[llvm] LoopVectorize: Set branch_weight for conditional branches (PR #72450)

[Matthias Braun via llvm-commits]

MatzeB wrote:

I agree that this is 3) in your categories and it's a guessing game where we likely can be way off depending on program and program inputs. However I also do not see any immediately better answers for this right now.

I am currently comparing estimated execution counts with actual counts and this one for example turns `llvm-test-suite/SingleSource/Benchmarks/Shootout/ary3.c` (which admittedly is a really small microbenchmark) from misjudging the counts by 2x to misjudging by 2% ...

https://github.com/llvm/llvm-project/pull/72450