v0.6.2

What's Changed

• ci: test target build on stable by @jvdd in #53
• feat: add 64-bit aarch64 SIMD implementation by @jvdd in #52
• perf: use uint MAX_INDEX on arm / aarch64 for uint SIMD by @jvdd in #54
• refactor: remove unnecessary repetition operator from macros by @jvdd in #57
• test: include argmin and argmax in integration tests by @jvdd in #58
• test: use full range for floats uniform by @jvdd in #60
• fix: remove stdsimd feature; add stdarch_x86_avx512 by @lazyky in #62
• chore: remove unused get_lane_size function from SIMDInstructionSet by @jvdd in #64
• enh: remove simd_nightly feature flag for aarch64 by @jvdd in #63

New Contributors

• @lazyky made their first contribution in #62

Full Changelog: v0.6.1...v0.6.2

v0.6.1

What's Changed

• feat: add "nightly_simd" feature to allow stable build by @jvdd in #50

Full Changelog: v0.6.0...v0.6.1

v0.6.0

What's Changed

• 🚀 add argmin & argmax by @jvdd in #42
• perf: tune scalar argmin & argmax by @jvdd in #43
• feat: optimize + add f16 scalar implementations by @jvdd in #44
• 🔥 impl argmin & argmax in public traits by @jvdd in #45
• chore: comply with breaking api change in arrow v36 by @jvdd in #47
• 🎉 add f16 simd support for ignore nan by @jvdd in #46
• 🎉 add optional arrow2 support by @jvdd in #48

Full Changelog: v0.5.0...v0.6.0

v0.5.0

This release adds nan-handling to the argminmax project 🚀

• ArgMinMax its argminmax ignores nans (= backwards compatible)
• NanArgMinMax (only implemented for floats) its nanargminmax propagates nans (return idx of first nan)

Other changes in this release:

• feature-gate floats with the (default) "float" feature
• add proper docs 📚
• refactoring of tests, benchmarks & internal traits (making code more flexible, limiting duplicate code)

_{P.S.: Since CodSpeedHQ/codspeed-rust#1 is resolved, we removed "half" from the default-features.}

What's Changed

• ♻️ major refactoring by @jvdd in #23
• 🥧 update NEON SIMD by @jvdd in #25
• ♻️ change nan default handling behavior to SkipNa by @jvdd in #28
• Add MacOS stuff to .gitignore by @varon in #29
• 🚀 float NaN handling by @jvdd & @varon in #21
• ♻️ parameterize tests by @jvdd in #31
• fix(test): clip shift size to 31 bits on arm by @jvdd in #32
• docs: describe transformation for uints by @jvdd in #33
• chore(bench): rename benches by @jvdd in #34
• refactor: architecture - use type-state pattern by @jvdd in #35
• refactor(test): pass struct to tests instead of method by @jvdd in #36
• ci: upgrade codspeed & make half non-default feature by @jvdd in #37
• feat: add new default float feature by @jvdd in #38
• refactor: add nanargminmax to NanArgMinMax trait by @jvdd in #39
• fix: allow double update in scalar ignore nan implementation by @jvdd in #40
• docs: add proper documentation 🎉 by @jvdd in #41

Full Changelog: v0.4.0...v0.5.0

v0.4.0

This release improves the underlying implementation by replacing ndarray::ArrayView1 with the more flexible (and zero-dependency) &[T] slice. This new implementation enhances the library's interoperability with other array-like data types and crates.

As of now importing the ArgMinMax trait adds the .argminmax function to:

• slice (&[T])
• Vec

Using optional features, the .argminmax function can be added to:

• "ndarray" feature: ndarray::ArrayView1 & ndarray::Array1 (allows backwards compatibility)
• "arrow" feature: arrow::array::PrimitiveArray

_{P.S.: To include the amazing @CodSpeedHQ continuous performance monitoring in our CI we had to add the "half" feature as default-feature (see CodSpeedHQ/codspeed-rust#1). As soon as this issue is resolved we plan to remove again "half" from the default-features.}

What's Changed

• 🐰 add codspeed benchmarking to ci/cd by @jvdd in #15
• 🍵 use slice internally + implement for various types by @jvdd in #13
• QOL improvements by @varon in #17

New Contributors

• @varon made their first contribution in #17

Full Changelog: v0.3.1...v0.4.0

v0.3.1

This is the latest release that uses ndarray::ArrayView1 as underlying and mandatory interface. In future releases, we will switch to using slices internally and offer optional compatibility with other data types, including ndarray::ArrayView1

_{P.S.: This release and tag were created after we had already implemented slices as the internal interface. The tag references the latest commit that includes the ndarray::ArrayView1 implementation, which corresponds to v0.3.1 of the Rust crate.}