[llvm] Globalopt pass produces invalid debug info (PR #100654)

[Jeremy Morse via llvm-commits]

jmorse wrote:

Interesting; we were aiming for identical behaviour between the two formats to avoid any uncertainty about it's correctness. I believe we're not aiming to keep support for intrinsic-format in the future, but if there's changing behaviour then this area wants testing and more examination (will examine it more on Monday).

https://github.com/llvm/llvm-project/pull/100654