Skip to content

[fix] fix CUTLASS_ROOT error when compiling nvidia_int8_gemm and simplify build setup#1152

Open
qinyiqun wants to merge 1 commit intoInfiniTensor:mainfrom
qinyiqun:fix_cutlass_root
Open

[fix] fix CUTLASS_ROOT error when compiling nvidia_int8_gemm and simplify build setup#1152
qinyiqun wants to merge 1 commit intoInfiniTensor:mainfrom
qinyiqun:fix_cutlass_root

Conversation

@qinyiqun
Copy link
Copy Markdown
Collaborator

@qinyiqun qinyiqun commented May 7, 2026

  • Remove standalone --cutlass option from xmake.lua; cutlass is now enabled automatically when CUTLASS_ROOT env var is set
  • Auto-add tools/util/include to include paths (no manual copy of cutlass/util needed for packed_stride.hpp)
  • Auto-detect CUDA arch from nvidia-smi when --cuda_arch is not specified, with fallback to native for unsupported archs (H100 → sm_90a, A100 → sm_80, etc.)

- Remove standalone --cutlass option from xmake.lua; cutlass is now
  enabled automatically when CUTLASS_ROOT env var is set
- Auto-add tools/util/include to include paths (no manual copy of
  cutlass/util needed for packed_stride.hpp)
- Auto-detect CUDA arch from nvidia-smi when --cuda_arch is not
  specified, with fallback to native for unsupported archs
  (H100 → sm_90a, A100 → sm_80, etc.)
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.

1 participant