Fix Gemma4 pure-decode full attention

[shsym]

Gemma4 31B full/global layers (32 Q / 4 KV heads) corrupted in pure decode because the manual full-attention path materialized GQA with ggml_repeat_4d instead of letting ggml_mul_mat apply grouped-query broadcast.

This keeps K/V grouped like the slow path and views the packed decode mask as a single-query mask for manual SDPA.

Verified with clean 31B Q4 long-gen, clean 31B Q8, and no 12B Q4 regression.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 9287120e-523d-4dbb-867c-797acb7e5e05

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch task/gemma4-pure-decode-full-attention

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}