Pynq-fud integration #1120
Pynq-fud integration #1120
Conversation
Previously, fud-pynq-script accepted a mapping and returned a string, now ittakes a path to an xclbin file and returns a dictionary conforming to json standards. The new execution.py retains all of the previous config settings while simplifying the kernel execution by running fud-pynq-script. execution.py is responsible for turning the script's output into a json file
…pynq-fud-integration
fud/fud/stages/xilinx/execution.py
Outdated
| def import_libs(): | ||
| """Import optional libraries""" | ||
| try: | ||
| import pyopencl as cl # type: ignore | ||
|
|
||
| self.cl = cl | ||
| except ImportError: | ||
| raise errors.RemoteLibsNotInstalled |
There was a problem hiding this comment.
Wasn't sure about this, should the pynq import from pyproject.toml go here instead?
There was a problem hiding this comment.
The reason this is inside a try-catch inside a function is because the fpga dependencies are not installed by default. The program only throws an error if you try to run the xilinx stage without installing the fpga dependencies in the flit configuration.
I would recommend moving the
from fud.stages.xilinx import fud_pynq_script
and using a self.pynq_script field to access it instead of importing it on the top which will cause errors for default fud installations
There was a problem hiding this comment.
Overall looks good! Address the comments about dependency to make sure that a fresh, default install of fud, when following these instructions does not break.
Once done, merge at your discretion
This PR can close #1114.
A new pynq-fud-script can take in arbitrary data, run a kernel on it, and output the relevant memories. Integrates into
fud e foo.xclbin --from xclbin --to fpga ...commands.This replaces the old OpenCL way of running a kernel that fud used to use.
Unfortunately this does not address #1037 #872 #856.