search

Xilinx / BNN-PYNQ

Quantized Neural Networks (QNNs) on PYNQ
https://xilinx.github.io/finn/
BSD 3-Clause "New" or "Revised" License
671 stars 306 forks source link

BNN-PyNQ on ZCU-104 #127

Closed kamilswitek closed 4 years ago

kamilswitek commented 4 years ago

Hello. Is it possible to create bitstream for the ZCU-104 board with one of those cnv networks from the repository?

giuliogamba commented 4 years ago

Yes, it is. I suggest to start from the project targeting the Ultra96 board and adapt the scripts available here to target the ZCU104.

kamilswitek commented 4 years ago

Thank you for the answer. I've found the following dependencies and have some questions:

giuliogamba commented 4 years ago

Hi, here some answers:

Thank you for the answer. I've found the following dependencies and have some questions:

  • bnn/libraries/ultra96/ There are .so files named *ultra96.so. Are they specific to the platform and have to be swapped(how?) or can I leave them as they are

You need to recompile on board (ZCU104) using the same compile scripts as for ultra96.

  • bnn/src/library/driver/platform-xlnk.cpp There are ADDR_REG_BASE defines - which addres base value is appropriate for ZCU104?

The same as Ultra96

  • bnn/src/library/script/ultra96/make-vivado-proj.tcl Is it enough to swap the part name here? (in set_proj_part)

I would not run the scripts, but manually do a new vivado project and add the HLS generated IP.

  • bnn/src/library/script/ultra96/ultra96.tcl Where can I get a proper file for ZCU104 from?
  • bnn/src/library/script/ultra96/ultra96.xdc I found the .xdc file for ZCU on XILINX page, is it ok to just swap it with the one for ultra96?

There's a board support package for ZCU104 in Vivado, you should use that.

  • bnn/src/network/make-hw.sh I've found the part name again as well as the target clock. Is the clock somehow constrained for the board or can I use the same as for ultra96?

Same clock as for Ultra96 should work.