[llvm] [SROA] Prevent poison propagation in insertInteger() (PR #109240)

[Nikita Popov via llvm-commits]

https://github.com/nikic requested changes to this pull request.

Our current de-facto model is that memory cannot contain poison values. I'm not willing to accept patches handling poison in memory before we have reached an actual consensus on the broader uninitialized memory question. Two of the more recent discussions of interest are https://discourse.llvm.org/t/rfc-add-nopoison-attribute-metadata/79833 and https://discourse.llvm.org/t/rfc-load-instruction-uninitialized-memory-semantics/67481.

Freezing all bit insertions in SROA is a very big hammer, and exactly one of the things we'd rather avoid doing.

https://github.com/llvm/llvm-project/pull/109240