Performance Estimation: IPC

Performance estimation of a processor can easily lead to a large saving in time, which would otherwise be spent on time consuming simulations. Poor choice of processor can lead to an expensive design due to either the choice of an expensive processor or due to a large hardware as the selected processor severely under performs. In this work, we estimate the IPC (instructions per cycle) of the processor for a particular application. We propose a technique to estimate the upper and lower limit of IPC for a given application. Experimental results indicate that IPC can be estimated within about 20% accuracy without using time consuming instruction set simulators.

1. Introduction

A typical embedded program runs on a processor with parts of the program running in hardware so that performance requirements are met. It is important to select a processor whose performance is close to the performance requirements of the application. Selecting a processor which under-performs by a huge margin means a very large piece of hardware have to be used to meet the requirements. On the other hand, an over-performing processor would be very expensive which would reflect in the higher unit cost of the product. The current estimating methods rely heavily on compiling and then executing the application on a instruction set simulator. This is a time consuming process and even more so when you have to consider a large number of processors and have to evaluate their performance. There is the added cost of the development tools itself.