Load Value Prediction
Predicting values loaded by machine instructions to expose more instruction-level parallelism.
Data dependencies can severely hinder Instruction Level Parallelism (ILP). They reduce ILP when long-latency instructions flow through the pipeline and there are not enough independent instructions to keep the processor busy; dependent instructions stall behind them, creating critical paths through the program.
Load Value Prediction (LVP) breaks the data dependency on load instructions by predicting the value the instruction will fetch. It exploits value locality, the likelihood of a previously-seen value recurring in a storage location. In this project we explore value locality, exploit it to perform load value prediction, and analyze performance on the SPEC2006 benchmarks.