All pull requests

Pull requests

• [cudax] Remove exhaustivity from mappings

  #12159·davebayer opened Oct 22, 2026Contributor·

  ·
  1

  4

• [libcu++] Optimize cuda::bit_fns

  #12157·davebayer opened Oct 22, 2026Contributor·

  ·
  7

• [CUB] Make DeviceScan implementation helpers private

  #12156·bernhardmgruber opened Oct 22, 2026Contributor·

  ·
  2

• [CUB] Only use compiled compute capabilities in the radix_sort_runs_on tests

  #12155·SUKUNA-AI opened Oct 22, 2026·2 tasks done·

  ·
  1

  2

• [libcu++][fp] Use cuda::bit_trunc

  #12153·davebayer opened Oct 22, 2026Contributor·

  ·
  1

• [libcu++] cuda::std::simd Fix Masked Reduction behavior

  #12152·fbusato opened Oct 22, 2026Contributor·

  ·
  4

  

• Switch the disabled shared_memory_barrier tests to c2h

  backport branch/3.6.x
  #12150·pciolkosz opened Oct 21, 2026Contributor·

  ·
  4

• Restore Thrust reduce_by_key load modifier to LOAD_DEFAULT

  backport branch/3.6.x
  #12148·bernhardmgruber opened Oct 21, 2026Contributor·

  ·
  2

• Apply cub::DeviceReduce::ReduceByKey tunings to all key, accumulator and operator types

  #12147·bernhardmgruber opened Oct 21, 2026Contributor·

  ·
  11

• [CUB] Don't vectorize DeviceTransform ublkcp stores for non-trivial output assignment

  #12146·bernhardmgruber opened Oct 21, 2026Contributor·

  ·
  6

• Flag all implicit misplaced widening casts as well

  #12145·Jacobfaib opened Oct 21, 2026Contributor·2 tasks·

  ·
  5

  

• Relax static assertion in cuda::std::is_sufficiently_aligned

  #12144·bernhardmgruber opened Oct 21, 2026Contributor·

  ·
  5

• [docs][fp] Add the FP optimization guide

  #12143·akolesov-nvidia opened Oct 21, 2026Contributor·1 of 2 tasks·

  ·
  2

• [cuda.compute] Make ReverseIterator + n move backward like the device advance

  #12142·v0ropaev opened Oct 21, 2026·2 tasks done·

  ·
  1

  2

• [cub] Use cuda::bit_trunc

  #12140·davebayer opened Oct 21, 2026Contributor·

  ·
  3

• Add terminal TempStorage.release() for CUTLASS

  #12139·tpn opened Oct 21, 2026Contributor·

  ·
  1

• [libcu++] Keep launch_transform storage alive until after the kernel launch on MSVC

  #12138·SUKUNA-AI opened Oct 21, 2026·2 tasks done·

  ·
  5

• Fix size overflow in vector_base::fill_insert and copy_insert

  #12137·SUKUNA-AI opened Oct 21, 2026·2 tasks done·

  ·
  1

  7

• Add terminal TempStorage.release() for Numba-CUDA-MLIR

  #12135·tpn opened Oct 21, 2026Contributor·

  ·
  5

• [cudax] Do not read managed memory from the host after starting work in the stream adaptor

  #12134·SUKUNA-AI opened Oct 21, 2026·2 tasks done·

  ·
  3

• [Rust] Add Block Adjacent Difference

  #12132·tpn opened Oct 21, 2026Contributor·2 tasks done·

  ·
  1

• [libcu++] Skip fabric status decoding when a shared_memory_barrier wait has no report

  backport branch/3.6.x
  #12131·SUKUNA-AI opened Oct 21, 2026·2 tasks done·

  ·
  1

  8

• [cuda.compute] Key cached operators by literal type, not just value

  #12129·v0ropaev opened Oct 21, 2026·2 tasks done·

  ·
  1

  2

• Fix BLOCK_LOAD/STORE_VECTORIZE docs for non-pointer contiguous iterators

  #12127·bernhardmgruber opened Oct 21, 2026Contributor·

  ·
  6

• [Rust] Add Block Shuffle with partial result validity

  #12126·tpn opened Oct 21, 2026Contributor·2 tasks done·

  ·
  1