New to this version:

• Fix x86-64 .so file
• Bring back precompiled library for aarch64
• Upgrade JUnit to 4.13.1 due to a potential vulnerability

New to this version:

• Fix computeSimilarity(v1, v2) method: the vectors are normalized if the similarity measure requires normalization (e.g. cosine similarity).
• No precompiled library for Debian Linux aarch64.

This version includes some new functions exposed from the native library:

• hasId(id): checks if an id is in the index;
• getData(id): retrieves the array for a specific id;
• computeSimilarity(v1,v2): computes the similarity on the native side taking advantage of SIMD, SS3, AVX instructions (when available);
• markDeleted(id): marks the item "id" as deleted;
• getEf(): returns the value of ef;
• getM(): returns the value of M;
• getEfConstruction: returns the value of efConstruction.

In addition, in QueryTuple, "labels" has been renamed to "ids" which will break a compatibility with previous versions.

This release includes:

• Support to aarch64 (ARM64) processors tested on AWS Graviton2 (debian 9);
• Some small changes in the dynamic library naming to better handling different architectures.

This release includes:

1. Hnswlib v0.4.0;
2. setEf() method exposed in the Index class;
3. Bug fix on loading the pre-generated DLL on Windows 10;
4. Documentation and examples updates.

Thanks @ashfaq92 for all suggestions. Thanks @yurymalkov and contributors for the new Hnswlib release.

In this release we include support to JNA v4.x.x via the hnswlib-jna-legacy. In addition, it also includes minor bug fixes.

This release includes an Index class with thread-safe add/query operations (ConcurrentIndex). It relies on ReentrantReadWriteLock with fairness enabled. Moreover, minor improvements for better stability were also included.