[bfgoodrich]

SGEMM / DGEMM using AMX2 (the first M1 has AMX2. The A14 has AMX1) is approximately 100% faster than the same running with NEON, which is already a specialized vector math system.