On August 7th, the release of GCC 16.2 — the second stable release of the GCC 16 branch. This new version is predominantly corrective in nature: the developers refer to it as a bug-fix release, containing fixes for regressions identified in GCC 16.1 compared to previous releases.
A total of 102 known bugshave been fixed since the April GCC 16.1 release. The fixes affect various components of the compiler and supported architectures; there is no major single change defining this release.
GCC 16.2 is not a new functional release: the main features of the GCC 16 branch were introduced in GCC 16.1, released on April 30. Among the most notable changes in this branch:
- C++20 has become the default C++ mode instead of GNU++17. To use the previous behavior, it is now necessary to explicitly specify the corresponding parameter -std=. Experimental support for C++20 modules must still be enabled separately.
- The implementation of C++26continues, including experimental static reflection, contracts, constexpr exceptions, and other proposals for the future standard.
- An experimental compiler for the language Algol 68 — ga68.
- has been included in GCC. Support for processorsAMD Zen 6 (-march=znver6),
- Intel Wildcat Lake, and Nova Lake has been added. Auto-vectorization with AVX-512 has been improved. Support forArmv9.6-A,
- Apple M4 and M5, NVIDIA GB10, and new Arm processor cores has been added in the AArch64 implementation; the possibility of applying Memory Tagging Extension via -fsanitize=memtag has also been included. For offloading to AMD GPUs, overhead has been significantly reduced when launching OpenMP/OpenACC regions, and experimental support for accelerators has been introduced..
AMD Instinct MI300 (gfx942). Thus, GCC 16.2 is primarily intended for users wishing to transition to the new branch after the regressions uncovered following the initial release have been addressed. The source code for GCC 16.2 is available onthe official project mirrors.
Source: linux.org.ru
