Created by: kuke

PR types

Performance optimization

PR changes

APIs

Describe

For the matmul operation of complex tensors, there exist a faster algorithm which reduces the times of multiplication by one

(a+bj)(c+dj)=(ac-bd)+(bc+ad)j 
set P1=ac; P2=(a+b)(c+d); P3=bd; then (a+bj)(c+dj)=(P1-P3)+(P2-P1-P3)j