Nice!

[udif]

It's good to see another SystemRDL tool to enhance the echosystem. (though I haven't tried your tool yet).

I'm curious why have you decided to write your own tool from scratch instead of using the Python-based PeakRDL.
After all, Python development is quicker, and a SystemRDL tool doesn't nead real-time performance.

[vowstar]

Thanks for taking a look.

You're right that performance isn't the reason. It's integration: qsoc is entirely C++, and I wanted the register library as a plain compile-and-link dependency rather than shipping a Python runtime beside it.

The one thing the library adds on top is reserved bit handling. Elaboration fills every gap with a reserved field so back end generators don't each have to redo that bookkeeping. The standard calls those gaps reserved and unused but doesn't ask a tool to create them, so it's a convenience rather than something PeakRDL is missing.

Edited: my earlier point about register width was wrong. The standard does require regwidth to be a power of two (10.6.1), and I'm removing that relaxation from the library rather than keeping it.