Managing the energy-performance tradeoff has become a major challenge with embedded systems. The cache hierarchy is a typical example where this tradeoff plays a central role. With the increasing level of integration density, a cache can feature millions of transistors, consuming a significant portion of the energy. At the same time however, a cache also permits to significantly improve performance. Configurable caches are becoming the "de-facto" solution to deal efficiently with these issues. Such caches are equipped with artifacts that enable one to resize it dynamically. With regard to embedded systems, however, many of these artifacts restrict the configurability at the application level. We propose in this paper to modify the structure of a configurable cache to offer embedded compilers the opportunity to reconfigure it according to a program dynamic phase, rather than on a per-application basis. We show in our experimental results that the proposed scheme has a potential for improving the compiler effectiveness to reduce the energy consumption, while not excessively degrading the performance.