Releases: shibatch/tlfloat
December 2024 Release
- Add all-integer square root
- Add BFloat16 math funcs
- Fix a bug in conversion from hex string
- Benchmark utility now works properly when LTO is enabled
- Add functionality to find latest version of clang to the toolchain file, which is used for cross compiling
- Fix detection of compiler support for float16 and bfloat16
November 2024 Release
- Experimental BFloat16 support is added.
- The library can now be built with both clang and clang-cl on Windows.
- Polymorphic functions are added to C++11 API.
- Cross-compilation support is improved.
- LTO option is added.
October 2024 Release
- Add sinpi, cospi and tanpi
- Add big endian support
- Add generation of pkgconfig file
Add benchmark tool
- New benchmark tool is added
- Adopted a new algorithm for faster division
- Execution speed of TLFloat in quad-precision is comparable to libquadmath.
Add octuple constants to C API
- Octuple constants are added to C API
- Add %% support to printf
- Support for newer version of clang for Windows
- More testing
- Minor bug fixes
Adopt Community Guidelines to Ensure Fair Rewards for OSS Projects
In this release, TLFloat adopts the Community Guidelines to Ensure Fair Rewards for OSS Projects as a code of conduct.
This is what I created, and TLFloat is the first project that adopts these guidlines.
See https://github.com/shibatch/gtefr for more detail.
Bug fixes and better portability
This release adds several small features while fixing bugs and improving portability.
New features : ++ and -- operators, fpclassify, and formatting in ostream << operator.
Fixed bugs : ldexp and frexp.
Improved portability : compiler feature detection is now improved and the library is now tested on macOS and 32-bit arm. Although not added to the CI configuration, it is confirmed that the library passes the self test on risc-v systems.
Add error functions, tgamma, etc.
This release adds erf, erfc, tgamma and remquo.
Mathematical constants are also added.
Add C++11 API for handling 128-bit integer
With this release, 128-bit integer can be used on MSVC with C++11 code.
Add C++11 API
In this release, tlfloat_quad and tlfloat_octuple types can be used as classes with overloaded operators.