Refactoring plan of SILVisitor for LTO

[schuett (Michael Schutt)]

If you have the time, have a look at:

2016 LLVM Developers’ Meeting: M. Amini & T. Johnson “ThinLTO: Scalable and Incremental LTO”

It gives detailed information about the cross-TU analysis and the overall architecture.

Feel free to ignore me.

I totally agree that the LTO backend does optimisations and codegen.

[kateinoigakukun (Yuta Saito)]

Thanks for your information!

[schuett (Michael Schutt)]

Thanks for your work. I am looking forward to use LTO in Swift.

[Andrew_Trick]

@kateinoigakukun

In my opinion, the Swift compiler driver should focus on single module compilation and build systems like SwiftPM should drive these kinds of cross-module cooperation things.

Yes.

@kateinoigakukun

Right, I know those issues. I'm sending patches to fix them now.

Great. First step is having a working serialized SIL.

@kateinoigakukun

Yes, as you said, this emulates LLVM thin-LTO architecture but not performs at link-time
Do you have any idea of name for this optimization?

To me, this is just cross-module optimization. You could call it thin-cross-module-optimization if you want to distinguish it from the less-scalable approach.

@kateinoigakukun

@Andrew_Trick

could (should?) be embedded within the .swiftmodule, but separating them allows for a single merged summary file

.sib/.sil and .swiftmodule.summary have same lifetime because they depends on internal things that are not exported as .swiftmodule. So I think that it should be better to embed summary info in .sib rather than .swiftmodule.

I don't understand that. An important distinction between .swiftmodule and .sib, is that the former needs to be loaded when its imported by other modules and the later is purely used for codegen within a module. There's nothing cross-module about the .sib. The module summary is only useful to export information that's already in the .sib to other modules. Both the .swiftmodule and the summary can embed assumptions about code within the .sib files.

I'm perfectly fine using separate files anyway. I think that makes debugging easier.