b10310
Breaking Changes📦 llama-cppView on GitHub →
⚠ 1 breaking🐛 2 fixes🔧 3 symbols
Summary
This release introduces aarch64 HWCAP fallbacks and fixes fp16 variant detection in ggml. It also renames the `has_fp16_va` field to `has_fp16` for better clarity on the FEAT_FP16 extension.
⚠️ Breaking Changes
- The `has_fp16_va` field has been renamed to `has_fp16`. This change gates the entire FEAT_FP16 extension, including scalar and vector half-precision arithmetic.
Migration Steps
- Update code that references `has_fp16_va` to use `has_fp16` instead.
🐛 Bug Fixes
- Added fallback definitions for missing aarch64 HWCAP bits.
- Required HWCAP_ASIMDHP for aarch64 fp16 CPU variants.