Analysis of memory accesses in embedded systems

The increasing number of multimedia applications for embedded systems shows designers the limits of classical memory systems when dealing with large vectors of data. Therefore, a new type of cache memory specifically targeted for processing arrays and large sets of data has been proposed. In order to justify a new cache memory architecture, two assessments are needed: an evaluation of the performance gain for specific code, and an indication of the modifications required in the software. This work presents theory and implementation details for a memory access analysis tool that enables evaluation of vector operations of particular software and identification of code segments that generate them. The tool has been developed assuming a MIPS 3000 microprocessor target system, and its capabilities have been benchmarked on execution traces from two multimedia software applications: djpeg image conversion and rawaudio audio file compression