אירועים והרצאות בפקולטה למדעי המחשב ע"ש הנרי ומרילין טאוב

Massively parallel, throughput-oriented processors are becoming increasingly common. Maximizing the benefit of these processors requires algorithms to be implemented differently than the sequential algorithms that most software developers are familiar with. This change is often very time-consuming and is not guaranteed to give an increase in performance matching the amount of effort. This work proposes an approach to predict the performance gain from porting an algorithm from the CPU to the GPU, based only on the measurements performed on the CPU implementation. CPU performance counters are measured on the sequential implementation and fed to a simple machine learning model trained on a given set of benchmarks. The experiments performed in this work have shown that the predictions' accuracy is similar or better than existing methods, that usually require converting the algorithm as a starting point. (Talk will be given in Hebrew)