OpenFASOC should expose the main generator flow as an importable python modules

[proppy]

In order to be able to orchestrate generation from other python codebase (or jupyter notebooks) it would be nice if OpenFASOC exposed a python interface over the main generators flows that can be imported, https://github.com/idea-fasoc/OpenFASOC/blob/main/generators/temp-sense-gen/tools/temp-sense-gen.py

Note: https://github.com/idea-fasoc/OpenFASOC/blob/main/generators/temp-sense-gen/tools/TEMP_netlist.py seems to already exposes functions some lower level function that could be imported, but it seems that there is a fair amount of work also done from the outer scripts.

[msaligane]

Thanks @proppy
This is really great!
So let's say if a generator is updated, what would be the procedure to re-package it as a python module? is there an automated way?

[proppy]

for conda packages that could happen as part of hdl/conda-eda#192 (the project has a continuous integration environment that rebuild package from HEAD every night) or thru dedicated https://github.com/features/actions in this repo if we wanted to publish those package on PyPI too.

[msaligane]

@erictaur has been working on github actions and somehow got it to work. @erictaur any updates?

[mithro]

There is some work on trying to get GitHub Actions to upload to PyPi -- https://github.com/SymbiFlow/actions/tree/main/includes

https://github.com/SymbiFlow/actions/blob/main/includes/actions/python/publish-to-pypi-src/action.yaml
https://github.com/SymbiFlow/actions/blob/main/includes/actions/python/check-upload-publish-packages/action.yaml

[mithro]

We could make this work now with https://github.blog/2021-11-29-github-actions-reusable-workflows-is-generally-available/ (rather than my older https://github.com/mithro/actions-includes).

[mithro]

I believe these are deployed on https://github.com/chipsalliance/fasm/actions

[joamatab]

See PR #43 for this

I can also add some jupyter notebooks for the documentation

[msaligane]

That would be great! let's schedule some time early this week. Thanks Joaquin!

[saicharan0112]

@proppy suggested a good idea to expose each stage inside the generator as a python module which can end up something like this
Eg:

from openfasoc import tempsense
netlist=tempsense.gen_netlist()
tempsense.run_synthesis(netllist...)

Next steps could be interesting after we have something. @proppy do you want to add something to this?

[msaligane]

Sounds great to me!

[saicharan0112]

Here is the idea:

1. Convert the three verilog netlists (TEMP_ANALOG_hv.v, TEMP_ANALOG_lv.v and counter_generic.v present in folder ) into python modules to generate TEMP_ANALOG_hv_nl.v, TEMP_ANALOG_lv_nl.v and counter.v files.. I could just think of using print statements to build these files because of their size. Any suggestions?

2. Convert TEMP_netlist.py, which generates the tempsense netlist using the function gen_temp_netlist(), into importable python module.

In the end, I think we would be having something like this from openfasoc import gen_temp_netlist. We could add other netlists in the future. For eg: from openfasoc import gen_ldo_netlist or from openfasoc import gen_pll_netlist.

Just having the netlist, I think we could represent the tempsense flow in notebook.

Looks simple, does simple thing. Do you people think that this is worth?

[proppy]

I could just think of using print statements to build these files because of their size. Any suggestions?

Curious if you're thinking of using an HDL for helping you maintain a structured view of the netlists, see: https://github.com/drom/awesome-hdl for some recommendations (https://github.com/amaranth-lang/amaranth, https://github.com/PyHDI/veriloggen comes to mind or https://github.com/PyHDI/Pyverilog and https://github.com/byuccl/spydrnet as lower leve option).

If you prefer to manipulate raw text, I would recommend to look at the templating language like https://jinja.palletsprojects.com/en/3.1.x/.

[saicharan0112]

Thanks.. Understood. I am not familiar with those projects who shared here though. I will take a look.