Bridging Inputs and Program Dynamic Behavior

Program optimizations have evolved from static to dynamic. However, runtime optimization often suffers from not knowing global behavior of a program's execution, and not affording sophisticated program analysis. On the other hand, offline profiling techniques overcome both obstacles but are oblivious to the effects of program inputs. This work attempts to address the problem by developing a framework including three components: an extensible Input Characterization Language (XICL), statistical feature selection, and regression model building. The first two components extract important attributes from program inputs, and the last one builds models between those attributes and program dynamic behavior.