Calendar

Current Events

3.30

Creating a Sustainable Society

3.16

Sustainable Energy Engineering

2.28

National Science Foundation Day

2.17

Reformulation-Based Approaches to Query Optimization

2.15

FREEDM Day

2.10

WEST: Cloning Data Cache Behavior using Stochastic Traces

2.09

Computer Vision and Image Analysis in the Study of Art

2.08

GSA Social

2.03

Skewed functional processes and their applications

EXACT: Explicit Dynamic-Branch Prediction with Active Updates

Abstract:
Many branch predictors use global branch or path history to implicitly distinguish between different dynamic instances of a branch, and specialize predictions to these different instances. State-of-the-art predictors are impressive in pushing the envelope of this implicit approach. A majority of mispredictions that remain, however, are due to this context not properly distinguishing different dynamic branches for which specialized outcomes are needed, especially memory-dependent branches. Explicitly specializing predictions for these dynamic branches is necessary but not sufficient, as they are not adapted as stores occur. This paper describes a new direction in branch predictor design for attacking these two interrelated problems. We propose a predictor that explicitly distinguishes dynamic branches based on their dependent memory addresses. The predictor constructs a dynamic-branch control-flow-graph (DB-CFG) whose nodes are these dynamic branches. Each node contains the outcome of the corresponding dynamic branch and two next-pointers to other nodes in the graph. The fetch unit simply replays the graph, following next-pointers according to the cached taken/not-taken outcomes. Stores to the memory addresses on which a dynamic branch depends, directly update its cached outcome. This novel “active update” concept avoids future mispredictions that might otherwise be incurred by conventional passive training and also automatically adapts the path that will be replayed through the graph.

Bio:
Muawya Al-Otoom completed his Bachelors degree in Computer Engineering from Jordan University of Science and Technology (JUST) in 2003, and Masters Degree in Computer Engineering from North Carolina State University (NCSU) in 2006. He is currently working under Prof. Eric Rotenberg toward his PhD degree at NCSU. His research topic is developing novel paradigms for branch prediction. During the last two years, Muawya interned with Intel and IBM where he got the chance to work on the development of future microprocessors and to co-author a patent on a novel technique to improve branch prediction accuracy. Muawya is a member of Phi Kappa Phi and IEEE.