NVIDIA/cutlass

View on GitHub

CUDA Templates and Python DSLs for High-Performance Linear Algebra

Stars: 10.5k
Forks: 2.1k
Open beginner issues: 16
Indexed issues: 472
Avg merge: 3d 11h
Merged PRs (30d): 7

Dominant language: C++
License: No license data
Last GitHub push: Sep 17, 2026
Latest indexed: Sep 19, 2026
Contributing guide: No contributing guide
Code of conduct: No code of conduct
Beginner labels: help wanted good first issue

Beginner friendly Everything

Open All

472 open issues indexed

[BUG] Operand initialized with a constant doesn't work with `inline_ptx` as an operand Open

? - Needs Triage bug CuTe DSL

Difficulty 4/5 3-5 days Newbie friendliness 52/100

NVIDIA/cutlass#3640 · 2 comments · Opened Sep 17, 2026
[BUG] CuTe DSL 4.7.1 TVM-FFI AOT objects reference symbols missing from `libcuda_dialect_runtime_static.a` Open

? - Needs Triage bug CuTe DSL

Difficulty 4/5 3-5 days Newbie friendliness 55/100

NVIDIA/cutlass#3635 · 1 comment · 1 reaction · Opened Sep 13, 2026
[BUG] Constexpr missing in dynamic control-flow regions Open

? - Needs Triage bug CuTe DSL

Difficulty 3/5 1-2 days Newbie friendliness 68/100

NVIDIA/cutlass#3632 · Opened Sep 13, 2026
[QST] Should `MMA_Atom::get_layoutC_TV()` mask or ignore `ThrK`? Open

? - Needs Triage CUTLASS C++ question

Difficulty 4/5 3-5 days Newbie friendliness 48/100

NVIDIA/cutlass#3612 · Opened Sep 11, 2026
[BUG] NVCC 13.3 cudafe++ explodes host code for structured bindings of CuTe TMA tuples Open

CUTLASS C++

Difficulty 5/5 Over a week Newbie friendliness 35/100

NVIDIA/cutlass#3606 · 2 comments · Opened Sep 9, 2026
[BUG] domain_offset overflows runtime Int32 coordinate with static stride Open

CuTe DSL

Difficulty 4/5 3-5 days Newbie friendliness 45/100

NVIDIA/cutlass#3604 · Opened Sep 9, 2026
[BUG] CuTe DSL: FloatNV8E5M3FNU.to(Float32) returns inf for the entire top exponent binade (codes 248-254) Open

? - Needs Triage bug CuTe DSL

Difficulty 3/5 1-2 days Newbie friendliness 68/100

NVIDIA/cutlass#3602 · 1 comment · Opened Sep 8, 2026
[BUG] Example 93 GQA initialization overwrites active remote contributions Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

NVIDIA/cutlass#3598 · Opened Sep 8, 2026
[BUG] Example 93 GQA reduction races when reusing shared scratch Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

NVIDIA/cutlass#3597 · Opened Sep 8, 2026
[BUG] SM90 blockwise FP8 GEMM reuses modified B scales across M waves Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

NVIDIA/cutlass#3596 · Opened Sep 8, 2026
[BUG] CuTe DSL 4.8 dev SIGABRTs in TVM-FFI launch for SM100 ragged SDPA (4.7 passes) Open

bug CuTe DSL

Difficulty 4/5 3-5 days Newbie friendliness 35/100

NVIDIA/cutlass#3595 · Opened Sep 8, 2026
[BUG] CuTe DSL 4.8 dev regresses SM80 SDPA backward compilation (NVVM constraint 'n') Open

? - Needs Triage bug CuTe DSL

Difficulty 4/5 3-5 days Newbie friendliness 45/100

NVIDIA/cutlass#3594 · Opened Sep 8, 2026
[BUG] SM100/SM103 ptr-array GEMM kernels size the per-SM tensormap workspace from the raw hw_info.sm_count, so a default KernelHardwareInfo causes out-of-bounds descriptor writes Open

CUTLASS C++

Difficulty 4/5 3-5 days Newbie friendliness 30/100

NVIDIA/cutlass#3585 · 1 comment · Opened Sep 4, 2026
[BUG] 02_dump_reg_shmem faults on sm_120: Fragment is 2-byte aligned but accessed through a 16-byte-aligned pointer Open

? - Needs Triage bug CUTLASS C++

Difficulty 4/5 3-5 days Newbie friendliness 55/100

NVIDIA/cutlass#3581 · 4 comments · Opened Sep 3, 2026
[BUG] `CUDA_CTA_RECONFIG_ACTIVATED` is not enabled for SM110a (Jetson AGX Thor) Open

