Open Source LLM

LLVM/Clang 20.1 Released With AMX-AVX512, AMX-FP8, AVX10.2, AMD GFX950 & Much More

LLVM 20.1 was just tagged in Git as the first stable version of the LLVM 20 compiler stack including sub-projects like the Clang 20 C/C++ Compiler.

LLVM/Clang 20.1 is the newest half-year feature release for this prominent open-source compiler stack that is widely used by many different vendors for a variety of purposes from traditional C/C++ compilation to being heavily used for GPU compute stacks, growing AI uses with MLIR, and much more.

Some of the many changes to find with the LLVM 20.1 stack include:

- AMD GFX950 support work for what is presumably going to be the Instinct MI350.

- AMX-AVX512 support was merged.

- AMX-FP8 support for Intel Diamond Rapids.

- Initial AVX10.2 support .

- More work on the latest C and C++ standards.

- The flang-new modern Fortran compiler was renamed back to "flang".

- TySan was merged as a sanitizer for type-based aliasing violations.

- The SPIR-V back-end is promoted to being "official" and enabled by default after previously carrying an "experimental" tag.

- Clang 20 now supports the Xtensa CPU target .

- An initial telemetry framework for LLVM .

- Tenstorrent TT-Ascalon-D8 RISC-V CPU support .

- IBM SystemZ arch15 support for what is presumably the IBM z17 / Telum II processor.


LLVM 20.1 sources can be downloaded from GitHub . The LLVM 20.1 release is arriving one week early due to no blocker bugs having come up in recent days.