llama.cpp Release b11045 Adds ROLL Operation Support for Qualcomm Hexagon DSPs
Open-source LLM inference engine llama.cpp has released build b11045, introducing support for the ROLL tensor operation on Qualcomm Hexagon DSPs (PR #29105). Expanding tensor operation support for Qualcomm Hexagon improves hardware-accelerated LLM inference performance on Snapdragon-powered mobile and edge platforms. This enables more efficient execution of neural network architectures directly on mobile DSP hardware. The release centers on PR #29105, which implements the ROLL tensor manipulation operator for the Hexagon execution backend. Pre-built release binaries were published across various platforms including Android arm64, Linux, Windows, macOS, CUDA 12/13, Vulkan, and SYCL.
## BACKGROUND
llama.cpp is a high-performance C/C++ library designed for running Large Language Models locally across diverse hardware architectures. Qualcomm Hexagon is a dedicated digital signal processor (DSP) integrated into Snapdragon SoCs that provides specialized hardware acceleration for vector math and machine learning tasks on mobile devices.