? - Needs Triage bug CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 67/100

NVIDIA/cutlass#3576 · Opened Sep 1, 2026
[BUG] CuTeDSL >= 4.6 cannot compile NCCL GIN device ops — ptxas rejects the generated PTX (4.5.2 works) Open

CuTe DSL

Difficulty 5/5 Over a week Newbie friendliness 35/100

NVIDIA/cutlass#3573 · 2 comments · Opened Aug 31, 2026
Blockscaled GEMV accumulates past K when floor(K/tile) is not a multiple of the stage count: example 91 fails for batch > 1 Open

CUTLASS C++

Difficulty 4/5 3-5 days Newbie friendliness 55/100

NVIDIA/cutlass#3568 · Opened Aug 29, 2026
[DOC] Request: Error handling documentation for CuTe DSL exported C API Open

? - Needs Triage CuTe DSL documentation

Difficulty 3/5 1-2 days Newbie friendliness 52/100

NVIDIA/cutlass#3562 · 2 comments · 1 reaction · Opened Aug 28, 2026
Conv device wrappers: update() leaves stale geometry and fused-output pointers, two wrappers miss the 2 GiB gate, negative output extents not rejected Open

CUTLASS C++

Difficulty 4/5 3-5 days Newbie friendliness 52/100

NVIDIA/cutlass#3545 · 3 comments · Opened Aug 25, 2026
Small profiler defects: device-ID validation off by one, kTensorNCHW mapped to TensorNHWC, sparse-meta host/device type mismatch Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 76/100

NVIDIA/cutlass#3544 · Opened Aug 25, 2026
Three host reference utility defects: wrong MRE formula, operator precedence in padded-diagonal fill, unseeded per-element RNG in gaussian fill Open

CUTLASS C++

Difficulty 4/5 3-5 days Newbie friendliness 64/100

NVIDIA/cutlass#3543 · 1 comment · Opened Aug 25, 2026
Validation gaps: grouped/split-K can_implement accept crashing configs; batch grid wraps at 65536; Sm80 EVT mode gates are assert-only Open

CUTLASS C++

Difficulty 4/5 3-5 days Newbie friendliness 55/100

NVIDIA/cutlass#3542 · Opened Aug 25, 2026
Several 2.x device wrappers have update()/operator() paths that never compile or silently apply stale state Open

CUTLASS C++

Difficulty 5/5 Over a week Newbie friendliness 38/100

NVIDIA/cutlass#3541 · Opened Aug 25, 2026
GemmUniversalWithBroadcast ColumnMajor-output transpose leaves the bias vector indexed in the wrong coordinate system Open

CUTLASS C++

Difficulty 4/5 3-5 days Newbie friendliness 45/100

NVIDIA/cutlass#3540 · Opened Aug 25, 2026
EllGemm ColumnMajor-output workspace sizing uses the unswapped problem shape and undersizes split-K semaphores Open

CUTLASS C++

Difficulty 2/5 1-3 hours Newbie friendliness 88/100

NVIDIA/cutlass#3539 · 2 comments · Opened Aug 25, 2026
GemmWithKReduction parallel split-K: workspace initialization is skipped, partials land in the user output buffer Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 70/100

NVIDIA/cutlass#3538 · 1 comment · Opened Aug 25, 2026
Blockwise-scaled split-K GEMM applies slice-0 scale factors to every K slice Open

CUTLASS C++

Difficulty 4/5 3-5 days Newbie friendliness 64/100

NVIDIA/cutlass#3537 · 1 comment · Opened Aug 25, 2026
Blockscaled GEMV drops the K tail when floor(K/tile) is a multiple of the stage count: example 91 fails at --k=1152 Open

CUTLASS C++

Difficulty 4/5 3-5 days Newbie friendliness 68/100

NVIDIA/cutlass#3536 · 2 comments · Opened Aug 25, 2026
smem_atom_layoutSFB branches on majorSFA instead of majorSFB Open

CUTLASS C++

Difficulty 1/5 Under an hour Newbie friendliness 88/100

NVIDIA/cutlass#3535 · 3 comments · Opened Aug 25, 2026
maximum_active_blocks wrappers pass int into CudaHostAdapter* and no longer compile Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 82/100

NVIDIA/cutlass#3534 · 2 comments · Opened Aug 25, 2026
Sparse split-K kernels offset the E operand without the kElementsPerElementE division Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 72/100

NVIDIA/cutlass#3533 · 1 comment · Opened Aug 25, 2026
rank_2k_universal and symm_universal Arguments constructors discard batch_stride_B Open

