Online Memory Access Pattern Analysis on an Application Profiling Tool

As memory subsystems have become complex in the state of the art system architectures, application program codes required to be optimized targeting to their deeper memory hierarchy for rewarding their performance. To support such optimizations, we are developing a memory access pattern analysis tool. In this paper, we present the methodology how we detect memory access patterns on-the-fly on an execution-driven application analysis tool called Exana. First, we implement an offline trace file based method using a Python script code and verify its functionalities. Then, in order to improve its analysis speed, the code is ported to C++ language programs and integrated in the Exana. We evaluate the time and memory usage for the analysis of each implementation. From the results, we confirmed our online implementation can process faster than the offline trace file based method.