|
      
|
|
|
|
As the current rate of improvement in processor performance far exceeds the rate of improvement in memory performance, memory latency can be the dominant overhead in many performance critical applications.
In many cases, automatic compiler-based optimizations to improve memory performance are limited and programmers frequently resort to manual optimization techniques. However this process is tedious and time-consuming.
Furthermore a diverse range of a rapidly evolving hardware makes the optimization process even more complicated. It is often hard to predict the potential benefit from different optimizations and there is no simple criteria to stop optimizations i.e. when optimal memory performance has been achieved or sufficiently approached.
Here we propose a platform independent technique that is both fast and reasonably accurate for estimating the memory performance upper-bound of many scientific applications and presents a platform independent optimisation approach based on feedback-directed program restructuring.
We have developed strategies that search the optimisation space by means of profiling to find the best possible program variant. These strategies have no a priori knowledge of the target machine and can be run on any platform. Before applying search strategy we estimate the memory performance upper-bound on a loop level to reduce the search space.
Those techniques have been tested on a wide range of programs from the SPEC benchmark suite, kernels and commercial applications and have been used to guide a manual optimization process and iterative compilation.
We show that our approach is able to give better or similar significant performance improvements over a state of the art high level restructurer based on static analysis and a platform specific profile feedback directed compiler.
