All pull requests

Pull requests

• [Vulkan] Clear the adstack sizer overflow word

  #972·hughperkins opened Oct 14, 2026Collaborator·3 tasks·

  ·
  11

• [Vulkan] Disable MoltenVK placement heaps by default on Apple

  #971·hughperkins opened Oct 13, 2026Collaborator·

  ·
  10

• [CI] Run Mac Vulkan tests in process batches to speedup CI by orders of magnitude

  #969·hughperkins opened Oct 12, 2026Collaborator·

  ·
  48

• [Linalg] Add buffer-backed CSR matrix multiplication in kernels

  awaiting-contributor-action
  #967·btgaskin opened Oct 9, 2026Contributor·

  ·
  41

• [Graph] Allow graph kernels to take ndarrays that own a gradient

  awaiting-contributor-action
  #953·arcusbuilds opened Oct 1, 2026·

  ·
  1

  8

• [Graph] CUDA: upload per-launch graph args with async HtoD on the launch stream

  awaiting-contributor-action
  #950·arcusbuilds opened Sep 30, 2026·

  ·
  15

• [Tensor] Support layout permutations for vector and matrix tensors

  #948·hughperkins opened Sep 30, 2026Collaborator·

  ·
  1

  4

• [Graph] Support checkpoints containing graph do-while loops

  #944·alanray-tech opened Sep 28, 2026Collaborator·

  ·
  1

  20

• [Perf] Speed up kernel and func decoration by skipping the sys.modules scan in class detection

  #940·Kashu7100 opened Sep 28, 2026·

  ·
  2

• [CPU] Configure scheduling and expose block indices

  #933·hughperkins opened Sep 26, 2026Collaborator·

  ·
  1

  39

• [Build] Use mimalloc for C++ operator new/delete on Windows

  #927·hughperkins opened Sep 24, 2026Collaborator·

  ·
  3

• [Perf] Guard operand-only IR walks and avoid CSE block copies

  #922·hughperkins opened Sep 23, 2026Collaborator·

  ·
  5

• [Caching] Fast-cache host-side Vector / Matrix values

  awaiting-contributor-action
  #911·tlequy-qc opened Sep 16, 2026·

  ·
  19

• [SPIRV] Negate unsigned integers rather than rejecting them

  awaiting-contributor-action
  #910·tlequy-qc opened Sep 16, 2026·

  ·
  13

• [Interop] Prototype Vulkan opaque-FD import for CUDA and AMDGPU

  #904·rglebovamd opened Sep 3, 2026·

  ·
  14

• [Caching] Ref 7: CPU per-task artifact cache (ORC object serialize/load)

  #901·hughperkins opened Sep 1, 2026Collaborator·3 of 4 tasks·

  ·
  69

• [AMDGPU] Non-terminating in-kernel assert + OOB index clamp (arch-independent)

  awaiting-contributor-action
  #898·paveltc opened Aug 31, 2026Contributor·

  ·
  6

• [DONOTMERGE] [AMDGPU] check_out_of_bound assert-state fix (parallel to #871)

  awaiting review passawaiting-ci-to-passneed-genesis-benchmarksneed-genesis-unit-tests
  #895·hughperkins opened Aug 28, 2026Collaborator·

  ·
  21

• [AMDGPU] Surface QuadrantsAssertionError after in-kernel assert (fix barrier hang)

  awaiting-contributor-action
  #871·paveltc opened Aug 14, 2026Contributor·

  ·
  22

• [Lang] Cut per-compilation Python overhead (sys.modules scan + per-node source formatting)

  #804·duburcqa opened Jul 21, 2026Contributor·

  ·
  13

• [AMDGPU] Add DPP row_shl fast path for shuffle_down reduction offsets

  awaiting-contributor-action
  #770·ZJLi2013 opened Jul 9, 2026Contributor·

  ·
  4

• [AMDGPU] Fix cross-half shuffle select for top-half lanes on wave64

  awaiting-contributor-action
  #769·ZJLi2013 opened Jul 9, 2026Contributor·

  ·
  5

• [Doc] Remove checkpoint fixme

  #755·hughperkins opened Jun 23, 2026Collaborator·

  ·
  4

• [Docs] Clarify top-level for-loops inside graph_do_while

  #727·hughperkins opened Jun 6, 2026Collaborator·

  ·
  2

• [Types] Support inheritance for @qd.dataclass

  #717·hughperkins opened Jun 2, 2026Collaborator·

  ·
  4