CHANGELOG.md (15955B)
1 # Changelog 2 3 ## 6.1.1 4 5 #### Changed 6 7 - DFT performance has been improved by 30-80% for ARM and ARM64 cpus 8 - DFT benchmark is now enabled for non-x86 builds 9 10 ## 6.1.0 11 12 #### Added 13 14 - C API now supports Multidimensional DFT 15 16 #### Changed 17 18 - Documentation update 19 - Update to latest CxxDox 20 - `Dims` template parameter in `dft_plan_md` and `dft_plan_md_real` now defaults to `dynamic_shape` 21 22 ## 6.0.4 23 24 #### Fixed 25 26 - Clang 19 compatibility fix 27 28 ## 6.0.3 29 30 #### Changed 31 32 - Added more Doxygen documentation for filter functions (@Jalmenara) 33 - Examples now include detailed comments 34 35 #### Fixed 36 37 - Fixed stateless parameter 38 - Fixed build with non-MSVC on Win32 (@jcelerier) 39 - Resolved Android building issues 40 - Removed deprecated atomic initialization 41 42 ## 6.0.2 43 44 #### Added 45 46 - Windows arm64 support 47 - Emscripten (wasm/wasm64) support 48 49 #### Changed 50 51 - `complex_size` now takes dft_pack_format as parameter 52 53 ## 6.0.0 54 55 - DFT performance has been improved up to 40% (backported to KFR 5.2.0 branch) 56 - C API for non x86 architectures 57 - DSP refactoring with easier initialization 58 - Multiarchitecture for resampling, FIR and IIR filters 59 - `matrix_transpose`: optimized matrix transpose (square/non-square, inplace/out-of-place, real/complex, scalar/vectors) 60 - CMake config file generation (`find_package(KFR CONFIG)` support, see [installation](installation.md)) 61 - `.npy` format support (reading/writing, v1/v2, c/fortran order, real/complex, bigendian/littleendian) 62 - Multidimensional DFT: real/complex 63 - `inline_vector` 64 65 #### Other changes 66 67 - CMake minimum version is 3.12 68 - Multidimensional reference DFT 69 - Easier cross compilation to ARM64 on x86_64 macOS 70 - Automated tests using GitHub Actions (previously Azure Pipelines) 71 - GCC 7 and 8: emulate missing avx-512 instrinsics 72 - `read_group` and `write_group` 73 - [❗breaking change] `reshape_may_copy` and `flatten_may_copy` in `tensor<>` allows copying by default 74 - `shape<>::transpose` function 75 - `tensor<>::transpose` function 76 - `convert_endianess` 77 - DFT, DSP and IO sources have been moved to `src/` directory 78 - Multiarchitecture is enabled by default 79 - `KFR_DFT_NO_NPo2` has been removed (assumed always enabled) 80 - Tests refactoring 81 - Some tests moved to `tests/internal/` 82 - [❗breaking change] Scalars are now passed by value in expressions (this fixes dangling references in some cases) 83 - Expression functions should return `expression_make_function` instead of `expression_function` 84 - `KFR_WITH_CLANG` 85 - `KFR_VERSION` CMake variable 86 - Functions to get module versions (`library_version_dft`, `library_version_dsp` etc) 87 - Exceptions are no longer enforced in MSVC 88 - `kfr::complex` removed (use `std::complex` instead). `KFR_STD_COMPLEX` cmake variable removed too 89 - `strides_for_shape` for fortran order 90 - AARCH and ARM emulation refactoring (dynamic libraries are now supported) 91 - `call_with_temp` 92 - `maximum_dims` is now 16 (was 8) 93 - `to_fmt`/`from_fmt` supports inplace 94 - `shape` refactoring: `rotate_left`, `rotate_right`, `remove_back`, `remove_front` 95 - temp argument can be `nullptr` for DFT (temporary buffer will be allocated on stack or heap) 96 - `dft_plan` and similar classes have now default and move constructors 97 - `-DCMAKE_POSITION_INDEPENDENT_CODE=ON` is required for building C API 98 - `ci/run.sh` can now build in a directory outside source tree 99 - [❗breaking change]`graphics/color.hpp` and `graphics/geometry.hpp` have been removed 100 - Simpler `CMT_CVAL` macro 101 - `/Zc:lambda` is now required for building KFR in MSVC 102 - `println` for `string_view` 103 - MSVC internal compiler error fixed 104 - Complex vector operators fixed 105 106 ## 5.2.0 107 108 2023-11-27 109 110 #### Changed 111 112 - The performance of DFT has been increased up to 40% compared to KFR 5 on x86 and x86_64 113 in single and double precision, inplace and out of place processing. 114 115 #### Added 116 117 - `KFR_NO_PERF_TESTS` define can now disable performance tests 118 - `CMT_CVAL` for extracting constexpr-enabled value from `cval_t` 119 - `fft_algorithm_selection` to select FFT algorithm for given FFT size. 120 121 #### Fixed 122 123 - Warnings in Clang 10 https://github.com/kfrlib/kfr/issues/198 124 - `std::is_pov` is peprecated in C++20 https://github.com/kfrlib/kfr/issues/190 125 - DFT sizes 0 and 1 were not processed correctly https://github.com/kfrlib/kfr/issues/195 126 - Internal compiler error in Visual Studio Compiler 19.37 https://github.com/kfrlib/kfr/issues/194 127 - Goertzel issue https://github.com/kfrlib/kfr/issues/121 128 - Bug in `nearest_real_or_complex` https://github.com/kfrlib/kfr/issues/137 129 - Fixed operators in `KFR_STD_COMPLEX` mode 130 - Testo library: typo in `epsilon_scope` 131 - Fix ambiguities with std::identity (c++20) 132 - Force linking in correct order for multi-architecture binaries 133 134 ## 5.1.0 135 136 2023-10-11 137 138 #### Added 139 140 - Tukey window function 141 - Subscript operator for `vec<>` 142 - Unary operators for `vec<>` 143 144 #### Fixed 145 146 - Inverse DCT has been fixed 147 - Allow C API to be built on non-x86 systems 148 - Tensor iteration range has been fixed 149 - `transpose(vec<vec<>>)` has been fixed 150 - `vec<bit<>>` bug on GCC and MSVC 151 - Internal constant `is_pod` has been removed 152 - `root` and `cbrt` have been fixed for negative values 153 - Fixed numerous warnings in MSVC 154 155 ## 5.0.3 156 157 2023-06-26 158 159 #### Added 160 161 - Planck-taper window function 162 163 #### Fixed 164 165 - Sample rate conversion: automatic zero padding 166 - DFT: incorrect result of real dft when input size != 4N https://github.com/kfrlib/kfr/pull/141 167 - Add options for installing libraries and headers https://github.com/kfrlib/kfr/pull/182 168 169 ## 5.0.2 170 171 2023-01-25 172 173 #### Performance 174 175 - ARM/ARM64 performance has been improved up to 2 times in various usage scenarios, including DFT. 176 177 #### Fixed 178 179 - Fix sine and other functions not accepting scalar references 180 - Fix possible name conflict with fmt 181 182 ## 5.0.1 183 184 2022-12-06 185 186 #### Changed 187 188 - Documentation updates 189 190 #### Fixed 191 192 - MSVC fixes 193 - Ambiguous assign operator fixes 194 195 ## 5.0.0 196 197 2022-11-30 198 199 #### Added 200 201 - New `tensor<T, dims>` class for multidimensional data (like numpy's nparray) 202 - Histogram computation 203 - Normal (gaussian) distribution for random number generator 204 - All builtin expressions support multiple dimensions 205 - Exception support (may be configured to call user-supplied function or std::abort) 206 - [changes required] CMake variables now have `KFR_` prefix 207 - Template parameter deduction for `vec`, so `vec{1, 2}` is the same as `vec<int, 2>{1, 2}` 208 - [changes required] `random_state` is now architecture-agnostic and defined in `kfr` namespace 209 - All expression classes have been moved from `kfr::CMT_ARCH_NAME::internal` to `kfr::CMT_ARCH_NAME` namespace 