How to initialize the deep learning ip core generated by deep learning hdl toolbox

29 visualizaciones (últimos 30 días)
Hi,
I’m currently working with the Deep Learning HDL Toolbox and would like to integrate the generated IP core into my own RTL design.
I used Wireshark to capture the traffic between MATLAB and the board, and I can now “run” the core by replaying those packets.
However, I haven’t been able to reverse-engineer the full initialization sequence: loading the network weights into external DDR, programming the configuration registers, and completing any hidden hand-shakes that happen before the first inference.
When I let MATLAB drive the board the same bit-stream works perfectly, so the missing piece is clearly the initialization flow.
Could you point me to the register map, the required power-on sequence, or any documentation that explains how to bring the Deep Learning IP out of reset and validate that it is ready to accept input data?
Thank you very much for your help.
Best regards.
Looking forward to your reply.
  1 comentario
dpb
dpb el 31 de Dic. de 2025 a las 16:54
In it, @Kiran Kintali seems to have provided an Answer that apparently got you along the right track and says to not hesitate to contact Tech Support, but I also note you've not seemed to respond with an Accept or acknowledgement of many of those prior posts.
For an external board, go to the manufacturer's website and get the product datasheet. It'll have all the info on communicating with it.

Iniciar sesión para comentar.

Respuesta aceptada

KH
KH hace alrededor de 3 horas
Hi,
I have found how to solve this problem.
Set the interface of dlhdl.Target to file.
hTarget = dlhdl.Target('Xilinx','Interface','File','Filename','******\fpga_init.dln');
After hW.deploy, a file will be generated, follow the commands in this file, the ipcore can be initialized.

Más respuestas (0)

Productos


Versión

R2024b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by