There's a lot on the table with this six-month update to LLVM. Some of the LLVM 19 release highlights include:
- C++17 support is now considered complete with remaining feature support addressed.
- Various additional C++20 and C++23 features have been implemented and even a few C++2c features.
- Clang 19 allows enabling C2y language support with the -std=c2y option.
- Support for C23's #embed .
- Clang deprecated the "-Ofast" option with users recommended to just use "-O3 -ffast-math" for achieving the same impact.
- Various Clang diagnostic improvements.
- Retiring of 3DNow! specific ISA intrinsics and code generation support.
- Intel Knights Mill and Knights Landing support was retired too.
- Support for Arm Cortex R82AE, A78AE, A520AE, A720AE, A725m X925, N3, V3, and V3AE processor cores.
- LLVM on Arm also now supports ELF pointer authentication and a few other smaller new features.
- LLVM 19 on RISC-V adds full support for the experimental Zabha (Byte and Halfword Atomic Memory Operations) extension and experimental support for the Ssnpm, Smnpm, Smmpm, Sspm, Ssqosid, and Supm 1.0.0 Pointer Masking extensions.
- RISC-V with LLVM 19 also now supports "-mcpu=native" handling when on Linux 6.4 and detecting the CPU features via hwprobe.
- The AMDGPU LLVM back-end supports new intrinsics as well as other features and GFX12 (RDNA4) preparations.
LLVM 19.1 downloads available from GitHub . More details on today's LLVM 19.1 release via LLVM.org .