Issues

• Deprecate cuda::cuda_error::status() in favor of status<cudaError_t>() and raw_code()

  NVIDIA/cccl#11887 · opened on Oct 6, 2026

• Bump the libcu++ ABI version after the cuda::cuda_error layout change

  NVIDIA/cccl#11886 · opened on Oct 6, 2026

• [BUG]: cuda-cccl 1.2.1 on Windows + CTK 12.4: NVVM link error, __half(__nv_bfloat16) multiply defined

  Bug

  NVIDIA/cccl#11885 · opened on Oct 6, 2026

• [clang-tidy] Check that internal headers include cuda/std/__meow/foo.h instead of cuda/std/meow

  NVIDIA/cccl#11868 · opened on Oct 5, 2026

• [spike]: Speed-of-Light Fancy Iterators

  NVIDIA/cccl#11840 · opened on Oct 3, 2026

• [BUG]: DeviceBatchedTopK deterministic tie-break ranks -0.0 below +0.0

  NVIDIA/cccl#11834 · opened on Oct 3, 2026
  1

• [BUG]: CUB env APIs silently ignore stream types whose conversion to cudaStream_t is not const-qualified

  NVIDIA/cccl#11824 · opened on Oct 2, 2026

• Make SASS diff CI job blocking, with commit-pinned maintainer override

  NVIDIA/cccl#11821 · opened on Oct 2, 2026

• [BUG]: reduce_by_keys with __half

  Bug

  NVIDIA/cccl#11816 · opened on Oct 2, 2026
  2

• Range histogram benchmarks generate zero-width bins for narrow integer samples

  NVIDIA/cccl#11814 · opened on Oct 2, 2026

• cub::(Multi)RangeHistogram I8 significantly regresses after negative bins bug fix

  NVIDIA/cccl#11809 · opened on Oct 2, 2026

• Replace topk inline ptx atomics with cluster scope atomic_ref

  NVIDIA/cccl#11803 · opened on Oct 1, 2026

• [FEA]: honor/exploit allocator hints for "locality" optimizations

  NVIDIA/cccl#11778 · opened on Oct 1, 2026
  1

• Unpin breathe<5 once breathe-doc/breathe#1085 is released

  NVIDIA/cccl#11777 · opened on Sep 30, 2026

• [BUG]: cub::DeviceTransform fill with a large power-of-two value type reserves gigabytes of local memory

  NVIDIA/cccl#11758 · opened on Sep 29, 2026

• [BUG] Tuning search might attribute a stale variant binary's timings to a new variant

  NVIDIA/cccl#11745 · opened on Sep 29, 2026

• Extend per-call overhead fixes to scan, segmented reduce, and segmented sort

  NVIDIA/cccl#11734 · opened on Sep 29, 2026

• Cache the validated stream handle by identity

  NVIDIA/cccl#11733 · opened on Sep 29, 2026
  1

• Add an explicit-invalidation call path for reduce to skip redundant per-call work

  NVIDIA/cccl#11732 · opened on Sep 29, 2026

• Improve and reenable compile time comparison results on PRs

  NVIDIA/cccl#11731 · opened on Sep 29, 2026

• Present on new group design

  NVIDIA/cccl#11729 · opened on Sep 29, 2026

• DeviceFor: migrate two-phase overloads to Env (multi-dimensional)

  NVIDIA/cccl#11714 · opened on Sep 29, 2026

• DeviceFor: migrate two-phase overloads to Env (core)

  NVIDIA/cccl#11713 · opened on Sep 29, 2026

• DeviceSegmentedSort: migrate two-phase overloads to Env (pairs)

  NVIDIA/cccl#11712 · opened on Sep 29, 2026

• DeviceSegmentedSort: migrate two-phase overloads to Env (keys)

  NVIDIA/cccl#11711 · opened on Sep 29, 2026