Skip to content

feat(mars): add native Mars kernels and htcc wrapper - #1

Closed
Ceng23333 wants to merge 1 commit into
base/v2026.08.18-refactorfrom
refactor-dev
Closed

Ceng23333 wants to merge 1 commit into
base/v2026.08.18-refactorfrom
refactor-dev

Conversation

@Ceng23333

@Ceng23333 Ceng23333 commented Aug 21, 2026

Copy link
Copy Markdown
Owner

Summary

  • Add Mars device backend under src/native/cuda/mars/ (mirrors src/native/cuda/metax/ layout).
  • Thin Operator<…, Device::Type::kMars> specializations reuse shared src/native/cuda/ops/*/kernel.h (same pattern as metax — not #include of metax headers).
  • HPCC-only glue: hcblas, scripts/htcc_wrapper.sh, CMake/device registration.

Motivation

Real delta vs product freeze Ops 94965ca (v2026.08.18-refactor) for --refactor-dev Mars/HPCC.

Layout policy

  • Mars impl stays under src/native/cuda/mars/.
  • Share generic CUDA ops via src/native/cuda/ops, matching metax.
  • Do not rewrite Mars to include metax device headers (wrong Device::Type).

Type of Change

  • feat

Port MetaX CUDA kernels to native/cuda/mars, compile .cc via htcc_wrapper,
and link HPCC libhtomp so InfiniOps can build WITH_MARS.
@Ceng23333

Copy link
Copy Markdown
Owner Author

Closing in favor of the upstream PR into InfiniTensor/InfiniOps: InfiniTensor#939 (Mars-only cherry-pick onto master, full template + smoke build evidence).

@Ceng23333 Ceng23333 closed this Aug 21, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant