[llvm] [CodeGenPrepare] Unmerging GEPs across indirect branches must respect types (PR #68587)

[Maurice Heumann via llvm-commits]

momo5502 wrote:

> High level comment: Would bailing out of the transform if the source element types don't match be "good enough"?
> 
> I'd prefer not to add the extra logic if we don't particularly care about making it work with mismatching GEP types. (Longer term this will be automatically fixed by migration to ptradd.)

I agree. I initially chose to bail out, but thought it could be interpreted as the "lazy way" and feared the PR might get rejected, hence the fix.
However, under the aspect of future migration to ptradd, it doesn't make sense to introduce a fix for something that is going to become obsolete sooner or later.
I will adjust it and revert to bailing out tomorrow.

https://github.com/llvm/llvm-project/pull/68587