All pull requests

Pull requests

• docs(cuda-core): locate the NVLS all-reduce example

  cutile-rs
  #1806·elibol opened Nov 13, 2026Collaborator·2 tasks done·

  ·
  1

• fix(benchmarks): use checkout dependencies and update Tokio

  cutile-rs
  #1805·elibol opened Nov 13, 2026Collaborator·2 tasks done·

  ·
  1

• fix(cutile-ir): serialize omitted overflow modes as none

  cutile-rs
  #1804·elibol opened Nov 13, 2026Collaborator·2 tasks done·

  ·
  1

• feat(cuda-core): add method for custom mem pool

  #1803·sephlor opened Nov 13, 2026·7 tasks done·

  ·
  1

  1

• fix(cuda-macros,mir-importer): define host stubs for #[device] extern fns

  #1801·physwkim opened Nov 13, 2026·

  ·
  1

  1

• docs(cutile): clarify versioned API compatibility

  #1799·elibol opened Nov 12, 2026Collaborator·

  ·

• fix(cutile-examples): add explicit B200 profiles and safe QKV tail loads

  #1798·elibol opened Nov 12, 2026Collaborator·3 tasks done·

  ·

• feat(cuda-device): add Bf16 and Bf16x2 value types

  #1797·sypsyp97 opened Nov 12, 2026·6 tasks done·

  ·
  1

  1

• feat(cuda-core): batched async copies with cuMemcpyBatchAsync

  cutile-rs
  #1796·YibaiMeng opened Nov 12, 2026·4 of 6 tasks·

  ·
  1

• feat(cuda-core): event record and stream wait flags

  cutile-rs
  #1795·YibaiMeng opened Nov 12, 2026·4 of 6 tasks·

  ·
  1

• feat(cuda-core): locality-domain VMM allocations

  cutile-rs
  #1794·YibaiMeng opened Nov 12, 2026·4 of 6 tasks·

  ·
  1

• feat(cuda-device): add math::sin_cos without local memory

  #1793·ssoj13 opened Nov 12, 2026·4 tasks done·

  ·
  1

• perf(cuda-async): pre-reserve AccessLease capacity in Submission::new

  cutile-rs
  #1792·ssoj13 opened Nov 12, 2026·3 tasks done·

  ·
  1

• fix(mir-importer,codegen): serialize concurrent writes of pipeline outputs

  #1791·ssoj13 opened Nov 12, 2026·3 tasks done·

  ·
  1

• fix(cuda-core): make no driver call on the retirement thread after exit

  cutile-rs
  #1788·physwkim opened Nov 12, 2026·

  ·
  1

  1

• Fix cuda-core allocation lease scaling

  cutile-rs
  #1787·SycamoreQ opened Nov 12, 2026·2 of 6 tasks·

  ·
  1

  1

• test: replace timing-based CI assertions with explicit synchronization

  cutile-rs
  #1782·elibol opened Nov 11, 2026Collaborator·

  ·
  1

• Report a clear error when preparing an uncontracted kernel

  #1771·efegokdemir opened Nov 10, 2026Contributor·3 of 6 tasks·

  ·
  1

• fix(cuda-oxide-codegen): accept sm_103a for Blackwell family features

  #1765·0z5a opened Nov 9, 2026Contributor·

  ·
  1

  1

• ci: build both cargo oxide new templates against the checkout

  #1764·vyncint opened Nov 9, 2026Contributor·

  ·
  1

  1

• ci: compile the cuda_* imports in markdown

  cutile-rs
  #1762·vyncint opened Nov 9, 2026Contributor·

  ·
  1

  1

• fix(cuda-async): preserve device state on duplicate registration

  cutile-rs
  #1754·fallintoplace opened Nov 8, 2026Contributor·

  ·
  2

• feat(cutile): pointer operands for foreign tile-library calls

  cutile-rs
  #1749·nvlcambier opened Nov 5, 2026Collaborator·

  ·
  6

• docs(cutile): demonstrate recorded autotune dispatch

  cutile-rs
  #1740·shuqi2077 opened Nov 4, 2026·

  ·
  1

• test(cuda-core): gate stream_query on observed state instead of elapsed time

  cutile-rs
  #1729·vyncint opened Nov 4, 2026Contributor·

  ·
  1