Locality driven dynamic cache write policy on graphics processing units

Locality Driven Dynamic Write Policy on GPU Caches switches to appropriate write policy by sampling locality densities at intervals of an application with the suggested locality detector. Different write policies being used in GPU architectures has their own advantages compared to other policies. The most appropriate cache write policy at runtime can vary depending on the type of application and hardware resources. Present day GPU architectures use a fixed cache policy which is decided to be used before run time. On this study, we showed that it is possible to increase the performance of applications by using a dynamic cache write policy instead of a fixed cache policy on a cycle level GPU performance simulator.