All pull requests

Pull requests

• [CuTeDSL] Guard SM100 UMMA peer arrival mask on the cluster v-extent

  #3661·yunweili3 opened Sep 22, 2026·

  ·
  1

• [CuTe] Mask ThrK in TiledMMA::get_layoutC_TV (#3612)

  #3660·LiRunGuo opened Sep 22, 2026·

  ·
  1

  2

• [C++] Fix SM120 blockscaled small-N tiles breaking cutlass_profiler builds (#3646)

  #3659·LiRunGuo opened Sep 22, 2026·

  ·
  1

• [CuTe] Reject Copy_Atom copies whose src/dst size does not match the CopyOp (#3488)

  #3656·LiRunGuo opened Sep 21, 2026·

  ·
  1

• docs: explain CuTe DSL AOT error handling

  #3655·efegokdemir opened Sep 21, 2026·

  ·
  1

• [examples/77] Order each load-warp lane's producer_acquire wait before the TMA it guards

  #3654·vnnm404 opened Sep 21, 2026·

  ·

• [examples/77] Wait for TMEM statistics stores before publishing them to the correction warpgroup

  #3653·vnnm404 opened Sep 21, 2026·

  ·

• [examples/93] Wait for the Acc2 tcgen05.ld before releasing TMEM to the MMA warp

  #3652·vnnm404 opened Sep 21, 2026·

  ·

• [C++] Fix EpilogueDirectStore warp index decomposition; reject sliced-K (#3496)

  #3651·LiRunGuo opened Sep 21, 2026·

  ·
  1

• [C++] Fix Stream-K broadcast epilogue reduction; document broadcast vector convention (#3540)

  #3650·LiRunGuo opened Sep 21, 2026·

  ·
  1

• [CuTeDSL] Fix logging of disk-cache load failures

  #3649·LittleYier opened Sep 21, 2026·

  ·
  1

• [CuTeDSL] Add BF16 support to Hopper dense GEMM examples

  #3648·LiRunGuo opened Sep 21, 2026·

  ·
  1

• [CuTe DSL] Fix cross-CTA barrier scope and missing SFB wait in 2cta_mma_blockscale_m128 example, issue #3643

  #3647·yunweili3 opened Sep 21, 2026·

  ·
  1

• [CuTeDSL] diagnostics: drop deprecated @dynamic_expr / @range_dynamic from PHASE_DYNAMIC_INDEX hint

  #3644·zkyue opened Sep 20, 2026·

  ·

• [CuTe DSL] Allow selected read-only inline_ptx operands to be materialized into a register

  #3642·0z5a opened Sep 18, 2026·

  ·
  8

• CuTeDSL: preserve outer comprehension generator_targets across nested comprehensions (#3499)

  #3641·SIDDARTHAREDDY8 opened Sep 17, 2026·

  ·
  1

• Fix maximum_active_blocks wrappers passing int to CudaHostAdapter*

  #3638·chrisXYZhang opened Sep 14, 2026·

  ·
  1

• Fix: enable CUDA_CTA_RECONFIG_ACTIVATED for SM110a (Jetson AGX Thor)

  #3637·roma5087 opened Sep 14, 2026·

  ·
  1

• [CuTeDSL] Compare missing runtime arguments by identity

  #3634·voipmonitor opened Sep 13, 2026·

  ·

• Fix misaligned local memory access in 02_dump_reg_shmem

  #3633·random25160765-collab opened Sep 13, 2026·

  ·
  1

• [Epilogue] Align fixed visitor workspace offsets

  #3631·XFDG opened Sep 13, 2026·

  ·
  1

• [CuTe DSL] Reject incomplete dynamic f-string formats

  #3630·XFDG opened Sep 13, 2026·

  ·
  1

• [CuTe DSL] Cache each distinct one-time warning

  #3629·XFDG opened Sep 13, 2026·

  ·
  1

• [Core] Saturate scalar int-to-8-bit conversions

  #3628·XFDG opened Sep 13, 2026·

  ·
  1

• [Epilogue] Fix VisitorRowReduction helper return type

  #3627·XFDG opened Sep 13, 2026·

  ·
  1