[llvm] [InferAttrs] malloc-family libcalls may write `errnomem` location (PR #164382)

[Antonio Frighetto via llvm-commits]

antoniofrighetto wrote:

I have been wondering whether it might be desirable to move this forward, notwithstanding PR #170290 possibly blocking this. Would it make sense to prioritize fixing the long-standing issues that stem from the current interaction between malloc/errno, at the slightly unfortunate cost of generating suboptimal code for the time being (at least, until the other PR is resumed)? The regressions from llvm-opt-benchmark do not appear to be excessively worrying, as far as I can tell. (Should this be discussed in the RFC instead?)

https://github.com/llvm/llvm-project/pull/164382