All pull requests

Pull requests

• [XPU] SYCL-TLA grouped GEMM for torch._grouped_mm, with a standalone tuner

  disable_distributeddisable_e2e
  #5923·githubsgi opened Oct 29, 2026·

  ·

• Enable testing test_sort_and_select_xpu.py after upstream test class name change.

  disable_autodisable_distributeddisable_e2edisable_win
  #5917·BBBela opened Oct 28, 2026Contributor·

  ·
  1

  

• Sync test_cpp_warnings_have_python_context with upstream jit warning filters

  disable_builddisable_distributeddisable_e2edisable_win
  #5915·PatrykWilczewski opened Oct 28, 2026Contributor·

  ·
  1

  1

• Fixed reduction issue #5858, test moved to upstream

  #5914·mwiktor-intel opened Oct 28, 2026Contributor·

  ·
  1

  3

• Return early in split_with_sizes_copy for empty inputs

  #5912·DamJanusz opened Oct 28, 2026Contributor·

  ·
  1

  1

• Align test_profiler_xpu with changes in upstream.

  #5911·kdrozd-dev opened Oct 28, 2026Contributor·

  ·
  2

• Vectorize and optimize the XPU distribution nullary kernel

  #5910·jianyizh opened Oct 28, 2026Contributor·

  ·

• [CI] Support xpu-smi 1.x and 2.x in GPU memory monitor

  disable_autodisable_builddisable_distributeddisable_ut
  #5904·mengfei25 opened Oct 28, 2026Contributor·

  ·
  1

• [Nightly] Add the microbench nightly regression detection

  disable_distributeddisable_e2edisable_win
  #5899·RUIJIEZHONG66166 opened Oct 28, 2026Contributor·

  ·

• Optimize foreach OP by reducing data transfer times

  #5894·orrangetabby17 opened Oct 28, 2026Contributor·

  ·
  3

• [DO NOT MERGE] try build fix for dpclang

  #5889·dvrogozh opened Oct 27, 2026Contributor·

  ·

• Re-enable test_broadcast_fn_map(2)_xpu in test_torch_xpu.py

  disable_distributeddisable_e2e
  #5888·jkosnox opened Oct 27, 2026Contributor·

  ·
  1

  2

• Optimize foreach occupacy by adjusting chunk sizing

  disable_distributed
  #5877·orrangetabby17 opened Oct 26, 2026Contributor·

  ·
  2

• [xccl] Make symmetric-memory barrier/signal timeouts fire on XPU

  sycl: experimentalsycl: new extension
  #5876·jkosnox opened Oct 26, 2026Contributor·

  ·
  4

• [UT] Skip tests in test/test_determination.py

  #5875·dominikjastrzebskix opened Oct 26, 2026Contributor·

  ·

• Skip test_passed_swizzle_arrays on XPU as upstream does

  disable_builddisable_distributeddisable_e2edisable_win
  #5874·PatrykWilczewski opened Oct 26, 2026Contributor·

  ·
  1

• [XPU] Optimize FP16 max_pool2d forward and backward on BMG

  #5873·yucai-pro opened Oct 26, 2026Contributor·

  ·
  6

• Use 64-bit sizes in XPU spatial softmax forward/backward

  disable_distributeddisable_e2e
  #5871·DamJanusz opened Oct 26, 2026Contributor·

  ·
  1

  3

• Enable LRU cache for XPU fft family operators

  disable_distributeddisable_e2e
  #5869·majing921201 opened Oct 26, 2026Contributor·

  ·
  1

• Update E2E model tiers P0/P1/P2

  disable_acceleratedisable_builddisable_distributeddisable_transformersdisable_utdisable_windisable_windows_ut
  #5868·mengfei25 opened Oct 26, 2026Contributor·

  ·
  5

• [XPU] Optimize FP16 upsample_nearest2d forward and backward on BMG

  disable_distributeddisable_e2e
  #5866·yucai-pro opened Oct 26, 2026Contributor·

  ·
  6

• [XPU] Optimize FP16 avg_pool2d backward on BMG

  disable_distributeddisable_e2e
  #5865·yucai-pro opened Oct 26, 2026Contributor·

  ·
  1

• ci: fix P0 UT deselect paths and collection errors

  #5856·libohao1201 opened Oct 25, 2026Contributor·

  ·

• test_dynamo_set_recursion_limit: skip test on Windows

  #5852·BartoszKokoszko opened Oct 23, 2026Contributor·

  ·
  1

  5

• Reduce: round kernel max work-group size down to a power of 2

  disable_distributeddisable_e2ehw: PTLos: Linuxos: Windowsos: WSL
  #5851·PatrykWilczewski opened Oct 23, 2026Contributor·

  ·
  1