Skip to content

Fix AArch64 NCCL EP linking with BOLT enabled - #3627

Merged
fheinecke merged 2 commits into
NVIDIA:mainfrom
aybchan:fix/ep-linker-flag
Oct 5, 2026
Merged

fheinecke merged 2 commits into
NVIDIA:mainfrom
aybchan:fix/ep-linker-flag

Conversation

@aybchan

@aybchan aybchan commented Oct 5, 2026 •

Copy link
Copy Markdown
Member

Description

BOLT NCCL EP submodule build fails on arm because GNU ld.bfd rejects --no-fix-cortex-a53-843419.

Pass the original -mno-fix-cortex-a53-843419 option to the host compiler driver through NVCC’s -Xcompiler, preserving the intended behavior of disabling the Cortex-A53 workaround.

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

Changes

Please list the changes introduced in this PR:

  • Fix NCCL EP linking on AArch64 by passing -mno-fix-cortex-a53-843419 through NVCC’s -Xcompiler instead of emitting an unsupported linker option.
  • Fix Transformer Engine’s CMake link options to pass the same flag directly to the compiler driver, removing the incorrect LINKER: prefix.

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

Signed-off-by: Alex Y. Chan <alechan@nvidia.com>
@greptile-apps

greptile-apps Bot commented Oct 5, 2026 •

Copy link
Copy Markdown
Contributor

RetriggerConfidence Score: 5/5

[Medium risk] Adjusts compiler flags for ARM64 NCCL linking.

The PR appears safe to merge based on the reviewed changes.

Summary

The PR changes how the AArch64 Cortex-A53 workaround flag reaches the compiler driver, avoiding an unsupported direct linker option.

  • The NCCL EP build now passes the flag through NVCC’s -Xcompiler.
  • The common CMake build passes the flag without the LINKER: prefix.

Reviews (2) · Last reviewed commit: "Merge branch 'main' into fix/ep-linker-f..."

@fheinecke

Copy link
Copy Markdown
Collaborator

Looks like our internal CI isn't triggering arm64 builds for PR CI runs, just amd64. CI pipeline run ID: 71705833

I'll approve/merge once this passes

@fheinecke fheinecke left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This has passed CI sufficiently enough to show that it works

@fheinecke
fheinecke merged commit f68a64a into NVIDIA:main Oct 5, 2026
12 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants