KFR 7.0.1

Added

• Variance and standard deviation functions: variance and stddev
• macOS universal binaries
• New module: audio
• New Wave, AIFF, FLAC, and Mp3 (via minimp3) encoders/decoders
• Apple ALAC support
• Apple CAF container support
• RF64/BW64 container support
• RISC-V support
• RISC-V prebuilt binaries
• New ARM and ARM64 toolchain implementation
• New RISC-V 64 toolchain for riscv64-linux-gnu
• macos-merge-binaries.cmake merges two directories into one with universal binaries
• Custom port to fix the macosforge/alac bug
• New elliptic function to design elliptic filters with ripple parameters (follows SciPy semantics)
• Version information added for each module (library_version_dsp, library_version_audio, etc.)
• samples_store and samples_load convert audio channels to/from stored formats (integer and floating-point, optional dithering)
• strided_channel class for reading/writing strided data in KFR expressions
• audio_data class providing a unified interface for interleaved/planar audio data
• Zero-copy construction from user-supplied pointers for audio_data
• Computing statistics for audio_data: Peak and RMS
• Per-channel processing for audio_data
• Basic dithering support: Rectangular and Triangular
• audio_encoder and audio_decoder base classes for encoding/decoding audio
• Reading RIFF chunks in audio_decoder
• Audio format detection from the first 16 bytes
• Audio format selection from file extension
• Raw decoder/encoder
• Audio decoder based on Media Foundation for Windows
• Automatic switch from standard Wave to RF64 if audio exceeds 4 GB (can be disabled with switch_to_rf64_if_over_4gb)
• encode_audio_file and decode_audio_file functions for simplified file handling
• arange function added (similar to NumPy’s)
• Added has_expression_traits, input_expression, output_expression, and input_output_expression concepts
• expr_element concept defines valid expression element types
• Added f_class, i_class, and u_class concepts (is_f_class, is_i_class, is_u_class)
• Added not_f_class, not_i_class, and not_u_class concepts
• has_data_size concept checks for std::data and std::size definitions
• Added filtfilt for forward-backward IIR filtering
• Added arrangement_speakers and arrangement_for_channels functions
• Added fopen_path for cross-platform UTF-8 file access; replaces fopen_portable
• memory_reader and memory_writer adapters for user memory buffers
• Added more tests for vector concatenation and slicing
• Minimal usage examples added under tests/usage-config and tests/usage-manual
• New RISC-V vector types: f32rvv, i32rvv, u32rvv, etc.
• Added simd_compat concept to verify SIMD element compatibility
• Added scoped_flush_denormals RAII class to enable flush-to-zero semantics on x86 and ARM
• test_matrix function introduced (replaces testo::matrix for Catch2 compatibility)

Changed

• Audio and IO modules can now be disabled during KFR build with KFR_ENABLE_AUDIO and KFR_ENABLE_IO options (both on by default) @jcelerier
• DSP module can be disabled too with KFR_ENABLE_DSP option (on by default)
• Documentation migrated to mkdocs + cxxdox 2
• Prebuilt binaries are tested after build
• Prebuilt binaries use Clang 21
• KFR_ENABLE_MULTIARCH is forcibly turned off for non-x86
• KFR_USE_BOOST_MATH option enables the use of standalone Boost.Math for elliptic filters (on by default)
• KFR_USE_BOOST option switches to using Boost.Math from a full Boost installation. Enable if you already use Boost
• C++20 standard is now handled via the cxx_std_20 CMake feature and inherited by dependent targets
• Refactored CMake config with separate targets and dependency handling
• add_kfr_library refactored to distinguish between private and public dependencies and definitions
• In multiarch builds, the base architecture is now named exactly as the module itself (was: kfr_dsp_sse2, kfr_dsp_avx2; now: kfr_dsp, kfr_dsp_avx2)
• Examples and tools migrated to the new audio I/O code
• All third-party libraries moved to src/thirdparty or include/kfr/thirdparty
• Old audio I/O handled by audio_reader and audio_writer is now deprecated
• reverse expression is now writable
• audio_sample_type now represents bit depth; negative if floating-point
• convert_endianness now accepts references to data types
• samplerate_converter is now default-, move-constructible, and move-assignable
• IO_SEEK_64 and IO_TELL_64 macros renamed to KFR_IO_SEEK_64 and KFR_IO_TELL_64
• get_nth_type optimized for Clang and GCC
• All constexpr variables now marked inline
• is_poweroftwo, next_poweroftwo, ilog2, etc., now use C++20 <bit> header
• Sample Rate Converter tool refactored for chunked resampling
• [Breaking change] CMake 3.16 required
• [Breaking change] cometa namespace merged into kfr; Cometa is no longer distributed separately. Replace cometa with kfr in your sources
• [Breaking change] All macros renamed from CMT_ to KFR_; replace all occurrences of CMT_ with KFR_
• [Breaking change] All macros renamed from TESTO_ to KFR_; replace all occurrences of TESTO_ with KFR_
• [Breaking change] cometa.hpp renamed to meta.hpp; kfr/cometa/*.hpp renamed to kfr/meta/*.hpp. Update your includes accordingly
• [Breaking change] kfr/testo/*.hpp renamed to kfr/test/*.hpp
• [Breaking change] Some classes and functions no longer accept a template parameter and now default to double for maximum precision. Related changes:
  • zpk<float_type> replaced with zpk
  • bessel<float_type> and butterworth<float_type> replaced with bessel and butterworth
  • bilinear<float_type>, lp2lp_zpk<float_type>, lp2bs_zpk<float_type>, lp2bp_zpk<float_type>, lp2hp_zpk<float_type>, and warp_freq<float_type> replaced with non-template equivalents
  • iir_lowpass<float_type>, iir_highpass<float_type>, iir_bandpass<float_type>, iir_bandstop<float_type> replaced with non-template equivalents
  • iir_params<float_type> still accepts a template parameter to control precision
  • to_sos now accepts float_type (default double) to return iir_params<float_type>
  • New elliptic function follows the same rule and always produces double-precision zpk
• intrinsics namespace renamed to intr
• Many functions from iir_design.hpp moved to src/iir_design.cpp
• Documentation updates: applying FIR/IIR filters, computing loudness, performing sample conversion
• [Breaking change] KFR_ACCEPT_EXPRESSIONS removed; replaced with expression_argument and expression_arguments concepts
• [Breaking change] identity removed; replaced with C++ std::type_identity_t
• [Breaking change] output_expression concept replaces enable_if_output_expression trait
• [Breaking change] 8-bit sample support removed
• [Breaking change] Fixed typo: convert_endianess → convert_endianness
• [Breaking change] CMT_NOEXCEPT removed; use noexcept
• [Breaking change] C API: kfr_size_t and kfr_int32_t removed; use size_t and int32_t
• [Breaking change] Speaker → speaker_type, SpeakerArrangement → speaker_arrangement
• [Breaking change] from_lambda parameter required to construct vec from a lambda generator
• ebu_r128 now accepts std::span instead of std::vector
• arraysize and carraysize removed; use std::size from C++17
• Removed C++17 compatibility definitions for non-compliant compilers
• More functions marked noexcept and constexpr
• CMT_ENABLE_IF mostly replaced with concepts and requires; KFR_ENABLE_IF still available
• CMT_CLANG_EXT renamed to KFR_VEC_EXT and now usable with GCC
• C API can now be built without exceptions
• numeric concept replaces is_numeric<T1> predicate
• aligned_size, aligned_force_free, aligned_release, and aligned_reallocate only defined if KFR_MANAGED_ALLOCATION is on (default off)

Fixed

• C++23 compatibility fixes
• Fixed extend(vec) function for compound vectors
• Custom port to fix the macosforge/alac bug
• Fixed generic fallback for bittestany
• Fixed u8/i8 shifts on x86 and related tests
• Various smaller fixes

KFR 7.0.0-rc

7.0.0-rc

Added

• macOS universal binaries
• New module: audio
• New Wave, AIFF, FLAC, and Mp3 (via minimp3) encoders/decoders
• Apple ALAC support
• Apple CAF container support
• RF64/BW64 container support
• RISC-V support
• RISC-V prebuilt binaries
• New ARM and ARM64 toolchain implementation
• New RISC-V 64 toolchain for riscv64-linux-gnu
• macos-merge-binaries.cmake merges two directories into one with universal binaries
• Custom port to fix the macosforge/alac bug
• New elliptic function to design elliptic filters with ripple parameters (follows SciPy semantics)
• Version information added for each module (library_version_dsp, library_version_audio, etc.)
• samples_store and samples_load convert audio channels to/from stored formats (integer and floating-point, optional dithering)
• strided_channel class for reading/writing strided data in KFR expressions
• audio_data class providing a unified interface for interleaved/planar audio data
• Zero-copy construction from user-supplied pointers for audio_data
• Computing statistics for audio_data: Peak and RMS
• Per-channel processing for audio_data
• Basic dithering support: Rectangular and Triangular
• audio_encoder and audio_decoder base classes for encoding/decoding audio
• Reading RIFF chunks in audio_decoder
• Audio format detection from the first 16 bytes
• Audio format selection from file extension
• Raw decoder/encoder
• Audio decoder based on Media Foundation for Windows
• Automatic switch from standard Wave to RF64 if audio exceeds 4 GB (can be disabled with switch_to_rf64_if_over_4gb)
• encode_audio_file and decode_audio_file functions for simplified file handling
• arange function added (similar to NumPy’s)
• Added has_expression_traits, input_expression, output_expression, and input_output_expression concepts
• expr_element concept defines valid expression element types
• Added f_class, i_class, and u_class concepts (is_f_class, is_i_class, is_u_class)
• Added not_f_class, not_i_class, and not_u_class concepts
• has_data_size concept checks for std::data and std::size definitions
• Added filtfilt for forward-backward IIR filtering
• Added arrangement_speakers and arrangement_for_channels functions
• Added fopen_path for cross-platform UTF-8 file access; replaces fopen_portable
• memory_reader and memory_writer adapters for user memory buffers
• Added more tests for vector concatenation and slicing
• Minimal usage examples added under tests/usage-config and tests/usage-manual
• New RISC-V vector types: f32rvv, i32rvv, u32rvv, etc.
• Added simd_compat concept to verify SIMD element compatibility
• Added scoped_flush_denormals RAII class to enable flush-to-zero semantics on x86 and ARM
• test_matrix function introduced (replaces testo::matrix for Catch2 compatibility)

Changed

• Prebuilt binaries are tested after build
• Prebuilt binaries use Clang 21
• KFR_ENABLE_MULTIARCH is forcibly turned off for non-x86
• KFR_USE_BOOST_MATH option enables the use of standalone Boost.Math for elliptic filters (on by default)
• KFR_USE_BOOST option switches to using Boost.Math from a full Boost installation. Enable if you already use Boost
• C++20 standard is now handled via the cxx_std_20 CMake feature and inherited by dependent targets
• Refactored CMake config with separate targets and dependency handling
• add_kfr_library refactored to distinguish between private and public dependencies and definitions
• In multiarch builds, the base architecture is now named exactly as the module itself (was: kfr_dsp_sse2, kfr_dsp_avx2; now: kfr_dsp, kfr_dsp_avx2)
• Examples and tools migrated to the new audio I/O code
• All third-party libraries moved to src/thirdparty or include/kfr/thirdparty
• Old audio I/O handled by audio_reader and audio_writer is now deprecated
• reverse expression is now writable
• audio_sample_type now represents bit depth; negative if floating-point
• convert_endianness now accepts references to data types
• samplerate_converter is now default-, move-constructible, and move-assignable
• IO_SEEK_64 and IO_TELL_64 macros renamed to KFR_IO_SEEK_64 and KFR_IO_TELL_64
• get_nth_type optimized for Clang and GCC
• All constexpr variables now marked inline
• is_poweroftwo, next_poweroftwo, ilog2, etc., now use C++20 <bit> header
• Sample Rate Converter tool refactored for chunked resampling
• [Breaking change] CMake 3.16 required
• [Breaking change] cometa namespace merged into kfr; Cometa is no longer distributed separately. Replace cometa with kfr in your sources
• [Breaking change] All macros renamed from CMT_ to KFR_; replace all occurrences of CMT_ with KFR_
• [Breaking change] All macros renamed from TESTO_ to KFR_; replace all occurrences of TESTO_ with KFR_
• [Breaking change] cometa.hpp renamed to meta.hpp; kfr/cometa/*.hpp renamed to kfr/meta/*.hpp. Update your includes accordingly
• [Breaking change] kfr/testo/*.hpp renamed to kfr/test/*.hpp
• [Breaking change] Some classes and functions no longer accept a template parameter and now default to double for maximum precision. Related changes:
  • zpk<float_type> replaced with zpk
  • bessel<float_type> and butterworth<float_type> replaced with bessel and butterworth
  • bilinear<float_type>, lp2lp_zpk<float_type>, lp2bs_zpk<float_type>, lp2bp_zpk<float_type>, lp2hp_zpk<float_type>, and warp_freq<float_type> replaced with non-template equivalents
  • iir_lowpass<float_type>, iir_highpass<float_type>, iir_bandpass<float_type>, iir_bandstop<float_type> replaced with non-template equivalents
  • iir_params<float_type> still accepts a template parameter to control precision
  • to_sos now accepts float_type (default double) to return iir_params<float_type>
  • New elliptic function follows the same rule and always produces double-precision zpk
• intrinsics namespace renamed to intr
• Many functions from iir_design.hpp moved to src/iir_design.cpp
• Documentation updates: applying FIR/IIR filters, computing loudness, performing sample conversion
• [Breaking change] KFR_ACCEPT_EXPRESSIONS removed; replaced with expression_argument and expression_arguments concepts
• [Breaking change] identity removed; replaced with C++ std::type_identity_t
• [Breaking change] output_expression concept replaces enable_if_output_expression trait
• [Breaking change] 8-bit sample support removed
• [Breaking change] Fixed typo: convert_endianess → convert_endianness
• [Breaking change] CMT_NOEXCEPT removed; use noexcept
• [Breaking change] C API: kfr_size_t and kfr_int32_t removed; use size_t and int32_t
• [Breaking change] Speaker → speaker_type, SpeakerArrangement → speaker_arrangement
• [Breaking change] from_lambda parameter required to construct vec from a lambda generator
• ebu_r128 now accepts std::span instead of std::vector
• arraysize and carraysize removed; use std::size from C++17
• Removed C++17 compatibility definitions for non-compliant compilers
• More functions marked noexcept and constexpr
• CMT_ENABLE_IF mostly replaced with concepts and requires; KFR_ENABLE_IF still available
• CMT_CLANG_EXT renamed to KFR_VEC_EXT and now usable with GCC
• C API can now be built without exceptions
• numeric concept replaces is_numeric<T1> predicate
• aligned_size, aligned_force_free, aligned_release, and aligned_reallocate only defined if KFR_MANAGED_ALLOCATION is on (default off)

Fixed

• Custom port to fix the macosforge/alac bug
• Fixed generic fallback for bittestany
• Fixed u8/i8 shifts on x86 and related tests
• Various smaller fixes

KFR 6.3.1

Bug-fix release.

Fixed

• Fix u8/i8 shifts in MSVC
• Refactor Bluestein's algorithm to make accuracy stable across all sizes
• Improve test coverage for large SIMD vector types

KFR 6.3.0

Added

• Progressive DFT execution

Changed

• Matrix transpose performance has been improved by 10-30%
• Optimize scalar to expression assignment for msvc (thanks to @zengdaiwei)

6.2.0

Added

• Android x86 and x86_64 builds
• Linux arm and aarch64 (arm64) builds

Changed

• Documentation update

Fixed

• Minor fixes to small_buffer and npy.hpp.

6.1.1

Changed

• DFT performance has been improved by 30-80% for ARM and ARM64 cpus
• DFT benchmark is now enabled for non-x86 builds

6.1.0

• It is now possible to perform multidimensional DFT via the C API.
• Documentation updated.
• [Internal] Updated to the latest CxxDox for enhanced documentation features.
• The Dims template parameter in dft_plan_md and dft_plan_md_real now defaults to dynamic_shape.
• A workaround has been applied to mitigate the MSVC Internal Compiler Error.
• Additional tests added.

Full Changelog: 6.0.4...6.1.0

6.0.4

This release includes fixes to ensure compatibility with Clang 19. No new features have been added.

6.0.3

KFR 6.0.3 fixes and improvements

• More documentation (thanks to @Jalmenara)
• Stateless parameter fixed
• Fix build with non-MSVC on Win32 (thanks to @jcelerier)
• Android fixes
• Removed deprecated atomic initialization

6.0.2

What's new in KFR 6

• DFT performance has been improved up to 40% (backported to KFR 5.2.0 branch)
• C API for non x86 architectures
• DSP refactoring with easier initialization
• Multiarchitecture for resampling, FIR and IIR filters
• matrix_transpose: optimized matrix transpose (square/non-square, inplace/out-of-place, real/complex, scalar/vectors)
• CMake config file generation (find_package(KFR CONFIG) support, see installation)
• .npy format support (reading/writing, v1/v2, c/fortran order, real/complex, bigendian/littleendian)
• Multidimensional DFT: real/complex
• inline_vector
• Windows arm64 support
• Emscripten (wasm/wasm64) support

Other changes

• CMake minimum version is 3.12
• Multidimensional reference DFT
• Easier cross compilation to ARM64 on x86_64 macOS
• Automated tests using GitHub Actions (previously Azure Pipelines)
• GCC 7 and 8: emulate missing avx-512 instrinsics
• read_group and write_group
• [❗breaking change] reshape_may_copy and flatten_may_copy in tensor<> allows copying by default
• shape<>::transpose function
• tensor<>::transpose function
• convert_endianess
• DFT, DSP and IO sources have been moved to src/ directory
• Multiarchitecture is enabled by default
• KFR_DFT_NO_NPo2 has been removed (assumed always enabled)
• Tests refactoring
• Some tests moved to tests/internal/
• [❗breaking change] Scalars are now passed by value in expressions (this fixes dangling references in some cases)
• Expression functions should return expression_make_function instead of expression_function
• KFR_WITH_CLANG
• KFR_VERSION CMake variable
• Functions to get module versions (library_version_dft, library_version_dsp etc)
• Exceptions are no longer enforced in MSVC
• kfr::complex removed (use std::complex instead). KFR_STD_COMPLEX cmake variable removed too
• strides_for_shape for fortran order
• AARCH and ARM emulation refactoring (dynamic libraries are now supported)
• call_with_temp
• maximum_dims is now 16 (was 8)
• to_fmt/from_fmt supports inplace
• shape refactoring: rotate_left, rotate_right, remove_back, remove_front
• temp argument can be nullptr for DFT (temporary buffer will be allocated on stack or heap)
• dft_plan and similar classes have now default and move constructors
• -DCMAKE_POSITION_INDEPENDENT_CODE=ON is required for building C API
• ci/run.sh can now build in a directory outside source tree
• [❗breaking change]graphics/color.hpp and graphics/geometry.hpp have been removed
• Simpler CMT_CVAL macro
• /Zc:lambda is now required for building KFR in MSVC
• println for string_view
• MSVC internal compiler error fixed
• Complex vector operators fixed