Disjoint Eager Execution (DEE) Animation

Reference: DEE Tutorial; also see: DEE Simulation Data

NOTE: It will take a minute or two for the images to load. (Next time, I'll just have it draw the lines. :-) )

An animation of the order of resource assignment in DEE is now presented. The basic assumptions are the same as those in the figure in the DEE Tutorial, especially: predicted paths go to the left, and not-predicted paths go to the right; and: a Branch Prediction Accuracy (BPA) of 70% is used. The latter was chosen for illustrative purposes; DEE ideas still hold for other, more common, BPA's. ILP speedups in the 10's have been demonstrated in simulations, assuming a BPA of about 90%. The data indicate that DEE is likely to improve upon SP execution even with better BPA's; there will always be some fraction of branches which are unpredictable, and due to an Amdahl's Law type of effect, their unpredictability will highly constrain ILP without the use of DEE.

A Java applet simulation of a DEE tree normally appears here.
There is some difficulty running it on your system.

Further Reading

A tutorial on branch effect reduction techniques is available as the paper: Branch Effect Reduction Techniques. A research paper on DEE is: Disjoint Eager Execution: An Optimal Form of Speculative Execution .
hits since August 5, 1996.

April 25, 1997 | Gus Uht | uht@ele.uri.edu