Value Prediction is one of the newest techniques used to break down ILP lim
its. Despite being under continuous study during the last few years, a few
aspects related to this emerging technique remain unanalysed in depth. Exha
ustively investigated in the context of control speculation, confidence est
imation has usually played a secondary role on value prediction and specula
tion. Closely linked to confidence estimation, value substitution also repr
esents a relegated subject of research. This paper is focussed on analysing
, in an isolated way, the respective impact on predictor performance of bot
h confidence estimation and value substitution mechanisms. By using detaile
d pipeline-level simulations, we prove that improvements in these mechanism
s are as important as reducing the predictor aliasing or even improving the
prediction model. (C) 2001 Elsevier Science B.V. All rights reserved.