0.5.0 - 2025-10-19

Using FMA to accelerate and reduce roundoff error in dot products.

Added

• Add fma feature that enables the use of fused multiply-add for dot products
  • Disabled by default because this causes a severe performance regression unless FMA is enabled for the target

Changed

• !Add trait bound on MulAdd<Output = T> for numeric types in FIR and IIR filters
• !Make std a default feature to enable publishing and docs to work properly

0.4.0 - 2025-10-03

Changed

• !Add bound on Num trait for generic fields
• Implement Default on all structs
• Update deps

0.3.0 - 2025-08-31

Various performance-related improvements.
~20% improvement in speed based on testing on an STM32H7 microcontroller.

Changed

• !Store filter coeffs and taps in opposite order (most-recent-last) to improve vectorization
• !Const-unroll compute intensive loops
  • !Limit size of FIR filters to <128 taps
  • !Limit order of polynomial fractional delay filters to <=10
• Add u value as initial for A*x dot product
  • (Potentially substantially) Improves float error for small values of latest measurement
    u at the expense of slightly worse float error for large u
• Update flop per eval count to 4N-1
• Update rust edition to 2024

0.2.5 - 2025-05-25

Changed

• Make Ring and AlignedArray pub to support read-only access to filter internals

Added

• Add FIR module with generic number field type
• Add Lagrange interpolator fractional delay filter construction
• Add functions for read-only access to IIR filter internals

0.2.4 - 2025-02-06

Changed

• Change large const arrays to static to avoid inlining excessively large data

0.2.3 - 2025-01-01

Changed

• Use ring buffer for IIR and median filter state storage
• Lower align of SisoIirFilter struct, since the internal storage is already aligned

Added

• Implement N-point median filter for odd N >= 3

0.2.2 - 2024-12-27

Changed

• Derive Clone and Copy on SisoIirFilter to allow faster construction of batches of identical filters
• Make generated module pub to provide access to min/max cutoff ratio bounds for generated filters
• Loosen version requirement on num_traits dep

0.2.1 - 2024-12-26

Changed

• Lower align of SisoIirFilter struct and internal state arrays to 8

Added

• Add SisoIirFilter.initialize(u) to set the internal state vector to the steady-state values associated with a given measurement u

0.2.0 - 2024-07-10

Changed

• Implement rescaling of C vector during initialization of interpolated filters to correct gain for interpolation error
• Tighten tolerance on gain of generated butterworth filters to 0.01% at minimum cutoff ratio
• Decrease range of validity for generated butterworth filters slightly to meet new tolerance