Issues

• [BUG] IKET: an IKET call in host-side @cute.jit code segfaults the compile when IKET is enabled (e.g. under run-iket)

  CuTe DSL
  NVIDIA/cutlass#3882 · Filter by author 0z5a opened on Nov 10, 2026

• [BUG] IKET: a boolean payload fails to compile ('iket.mark' op operand #0 ... but got 'i1')

  bugCuTe DSL
  NVIDIA/cutlass#3881 · Filter by author 0z5a opened on Nov 10, 2026

• cute.compile can't use the file cache: compile-only builds force no_cache, and the cache key predates the TVM-FFI arg spec

  NVIDIA/cutlass#3880 · Filter by author matteso1 opened on Nov 10, 2026

• [BUG] CuTe TMA kernels fail to compile on Windows with CUTLASS's MSVC flags: C2719 in nvcc's cudafe1.stub.c (CUtensorMap alignas(128) under /Zc:__cplusplus)

  NVIDIA/cutlass#3878 · Filter by author zbrad opened on Nov 10, 2026

• [BUG] CuTe DSL: while n: with a runtime Int32 or Float32 fails with an internal IR error

  NVIDIA/cutlass#3874 · Filter by author ruiling-smartbear opened on Nov 9, 2026

• [BUG] CuTe DSL: indexing a layout with dynamic shape at a small static 1-D index fails with "unrealized_conversion_cast"

  NVIDIA/cutlass#3873 · Filter by author ruiling-smartbear opened on Nov 9, 2026

• [BUG] CuTe DSL: integer TensorSSA "/" returns an f32 vector labeled with the integer type

  NVIDIA/cutlass#3871 · Filter by author ruiling-smartbear opened on Nov 9, 2026

• [BUG] CuTe DSL: TensorSSA slicing with an integer into a nested mode returns the wrong elements

  NVIDIA/cutlass#3869 · Filter by author ruiling-smartbear opened on Nov 9, 2026

• [BUG] CuTe DSL: BFloat16 scalar kernel arguments are truncated instead of rounded to nearest even

  NVIDIA/cutlass#3867 · Filter by author ruiling-smartbear opened on Nov 9, 2026

• [BUG] CuTe DSL: make_tiled_tma_atom accepts global strides that are not 16-byte aligned; TMA silently rounds them down

  NVIDIA/cutlass#3866 · Filter by author ruiling-smartbear opened on Nov 9, 2026
  1

• [BUG] CuTe DSL: cute.arch.red() fails to compile with its default sem=None

  NVIDIA/cutlass#3864 · Filter by author ruiling-smartbear opened on Nov 9, 2026

• [BUG] CuTe DSL: TensorSSA.reduce() with MAX or MIN fails for integer tensors

  NVIDIA/cutlass#3862 · Filter by author ruiling-smartbear opened on Nov 9, 2026

• [BUG] CuTe DSL: reading leading_dim keeps from_dlpack tensors alive

  NVIDIA/cutlass#3860 · Filter by author ruiling-smartbear opened on Nov 9, 2026

• [BUG] CuTe DSL: range() with a runtime step fails to compile with Int64 bounds

  bugCuTe DSL
  NVIDIA/cutlass#3858 · Filter by author ruiling-smartbear opened on Nov 9, 2026

• [BUG] CuTe DSL: Hopper dense_gemm example fails its fp8 reference check

  NVIDIA/cutlass#3855 · Filter by author ruiling-smartbear opened on Nov 9, 2026

• [BUG] CuTe DSL: Uint8 pointers become Int8 inside kernels

  NVIDIA/cutlass#3854 · Filter by author ruiling-smartbear opened on Nov 8, 2026

• [BUG] CuTe DSL: integer to Float8E4M3FN/E5M2 is a bit reinterpretation for Int8 and fails to compile for wider ints

  NVIDIA/cutlass#3846 · Filter by author Zhonglin1231 opened on Nov 6, 2026
  1

• [BUG] CuTe DSL: -0.0 and 0.0 constants in one kernel come out as the same zero

  NVIDIA/cutlass#3845 · Filter by author Zhonglin1231 opened on Nov 6, 2026

• [BUG] CuTe DSL: abs() of an unsigned runtime value treats it as signed

  NVIDIA/cutlass#3844 · Filter by author Zhonglin1231 opened on Nov 6, 2026

• [BUG] CuTe DSL: signed to wider unsigned conversion zero-extends at runtime

  NVIDIA/cutlass#3843 · Filter by author Zhonglin1231 opened on Nov 6, 2026

• [BUG] cute::tma_descriptor_replace_dims_strides_in_shared_mem writes to a read-only inline-asm input operand

  ? - Needs TriagebugCUTLASS C++
  NVIDIA/cutlass#3831 · Filter by author Marisa1885 opened on Nov 5, 2026

• [BUG] cutlass.operators.providers keeps the provider ImportError traceback alive, pinning the importer's frames and locals (GPU memory leak)

  NVIDIA/cutlass#3822 · Filter by author janbernloehr opened on Nov 4, 2026

• [Bug][CuTeDSL][Fuzz] utils.gemm.sm100.epilogue releases the accumulator before tcgen05.ld completes: no tcgen05.wait::ld before the arrive that lets the next MMA overwrite TMEM

  bugCuTe DSL
  NVIDIA/cutlass#3805 · Filter by author jeffchen006 opened on Oct 31, 2026

• [BUG] CuTe DSL 4.7.1: the same MLIR compiles to different PTX from run to run (address-dependent codegen in the embedded libNVVM 13.3); not reproduced in 4.8.1

  bugCuTe DSL
  NVIDIA/cutlass#3797 · Filter by author 16bit-ykiko opened on Oct 28, 2026
  1

• [QST] Remove unused Python protobuf dependency from nvidia-cutlass-dsl

  ? - Needs Triagequestion
  NVIDIA/cutlass#3796 · Filter by author i-Pear opened on Oct 27, 2026
  2