All pull requests

Pull requests

• fix: align weekly-test operator ids with aten names and restore isneginf_out/signbit_out perf

  benchmarksize/Mediumtests
  #7233·kkkwb opened Oct 16, 2026Contributor·3 tasks·

  ·

• [kunlunxin] Add flat 1-D kernel for _thnn_fused_gru_cell (aggregate 0.4217 -> 0.8373)

  size/Largevendor/Kunlunxin
  #7232·llaboon opened Oct 16, 2026Contributor·

  ·

• [KernelGen][Nvidia] Add copy_sparse_to_sparse_ operator with Triton kernel

  benchmarkcoreKernelGenops/atensize/Mediumtests
  #7231·bwbwzzz opened Oct 16, 2026Contributor·

  ·

• [KernelGen][Nvidia] Add _reshape_from_tensor operator with Triton kernel

  benchmarkcoreKernelGenops/atensize/Largetests
  #7230·bwbwzzz opened Oct 16, 2026Contributor·

  ·

• [KernelGen][Nvidia] Add flatten_dense_tensors operator with Triton kernel

  benchmarkcoreKernelGenops/atensize/Largetests
  #7229·bwbwzzz opened Oct 16, 2026Contributor·

  ·

• [KernelGen][Nvidia] Add _new_zeros_with_same_feature_meta operator with Triton kernel

  benchmarkcoreKernelGenops/atensize/Mediumtests
  #7228·bwbwzzz opened Oct 16, 2026Contributor·

  ·

• [Tests] Skip unsupported Ascend reference cases in generated suites

  size/XLargetests
  #7227·lukatao opened Oct 16, 2026Collaborator·1 of 2 tasks·

  ·

• [KernelGen][Nvidia] Add _make_dual_copy operator with Triton kernel

  benchmarkcoreKernelGenops/atensize/Mediumtests
  #7226·Yukun-Cui opened Oct 16, 2026Contributor·

  ·

• [KernelGen][Nvidia] Add _fw_primal_copy operator with Triton kernel

  benchmarkcoreKernelGenops/atensize/Mediumtests
  #7225·Yukun-Cui opened Oct 16, 2026Contributor·

  ·

• [KernelGen][Nvidia] Add _values_copy operator with Triton kernel

  benchmarkcoreKernelGenops/atensize/Mediumtests
  #7224·Yukun-Cui opened Oct 16, 2026Contributor·

  ·

• [KernelGen][Nvidia] Add _indices_copy operator with Triton kernel

  benchmarkcoreKernelGenops/atensize/Mediumtests
  #7223·Yukun-Cui opened Oct 16, 2026Contributor·

  ·

• [KernelGen][Nvidia] Add indices_copy operator with Triton kernel

  benchmarkcoreKernelGenops/atensize/Mediumtests
  #7222·Yukun-Cui opened Oct 16, 2026Contributor·

  ·

• [Iluvatar] optimize conv2d: cache packed weight, fused channels_last+…

  size/Mediumvendor/Iluvatar
  #7220·ph0375 opened Oct 16, 2026Contributor·

  ·

• [Benchmark] Add --base-data for external NVIDIA baseline comparison

  benchmarkdocumentationsize/XLargetests
  #7219·lukatao opened Oct 16, 2026Collaborator·2 of 3 tasks·

  ·

• [KernelGen][Nvidia] Add _copy_from operator with Triton kernel

  benchmarkcoreKernelGenops/atensize/Mediumtests
  #7216·bwbwzzz opened Oct 16, 2026Contributor·

  ·

• perf(kunlunxin): cache the 1D-tile geometry across pointwise launches

  size/Smallvendor/Kunlunxin
  #7212·llaboon opened Oct 16, 2026Contributor·

  ·
  1

• [KMCompiler][Thead] Fix GRU barrier deadlock on T-Head backend

  size/XLargevendor/THead
  #7203·jhvghjjn opened Oct 16, 2026Contributor·3 tasks·

  ·

• [Kunlunxin] Optimize scatter_reduce with XTDK O(N) native kernel

  size/Smallvendor/Kunlunxin
  #7201·6nz5yttj5z-eng opened Oct 16, 2026Contributor·3 tasks·

  ·

• fix(kunlunxin): sgn/sgn_ large-numel path — cheaper body + configurable 1D-tile grid

  size/Smallvendor/Kunlunxin
  #7200·llaboon opened Oct 16, 2026Contributor·

  ·
  1

• cleanup: remove first batch of mis-ported CPU operators

  benchmarkcoreops/atensize/XLargetestsvendor/Cambriconvendor/Enflamevendor/Hygonvendor/Iluvatarvendor/Kunlunxinvendor/MetaXvendor/MooreThreadsvendor/THeadvendor/TsingMicro
  #7199·qcxl99 opened Oct 16, 2026·

  ·
  1

• tensorview 9.1.0 -beta3

  ops/fusedsize/XLargevendor/Ascend
  #7198·jia-heng opened Oct 16, 2026Collaborator·3 tasks·

  ·

• [Ascend] Reduce pack_seq host overhead for small batches

  benchmarkops/fusedsize/Smalltests
  #7197·macqueen09 opened Oct 15, 2026·1 of 3 tasks·

  ·
  1

• [KunlunXin][XPU3] w8a8_block_fp8_matmul: FP8 capability gate for the accuracy cases (no marker claim)

  size/XSmalltests
  #7190·llaboon opened Oct 15, 2026Contributor·

  ·

• build(deps): bump the actions-minor group across 1 directory with 4 updates

  ci/cddependenciesgithub_actionssize/XSmall
  #7186·dependabot[bot] opened Oct 15, 2026Bot·

  ·

• Fix：add unique_dim_consecutive operator and fix weekly-test failures

  benchmarkcoreops/atensize/Largetests
  #7185·kkkwb opened Oct 15, 2026Contributor·3 tasks·

  ·