Perf/q6k neon kernel#787

Merged

michalharakal merged 3 commits into

perf/q6k-neon-kernel

Jul 2, 2026

Commits on Jul 2, 2026

perf(native q5k,q8_0): block-outer loop order for sequential weight reads

michalharakal
and
claude
committed
perf(native q6k): block-outer loop order (sequential weight reads)

michalharakal
and
claude
committed
perf(native q6k): fused Q8 int8 dot path (dotprod)

michalharakal
and
claude
committed