16:54 <mabl[m]> <zyp[m]> "speaking of cxxrtl, how does the..." <- Is there data on how much faster it would be? I know there is an old branch in the repo. Would rebasing it on the latest simulation semantics be something where external contributions are welcome?

16:59 <whitequark[cis]> it depends on the design and testsuite

16:59 <whitequark[cis]> cxxrtl has a high startup cost, so re-using simulator instances and/or netlists are basically mandatory

16:59 <whitequark[cis]> a C++ compiler is needed, which would probably be provided by the ziglang pypi package, but integration needs to be written

17:00 <whitequark[cis]> also, there will be a significant additional engineering effort on top of rebasing that old branch, because the internals of the new simulator aren't quite compatible with CXXRTL yet (this was known at the time), nor is CXXRTL ideal for combining the two yet (also a known issue)

17:01 <whitequark[cis]> the performance of even long-running testsuites were limited by how the clock generation was done in Python by add_clock, with the delay of that dominating the time spent in C++

17:01 <whitequark[cis]> so something like a reactor would have to be added to CXXRTL to aid dynamic scheduling required by e.g. the Amaranth testsuite

17:01 <whitequark[cis]> in short, it's a lot more tricky than just rebasing it, which is why it's currently sitting there idle

17:16 <mabl[m]> That does indeed sound more than a mouth full. Thanks for taking the time to outline that work.