Debug info in optimized code - how far can we go? Improving LLVM debug info with function entry val…