ask for matrix multiply

hi, I have a question

[4x4] * [4x1] matrix multipying benchmark is 20ns(speed at 400MHz) in adsp-21489' data sheet.

but using matmmltf function [4x4] * [4x1] matrix multipying is measured 338ns(using cces)

how can I use matrix multiplying like benchmark?

best regard.

Edit Notes
fix

[edited by: booo at 7:56 AM (GMT 0) on 20 Feb 2020]

cces

benchmark

sharc

simd

21489