Writing a dead code elimination pass

In this recipe, you will learn how to eliminate dead code from the program. By dead code elimination, we mean removing the code that has no effect whatsoever on the results that the source program outputs on executing. The main reasons to do so are reduction of the program size, which makes the code quality good and makes it easier to debug the code later on; and improving the run time of the program, as the unnecessary code is prevented from being executed. In this recipe, we will show you a variant of dead code elimination, called aggressive dead code elimination, that assumes every piece of code to be dead until proven otherwise. We will see how to implement this pass ourselves, and what modifications we need to make so that the pass can run just like other passes in the lib/Transforms/Scalar folder of the LLVM trunk.

Getting ready

To show the implementation of dead code elimination, we will need a piece of test code, on which we will run the aggressive...