All pull requests

Pull requests

• Attention gym gb300 upstream

  CLA Signed
  #4053·yushangdi opened Oct 16, 2026Contributor·

  ·

• [autotuner] Fill short per-loop lists in user seed configs

  CLA Signed
  #4045·yashb98 opened Oct 16, 2026Contributor·

  ·

• [no land] Keep the fused RMSNorm kernel and twelve GB300 configs

  CLA Signed
  #4044·yushangdi opened Oct 16, 2026Contributor·

  ·

• Isolate retained in-process benchmark arguments

  CLA Signed
  #4043·yushangdi opened Oct 16, 2026Contributor·

  ·

• Seed bounded cluster and full-row register packets

  CLA Signed
  #4042·yushangdi opened Oct 16, 2026Contributor·

  ·

• Reject rolled-cluster padding beyond signed Int32 bounds

  CLA Signed
  #4041·yushangdi opened Oct 16, 2026Contributor·

  ·

• Admit block-dimension geometry in consumer packet prefetch

  CLA Signed
  #4040·yushangdi opened Oct 16, 2026Contributor·

  ·

• Support safe affine one-trip consumer packet prefetch

  CLA Signed
  #4039·yushangdi opened Oct 16, 2026Contributor·

  ·

• Add backend capability hooks for out-of-tree Triton drivers

  CLA Signed
  #4035·yzchang-plus opened Oct 16, 2026·

  ·

• Make benchmarking and launcher robust for out-of-tree Triton drivers

  CLA Signed
  #4034·yzchang-plus opened Oct 16, 2026·

  ·

• [AOT] Fix collect hash sampling for in-place kernels

  CLA Signed
  #4033·NaOHCC opened Oct 16, 2026Contributor·

  ·

• Add backend device-type routing for non-Triton accelerators

  CLA Signed
  #4032·yzchang-plus opened Oct 16, 2026·

  ·

• [CD] Add XPU pretuned benchmark job to the dispatch workflow

  CLA Signed
  #4031·NaOHCC opened Oct 16, 2026Contributor·

  ·

• [compiler] Reduce over multiple dims as one merged reduction dim

  CLA Signed
  #4030·coder-2011 opened Oct 16, 2026Contributor·

  ·
  2

• [cutedsl] Prioritize async matmul seeds

  CLA Signed
  #4027·coder-2011 opened Oct 15, 2026Contributor·

  ·
  1

• [cutedsl] Interleave complete matmul seed schedules

  CLA Signed
  #4026·coder-2011 opened Oct 15, 2026Contributor·

  ·
  1

• [benchmark] Add reproducible K2 scaled-mm comparisons

  CLA Signed
  #4024·yushangdi opened Oct 15, 2026Contributor·

  ·

• [noland] Record paused K2 hillclimb results and follow-up plan

  CLA Signed
  #4023·yushangdi opened Oct 15, 2026Contributor·

  ·

• [cutedsl] Add measured GB300 scaled-mm pretuned configurations

  CLA Signed
  #4022·yushangdi opened Oct 15, 2026Contributor·

  ·

• [cutedsl] Add guarded dense tcgen05 schedules and output variants

  CLA Signed
  #4021·yushangdi opened Oct 15, 2026Contributor·

  ·

• [cutedsl] Seed short persistent row reductions and admit them to proven bounds

  CLA Signed
  #4018·yushangdi opened Oct 15, 2026Contributor·

  ·

• [cutedsl] Add proven nc_l2_first packet loads and an l2_evict_last packet store policy

  CLA Signed
  #4016·yushangdi opened Oct 15, 2026Contributor·

  ·

• [cutedsl] Pack FP8 persistent reduction stores

  CLA Signed
  #4015·yushangdi opened Oct 15, 2026Contributor·

  ·

• [autotuner] Cap compiler seeds in best-available initial populations

  CLA Signed
  #4014·coder-2011 opened Oct 15, 2026Contributor·

  ·
  1

• [cutedsl] Improve narrow pointwise tile seeds

  CLA Signed
  #4013·coder-2011 opened Oct 15, 2026Contributor·

  ·
  1