Closed ezrosent closed 6 months ago
However, I'd love to see at least a few more tests before we merge. Perhaps more unstructured bril programs that don't involve pointers? In fact, we could randomly generate some cfgs and see what happens
There are 3 or 4 more failing CFGs in the issues that I'm hoping to fix in follow-ups, but I can look at adding some more.
I'm not sure where the failing test is from either... that snapshot is present on this branch.
This PR adapts the algorithm from the original RVSDG papers for restructuring branches. This will let us translate more programs successfully.
I suspect there is more work to do on the translation back to CFGs now that this is up.
This PR also includes a few cleanups around making LLVM tests optional, and adding some switches to dump CFG visualizations to
/tmp
; both of those use cargo features.