- Add load_q_mma_frags<KD>() shared template in attn_mma_utils.cuh - Replace ~15 duplicated Q-load lines in 3 MMA kernels - Unify section header comment style to // ---- Section ---- - Remove duplicate separator line in attn_mma_utils.cuh
- Add load_q_mma_frags<KD>() shared template in attn_mma_utils.cuh - Replace ~15 duplicated Q-load lines in 3 MMA kernels - Unify section header comment style to // ---- Section ---- - Remove duplicate separator line in attn_mma_utils.cuh