P. Amestoy et al. (Eds.): Euro-Par’99, LNCS 1685, pp. 1291-1295, 1999.  Springer-Verlag Berlin Heidelberg 1999 Implementation of Hybrid Context Based Value Predictors Using Value Sequence Classification † Luis Piñuel, Rafael A. Moreno, and Francisco Tirado Departamento de Arquitectura de Computadores y Automática, Universidad Complutense, Madrid 28040, Spain {lpinuel,rmoreno,ptirado}@dacya.ucm.es Abstract. Value prediction is as yet a very novel technique, whose efficiency has still to be proved. To take advantage of this emerging technique in the short term it is essential to design accurate and low cost value predictors. This work presents a new approach of implementing hybrid predictors that allows the maximum sharing of information between predictors. We show that the new hybrid predictor outperforms not only the accuracy of the others predictors, but also their hardware utilization. 1 Introduction Over the last few years, many of the efforts in microarchitecture research have been focussed on attempting to counteract the program dependencies and thus improve the extractable instruction-level parallelism (ILP). Several techniques for eliminating control and data dependencies have been proposed, based mainly on branch and data prediction and speculative execution. While branch prediction [1], can be considered today as a widely accepted classical technique, value prediction, however, is as yet a very novel technique, whose efficacy has still to be proved. The value prediction technique, like branch prediction, allows temporal violation of the program constraints without affecting its semantics. However, if the predictions are not accurate, the use of speculation is detrimental for processor performance. It is obvious that the more history the predictor captures the more accuracy it has. However, the use of large history tables is not realistic for the next generations of processors. Consequently, to take advantage of the value prediction in the short term it is essential to design accurate and low cost value predictors. In this paper, we first study the relationship between accuracy and cost for different value predictors. Subsequently, we propose a new methodology for designing cost- effective hybrid data value predictors that allows us to combine the benefits of different predictors, without increasing the amount of hardware. The rest of the paper is organized as follows. Section 2 summarizes the previous work on value prediction. Section 3 describes the experimental framework. Section 4 presents a comparative analysis of actual value predictors. Section 5 introduces the low-cost hybrid predictor proposed by the authors. Finally, section 6 presents the conclusions. † This work has been supported by the Spanish Ministry of Education under grant TIC96-1071.