To separate code from external data, we are providing statically-built binaries for Dashing2.
These are kept in osx/ and linux/ directories, within which they are grouped by release.
For each of these, there are versions for SSE2, AVX, AVX2, AVX512F, and AVX512BW.
For Intel CPUs, simply select the version compiled for your hardware.
This is commonly AVX2 _savx2, and almost certainly SSE2 _s128.
Recent hardware with AVX512F and AVX512, such as Knight's Landing, Sky Lake, or Coffee Lake, use newer vectorized operations where available.
To fully benefit from these, you would use dashing2_s512 (for KNL) or dashing2_s512bw (for Skylake/CoffeeLake).
We will continue to update this with a list of statically-compiled binaries.
v2.0.4 - First 2.0 version. OSX + Linux. v2.1.0 - First 2.1 version. OSX + Linux v2.1.1 - OSX + Linux v2.1.2 - Bug fix release, skipped in favor of 2.1.3 v2.1.3 - OSX + Linux, improved memory management v2.1.5 - OSX + Linux. Faster formatting, better hashing, and differently-parallelized linear-time clustering. v2.1.6 - OSX + Linux. Corrections/speedups to KNN/similarity-thresholded graph generation. v2.1.7 - OSX + Linux. Add fasta-output for --parse-by-seq deduplication (--greedy). v2.1.8 - OSX + Linux. Add support for direct sketching in SetSketch with pre-selected parameters. (via --setsketch-ab <a,b> --fastcmp <nb>, or --fastcmp-{bytes,shorts,words,nibbles}) v2.1.13 - OSX + Linux. Bug fixes and update hash map. v2.1.14 - OSX + Linux. Bug fixes! v2.1.15 - OSX + Linux. Bug fixes! v2.1.16 - OSX + Linux. Bug fixes, small clean ups. Bump to c++20. v2.1.17 - OSX + Linux. Bug fix: --parse-by-seq with packed truncated registers. v2.1.18 - OSX up; Skipping linux upload. Skip this release. v2.1.19 - OSX + Linux.
Warning: v2.1.14 has a known performance regression with erroneous logging. We highly recommend avoiding it. v2.1.15 and later do not have this issue. v2.1.18 has an error in --parse-by-seq mode due to the new memory-reducing implementation. This is fixed in v2.1.19.