llama.cpp Release b11167 Adds Hexagon Backend Support for I32 CPY and CONT
llama.cpp release b11167 introduces support for 32-bit integer (I32) copy (CPY) and contiguous (CONT) operations on the Qualcomm Hexagon backend. This update expands tensor operation coverage for Snapdragon hardware acceleration. Expanding tensor operator coverage on Qualcomm's Hexagon architecture enables smoother execution of large language models on mobile and edge devices. It improves compatibility and hardware acceleration for Snapdragon-powered Android and Linux systems. The patch implements pull request #29379 to handle I32 tensor memory formatting and copying in the Hexagon backend. Pre-compiled binaries with Snapdragon Hexagon NPU support are provided for Linux arm64 and Android arm64 platforms.
## BACKGROUND
llama.cpp is a popular open-source inference engine designed to execute large language models efficiently across a wide variety of hardware architectures. Qualcomm Hexagon is a programmable digital signal processor (DSP) and NPU integrated into Snapdragon System-on-Chips to accelerate AI workloads on mobile devices.