From Sequences of Dependent Instructions to Functions

Publisher Website
Google Scholar
Abstract
In this article, we present an approach for improving the performance of sequences of dependent instructions. We observe that many sequences of instructionscan be interpreted as functions. Unlike sequences of instructions, functions can be translated into very fast butexponentially costly two-level combinational circuits. Wepresent an approach that exploits this principle, speeds upprograms thanks to circuit-level parallelism/redundancy,but avoids the exponential costs.We analyze the potential of this approach, and thenwe propose an implementation that consists of a superscalar processor with a large specific functional unit associated with specific back-end transformations. The performance of the SpecInt2000 benchmarks and selectedprograms from the Olden and MiBench benchmark suitesimproves on average from 2.4% to 12% depending on thelatency of the functional units, and up to 39.6%; moreprecisely, the performance of optimized code sections improves on average from 3.5% to 19%, and up to 49%.

This publication has 15 references indexed in Scilit: