[llvm] GlobalOpt: Handle threadlocal.address intrinsic (PR #88454)

[Matthias Braun via llvm-commits]

MatzeB wrote:

For the record: I believe there is no additional problems with coroutines (functions changing thread-id) with this optimization given `GlobalOpt` computes summaries for the whole module independent of control flow anyway.

https://github.com/llvm/llvm-project/pull/88454