Documentation of all notable changes to the intx project.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

• Fix MSVC build in preprocessor conformance mode #327

• Consequent type aliases and literal suffixes: #317
  • uint128, 1_u128
  • uint192, 1_u192
  • uint256, 1_u256
  • uint320, 1_u320
  • uint384, 1_u384
  • uint448, 1_u448
  • uint512, 1_u512

• Restructure of the code by using friend operator and if constexpr. #318 #319 #320 #321 #322 #324

• Support constexpr context execution for all arithmetic including division. #305 #311 #314

• Literal operators changed to consteval. #298
• Some C++20 modernization. #297 #308

• Fixes to division for constexpr context. #309 #310

• Fixed buggy __builtin_subcll in Xcode 14.3.1 on arm64. 294

• C++20 is now required to use intx. #287 #290

• Fixed buggy __builtin_subcll in Xcode 14.3.1 on arm64. 294

• Increase tolerance to some compiler warnings (e.g. unknown pragma, unknown attribute) in case intx.hpp is used without CMake assistance. #286

• Fixed 32-bit Windows builds: the _umul128 intrinsic is not available there. #283

• Optimizations for GCC compiler. #279 #281
• Required CMake version increased from 3.10 to 3.16 #276
• Simplification of many operators declarations by using friend inline functions. #270 #271
• Benchmarks updates and fixes. #274 #275 #277

• Added support for big-endian architectures. #257
• Extend endian-specific load/store helpers to work with builtin integer types. #265

• The addc() and subc() have been optimized with compiler's builtins if available. #250 #251 #253
• Fixed and enabled CI testing for architectures other than x86. #255
• Small multiplication optimization. #261
• Small division optimization. #263
• Small comparison operators optimization for the uint256 type. #264

• Added support for bswap() of small unsigned types: uint8_t, uint16_t and uint32_t. Previously arguments where promoted to uint64_t what produced invalid results. #247

• The int128.hpp and intx.hpp header files have been merged. Now the whole library is included in single intx/intx.hpp file. #242
• The addmod() implementation has optimized path for elliptic curve context. #206
• The implementation of operator== now explicitly performs XOR folding. #245

• The internal representation of unsigned integers has been changed to be an array of 64-bit words. This provides types composed of any number of words (e.g. uint384). Previously only power-of-two numbers of words were allowed. This is big change that affects implementation of many operators. #212 #213
• Subtraction has been slightly rewritten to make it easier for compilers to optimize it. The performance now matches the addition. #189
• Multiplication API has been cleaned up by using std::is_constant_evaluated-like helpers internally. #195 #205
• Shift API has been changed to use uint64_t as shift amount on the fundamental API level. #202

• The utility macros (e.g. INTX_UNREACHABLE()) have been redefined. #182
• Test and benchmarks improvements. #175

• Support for using intx dependency as a CMake subproject (e.g. by git submodules or FetchContent). #160
• Support for building intx exceptions handling disabled (-fno-exceptions) and without RTTI (-fno-rtti). #164

• Converted to header-only library. #155 #156
• Addition / subtraction optimizations. #115 #117 #130
• Division optimizations. #129 #131 #134 #141 #147 #171
• Exponentiation optimizations. #128 #146

• The unused builtins.h header has been deleted. #173

• Added the as_bytes() casting helper. #106

• The endian-specific API for converting intx types to/from bytes has been reworked. #107
• The clz() is now constexpr and produces correct answer for zero inputs. #108

• New addmod() and mulmod() procedures have been added for the uint256 type. #101

• Pedantic compiler warnings have been fixed. #98
• Performance of the division algorithm increased up to 40% when dividing 256-bit values by 128-bit and 64-bit ones. #99