210 - `expression_traits<T>` introduced to support interpreting any object as kfr expression 211 - [changes required] User-defined expressions should be rewritten to be used in KFR5 212 - Out-of-class assign operators for all input & output expressions 213 - `round.hpp`, `clamp.hpp`, `select.hpp`, `sort.hpp`, `saturation.hpp`, `min_max.hpp`, `logical.hpp`, `abs.hpp` headers have been moved to `simd` module 214 - `state_holder.hpp` has been moved to `base` module 215 - All code related to expressions have been moved to `base` module 216 - `vec<T, N>::front()` and `vec<T, N>::front()` are now writable 217 - `set_elements` functions for output expressions like `get_elements` for input expressions 218 219 #### Changed 220 221 - Documentation updates 222 223 ## 4.3.1 224 225 2022-11-23 226 227 #### Fixed 228 229 - C++20 compatibility fixes 230 231 ## 4.3.0 232 233 2022-10-14 234 235 #### Changed 236 * Compile times improved and memory usage reduced for MSVC and GCC 237 * `cxxdox` version updated 238 * Tests for latest Clang, Azure Pipelines images are updated 239 * .editorconfig file 240 241 #### Fixed 242 * Fixed incompatibility with latest GCC 243 * Fixed various Internal Compiler Error in latest MSVC2019 244 * Fixed tests for Clang 14 245 * Fixed bugs in gather/scatter and read/write functions 246 247 ## 4.2.1 248 249 2021-04-29 250 251 #### Fixed 252 253 - AVX512 intrinsics in MSVC and GCC 254 - `carg` 255 - Python 3.7 compatibility 256 257 #### Changed 258 259 - Global constants are now inline 260 261 ## 4.2.0 262 263 2020-02-17 264 265 #### Added 266 267 - `ENABLE_DFT_MULTIARCH` cmake option can be used to build `kfr_dft` with multiple architectures support (x86/x86_64 only) 268 - `config.h` is generated during install step with all `#define`s needed for correct usage of installed libraries 269 270 #### Changed 271 272 - `CMAKE_INSTALL_PREFIX` is reset to empty on Win32 (can be overriden in cmake command line) 273 - C API binary is now installed using install command (`make install`, `ninja install` or `cmake --build . --target install`) 274 275 ## 4.1.0 276 277 2020-03-04 278 279 #### Added 280 281 - MSVC2019 support (16.4.4 is required) 282 - Stateless short_fir expression (thanks to https://github.com/slarew) 283 - `biquad_filter` constructor taking `std::vector<biquad_params<T>>` (thanks to https://github.com/BenjaminNavarro) 284 - `cabssqr` function (thanks to https://github.com/slarew) 285 - `moving_sum` expression (thanks to https://github.com/slarew) 286 - `gen_expj` generator (thanks to https://github.com/slarew) 287 288 #### Changed 289 - `cdirect_t{}` is now allowed in real dft plan methods for compatibility 290 - complex support for `convolve_filter` (thanks to https://github.com/slarew) 291 292 #### Fixed 293 294 - GCC debug build (thanks to https://github.com/BenjaminNavarro) 295 - `is_invocable_r_impl` fallback for missing C++17 feature 296 - `std::complex` compatibility (thanks to https://github.com/slarew) 297 - Various CI fixes 298 299 ## 4.0.0 300 301 2019-12-05 302 303 #### Added 304 305 - IIR filter design 306 - Butterworth 307 - Chebyshev type I and II 308 - Bessel 309 - Lowpass, highpass, bandpass and bandstop filters 310 - Conversion of arbitrary filter from Z,P,K to SOS format (suitable for biquad function and filter) 311 - Discrete Cosine Transform type II and III 312 - cmake uninstall target 313 - C API: 314 - DFT 315 - real DFT 316 - DCT 317 - FIR 318 - IIR 319 - Convolution 320 - Aligned memory allocation 321 - Built for SSE2, SSE4.1, AVX, AVX2, AVX512, x86 and x86_64, architecture