question about PDGEMM

[libuxiao]

Recently,I am using scalapack to compute a problem D-B*inverse(A)*C. I use PDGEMV to compute inverse(A)*C,and then call the subroutine PDGEMM to compute D-B*temp,there temp is the result of inverse(A)*C.It seems that the compute time taken by PDGEMM is too long,and the parallel efficiency is not very good.I think it's hard to imagine.Is there any other way to make it faster?