[benrichard-amd]

Motivation

Part of work for ROCm/TheRock#5518

xml-test.cc (added in #13605) unconditionally includes <cpuid.h> and uses __cpuid() x86 inline asm, causing a compile error on non-x86 architectures. Two tests also hardcode EXPECT_EQ("x86_64", ...) for the arch attribute, which fails on ppc64le where the SUT correctly returns "ppc64".

Technical Details

• Guard #include <cpuid.h>, the CpuId struct, HostCpuId() function, and the GetXmlFromCpu_FillsAffinityArchAndCpuid test behind #if defined(__x86_64__) || defined(_M_X64), matching the convention used in xml.cc, init.cc, and init-test.cc.
• Add a NCCL_HOST_ARCH preprocessor macro that mirrors the arch detection logic in src/graph/xml.cc ("x86_64" / "ppc64" / "arm64"). Replace the two remaining hardcoded "x86_64" assertions in GetXmlFromSys_NoSysfsAttachesToTheUnknownCpu and GetXmlFromSys_ExistingParentIsFilledInTurn with NCCL_HOST_ARCH.
• The GetXmlFromCpu_FillsAffinityArchAndCpuid test remains x86-only since it validates cpuid vendor/family/model fields that only exist on x86.

Test Plan

Built RCCL and ran the host microtest suites (ctest -R "Micro|HostOnly|MicroTopo") on a POWER10 ppc64le machine.

Test Result

137/153 ctest targets pass. The 2 previously failing XmlSysfsMicrotest arch tests now pass. The 16 remaining failures (2 unique tests × suite/config variants) are pre-existing and unrelated:

• P2pSchedule_ZeroGroupSizeParam_DiesOnDivideByZero — death test expects SIGFPE, but POWER hardware returns 0 for integer divide-by-zero instead of trapping.
• GroupJobUnlinkCommsTest.EmptyAsyncJobsStillClearsTheCommHeadChain — groupJob pointer not being cleared; logic bug in the new test.

[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/xml-test.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

3 more finding(s), not posted inline

• projects/rccl/test/host/xml-test.cc:658 (Low) x86 guard wraps the whole of GetXmlFromCpu_FillsAffinityArchAndCpuid, so off x86 nothing asserts the non-zero affinity_offset branch (xml.cc:529); guard only the cpuid parts.
• projects/rccl/test/host/xml-test.cc:18 (Nit) NCCL_HOST_ARCH tests PPC64 variants where xml.cc:535 tests PPC; use PPC so the test copy matches production.
• projects/rccl/test/host/xml-test.cc:17 (Nit) Two adjacent blocks repeat the same x86 guard; fold the <cpuid.h> include into the x86_64 branch of the NCCL_HOST_ARCH chain.

Bottom line: LGTM.

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

[amd-argus]

ARGUS

Bottom line: LGTM.

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