1 / 14

An Ahead Pipelined Alloyed Perceptron with Single Cycle Access Time

An Ahead Pipelined Alloyed Perceptron with Single Cycle Access Time. David Tarjan (CS) Kevin Skadron (CS) Mircea R. Stan (ECE) University of Virginia. The whole presentation in one slide. Perceptron: Best predictor yet, basic unit is table plus adder

leif
Download Presentation

An Ahead Pipelined Alloyed Perceptron with Single Cycle Access Time

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. An Ahead Pipelined Alloyed Perceptron with Single Cycle Access Time David Tarjan (CS) Kevin Skadron (CS) Mircea R. Stan (ECE) University of Virginia

  2. The whole presentation in one slide • Perceptron: Best predictor yet, basic unit is table plus adder • Ahead pipeline it and precompute, so it has eff. latency 1 and shorter pipeline • Accuracy does not suffer:

  3. Talk Outline • Motivation & Perceptron Theory • Related Work • Ahead Pipelining a Perceptron • Precomputing a local hist. perceptron • Results • Conclusion

  4. Motivation Main Problems in Branch Prediction: • Accuracy (larger tables, more logic) • Latency (smaller tables, less logic) • Multiple Branch/Trace/Stream/etc. per cycle We address these points Tradeoff!

  5. Motivation • Two recent innovations (both by Jiménez): • Perceptrons: simplest neural network, better accuracy than any previously known predictor(MICRO’01) • Ahead pipelining: Start prediction early, hide latency of prediction, allows access to large tables(HPCA’03) • Use of alloyed history: reduces wrong history branch mispredictions(PACT’00)

  6. What’s a Perceptron?: The short version

  7. Related Work • Perceptron Branch Predictor: lots of work by Jiménez(HPCA’01, MICRO’03) • Perceptron continued: new work by Seznec(TR 1554, 1620) and Ipek (Cornell) • Ahead: Gshare.fast(Jiménez, HPCA’03), EV8 Bpred Adaption(Seznec, ISCA’03) • Precomputation: Original 2-level Branch Predictor Paper(Yeh & Patt, Micro’91)

  8. Normal pipelined perceptron

  9. Ahead pipelined perceptron

  10. Precomputing a local hist. perceptron

  11. Results

  12. Results

  13. Conclusion • Ahead Pipelining solves latency problem • Precomputation allows use of local history in a fast predictor • Alloyed Perceptron may offer better accuracy

  14. Questions?

More Related