Do optimizations based on if we know an expression is 0 or !=0, even though we don't know anything else.
Build DAGs (basically find all the common subexpressions). Must be done after all other optimizations, because most of them depend on the trees being trees, not DAGs. The general strategy is: Compute available expressions (AEs) For each block stick together nodes that match, keeping AEs up to date For each block unstick unprofitable common subexpressions (this is generally target-dependent)
Directed acyclic graphs and global optimizer common subexpressions
Compiler implementation of the D programming language.