Gleipnir: A memory tracing and profiling tool

Embedded and high performance applications often require fine-tuning to improve their performance. This is achieved using analysis tools that provide insights into the application’s behavior. A common approach is to instrument the application code and observe its behavior during an actual execution on a target system. In this paper we describe a program profiling and tracing tool called Gleipnir. Gleipnir is built as a plug-in tool to a widely used binary instrumentation framework called Valgrind. Gleipnir can be used to trace memory accesses and associate each access with a specific program internal structure such as threads, functions, local, global, and dynamic data structures, and scalar variables.This ability makes Gleipnir a good candidate for advanced memory performance tuning. The data provided by Gleipnir may be used by trace-driven simulators, such as cache simulators to analyze accesses to data structure elements so that programmers can understand how the memory access patterns are impacting the execution time of the application. The programmer may then be able to change data layouts or reorder code to change the access patterns and eliminate performance bottlenecks. The goal of Gleipnir is to give information rich traces that can be used by any number of advanced memory analysis tools, particularly cache simulators. Our claim is that despite advances in allocation techniques and data reordering, detailed dynamic and static memory behavior of applications is often not readily available or available only in terms of statistical average accesses and cache miss rates. It is our hypothesis that optimizing cache performance at all levels is very important to improving the performance of applications running on singlecore and multi-core processors. In this paper we will describe Gleipnir and provide examples of on how the output of Gleipnir can be used by cache