HyperLink   A Framework for Profile-Driven Optimization in the IMPACT Binary Reoptimization System.
Publication Year:
  Matthew C. Merten
  MS thesis, Department of Electrical and Computer Engineering, University of Illinois, Urbana IL, June 1999

Since the invention of the computer, higher application performance has been in incessant demand. Even orders-of-magnitude performance improvement has not appeased users as they continue to dream of new, more demanding applications. This pursuit has led to aggressive development at every level of computer design, from semiconductor to compiler technologies. Improved semiconductor technology contributes higher die densities, which allows for an increase in the number of transistors while also allowing higher clock frequencies. Computer architectures now may allow coarse-grained parallelization, speculation, and predication. Advanced computer microarchitectures are able to exploit one-grained parallelism more effectively while also executing instructions more efficiently than in previous microarchitectures. Memory systems have improved to reduce the average latency of accesses to memory by including multiple levels of caching, and by adding new types of caches, such as victim caches. Compilers provide aggressive classical and instruction-level-parallelism optimizations and pointer analysis to reduce the number of instructions executed, and to provide in-depth analysis of the application. These and other improvements have all lead to exponential growth of computer performance.