In the particular out-of-tree code I'm trying to fix up the issue isn't that you don't know what function the blocks will go into, more that you don't know at the point when the blocks are created whether they will end up getting inserted into the function at all, or where. Imagine something like:

I did some refactoring here years ago but I'm not that familiar with the ABIs or the handling in clang.

LGTM.

Accidentally created a new revision rather than updating https://reviews.llvm.org/D39339.

Address review feedback regarding using void* as a key.

Apply the original changes to top of tree.

I'll take a look at getting this or something like it in.

• [GISel] Address review feedback on passing MD_callees to lowerCall.

@mssimpso Is this something you're still interested in pushing forward?

LGTM. Thanks for following up with this!

Update the test to ensure both functions exist rather that grepping the output of -stats.

Thanks for taking a look!

I don't think this ever went out to llvm-commits, so I'm going to re-open with llvm-commits as a subscriber when opening.