubs.cc; no test file found
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—

🎉 All policy checks passed!

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

🙋 Wish to Override Policy?

[amd-argus]

ARGUS

Findings

1 finding(s) that fall outside the diff and could not be anchored to a line

• projects/rccl/test/host/CMakeLists.txt:271 (Low) Comment at CMakeLists.txt:267-276 now misdescribes netTransport

Bottom line: LGTM once #include "transport.h" moves up to the file's existing include block. That's mechanical, so it doesn't need another pass from me.

_{Comment @argus review to re-run, or @argus re-review after pushing.}

[benrichard-amd]

Hi @pvallem we need 1 more review in order to merge this. Thanks!

[amd-argus]

ARGUS

Findings

1 finding(s) that could not be posted as inline comments

• projects/rccl/test/host/CMakeLists.txt:384 (Low) Defining netTransport makes the undefined-symbol inventory comment in test/host/CMakeLists.txt stale, including three hard counts and the recount recipe below them Defining netTransport makes the undefined-symbol inventory in projects/rccl/test/host/CMakeLists.txt stale, and the recount recipe at lines 392-393 of that file will now report a number the comment contradicts: ''' line 382 "leaves 34 undefined" -> 33 line 383 "33 ordinary externals" -> 32 line 384 drop netTransport from the named symbols line 388 "All 34 that remain" -> 33 ''' The standalone block needs no edit, since line 966 of that file only cross-references this one.

Bottom line: LGTM once the symbol inventory in test/host/CMakeLists.txt drops netTransport and its three counts come down by one. That one is mechanical, so it does not need another pass from me.

_{Comment @argus review to re-run, or @argus re-review after pushing.}

[systems-assistant]

TheRock Submodule Bump Activity

Newest first

• 2026-09-21 — Included in TheRock via ROCm/TheRock#8378.