This PR Implement vector distance using Google's highway library, which supports SIMD on ARM. Fixes https://github.com/1yefuwang1/vectorlite/issues/13.
Besides, it would be lot easier to add float16, bfloat16, int8 support in the future.
Based on the benchmark on my PC(i5-12600KF with AVX2 support),
this implementation is 1.5x-3x faster than HNSWLIB's SIMD implementation
when dealing with vectors with 256 elements or more.
This PR Implement vector distance using Google's highway library, which supports SIMD on ARM. Fixes https://github.com/1yefuwang1/vectorlite/issues/13. Besides, it would be lot easier to add float16, bfloat16, int8 support in the future.
Based on the benchmark on my PC(i5-12600KF with AVX2 support), this implementation is 1.5x-3x faster than HNSWLIB's SIMD implementation when dealing with vectors with 256 elements or more.