[Common] Fix Build: NCCL EP build to respect MAX_JOBS

[phu0ngng]

Description

NCCL EP build to respect MAX_JOBS.

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refactoring

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[phu0ngng]

/te-ci L0

[greptile-apps]

Greptile Summary

This PR updates the NCCL EP build in setup.py to respect the MAX_JOBS environment variable (and NVTE_BUILD_MAX_JOBS) by replacing the hardcoded os.cpu_count() or 8 with the shared get_max_jobs_for_parallel_build() utility already used by the CMake build path.

• get_max_jobs_for_parallel_build() returns 0 as a sentinel for "no limit", but the make invocation in setup.py unconditionally passes str(nproc) to -j, producing make -j 0 when no job limit is configured. The parallel build_ext.py call site correctly guards this with if max_jobs > 0; the same pattern is needed here.
• When a valid job count is provided via MAX_JOBS, the change works as intended and aligns the NCCL EP build with the rest of the TE build system.

Confidence Score: 4/5

Safe to merge if the make -j 0 edge case is addressed; builds that explicitly set MAX_JOBS work correctly today.

When no job-limit environment variable is set, get_max_jobs_for_parallel_build() returns 0 and the NCCL EP make invocation becomes make -j 0, whose behavior is undefined across GNU make versions. Builds with an explicit MAX_JOBS are unaffected, but the default (no-limit) path is broken in a way that depends on the host's make version.

setup.py — the make invocation needs the same if nproc > 0 guard that build_ext.py already uses for its CMake parallel argument.

Important Files Changed

Filename	Overview
setup.py	Replaces hardcoded os.cpu_count() or 8 with get_max_jobs_for_parallel_build(), but does not guard against the 0 sentinel (unlimited) returned by that function, resulting in make -j 0 when no job limit is configured.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[build_nccl_ep_submodule] --> B[get_max_jobs_for_parallel_build]
    B --> C{env var set?}
    C -->|NVTE_BUILD_MAX_JOBS or MAX_JOBS| D[return N gt 0]
    C -->|none set| E[return 0]
    D --> F["make -j N (correct)"]
    E --> G["make -j 0 (undefined behavior)"]
    style G fill:#f66,color:#fff

Loading

_{Reviews (3): Last reviewed commit: "Merge branch 'main' into te_ep/fix_build" | Re-trigger Greptile}

[timmoon10]

The fix at #3034 (comment) will reuse our existing infrastructure.

[phu0ngng]

/te-ci L0

[timmoon10]