Changelog

All notable changes to this project will be documented in this file.

[0.0.9] - 2022-07-25

Features

Rework `TwoSimdBlocks` into `Twice`

TwoSimdBlocks got removed
Twice<A> gives an alignment guarantee of two blocks of A
halves on aligned block of Twice<A> splits a block into two smaller alignments

[0.0.8] - 2022-07-09

Features

Make `Default` conform to strict provenance

Related: #34

Made `Alignment` types empty

Related: #31

Bug Fixes

Invalid `Hash` impl for `AlignedBytes`

Related: #33

[0.0.7] - 2022-06-03

Features

`alignment_size` function for aligned types

Allows getting the alignment size as a method instead of static.

Related: #29

[0.0.6] - 2022-05-28

Bug Fixes

Take `FnMut` in `new_initialize`

Fn is unnecessarily prohibitive, FnMut allows more initialisation functions.

[0.0.5] - 2022-05-27

Features

Add support for AVX512

When the avx512f target feature is enabled the SimdBlock is set to 64 bytes.

Related: #23

Relax AVX2 requirement to AVX

AVX already includes 256-bit SIMD vectors, so using it as the trigger for SimdBlock instead of AVX2 makes more sense

Related: #22

Reliability

Enable local simd-size-matrix tests

One can use the test_simd_sizes.py script to run the simd-size test matrix locally.

[0.0.4] - 2022-05-25

Features

Support `sse` target feature

When the sse target feature is enabled on x86/x86_64 the SimdBlock size is 16 bytes, unless a target feature with a larger vector is also enabled.

Related: #18

Reliability

Miri test on windows targets

Added `simd_alignment_test`

Test whether the size of SimdBlock agrees with the expected one for a given target.

Added simd-size-matrix test

Use simd_alignment_test to check whether SimdBlock sizes remain consistent across targets.

[0.0.3] - 2022-05-18

Bug Fixes

Impl of `Default` returning unaligned ptrs

Implementations of Default for
AlignedBytes
&AlignedSlice
&mut AlignedSlice
were incorrect by using NonNull::dangling as the pointer.
That pointer need not be aligned to the specified Alignment.
This was fixed by using std's dangling implementation tweaked to
the required alignment, while still not allocating.

Related: #11

Reliability

Allow miri symbolic-alignment-check

Miri's -Zmiri-symbolic-alignment-check breaks align_offset to always return usize::MAX (that is by design), so we need to fallback to regular cast-then-modulo alignment check for tests when running under Miri.

Related: #12

[0.0.2] - 2022-05-17

Features

Added `alignment::TwoTo<N>`

Related: #5

[0.0.1] - 2022-05-17

Features

Initial API

Alignment types driven by the alignment module.
AlignedBytes -- aligned container
AlignedSlice -- type guarantee on alignment of &[u8]
AlignedBlockIterator -- iterate aligned blocks of a slice.
Quality of life implementations of Deref and std::cmp traits.

Reliability

Automatically generate changelog (#1)

Related: #1

Use cargo-hack for feature powerset

Related: #3