Value Reuse Optimization: Reuse of Evaluated Math

Value reuse technique eliminates the redundant evaluation of expressions, using the support of hardware at runtime to eliminate them. The potential performance of a value reuse mechanism not only depends on the number of instances it has eliminated, but as well as on the time spent in detecting the instance to be eliminated. In order to have better value reuse, we move from instruction level reuse to function reuse, so that one can detect more eliminative instances with reasonable e#ort. This removes the extra time spent on detecting the possibility of instruction execution elimination. The results of the mathematical library functions evaluated can be reused. Since they can be treated as instructions, i.e, functions which do not have side-e#ects with the application, one can eliminate the function call itself. In this paper, we propose a technique to re-use the evaluated values for math library functions. Execution time is decreased up to 6%. Reduction in the number of calls due to instrumentation of code for function reuse for math library routines is up to 99% at a particular call-site in the application