Runtime extraction of memory access information from the application source code

The shift towards using increasing numbers of processing elements has placed new burdens on the programming community to fully exploit the potential performance gain of multiprocessor systems. The programming problem is even more complex in the case of systems that utilize reconfigurable devices. The increased complexity of programming necessitates the use of tools that can support programmers in migrating existing applications to these emerging systems. Programmers need increasingly sophisticated tools for profiling and analysis of applications. Particularly, tools to inspect the memory access behavior of applications become crucial due to the processor/memory communication bottleneck. In this paper, we present xQUAD, a unique extension to the QUAD dynamic profiling toolset, which augments the memory access analysis of an application by providing detailed, fine-grained intra-function information. xQUAD provides detailed memory access information on the application source code data object granularity. This information can help programmers for application optimization and revision. We also present a ranking method based on the memory access intensity of a function, which reveals more accurately the suitability of the function for hardware implementation. xQUAD is tested on a real application from the multimedia domain to describe the capabilities of the proposed toolset.