Change8

b9928

📦 llama-cppView on GitHub →
3 features🐛 3 fixes🔧 9 symbols

Summary

This release focuses heavily on internal optimizations and robustness improvements for the hexagon backend, particularly for matrix multiplication and attention kernels, alongside asynchronous queue handling and workpool enhancements.

✨ New Features

  • Introduced new VTCM layouts and improved pipelines for MUL_MAT, MUL_MAT_ID, and FLASH_ATTN_EXT operations within the hexagon backend.
  • Added support for queue wakeup and made suspend asynchronous in hmx-queue to reduce hmx-lock latency.
  • Implemented hybrid polling in hex-threads workpool.

🐛 Bug Fixes

  • Fixed trailing spaces in hex-mm.
  • Disabled KleidiAI support for macOS Apple Silicon (arm64) builds.
  • Disabled openEuler builds.

Affected Symbols