As the complexity requirements for embedded applications increase, the performance
demands of embedded compilers also increase. Compiler optimizations, such as software
pipelining and recurrence elimination, can significantly reduce execution time for applications, but these transformations require the use of additional registers to hold data values across one or more loop iterations. Compilers for embedded systems have difficulty exploiting these optimizations since they typically do not have enough registers on an embedded processor to be able to apply the transformations. In this paper, we evaluate a new application configurable processor utilizing several different register structures which can enable these optimizations without increasing the architecturally addressable register storage requirements. Using this approach can lead to an improved execution time through enabled optimizations and reduced register pressure for embedded architectures.
