Skip to content

hexagon: add alternative Hexagon NPU backend implementation - #27535

Open
zhouwg-jeffzhou wants to merge 1 commit into
ggml-org:masterfrom
zhouwg-jeffzhou:pr_to_upstream_v2
Open

hexagon: add alternative Hexagon NPU backend implementation#27535
zhouwg-jeffzhou wants to merge 1 commit into
ggml-org:masterfrom
zhouwg-jeffzhou:pr_to_upstream_v2

Conversation

@zhouwg-jeffzhou

Copy link
Copy Markdown

Overview

Follow‑up to #26373

This PR adds FastRPC-based ggml-hexagon, an alternative backend for Qualcomm Hexagon NPU on Android, which can coexist with the existing official dspqueue-based ggml-hexagon implementation.

Additional information

details can be found at RFC-26227

Requirements

  • I have read and agree with the contributing guidelines
  • AI usage disclosure: YES - used AI assistance for implementation help. I reviewed every changed line, tested locally, and take full responsibility for this PR.

@zhouwg-jeffzhou
zhouwg-jeffzhou requested a review from a team as a code owner August 22, 2026 11:09
@github-actions github-actions Bot added ggml changes relating to the ggml tensor library for machine learning Hexagon labels Aug 22, 2026
@ggml-gh-bot

ggml-gh-bot Bot commented Aug 22, 2026

Copy link
Copy Markdown

Hi @zhouwg-jeffzhou, thanks for your contribution!

Per our contribution guidelines, the automated PR checker found the following issue(s) that need your attention:

  • Multiple open PRs from a new contributor: We limit new contributors (those without a previously merged PR) to 1 open PR at a time. You currently have 2 open PRs.

  • Large PR: Large changes require prior discussion (e.g. an issue or RFC) and maintainers may not be able to review this PR as-is. Consider splitting it into smaller, focused PRs.


Please note that maintainers reserve the right to make final decisions on PRs. If you believe there is a mistake, please comment below.

@zhouwg-jeffzhou

Copy link
Copy Markdown
Author

@ggml-gh-bot, thanks for your kind reminder. I have already converted my opening PR to drat.

@zhouwg-jeffzhou

Copy link
Copy Markdown
Author

@max-krasnyansky, sorry to bother you. I have carefully studied your feedback in #26373, thank you very much for providing many valuable suggestions. This new PR addresses your suggestions, and I would like to make some solid contributions to ggml‑hexagon.

@zhouwg-jeffzhou
zhouwg-jeffzhou force-pushed the pr_to_upstream_v2 branch 4 times, most recently from 3b07df8 to e5001b2 Compare August 22, 2026 15:12

@zhouwg-jeffzhou zhouwg-jeffzhou Aug 22, 2026

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

after submitting the PR, I spent several hours reviewing source files and scripts on the GitHub page, trying to strictly align with Qualcomm's original behavior.

It sees that ggml/src/ggml‑hexagon/CMakeLists.txt and ggml/src/ggml‑hexagon/htp/CMakeLists.txt are already aligned with upstream.

I am not sure whether I should modify the backend name from Hexagon‑cDSP to HTP, so that the backend name is also strictly aligned with Qualcomm's original behavior.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ggml changes relating to the ggml tensor library for machine learning Hexagon

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants