Improve subsampling support

What's Changed

• Add support for subsampling without replacement by @sfiligoi in #43
• Make subsampling with replacement multi-threaded by @sfiligoi in #44
• Clean up the build process by @emollier in #45 and by @mr-c in #46

Full Changelog: v1.3.2...v1.4

Auto-detect CPU capabilities at runtime in Linux

What's Changed

• Tune the build procedure, use AVX2 for performance build by @sfiligoi in #41
• Auto-detect CPU capabilities at runtime in Linux by @sfiligoi in #42

Full Changelog: v1.3.1...v1.3.2

Fix memory leak and improve build procedure

What's Changed

• Fix memory leak by @sfiligoi in #39
• Add option to disable AVX optimization by @sfiligoi in #37 and #40
• Add one_off_matrix_inmem_v2 by @sfiligoi in #38

Full Changelog: v1.3...v1.3.1

Add support for PERMANOVA, subsampling and the multi mode

What's Changed

• Allow to set random seed by @sfiligoi in #31
• Add support for permanova by @sfiligoi in #32
• Add support for subsampling with replacement by @sfiligoi in #33
• Add multi mode by @sfiligoi in #36

Full Changelog: v1.2.1...v1.3

Better GPU detection and PCoA-only output

What's Changed

• Use OpenACC to detect GPU presence by @sfiligoi in #25
• Add support for hdf5_nodist by @sfiligoi in #27
• Bug fix: Enforce int64, to avoid overflow by @sfiligoi in #28

Full Changelog: v1.2...v1.2.1

Change default precision and deprecate -n

What's Changed

• Fix static hdf5 linking by @sfiligoi in #15
• Reprecate -n and add --n-substeps by @sfiligoi in #18
• Rename some of the variables to clarify semantics. No change in behavior by @sfiligoi in #19
• Change default method precision to fp32, and add explicit _fp64 equivalent by @sfiligoi in #20
• add citation by @wasade in #16
• Request build on linux-gpu-cuda and cleanup code by @sfiligoi in #22
• Reduce disk usage of the GPU compiler by @sfiligoi in #23

Full Changelog: v1.1.3...v1.2

Bug fix and speed improvements

A buffer overflow bug was fixed.

Moreover, the Unweighted Unifrac has been significantly sped up and there has been additional CPU speedups across the board, too.

What's Changed

• Add precomputed sums for Unweighted + bug fix by @sfiligoi in #13
• Add tiling to speed CPU code by @sfiligoi in #14

Full Changelog: v1.1.2...v1.1.3

CPU optimizations

CPU optimizations yielding 1.5x speedup across the board on CPUs, no change to GPU performance.

What's Changed

• Add CPPFLAGS to R compile by @sfiligoi in #7
• Add support for UNIFRAC_TIMING_INFO env variable by @sfiligoi in #8
• Vectorize Weighted unifrac for CPUs by @sfiligoi in #9
• Add larger unifrac tests by @sfiligoi in #10
• Use plain mkstemp() if O_NOATIME is not available by @jmarshall in #6
• More aggressive vectorization of Weighted by @sfiligoi in #11
• Add zero checking to Unweighted for CPU code by @sfiligoi in #12

New Contributors

• @jmarshall made their first contribution in #6

Full Changelog: v1.1.1...v1.1.2

Address performance regression

The switch from 0.20.3 -> 1.0.0 introduced a 10% performance regression on CPUs by compiling for general targets. This minor release addresses the build change.

Expose in-memory API options

With this release, we introduce the ability to call the API using in-memory objects instead of requiring on-disk representations. We also parallelize a few components of the BIOM constructor.