What's Changed

• Use u64 in Raders twiddle calculations by @HEnquist in #164
• Release v6.4.1 by @ejmahler in #165

Full Changelog: 6.4.0...6.4.1

What's Changed

• Disable doctests for wasm by @ejmahler in #160
• Adding process_immutable_with_scratch method by @michaelciraci in #157
• Refactor FFT boilerplate by @ejmahler in #161
• Move MSRV to cargo.toml and remove build.rs by @lucab in #143
• Release 6.4 by @ejmahler in #162

New Contributors

• @michaelciraci made their first contribution in #157
• @lucab made their first contribution in #143

Full Changelog: 6.3.0...6.4.0

Changed

• Miscellaneous improvements to performance for non-power-of-two FFT lengths, especially on NEON (#131, #132, #134, #136, #137)

Fixed

• Fixed FftPlanner not being Sync (#153)

[6.1]

Released 7th Novemeber 2022

Added

• Implemented a code path for Neon-optimized FFTs on AArch64 (Thanks to Henrik Enquist!) (#84 and #78)

Changed

• Improved performance of power-of-3 FFTs when not using SIMD-accelerated code paths (#80)
• Reduced memory usage for some FFT sizes (#81)

[6.0.1]

Released 10 May 2021

Fixed

• Fixed a compile-time divide by zero error on nightly Rust in stdarch\crates\core_arch\src\macros.rs (#75)
• Increased the minimum version of strength_reduce to 0.2.3

[5.1.1]

Released 10 May 2021

Fixed

• Fixed a compile-time divide by zero error on nightly Rust in stdarch\crates\core_arch\src\macros.rs (Backported from v6.0.1)
• Increased the minimum version of strength_reduce to 0.2.3 (Backported from v6.0.1)

Released 16 April 2021

Breaking Changes

• Increased the version of the num-complex dependency to 0.4.
  • This is a breaking change because we have a public dependency on num-complex.
  • See the num-complex changelog for a list of breaking changes in num-complex 0.4
  • As a high-level summary, most users will not need to do anything to upgrade to RustFFT 6.0: num-complex 0.4 re-exports a newer version of rand, and that's num-complex's only documented breaking change.

Released 16 April 2021

Added

• Implemented a code path for SSE-optimized FFTs (Thanks to Henrik Enquist!) (#60)
  • Plan a FFT using the FftPlanner (or the new FftPlannerSse) on a machine that supports SSE4.1 (but not AVX) and you'll see a 2-3x performance improvement over the default scalar code.

Fixed

• Fixed underflow when planning an AVX FFT of size zero (#56)
• Fixed the FFT planner not being Send, due to internal use of Rc<> (#55)
• Fixed typo in documentation (#54)
• Slightly improved numerical precision of Rader's Algorithm and Bluestein's Algorithm (#66, #68)
• Minor optimizations to Rader's Algorithm and Bluestein's Algorithm (#59)
• Minor optimizations to MixedRadix setup time (#57)
• Optimized performance of Radix4 (#65)

[5.0.1]

Released 8 January 2021

Fixed

• Fixed the FFT planner not choosing an obviously faster plan in some rare cases (#46)
• Documentation fixes and clarificarions (#47, #48, #51)

Released 4 January 2021

Breaking Changes

• Several breaking changes. See the Upgrade Guide for details.

Added

• Added support for the Avx instruction set. Plan a FFT with the FftPlanner on a machine that supports AVX, and you'll get a 5x-10x speedup in FFT performance.

Changed

• Even though the main focus of this release is on AVX, most users should see moderate performance improvements due to a new internal architecture that reduces the amount of internal copies required when computing a FFT.