Single SimPoints
 

Home
Up
Contents

 

 

The goal of Single SimPoint is to find the best single  simulation point that accurately representatives the complete execution of a program.   The single simulation point typically represents a phase transition, so the single simulation point typically captures the behavior of two dominate phases.  Since programs have many phases, the single simulation point will have a higher error rate than if multiple points are used.  None the less, the single simulation point is fairly accurate and is used by many researchers for guiding their architecture design tradeoffs. 

Another option for using a single simulation point is to use the largest weighted simulation point from the multiple simulation points below.  This can provide an accurate representation of a large fraction of the program's execution.

The interval size used to gather these results is 100 million execution instructions, each simulation point chosen is given as the number of the intervals, starting with 0. For example, SimPoint of 0 represents instructions 1 - 100 million in the programs execution.

The interval size used to create these simulation points was 100 million instructions.  Each simulation point provided is given as the number of the intervals from the start of execution. The first interval of execution is indicated by a Simulation Point of 0 (this is an important change from SimPoint 2.0). For example, if the Simulation Point is 90, then you start simulating at instruction 90 * 100 million (9 billion) and stop simulating at instruction 9.1 billion assuming a 100 million interval size was used to create the simulation points.

Also given for each Simulation Point is the PC of the executed instruction that started that interval to be simulated.  You then only need to fast-forward until that instruction PC is encountered the number of times shown in the last column (xBB).  Then start simulating for the interval length of instructions. Using the simulation point starting PC and execution count is the recommended approach for using SimPoint.

Standard Single Simulation Points

bullet

The 2nd column shows the Simulation Point chosen using the original ASPLOS 2002 SimPoint 1.1 algorithm. 

Early Single Simulation Points

bullet

A single simulation point for 100 million instructions was chosen early in the program. The above table shows the Early Single SimPoint 2.0 algorithm, and the percent IPC error of that point when compared to the full program's execution. The last column shows the percent of execution you would need to fast forward through to get to the single SimPoint.

 

Home ] Up ]

Send mail to calder@cs.ucsd.edu with questions or comments about this web site.
Last modified: 06/22/05