Haupt-Navigation

Publications

Interprocedural Speculative Optimization of Memory Accesses to Global Variables
Citation key	gesellensetter08interprocedural0
Author	Lars Gesellensetter and Sabine Glesner
Title of Book	European Conference on Parallel and Distributed Computing (Euro-Par 2008)
Year	2008
Publisher	LNCS
Abstract	The discrepancy between processor and memory speed, also known as memory gap, is steadily increasing. This means that execution speed is more and more dominated by memory accesses. We investigate the use of globals, which reside inherently in memory, in standard applications and present an approach to reduce the number of memory accesses, thereby reducing the effect of the memory gap. Our approach can explicitly deal with uncertain information and, hence, optimize more aggressively with the help of speculative techniques while not changing the semantics of the optimized programs. We present an implementation of the proposed optimization in our compiler framework for the Intel Itanium and show that our techniques lead to an increased performance for the SPEC CPU2006 benchmarks, thus showing that the impact of the memory gap can be effectively mitigated with advanced speculative optimization.
Bibtex Type of Publication	Conference Article