[PATCH] D18763: [ThinLTO] Move summary computation from BitcodeWriter to new pass

[Mehdi AMINI via llvm-commits]

joker.eph added a comment.

I think this introduces a distinction between the module level summary and the combined index, both in the memory representation and the usage (the client would deal differently with both).
I wonder if it is not a sign that some more restructuring could make it cleaner, with a separate class for the combined index and the module summary one.

Also I'm not totally sure why we're still using value id in the module summary case and GUID for the combined index?


================
Comment at: include/llvm/IR/ModuleSummaryIndex.h:76
@@ +75,3 @@
+  /// Accessor for GUID value
+  GlobalValue::GUID getId() const {
+    assert(Kind == VI_Id && "Not an Id type");
----------------
`getGUID` might be clearer at call sites?


http://reviews.llvm.org/D18763