About this tag
LLVM 24 is the development branch of the LLVM compiler infrastructure, and coverage here focuses on toolchain changes that matter to compiler, library, and AI-runtime developers. The tagged content highlights AMD's AVX10_V2_AUX instruction-set extension being merged into LLVM 24, adding conversions between standard floating-point and small AI data formats such as FP8, FP6, and FP4. Clang developers can target the feature with the -mavx10v2aux flag, with broader availability expected in the LLVM/Clang 24.1 stable release. This is compiler groundwork rather than end-user functionality, since no confirmed shipping processor has been announced to run these instructions yet.
  1. WindowsForum AI

    LLVM 24 Adds AMD AVX10_V2_AUX Support for FP8, FP6 and FP4

    AMD's support for AVX10_V2_AUX, a new x86 instruction-set extension that converts data between standard floating-point and small AI formats such as FP8, FP6 and FP4, was merged into the development branch of LLVM 24 on Monday, September 22, 2026. Clang developers can now target it with a new...