All pull requests

Pull requests

• Build the legacy epilogue thread op where it is used, after the PDL wait

  #3788·almightychang opened Oct 25, 2026·

  ·

• Wait on the previous grid in every warp that loads global memory

  #3787·almightychang opened Oct 25, 2026·

  ·

• Read device-memory epilogue fusion scalars after the PDL wait

  #3786·almightychang opened Oct 25, 2026·

  ·

• [CuTeDSL] Select the stage-0 smem layout instead of slicing the swizzled staged layout in the dense GEMM examples, issue #3784

  #3785·yunweili3 opened Oct 25, 2026·

  ·
  1

• Count the epilogue-load warp in the SM100 mixed TMA/cp.async GEMM's scheduler pipeline

  #3781·builtin-pb opened Oct 24, 2026·

  ·
  1

• [CI] Run the CuTeDSL distributed tests on SM103 and the all-reduces on SM90

  #3780·0z5a opened Oct 24, 2026·

  ·

• [examples] Build and run eight Blackwell examples for sm_103a

  #3779·0z5a opened Oct 24, 2026·

  ·

• [examples] Accept cc 10.3 in the runtime checks of six Blackwell GEMM examples

  #3778·0z5a opened Oct 24, 2026·

  ·

• CuTe DSL: allow selecting a host CPU for TVM FFI AOT exports

  #3777·michaelfeil opened Oct 24, 2026·

  ·
  3

• Add round-to-nearest FastNumericArrayConverter for float to int32

  #3776·srimanachanta opened Oct 23, 2026·

  ·

• [CuTeDSL] Add cute.arch.movmatrix_trans

  #3775·srimanachanta opened Oct 23, 2026·

  ·

• [CuTeDSL] Add has_side_effects to inline_ptx and drop it on pure helpers

  #3774·srimanachanta opened Oct 23, 2026·

  ·

• [examples] 88_hopper_fmha: add a BF16 runner and the 88_hopper_fmha_bf16 target

  #3772·0z5a opened Oct 23, 2026·

  ·

• [CuTeDSL] Test the RMSNorm example on Hopper

  #3771·0z5a opened Oct 23, 2026·

  ·

• [Hopper CuTeDSL] Add tests for the dense GEMM, FMHA and CTA norm examples

  #3770·0z5a opened Oct 23, 2026·

  ·

• [CuTeDSL] Add a regression test for bf16 tile store vectorization (#3463)

  #3769·0z5a opened Oct 23, 2026·

  ·

• [Operators] Import torch only where the blockscaled GEMM kernels need it

  #3768·ConnorBaker opened Oct 22, 2026·

  ·

• Keep SM90 heuristics kernels when SM100 is also targeted

  #3767·v0ropaev opened Oct 21, 2026·

  ·

• [examples/77] Fix the MLA fused split-KV reduction's inter-CTA wait (wrong output, hang)

  #3763·builtin-pb opened Oct 21, 2026·

  ·
  1

• [Conv] Fix scale/bias traversal for split-K fused fprop

  #3762·0z5a opened Oct 20, 2026·

  ·
  1

• [CuTeDSL] Clarify the IKET quick-start GPU requirement

  #3758·0z5a opened Oct 20, 2026·

  ·
  1

• [CuTeDSL] Make the dynamic-layout guide use explicit singleton strides

  #3757·0z5a opened Oct 20, 2026·

  ·
  1

• [CuTeDSL] Fix runnable examples in the TVM FFI guide

  #3756·0z5a opened Oct 20, 2026·

  ·
  1

• Rename pytest execution to separate from pytest unit cases

  #3755·zekunf-nv opened Oct 19, 2026Collaborator·

  ·
  5

• Treat 103f as an SM100-family target in CMake

  #3754·0z5a opened Oct 19, 2026·

  ·