# llama.cpp b11976 adds a Hexagon IM2COL improvement

> The pre-release adds block-based HVX parallelization for Hexagon IM2COL and lists builds for several CPU, GPU and mobile platforms.

Oossa · 2026-10-29 · https://oossa.com/en/llama-cpp-b11976-adds-a-hexagon-im2col-improvement

The llama.cpp project published pre-release b11976 on October 29, 2026. Its main change is a Hexagon IM2COL update that processes output-column blocks across multiple threads using HVX parallelization, a way to run work across Hexagon’s vector hardware.

The release listing includes Linux and Android builds for Snapdragon devices with CPU, Adreno GPU and Hexagon NPU support. The notes do not report a speedup or say which models benefit.

## The facts

- Version: b11976, marked as a pre-release.
- The notes credit Max Krasnyansky as a co-author of the Hexagon change.

## Why it matters

Developers testing llama.cpp on supported Snapdragon hardware can try the updated Hexagon path, but the release notes give no measured performance gains.

## Sources & references

1. [ggml-org/llama.cpp b11976](https://github.com/ggml-org/llama.cpp/releases/tag/b11976) – llama.cpp, 2026-10-29

Last updated: 2026-10-29