All pull requests

Pull requests

• Optimize FindChild for AVX2/SSE4/SSE2 since they dont have native unsigned Le(<=) instruction and is implemented as Not(Gt()) in highway where unsigned Gt() is also not a native instruction on these archs so we flip the MSB of both operands to use signed Gt().

  #3518·copybara-service[bot] opened Oct 26, 2026Bot·

  ·

• Optimize FindChild by removing the std::min() at the end which can cause stalling since num_keys is present at a different cache line.

  #3516·copybara-service[bot] opened Oct 26, 2026Bot·

  ·

• Fix LoadNOr filling the tail with the wrong lanes of no

  ready to pull
  #3515·Arthur031221 opened Oct 26, 2026Contributor·

  ·
  1

• Fix signed int64 conversion under fast math

  #3514·MokiMeow opened Oct 24, 2026·

  ·
  1

  1

• Fix float16_t and target selection in the CUDA device pass for Arm hosts.

  #3512·copybara-service[bot] opened Oct 23, 2026Bot·

  ·
  1

• Fix NaN and Inf division by zero on EMU128 and SCALAR

  #3511·reginaldalfret opened Oct 23, 2026·

  ·

• Avoid relinking more tests after unused library changes

  #3509·kennethassogba opened Oct 22, 2026Contributor·

  ·
  1

• Optimize Highway SIMD Exp/Exp2/Expm1/Sinh/Cosh/Tanh in math-inl.h:

  #3505·copybara-service[bot] opened Oct 20, 2026Bot·

  ·
  1

• Use one minpos for x86 u8, i8 and i16 lane min and max

  #3504·advitrocks9 opened Oct 20, 2026Contributor·

  ·

• G3docs: Replaced Intel intrinsics guide with a working link

  ready to pull
  #3489·githubskribble24 opened Oct 14, 2026·

  ·
  2

• base64: guard against size_t overflow in Base64EncodedSize and Base64Encode

  ready to pull
  #3472·jdymitarai opened Oct 10, 2026Contributor·

  ·
  3

• Preserve the sign of a zero in the emulated Round, Ceil, Floor and Trunc

  ready to pull
  #3447·itz-puneet opened Oct 2, 2026·

  ·
  1

  5

• Add argsort, partial argsort and argselect to contrib/sort

  ready to pull
  #3430·Wint3rNight opened Sep 28, 2026Contributor·

  ·
  43

• N/A

  #3418·copybara-service[bot] opened Sep 22, 2026Bot·

  ·

• Add ShuffleSpan to hwy/contrib/algo

  #3392·Wint3rNight opened Sep 16, 2026Contributor·

  ·
  12

• internal change for now

  #3390·copybara-service[bot] opened Sep 16, 2026Bot·

  ·
  1

• tests: add differential harness for cross-target agreement

  #3375·tvost2 opened Sep 11, 2026Contributor·

  ·

• Add hwy/contrib/multiprec: fixed-width unsigned multiply (MulAdd52)

  #3372·tvost2 opened Sep 11, 2026Contributor·

  ·
  37

• Add IEEE 754 special value (+/-Inf, +/-NaN) support and tests to Highway math functions

  #3363·copybara-service[bot] opened Sep 8, 2026Bot·

  ·

• Internal build system change.

  #3359·copybara-service[bot] opened Sep 7, 2026Bot·

  ·
  1

• Add attention kernels optimized for arm's i8mm instruction.

  #3168·copybara-service[bot] opened Jun 30, 2026Bot·

  ·

• n/a

  #3127·copybara-service[bot] opened Jun 18, 2026Bot·

  ·

• Add implementations of Reduce2/4/8 to quickly compute simultaneous sum/min/max of N registers + reusable template for those functions.

  #3115·copybara-service[bot] opened Jun 12, 2026Bot·

  ·

• Helper functions to use arm matrix multiplication instructions in a way that is testable on other platofrms

  #3101·copybara-service[bot] opened Jun 10, 2026Bot·

  ·

• No public description

  #3088·copybara-service[bot] opened Jun 7, 2026Bot·

  ·
  1