Change8

b8361

📦 llama-cppView on GitHub →
2 features🐛 1 fixes🔧 5 symbols

Summary

This release resolves a critical crash issue on AMD APU/iGPU systems by making hipMemAdviseSetCoarseGrain calls non-fatal. It also introduces enhanced debugging logs for memory diagnostics on APU hardware.

✨ New Features

  • Added pre-allocation debug logging (GGML_LOG_DEBUG) to help diagnose memory issues on APU systems.
  • Store totalGlobalMem in device info for APU systems.

🐛 Bug Fixes

  • Fixed APU compatibility issue on AMD APU/iGPU devices by treating hipMemAdviseSetCoarseGrain as an optional performance hint (calling it without error checking and clearing the error with hipGetLastError()). This prevents crashes on systems like AMD Strix Halo.

Affected Symbols