is selected at runtime 322 - New vector based types: 323 - color 324 - rectangle 325 - point 326 - size 327 - border 328 - geometric vector 329 - 2D matrix 330 - Color space conversion: 331 - sRGB 332 - linear RGB 333 - XYZ 334 - Lab 335 - LCH 336 - HSV 337 - MP3 audio reading 338 - `aligned_reallocate` function 339 - `zip` and `column` functions 340 - `vec<vec<vec<T>>>` support 341 - New example: `iir.cpp` 342 - `biquad` that gets `std::vector` of `biquad_param`s 343 - `csqr` function 344 - `factorial` function 345 - `isreal` function 346 - `make_complex` function for expressions 347 - New optimization technique for vector element shuffle 348 - `std::get<>` support for `vec<>` 349 - C++17 structured bindings support for `vec<>` 350 - comparison operators for `cvals_t` 351 - compile time `cminof`, `cmaxof` functions 352 - `vector_width_for` constant to get vector width for specific architecture 353 - make_univector for containers and arrays 354 - univector: copy constructor optimization 355 - Custom `assertion_failed` 356 357 #### Changed 358 359 - C++17 compiler is required. Some of C++17 library features may be missing, in this case KFR uses custom implementation 360 - `cast` function now works with both vectors and expressions (expression versions was previously called `convert`) 361 - Memory alignment can be up to 32768 362 - New implementation for `cometa::function` using shared pointer 363 - `memory.hpp` allocation functions has been moved to cometa 364 - `cconj` has been moved to simd module (instead of `math`) 365 - dr_libs version updated 366 - All global constants are now `inline constexpr` 367 - enums moved out of architecture namespace 368 - `special_constant::undefined` removed 369 370 #### Fixed 371 372 - Fixed real DFT with raw poitners 373 - Bug with generators 374 - Fixed grid issues in dspplot 375 - Wrong index in `concatenate` 376 - `csqrt` function 377 - Fixed NaNs in `amp_to_dB` 378 - GCC 9 support 379 - Workaround for Clang 8.0 FMA code generator bug 380 - Flat top window 381 - SSE4.2 detection (previously detected as SSE4.1) 382 383 #### Notes 384 385 * MSVC support is limited to MSVC2017 due to ICE in MSVC2019. Once fixed, support will be added 386 * DFT is limited to Clang due to ICE in MSVC and broken AVX optimization in GCC 8 and 9. Once fixed, support will be added 387 388 ## 3.0.9 389 390 2019-04-02 391 392 #### Added 393 394 - `reduce` supports different types and containers other than `univector` 395 - Assignment operators for `univector`: `+=`, `*=` etc 396 - `concatenate` function to concatenate two expressions sequentially 397 - Audio file IO: `read_channels`/`write_channels` to read channels data directly without interleaving/deinterleaving 398 - `as_string`: support for `std::vector` 399 400 #### Changed 401 402 - `expression_scalar`: support for `vec<T>` 403 404 #### Fixed 405 406 - CPU detection in cmake subdirectory 407 - MSVC 2017 32-bit intrinsics 408 409 ## 3.0.8 410 411 2019-03-15 412 413 #### Added 414 415 - Ability to pass `random_bit_generator` by reference 416 - Tests for iOS ARM and ARM64 417 418 #### Changed 419 420 - `kfr::complex` is placed in `kfr` namespace 421 422 #### Fixed 423 424 ## 3.0.7 425 426 2019-03-13 427 428 #### Added 429 430 - Detected CPU is now saved to CMake cache 431 - Linux/AArch64 tests have been added to CI 432 433 #### Changed 434 435 - `mask<>` is now a specialization of `vec<>`. This allows using many `vec` functions for masks 436 - `short_fir` performance has been increased by around 50%-60% 437 438 #### Fixed 439 440 - Fixed the bug with infinitely loading in Intellisense 441 442 ## 3.0.6 443 444 2019-03-07 445 446 #### Added 447 448 - Android arm & arm64 tests have been added to CI 449 450 #### Fixed 451 452 - Fixed Android support (thanks to https://github.com/Jman420) 453 - Fixed building cpu detection tool in CMake subdirectory (thanks to https://github.com/Jman420) 454 455 ## 3.0.5 456 457 2019-02-21 458 459 #### Added 460 461 - DFT speeds have been improved by up to 15% on most modern cpus 462 - Support for MSVC 2017 463 - Support for GCC 7.3 464 - Support for GCC 8.2 465 - Support for resampling complex vectors (Thanks to https://github.com/ermito) 466 - Tests for various math functions no longer depend on MPFR 467 468 #### Changed 469 470 - Testo now allocates much less memory during long tests (x3 less than previously) 471 472 #### Fixed 473 474 - Building generators (Thanks to https://github.com/ermito) 475 476 ## 3.0.4 477 478 2019-01-08 479 480 #### Added 481 482 #### Changed 483 484 - `KFR_READCYCLECOUNTER` may be redefined to point to any function returning (pseudo-)random value 485 - Ability to disable random number initialization functions 486 487 #### Fixed 488 489 ## 3.0.3 490 491 2018-12-27 492 493 #### Added 494 495 - Partial compatibility for Visual Studio 2017 496 - Support for `KFR_USE_STD_ALLOCATION` 497 - `univector` support for `abstract_reader`/`abstract_writer` 498 499 #### Changed 500 501 #### Fixed 502 503 - Paths in `CMakeLists.txt` 504 505 ## 3.0.2 506 507 2018-12-19 508 509 #### Added 510 511 - More documentation 512 - dspplot: `freqticks` parameter 513 - dspplot: `freq_lim` parameter 514 - dspplot: `freq_dB_lim` parameter 515 - dspplot: `phasearg` parameter 516 - More tests for Sample Rate Converter 517 518 #### Changed 519 520 - Sample Rate Converter: now computes the transition width of the filter. This makes cutoff frequency more precise 521 - Sample Rate Converter: now uses Kaiser window with different β parameter for different quality. This leads to a better balance between transition width and sidelobes attenuation. 522 - Sample Rate Converter: `quality` parameter is now passed as runtime parameter rather than compile time parameter 523 - `fir.cpp` example has been extended to include examples of applying FIR filters 524 525 #### Fixed 526 527 - `amp_to_dB` function 528 - 24-bit audio reading 529 530 ## 3.0.1 531 532 2018-11-28 533 534 #### Added 535 536 - WAV file reading/writing 537 - FLAC file reading 538 - Audio sample conversion 539 - Interleaving/deinterleaving 540 - sample_rate_converter example 541 - Sample Rate Converter: `process`() function 542 543 #### Changed 544 545 - Assignment to an empty vector resizes it 546 - New documentation 547 - KFR IO is now built separately (only needed for audio file support) 548 549 #### Fixed 550 551 - Resampler bug: sequential calls to `resampler::operator()` may fail 552 553 ### 3.0.0 554 555 #### Added 556 557 - Optimized non-power of two DFT implementation 558 - Full AVX-512 support 559 - EBU R128 560 - Ability to include KFR as a subdirectory in cmake project 561 - Number of automatic tests has been increased 562 - C API for DFT 563 - Partial GCC 8.x support 564 - Ability to make sized linspace 565 - `fraction` type 566 567 #### Changed 568 569 - GPL version changed from 3 to 2+ 570 - Refactoring of DFT code 571 - KFR DFT is now built separately 572 - FIR filter now supports different tap and sample types (thanks to @fbbdev) 573 574 #### Fixed 575 576 - Various fixes for KFR functions 577 - FFT with size=128 on architectures with SSE only 578 - Small kfr::complex type fixes