All pull requests

Pull requests

• Add CuTe conv2d fprop tutorial using TMA im2col on SM90

  #3700·chrisXYZhang opened Sep 29, 2026·

  ·

• [CuTeDSL] Add compilation/launch listener hooks for external tracers

  #3697·FindHao opened Sep 29, 2026·

  ·

• Quote source paths in generated manifest.cmake

  #3696·SammyTourani opened Sep 28, 2026·

  ·
  1

• [CuTeDSL] examples: fix N clobber in 01_array_concepts.ipynb

  #3695·zkyue opened Sep 28, 2026Contributor·

  ·

• test: cover SM101/SM110 enum alias contracts

  #3694·0z5a opened Sep 28, 2026·

  ·

• Keep plain pytest functions in separate sharding groups

  #3693·jakezwang opened Sep 28, 2026·

  ·

• Honor rank_2k in library operation filters

  #3692·jakezwang opened Sep 28, 2026·

  ·

• Fix scale-factor columns in kernel audit CSV

  #3691·jakezwang opened Sep 28, 2026·

  ·

• [CuTeDSL] Work around lost D when MMA C is a separate register tensor

  #3690·yunweili3 opened Sep 27, 2026·

  ·
  1

  1

• [CuTeDSL] Make dynamic % and float // match Python semantics

  #3688·yunweili3 opened Sep 27, 2026·

  ·
  1

• Stop CUDA error messages from reentering GPU detection

  #3686·Hashim1999164 opened Sep 26, 2026·

  ·
  1

• [C++] Fix non-packed Tensor NHWC and add unit tests

  #3685·MaybeWilli opened Sep 26, 2026·

  ·
  1

• [CuTeDSL] Only parse -diagnostic itself from the command line

  #3684·SammyTourani opened Sep 26, 2026·

  ·
  1

• Load runtime libraries before dumping the export object file

  #3683·SammyTourani opened Sep 26, 2026·

  ·
  1

• [CuTeDSL] Restrict pickle.loads in signature metadata decoding

  #3682·alexxony opened Sep 26, 2026·

  ·

• Fix EVT testbed host/device argument size mismatch for DAG nodes

  #3679·SammyTourani opened Sep 25, 2026·

  ·
  1

• [CuTeDSL] Raise on unsupported tma_desc_ptr types in TMA copies

  #3678·SammyTourani opened Sep 25, 2026·

  ·
  1

• Fix ArraySubbyte::back() and PredicateVector::is_zero()

  #3677·rahul05ranjan opened Sep 25, 2026·

  ·
  2

  1

• [CuTeDSL] Convert gmem TMA descriptor pointers to generic in tma_desc_ptr

  #3676·yunweili3 opened Sep 24, 2026·

  ·
  1

• [CuTeDSL] Honour cutlass.Pointer TMA descriptors passed via tma_desc_ptr

  #3675·yunweili3 opened Sep 24, 2026·

  ·
  1

• [CuTeDSL] Add FP32 residual and Affine fusion example

  #3674·0z5a opened Sep 24, 2026·

  ·

• [CuTeDSL] Add non-TMA bulk reduction CopyOp for FP16, BF16, FP32 and FP64

  #3669·0z5a opened Sep 24, 2026·

  ·

• [CuTeDSL] Explain stride alignment after partitioning

  #3668·0z5a opened Sep 24, 2026·

  ·

• [CuTeDSL] Reject incomplete copy-atom thread groups

  #3667·0z5a opened Sep 24, 2026·

  ·

• [CuTeDSL] docs: fix cutlass.range unroll=0 and unroll_full descriptions

  #3666·zkyue opened Sep 24, 2026Contributor·

  ·
  3