Change8

b9128

📦 llama-cppView on GitHub →
2 features🔧 4 symbols

Summary

This release focuses on internal optimizations within the Hexagon and HMX backends, specifically eliminating scalar VTCM loads and improving scale handling.

✨ New Features

  • Introduced hvx_vec_repl helpers in hexagon backend for splat-from-vtcm use cases.
  • Added hvx_vec_repl_2x_f16 helper and consolidated existing replication helpers in hexagon.

Affected Symbols