All pull requests

Pull requests

• Initialize the default XNNPACK allocator before querying packed weight

  #11840·copybara-service[bot] opened Oct 29, 2026Bot·

  ·

• Add ARM SME2 ACLE GEMM for pf16 and pf32 and LHS packing microkernels, along with unit tests and benchmarks.

  #11837·copybara-service[bot] opened Oct 28, 2026Bot·

  ·

• Return unsupported_parameter for unsupported datatypes in xnn_define_argmax_pooling_2d.

  #11836·copybara-service[bot] opened Oct 28, 2026Bot·

  ·

• Return unsupported_parameter for unsupported datatypes in xnn_define_static_resize_bilinear_2d.

  #11835·copybara-service[bot] opened Oct 28, 2026Bot·

  ·

• Return unsupported_parameter for unsupported datatypes in xnn_define_max_pooling_2d.

  #11834·copybara-service[bot] opened Oct 28, 2026Bot·

  ·

• Fix/extend per channel host dequantization to i2/i4 for the XNNPACK backend.

  #11832·copybara-service[bot] opened Oct 28, 2026Bot·

  ·

• xnn_define_static_constant_pad_v2 converted the padding value after adding the

  #11830·copybara-service[bot] opened Oct 28, 2026Bot·

  ·

• Map XNNPACK statuses to absl status codes.

  #11828·copybara-service[bot] opened Oct 28, 2026Bot·

  ·

• Validate input ranks in even-split reshape and unary resize

  #11826·professor-moody opened Oct 28, 2026·

  ·

• Fix Windows AMX IGEMM crash by restricting direct tile load to mr == 1.

  #11824·copybara-service[bot] opened Oct 28, 2026Bot·

  ·

• Extend Hexagon HVX c4-hvx.c.in GEMM and IGEMM templates to support

  #11822·copybara-service[bot] opened Oct 28, 2026Bot·

  ·

• Copybara import of the project:

  #11821·copybara-service[bot] opened Oct 28, 2026Bot·

  ·

• Fix slow random shapes and increase shard count in YNNPACK dot test

  #11819·copybara-service[bot] opened Oct 27, 2026Bot·

  ·

• Automated Code Change

  #11818·copybara-service[bot] opened Oct 27, 2026Bot·

  ·

• Only start PR CI on approvals from maintainers

  #11817·mab2500 opened Oct 27, 2026·

  ·
  2

• Add ResNet-50 v1.5 subgraph benchmark

  #11808·copybara-service[bot] opened Oct 27, 2026Bot·

  ·

• Fix Hexagon HVX FP16 subgraph fallback and optimize f16<->f32 vcvt

  #11807·copybara-service[bot] opened Oct 26, 2026Bot·

  ·

• Tweak choose_split_factors to improve performance for large dots

  #11805·copybara-service[bot] opened Oct 26, 2026Bot·

  ·

• Initialize YNNPACK delegate subgraphs in parallel on the thread pool.

  #11803·copybara-service[bot] opened Oct 26, 2026Bot·

  ·

• Fix RISC-V Vector build errors, crashes, and test bugs across Coral

  #11800·copybara-service[bot] opened Oct 25, 2026Bot·

  ·

• reject dynamic qcint8 bmm input whose channel dim is not N

  #11799·aizu-m opened Oct 25, 2026Contributor·

  ·

• Add Qualcomm Hexagon HVX c2 f16-gemm and x16-packw microkernels

  #11798·copybara-service[bot] opened Oct 25, 2026Bot·

  ·

• disable widen_fp16_accumulators rewrite

  #11797·aizu-m opened Oct 25, 2026Contributor·

  ·

• [XNNPACK] Add F16 raddstoreexpminusmax SIMD microkernels

  #11796·copybara-service[bot] opened Oct 24, 2026Bot·

  ·

• Add _zip and neondot M=1 in NEON I8MM GEMM and use vld2q_dup_u64.

  #11794·copybara-service[bot] opened Oct 23, 2026Bot·

  ·