CUTLASS C++

Difficulty 2/5 1-3 hours Newbie friendliness 82/100

NVIDIA/cutlass#3532 · 3 comments · Opened Aug 25, 2026
[BUG | QST] CPU overhead regression with TVM-FFI Open

? - Needs Triage bug CuTe DSL

Difficulty 4/5 3-5 days Newbie friendliness 35/100

NVIDIA/cutlass#3527 · 1 comment · 3 reactions · Opened Aug 24, 2026
Two small example bugs: wgrad split-k dumps the wrong tensor as Computed; FMHA diff-checker launches with block/grid swapped Open

CUTLASS C++

Difficulty 2/5 1-3 hours Newbie friendliness 88/100

NVIDIA/cutlass#3525 · Opened Aug 24, 2026
Example 64 (Ada FP8 grouped GEMM) refuses capable GPUs via exact CC 8.9 gate Open

CUTLASS C++

Difficulty 1/5 Under an hour Newbie friendliness 91/100

NVIDIA/cutlass#3524 · 1 comment · Opened Aug 24, 2026
NonFusedGemmRun verification compares GEMM1 twice; GEMM0 output is never checked Open

CUTLASS C++

Difficulty 1/5 Under an hour Newbie friendliness 92/100

NVIDIA/cutlass#3523 · Opened Aug 24, 2026
TensorNHWC::inverse device path diverges from host path when strides are not w-multiples Open

CUTLASS C++

Difficulty 2/5 1-3 hours Newbie friendliness 84/100

NVIDIA/cutlass#3519 · Opened Aug 24, 2026
SubbyteReference breaks its offset invariant for negative offsets (constructor and operator+=) Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 76/100

NVIDIA/cutlass#3518 · 1 comment · Opened Aug 24, 2026
TensorView::operator- does not compile; TensorViewPlanarComplex::reset ignores its size parameter Open

CUTLASS C++

Difficulty 2/5 1-3 hours Newbie friendliness 84/100

NVIDIA/cutlass#3517 · Opened Aug 24, 2026
Matrix<R,C>::at() strides by row count: wrong values and out-of-bounds access for all non-square shapes (diagonal/hcat/vcat affected) Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 78/100

NVIDIA/cutlass#3516 · Opened Aug 24, 2026
ArraySubbyte::back() returns the wrong element when N does not fill its last storage unit Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 76/100

NVIDIA/cutlass#3515 · Opened Aug 24, 2026
zip2_by does not handle Underscore tiler entries: kept modes are silently split or fail to compile Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 78/100

NVIDIA/cutlass#3514 · 1 comment · Opened Aug 24, 2026
zipped/tiled divide with a dynamic tiler skips divisibility checks on static layouts and returns a non-size-preserving layout Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 72/100

NVIDIA/cutlass#3513 · 1 comment · Opened Aug 24, 2026
PredicateVector::is_zero ignores the last storage word when kBytes is a multiple of sizeof(Storage) Open

CUTLASS C++

Difficulty 2/5 1-3 hours Newbie friendliness 82/100

NVIDIA/cutlass#3512 · Opened Aug 24, 2026
GEMM profiler verification compares only batch 0 when batch-count > 1 Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 55/100

NVIDIA/cutlass#3511 · Opened Aug 24, 2026
uint1b_t ReduceArray specializations are broken (logical_or never compiles / always true; logical_and tests bytes not bits) Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 78/100

NVIDIA/cutlass#3510 · Opened Aug 24, 2026
LinearCombinationClamp with float output does not compile: platform numeric_limits<float> has no lowest() Open

CUTLASS C++

Difficulty 2/5 1-3 hours Newbie friendliness 82/100

NVIDIA/cutlass#3509 · 1 comment · Opened Aug 24, 2026
numeric_limits constants wrong for float_ue8m0_t::lowest, tfloat32_t::epsilon and half_t::epsilon Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 82/100

NVIDIA/cutlass#3508 · 1 comment · Opened Aug 24, 2026
CuTeDSL: staged bool() with no arguments crashes the preprocessor with a raw IndexError Open

bug CuTe DSL

Difficulty 2/5 1-3 hours Newbie friendliness 82/100

NVIDIA/cutlass#3506 · 1 comment · 1 reaction · Opened Aug 24, 2026
Five conv filter tile access iterators invert the byte scaling in add_pointer_offset (8/sizeof_bits instead of sizeof_bits/8) Open

CUTLASS C++

Difficulty 3/5 1-2 days Newbie friendliness 74/100

NVIDIA/cutlass#3504 · 1 comment · Opened Aug 24, 2026

NVIDIA/cutlass

Get new issues in your inbox