[clang] [llvm] [CGData][ThinLTO] Global Outlining with Two-CodeGen Rounds (PR #90933)

[Nuri Amari via cfe-commits]

================
@@ -1558,6 +1562,60 @@ class InProcessThinBackend : public ThinBackendProc {
     return BackendThreadPool.getMaxConcurrency();
   }
 };
+
+/// This Backend will run ThinBackend process but throw away all the output from
+/// the codegen. This class facilitates the first codegen round.
+class NoOutputThinBackend : public InProcessThinBackend {
----------------
NuriAmari wrote:

If I'm following correctly, it's not as if this backend really throws results away, but it writes the produced object files into the scratch buffer rather than to files. It also writes the optimized bitcode files to disk right?

I don't have a strong opinion, but maybe we could name the backend / adjust the comment to reflect this. I wouldn't even be opposed to naming the backends something specific to two-codegen rounds, as I don't see anyone using them for other purposes.

Lastly just an idea:  We could just hold the optimized bitcode in a similar buffer in memory, rather than writing them to disk and reading them again between rounds. Might run a bit faster.

https://github.com/llvm/llvm-project/pull/90933