Search code examples
carmlinear-algebraxilinxzynq

Efficient implementation of matrix multiplication ARM cortex A9 - Xilinx SDK

Is there any simple way-library to efficient (max possible speed) implement linear algebra on an ARM CortexA9 dual core using Xilinx SDK?

I am using a zybo z7 developememt board with a dual core Arm proccesor and i want to implement a simple neural network with one convolution layer followed by a dense one, on Xilinx SDK. Specificaly, to tranfer a python numpy based model on Arm. I read some manuals for ARM and SIMD library but i don't want to dive so deep.

An easy way for me is to use a library and do the multiplication/dot product/convolve etc by itself (fast) like numpy in python and avoid pure for...loop syntax. An example would be nice!

Thank for your time

Solution

  • You can try the Eigen library used by Tensorflow to implement the matrix calculations, or you can even try to use TensorFlow lite which is already tested with the ARM-Cortex M series of processors.