[GSoC 2021] Alive2 for SIL

[justinfargnoli]

@codafi, here's my proposal! I have also submitted it through the GSoC portal. Anyone should feel welcome to check it out and comment on it

---

@codafi

it would be interesting to submit constraints that roughly equate to "this pass doesn't unnecessarily introduce refcounting traffic"

To do this, are you anticipating needing to modify/extend Alive2 so that it understands ownership?

---

@codafi

As for the second question, Alive2's IR is pretty well tied to LLVM, but then again SIL has a battle-tested lowering to LLVM IR already written. It seems to me we could exploit that and get things up and running quicker. But it's just one way of doing things.

Based on this, what do you think about building the system by: (This is different than what I currently have in my project proposal.)

1. Writing a program that takes SIL as input, passes that SIL into a SIL optimization pass to get the target SIL, uses the Swift compiler to lower the source and target SIL to LLVM IR, and invokes Alive2 with that source and target LLVM IR.
2. Extend that program to lower the source and target SIL directly to Alive2's IR, instead of using the Swift compiler to lower the SIL to LLVM IR.

Number 1 could be an alternative to the second deliverable that I described in my project proposal.