PARROT: Power Awareness Through Selective Dynamically Optimized Traces

Trace cache sampling filter

“…This result has significance of its own. Several works have proposed to use hardware to optimize frequent code on the fly [10] [15]. By increasing the TCU, the sampling filter ensures that optimized code will be reused many times prior to its replacement.…”

Specialized dynamic optimizations for high-performance energy-efficient microarchitecture

“…For more details on the simulation methodology and comparative results (e.g. breakdown of energy/power consumption of different microarchitectural components) we refer the reader to [29].…”

“…The essential challenge within this domain is the increasingly poor scaling of performance with power consumption. The PARROT (the Power-Aware aRchitecture Running Optimized Traces) microarchitecture proposes a trace-cachebased, decoupled mechanism for handling the frequently executed code -as initially reported in [29]. Within this microarchitectural framework we focus on the performance and energy-saving potential of dynamic optimizations performed on the most frequent traces in program execution.…”