All pull requests

Pull requests

• fix setup for sm90a

  #25·yiakwy-xpu-ml-framework-team opened Oct 4, 2026·

  ·

• fix(interface): reject out-of-range topk before allocating outputs

  #24·Afloat16 opened Oct 1, 2026·

  ·

• Fix GCC 11 compatibility in stride error

  #22·dougbtv opened Sep 22, 2026·

  ·
  1

• Fix cluster gather completion barriers targeting the wrong CTA

  #16·lucifer1004 opened Sep 15, 2026·

  ·

• Keep interface validation active under python -O

  #15·adenzhou1350 opened Sep 14, 2026·

  ·

• [SM90] Add optimized SM90 support.

  #14·yuyu5333 opened Sep 14, 2026·

  ·
  6

• fp32 kernel main loop improvements

  #12·ylh440104 opened Sep 11, 2026·

  ·

• Fall back to torch when lspci is unavailable

  #11·PerryLink opened Sep 10, 2026·

  ·

• Reduce redundant Top-K host setup

  #8·morluto opened Sep 10, 2026·

  ·
  1

• Bound Top-K output stores and validate input padding

  #6·morluto opened Sep 10, 2026·

  ·
  1

• Honor the input CUDA device and stream in Top-K

  #4·morluto opened Sep 10, 2026·

  ·
  1

• Prevent overflow in sorted Top-K index offsets

  #2·morluto opened Sep 10, 2026·

  ·
  1