Value-Driven Code Motion

this paper focuses on lazy code motion as proposed by Knoop, Ruthing, and Steffen and modified by Drechsler and Stadel [14, 15, 11]. That algorithm is provably optimal; this paper shows that by changing our assumptions about the shape of the input program, we can produce a technique that both eliminates more redundancies and runs more efficiently. We present experimental data that shows both these effects. 1 Introductio