Issues

• Bug: f16 ULP tests count float32 spacing

  ashvardanian/NumKong#397 · Filter by author ternaus opened on Oct 10, 2026
  1

• Bug: NEON-half f16 FMA exceeds the precision limit

  ashvardanian/NumKong#396 · Filter by author ternaus opened on Oct 10, 2026
  2

• Performance: float16 nk.scale and nk.blend are much slower than float32 on ARM64

  ashvardanian/NumKong#395 · Filter by author ternaus opened on Oct 10, 2026
  5

• Bug: 7.8.5 PyPI release is missing CPython 3.13 Linux x86_64 wheels

  ashvardanian/NumKong#394 · Filter by author ternaus opened on Oct 10, 2026

• f32 kernels accumulate in f64: 22% fewer HNSW searches per second on Graviton4 than simsimd 6.5's f32 kernels

  ashvardanian/NumKong#392 · Filter by author swasik opened on Oct 8, 2026
  2

• Feature: Add a truncation mode to nk.astype for float-to-integer casts

  ashvardanian/NumKong#377 · Filter by author ternaus opened on Aug 13, 2026

• Feature request: CPU separable 3D linear resampling for DHWC volumes

  ashvardanian/NumKong#372 · Filter by author ternaus opened on Aug 3, 2026

• Feature: N-D lerp with broadcast weights for alpha blending

  ashvardanian/NumKong#369 · Filter by author ternaus opened on Jul 24, 2026

• Feature: N-D minimum, maximum, median, and mode neighborhood filters

  ashvardanian/NumKong#367 · Filter by author ternaus opened on Jul 24, 2026

• Feature: N-D elementary math ufuncs, including atan2 and hypot

  ashvardanian/NumKong#365 · Filter by author ternaus opened on Jul 24, 2026

• Feature: N-D comparisons, minimum, maximum, clip, and where

  ashvardanian/NumKong#366 · Filter by author ternaus opened on Jul 24, 2026

• Feature: N-D convolve1d and separable filters for box and Gaussian blur

  ashvardanian/NumKong#364 · Filter by author ternaus opened on Jul 24, 2026

• Feature: SIMD histogram and bincount for uint8 data and integer labels

  ashvardanian/NumKong#363 · Filter by author ternaus opened on Jul 24, 2026
  1

• Feature: Batched 2-D and 3-D resampling for resize, warp, and remap

  ashvardanian/NumKong#362 · Filter by author ternaus opened on Jul 24, 2026

• SwiftPM: header-only CNumKong target breaks transitive linking in Xcode 16+

  ashvardanian/NumKong#353 · Filter by author beshkenadze opened on May 7, 2026
  1

• Feature: Allow building nk_shared as a static archive (for runtime dispatch under static linkage)

  ashvardanian/NumKong#351 · Filter by author kyusic opened on Apr 28, 2026
  1

• Feature: Per-function target pragmas for LoongArch LASX kernels (blocked on GCC 15 / Clang 22 availability)

  enhancement
  ashvardanian/NumKong#348 · Filter by author ashvardanian opened on Apr 15, 2026

• Feature: ARM32 NEON Kernel Portability

  enhancement
  ashvardanian/NumKong#339 · Filter by author ashvardanian opened on Apr 5, 2026

• Feature: Poison misaligned tails in PyTest

  enhancement
  ashvardanian/NumKong#334 · Filter by author ashvardanian opened on Apr 2, 2026

• Feature: Shape-preserving N-D nk.scale with documented integer rounding

  enhancement
  ashvardanian/NumKong#327 · Filter by author ternaus opened on Mar 23, 2026

• Feature: Require "pre-packing" for symmetric GEMM-like kernels

  enhancementv8
  ashvardanian/NumKong#323 · Filter by author ashvardanian opened on Mar 22, 2026

• Feature: Split MinMax and ArgMinMax reductions

  enhancementv8
  ashvardanian/NumKong#322 · Filter by author ashvardanian opened on Mar 22, 2026

• Feature: Mixed-dtype broadcasted affine transform for normalization

  enhancement
  ashvardanian/NumKong#315 · Filter by author ternaus opened on Mar 18, 2026
  2

• Tracking: N-D numerical, image, and volume kernels requested by AlbumentationsX

  enhancement
  ashvardanian/NumKong#313 · Filter by author ternaus opened on Mar 18, 2026
  1

• Feature: Direct matmul for ephemeral and small right-hand matrices

  enhancement
  ashvardanian/NumKong#312 · Filter by author ternaus opened on Mar 18, 2026
  1