quill-fastsort-simd 0.1.0

x86-simd-sort (Intel AVX-512) native extension for Quill

quill-fastsort-simd

Intel x86-simd-sort (AVX-512 / AVX2) native extension for QuillSort.

Wraps the same AVX-512 sort kernel that modern NumPy uses internally, exposed directly via pybind11 to skip np.sort dispatch overhead (~1-2 ms saved per call at 1M-10M element sizes) so QuillSort's self-tuning DB can measure it independently from the np.sort fallback.

Public Python API

In-place sort, GIL released:

• sort_i64(arr), sort_u64(arr)
• sort_i32(arr), sort_u32(arr)
• sort_f64(arr), sort_f32(arr)

Runtime CPU detection (CPUID) picks AVX-512 when available and falls back to the AVX2 path otherwise.

Build

export PYO3_USE_ABI3_FORWARD_COMPATIBILITY=1
python -m pip wheel . -w dist --no-deps --no-build-isolation

Vendored

src/x86-simd-sort/ — upstream https://github.com/intel/x86-simd-sort (commit fa944ef, 2026-06-25).