help: bypass a type of loop optimization in LLVM front end

Tim and guys,

let's look at the following example: