ggml-hip: enable -ffast-math for HIP builds by a-huk · Pull Request #23862 · ggml-org/llama.cpp

Benchmarks Benchmarked on gfx1151 (RDNA3.5/Strix Halo, 40 CUs) against latest master (2f6c815). These two models cover both the FA paths. Qwen3.5-27B Q4_K_M — head_dim=256 (TILE path), FA=1, prompt t/s: pp upstream + fast-math delta 512 307 321 +4.6% 2048 302 323 +7.0% 4096 300 312 +4.0% 8192 288 30