All pull requests

Pull requests

• perf: stop re-stat'ing the assembler on every launch-site miss

  #307·elibol opened Sep 23, 2026·

  ·

• perf: cut the per-launch host cost of access tracking on the await path

  #306·elibol opened Sep 22, 2026·

  ·

• cuda-async: non-blocking drop for in-flight device futures (reaper)

  #305·elibol opened Sep 22, 2026·

  ·
  1

• Add bytecode reader. Allows cutile-rs to run existing cuda-tile kernels

  #300·ivarflakstad opened Sep 21, 2026·

  ·
  1

• fix: FP8 tests are expected to fail on ampere+ada

  #299·lukeyeager opened Sep 19, 2026·

  ·
  1

• reactor: adaptive scanner spin budget

  #296·0z5a opened Sep 18, 2026·

  ·
  1

• bench(jit): measure cross-process cold-cache compilation stampedes

  #294·0z5a opened Sep 18, 2026·

  ·
  2

• fix(compiler): preserve control dependence when placing bounds checks

  #293·0z5a opened Sep 18, 2026·

  ·
  1

• fix(compiler): validate optimization hint values before backend compilation

  #292·0z5a opened Sep 18, 2026·

  ·
  2

• Cutile frontend

  #287·ivarflakstad opened Sep 15, 2026·

  ·
  1

• Extract cutile-macro concepts into cutile-expand and cutile-syn-utils crates

  #286·ivarflakstad opened Sep 15, 2026·

  ·
  1

• fix: round f8 constants to nearest even

  #283·arcusbuilds opened Sep 14, 2026·

  ·
  1

  2

• feat(cutile): add reduce_xor, reduce_and, and reduce_or bitwise reductions

  #280·emersonbusson opened Sep 13, 2026·

  ·
  1

• feat(cuda-core): add zero-copy PinnedHostMapping and DeviceAllocation bridge

  #279·emersonbusson opened Sep 13, 2026·

  ·
  1

• fix(cutile): synchronize stream in to_host_vec before dropping owned tensor

  #278·emersonbusson opened Sep 13, 2026·

  ·
  1

  1

• fix: lifetime-brand CudaGraph so captured buffers cannot dangle or race

  #270·elibol opened Sep 7, 2026·

  ·

• fix: round f8 constants to nearest even

  #262·w3lld1 opened Sep 3, 2026·

  ·
  1

  1

• fix(simt): soundness and correctness fixes for the cuda-oxide bridge and VMM

  #257·elibol opened Sep 2, 2026·

  ·
  1

• feat: expose CUDA device attributes

  #255·honeyspoon opened Sep 1, 2026·

  ·
  1

• feat: add CUDA graph capture modes

  #254·honeyspoon opened Sep 1, 2026·

  ·
  1

• feat: model full debug and add TileIR JIT overrides

  #237·almightychang opened Aug 23, 2026·

  ·
  4

• feat(cutile): legalize mixed FP8xFP4 mmaf_scaled for Tile IR 13.3

  #230·lucifer1004 opened Aug 20, 2026·

  ·
  1

• feat(cutile): add raw-offset memory ops with generic forwarding support

  #229·lucifer1004 opened Aug 20, 2026·

  ·
  1

• feat(launch): support programmatic dependency ordering

  #228·lucifer1004 opened Aug 20, 2026·

  ·
  1

• feat(cutile): support scalar specialization overrides

  #227·lucifer1004 opened Aug 20, 2026·

  ·
  1