llama.cpp icon indicating copy to clipboard operation
llama.cpp copied to clipboard

Feature Request: Compile bug: QCM6490 Platform Support

Open hiwudery opened this issue 1 month ago • 1 comments

Prerequisites

  • [x] I am running the latest code. Mention the version if possible as well.
  • [x] I carefully followed the README.md.
  • [x] I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed).
  • [x] I reviewed the Discussions, and have a new and useful enhancement to share.

Feature Description

Hi @chraac

I'm using the QCM6490 platform and noticed that the current codebase doesn't support the Hexagon v68 DSP. Could you let me know if there are any plans to add support for this SoC in the future?

Thank you Jack

Motivation

Market relevance: QCM6490 is widely used in automotive, robotics, and industrial IoT applications, so supporting this platform would benefit a large user base

Possible Implementation

No response

hiwudery avatar Nov 08 '25 14:11 hiwudery

Hi @hiwudery

Sorry, Hexagon v68 is a bit old now—I’m focusing on v73 and newer architectures. If the public API differences between v68 and the newer toolchains are small, you might be able to add support yourself by adapting the existing code paths. Most of the core ideas should carry over with minor tweaks (for example: add CMake targets to build the v68 libs, then extend the loader to select the right library/path).

Also, I’ve migrated my work toward the official ggml-hexagon backend (recently merged in this PR: https://github.com/ggml-org/llama.cpp/pull/16547). They have a faster matmul using int8 rmpy internally, so it’s worth a look—both for performance ideas and a more future‑proof base.

If you try a v68 adaptation, feel free to open a PR—I’m happy to take a look.

chraac avatar Nov 09 '25 16:11 chraac