All pull requests

Pull requests

• Copybara import of the project:

  #11404·copybara-service[bot] opened Sep 25, 2026Bot·

  ·

• [operators] Guard convolution-nchw dimension arithmetic overflow

  #11403·anagnorisis2peripeteia opened Sep 25, 2026Contributor·

  ·

• [batch-matrix-multiply] Guard residual stride rounding

  #11402·anagnorisis2peripeteia opened Sep 25, 2026Contributor·

  ·

• [fully-connected] Guard residual GEMM stride rounding

  #11401·anagnorisis2peripeteia opened Sep 25, 2026Contributor·

  ·

• [packq] Reject qpint8 packed-size overflow

  #11400·anagnorisis2peripeteia opened Sep 25, 2026Contributor·

  ·

• Fold ynnpack/kernels/dot:pack into ynnpack/kernels/dot:dot

  #11399·copybara-service[bot] opened Sep 25, 2026Bot·

  ·

• [configs] Harden GEMM mr/nr dynamic bounds checks against out-of-bounds array writes

  #11393·anagnorisis2peripeteia opened Sep 25, 2026Contributor·

  ·

• Copybara import of the project:

  #11387·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  1

• Fix WASM test/benchmark macro guards in xnncommon.py and regenerate NEONv8 vcvt microkernels.

  #11369·copybara-service[bot] opened Sep 25, 2026Bot·

  ·

• [configs] Harden hardware config queries and guard vector configs against NULL/zero vlenb

  #11366·anagnorisis2peripeteia opened Sep 24, 2026Contributor·

  ·

• [reference] Harden integer elementwise math against overflow and hardware traps

  #11364·anagnorisis2peripeteia opened Sep 24, 2026Contributor·

  ·

• [enums] Return '<unknown>' for unrecognized enum values instead of triggering XNN_UNREACHABLE

  #11363·anagnorisis2peripeteia opened Sep 24, 2026Contributor·

  ·

• [operators] Harden fully-connected-nc static bias conversion against buffer overflow

  #11362·anagnorisis2peripeteia opened Sep 24, 2026Contributor·

  ·

• Copybara import of the project:

  #11361·copybara-service[bot] opened Sep 24, 2026Bot·

  ·

• Copybara import of the project:

  #11360·copybara-service[bot] opened Sep 24, 2026Bot·

  ·

• Use vshll_n_u16 and XNN_INTRINSIC in NEON BF16 unary microkernels.

  #11355·copybara-service[bot] opened Sep 24, 2026Bot·

  ·

• Optimize LiteRT Tensor API WebGPU backend GPU performance and add ML Drift head-to-head benchmark

  #11354·copybara-service[bot] opened Sep 24, 2026Bot·

  ·

• Implement WebGPU HF tiled MatMul (FullyConnected, BatchMatMul) and reduction operations for LiteRT Tensor API.

  #11353·copybara-service[bot] opened Sep 24, 2026Bot·

  ·

• Implement WebGPU unary, activation, cast, and shape-aliasing operations for LiteRT Tensor API.

  #11352·copybara-service[bot] opened Sep 24, 2026Bot·

  ·

• Implement remaining WebGPU operations for 100% arithmetic_graph.h coverage

  #11351·copybara-service[bot] opened Sep 24, 2026Bot·

  ·

• Implement WebGPU shape/indexing, spatial conv/pool, and LSTM operations and enable GPU NumericalTestSuite.

  #11350·copybara-service[bot] opened Sep 24, 2026Bot·

  ·

• Implement WebGPU binary, comparison, logical, and ternary (Select/SelectV2) operations for LiteRT Tensor API.

  #11349·copybara-service[bot] opened Sep 24, 2026Bot·

  ·

• [microparams] Harden scale initialization against invalid arguments

  #11346·anagnorisis2peripeteia opened Sep 24, 2026Contributor·

  ·

• [mutex] Harden xnn_mutex functions against NULL pointers and double release

  #11345·anagnorisis2peripeteia opened Sep 24, 2026Contributor·

  ·

• [pack-lh] Harden size, offset, and packing routines against divide-by-zero and null pointers

  #11344·anagnorisis2peripeteia opened Sep 24, 2026Contributor·

  ·